filename
stringlengths
17
38
gt_das
stringlengths
41
7.52k
genshare_das
stringlengths
10
46.3k
md_content
stringlengths
0
359k
healing_error
stringclasses
4 values
10.34133_2022_9870149.pdf
Code Availability. The source codes for our CS framework are available at https://github.com/bianzhiyu/ContinuityScaling.
Additional Points Code Availability. The source codes for our CS framework are available at https://github.com/bianzhiyu/ContinuityScaling .
AAAS Research Volume 2022, Article ID 9870149, 10 pages https://doi.org/10.34133/2022/9870149 Research Article Continuity Scaling: A Rigorous Framework for Detecting and Quantifying Causality Accurately Xiong Ying,1,2,3 Si-Yang Leng ,2,4 Huan-Fei Ma ,5 Qing Nie and Wei Lin 1,2,3,8 ,6 Ying-Cheng Lai ,7 1School of Mathematical Sciences, SCMS, and SCAM, Fudan University, Shanghai 200433, China 2Research Institute for Intelligent Complex Systems, CCSB, and LCNBI, Fudan University, Shanghai 200433, China 3State Key Laboratory of Medical Neurobiology and MOE Frontiers Center for Brain Science, Institutes of Brain Science, Fudan University, Shanghai 200032, China 4Institute of AI and Robotics, Academy for Engineering and Technology, Fudan University, Shanghai 200433, China 5School of Mathematical Sciences, Soochow University, Suzhou 215006, China 6Department of Mathematics, Department of Developmental and Cell Biology, And NSF-Simons Center for Multiscale Cell Fate Research, University of California, Irvine, CA 92697-3875, USA 7School of Electrical, Computer, And Energy Engineering, Arizona State University, Tempe, Arizona 85287-5706, USA 8Shanghai Artificial Intelligence Laboratory, Shanghai 200232, China Correspondence should be addressed to Wei Lin; wlin@fudan.edu.cn Received 7 March 2022; Accepted 24 March 2022; Published 4 May 2022 Copyright © 2022 Xiong Ying et al. Exclusive Licensee Science and Technology Review Publishing House. Distributed under a Creative Commons Attribution License (CC BY 4.0). Data-based detection and quantification of causation in complex, nonlinear dynamical systems is of paramount importance to science, engineering, and beyond. Inspired by the widely used methodology in recent years, the cross-map-based techniques, we develop a general framework to advance towards a comprehensive understanding of dynamical causal mechanisms, which is consistent with the natural interpretation of causality. In particular, instead of measuring the smoothness of the cross-map as conventionally implemented, we define causation through measuring the scaling law for the continuity of the investigated dynamical system directly. The uncovered scaling law enables accurate, reliable, and efficient detection of causation and assessment of its strength in general complex dynamical systems, outperforming those existing representative methods. The continuity scaling-based framework is rigorously established and demonstrated using datasets from model complex systems and the real world. 1. Introduction Identifying and ascertaining causal relations are a problem of paramount importance to science and engineering with broad applications [1–3]. For example, accurate detection of causation is the key to identifying the origin of diseases in precision medicine [4] and is important to fields such as psychiatry [5]. Traditionally, associational concepts are often misinterpreted as causation [6, 7], while causal analysis in fact goes one step further beyond association in a sense that, instead of using static conditions, causation is induced under changing conditions [8]. The principle of Granger causality formalizes a paradigmatic framework [9–11], quantifying causality in terms of prediction improvements, but, because of its linear, multivariate, and statistical regression nature, the various derived methods require extensive data [12]. Entropy-based methods [13–20] face a similar difficulty. Another issue with the Granger causality is the fundamental requirement of separability of the underlying dynamical var- iables, which usually cannot be met in the real world sys- tems. To overcome these difficulties, the cross-map-based techniques, paradigms tailored to dynamical systems, have been developed and have gained widespread attention in the past decade [21–36]. 2 Research The cross-map is originated from nonlinear time series analysis [37–42]. A brief understanding of such a map is as follows. Consider two subsystems: X and Y. In the recon- structed phase space of X, if for any state vector at a time a set of neighboring vectors can be found, the set of the cross-mapped vectors, which are the partners with equal time of X, could be available in Y. The cross-map underlying the reconstructed spaces can be written as Y t = ΦðXtÞ (where Xt and Y t are delay coordinates with sufficiently large dimensions) for the case of Y unidirectionally causing X while, mathematically, its inverse map does not exist [34]. In practice, using the prior knowledge on the true cau- sality in toy models or/and the assumption on the expanding property of Φ (representing by its Jacobian’s singular value larger than one in the topological causality framework [24]), scientists developed many practically useful tech- niques based on the cross-map for causality detection. For instance, the “activity” method, originally designed to mea- sure the continuity of the inverse of the cross-map, com- pares the divergence of the cross-mapped vectors to the state vector in X with the divergence of the independently- selected neighboring vectors to the same state vector [22, 23]. The topological causality measures the divergence rate of the cross-mapped vectors from the state vectors in Y [24], and the convergent cross-mapping (CCM), increasing the length of time series, compares the true state vector Y with the average of the cross-mapped vectors, as the estima- tion of Y [21, 25–36]. Then, the change of the divergence or the accuracy of the estimation is statistically evaluated for determining the causation from Y to X. Inversely, the causa- tion from X to Y can be evaluated in an analogous manner. The above evaluations [21, 24, 26–36] can be understood at a conceptional and qualitative level and perform well in many demonstrations. In this work, striving for a comprehensive understanding of causal mechanisms and inspired by the cross-map-based techniques, we develop a mathematically rigorous frame- work for detecting causality in nonlinear dynamical systems, turning eyes towards investigating the original systems from their cross-maps, which is also logically consistent with the natural interpretation of causality as functional dependences [2, 8]. The skills used in cross-map-based methods are assimilated in our framework, while we directly study the original dynamical systems or the reconstructed systems instead of the cross-maps. The foundation of our framework is the scaling law for the changing relation of ε with δ arising from the continuity for the investigated system, henceforth the term “continuity scaling”. In addition to providing a the- ory, we demonstrate, using synthetic and real-world data, that our continuity scaling framework is accurate, computa- tionally efficient, widely applicable, showing advantages over the existing methods. 2. Continuity Scaling Framework To explain the mathematical idea behind the development of our framework, we use the following class of discrete time dynamical systems: xt+1 = fðxt, ytÞ and yt+1 = gðxt, ytÞ for t ∈ ℕ, where the state variables fxtgt∈ℕ, fytgt∈ℕ evolve in the compact manifolds M, N of dimension DM, DN under sufficiently smooth map f, g, respectively. We adopt the common recognition of causality in dynamical systems. Definition 1. If the dependence of fðx, yÞ on y is nontrivial (i. e., a directional coupling exists), a variation in y results in a change in the value of fðx, yÞ for any given x, which, accord- ing to the natural interpretation of causality [2, 43], admits that y : fytgt∈ℕ has a direct causal effect on x : fxtgt∈ℕ, denoted by y↪x, as shown in the upper panel of Figure 1(a). We now interpret the causal relationship stipulated by ð·Þ ≜ fðxg, ·Þ for a given the continuity of a function. Let fxg ∈ N , we denote its image under ∈ M. For any yP point xg ≜ fxg the given function by xI ðyPÞ. Applying the logic state- ment of a continuous function to fxg ð·Þ, we have that, for any neighborhood OðxI, εÞ centered at xI and of radius ε > 0, there exists a neighborhood OðyP, δÞ centered at yP of radius δ > 0, such that fxg ðOðyP, δÞÞ ⊂ OðxI, εÞ. The neigh- borhood and its radius are defined by O p, hð Þ = s ∈ S distS f j s, pð Þ < h g, p ∈ S, h > 0, ð1Þ where distSð·, · Þ represents an appropriate metric describ- ing the distance between two given points in a specified manifold S with S = M or N . The meaning of this mathe- matical statement is that, if we have a neighborhood of the resulting variable xI first, we can then find a neighborhood for the causal variable yP to satisfy the above mapping and inclusion relation. This operation of “first-ε-then-δ” pro- vides a rigorous base for the principle that the information about the resulting variable can be used to estimate the information of the causal variable and therefore to ascertain causation, as indicated by the long arrow in the middle panels of Figure 1(a). Note that, the existence of the δ > 0 neighborhood is always guaranteed for a continuous map . In fact, due to the compactness of the manifold N , a fxg largest value of δ exists. However, if yP does not have an explicit causal effect on the variable xI, i.e., fxg is independent of yP, the existence of δ is still assured but it is independent of the value of ε, as shown in the upper panel of Figure 1(b). This means that merely determining the existence of a δ- neighborhood is not enough for inferring causation - it is necessary to vary ε systematically and to examine the scaling relation between δ and ε. In the following we discuss a num- ber of scenarios. Case I. Dynamical variables fðxt, ytÞgt∈ℕ are fully measur- x > 0, the set fxτ ∈ Mjτ ∈ It able. For any given constant ε ðε xÞg can be used to approximate the neighborhood Oðxt+1, ε xÞ, where the time index set is x It x ε xð Þ ≜ τ ∈ ℕ distMj f ð xt+1, xτ Þ < ε x g: ð2Þ The radius δt y = δt yðε xÞ of the neighborhood Oðyt, δt yÞ Research 3 (a) (b) Figure 1: Illustration of causal relation between two sets of dynamical variables. (a) Existence of causation from y in N to x in M, where each correspondence from xt+1 to yt is one-to-one, represented by the line or the arrow, respectively, in the upper and the middle panels. In y (the lower panel) with ε this case, a change in ln ε y denoting the neighborhood size of the resulting variable x and of the causal variable y, respectively. (b) Absence of causation from y to x, where every point on each trajectory, fytg, in N could be the correspondent point from xt+1 in M (the upper panel) and thus every point in N belongs to the largest δ-neighborhood of yt (the middle panel). In this case, δ x (the lower panel). Also refer to the supplemental animation for illustration. x results in a direct change in δ y does not depend on ε x and δ satisfying fxg=xt ðOðyt, δt yÞÞ ⊂ Oðxt+1, ε xÞ can be estimated as n h Þ ≜ # (cid:2)It x ε xð Þ δt y ε xð i o−1 〠 τ∈(cid:2)It x ε xð Þ distN yt, yτ−1 ð Þ, ð3Þ xg. xðε xÞ ≜ fτ ∈ It where #½·(cid:2) is the cardinality of the given set and the index set is (cid:2)It xÞjdistMðxt, xτ−1Þ < ε xðε The strict mathematical steps for estimating δt y are given in Section II of Supplementary Information (SI). We empha- size that here correspondence between xt+1 and yt is investi- gated, differing from the cross-map-based methods, with one-step time difference naturally arising. This consider- ation yields a key condition [DD], which is only need when considering the original iteration/flow and whose detailed description and universality are demonstrated in SI. We reveal a linear scaling law between hδt yit∈ℕ and ln ε x, as shown in the lower panels of Figure 1, whose slope s y↪x is an indicator of the correspondent relation between ε and δ and hence the causal relation y↪x. Here, h·it∈ℕ denotes the average over time. In particular, a larger slope value implies a stronger causation in the direction from y to x as represented by the map functions fðxt, ytÞ (Figure 1(a)), while a near zero slope indicates null causation in this direc- tion (Figure 1(b)). Likewise, possible causation in the reversed direction, x↪y, as represented by the function gð xt, ytÞ, can be assessed analogously. And the unidirectional case when fðx, yÞ = f0ðxÞ independent of y is uniformly con- sidered in Case II. We summarize the consideration below and an argument for the generic existence of the scaling law is provided in Section II of SI. Theorem 2. For dynamical variables fðxt, ytÞgt∈ℕ measured directly from the dynamical systems, if the slope s y↪x defined above is zero, no causation exists from y to x. Otherwise, a directional coupling can be confirmed from y to x and the slope s y↪x increases monotonically with the coupling strength. Case II. The dynamical variables fðxt, ytÞgt∈ℕ are not directly accessible but measurable time series futgt∈ℕ and fvtgt∈ℕ are available, where ut = uðxtÞ and vt = vðytÞ with u: M ⟶ ℝru and v: N ⟶ ℝrv being smooth observational functions. To assess causation from y to x, we assume one- dimensional observational time series (for simplicity): ru = rv = 1, and use the classical delay-coordinate embedding method [37–42, 44] to reconstruct the phase space: ut = T ðut, ut+τu,⋯,ut+ðdu−1ÞτuÞ and vt = ðvt, vt+τv ,⋯,vt+ðdv−1Þτv Þ , where τu,v is the delay time and du,v > 2ðDM + DN Þ is the embedding dimension that can be determined using some standard criteria [45]. As illustrated in Figure 2, the dynam- ical evolution of the reconstructed states fðut, vtÞgt∈ℕ is gov- erned by T ut+1 = ~ f ut, vt ð Þ, vt+1 = ~g ut, vt ð Þ: ð4Þ The map functions can be calculated as ∘ E−1 ~ fðu, vÞ ≜ Eu ∘ ∘ E−1 ∘ E−1 u ðuÞ, Π u 1 ∘ E−1 v ðvÞÞ, where the embedding (diffeomorphism) v ðvÞÞ, ~gðu, vÞ ≜ Ev ∘ ½f, g(cid:2)ðΠ 1 2 ½ f, g(cid:2)ðΠ ðuÞ, Π 2 4 Research (f(x, y), g(x, y)) M × N M × N y↪x or x↪y ) u ( 1 u – E ° 1 П ) v ( 1 – v E ° 2 П ) y , x ( u E ) y , x ( v E ˜ ˜ Lu × Lv ˜ ˜ Lu × Lv v↪u or u↪v ˜ ˜ (f(u,v),g(u,v)) {ut(x)}t𝜖N {vt(y)}t𝜖N Figure 2: Illustration of system dynamics before and after embedding for Case II. In the left panel, the arrows describe how ~ f, ~gÞ after the original systems ðf, gÞ is equivalent to the system ð embedding. In the right panel, causation between the internal variables x and y can be ascertained by detecting the causation between the variables u and v reconstructed from measured time series. Es: M × N ⟶ ~L s ⊂ ℝds with given by ~L s ≜ EsðM × N Þ, s = u or v, is Eu x, y ð   f, g½ Ev x, y ð   f, g½ ∘ 1 ∘ f, g½ Þ ≜ uð xð Þ, u ∘ Π (cid:2)2τu x, y ð Þ ≜ vð yð Þ, v ∘ Π 2 (cid:2)2τv x, y ð τu x, y (cid:2) ð du−1 Þ, ⋯, u ∘ Π ∘ f, g½ (cid:2) 1 τv x, y ∘ f, g½ (cid:2) ð dv−1 ∘ f, g½ ð Þ, ⋯, v ∘ Π Þ, u ∘ Π 1 Þτu x, y ð Þ, v ∘ Π 2 Þτv x, y ð (cid:2) ð ÞÞ, ∘ ÞÞ, 2 ð5Þ k ~L s, ½f, g(cid:2) s defined on 1ðx, yÞ = x and Π with the inverse function E−1 represent- ing the kth iteration of the map and the projection mappings as Π 2ðx, yÞ = y. Case II has now been reduced to Case I, and our continuity scaling framework can be used to ascertain the causation from v to u based on the measured time series with the indices It uÞ and s v↪u (equations (2) and (3)). uÞ, δt uðε vðε ~ f0ðutÞ and vt+1 = ~gðut, vtÞ, where Does the causation from v to u imply causation from y to x? The answer is affirmative, which can be argued, as follows. If the original map function f is independent of y: fðx, yÞ = f0ðxÞ, there is no causation from y to x. In this case, the embedding Euðx, yÞ becomes independent of y, degenerating into the form of Euðx, yÞ = Eu0ðxÞ, a diffeomorphism from M ~L u0 = Eu0ðMÞ only. As a result, equation (4) becomes to ~ ∘ f ∘ E−1 ut+1 = f0ðuÞ = Eu0 u0ð ~ f0 is independent of v. The uÞ and the resulting mapping independence can be validated by computing the slope v↪u associated with the scaling relation between hδt s vit∈ℕ and ln ε u, where a zero slope indicates null causation from v to u and hence null causation from y to x. Conversely, a finite slope signifies causation between the variables. Thus, any type of causal relation (unidirectional or bi-directional) detected variables fðut, vtÞgt∈ℕ implies the same type of causal relation between the internal but inaccessible variables x and y of the original system. reconstructed between state the T Case III. The structure of the internal variables is completely unknown. Given the observational functions ~u, ~v: M × N ⟶ ℝ with ~ut = ~uðxt, ytÞ and ~vt = ~vðxt, ytÞ, we first recon- struct the state space: ~ut = ð~ut, ~ut+τ,⋯,~ut+ðd−1ÞτÞ and ~vt = ð~vt, ~vt+τ,⋯,~vt+ðd−1ÞτÞ . To detect and quantify causation from ~v to ~u (or vice versa), we carry out a continuity scaling analysis with the modified indices It ~vðε~uÞ and s~v↪~u. Differing from Case II, here, due to the lack of knowledge about the correspondence structure between the internal and observational variables, a causal relation for the latter does not definitely imply the same for the former. ~uðε~uÞ, δt T Case IV. Continuous-time dynamical systems possessing a sufficiently smooth flow fSt ; t ∈ ℝg on a compact manifold H : dStðu0Þ/dt = χðStðu0ÞÞ, where χ is the vector field. Let f̂ut=ωn+νgn∈ℤ and f̂vt=ωn+νgn∈ℤ be two respective time series from the smooth observational functions ̂u, ̂v: H ⟶ ℝ with ̂ut = ̂uðStÞ and ̂vt = ̂vðStÞ, where 1/ω is the sampling rate and ν is the time shift. Defining Ξ ≜ Sω: H ⟶ H and ̂Sn ≜ Sωn+νðu0Þ, we obtain a discrete-time system as ̂Sn+1 = Ξð̂SnÞ with the observational functions as ̂un = ̂uð̂SnÞ and ̂vn = ̂vð ̂SnÞ, reducing the case to Case III and rendering applicable our continuity scaling analysis to unveil and quantify the causal relation between f̂ut=ωn+νgn∈ℤ and f̂vt=ωn+νgn∈ℤ. If the domains of ̂u and ̂v have their own restrictions on some particular subspaces, e.g., ̂u: H u ⟶ ℝ and ̂v: H v ⟶ ℝ with H = H u ⊕ H v, the case is further reduced to Case II, so the detected causal relation between the observational variables imply causation between the internal variables belonging to their respective subspaces. 3. Demonstrations: From Complex Dynamical Models to Real-World Networks To demonstrate the efficacy of our continuity scaling frame- work and its superior performance, we have carried out extensive numerical tests with a large number of synthetic and empirical datasets, including those from gene regulatory networks as well as those of air pollution and hospital admission. The practical steps of the continuity scaling framework together with the significance test procedures are described in Methods. We present three representative examples here, while leaving others of significance to SI. 2,t+1 = x The first example is an ecological model of two unidirec- tionally interacting species: x 1,tð3:8 − 3:8x 1,t − μ 1,t+1 = x 12 x 2,t − μ 2,tÞ and x x 2,tð3:7 − 3:7x 1,tÞ. With time series 2,tÞgt∈ℕ obtained from different values of the cou- 1,t, x fðx pling parameters, our continuity scaling framework yields correct results of different degree of unidirectional causa- tion, as shown in Figures 3(a) and 3(b). In all cases, there exists a reasonable range of ln εx (neither too small nor too large) from which the slope sx of the linear scaling can be extracted. The statistical significance of the estimated slope values and consequently the strength of causation can be assessed with the standard p-value test [46] (Methods and SI). An ecological model with bidirectional coupling has also been tested (see Section III of SI). Figures 3(c) and 3(d) ↪x 21 2 1 2 Research 5 0.6 0.4 t 〉 1 x 𝛿 〈 t 0.2 0.6 0.4 t 〉 2 x 𝛿 〈 t 0.2 0 –8 –6 –2 0 –4 ln 𝜀x2 0 –8 –6 –2 0 –4 ln 𝜀x1 𝜇21 = 0.00 𝜇21 = 0.05 𝜇21 = 0.10 𝜇21 = 0.15 (a) 𝜇12 = 0.00 𝜇12 = 0.00 𝜇12 = 0.00 𝜇12 = 0.00 (b) t c e ff E x5 x4 x3 x2 x1 j x ↪ x s i e p o l S t c e ff E x5 x4 x3 x2 x1 0.12 0.1 0.08 0.06 0.04 0.02 0 j x ↪ x s i e p o l S 0.15 0.12 0.09 0.06 0.03 0 x1 x2 x4 x5 x3 Cause (d) x1 x2 x4 x5 x3 Cause (c) Figure 3: Ascertaining and characterizing causation in various ecological systems of interacting species. (a, b) Unidirectional causation of two coupled species. In (a), the values of the slope sx 2 are approximately 0.0004, 0.1167, 0.1203, and 0.1238 for four different values of the coupling parameter μ indicating its nonexistence. (c, d) Inferred causal network of five species whose interacting structure is, respectively, that of a ring: xi↪xi+1ðmod 5Þ (i = 1, ⋯, 5) and of a tree: x j↪xj+1,j+3 (j = 1, 2), where the estimated slope values are color-coded. Results of a statistical analysis of the accuracy and reliability of the determined causal interactions are presented in SI Section III. Time series of length 5000 are used in all these simulations. The embedding parameters are τs = 1 and ds = 3 with s = x 21. (b) Near zero slope values sx associated with the causal relation x for x ↪x ↪x 1, ↪x ↪x 1 2 2 1 1 2 1, ⋯, x 5. show the results from ecological networks of five mutually interacting species on a ring and on a tree structure, respec- tively, where the color-coded slope values reflect accurately the interaction patterns in both cases. The second example is the coupled Lorenz system: _xi = σiðyi − xiÞ + μijx j, _yi = xiðρi − ziÞ − yi, _zi = xiyi − βizi with i, j = 1, 2 and i=j. We use time series fy 2,tgt=nω for detecting different configurations of causation (see Section III of SI). Figure 4 presents the overall result, where the color-coded estimated values of the slope from the continu- ity scaling are shown for different combinations of the sam- pling rate 1/ω and coupling strength. Even with relatively low sampling rate, our continuity scaling framework can successfully detect and quantify the strength of causation. Note that the accuracy does not vary monotonously with the sampling rate, indicating the potential of our framework 1,t, y to ascertain and quantify causation even with rare data. Moreover, the proposed index can accurately reflect the true causal strength (denoting by the coupling parameter), which is also evidenced by numerical tests in Sections III and IV of SI. Robustness tests against different noise perturbations are provided in Section III of SI demonstrating the practical usefulness of our framework. Additionally, analogous to the first example, we present in SI several examples on cau- sation detection in the coupled Lorenz system with nonlin- ear couplings, and the Rössler-Lorenz system, etc., which further demonstrates the generic efficacy of our framework. In addition, we present study on several real-world data- set, which brings new insights to the evolutionary mecha- nism of underlying systems. We study gene expression data from DREAM4 in silico Network Challenge [47, 48], whose intrinsic gene regulatory networks (GRNs) are known for verification (Figure 5(a) and Figure S17 of SI). Applying 6 Research 1 2 𝜇 6 4 2 0 5 4 3 2 1 0 2 y ↪ 1 y s e p o l S 1 2 𝜇 6 4 2 0 1 y ↪ 2 y s e p o l S 5 4 3 2 1 0 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Sampling duration/0.001 (a) 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Sampling duration/0.001 (b) Figure 4: Detecting causation in the unidirectionally coupled Lorenz system. The results are for different values of μ sampling rate 1/ω. (a, b) Color-coded values of the slopes sy embedding parameters are ds = 7, τs ≈ 0:05 with ωjτs (s = y including the time series lengths used in the simulations. 12 = 0) and , respectively. The integration time step is 10−3 and the and sy 2). See Section III and Table S9 of SI for all the other parameters 1 or y 21 (μ ↪y ↪y 1 2 1 2 100 45 44 43 1 0.8 0.6 0.4 0.2 e t a r e v i t i s o p e u r T 0 0 75 36 69 37 67 25 85 38 10 62 72 96 23 83 87 73 Enhancer Inhibitor (a) 0.2 0.4 0.6 0.8 1 False positive rate 1, AUROC = 0.765 2, AUROC = 0.667 3, AUROC = 0.693 4, AUROC = 0.693 5, AUROC = 0.868 (b) Figure 5: Detecting causal interactions in five GRNs. (a) One representative GRN containing 20 randomly selected genes. Other four structures can be found in Figure S17 of SI. (b) The ROC curves as well as their AUROC values demonstrate the efficacy of our framework. our framework to these data, we ascertain the causations between each pair of genes by using the continuity scaling framework. The corresponding ROC curves for five different networks as well as their AUROC values are shown in Figure 5(b), which indicates a high detection accuracy in dealing with real-world data. We then test the causal relationship in a marine ecosys- tem consisting of Pacific sardine landings, northern anchovy landings and sea surface temperature (SST). We reveal new findings to support the competing relationship hypothesis stated in [49] which cannot be detected by CCM [25]. As pointed out in Figure 6, while common influence from SST to both species is verified with both methods, our continuity scaling additionally illuminates notable influence from anchovy to sardine with its reverse direction being less sig- nificant. While competing relationship plays an important role in ecosystems, continuity scaling can reveal more essen- tial interaction mechanism. See Section III.E of SI for more details. Moreover, we study the transmission mechanism of the recent COVID-19 pandemic. Particularly, we analyze the daily new cases of COVID-19 of representative countries for two stages: day 1 (January 22 nd 2020) to day 100 (April 30 th 2020) and day 101 (May 1 st 2020) to day 391 (February 15 th 2021). Our continuity scaling is pairwisely applied to reconstruct the transmission causal network. As shown in Figure 7, China shows a significant effect on a few countries at the first stage and this effect disappears at the second stage. However, other countries show a different situation with China, whose external effect lasts as shown in Section III.E and Figure S18 of SI. Our results accord with that China holds stringent epidemic control strategies with Research 7 SST SST Sardine Anchovy Sardine Anchovy Continuity scaling CCM Figure 6: The comparison of causal network structure detected by continuity scaling and CCM among sea surface temperature, sardine, and anchovy. Figure 7: The causal effect from China to other countries of the COVID-19 pandemic detected by continuity scaling between stages 1 and 2. Here, stage 1 is from January 22 nd 2020 to April 30 th 2020, and stage 2 is from May 1 st 2020 to February 15 th 2021. For those detected causal links between all countries, refer to Section III.E and Figure S18 of SI. These maps are for illustration only. sporadic domestic infections, as evidenced by official daily briefings, demonstrating the potential of continuity scaling in detecting causal networks for ongoing complex systems. Additionally, We emphasize that day 100 is a suitable critical day to distinguish the early severe stage and the late well-under-control stage of the pandemic (see Figure S18(a) of SI), while slight change of the critical day will not nullify our result. As shown in Figure S18(b) of SI, when the critical day varies from day 94 to day 106, no significant change (less than 5%) of the detected causal links occurs at both stages, and the number of countries under influence of China at Stage 2 remains zero. See more details in Section III.E of SI. Additional real world examples including air pollutants and hospital admission record from Hong Kong are also shown in Section III of SI. 4. Discussion To summarize, we have developed a novel framework for data-based detection and quantification of causation in com- plex dynamical systems. On the basis of the widely used cross-map-based techniques, our framework enjoys a rigor- ous foundation, focusing on the continuity scaling law of the concerned system directly instead of only investigating the continuity of its cross-map. Therefore, our framework is consistent with the standard interpretation of causality, and works even in the typical cases where several existing typical methods do not perform that well or even they fail (see the comparison results in Section IV of SI). In addition, the mathematical reasoning leading to the core of our frame- work, the continuity scaling, helps resolve the long-standing issue associated with techniques directly using cross-map that information about the resulting variables is required to project the causal variables, whereas several works in the literature [50], which directly studied the continuity or the smoothness of the cross-map, likely yielded confused detected results on causal directions. Computational complexity. The computational com- plexity of the algorithm is OðT 2NεÞ, which is relatively smaller than the CCM method, whose computational com- plexity is OðT 2 log TÞ. the dynamical behavior of Limitations and future works. Nevertheless, there are still some spaces for improving the presently proposed framework. First, currently, only bivariate detection algo- rithm is designed, so generalization to multivariate network inference requires further considerations, as analogous to those works presented in Refs. [51–53]. Second, the causal time delay has not been taken into account in the current framework, so it also could be further investigated, similar to the work reported in Ref. [33]. Also, more advanced algo- rithms, such as the one developed in Ref. [54], could be 8 Research integrated into this framework for detecting those temporal causal structures. Definitely, we will settle these questions in our future work. Detecting causality in complex dynamical systems has broad applications not only in science and engineering, but also in many aspects of the modern society, demanding accurate, efficient, and rigorously justified and hence trust- worthy methodologies. Our present work provides a vehicle along this feat and indeed resolves the puzzles arising in the use of those influential methods. 5. Methods Continuity scaling framework: a detailed description of algo- rithms. Let futgt=1,2,⋯,T and fvtgt=1,2,⋯,T be two experimen- tally measured time series of internal variables fðxt, ytÞgt∈ℕ. Typically, if the dynamical variables fðxt, ytÞgt∈ℕ are accessi- ble, fðut, vtÞg reduce to one-dimensional coordinate of the internal system. The key computational steps of our conti- nuity scaling framework are described, as follows. We reconstruct the phase space using the classical method of delay coordinate embedding [37] with the opti- mal embedding dimension dz and time lag τz determined by the methods in Refs. [55, 56] (i.e., the false nearest neigh- bors and the delayed mutual information, respectively): n (cid:2) L z ≜ z tð Þ = zt, zt+τz , ⋯, zt+ dz−1 ð Þτz (cid:3) j o t = 1, ⋯, T 0 , ð6Þ z = u, v, T where Euclidean distance is used for both L u,v. 0 = min fT − ðdz − 1Þτzjz = u, vg, and We present the steps for causation detection using the case of v↪u as an example. We calculate the respective diameters for L u,v as (cid:4) Dz ≜ max distLz z tð Þ, z τð Þ ð Þ 1 ≤ t, τ ≤ T j (cid:5) , 0 ð7Þ where z = u, v, and z = u, v. We set up a group of numbers, fε u,N ε = Du, with the other ele- u,jgj=1,⋯,N ε ments satisfying u,1 = e · Du, ε , as ε ln ε u,j − ln ε j − 1 u,1 = ln ε − ln ε u,N ε Nε − 1 u,1 , ð8Þ for j = 2, ⋯, N ε − 1. Then, in light of (2) with (3), we have δt v ε uð (cid:6) Þ = # It u ε uð Þ (cid:7)−1 〠 τ∈It u ε uð Þ distL v v tð Þ, v τ − 1 ð ð Þ Þ, ð9Þ with It u ε uð (cid:4) (cid:8) (cid:8) Þ = τ ∈ ℕ distLu u t + 1 ð Þ, u τð Þ ð Þ < ε u, t + 1 − τ j (cid:5) j > E ð10Þ where numerically, ε set fε u,jgj=1,⋯,N ε u alters its value successively from the , and the threshold E is a positive number 0 0 vðε chosen to avoid the situation where the nearest neighboring points are induced by the consecutive time order only. u,jÞg vðε u, where ℕT As defined, hδt uÞit∈ℕ is the average of δt vðε uÞ over all t. We use a finite number of pairs possible time fðhδt , ln ε vðε u,jÞit∈ℕT to approximate the scaling j=1,⋯,N ε relation between hδt uÞit∈ℕ and ln ε = f1, 2,⋯,T 0g. Theoretically, a larger value of N ε and a smaller value of e will result in a more accurate approximation of the scaling relation. In practice, the accuracy is determined by the length of the observational time series, the sampling duration, and different types of noise perturbations. In numerical simulations, we set e = 0:001 and N ε = 33. In addi- tion, a too large or a too small value of ε u can induce insuffi- cient data to restore the neighborhood and/or the entire manifold. We thus set δt u,jÞ = δt u,j+1Þ as a practical tech- nique as the number of points is limited practically in a small neighborhood. As a result, near zero slope values would appear on both sides of the scaling curve hδt uÞit∈ℕ-ln ε u, as demonstrated in Figure 3 and in SI. In such a case, to esti- mate the slope of the scaling relation, we take the following approach. vðε vðε vðε Define a group of numbers by (cid:11) (cid:9) − δt v − ln ε Sj ≜ ln ε u,j+1 δt v t∈ℕT (cid:11) (cid:12) (cid:10) ε 0 u,j+1 u,j (cid:9) (cid:10) (cid:12) ε u,j t∈ℕT 0 , ð11Þ where j = 1, ⋯, Nε − 1, sort them in a descending order, from which we determine the ½Nε + 1/2(cid:2) largest numbers, collect their subscripts - j’s together as an index set ̂J, and set H ≜ fj, j + 1jj ∈ ̂Jg. Applying the least squares method to the linear regression model: (cid:11) (cid:12) δt v ε uð Þ t∈ℕ = S · ln ε u + b ð12Þ with the dataset fðhδt optimal values ð̂S, finally obtain the slope of the scaling relation as s , we get the ̂bÞ for the parameters ðS, bÞ in (12) and u,jÞit∈ℕT , ln ε ≜ ̂S. u,jÞg vðε j∈H 0 v↪u For the other causal direction from u to v, these steps are equally applicable to estimating the slope s u↪v. 0 To assess the statistical significance of the numerically determined causation, we devise the following surrogate test using the case of v causing u as an illustrative example. Divide the time series fuðtÞgt∈ℕT into NG consecutive segments of equal length (except for the last segment - the shortest segment). Randomly shuffle these segments and then regroup them into a surrogate sequence f̂uðtÞgt∈ℕT . Applying such a random permutation method to fvðtÞgt∈ℕT generates another surrogate sequence f̂vðtÞgt∈ℕT . Carrying out the slope computation yields ŝv↪̂u. The procedure can be repeated for a sufficient number of times, say Q, which consequently yields a group of estimated slopes, denoted as fsq ̂v↪̂u is set as s v↪u obtained from the original time series. For all the estimated slopes, we calculate ̂v↪̂ugq=0,1⋯,Q, where s0 0 0 0 Research 9 their mean bμ -value for s v↪u is calculated as v↪u and the standard deviation bσ v↪u. The p (cid:13) s ps v↪u ≜ 1 − normcdf (cid:14) , v↪u ð13Þ v↪u bσ − bμ v↪u where normcdf ½·(cid:2) is the cumulative Gaussian distribution function. The principle of statistical hypothesis testing guar- antees the existence of causation from v to u if ps < 0:05. In simulations, we set the number of segments to be N G = 25 and the number of times for random permutations to be Q ≥ 20. v↪u Additional Points Code Availability. The source codes for our CS framework are available at https://github.com/bianzhiyu/ContinuityScaling. Conflicts of Interest The authors declare no competing interests. Authors’ Contributions W.L. conceived idea. X.Y., S.-Y.L., and W.L. designed and performed the research. X.Y., S.-Y.L., H.-F.M., and W.L. analyzed the data. H.-F.M., Y.-C.L., and Q.N. contributed data and analysis tools, and all the authors wrote the paper. X.Y. and S.-Y.L. equally contributed to this work. Acknowledgments W.L. is supported by the National Key R&D Program of China (Grant No. 2018YFC0116600), by the National Natu- ral Science Foundation of China (Grant Nos. 11925103 and 61773125), by the STCSM (Grant No. 18DZ1201000), and by the Shanghai Municipal Science and Technology Major Project (No. 2021SHZDZX0103). Y.-C.L. is supported by AFOSR (Grant No. FA9550-21-1-0438). S.-Y.L. is supported by the National Natural Science Foundation of China (No. 12101133) and “Chenguang Program” supported by Shang- hai Education Development Foundation and Shanghai Municipal Education Commission (No. 20CG01). Q.N. is partially supported by NSF (Grant No. DMS1763272) and the Simons Foundation (Grant No. 594598). H.-F.M. is sup- ported by the National Natural Science Foundation of China (Grant No. 12171350) and by the National Key R&D Pro- gram of China (Grant No. 2018YFA0801100). Supplementary Materials Supplementary materials: SI.pdf (where we include analytic and computational details of the results in the main text. This SI is helpful but not essential for understanding the main results of the paper.) (Supplementary Materials) References [1] M. Bunge, Causality and Modern Science, Routledge, 2017. [2] J. Pearl, Causality, Cambridge university press, 2013. [3] J. Runge, S. Bathiany, E. Bollt et al., “Inferring causation from time series in earth system sciences,” Nature Communications, vol. 10, no. 1, p. 2553, 2019. [4] F. S. Collins and H. Varmus, “A new initiative on precision medicine,” New England Journal of Medicine, vol. 372, no. 9, pp. 793–795, 2015. [5] G. N. Saxe, A. Statnikov, D. Fenyo et al., “A complex systems approach to causal discovery in psychiatry,” PloS One, vol. 11, no. 3, article e0151174, 2016. [6] D. R. Cox and D. V. Hinkley, Theoretical Statistics, CRC Press, 1979. [7] T. M. Cover, Elements of Information Theory, John Wiley & Sons, 1999. [8] J. Pearl, “Causal inference in statistics: an overview,” Statistics Surveys, vol. 3, pp. 96–146, 2009. [9] N. Wiener, The Theory of Prediction, Modern mathematics for engineers, 1956. [10] C. W. Granger, “Investigating causal relations by econometric models and cross-spectral methods,” Econometrica: Journal of the Econometric Society, vol. 37, no. 3, pp. 424–438, 1969. [11] S. Haufe, V. V. Nikulin, K. R. Müller, and G. Nolte, “A critical assessment of connectivity measures for EEG data: a simula- tion study,” NeuroImage, vol. 64, pp. 120–133, 2013. [12] M. Ding, Y. Chen, and S. L. Bressler, “Granger causality: basic theory and application to neuroscience,” Handbook of Time Series Analysis: recent theoretical developments and applica- tions, vol. 437, 2006. [13] T. Schreiber, “Measuring information transfer,” Physical Review Letters, vol. 85, no. 2, pp. 461–464, 2000. [14] S. Frenzel and B. Pompe, “Partial mutual information for cou- pling analysis of multivariate time series,” Physical Review Let- ters, vol. 99, no. 20, article 204101, 2007. [15] R. Vicente, M. Wibral, M. Lindner, and G. Pipa, “Transfer entropy–a model-free measure of effective connectivity for the neurosciences,” Journal of Computational Neuroscience, vol. 30, no. 1, pp. 45–67, 2011. [16] J. Runge, J. Heitzig, V. Petoukhov, and J. Kurths, “Escaping the curse of dimensionality in estimating multivariate transfer entropy,” Physical Review Letters, vol. 108, no. 25, article 258701, 2012. [17] J. Sun, C. Cafaro, and E. M. Bollt, “Identifying the coupling structure in complex systems through the optimal causation entropy principle,” Entropy, vol. 16, no. 6, pp. 3416–3433, 2014. [18] C. Cafaro, W. M. Lord, J. Sun, and E. M. Bollt, “Causation entropy from symbolic representations of dynamical systems,” Chaos: An Interdisciplinary Journal of Nonlinear Science, vol. 25, article 043106, 2015. [19] J. Sun, D. Taylor, and E. M. Bollt, “Causal network inference by optimal causation entropy,” SIAM Journal on Applied Dynam- ical Systems, vol. 14, no. 1, pp. 73–106, 2015. [20] M. Solyanik-Gorgone, J. Ye, M. Miscuglio, A. Afanasev, A. E. Willner, and V. J. Sorger, “Quantifying information via Shan- non entropy in spatially structured optical beams,” Research, vol. 2021, article 9780760, 2021. [21] Y. Hirata and K. Aihara, “Identifying hidden common causes from bivariate time series: a method using recurrence plots,” Physical Review E, vol. 81, no. 1, article 016203, 2010. 10 Research [22] R. Q. Quiroga, J. Arnhold, and P. Grassberger, “Learning driver-response relationships from synchronization patterns,” Physical Review E, vol. 61, no. 5, pp. 5142–5148, 2000. [23] J. Arnhold, P. Grassberger, K. Lehnertz, and C. E. Elger, “A robust method for detecting interdependences: application to intracranially recorded eeg,” Physica D: Nonlinear Phenomena, vol. 134, no. 4, pp. 419–430, 1999. [24] D. Harnack, E. Laminski, M. Schünemann, and K. R. Pawelzik, “Topological causality in dynamical systems,” Physical Review Letters, vol. 119, no. 9, article 098301, 2017. [25] G. Sugihara, R. May, H. Ye et al., “Detecting causality in com- plex ecosystems,” Science, vol. 338, no. 6106, pp. 496–500, 2012. [26] E. R. Deyle, M. Fogarty, C. H. Hsieh et al., “Predicting climate effects on pacific sardine,” Proceedings of the National Acad- emy of Sciences, vol. 110, no. 16, pp. 6430–6435, 2013. [27] X. Wang, S. Piao, P. Ciais et al., “A two-fold increase of carbon cycle sensitivity to tropical temperature variations,” Nature, vol. 506, no. 7487, pp. 212–215, 2014. [28] H. Ma, K. Aihara, and L. Chen, “Detecting causality from non- linear dynamics with short-term time series,” Scientific Reports, vol. 4, pp. 1–10, 2014. [29] J. M. McCracken and R. S. Weigel, “Convergent cross- mapping and pairwise asymmetric inference,” Physical Review E, vol. 90, no. 6, article 062903, 2014. [30] H. Ye, E. R. Deyle, L. J. Gilarranz, and G. Sugihara, “Distin- guishing time-delayed causal interactions using convergent cross mapping,” Scientific Reports, vol. 5, no. 1, article 14750, 2015. [31] A. T. Clark, H. Ye, F. Isbell et al., “Spatial convergent cross mapping to detect causal relationships from short time series,” Ecology, vol. 96, no. 5, pp. 1174–1181, 2015. [32] J.-J. Jiang, Z.-G. Huang, L. Huang, H. Liu, and Y.-C. Lai, “Directed dynamical influence is more detectable with noise,” Scientific Reports, vol. 6, no. 1, article 24088, 2016. [33] H. Ma, S. Leng, C. Tao et al., “Detection of time delays and directional interactions based on time series from complex dynamical systems,” Physical Review E, vol. 96, no. 1, article 012221, 2017. [34] J. M. Amigó and Y. Hirata, “Detecting directional couplings from multivariate flows by the joint distance distribution,” Chaos: An Interdisciplinary Journal of Nonlinear Science, vol. 28, article 075302, 2018. [35] Y. Wang, J. Yang, Y. Chen, P. De Maeyer, Z. Li, and W. Duan, “Detecting the causal effect of soil moisture on precipitation using convergent cross mapping,” Scientific Reports, vol. 8, no. 1, pp. 1–8, 2018. [36] S. Leng, H. Ma, J. Kurths et al., “Partial cross mapping elimi- influences,” Nature Communications, nates indirect causal vol. 11, no. 1, pp. 1–9, 2020. [37] F. Takens, “Detecting strange attractors in turbulence,” in Dynamical Systems and Turbulence, Warwick 1980, Springer, 1981. [38] N. H. Packard, J. P. Crutchfield, J. D. Farmer, and R. S. Shaw, “Geometry from a time series,” Physical Review Letters, vol. 45, no. 9, pp. 712–716, 1980. [39] T. Sauer, J. A. Yorke, and M. Casdagli, “Embedology,” Journal of Statistical Physics, vol. 65, no. 3-4, pp. 579–616, 1991. [40] J. Stark, “Delay embeddings for forced systems. I. Determinis- tic forcing,” Journal of Nonlinear Science, vol. 9, no. 3, pp. 255– 332, 1999. [41] J. Stark, D. S. Broomhead, M. E. Davies, and J. Huke, “Delay embeddings for forced systems. II. Stochastic forcing,” Journal of Nonlinear Science, vol. 13, no. 6, pp. 519–577, 2003. [42] M. R. Muldoon, D. S. Broomhead, J. P. Huke, and R. Hegger, “Delay embedding in the presence of dynamical noise,” Dynamics and Stability of Systems, vol. 13, no. 2, pp. 175– 186, 1998. [43] P. Spirtes, C. N. Glymour, R. Scheines, and D. Heckerman, Causation, Prediction, and Search, MIT press, 2001. [44] B. Cummins, T. Gedeon, and K. Spendlove, “On the efficacy of state space reconstruction methods in determining causality,” SIAM Journal on Applied Dynamical Systems, vol. 14, no. 1, pp. 335–381, 2015. [45] H. Kantz and T. Schreiber, Nonlinear Time Series Analysis, vol. 7, Cambridge university press, 2004. [46] G. Lancaster, D. Iatsenko, A. Pidde, V. Ticcinelli, and A. Stefanovska, “Surrogate data for hypothesis testing of phys- ical systems,” Physics Reports, vol. 748, pp. 1–60, 2018. [47] D. Marbach, T. Schaffter, C. Mattiussi, and D. Floreano, “Gen- erating realistic in silico gene networks for performance assess- ment of Journal Computational Biology, vol. 16, no. 2, pp. 229–239, 2009. [48] D. Marbach, R. J. Prill, T. Schaffter, C. Mattiussi, D. Floreano, and G. Stolovitzky, “Revealing strengths and weaknesses of methods for gene network inference,” Proceedings of the National Academy of Sciences, vol. 107, no. 14, pp. 6286– 6291, 2010. engineering methods,” reverse of [49] R. Lasker and A. Mac Call, “New ideas on the fluctuations of the clupeoid stocks off California,” Proceedings of the Joint Oceanographic Assembly, Halifax, August 1982: General Sym- posia, Department of Fisheries and Oceans, Ontario, 1983. [50] M. L. V. Quyen, J. Martinerie, C. Adam, and F. J. Varela, “Non- linear analyses of interictal EEG map the brain interdepen- dences in human focal epilepsy,” Physica D: Nonlinear Phenomena, vol. 127, no. 3-4, pp. 250–266, 1999. [51] J. Peters, D. Janzing, and B. Schölkopf, Elements of Causal Inference: Foundations and Learning Algorithms, MIT Press, 2017. [52] J. Runge, “Causal network reconstruction from time series: from theoretical assumptions to practical estimation,” Chaos: An Interdisciplinary Journal of Nonlinear Science, vol. 28, arti- cle 075310, 2018. [53] Y. Lou, L. Wang, and G. Chen, “Enhancing controllability through redirecting robustness of q-snapback networks edges,” Research, vol. 2019, article 7857534, 23 pages, 2019. [54] J.-W. Hou, H.-F. Ma, D. He, J. Sun, Q. Nie, and W. Lin, “Har- vesting random embedding for high-frequency change-point detection in temporal complex,” National Science Review, vol. 9, article nwab228, 2022. [55] A. M. Fraser and H. L. Swinney, “Independent coordinates for strange attractors from mutual information,” Physical Review A, vol. 33, no. 2, pp. 1134–1140, 1986. [56] M. B. Kennel, R. Brown, and H. D. Abarbanel, “Determining embedding dimension for phase-space reconstruction using a geometrical construction,” Physical Review A, vol. 45, no. 6, pp. 3403–3411, 1992.
null
10.1088_1361-6501_ad180c.pdf
Data availability statement The data cannot be made publicly available upon publication because they are owned by a third party and the terms of use prevent public distribution. The data that support the findings of this study are available upon reasonable request from the authors.
Data availability statement The data cannot be made publicly available upon publication because they are owned by a third party and the terms of use prevent public distribution. The data that support the findings of this study are available upon reasonable request from the authors.
Meas. Sci. Technol. 35 (2024) 035605 (12pp) Measurement Science and Technology https://doi.org/10.1088/1361-6501/ad180c Automated defect detection in precision forging ultrasonic images based on deep learning Jianjun Zhao1, Yuxin Zhang1, Xiaozhong Du1,3,∗ and Xiaoming Sun2 1 School of Mechanical Engineering, Taiyuan University of Science and Technology, Taiyuan 030024, People’s Republic of China 2 College of Mechanical and Electrical Engineering, Central South University, Changsha 410083, People’s Republic of China 3 School of Energy and Materials Engineering, Taiyuan University of Science and Technology, Jincheng 048000, People’s Republic of China E-mail: xiaozhong_d@163.com Received 26 July 2023, revised 6 December 2023 Accepted for publication 21 December 2023 Published 29 December 2023 Abstract Ultrasonic testing is a widely used non-destructive testing technique for precision forgings. However, assessing defects in ultrasonic B-scan images can be prone to errors, misses, and inefficiencies due to human judgment. To address these challenges, we propose a method based on deep learning to automate the evaluation of such images. We started by creating a dataset comprising 8000 images, each measuring 224 × 224 pixels. These images were cropped from ultrasonic B-scan images of 7 specimens, each featuring different sizes and locations of holes and crack defects. We then used state-of-the-art deep learning models to benchmark the dataset and identified YOLOv5s as the best-performing baseline model for our study. To address the challenges of deploying deep learning models and the issue of small defects being easily confused with the background in ultrasonic B-scan images, we made lightweight improvements to the deep learning model. Additionally, we enhanced the quality of data labels through data cleaning. Our experiments show that our method achieved a precision of 97.8%, a recall of 98.1%, mAP@0.5 of 99.0%, and mAP@.5:.95 of 67.6%, with a frames per second (FPS) of 74.5. Furthermore, the number of model parameters was reduced by 43.2%, while maintaining high detection accuracy. Overall, our proposed method offers a significant improvement over the original model, making it a more reliable and efficient tool for automated defect assessment in ultrasonic B-scan images. Keywords: deep learning, ultrasonic testing, automated detection, lightweight improvement 1. Introduction The non-destructive testing of precision machined complex forgings is crucial, as they are irreplaceable core components in mechanical equipment. Ultrasonic testing is widely utilized in non-destructive testing of precision forgings due to its ease of use and ability to accurately locate defects [1]. While the acquisition of ultrasonic data is largely automated, the analysis ∗ Author to whom any correspondence should be addressed. of the acquired data is predominantly conducted manually by professionally trained experts. The quality of the analysis res- ults depends entirely on the knowledge and experience of the analysts, which can lead to issues such as missed detections, incorrect detections, and lengthy consumption times. As a res- ult, numerous researchers have made efforts to develop auto- mated methods for defect detection to streamline the analysis process. Figure 1 illustrates the basic scanning methods used in ultrasonic testing imaging, including A-scans, B-scans, and C-scans. In the past, most studies focused on using A-scan 1 © 2023 IOP Publishing Ltd Meas. Sci. Technol. 35 (2024) 035605 J Zhao et al Figure 1. Basic way of imaging with ultrasonic testing. data for automated analysis of ultrasonic data due to the poor imaging quality of B-scans images. For instance, in 2004, Bettayeb et al [2] proposed an automated ultrasonic NDE system that utilizes wavelet transform to suppress noise and enhance defect localization in ultrasonic signals, along with an artificial neural network classification algorithm, which achieved defect classification. In 2006, Matz et al [3] util- ized an ultrasonic signal filtering method with discrete wave- let transform and a pattern recognition method with support vector machines (SVMs) to classify A-scans signals into three categories: fault echoes, weld echoes, and back wall echoes. In 2007, Khelil et al [4] employed the principal component analysis method to optimize the wavelet parameters extrac- ted from ultrasonic echoes and establish a SVM algorithm to classify planar and volumetric defects. In 2011, Sambath et al [5] chose 12 coefficients from the wavelet representa- tion of the echo signal as features, such as mean, variance, energy, and amplitude, and inputted them to a neural net- work with two hidden layers for training to identify four types of defects with a 94% accuracy rate. In 2014, Chen et al [6] proposed a hierarchical multiclass SVM (LMSVM) with parameter optimization and feature selection using BA. The method was proven to be robust, accurate, and reliable for ultrasonic defect classification through experiments conducted on a welding defect dataset. In 2017, Cruz et al [7] used three different feature extraction methods, including the discrete Fourier transform, wavelet transform, and cosine transform, as well as two different artificial neural network architectures to automatically classify welding defects. They achieved effi- cient identification of defects using this approach. Meng et al [8] proposed a deep convolutional neural network (CNN) with a linear SVM top layer to classify cavity-layered ultrasonic signals in carbon fiber-reinforced polymer (CFRP) samples. In 2018, Munir et al [9] evaluated the performance of deep and shallow neural networks for automatic classification of weld defect ultrasonic signal data. They found that deep neural networks had better performance, achieving the highest accur- acy of 91.89% on a mixed-frequency dataset. In 2019, Munir et al [10] applied CNNs to noisy ultrasonic features to improve the classification performance and applicability of defects in welded parts. Their experimental results showed that CNNs can achieve fairly high defect classification accuracy even for noisy signals. Guo et al [11] combined principal component analysis (PCA) on adaptive enhancement (Adaboost), extreme gradient enhancement (XGBboost), and SVM—three machine learning models widely used in NDT—to compare their per- formance on 220 laser ultrasonic signal data collected from 22 samples with different subsurface defect sizes. PCA XGBoost achieved the highest recognition rate of 98.48%. While many researchers have achieved good results with automated analysis of A-scans, the evaluation datasets used have only contained a few hundred or a few thousand A-scans. Such datasets are hardly a complete representation of all pos- sible shape variations when compared to the millions of A- scans used in actual inspection tasks. Moreover, the lack of surrounding information in the A-scans makes it challenging to distinguish between noise and defect signals, which is not conducive to defect classification. Ultrasonic B-scan images provide valuable spatial inform- ation for automated analysis of ultrasonic testing, and recent advances in ultrasonic testing equipment have significantly improved their imaging quality. In 2019, Posilovic et al [12] 2 Meas. Sci. Technol. 35 (2024) 035605 J Zhao et al Figure 2. System framework. tested the performance of YOLO and SSD models in detecting defects from 98 real B-scan data by means of data augmenta- tion, and YOLO achieved better detection results with an aver- age accuracy (AP) of 89.7%. Virupakshappa et al [13] simu- lated a total of 400 ultrasonic B-scan images of various defect types and demonstrated that deep learning methods, such as CNN, can be used for defect identification in ultrasonic NDT with high classification accuracies. In 2021, Ye et al [14] cre- ated a new ‘USimgAIST’ dataset of over 7000 real ultrasonic testing images of 17 types of defects and benchmarked the dataset using state-of-the-art deep learning models. DenseNet achieved the best result with an f1 score of 95.33% in the work of Virkkunen et al [15], who used data augmentation to bring the deep learning network to human evaluation levels of per- formance in identifying cracks in pipe welds. Latete et al [16] used simulated and experimental data to train the Faster R- CNN, which allowed accurate identification, localization and sizing of flat bottom holes and side drilled holes in the speci- men. Medak et al [17] achieved 89.6% mAP for the detection of extreme aspect ratio defects commonly found in ultrasonic images by training the EfficientDet deep learning framework with adjusted hyperparameters. These studies demonstrate that deep learning has great potential for ultrasonic testing image recognition, and the com- bination of deep learning algorithms with ultrasonic testing B- scan images recognition has considerable significance in terms of improving detection efficiency and ensuring discriminatory accuracy. However, most of the research has been conducted based on simulation data, and there are no scholars who have systematically studied various aspects of data acquisition, data cleaning, deep learning model selection and model improve- ment in practical industrial applications. In this study, we have conducted comprehensive research on Dataset creation, benchmark selection for deep learn- ing models and model improvement methods. The overall framework of our approach is illustrated in figure 2. We have evaluated the performance of the latest deep learning models for automated defect evaluation of ultrasonic images in precision forgings using our self-built datasets. We have used YOLOv5s as the baseline and fine-tuned the automated ultrasonic image detection process through lightweight model improvements and data cleaning methods. Our study provides some insight into the deployment of deep learning models for automated defect assessment applications in ultrasonic images. 2. Construction of ultrasonic image dataset There is currently a lack of publicly available large-scale data- sets for ultrasonic testing due to the diversity and variability of defects and the difficulty in obtaining sample data. To address this gap and develop an automated evaluation algorithm for ultrasonic testing images, we collected data from 7 samples containing hole defects ranging from 0.5 mm to 1 mm in dia- meter and crack defects less than 1 mm in width. Data was acquired using an AOS phased-array ultrasonic real-time total focus imaging system, as shown in figure 3. The phased-array transducer utilized had a frequency of f = 5 MHz, 128 arrays, a width of e = 0.65 mm, a gap of g = 0.1 mm between arrays, a center distance of p = 0.75 mm, and a height of w = 10 mm. A total of 30 ultrasonic B-scan images were acquired using the Total Focus Imaging Algorithm (TFM), comprising 28 images with defects and 2 images without defects. To enhance the sample size for improved training of the deep learning model, the images were randomly cropped to generate 8000 B-scan images of 224 × 224 pixels. These images were ana- lyzed and labeled by multiple engineers to identify the types and locations of defects, as illustrated in table 1. The dataset was divided into a training set, validation set, and test set in a 3:1:1 ratio. 3 Meas. Sci. Technol. 35 (2024) 035605 J Zhao et al Figure 3. Phased array ultrasonic real-time total focus imaging system. Table 1. Dataset division. Number of hole defects Number of crack defects Number of defective images Number of defect-free images Total number of images Training set Validation set Test set Total 7543 2471 2465 12 470 1190 362 368 1920 4088 1379 1380 6847 692 231 230 1153 4780 1610 1610 8000 Figure 4 displays ultrasonic B-scan images of the defect- ive and healthy samples we acquired, which contain holes and cracks. While crack defects are clearly visible through inspection, some tiny hole defects are not eas- visual ily distinguishable from the background. This difficulty in detection can lead to missed or false detections during analysis. Due to the small size of hole defect targets and the pres- ence of some background noise in B-scan images, engineers may encounter problems during annotation, such as miss- ing defect annotations, mislabeled types, oversized bound- ing boxes, and bounding boxes with center points outside the image. To address these issues, we used the method shown in figure 5 to automatically retrieve data and obtained over 200 images with problematic annotations. We then re-annotated these images. The dataset after data cleaning was divided as shown in table 2. 3. Baseline testing trained all models on a machine equipped with a single NVIDIA GTX 1070 (8G) graphics card, using the default hyperparameter settings and the maximum batch size accept- able for that card. To evaluate the performance of each model, we used four statistics: TP (True Positive), FP (False Positive), TN (True Negative), and FN (False Negative). Based on these statist- ics, we introduced six evaluation metrics: Precision (P), Recall (R), mAP@0.5, mAP@.5:.95, number of model parameters (Params), and frames per second (FPS). Precision, which refers to the probability of detecting the correct target in all detected targets, can be calculated using equation (1). Recall, which refers to the probability of correct recognition among all positive samples, can be derived from equation (2) Precision = TP TP + FP Recall = TP TP + FN . (1) (2) In this study, we aimed to evaluate the performance of various deep learning models on our acquired ultrasonic defect dataset. Specifically, we compared YOLOv3 [18], YOLOv5, YOLOv7 [19], YOLOR [20], EffcientDet [21], and the two-stage detec- tion model Faster-RCNN [22], which are among the most commonly used models in the field of object detection. We A Precision-Recall curve can be plotted from an array con- taining Precision and Recall values, and the average precision AP is the area under the curve, calculated as follows: 1ˆ AP = p (r) dr. (3) 0 4 Meas. Sci. Technol. 35 (2024) 035605 J Zhao et al Figure 4. Ultrasonic B-scan images of (a) holes, (b) cracks, (c) no defects. mAP is the mean of all classes of AP: the YOLOv5s model was chosen as the baseline for further research. mAP = 1 n n∑ i =1 APi (4) 4. Methods where n represents the type of defect, mAP@0.5 represents the mAP value when IoU is set to 0.5, and mAP@.5:.95 represents the average mAP at different IoU thresholds (from 0.5 to 0.95, step size 0.05). Table 3 shows that among the deep learning models tested, the YOLOv5s model achieved the highest levels of preci- sion, mAP@.5:.95, and FPS. While the recall rate was only 3.0% lower than the best-performing YOLOR-P6 model. Compared to the best performing EfficientDet d0, the differ- ence in mAP@0.5 was only 0.7%. As the localization effect of defects and re-al-time detection are of more importance, Although the YOLOv5s model already exhibits good infer- ence speed and detection performance, this study still faces some challenges that need to be addressed. Firstly, during the detection of hole defects, the small size of the target leads to a low Precision and Recall of YOLOv5s, resulting in missed detections. Secondly, the large number of convolutional and deep neural network structures can lead to excessive model complexity, which is not suitable for deployment to mobile or embedded systems. To address these issues, this study optimized the model structure and used data cleaning methods to improve the 5 Meas. Sci. Technol. 35 (2024) 035605 J Zhao et al Figure 5. Data cleaning process. Table 2. Data cleaning results. Number of hole defects Number of crack defects Number of defective images Number of defect-free images Total number of images Training set Validation set Test set Total 7502 2481 2458 12 441 1190 362 368 1920 4082 1383 1371 6836 698 227 239 1164 4780 1610 1610 8000 Table 3. Benchmarking experiments. Model Faster-RCNN(resnet50) YOLOv3 spp YOLOv5s YOLOv7 YOLOR-P6 EffcientDet d0 Size 224 512 640 640 640 512 P 0.899 0.864 0.937 0.882 0.653 0.941 R 0.902 0.871 0.936 0.901 0.966 0.937 6 mAP@0.5 mAP@.5:.95 FPS Params 0.936 0.901 0.96 0.925 0.933 0.967 0.558 0.471 0.591 0.471 0.506 0.537 17 21 72 23 35 29 41.8M 62.5M 7.02M 36.9M 36.8M 3.9M Meas. Sci. Technol. 35 (2024) 035605 J Zhao et al Figure 6. Improved YOLOv5s model. Figure 7. CBS_CBAM module. efficiency of automated defect detection in ultrasonic images. The optimized YOLOv5s model structure is shown in figure 6. 4.1. YOLOv5s model improvement To improve detection efficiency and enhance the network’s focus on the target, we incorporated CBAM into the CBS module in layers 1, 3, and 5 of the backbone network. As shown in figure 7, the CBAM module [23] sequen- tially infers the attention graph through the channel atten- tion module and the spatial attention module. The chan- nel attention module leverages the information between fea- ture channels, while the spatial attention module leverages the information between feature spaces. The attention graph is then multiplied with the input feature graph for adapt- ive feature optimization, which effectively attends to small targets. To fulfill the requirements of industrial applications, we introduced the Ghost module [24] for a lightweight net- work design, by replacing the CBS module at layer 7 with GhostConv. In figure 8(a), GhostConv is performed in two steps: firstly, using normal convolution to obtain fewer fea- ture maps, then applying a second convolution on top of it to obtain more feature maps, and finally concatenating the different feature maps together to produce a new output. We replaced the C3 module in Backbone with C3Ghost, whose structure is shown in figure 8(c), mainly consisting of ordinary convolution and the GhostBottleneck as shown in figure 8(b). 7 Meas. Sci. Technol. 35 (2024) 035605 J Zhao et al Figure 8. (a) GhostConv, (b) GhostBottleneck and (c) C3Ghost modules. The GhostBottleneck module allows sufficient or redundant information to be provided in the feature layer, to always ensure the model’s understanding of the input data. To address the small and medium-sized target defects in this study, which are mainly concentrated in the shallow part of the neural network, we reduced the number of C3Ghost modules in layers 2, 4, and 6 of the backbone network from the ori- ginal [3, 6, 9] to [2, 4, 6]. This adjustment effectively improves the network’s detection capability for small and medium-sized targets. To reduce computational complexity and network structure while maintaining accuracy, the CBS module of the neck net- work was replaced with GhostConv, and the C3 module was replaced with C3Ghost, effectively compressing the network parameters of YOLOv5s. IOU = B1 ∩ B2 B1 ∪ B2 (6) where ρ2 (b, bgt) represents the Euclidean distance between the centroids of the prediction frame and the true frame. c repres- ents the diagonal distance of the smallest closed area that can contain both the prediction box and the true box, B1 for the true box and B2 for the prediction box. a = v 1 − IOU + v ( v = 4 π 2 arctan wgt hgt − arctan ) 2 w h LOSSCIOU = 1 − IOU + ρ2 (b, bgt) c2 + av (7) (8) (9) 4.2. Comparison of loss functions YOLOv5 defaults to using the CIOU loss function, and CIOU Loss takes into account the overlap area, centroid distance and aspect ratio of the bounding box regression though. As shown in equation (5): CIOU = IOU − ρ2 (b, bgt) c2 − av (5) where a is the weight function and v reflects the difference in aspect ratio rather than the true difference between the aspect ratio and its confidence level respectively, so this can some- times prevent the model from optimizing similarity effect- ively, for which we introduced the EIOU loss function and compared it with the CIOU loss function. EIOU loss was pro- posed by Zhang et al [25] in 2021, which minimizes the differ- ence between the width and height of the target frame and the 8 Meas. Sci. Technol. 35 (2024) 035605 J Zhao et al Anchor, producing faster convergence and better localization results. in Precision, a 0.8% increase in Recall, a 43.2% reduction in the amount of parameters, and an increase in FPS to 75.1. The comparison of solutions A, D, E, and G reveals that although optimizing the backbone network can reduce the model’s parameter count, it does not significantly improve the FPS. This is because the introduction of the CBAM attention mechanism in the improved backbone network increased the network layers, thus affecting the inference speed. Solution D involved lightweight design specifically for the backbone and neck networks, achieving a superior bal- ance between mAP and params. This approach resulted in the best detection performance. While ensuring detec- tion accuracy, it significantly reduced the model’s para- meter count, enhancing detection efficiency. It meets the requirements of accuracy and real-time performance in the lightweight industrial defect detection. Additionally, model is well-suited for practical deployment in production environments. In order to demonstrate the impact of data quality on the model’s detection accuracy, this study trained and tested the original YOLOv5s model and the improved model on the cleaned dataset. The results are presented in table 5, which clearly shows that data cleaning can effectively improve all aspects of model metrics. Precision improved by 4.5% and mAP@0.5 by 3.2%. Moreover, compared to the original YOLOv5s model, the improved model has only 0.1% lower mAP@.5:.95, but the number of model parameters is reduced by 43.2%, FPS is also slightly improved, and several other indicators have not changed significantly. Therefore, in prac- tical applications, it is more productive to identify the deep learning model first and then look for ways to improve the data. Figure 9 illustrates a comparison of the mAP between the improved algorithm and the original algorithm during the training phase. It can be observed that the convergence rate of the improved model is similar to that of the original model, indicating that the improvements made in this study do not affect the model’s convergence. Figure 10 illustrates the detection results of the YOLOv5s model before and after the improvement. The green circles represent the defects that were missed. Compared to the YOLOv5s model, the improved method in this study still has some cases of missing detection for small defects that are difficult to distinguish with the naked eye. However, it achieves a high accuracy rate of detection overall. This indic- ates that the lightweight model proposed in this study has excellent detection performance and can meet the accuracy and real-time performance requirements in industrial defect detection. The equation for EIOU loss is as follows: LOSSEIOU = LOSSIOU + LOSSdis + LOSSasp = 1 − IOU + ρ2 (b, bgt) c2 + ρ2 (w, wgt) C2 w + ρ2 (h, hgt) C2 h (10) where C2 rectangle of the predicted and real boxes. w and C2 h are the width and height of the smallest outer The EIOU equation consists of three parts. LOSSIOU is the loss of overlap between the predicted and true frames, LOSSdis is the loss of distance between the center of the predicted and true frames, which is the same as that of CIOU, and LOSSasp is the loss of width and height of the predicted and true frames. 5. Experimental results and discussion To validate the impact of the improvements described in this study on the detection performance of the model, an evaluation was carried out on the ultrasound B-scan dataset that we col- lected. We set up 7 solutions to analyze the different improve- ment components, each using the same training parameters. Table 4 shows the results of the evaluation. Solution A optimizes the backbone network by using the CBS_CBAM structure, adding a channel attention module and a spatial attention module to enhance the detection of small targets, replacing the Conv module at layer 7 with GhostConv, and replacing the C3 module with C3Ghost. The mAP@.5:.95 only reduced by 0.6%, while the amount of model paramet- ers was reduced by 24.8%, and the FPS increased to 73.6. Solution B improves the neck network by replacing the Conv module with GhostConv and the C3 module with C3Ghost, reducing the amount of model parameters by 18.5% and increasing the FPS considerably to 85.7. Solution C uses the EIOU loss function to replace the original CIOU loss to improve the localization accuracy of the model bounding box, with effective improvements in Precision, Recall, and FPS. Solutions D, E, and F were subjected to two-by-two cross- validation, and the comparison revealed that improving the backbone and neck networks could significantly reduce the number of model parameters. Improving the IOU loss could improve the Recall rate and reduce defect misses. Solution G has been improved for all three areas. Although mAP@0.5 is reduced by 0.2% and mAP@.5:.95 by 2.2% com- pared to the original YOLOv5s model, there is a 0.6% increase 9 Meas. Sci. Technol. 35 (2024) 035605 J Zhao et al Model Backbone Neck EIOU P R mAP@0.5 mAP@.5:.95 Params FPS Table 4. Ablation experiments. YOLOv5s Solution A Solution B Solution C Solution D Solution E Solution F Solution G 3 3 3 3 3 3 3 3 3 3 3 3 0.937 0.936 0.935 0.945 0.941 0.943 0.943 0.942 0.936 0.936 0.931 0.944 0.936 0.944 0.943 0.944 0.96 0.957 0.958 0.961 0.961 0.962 0.96 0.958 0.591 0.585 0.586 0.589 0.572 0.588 0.582 0.569 7.02M 5.28M 5.72M 7.02M 3.99M 5.28M 5.72M 3.99M Table 5. Data cleaning test results. Model YOLOv5s YOLOv5s + Data cleaning Solution D + Data cleaning P 0.937 0.982 0.978 R 0.936 0.981 0.981 mAP@0.5 mAP@.5:.95 0.96 0.992 0.990 0.591 0.677 0.676 Params 7.02M 7.02M 3.99M 72.6 73.6 85.7 86.1 74.5 74.4 85.5 75.1 FPS 72.6 72.6 74.5 Figure 9. Comparison of (a) mAP@0.5 and (b) mAP@.5:.95 for Solution D and YOLOv5s. Figure 10. Detection results of (a) YOLOv5s and (b) Solution D. 10 Meas. Sci. Technol. 35 (2024) 035605 J Zhao et al 6. Conclusion Consent for publication To achieve effective automation in the detection of ultrasound B-scan images, this study proposes an improved YOLOv5 model, making the following contributions: All authors have consented to have this work published and have approved of submission to the Measurement Science and Technology. (1) A baseline test was conducted on the constructed data- set, comparing the performance of YOLOv3, YOLOv5, YOLOv7, YOLOR, EfficientDet, and Faster-RCNN. The YOLOv5 model was identified as the most efficient method for analyzing ultrasound B-scan images currently available. (2) The YOLOv5 model was enhanced by incorporating the CBAM attention mechanism and GhostConv light- weight convolution, simplifying the model complexity and improving detection efficiency. (3) From a data nificantly accuracy. data-centric perspective, an cleaning method was employed enhance the algorithm’s automated to sig- detection The primary objective of this study is to validate the feasib- ility and effectiveness of the developed method. It is acknow- ledged that defects in real-world applications might be even more intricate. The breadth of collected ultrasound data and the extent of automated quantitative analysis will be explored in future work. Data availability statement The data cannot be made publicly available upon publication because they are owned by a third party and the terms of use prevent public distribution. The data that support the findings of this study are available upon reasonable request from the authors. Acknowledgments The author would like to thank the anonymous reviewers for constructive comments and suggestions which led to an improved presentation. Funding This work was supported by the General Project of the National Natural Science Foundation of China, Project No. in 51875501; Postgraduate Education Innovation Project Shanxi Province of China, Project No. 2022Y671, and the Natural Science Foundation of Shanxi Province, China, Project No. 202103021224273. Conflict of interest The authors declare that we have no competing interest. 11 Authors’ contributions Z J and D X wrote most of the manuscript text and gener- ated all the graphics. Z J and Z Y developed the algorithm and wrote the program. S X wrote the introduction and provided background information and references. ORCID iD Jianjun Zhao  https://orcid.org/0009-0003-2931-0379 References [1] Zhao J, Zhang Z, Zhang M and Du X 2022 Scanning path planning of ultrasonic testing robot based on deep image processing Russ. J. Nondestruct. Test. 58 167–75 [2] Bettayeb F, Rachedi T and Benbartaoui H 2004 An improved automated ultrasonic NDE system by wavelet and neuron networks Ultrasonics 42 853–8 [3] Matz V, Kreidl M and Smid R 2006 Classification of ultrasonic signals Int. J. Mater. Prod. Technol. 27 145–55 [4] Khelil M, Boudraa M, Kechida A and Drai R 2007 Classification of defects by the SVM method and the principal component analysis (PCA) Int. J. Electr. Comput. Eng. 1 1–6 [5] Sambath S, Nagaraj P and Selvakumar N 2011 Automatic defect classification in ultrasonic NDT using artificial intelligence J. Nondestruct. Evaluation 30 20–28 [6] Chen Y, Ma H and Zhang G 2014 A support vector machine approach for classification of welding defects from ultrasonic signals Case Stud. Nondestruct. Test. Eval. 29 243–54 [7] Cruz F C, Filho E F S, Albuquerque M C S, Silva I C, Farias C T T and Gouvea L L 2017 Efficient feature selection for neural network based detection of flaws in steel welded joints using ultrasound testing Ultrasonics 73 1–8 [8] Meng M, Chua Y J, Wouterson E and Ong C P K 2017 Ultrasonic signal classification and imaging system for composite materials via deep convolutional neural networks Neurocomputing 257 128–35 [9] Munir N, Kim H, Song S and Kang S 2018 Investigation of deep neural network with drop out for ultrasonic flaw classification in weldments J. Mech. Sci. Technol. 32 3073–80 [10] Munir N, Kim H, Park J, Song S and Kang S 2019 Convolutional neural network for ultrasonic weldment flaw classification in noisy conditions Ultrasonics 94 74–81 [11] Lv G, Guo S, Chen D, Feng H, Zhang K, Liu Y and Feng W 2023 Laser ultrasonics and machine learning for automatic defect detection in metallic components NDT & E Int. 133 102752 [12] Posilovic L, Medak D, Subasic M, Petkovic T, Budimir M and Loncaric S 2019 Flaw detection from ultrasonic images using YOLO and SSD Proc. 11th Int. Symp. Image Signal Process. Anal (ISPA) pp 163–8 Meas. Sci. Technol. 35 (2024) 035605 J Zhao et al [13] Virupakshappa K and Oruklu E 2019 Multi-class classification [20] Wang C, Yeh I and Liao H M 2021 You only learn one of defect types in ultrasonic NDT signals with convolutional neural networks Proc. IEEE Int. Ultr. Symp. (IUS) pp 1647–50 [14] Ye J and Toyama N 2021 Benchmarking deep learning models for automatic ultrasonic imaging inspection IEEE Access 9 36986–94 [15] Virkkunen I, Koskinen T, Jessen-Juhler O and Rinta-Aho J 2021 Augmented ultrasonic data for machine learning J. Nondestruct. Evaluation 40 1–11 [16] Latete T, Gauthier B and Belanger P 2021 Towards using representation: unified network for multiple tasks (arXiv: 2105.04206) [21] Tan M, Pang R and Le Q V 2020 EfficientDet: scalable and efficient object detection 2020 IEEE/CVF Conf. on Computer Vision and Pattern Recognition (CVPR) (Seattle, WA, USA) pp 10778–87 [22] Ren S, He K, Girshick R and Sun J 2017 Faster R-CNN: towards real-time object detection with region proposal networks IEEE Trans. Pattern Anal. Mach. Intell. 39 1137–49 convolutional neural network to locate, identify and size defects in phased array ultrasonic testing Ultrasonics 115 106436 [23] Woo S, Park J, Lee J and Kweon I S 2018 CBAM: convolutional block attention module Proc. European Conf. on Computer Vision pp 3–19 [17] Medak D, Posilovic L, Subasic M, Budimir M and Loncariic S [24] Han K, Wang Y, Tian Q, Guo J, Xu C and Xu C 2021 Automated defect detection from ultrasonic images using deep learning IEEE Trans. Ultrason. Ferroelectr. Freq. 68 3126–34 [18] Redmon J and Farhadi A 2018 YOLOv3: an incremental 2020 GhostNet: more features from cheap operations 2020 IEEE/CVF Conf. on Computer Vision and Pattern Recognition (CVPR) (Seattle, WA, USA) pp 1577–86 improvement (arXiv:1804.02767) [25] Zhang Y, Ren W, Zhang Z, Jia Z, Wang L and [19] Wang C, Bochkovskiy A and Liao H M 2022 YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors (arXiv: 2207.02696) Tan T 2022 Focal and efficient IOU loss for accurate bounding box regression Neurocomputing 506 146–57 12
null
10.1038_s41467-022-30406-4.pdf
Data availability The cryo-EM particle stacks, maps and models generated in this study have been deposited in EMPIAR image archive, EMDB database and the Protein Data Bank, respectively, under accession codes EMPIAR-10969, EMD-25757 and PDB-7T9G) for VcINDY-Na+ (300 mM) structure and under accession codes EMPIAR-10970, EMD- 25756 and PDB-7T9F) for VcINDY-Ch+ structure. Source Data for Fig. 4 are available with the paper.
Data availability The cryo-EM particle stacks, maps and models generated in this study have been deposited in EMPIAR image archive, EMDB database and the Protein Data Bank, respectively, under accession codes EMPIAR-10969, EMD-25757 and PDB-7T9G ) for VcINDY-Na + (300 mM) structure and under accession codes EMPIAR-10970, EMD- 25756 and PDB-7T9F ) for VcINDY-Ch + structure. Source Data for Fig. 4 are available with the paper.
ARTICLE https://doi.org/10.1038/s41467-022-30406-4 OPEN Structural basis of ion – substrate coupling in the Na+-dependent dicarboxylate transporter VcINDY David B. Sauer1,2,4, Jennifer J. Marden1,2, Joseph C. Sudar Da-Neng Wang 1,2✉ 2, Jinmei Song1,2, Christopher Mulligan 3✉ & ; , : ) ( 0 9 8 7 6 5 4 3 2 1 The Na+-dependent dicarboxylate transporter from Vibrio cholerae (VcINDY) is a prototype for the divalent anion sodium symporter (DASS) family. While the utilization of an electro- chemical Na+ gradient to power substrate transport is well established for VcINDY, the structural basis of this coupling between sodium and substrate binding is not currently understood. Here, using a combination of cryo-EM structure determination, succinate binding and site-directed cysteine alkylation assays, we demonstrate that the VcINDY protein couples sodium- and substrate-binding via a previously unseen cooperative mechanism by conformational selection. In the absence of sodium, substrate binding is abolished, with the succinate binding regions exhibiting increased flexibility, including HPinb, TM10b and the substrate clamshell motifs. Upon sodium binding, these regions become structurally ordered and create a proper binding site for the substrate. Taken together, these results provide strong evidence that VcINDY’s conformational selection mechanism is a result of the sodium-dependent formation of the substrate binding site. 1 Department of Cell Biology, New York University School of Medicine, New York, NY 10016, USA. 2 Skirball Institute of Biomolecular Medicine, New York University School of Medicine, New York, NY 10016, USA. 3 School of Biosciences, University of Kent, Canterbury, Kent, UK. 4Present address: Centre for Medicines Discovery, Nuffield Department of Medicine, University of Oxford, Oxford, UK. email: c.mulligan@kent.ac.uk; da-neng.wang@med.nyu.edu ✉ NATURE COMMUNICATIONS | (2022) 13:2644 | https://doi.org/10.1038/s41467-022-30406-4 | www.nature.com/naturecommunications 1 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-022-30406-4 VcINDY is a Na+-dependent dicarboxylate transporter that imports TCA cycle intermediates across the inner mem- brane of Vibrio cholerae1,2. The detailed structural and mechanistic understanding of VcINDY1–4 has made the protein a prototype of the divalent anion sodium symporter (DASS) family (Supplementary Fig. 1a, b)5. Within the human genome, the SLC13 genes encode for DASS members including the Na+-dependent, citrate transporter (NaCT) and dicarboxylate transporters 1 and 3 (NaDC1 and NaDC3)6. Besides functioning as TCA cycle intermediates, DASS-imported substrates are cen- tral to a number of cellular processes. In bacteria C4-carboxylates can serve as the sole carbon source for growth7, while imported citrate and tartrate are electron acceptors during fumarate respiration8. Citrate is also a precursor for both fatty acid bio- synthesis and histone acetylation in mammals9,10. Dicarboxylates such as succinate and α-ketoglutarate act as signaling molecules that regulate the fate of naive embryonic stem cells and certain types of cancer cells11,12. As a result of these roles in regulating cellular di- and tricarboxylate levels, mutations in DASS trans- porters have dramatic physiological consequences. Deletion of bacterial DASS transporters can abolish growth on particular dicarboxylates7,8. Mutations in the human NaCT transporter cause SLC13A5-Epilepsy in newborns13, whereas variants in the dicarboxylate transporter NaDC3 lead to acute reversible leukoencephalopathy14. In mice, knocking out NaCT results in protection from obesity and insulin resistance15. Such roles of SLC13 proteins in cell metabolism have made them attractive targets for the treatment against obesity, diabetes, cancer and epilepsy16–18. Therefore, mechanistic characterization of the prototype transporter VcINDY will help us to better understand the transport mechanism of the entire DASS family, including the human di- and tricarboxylate transporters. The VcINDY protein is a homodimer consisting of a scaffold domain and a transport domain (Supplementary Fig. 1b–f)1. The conservation of this architecture throughout the DASS/SLC13 family has been confirmed by X-ray crystallography and cryo- electron microscopy (cryo-EM) structures of VcINDY, LaINDY, a dicarboxylate exchanger from Lactobacillus acidophilus, and the human citrate transporter NaCT1,4,19,20. Comparison of VcINDY in its inward-facing (Ci) conformation with the outward-facing (Co) structure of LaINDY, along with MD simulations, reveals that an elevator-type movement of the transport domain, through an ~12 Å translation along with an ~35° rotation, facilitates translocation of the substrate from one side of the membrane to the other19. In fact, the structural and mechanistic conserva- tion may extend beyond DASS to the broader Ion Transport Superfamily (ITS)5,21. Substrate transport of VcINDY is driven by the inwardly- directed Na+ gradient, with dicarboxylate import coupled to the co- transport of three sodium ions (Supplementary Fig. 1a, b)1,2,22. The binding sites for the substrate and two central Na+s have been identified in the structures of VcINDY in its Na+- and substrate- bound inward-facing (Ci-Na+-S) state (Supplementary Fig. 1e, f)1,4. The Na1 site on the N-terminal half of the transport domain is defined by a clamshell formed by loop L5ab and the tip of hairpin HPin. A second clamshell encloses Na2, related to Na1 by inverted- repeat pseudo-symmetry in the sequence and structure, and formed by L10ab and the tip of hairpin HPout (Supplementary Fig. 1c). In addition to binding the Na+s, both hairpin tips also form parts of located between the Na+ sites. Each the substrate-binding site, hairpin tip consists of a conserved Ser-Asn-Thr (SNT) motif, and the two SNT motifs form part of the substrate-binding site, making direct contact with carboxylate groups of the substrate. Whereas these two SNT signature motifs are responsible for recognizing carboxylate, additional residues in neighboring loops have been proposed to distinguish between different kinds of substrates4. Furthermore, VcINDY’s structure with sodium in the absence of a substrate (the Ci-Na+ state), determined in 100 mM Na+, is very similar to that of the Na+- and substrate-bound state Ci-Na+-S19. While the Na+- and substrate-binding sites in VcINDY have been well-characterized1,4,23, the coupling mechanism between the electrochemical gradient and substrate transport24 is less well understood. There is strong evidence that charge compensation by sodium ions is essential in lowering the energy barrier for transporting the di- and trivalent anionic substrates across the membrane19. However, such charge compensation alone does not necessarily result in substrate binding as Li+ is able to bind to VcINDY similarly to Na+, but results in a lower affinity substrate binding site and considerably reduced transport rates2,23. More importantly, charge neutralization cannot explain the sequential binding observed for VcINDY. As is the case for other DASS proteins25–28, all available experimental evidence from both whole cells and reconstituted systems supports the notion that in VcINDY sodium ions and substrate bind in a sequential manner, with Na+s binding first, followed by dicarboxylate2,3,23,29. As a secondary-active transporter can transport substrate in either direction, it follows that the release of the substrate and Na+s is also ordered, with the substrates being released first. Structures of VcINDY in the Na+- and substrate-bound state Ci-Na+-S, in which the Na+ sites share residues with the sub- strate site in their center, allowed us to propose that substrate binding in VcINDY follows a cooperative binding mechanism via conformational selection1,30. In this mechanism, the binding of sodium ions helps to induce a proper binding site for the substrate (Supplementary Fig. 1a, b). Conversely, in the absence of bound sodium ions the substrate-binding site will change significantly, such that the substrate cannot bind. Not only can such a mechanism be part of Na+—substrate coupling, it may also explain the sequential binding observed for VcINDY. This conformational selection mechanism of substrate binding enables us to make two explicit, experimentally testable predic- tions. First, the affinity of the transporter to a substrate must be much higher in the presence of Na+ than in its absence. Second, substantial structural changes will occur at the Na+ sites in the absence of sodium, affecting substrate binding. In this work, we aim to test these two predictions using a combination of structure determination by single-particle cryo-EM, substrate-binding affinity measurements by intrinsic tryptophan fluorescence quenching, and position accessibility quantification via a newly-developed site-directed cysteine alkylation assay29. In particular, we characterize VcINDY in sodium-saturating and sodium-free conditions, including structures in Ci-Na+ and Ci-apo states. These experimental results allow us to directly test the conformational selection binding model of VcINDY. Results Succinate binding depends on the presence of Na+. To test the first of the predictions generated from our conformational selection hypothesis, we measured VcINDY’s binding affinity for the model substrate, succinate, in both the presence and absence of Na+ (Supplementary Fig. 2). We reasoned that VcINDY’s tryptophans, particularly Trp148 located at the tip of HPin of the Na1 site, may change its position or environment upon Na+-/substrate-binding. Thus, we used intrinsic tryptophan fluorescence quenching, a technique that has been successfully applied to measure substrate binding for various membrane transporters20,31–36. In the presence of 100 mM Na+, detergent- purified VcINDY was found to bind succinate with an apparent Kd of 92.2 ± 47.4 μM (Fig. 1a, Supplementary Fig. 2b). For comparison, the human NaCT in the same protein family binds its substrate citrate at an apparent Kd of 148 ± 28 μM20. 2 NATURE COMMUNICATIONS | (2022) 13:2644 | https://doi.org/10.1038/s41467-022-30406-4 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-022-30406-4 ARTICLE Fig. 1 Cryo-EM structure of VcINDY in the Ci-Na+ state determined in 300 mM Na+. a Measurements of succinate binding to detergent-purified VcINDY in the presence of 100 mM NaCl, using intrinsic tryptophan fluorescence quenching (N = 4). Data are presented as mean values ± SEM. The apparent Kd was determined to be 92.2 ± 47.4 mM. When NaCl was replaced with Choline chloride, no binding of succinate to VcINDY could be measured (N = 4). b Cryo-EM map of VcINDY determined in the presence of 300 mM NaCl. The map is colored by local resolution (Å) and contoured at 5.1 σ. The overall map resolution is 2.83 Å. c Structure of VcINDY in the Ci-Na+ state. The structure is colored by the B-factor. d Na1 site structure and Coulomb map. e Na2 site structure and Coulomb map. f Overlay of VcINDY structures around the substrate and sodium binding sites in the Ci-Na+ state (green) and Ci- Na+-S state (PDB ID: 5UL7, blue). There is very little structural change observed between the two states. To measure the binding affinity of succinate to VcINDY with empty Na+ binding sites, we searched for a cation to replace Na+ in the purification buffer. This ion should not occupy the Na1 or Na2 sites while still allowing the transporter protein to remain stable in the solution. K+ is unable to power substrate transport in VcINDY, but was found to be unsuitable as the protein precipitated when purified in the presence of 100 mM KCl. We next tested the organic cation choline (C5H14NO+, Ch+ in abbreviation). We reasoned that Ch+ would be more stabilizing than K+ based on its position in the Hofmeister series37, and that its size would preclude it from occupying Na+ binding sites38. Indeed, VcINDY purified in 100 mM NaCl remained soluble at 0.5–1.0 mg/mL after diluting the sample 11,000-fold in 100 mM ChCl. VcINDY was therefore purified in the presence of 100 mM Ch+ as the only monovalent cation. The protein eluted as a sharp, symmetrical peak on a size- exclusion chromatography column (Supplementary Fig. 2a), con- firming its stability and structural homogeneity. intrinsic tryptophan fluorescence quenching with VcINDY purified and assayed in the presence of 100 mM Ch+ revealed no succinate binding (Fig. 1a, Supplementary Fig. 2c). Thus, the binding measurements in the presence and absence of Na+ are consistent with a conformational selection model where bound sodium ions are necessary to VcINDY forming a proper binding site for succinate. Encouraged by these findings and our ability to produce stable, structurally homogeneous and Na+-free VcINDY, we next sought to uncover the structural basis of this Na+—substrate transporter’s structures using cryo-EM in different states. coupling by determining the Notably, Structure of VcINDY in 300 mM Na+. Generally speaking, the transport mechanism of a secondary-active transporter is rever- sible, in which the direction of substrate translocation depends on the direction and magnitude of the driving force (Supplementary Fig. 1a, b). Consequently, substrate binding is structurally equivalent to substrate release. Therefore, to provide structural insights into VcINDY’s binding process, we aimed to characterize the substrate release process in the inward-facing (Ci) con- formations by capturing the structures of VcINDY in the fol- lowing states: its Na+-and substrate-bound state (Ci-Na+-S), its Na+-bound state (Ci-Na+) and its Na+- and substrate-free state (Ci-apo). The Ci-Na+-S structure of VcINDY has previously been solved using X-ray crystallography1,4. Additionally, we had characterized the Ci-Na+ state using a cryo-EM structure of VcINDY purified in 100 mM Na+ without substrate19. However, as the apparent K0.5 for Na+ for VcINDY was measured to be 41.7 mM2, our earlier VcINDY sample in 100 mM Na+ likely represents a mixture of the Ci-Na+ and Ci-apo states. It is unclear whether the subsequent cryo-EM image processing of the particles was able to exclude all particles of the Na+-free Ci-apo state. To more clearly and definitively resolve the Ci-Na+ state structure, in the current work we purified and determined a structure of VcINDY in 300 mM Na+. This ion concentration was optimized to increase the Na+ occupancy, while, at the same time, ensuring a low enough noise level in the cryo-EM images to determine a Ci-Na+ state structure of this small membrane protein (total dimer mass: 126 kDa) at 2.83 Å resolution (Fig. 1b, c, Supplementary Figs. 3 and 4a–c, Table 1). Compared with the two previously determined cryo-EM structures of VcINDY in the presence of 100 mM NaCl19, the herein reported structure in 300 mM Na+ (Fig. 1c, Supplementary Fig. 4b) is identical to the one bound to a Fab and embedded in lipid nanodisc (PDB ID: 6WW5)19 (r.m.s.d. of 0.460 Å for all the non-hydrogen atoms), except for the position of the last three residues at the C-terminus, which interact with the Fab molecule used for structure determination (Supplementary Fig. 4d). Further- though the map obtained in 300 mM Na+ conditions more, clarified the loop connecting HPoutb and TM10b, the model in 300 mM NaCl is effectively identical to the other previous Ci-Na+ structure in 100 mM NaCl, determined in amphipol and without Fab (PDB ID: 6WU3)19, with an r.m.s.d of 0.358 Å after excluding Val392 – Pro400 (Supplementary Fig. 4d). NATURE COMMUNICATIONS | (2022) 13:2644 | https://doi.org/10.1038/s41467-022-30406-4 | www.nature.com/naturecommunications 3 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-022-30406-4 As expected from the higher Na+ occupancy in the 300 mM sample, better-defined densities appeared within both the Na1 and Na2 clamshells (Fig. 1d, e), which were absent in the previous 100 mM Na+ maps19. In addition to coordination by side chains and backbone carbonyl oxygens, the sodium ion at the Na1 site is stabilized by the helix dipole moments from HPinb and TM5b (Fig. 1f; Supplementary Fig. 1e, f), as previously observed in other membrane proteins39,40. Similarly, the Na+ ion in the Na2 site is stabilized by HPoutb and TM10b. Finally, this higher resolution map confirmed our earlier observations that succinate release caused only limited changes at the substrate-binding site without relaxing the two Na+ clamshells19. Both the overall structure and the sodium- and substrate-binding sites in the Ci-Na+ state are similar to those in the sodium- and substrate-bound Ci-Na+-S state (Fig. 1e, f, Supplementary Fig. 4e). The similarity of these structures agrees with our conformational selection model of Na+ – substrate coupling, which requires sodium-binding induce a Ci-Na+ state structure that can bind substrate directly as in the Ci-Na+-S state (Supplementary Fig. 1a, b). Apo structure of VcINDY in Choline+. With the structures of sodium- and succinate-bound1,4 and Na+-only bound (Fig. 1) states in hand, the missing piece of the puzzle to validate the Na+ con- formational selection mechanism was the Ci-apo state structure of the transporter protein. As a Ch+ ion is too large to fit into a Na+ binding site36,38, and VcINDY was stable and monodisperse in the presence of 100 mM ChCl (Supplementary Fig. 2a), such a pre- paration allowed us to obtain cryo-EM maps of the Ci-apo state (Fig. 2, Supplementary Figs. 5 and 6, Table 1). Unlike the VcINDY map in 300 mM Na+ for which 3D classification converged to a single map (Supplementary Fig. 3), the VcINDY-choline dataset yielded four distinct classes at a resolution range of 3.6—4.4 Å resolution (Supplementary Figs. 5 and 7). The 3D class with the highest resolution was further refined to 3.23 Å resolution (Fig. 2a, b, Supplementary Fig. 6c). The least well-resolved regions of the map, and highest B-factors of the model, are found in L4-HPin and L9- HPout, two previously-identified hinge regions that facilitate move- ment of the transport domain19. Whereas the overall fold of the protein in Ch+ remains the same (Supplementary Fig. 6d, f), Na+ densities within the Na1 and Na2 clamshells are totally absent. Additional local changes are observed for the protein parts near the Na1 and Na2 sites (Fig. 2c), with a loss of density in each Ci-apo map at the HPinb and TM10b helices (Fig. 3b), indicating increased local structural flexibility. Flexibility of Apo VcINDY near the Na1 and Na2 sites. While the VcINDY Ci-apo state overall structure is similar to those in the 300 mM Na+ (r.m.s.d of 0.672 Å) (Supplementary Fig. 6f), the model exhibited significant changes near the Na1 and N2 sites (Figs. 2c and 3a, b). The tip of HPin and the L10a-b loop have moved away from the Na1 and Na2 sites, respectively, with the carbonyls of Ala376 and Ala420, and the side chain of Asn378 also rotated away from the Na2 site. Notably, HPinb near the Na1 site and TM10b near the Na2 site and their connecting loops showed marked decreased density in the cryo-EM map, corre- sponding to the increased flexibility of these regions (Fig. 3a, b). Correspondingly, the model exhibited significantly higher relative B-factors in the same regions compared to the rest of the model (Fig. 3c, d). However, we recognized that such a single, averaged model might not fully describe the true structural ensemble, and sought a method to describe the Ci-apo state’s mobility. To further analyze such local flexibility, we used simulated annealing41,42 in a model refinement protocol analogous to protein structure determination by NMR spectroscopy43. We reasoned that in multiple, separate refinements with simulated annealing the rigid parts of the VcINDY would converge to the same coordinates, while mobile portions of the protein would arrive at distinct atomic positions in each run. We term this as NMR-style analysis in recognition of NMR’s power to characterize protein dynamics, though in cryo-EM the constraints are Coulomb potential maps rather than distances. Most parts of the VcINDY structure exhibit no variation in the Ci-Na+ state, including HPinb and TM10b in the 300 mM Na+ condition (Fig. 3e, Supplementary Fig. 8a). In contrast, in the Ci- apo state the NMR-style analysis clearly illustrated the structural heterogeneity near the Na1 and Na2 sites (Fig. 3f, Supplementary Fig. 8b). Instead of converging to one structure, the simulated annealing resulted in an ensemble of structures, with the greatest variations occurring in the HPinb and TM10b regions. The mean r.m.s.d. of the transport domain’s backbone atoms for the Ci-apo protomers is 0.589 Å, as opposed to 0.099 Å among Ci-Na+ protomers refined using the same protocol to the same resolution. As the 3.23 Å apo map imposes C2 symmetry on one of four classes of particles in ChCl (Supplementary Fig. 5), and all four classes are different the degree of flexibility of these helices in the Ci-apo state is likely to be even greater. Such helix flexibility results from the absence of Na+ interactions with residues in the clamshells and with the dipoles of HPinb and TM10b44,45. (Supplementary Fig. 7), Site-directed alkylation supports structural changes to Na1 and Na2 sites. To confirm the local conformational changes and helix flexibility observed in our VcINDY structures, we implemented a site-directed cysteine alkylation strategy that can directly assess the solvent accessibility of specific positions in a protein. In this a L4-HPout b c 4.0 3.5 3.0 2.5 L9-HPin N378 A376 Na1 I149 N151 Na2 A420 P422 HPinb TM10b Fig. 2 Cryo-EM structure of VcINDY in the Ci-apo state determined in Choline+. a Cryo-EM map of VcINDY preserved in amphipol determined in the presence of 100 mM Choline Chloride. The map is colored by local resolution (Å) on the same scale as Fig. 1b and contoured at 4.8 σ. The overall map resolution is 3.23 Å. The two previously-identified hinge regions which facilitate movement of the transport domain19, L4-HPin and L9-HPout, are found to be most flexible. b Structure of VcINDY in the Ci-apo state. The structure is colored by the B-factor on the same scale as Fig. 1b. c Overlay of VcINDY structures around substrate and sodium binding sites in the sodium-bound Ci-Na+ state (green) and the Ci-apo state (pink). The structures of the two Na1 and Na2 clamshells have changed in the absence of sodium ions, particularly around residues Ile149, Asn151, Ala376, Asn378, Ala420 and Pro422. 4 NATURE COMMUNICATIONS | (2022) 13:2644 | https://doi.org/10.1038/s41467-022-30406-4 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-022-30406-4 ARTICLE a c e Na1 Na2 TM10b HPinb b Na1 Na2 TM10b HPinb d Na1 Na2 Na1 Na2 TM10b HPinb TM10b HPinb f Na1 Na2 Na1 Na2 TM10b HPinb TM10b HPinb Fig. 3 VcINDY flexibility changes near the Na1 and Na2 sites between the Ci-Na+ state and Ci-apo states. a Cryo-EM density map in 300 mM NaCl. b. Cryo-EM density map in 100 mM Choline Chloride. In a and b, the respective protein models’ backbones are fitted into the densities. Maps are contoured such that the scaffold domains have equal volume. c. Structure of VcINDY in its Ci-Na+ state. d. Structure of VcINDY in its Ci-apo state. In c and d, the structures are colored by normalized B-factors. e NMR-style analysis of the VcINDY structure in Na+. f NMR-style analysis of the VcINDY structure in Choline+. The resolution for refinement of both structures in e and f was truncated to 3.23 Å. In the absence of sodium, the helices on the cytosolic side of Na1 and Na2, particularly HPinb and TM10b and their connecting loops, show markedly increase flexibility. Instead of a single structure, the Ci-apo model consists of an ensemble of structures. approach, single cysteines are introduced into a Cys-less version of VcINDY, which is capable of robust Na+-driven transport2,3. Following purification, the cysteine mutants of VcINDY are incubated with the thiol-reactive methoxypolyethylene glycol maleimide 5 K (mPEG5K). This tag reacts with solvent-accessible cysteines and increases the protein mass by ~5 kDa, which is separable from unmodified protein on an SDS-PAGE gel. As mPEG5K will react faster with cysteines that are more accessible, monitoring PEGylation over time provides us with the ability to follow changes in the accessibility of particular parts of the pro- tein under different conditions29. To test our conformational selection model using biochemical approaches, we designed a panel of single-cysteine mutants of VcINDY that would report on the Na+-dependent accessibility changes at the Na1 and Na2 sites predicted from structures (Fig. 4a, Supplementary Fig. 8d). We selected residues that, if our cooperative binding model is accurate, will exhibit a higher rate of PEGylation in the absence of Na+ compared to its presence due to the increased mobility of HPin and TM10b. To create our panel proximal to the Na1 site, we purified four cysteine mutants whose reactive thiol groups are buried in the Ci-Na+ state behind HPin (L138C on HPina, A155C and V162C on HPinb and A189C on TM5a). However, similar cysteine substitutions near Na2 (Val427, Ile433, Gly442 and Met438) resulted in diminished expression levels, likely indicating the importance of these residues to the stability of the protein. Fortunately, cysteine mutation of Val441 to cysteine, a residue located on TM11 and behind TM10b (Fig. 4a, Supplementary Fig. 8d), expressed well and allowed for purifica- tion. Typically, well-expressing single cysteine VcINDY mutants that can be purified are capable of Na+-driven succinate transport29. We monitored the PEGylation of each mutant in the presence and absence of Na+. Under these reaction conditions there is no PEGylation of the Cys-less variant, demonstrating no background labelling that could hinder analysis (Fig. 4b, top row). In the presence of 300 mM Na+ we observed complete inhibition of PEGylation at every position (Leu138, Ala155, Val162, Ala189 and Val411) over the time course of 60 min (Fig. 4b, left panels), in agreement with our model that these residues are buried in the Na+-bound state. However, in the absence of Na+ (but with 300 mM Ch+), every mutant showed escalated levels of PEGyla- tion over time (Fig. 4b, right panels), indicating the increased flexibility of HPinb and TM10b. To ensure that the change in PEGylation rate that we observed was due to changes in residue accessibility and not caused by an unforeseen effect the cations may have on the PEGylation reaction, we monitored the reaction rate of a position for which we observed no accessibility change in the structural analysis. A cysteine mutant at Ser436, positioned at the periphery of the transporter protein (Fig. 4a), exhibited minimal Na+-dependent accessibility changes (Fig. 4b, bottom row). These accessibility measurements, along with our previous PEGylation results on three other VcINDY residues near the Na1 site (T154C, M157C and T177C, Supplementary Fig. 8e)29, fully support the changes in protein dynamics predicted upon occupation of the Na1 and Na2 sites, and are consistent with a conformational selection coupling model. Structural comparison of Ci-Na+-S, Ci-Na+ and Ci-apo states. The VcINDY structures determined in 300 mM Na+ and apo as reported here, together with previously-determined X-ray struc- ture of the protein with both sodium and substrate bound1,4, allowed us to examine the structural changes of the transporter between the Ci-Na+-S, Ci-Na+ and Ci-apo states. In addition to the flexibility observed in HPinb and TM10b, we observed amino acid sidechain movements both at the interface between the scaffold domain and the transport domain, as well as on the periplasmic surface of the protein. At the scaffold–transport domain interface, side chains of several bulky amino acids rotated or shifted between the three states, including Phe100, His111 and Phe326 (Fig. 5a). On the periplasmic surface, Trp461 at the C-terminus is buried in the apo and Na+-bound structures (Fig. 5b). However, in the Ci-Na+-S structure, the ring of the nearby Phe220 was rotated by ~90°, pushing out the side chain of Trp461, leaving the C-terminus pointing to the periplasmic space. Accordingly, the loop between HPoutb and TM10a moved into the periplasmic space of the apo VcINDY structure, displacing Glu394 and breaking its salt bridge with Lys337. Whereas no single switch was identified that can trigger conformational exchanges between the inward- and outward-facing states, local structural changes observed here suggest that small changes at multiple locations are required for inter-conformation transitions in VcINDY. In comparing maps of the three states, we noted the VcINDY Ci-Na+ map reported herein was sufficiently detailed to identify NATURE COMMUNICATIONS | (2022) 13:2644 | https://doi.org/10.1038/s41467-022-30406-4 | www.nature.com/naturecommunications 5 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-022-30406-4 a Na1 HPina 138 HPinb Na2 441 TM10b TM5a 189 155 162 90 HPina 138 162 Na1 HPinb 441 TM5a 189 155 TM10b 436 Na2 b kDa 55 + Na+ Na+ Cys-less L138C 436 A155C 35 25 55 35 25 55 35 25 55 35 25 55 35 25 55 35 25 55 35 25 P U P U P U P U P U P U P U 0 5 10 30 60 0 5 10 30 60 min V162C A189C V441C S436C Fig. 4 Cysteine alkylation with mPEG5K of VcINDY near the Na1 and Na2 sites in the presence and absence of Na+. a Location of cysteine mutations. Our structures suggested that HPinb and TM10b become flexible in the absence of sodium, increasing the solvent accessibility of Leu138, Ala155, Val162 and Ala189 near the Na1 site, and Val441 near the Na2 site. Position Ser436, for which no accessibility change was observed between our structures, is used as a control. On a Cys-less background, residues at these positions were individually mutated to a cysteine for mPEG5K labeling. For clarity, only amino acid numbers are labeled and the types are omitted. b Coomassie Brilliant Blue-stained non-reducing polyacrylamide gels showing the site-directed PEGylation of each cysteine mutant over time in the presence and absence of Na+. P: PEGylated protein; U: Un-PEGylated protein. Each reaction was performed on two separate occasions with the same result. Source data is provided as Source Data file. a b F326 F100 H111 five ordered water molecules buried at the dimer interface (Supplementary Fig. 8c). The water molecules are not visible in previous maps, or the VcINDY-apo map, indicating the high- the VcINDY Ci-Na+ map reported here was resolution of necessary for their identification. These waters are arranged in a square pyramidal configuration in the largely hydrophobic pocket, coordinated by only the symmetry-related carbonyls of Phe92 and inter-water hydrogen bonds. The role of these waters in VcINDY folding or transport are unclear, though protein folding defects underlie several pathogenic mutations on the equivalent dimerization interface of NaCT5. F220 W461 TM10a HPoutb E394 K337 Fig. 5 Movement of VcINDY’s amino acid side chains between its Ci-Na+-S, Ci-Na+ and Ci-apo states. VcINDY structures in three states are overlaid: Ci-Na+-S (blue), Ci-Na+ (green) and Ci-apo (pink) states. a At the scaffold-transport domain interface, the side chains of Phe100, His111 and Phe326 rotate between states. b On the periplasmic surface, some loops and side chains move between the states, including Phe220, Lys337, Glu394 and Trp462. Discussion Despite great advances in structural and mechanistic studies on the membrane transporters over the past ion–substrate coupling mechanism is well characterized for only very few co-transporters, this fundamental aspect of the secondary-active transport mechanism. Here, we have described the structural basis of ion–substrate coupling for VcINDY, which reveals a distinct conformational selection mechanism that ensures obligatory coupling. limiting our understanding of twenty years46–50, While Na+ sites in some other Na+-dependent transporters are buried in the middle of the protein38,48,49,51, the sodium sites in VcINDY are directly accessible from the extramembrane space. Previous experimental data support that Na+-driven DASS co- transporters operate via an ordered binding and release2,25–29. Specifically, Na+ binding occurs before substrate binding, while substrate release precedes Na+ release. For VcINDY, we have now observed that sodium release from the Na1 and Na2 sites in the cytoplasm allows increased conformational diversity going from the Ci-Na+ to the Ci-apo states, whereas the Ci-Na+ and Ci- 6 NATURE COMMUNICATIONS | (2022) 13:2644 | https://doi.org/10.1038/s41467-022-30406-4 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-022-30406-4 ARTICLE Co Co-Na+ Co-Na+-S TM10b HPin Ci TM10b HPin TM10b HPin Ci-Na+ Ci-Na+-S Fig. 6 Schematic model of conformational selection mechanism for sodium—substrate coupling in VcINDY. In the absence of sodium ions, HPinb and TM10b, along with their connecting loops responsible for sodium and substrate binding, are flexible. From the ensemble of flexible structures, the binding of sodium ions (blue circles) selects a conformation with a proper binding site for the substrate, allowing its binding (red oval). The scaffold and transport domains in each protomer are colored as green and pink, respectively. Only the Na1 and Na2 sites are illustrated. Transport domain movements in the two protomers are shown as symmetric for simplicity but are functionally independent. Na+-S states are structurally similar (Fig. 6). Specifically, the movement of helices HPinb and TM10b is tightly coupled to Na+ binding. At the Na1 and Na2 sites, the sodium ions are stabilized via direct and ion—dipole interaction with the two helices. Therefore, upon Na+-release, the elimination of these interac- tions caused the relaxation of the HPinb and TM10b helices44,45, leading to increased mobility in the connected loops responsible for substrate binding. In the reverse reaction, ions binding at the Na1 and Na2 sites, concurrent with helix re-ordering, select from the ensemble a structure with the proper binding site for the substrate. While the effects of VcINDY’s cryptic third Na+ are still to be determined, we now have established a structural understanding of the Na+—substrate coupling mechanism for this co-transporter. By extension, other DASS transporters may utilise a similar structural mechanism for Na+—substrate cou- pling (Fig. 6). The structural basis of Na+-substrate coupling for VcINDY is distinct from that of GltPh/GltTk from the dicarboxylate/ amino acid:cation symporter family which otherwise share several commonalities with VcINDY including the presence of re-entrant hairpin elevator-like mechanism1,3,4,48,52–56. In addition, as we have shown here for VcINDY, a cooperative binding mechanism has been suggested for both GltPh and GltTk, which requires the initial binding of Na+ in order to prime the binding site for the substrate, aspartate38,57. However, the structural basis of Na+-substrate coupling in GltPh/ GltTk differs substantially from the coupling mechanism we observe for VcINDY. Rather than the general relaxation of a helix governing substrate-binding site formation, the binding of Na+ to GltPh/GltTk induces discrete conformational changes of a small number of amino acid residues centered on the highly conserved NMDGT motif53,57,58. As is the case here for VcINDY (Fig. 6), the fully loaded and Na+-only bound structures of GltPh/GltTk are largely identical52,53,57,58, demonstrating that Na+ binding drives the for- mation of the substrate-binding site, and not the substrate itself. utilization loops and the an of In addition to conformational selection, another mechanism for ion–substrate coupling of co-transporters has been proposed to be charge compensation5,19. Such a mechanism can greatly minimize the energy penalty for translocating charged substrates across the hydrophobic lipid bilayer59,60. Unlike for DASS exchangers19, where charge compensation is the major force for overcoming the ←→ Ci transition, both local structural energy barrier in the Co ordering and charge balance are needed for Na+-coupled co- transporters within the DASS family. Comparison of the VcINDY structures reported here with those determined earlier1,4,19, of three states in total, also sheds new light on the mechanism of the transporter’s conformational switching between the two sides of the membrane. As the Ci-apo structure is significantly different from that of the Ci-Na+-S state, their corresponding transitions to the outward-facing state: Ci-apo to Co-apo and Ci-Na+-S to Co-Na+-S, are different at the transport-scaffold domain interface (Fig. 6). Whereas the transi- tion between Ci-Na+-S and Co-Na+-S state can be described as rigid-body movement, as was seen in the DASS exchangers5, the co-transporters’ Co-apo ←→ Ci-apo state transition likely involves large structural rearrangements of the transport domain. Considering the pseudo-symmetry of the DASS fold, the Ci-apo → Co-apo movement would require refolding of TM10b to pack against the scaffold domain, and possibly the concurrent unfolding of TM5b. This potential asymmetry between the apo- state transition (Co-apo ←→ Ci-apo) and transition of the fully- loaded transporter (Ci-Na+-S ←→ Co-Na+-S) needs further investigation. Finally, the pseudo-symmetry within the DASS fold and sequence1,3,19 and Na+ dependent solvent accessibility of the S381C mutant on HPoutb of VcINDY, which we investigated previously29, seem to indicate the Co state also undergoes Na+ dependent conformational selection to enable substrate binding. However, verifying this hypothesis will require structural char- acterization of a DASS symporter’s outward-facing state. Methods VcINDY expression and purification. Expression and purification of VcINDY were carried out according to our previous protocol1. Briefly, E. coli BL21-AI cells (Invitrogen) were transformed with a modified pET vector61 encoding N-terminal 10x His tagged VcINDY. Cells were grown at 32 °C until OD595 reached 0.8, protein expression occurred at 19 °C following IPTG induction, and cells were harvested 16 h post-induction. Cell membranes were solubilized in 1.2 % DDM and the protein was purified on a Ni2+-NTA column. For cryo-EM and substrate binding experiments, the protein was purified using size-exclusion chromatography (SEC) in different buffers. The protein used for the cysteine alkylation assays was produced as described previously29. Tryptophan fluorescence quenching assay. Tryptophan fluorescence quenching was used to measure affinity of succinate to purified VcINDY in detergent, using a protocol adapted from earlier work on other membrane transporters20,31–34. VcINDY purified by SEC in a buffer of 25 mM Tris pH 7.5, 100 mM NaCl and 0.05% DDM was used to measure succinate affinity, while the 100 mM NaCl was replaced by 100 mM ChCl for affinity measurements in the absence of sodium. Protein was diluted to a final concentration of 4 μM in SEC buffer. Using a Horiba FluoroMax-4 fluorometer (Kyoto, Japan) at 22 °C and a 280 nm excitation wave- length, the emission spectrum was recorded between 290 and 400 nm. The emis- sion maximum was determined to be 335 nm. Subsequently, the change in fluorescent emission at 335 nm was monitored with increasing concentrations of succinic acid (pH 7.5), from 0.1 μM to 1 mM. Each experimental condition was repeated 4 times. The binding curve was fit in Prism using a quadratic binding equation to account for bound substrate62. Amphipol exchange and cryo-EM sample preparation. From Ni2+-NTA pur- ified VcINDY, DDM detergent was exchanged to PMAL-C8 (Anatrace, Maumee, OH) as previously described19,63. Following further purification by SEC in buffer containing 25 mM Tris pH 7.5, 100 mM NaCl and 0.2 mM TCEP, the NaCl con- centration was increased to 300 mM and the protein sample was concentrated to 1.3 mg/mL. For the apo protein preparation, NaCl in the abovementioned SEC buffer was replaced with 100 mM ChCl, and the protein sample was concentrated to 1.3 mg/mL. Cryo-EM grids were prepared by applying 3 μL of protein to a glow-discharged QuantiAuFoil R1.2/1.3 300-mesh grid (Quantifoil, Germany) and blotted for 2.5 to 4 s under 100% humidity at 4 °C before plunging into liquid ethane using a Mark IV Vitrobot (FEI). Cryo-EM data collection. Cryo-EM data were acquired on a Titan Krios micro- scope with a K3 direct electron detector, using a GIF-Quantum energy filter with a NATURE COMMUNICATIONS | (2022) 13:2644 | https://doi.org/10.1038/s41467-022-30406-4 | www.nature.com/naturecommunications 7 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-022-30406-4 20-eV slit width. SerialEM was used for automated data collection64. Each micro- graph was dose-fractioned over 60 frames, with an accumulated dose of 65 e-/Å2. Cryo-EM image processing and model building. Motion correction, CTF esti- mation, particle picking, 2D classification, ab initio model generation, hetero- genous and non-uniform refinement, and per-particle CTF refinement were all performed with cryoSPARC65. Each dataset was processed using the same protocol, except as noted. Micrographs underwent patch motion correction and patch CTF estimation, and those with an overall resolution worse than 8 Å were excluded from subsequent steps. An ellipse-based particle picker identified particles used to generate initial 2D classes. These classes were used for template-based particle picking. Template identified particles were extracted and subjected to 2D classification. A subset of well-resolved 2D classes were used for the initial ab initio model building, while all picked particles were subsequently used for heterogeneous 3D refinement. After multiple rounds of 3D classification (ab initio model generation and heterogeneous 3D refinement with two or more classes), a single class was selected for nonuniform 3D refinement with C2 symmetry imposed, resulting in the final map. All Cryo-EM maps were sharpened using Auto-sharpen Map in Phenix66, models were built in Coot67, and refined in Phenix real space refine68. The model for VcINDY in NaCl was built using the structure of VcINDY embedded in a lipid nanodisc (PDB: 6WW5) as an initial model, with lipid and antibody fragments removed. The VcINDY model in choline used the structure of VcINDY in 300 mM NaCl, with ions and waters removed, as the starting model. The NMR-style analysis used 5 independent runs of phenix.real_space_refine66 to refine the models of VcINDY in apo and in 300 mM NaCl, with ions and waters removed, using unique computational seeds for each run. Each refinement was performed with simulated annealing, without NCS constraints or secondary structure restraints, and a refinement resolution limit of 3.23 Å for both maps. Analysis with or without map sharpening, or randomizing initial atomic positions using phenix.pdbtools, gave similar results. Transport domain maps were scaled to equivalent contours using the scaffold domain’s volume as an internal standard after extracting with phenix.map_box. Figures were made using UCSF Chimera69 and PyMOL70. Cysteine alkylation assay. For the cysteine alkylation experiments, each purified cysteine mutant was exchanged into reaction buffer containing 50 mM Tris, pH 7, 5% glycerol, 0.1% DDM and either 300 mM NaCl or 300 mM ChCl (Na+-free conditions). Protein samples were incubated with 6 mM mPEG5K and samples were taken at the indicated timepoints and immediately quenched by addition of SDS-PAGE samples buffer containing 100 mM methyl methanesulfonate (MMTS). Samples were analyzed with Coomassie Brilliant Blue-stained non-reducing polyacrylamide gels. Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article. Data availability The cryo-EM particle stacks, maps and models generated in this study have been deposited in EMPIAR image archive, EMDB database and the Protein Data Bank, respectively, under accession codes EMPIAR-10969, EMD-25757 and PDB-7T9G) for VcINDY-Na+ (300 mM) structure and under accession codes EMPIAR-10970, EMD- 25756 and PDB-7T9F) for VcINDY-Ch+ structure. Source Data for Fig. 4 are available with the paper. Received: 11 January 2022; Accepted: 28 April 2022; References 1. Mancusso, R., Gregorio, G. G., Liu, Q. & Wang, D. N. Structure and mechanism of a bacterial sodium-dependent dicarboxylate transporter. Nature 491, 622–626 (2012). 2. Mulligan, C., Fitzgerald, G. A., Wang, D. N. & Mindell, J. A. Functional characterization of a Na+-dependent dicarboxylate transporter from Vibrio cholerae. J. Gen. Physiol. 143, 745–759 (2014). 3. Mulligan, C. et al. The bacterial dicarboxylate transporter VcINDY uses a two- domain elevator-type mechanism. Nat. Struct. Mol. Biol. 23, 256–263 (2016). 4. Nie, R., Stark, S., Symersky, J., Kaplan, R. S. & Lu, M. Structure and function of the divalent anion/Na+ symporter from Vibrio cholerae and a humanized variant. Nat. Commun. 8, 15009 (2017). Sauer, D. B. et al. The ups and downs of elevator-type di-/tricarboxylate membrane transporters. FEBS J 289, 1515–1523 (2021). 5. 6. Pajor, A. M. Sodium-coupled dicarboxylate and citrate transporters from the SLC13 family. Pflug. Arch. 466, 119–130 (2014). 7. Youn, J. W., Jolkver, E., Kramer, R., Marin, K. & Wendisch, V. F. Identification and characterization of the dicarboxylate uptake system DccT in Corynebacterium glutamicum. J. Bacteriol. 190, 6458–6466 (2008). 8. Kim, O. B. & Unden, G. The L-tartrate/succinate antiporter TtdT (YgjE) of L-tartrate fermentation in Escherichia coli. J. Bacteriol. 189, 1597–1603 (2007). 9. Wellen, K. E. et al. ATP-citrate lyase links cellular metabolism to histone acetylation. Science 324, 1076–1080 (2009). 10. Ryan, D. G., Frezza, C. & O’Neill, L. A. TCA cycle signalling and the evolution of eukaryotes. Curr. Opin. Biotechnol. 68, 72–88 (2021). 11. Carey, B. W., Finley, L. W., Cross, J. R., Allis, C. D. & Thompson, C. B. Intracellular alpha-ketoglutarate maintains the pluripotency of embryonic stem cells. Nature 518, 413–416 (2015). 12. Morris, J. P. T. et al. alpha-Ketoglutarate links p53 to cell fate during tumour suppression. Nature 573, 595–599 (2019). 13. Klotz, J., Porter, B. E., Colas, C., Schlessinger, A. & Pajor, A. M. Mutations in the Na+/citrate cotransporter NaCT (SLC13A5) in pediatric patients with epilepsy and developmental delay. Mol. Med 22, 310–321 (2016). 14. Dewulf, J. P. et al. SLC13A3 variants cause acute reversible leukoencephalopathy and alpha-ketoglutarate accumulation. Ann. Neurol. 85, 385–395 (2019). 15. Birkenfeld, A. L. et al. Deletion of the mammalian INDY homolog mimics aspects of dietary restriction and protects against adiposity and insulin resistance in mice. Cell Metab. 14, 184–195 (2011). 16. Huard, K. et al. Discovery and characterization of novel inhibitors of the sodium-coupled citrate transporter (NaCT or SLC13A5). Sci. Rep. 5, 17391 (2015). 17. Pajor, A. M. et al. Molecular basis for inhibition of the Na+/citrate transporter NaCT (SLC13A5) by dicarboxylate Inhibitors. Mol. Pharmacol. 90, 755–765 (2016). 18. Willmes, D. M. et al. The longevity gene INDY (I’m Not Dead Yet) in metabolic control: Potential as pharmacological target. Pharm. Ther. 185, 1–11 (2018). 19. Sauer, D. B. et al. Structural basis for the reaction cycle of DASS dicarboxylate transporters. Elife 9, e61350 (2020). 20. Sauer, D. B. et al. Structure and inhibition mechanism of the human citrate transporter NaCT. Nature 591, 157–161 (2021). 21. Vergara-Jaque, A., Fenollar-Ferrer, C., Mulligan, C., Mindell, J. A. & Forrest, L. R. Family resemblances: A common fold for some dimeric ion-coupled secondary transporters. J. Gen. Physiol. 146, 423–434 (2015). 22. Fitzgerald, G. A., Mulligan, C. & Mindell, J. A. A general method for determining secondary active transporter substrate stoichiometry. Elife 6, e21016 (2017). 23. Sampson, C. D. D., Fabregas Bellavista, C., Stewart, M. J. & Mulligan, C. Thermostability-based binding assays reveal complex interplay of cation, substrate and lipid binding in the bacterial DASS transporter, VcINDY. Biochem J. 478, 3847–3867 (2021). 24. Stein, W. D. Transport and Diffusion across Cell Membranes, 374–404 (Academic Press, New York, 1986). 25. Wright, S. H., Hirayama, B., Kaunitz, J. D., Kippen, I. & Wright, E. M. Kinetics of sodium succinate cotransport across renal brush-border membranes. J. Biol. Chem. 258, 5456–5462 (1983). 26. Yao, X. & Pajor, A. M. The transport properties of the human renal Na+-dicarboxylate cotransporter under voltage-clamp conditions. Am. J. Physiol. Ren. Physiol. 279, F54–F64 (2000). 27. Hall, J. A. & Pajor, A. M. Functional characterization of a Na+-coupled dicarboxylate carrier protein from Staphylococcus aureus. J. Bacteriol. 187, 5189–5194 (2005). 28. Pajor, A. M., Sun, N. N. & Leung, A. Functional characterization of SdcF from Bacillus licheniformis, a homolog of the SLC13 Na+/dicarboxylate transporters. J. Membr. Biol. 246, 705–715 (2013). 29. Sampson, C. D. D., Stewart, M. J., Mindell, J. A. & Mulligan, C. Solvent accessibility changes in a Na+-dependent C4-dicarboxylate transporter suggest differential substrate effects in a multistep mechanism. J. Biol. Chem. 295, 18524–18538 (2020). 30. Hammes, G. G., Chang, Y. C. & Oas, T. G. Conformational selection or induced fit: a flux description of reaction mechanism. Proc. Natl Acad. Sci. USA 106, 13737–13741 (2009). 31. Boulter, J. M. & Wang, D. N. Purification and characterization of human erythrocyte glucose transporter in decylmaltoside detergent solution. Prot. Expr. Purif. 22, 337–348 (2001). 32. Li, X. D. et al. Monomeric state and ligand binding of recombinant GABA transporter from Escherichia coli. FEBS Lett. 494, 165–169 (2001). 33. Law, C. J., Yang, Q., Soudant, C., Maloney, P. C. & Wang, D. N. Kinetic evidence is consistent with the rocker-switch mechanism of membrane transport by GlpT. Biochemistry 46, 12190–12197 (2007). 34. Law, C. J., Enkavi, G., Wang, D. N. & Tajkhorshid, E. Structural basis of substrate selectivity in the glycerol-3-phosphate:phosphate antiporter. Biophys. J. 97, 1346–1353 (2009). 8 NATURE COMMUNICATIONS | (2022) 13:2644 | https://doi.org/10.1038/s41467-022-30406-4 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-022-30406-4 ARTICLE 35. Erkens, G. B. & Slotboom, D. J. Biochemical characterization of ThiT from Lactococcus lactis: a thiamin transporter with picomolar substrate binding affinity. Biochemistry 49, 3203–3212 (2010). 36. Canul-Tec, J. C. et al. The ion-coupling mechanism of human excitatory amino acid transporters. EMBO J, 41, e108341 (2021). 37. Kang, B., Tang, H., Zhao, Z. & Song, S. Hofmeister Series: Insights of Ion Specificity from Amphiphilic Assembly and Interface Property. ACS Omega 5, 6229–6239 (2020). 38. Boudker, O., Ryan, R. M., Yernool, D., Shimamoto, K. & Gouaux, E. Coupling substrate and ion binding to extracellular gate of a sodium-dependent aspartate transporter. Nature 445, 387–393 (2007). 39. Dutzler, R., Campbell, E. B., Cadene, M., Chait, B. T. & MacKinnon, R. X-ray structure of a ClC chloride channel at 3.0 Å reveals the molecular basis of anion selectivity. Nature 415, 287–294 (2002). 40. Sippel, K. H. & Quiocho, F. A. Ion-dipole interactions and their functions in proteins. Protein Sci. 24, 1040–1046 (2015). 41. Adams, P. D., Pannu, N. S., Read, R. J. & Brunger, A. T. Cross-validated maximum likelihood enhances crystallographic simulated annealing refinement. Proc. Natl Acad. Sci. USA 94, 5018–5023 (1997). 42. Brunger, A. T. & Adams, P. D. Molecular dynamics applied to X-ray structure refinement. Acc. Chem. Res 35, 404–412 (2002). 43. Vranken, W. F., Vuister, G. W. & Bonvin, A. M. NMR-based modeling and refinement of protein 3D structures. Methods Mol. Biol. 1215, 351–380 (2015). 44. Chakrabarti, P. Does helix dipole have any role in binding metal ions in protein structures? Arch. Biochem Biophys. 290, 387–390 (1991). 45. Ganesan, S. J. & Matysiak, S. Role of backbone dipole interactions in the formation of secondary and supersecondary structures of proteins. J. Chem. Theory Comput 10, 2569–2576 (2014). 46. Huang, Y., Lemieux, M. J., Song, J., Auer, M. & Wang, D. N. Structure and mechanism of the glycerol-3-phosphate transporter from Escherichia coli. Science 301, 616–620 (2003). 47. Abramson, J. et al. Structure and mechanism of the lactose permease of Escherichia coli. Science 301, 610–615 (2003). 48. Yernool, D., Boudker, O., Jin, Y. & Gouaux, E. Structure of a glutamate transporter homologue from Pyrococcus horikoshii. Nature 431, 811–818 (2004). 49. Yamashita, A., Singh, S. K., Kawate, T., Jin, Y. & Gouaux, E. Crystal structure of a bacterial homologue of Na+/Cl−-dependent neurotransmitter transporters. Nature 437, 215–223 (2005). 50. Drew, D. & Boudker, O. Shared molecular mechanisms of membrane transporters. Annu Rev. Biochem 85, 543–572 (2016). 51. Zhou, Z. et al. LeuT-desipramine structure reveals how antidepressants block neurotransmitter reuptake. Science 317, 1390–1393 (2007). 52. Reyes, N., Ginter, C. & Boudker, O. Transport mechanism of a bacterial homologue of glutamate transporters. Nature 462, 880–885 (2009). 53. Arkhipova, V., Guskov, A. & Slotboom, D. J. Structural ensemble of a glutamate transporter homologue in lipid nanodisc environment. Nat. Commun. 11, 998 (2020). 54. Hanelt, I., Wunnicke, D., Bordignon, E., Steinhoff, H. J. & Slotboom, D. J. Conformational heterogeneity of the aspartate transporter GltPh. Nat. Struct. Mol. Biol. 20, 210–214 (2013). 55. Georgieva, E. R., Borbat, P. P., Ginter, C., Freed, J. H. & Boudker, O. Conformational ensemble of the sodium-coupled aspartate transporter. Nat. Struct. Mol. Biol. 20, 215–221 (2013). 56. Erkens, G. B., Hanelt, I., Goudsmits, J. M., Slotboom, D. J. & van Oijen, A. M. Unsynchronised subunit motion in single trimeric sodium-coupled aspartate transporters. Nature 502, 119–123 (2013). Jensen, S., Guskov, A., Rempel, S., Hanelt, I. & Slotboom, D. J. Crystal structure of a substrate-free aspartate transporter. Nat. Struct. Mol. Biol. 20, 1224–1226 (2013). 57. 58. Wang, X. & Boudker, O. Large domain movements through the lipid bilayer mediate substrate release and inhibition of glutamate transporters. Elife 9, e58417 (2020). 59. Grewer, C. et al. Charge compensation mechanism of a Na+-coupled, secondary active glutamate transporter. J. Biol. Chem. 287, 26921–26931 (2012). 60. Nath, S. Charge transfer across biomembranes: A solution to the conundrum of high desolvation free energy penalty in ion transport. Biophys. Chem. 275, 106604 (2021). 61. Love, J. et al. The New York Consortium on Membrane Protein Structure (NYCOMPS): a high-throughput platform for structural genomics of integral membrane proteins. J. Struct. Funct. Genomics 11, 191–199 (2010). Jarmoskaite, I., AlSadhan, I., Vaidyanathan, P. P. & Herschlag, D. How to measure and evaluate binding affinities. Elife 9, e57264 (2020). 62. 63. Huynh, K. W. et al. CryoEM structure of the human SLC4A4 sodium-coupled acid-base transporter NBCe1. Nat. Commun. 9, 900 (2018). 64. Schorb, M., Haberbosch, I., Hagen, W. J. H., Schwab, Y. & Mastronarde, D. N. Software tools for automated transmission electron microscopy. Nat. Methods 16, 471–477 (2019). 65. Punjani, A., Rubinstein, J. L., Fleet, D. J. & Brubaker, M. A. cryoSPARC: algorithms for rapid unsupervised cryo-EM structure determination. Nat. Methods 14, 290–296 (2017). 66. Afonine, P. V. et al. Real-space refinement in PHENIX for cryo-EM and crystallography. Acta Crystallogr D. Struct. Biol. 74, 531–544 (2018). 67. Emsley, P. & Cowtan, K. Coot: model-building tools for molecular graphics. Acta Crystallogr D. Biol. Crystallogr 60, 2126–2132 (2004). 68. Adams, P. D. et al. PHENIX: a comprehensive Python-based system for macromolecular structure solution. Acta Crystallogr D. Biol. Crystallogr 66, 213–221 (2010). 69. Pettersen, E. F. et al. UCSF Chimera–a visualization system for exploratory research and analysis. J. Comput Chem. 25, 1605–1612 (2004). 70. DeLano, W. L. The PyMOL User’s Manual, 1 (DeLano Scientific, San Carlos, CA, 2002). Acknowledgements This work was financially supported by the NIH (R01NS108151, R01GM121994 and R01- DK099023), the G. Harold & Leila Y. Mathers Foundation and the TESS Research Foun- dation (to D.N.W); and Wellcome Trust (210121/Z/18/Z) and BBSRC (BB/V007424/1) (to C.M.). D.B.S. was supported by the American Cancer Society Postdoctoral Fellowship (129844-PF-17-135-01-TBE) and Department of Defense Horizon Award (W81XWH-16-1- 0153). We thank the following colleagues for helpful discussions: N. Coudray, R. Gonzalez Jr., M. Lopez Redondo, J.A. Mindell and E. Tajkhorshid. We are also grateful to colleagues at the Biophysics Colab, C. Grewer, R.M. Ryan and X. Wang, for commenting on the manuscript. We thank the staff at the NYU Cryo-EM Facility and the NYU Microscopy Core for assistance in grid screening and the Pacific Northwest Center for Cryo-EM in data collection. EM data processing used computing resources at the HPC Facility of NYULMC. Author contributions J.J.M., J.S. and C.M. purified the proteins. J.J.M., J.C.S. and D.B.S. collected and analyzed the substrate-binding data. C.M. did all the cysteine PEGylation experiments. D.B.S collected and processed the cryo-EM images and built the atomic models. D.B.S and D.N.W. analyzed the structures. D.B.S., C.M. and D.N.W. wrote the manuscript. All authors participated in the discussion and manuscript editing. C.M. and D.N.W. supervised the research. Competing interests The authors declare no competing interests. Additional information Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41467-022-30406-4. Correspondence and requests for materials should be addressed to Christopher Mulligan or Da-Neng Wang. Peer review information Nature Communications thanks Jeff Abramson, Reinhart Reithmeier and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available. Reprints and permission information is available at http://www.nature.com/reprints Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/ licenses/by/4.0/. © The Author(s) 2022 NATURE COMMUNICATIONS | (2022) 13:2644 | https://doi.org/10.1038/s41467-022-30406-4 | www.nature.com/naturecommunications 9
null
10.1085/jgp.202012858
null
null
ARTICLE EVAP: A two-photon imaging tool to study conformational changes in endogenous Kv2 channels in live tissues Parashar Thapa1*, Robert Stewart1*, Rebecka J. Sepela1, Oscar Vivas2, Laxmi K. Parajuli2, Mark Lillya1, Sebastian Fletcher-Taylor1,3, Bruce E. Cohen3,4, Karen Zito2, and Jon T. Sack1,5* A primary goal of molecular physiology is to understand how conformational changes of proteins affect the function of cells, tissues, and organisms. Here, we describe an imaging method for measuring the conformational changes of the voltage sensors of endogenous ion channel proteins within live tissue, without genetic modification. We synthesized GxTX-594, a variant of the peptidyl tarantula toxin guangxitoxin-1E, conjugated to a fluorophore optimal for two-photon excitation imaging through light-scattering tissue. We term this tool EVAP (Endogenous Voltage-sensor Activity Probe). GxTX-594 targets the voltage sensors of Kv2 proteins, which form potassium channels and plasma membrane–endoplasmic reticulum junctions. GxTX-594 dynamically labels Kv2 proteins on cell surfaces in response to voltage stimulation. To interpret dynamic changes in fluorescence intensity, we developed a statistical thermodynamic model that relates the conformational changes of Kv2 voltage sensors to degree of labeling. We used two-photon excitation imaging of rat brain slices to image Kv2 proteins in neurons. We found puncta of GxTX-594 on hippocampal CA1 neurons that responded to voltage stimulation and retain a voltage response roughly similar to heterologously expressed Kv2.1 protein. Our findings show that EVAP imaging methods enable the identification of conformational changes of endogenous Kv2 voltage sensors in tissue. image the conformational changes of endogenous Kv2 voltage- sensitive proteins. Introduction To move the field of voltage-sensitive physiology forward, we need new tools that indicate when and where voltage-sensitive Kv2 proteins form voltage-gated K+ channels (Frech et al., conformational changes in endogenous proteins occur. Many classes of transmembrane proteins have been found to be voltage 1989), bind endoplasmic reticulum proteins to form plasma membrane–endoplasmic reticulum junctions (Johnson et al., sensitive (Bezanilla, 2008). One important class of voltage-sensitive proteins is the voltage-gated ion channels. Electrophysiological 2018; Kirmiz et al., 2018a), regulate a wide variety of physio- techniques have enabled remarkably precise studies of the voltage logical responses in tissues throughout the body (Bocksteins, sensitivity of ionic conductances, primarily under reduc- 2016), and integrate their response to voltage with many other tionist experimental conditions where the channels have been cellular processes, including phosphorylation (Murakoshi et al., removed from their native tissue. In addition to their canon- 1997), SUMOylation (Plant et al., 2011), oxidation (MacDonald ical function as ion-conducting channels, voltage-gated ion et al., 2003), membrane lipid composition (Ramu et al., 2006), channel proteins have nonconducting functions that are in- and auxiliary subunits (Gordon et al., 2006; Peltola et al., 2011). dependent of their ion conducting functions (Tanabe et al., Kv2 proteins are members of the voltage-gated cation channel 1988; Kaczmarek, 2006). These nonconducting protein func- superfamily. The voltage sensors of proteins in this superfamily comprise a bundle of four transmembrane helices termed S1–S4 tions are largely inaccessible to study by electrophysiology and are more poorly understood. Novel approaches are needed (Long et al., 2005, 2007). The S4 helix contains positively to learn more about voltage sensing in intact tissues and to charged arginine and lysine residues, gating charges, that re- unlock the mysterious realm of nonconducting voltage- spond to voltage changes by moving through the transmem- sensitive physiology. Here, we present a new approach to brane electric field (Aggarwal and MacKinnon, 1996; Seoh et al., ............................................................................................................................................................................. 1Department of Physiology and Membrane Biology, University of California, Davis, Davis, CA; 3The Molecular Foundry, Lawrence Berkeley National Laboratory, Berkeley, CA; Laboratory, Berkeley, CA; 5Department of Anesthesiology and Pain Medicine, University of California, Davis, Davis, CA. 4Division of Molecular Biophysics and Integrated Bioimaging, Lawrence Berkeley National 2Center for Neuroscience, University of California, Davis, Davis, CA; *P. Thapa and R. Stewart contributed equally to this paper; Correspondence to Jon T. Sack: jsack@ucdavis.edu. © 2021 Thapa et al. This article is distributed under the terms of an Attribution–Noncommercial–Share Alike–No Mirror Sites license for the first six months after the publication date (see http://www.rupress.org/terms/). After six months it is available under a Creative Commons License (Attribution–Noncommercial–Share Alike 4.0 International license, as described at https://creativecommons.org/licenses/by-nc-sa/4.0/). Rockefeller University Press J. Gen. Physiol. 2021 Vol. 153 No. 11 e202012858 https://doi.org/10.1085/jgp.202012858 1 of 24 1996; Islas and Sigworth, 1999). When voltage sensor domains encounter a transmembrane voltage that is more negative on the inside of the cell membrane, voltage sensors are biased toward resting conformations, or down states, in which gating charges are localized intracellularly. When voltage becomes more posi- tive, gating charges translate toward the extracellular side of the membrane, and voltage sensors are progressively biased toward up states in a process of voltage activation (Armstrong and Bezanilla, 1973; Zagotta et al., 1994; Tao et al., 2010; Xu et al., 2019). Channel pore opening is distinct from, but coupled to, voltage sensor movement. In some voltage-gated ion channel proteins, voltage sensor movement is coupled to nonconducting protein functions (Tanabe et al., 1988; Kaczmarek, 2006). To study the functional outputs of voltage sensors, it is essential to measure voltage sensor activation itself. Conformational changes of voltage sensors have been detected with electro- physiological measurements of gating currents (Armstrong and Bezanilla, 1973; Schneider and Chandler, 1973; Bezanilla, 2018) or by optical measurements from fluorophores inserted near voltage sensors by genetic encoding (Lin and Schnitzer, 2016) or chemical modification (Zhang et al., 2015). However, the following experimental limitations prevent these existing techniques from measuring conformational changes of voltage sensors of most endogenous proteins: gating currents can only be measured when the proteins are expressed at high density in a voltage-clamped membrane; engineered proteins differ from endogenous channels; most chemical modification strategies result in off-target labeling; and conjugation of fluorophores into voltage sensors irreversibly alters structure and function. Here, we develop a different strat- egy to reveal conformational states of Kv2 proteins. To image where in tissue the voltage sensors of Kv2 pro- teins adopt a specific resting conformation, we exploited the conformation-selective binding of the tarantula peptide guang- xitoxin (GxTX)-1E, which can be conjugated to fluorophores to report Kv2 conformational changes (Tilley et al., 2014; Fletcher- Taylor et al., 2020; Stewart et al., 2021). Here, we synthesize GxTX-594, a Ser13Cys GxTX variant conjugated to Alexa Fluor 594, a fluorophore compatible with two-photon excitation imaging through light-scattering tissue. GxTX-594 dynamically binds Kv2 channels in living tissue. When GxTX-594 binds, it becomes immobilized and fluorescently labels Kv2 proteins at the cell surface. When Kv2 channels become voltage activated, GxTX-594 unbinds, resulting in unlabeling (see illustration). This labeling/unlabeling dynamic is similar to a recently re- ported point accumulation for imaging of nanoscale topology superresoluton imaging method (Legant et al., 2016), yet the method reported here is sensitive to changes in protein con- formation. GxTX-594 labeling of Kv2 proteins equilibrates on the time scale of seconds, revealing the probability (averaged over time) that unbound voltage sensors are resting or active. Here, we develop a method to calculate the average conforma- tional status of unlabeled Kv2 proteins from images of GxTX-594 fluorescence and deploy the GxTX-594 probe in brain slices to image voltage-sensitive fluorescence changes that reveal con- formational changes of endogenous neuronal Kv2 proteins. We refer to this type of imaging tool as an Endogenous Voltage- sensor Activity Probe (EVAP). This EVAP approach provides an imaging technique to study conformational changes of en- dogenous voltage-sensitive Kv2 proteins in samples that have not (or cannot) be genetically modified. Materials and methods GxTX-594 synthesis We used solid-phase peptide synthesis to generate a variant of GxTX, an amphiphilic 36-amino acid cystine knot peptide. We synthesized the same peptide used for GxTX-550, Ser13Cys GxTX, where a free thiolate side chain of cysteine 13 is predicted to extend into extracellular solution when the peptide is bound to a voltage sensor (Tilley et al., 2014). GxTX-1E folds by for- mation of three internal disulfides, and cysteine 13 was differen- tially protected during oxidative refolding to direct chemoselective conjugation. Following refolding and thiol deprotection, Alexa Fluor 594 C5 maleimide was condensed with the free thiol, and Ser13Cys (Alexa Fluor 594) GxTX-1E (called GxTX-594) was purified (Fig. S1). The Ser13Cys GxTX peptide was synthesized as previously described (Tilley et al., 2014). Methionine 35 of GxTX was re- placed by the oxidation-resistant noncanonical amino acid norleucine to avoid complications from methionine oxidation, and serine 13 was replaced with cysteine to create a spinster thiol. Ser13Cys GxTX was labeled with a Texas Red derivative (Alexa Fluor 594 C5 maleimide, cat. #10256; Thermo Fisher Scientific) to form GxTX-594. Ser13Cys GxTX lyophilisate was brought to 560 μM in 50% acetonitrile (ACN) + 1 mM Na2EDTA. 2.4 μl of 1M Tris (pH 6.8 with HCl), 4 μl of 10 mM Alexa Fluor 594 C5 maleimide in DMSO, and 17.9 μl of 560 μM Ser13Cys GxTX were added for a final solution of 100 mM Tris, 1.6 mM Alexa Fluor 594 C5 maleimide, and 0.4 mM GxTX in 24 μl of reaction solution. Reactants were combined in a 1.5-ml low- protein–binding polypropylene tube (LoBind, cat. #022431081; Eppendorf) and mixed at 1,000 rpm at 20°C for 4 h (Thermo- mixer 5355 R; Eppendorf). After incubation, the tube was centrifuged at 845 RCF for 10 min at room temperature. A purple pellet was observed after centrifugation. The supernatant was transferred to a fresh tube and centrifuged at 845 RCF for 10 min. After this second centrifugation, no visible pellet was seen. The supernatant was injected onto a reverse-phase HPLC C18 column (Biobasic 4.6-mm RP-C18 5 μm, cat. #2105-154630; Thermo Fisher Scientific) equilibrated in 20% ACN, 0.1% tri- fluoroacetic acid (TFA) at 1 ml/min, and eluted with a protocol holding in 20% ACN for 2 min, increasing to 30% ACN over 1 min, then increasing ACN at 0.31% per minute. HPLC effluent was monitored by fluorescence and an absorbance array detec- tor. 1-ml fractions were pooled based on fluorescence (280-nm Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 2 of 24 excitation, 350-nm emission) and absorbance (214 nm, 280 nm, and 594 nm). GxTX-594 peptide–fluorophore conjugate eluted at ∼35% ACN, and mass was confirmed by mass spectrometry using a Bruker ultrafleXtreme matrix-assisted laser desorption ionization time-of-flight (MALDI-TOF; Fig. S1). Samples for identification from HPLC eluant were mixed 1:1 in an aqueous solution of 25% MeOH and 0.05% TFA saturated with α-cyano-4- hydrocinnamic acid, pipetted onto a ground-steel plate, dried under vacuum, and ionized with 60–80% laser power. Molecular species were detected using a reflector mode protocol and quantitated using Bruker Daltonics flexAnalysis 3.4 software. Lyophilizate containing GxTX-594 conjugation product was dissolved in cell external (CE) buffer (defined below) and stored at −80°C. GxTX-594 concentration was determined by 280-nm absorbance using a calculated molar attenuation coefficient of 18,900 M−1 cm−1. Chinese hamster ovary (CHO) cell methods CHO cell culture and transfection The CHO-K1 cell line (American Type Culture Collection) and a subclone transfected with a tetracycline-inducible rat Kv2.1 construct (Kv2.1-CHO; Trapani and Korn, 2003) were cultured as described previously (Tilley et al., 2014). To induce Kv2.1 expression in Kv2.1-TREx-CHO cells, 1 μg/ml minocycline (cat. #ALX-380-109-M050; Enzo Life Sciences), prepared in 70% ethanol at 2 mg/ml, was added to the maintenance media to induce Kv2.1 expression. Minocycline was added 40–48 h before imaging and voltage-clamp fluorometry experiments. Minocy- cline was added 1–2 h before whole-cell ionic current recordings to limit K+ conductance such that voltage clamp could be maintained. Transfections were achieved with Lipofectamine 2000 (cat. #1668027; Life Technologies). 1.1 μl Lipofectamine was diluted, mixed, and incubated in 110 μl of Opti-MEM (pro- duct no. 31985062, lot no. 1917064; Gibco-BRL) in a 1:100 ratio for 5 min at room temperature. Concurrently, 1 μg of plasmid DNA and 110 μl of Opti-MEM were mixed in the same fashion. DNA and Lipofectamine 2000 mixtures were mixed and left at room temperature for 20 min. Then, the transfection cocktail mixture was added to 2 ml of culture media in a 35-mm cell culture dish of CHO cells at ∼40% confluency and allowed to settle at 37°C in 5% CO2 for 4–6 h before the media were replaced. Cells were given 40–48 h recovery following transfection before being used for experiments. Rat Kv2.1-GFP (Antonucci et al., 2001), rat Kv2.2-GFP (Kirmiz et al., 2018b), rat Kv1.5-GFP (Li et al., 2001), rat Kv4.2-GFP (Shibata et al., 2003), mouse BK-GFP, rat KvBeta2 (Shibata et al., 2003), and rat KCHiP2 (An et al., 2000) plasmids were all gifts from James Trimmer (University of California, Davis, Davis, CA). Identities of constructs were confirmed by sequencing from their cytomegalovirus promoter. To minimize any day-to-day variances, the cells in experiments shown in Fig. 4 or Fig. S2 were each plated for all transfections from a single-cell suspension, transfected in parallel, and imaged 2 d later using the same thawed aliquot of GxTX-594. Confocal and Airy disk imaging Confocal images were obtained with an inverted confocal system (LSM 880 410900-247-075; Zeiss) run by ZEN Black 2.1 software. A 63×/1.40 NA oil DIC objective (420782-9900-799; Zeiss) was used for most imaging experiments; a 63×/1.2 NA water DIC objective (441777-9970-000; Zeiss) was used for voltage clamp fluorometry experiments. GFP and YFP were excited with the 488-nm line from an argon laser (3.2 mW at installation) powered at 0.5% unless otherwise noted. GxTX-594 was excited with a 594-nm helium–neon laser (0.6 mW at in- stallation) powered at 10% unless otherwise noted. Wheat germ agglutinin (WGA)-405 was excited with a 405-nm diode laser (3.5 mW at installation) powered at 1% unless otherwise noted. Temperature inside the microscope housing was 27–30°C. In confocal imaging mode, fluorescence was collected with the microscope’s 32-detector gallium arsenide phosphide de- tector array arranged with a diffraction grating to measure 400–700-nm emissions in 9.6-nm bins. Emission bands were 495–550 nm for GFP and YFP, 605–700 nm for GxTX-594, and 420–480 nm for WGA-405. Point spread functions were calcu- lated using ZEN Black software using emissions from 0.1-μm fluorescent microspheres prepared on a slide according to manufacturer’s instructions (cat. #T7279; Thermo Fisher Sci- entific). The point spread functions for confocal images with the 63×/1.40 NA oil DIC objective in the x–y direction were 228 nm (488-nm excitation) and 316 nm (594-nm excitation). In Airy disk imaging mode, fluorescence was collected with the microscope’s 32-detector gallium arsenide phosphide de- tector array arranged in a concentric hexagonal pattern (Airy- scan 410900-2058-580; Zeiss). After deconvolution, the point spread functions for the 63×/1.40 NA oil objective with 488-nm excitation was 124 nm in x–y and 216 nm in z and with 594-nm excitation, 168 nm in x–y and 212 nm in z. For the 63×/1.2 NA water objective, the point spread function with 488-nm excita- tion was 187 nm in x–y and 214 nm in z and with 594-nm ex- citation, 210 nm in x–y and 213 nm in z. Unless stated otherwise, cells were plated in uncoated 35-mm dishes with a 7-mm inset no. 1.5 coverslip (cat. #P35G-1.5-20-C; MatTek). The CHO CE solution used for imaging and electro- physiology contained (in mM): 3.5 KCl, 155 NaCl, 10 HEPES, 1.5 CaCl2, and 1 MgCl2, adjusted to pH 7.4 with NaOH. Measured osmolality was 315 mOsm/liter. When noted, solution was sup- plemented with either 10 mM glucose (CEG) or 10 mM glucose and 1% BSA (CEGB). For time lapse, GxTX-594 concentration–effect experiments, Kv2.1-CHO cells were plated onto 22 × 22-mm no. 1.5H cover- glass (Deckglaser), and Kv2.1 expression was induced with minocycline 48 h before experiments such that all Kv2.1-CHO cells expressed Kv2.1. Prior to imaging, cell maintenance media were removed and replaced with CEGB, then the coverslip was mounted on an imaging chamber (cat. #RC-24E; Warner In- struments) with vacuum grease. We performed three 10-fold serial dilutions of 1,000 nM GxTX-594 to generate the range of concentrations used for this concentration–effect experiment and applied each concentration of GxTX-594 to Kv2.1-CHO cells for 15 min followed by 15 min of washout before the next con- centration of GxTX-594 was applied. Solutions were added to the imaging chamber perfusion via a syringe at a flow rate of ∼1 ml per 10 s. Images were taken every 5 s. Laser power was set to 0.5% for the 488-nm laser and 1.5% for the 594-nm laser. For Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 3 of 24 colocalization experiments with GFP-tagged proteins, cells were incubated in 100 nM GxTX-594 for 5 min and then washed with 1 ml CEGB three times before imaging. Whole-cell voltage clamp for CHO cell imaging Kv2.1-CHO cells plated in glass-bottom 35-mm dishes were in- cubated in minocycline for 48 h to induce Kv2.1 channel expression. Cells were washed with CEGB, placed in the microscope, and then incubated in 100 μl of 100 nM GxTX-594 for 5 min to label cells. Before patch clamp, the solution was diluted with 1 ml of CEG for a working concentration of 9 nM GxTX-594 during experiments. We determined the time re- quired for GxTX-594 to reach a stable fluorescence after dilution from 100 nM to 9 nM by time-lapse imaging during dilution (Fig. 5 A). The rate of fluorescence change (kΔF) indicated that, on average, equilibration was 85% complete 9 min after dilution to 9 nM. After dilution to 9 nM, the mean fluorescence intensity decreased by 39.1 ± 8.4% and remained stable (Fig. 5 B). Cells with obvious GxTX-594 surface staining were voltage clamped in whole-cell mode with an EPC-10 patch-clamp amplifier (HEKA Elektronik) run by Patchmaster software (v2 × 90.2; HEKA Elektronik). The patch pipette contained the following potassium-deficient Cs+ internal pipette solution to reduce voltage error by limiting outward current: 70 mM CsCl, 50 mM CsF, 35 mM NaCl, 1 mM EGTA, and 10 mM HEPES, brought to pH 7.4 with CsOH. Osmolality was 310 mOsm/liter. The liquid junction potential was calculated to be 3.5 mV and was not corrected. Borosilicate glass pipettes (cat. #BF150-110-10HP; Sutter Instruments) were pulled with blunt tips to resistances <3.0 MΩ in these solutions. Cells were held at −80 mV (unless noted otherwise) and stepped to indicated voltages. The voltage step stimulus was maintained until any observed change in fluorescence was complete. Cells were stepped to −80 mV for at least 1 min and were visually inspected to determine sufficient fluorescence recovery before being stepped to another voltage. This recovery time did not always allow GxTX-594 labeling to equilibrate fully. For stimulus frequency dependence experi- ments, cells were given 2-ms steps to 40 mV at the stated fre- quencies (0.02, 5, 10, 20, 50, 100, 150, or 200 Hz). Images for voltage clamp fluorometry were taken in Airy disk imaging mode with the settings described above. During time-lapse imaging of voltage-clamped cells, we no- ticed that GxTX-594 fluorescence in the center of the glass- adhered surface responded more slowly to voltage changes than the periphery. At −80 mV, at the center of the glass- adhered cell surface, relabeling was incomplete after 500 s; however, at the cell periphery, relabeling neared completion within 200 s (Fig. S4 A). To quantify this observation, concentric circular regions of interest (ROIs) were drawn, with the smallest ROI in the center of the glass-adhered surface. The average fluorescence intensities from each of these ROIs were compared with each other and an ROI at the cell periphery (Fig. S4, B and C). We quantified kΔF by fitting the average fluorescence in- tensities from each ROI with a monoexponential function (Eq. 1 in Image analysis). In response to voltage change, the kΔF was consistently slower at the center of the glass-adhered surface and faster toward the outer periphery (Fig. S4, D and E). While kΔF was consistently slower at the center of the glass-adhered surface, we observed that kΔF was more pronounced during la- beling at −80 mV than unlabeling at 40 mV. When cells were held at a membrane potential of −80 mV, kΔF at the periphery was ∼10-fold faster than the kΔF at the center of the cell. In comparison, when the membrane potential was held at 40 mV, kΔF at the periphery was approximately threefold faster than the kΔF at the center of the cell. The gradual change in fluorescence intensity over many seconds after a voltage step is inconsistent with a fast electrochromic effect leading to fluorescence change as the change in fluorescence intensity does not occur instan- taneously when the membrane potential is stepped. Addition- ally, the slowing of kΔF in subcellular regions farther from the periphery of cells is inconsistent with a slower electrochromic effect as whole-cell voltage clamp should render the membrane surface isopotential. K+ channel–GFP ionic current recordings Whole-cell voltage clamp was used to measure currents from CHO cells expressing Kv2.1-GFP, Kv2.2-GFP, Kv1.5-GFP, Kv4.2- GFP, BK-GFP, or GFP that was transfected as described above. Cells were plated on glass-bottom dishes. Cells were held at −80 mV, then 100-ms voltage steps were delivered ranging from −80 mV to +80 mV in +5-mV increments. Pulses were repeated every 2 s. The external (bath) solution contained CE solution. The internal (pipette) solution contained (in mM) 35 KOH, 70 KCl, 50 KF, 50 HEPES, and 5 EGTA, adjusted to pH 7.2 with KOH. Liquid junction potential was calculated to be 7.8 mV and was not corrected. Borosilicate glass pipettes (cat. #BF150-110-10HP; Sutter Instruments) were pulled into pipettes with resistance <3 MΩ for patch clamp recording. Recordings were at room tem- perature (22–24°C). Voltage clamp was achieved with an Axon Axopatch 200B Amplifier (Molecular Devices) run by Patchmaster software (v2 × 90.2; HEKA Elektronik). Holding potential was −80 mV capacitance and ohmic leak were subtracted using a P/5 protocol. Recordings were low-pass filtered at 10 kHz and digitized at 100 kHz. Voltage clamp data were plotted with Igor Pro 7 (WaveMetrics). As the experiments plotted in Fig. S2 A were merely to confirm functional expression of ion channels at the cell surface, series resistance compensation was not used, and sub- stantial cell voltage errors are predicted during these experiments. Kv2.1 ionic current recordings Prior to patching, Kv2.1-CHO cells were washed in divalent-free PBS and then harvested in Versene (cat. #15040066; Gibco-BRL). Cells were scraped and transferred to a polypropylene tube, pelleted, and washed three times at 1,000 g for 2 min and then resuspended in the same external solution as used in the re- cording chamber bath. Cells were rotated in a polypropylene tube at room temperature (22–24°C) until use. Cells were then pipetted into a 50-μl recording chamber (RC-24N; Warner In- struments) prefilled with external solution and allowed to settle for ≥5 min. After adhering to the bottom of the glass recording chamber, cells were thoroughly rinsed with external solution using a gravity-driven perfusion system. Cells showing uniform intracellular GFP expression of intermediate intensity were se- lected for patching. Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 4 of 24 Voltage clamp was achieved with a patch clamp amplifier (Axon Axopatch 200B; Molecular Devices) run by Patchmaster software. Borosilicate glass pipettes (BF150-110-7.5HP; Sutter Instruments) were pulled with blunt tips, coated with silicone elastomer (Sylgard 184; Dow Corning), heat cured, and tip fire- polished to resistances <4 MΩ. Capacitance and ohmic leak were subtracted using a P/5 protocol. Recordings were low-pass fil- tered at 10 kHz using the amplifier’s built-in Bessel function and digitized at 100 kHz. For whole-cell ionic current measurements in Kv2.1-CHO cells, the external patching solution contained (in mM) 3.5 KCl, 155 NaCl, 10 HEPES, 1.5 CaCl2, and 1 MgCl2, adjusted to pH 7.4 with NaOH. The internal (pipette) solution contained (in mM) 70 KCl, 5 EGTA, 50 HEPES, 50 KF, and 35 KOH, adjusted to pH 7.4 with KOH. The osmolality was 315 mOsm/liter for the ex- ternal solution and 310 mOsm/liter for the internal solution measured by a vapor pressure osmometer. Following estab- lishment of the whole-cell seal, ionic K+ current recordings were taken in the presence of a vehicle, which consisted of 100 nM tetrodotoxin, 10 mM glucose, and 0.1% BSA prepared in external solution. Cells were held at −100 mV with channel activation steps ranging from −80 mV to +120 mV in increments of +5 mV (100 ms) before being returned to 0 mV (100 ms) to record tail currents. The intersweep interval was 2 s. To determine the bioactivity of GxTX-594, Kv2.1 ionic currents were recorded once more, 5 min following the wash-in of bath solution also containing 100 nM GxTX-594. Wash-ins were performed while holding at −100 mV; 100 μl was washed through the chamber and removed distally through vacuum tubing to maintain con- stant bath fluid level. Ionic current analysis The average current in the 100 ms before voltage step was used to zero subtract the recording. Outward current taken as the mean value between 90 and 100 ms of the channel activation step was used to calculate and correct for series resistance- induced voltage error. Tail current values were derived from the mean value between 0.2 and 1.2 ms of the 0-mV tail current step. Tail current was normalized by the mean activation step current from 50 to 80 mV and plotted against the estimated membrane potential, which had been corrected for voltage error and the calculated liquid junction potential of 8.5 mV. These tail GV plots were fit with a fourth-power Boltzmann function (Sack et al., 2004), and the fit parameters were used for statistical analysis. Brain slice methods Hippocampal slice culture preparation and transfection All experimental procedures were approved by the University of California, Davis, institutional animal care and use committee and were performed in strict accordance with the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. Animals were maintained under standard light–dark cycles and had ad libitum access to food and water. Organotypic hippocampal slice cultures were prepared from postnatal day 5–7 rats, as previously described (Stoppini et al., 1991) and detailed in a video protocol (Opitz-Araya and Barria, 2011). DIV15–30 neurons were transfected 2–6 d before imaging via biolistic gene transfer (160 psi, Helios gene gun; Bio-Rad) as described in a detailed video protocol (Woods and Zito, 2008). 10 μg of plasmid was coated to 6–8 mg of 1.6-μm gold beads. Two-photon excitation slice imaging Image stacks (512 × 512 pixels, 1-mm Z-steps, 0.035 μm/pixel) were acquired using a custom two-photon excitation microscope (LUMPLFLN 60XW/IR2 objective, 60×/1.0 NA; Olympus) with two pulsed Ti:sapphire lasers (Mai Tai; Spectra Physics) tuned to 810 nm (for GxTX-594 imaging) and 930 nm (for GFP imaging) and controlled with ScanImage software (Pologruto et al., 2003). After identifying a neuron expressing Kv2.1-GFP, perfusion was stopped, and GxTX-594 was added to the static bath solution to a final concentration of 100 nM. After 5-min incubation, perfu- sion was restarted, leading to washout of GxTX-594 from the slice bath. Red and green photons (565dcxr, BG-22 glass, HQ607/ 45; Chroma Technology) emitted from the sample were collected with two sets of photomultiplier tubes (R3896; Hamamatsu). Whole-cell voltage clamp for brain slice imaging Organotypic hippocampal slice cultures (6–7 DIV, not trans- fected) were transferred to an imaging chamber with re- circulating artificial cerebrospinal fluid (ACSF) maintained at 30°C. To hold the slice to the bottom of the chamber, a horseshoe-shaped piece of gold wire was used to weight the membrane holding the slice. ACSF contained (in mM) 127 NaCl, 25 NaHCO3, 25 D-glucose, 2.5 KCl, 1.25 NaH2PO4, 1 MgCl2, 2 CaCl2, and 200 nM tetrodotoxin, pH 7.3, and aerated with 95% O2/5% CO2 (∼310 mOsm). 4 ml of 100 nM GxTX-594 in ACSF was used in the circulating bath to allow the toxin to reach the slice and reach the desired concentration of 100 nM throughout the circulating bath. Images were acquired beginning 3 min after GxTX-594 was added. Apparent CA1 neurons with GxTX-594 labeling in a Kv2-like pattern were selected for whole-cell patch clamp. Voltage clamp was achieved using an Axopatch 200B Amplifier (Molecular Devices) controlled with custom software written in MATLAB (MathWorks, Inc.). Patch pipettes (5–7 MΩ) were filled with intracellular solution containing (in mM) 135 Cs-methanesulfonate, 10 Na2-phosphocreatine, 3 Na-L- ascorbate, 4 NaCl, 10 HEPES, 4 MgCl2, 4 Na2ATP, and 0.4 NaGTP, pH 7.2. Neurons were clamped at −70 mV. Input resistance and holding current were monitored throughout the experiment. Cells were excluded if the pipette series resistance was >25 MΩ or if the holding current exceeded −100 pA. To activate Kv2 channels, a 50- s depolarizing step from −70 mV to 0 mV was given. Image analysis Fluorescence images were analyzed using ImageJ 1.52n software (Schneider et al., 2012). ROIs encompassed the entire fluorescent region of an individual cell or neuron unless mentioned other- wise. ROIs were drawn manually. Analysis of images was con- ducted independently by multiple researchers who produced similar results, but analysis was not conducted in a blinded or randomized fashion. Fluorescence intensity (F) was background subtracted using the mean F of a region that did not contain cells. In experiments with CHO cells where the bath solution contained 9 nM GxTX-594, the apparent surface membrane of most cells (40 Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 5 of 24 of 47) lacking Kv2.1-GFP protein had lower F than the background region that did not contain cells (Fig. S3 B, horizontal black dashed line), indicating that the background F was overestimated. Based on the mean F from the apparent surface membrane of cells lacking Kv2.1 protein, the background was overestimated by 24%. The signal from the surface membrane of cells stably transfected with rat Kv2.1 (Kv2.1-CHO) was, on average, 18× higher than re- gions that did not contain cells, making the error generated by our overestimation of the background ∼1%. As the error produced by this background subtraction method was relatively small, it was not corrected. For F/Finit normalization, Finit was the mean fluo- rescence preceding the indicated voltage stimuli, or the max ob- served intensity in concentration–effect experiments. Further details of specific normalization and background subtraction procedures are provided in the figure legends. Time dependence of fluorescence intensity was fit with a monoexponential decay: F (cid:2) F∞ + (F0 − F∞) e −(t−t0 τ ) , (1) where τ = 1/kΔF, t = time, t0 = time at start of fit, F = fluorescence intensity, F0 = fluorescence at start of fit, and F∞ = fluorescence after infinite time. The Kv2.1–GxTX-594, fluorescence–voltage (FV) responses were fit with a Boltzmann function: f (V) (cid:2) offset + A 8 >< >: − 1 + e 9 >= >; , 1 h(cid:2) (cid:3) i V−V1 2 ∗ zF RT (2) where V = voltage, offset = the offset from zero of the Boltzmann distribution, A = the amplitude, z = number of elementary charges, F = Faraday’s constant, R = the universal gas constant, and T = temperature (held at 295°K). For colocalization analyses, the Pearson coefficient was calcu- lated using the JACoP plugin (Bolte and Cordelières, 2006). Co- localization analyses were conducted within ROIs defining individual cells. A Pearson correlation coefficient value of 0 sig- nifies uncorrelated pixels between two images, and a Pearson correlation coefficient value of 1 signifies complete correlation between pixels from two images. Correlation between pixels from two fluorescent recordings of an image suggests spatial colocali- zation of proteins. Plotting and curve fitting was performed with Igor Pro 7 or 8 (WaveMetrics), which performs nonlinear least squares fits using a Levenberg–Marquardt algorithm. Sample sizes of n ≥ 3 were selected to confirm reproducibility. Sample sizes of n ≥ 6 were selected to power nonparametric statistical compar- isons to discern P < 0.01. The α for statistical significance in nonparametric statistical comparisons was adjusted for multiple comparisons using the Bonferroni method. The Bonferroni- corrected P value = α , where α = 0.01 and n represents the number n of comparisons made in an experiment. Error values from indi- vidual curve fittings are SDs. All other errors, including error bars, indicate SEs. Arithmetic means are reported for intensity measurements and correlation coefficients. As the distributions underlying variability in results are unknown, nonparametric statistical comparisons were conducted with Mann–Whitney U tests, and two-tailed P values were reported individually if P > 0.0001. Parametric statistical tests, which include the Student’s t test, ANOVA, and Tukey’s post hoc test, were performed with paired data and on sample sizes of n ≤ 6 due to the weak statistical power of nonparametric tests when comparing small sample sizes. EVAP model In the EVAP model, at any given voltage, there is a probability that a voltage sensor is either in its resting conformation (Presting) or in its activated conformation (Pactivated) such that Pactivated = (1 – Presting). The equilibrium for voltage sensor activation is then a ratio of ac- tivated-to-resting voltage sensors (Pactivated/Presting) in which Pactivated Presting Pactivated Presting unlabeled (cid:2) e (V−V1/2,unlabeled ) zF RT labeled (cid:2) e (V−V1/2,labeled ) zF RT, (3a) (3b) where V1/2 is the voltage where Pactivated/Presting = 1. In a prior study, our analysis of the conductance–voltage relation of Kv2.1 yielded a V1/2 = −32 mV with z = 1.5 elementary charges (e0) for the early movement of four independent voltage sensors, and we found that with a saturating concentration of GxTX, the V1/2 = +42 mV (Fig. 1 C; Tilley et al., 2019). These values were used for V1/2,unlabeled, z, and V1/2,labeled, respectively (Table 1). To relate voltage sensor activation to transient labeling and unlabeling, we used microscopic binding (kon[EVAP]) and unbinding (koff) rates that are distinct for resting and activated voltage sensors. We estimated values for these rates assuming and kΔF (cid:2) kon[EVAP] + koff Kd (cid:2) koff kon . (4) (5) To calculate the kon,resting and koff,resting values reported in Table 1, – we used the saturating value at negative voltages of the kΔF voltage relation (see Fig. 6 E), and Kd from concentration–effect imaging (Fig. 2 D). In 9 nM GxTX-594, at greater than +40 mV, voltage-dependent unlabeling was nearly complete, indicating that koff,activated >> kon,activated[EVAP]. The model does not include EVAP signal that is insensitive to voltage (Fig. 6 C). We input the saturating amplitude of the Boltzmann fit to the kΔF at positive voltages as koff,activated (Fig. 6 E). The slow labeling of activated voltage sensors confounded attempts to measure kon,activated di- rectly, and we used the statistical thermodynamic principle of microscopic reversibility (Lewis, 1925) to constrain kon,activated: Pactivated Presting Pactivated Presting , labeled , unlabeled (cid:2) koff ,resting kon,resting koff ,activated kon,activated . (6) The EVAP model depicted in Scheme 2 has only a single microscopic binding rate, kon,total[EVAP], and unbinding rate, koff,total. kon,total is a weighted sum of both kon,resting and kon,activated from Scheme 1. The weights for kon,total are the relative probabilities that unlabeled voltage sensors are resting or activated, which is determined at any static voltage by an equilibrium constant, Pactivated Presting unlabeled: Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 6 of 24 kon,total (cid:2) kon,resting[EVAP] · 1 1 + Pactivated Presting + unlabeled ratio of fluorescence at a test voltage to fluorescence at a prior voltage (F/Finit) is equal to the probability that a Kv2 subunit is reversibly labeled by GxTX-594 (Plabeled): kon,active[EVAP] · 1 + 1 1 . Pactivated Presting unlabeled (7) F Finit (cid:2) Plabeled Plabeled,init . (9) The equilibrium Plabeled at any voltage can be determined from microscopic binding rates associated with Scheme 2 where Similarly, koff,total is determined by the unbinding rate from resting voltage sensors (koff,resting) and the unbinding rate from activated voltage sensors (koff,activated) and weighted such that Plabeled (cid:2) 1 1 + Kd,total [EVAP] (cid:2) 1 1 + koff ,total kon,total·[EVAP] . (10) koff ,total (cid:2) koff ,resting · 1 1 + Pactivated Presting labeled + koff ,active · 1 + . 1 1 Pactivated Presting labeled Predictions of F / Finit and kΔF during trains of 2-ms voltage steps from −80 mV to +40 mV were made from the EVAP model by summing the products of time-averaged probability of being at each voltage (PVn) and the fluorescence change predicted at that voltage (ΔFVn) : (8) (cid:4) F Finit (cid:2) (PV1 · ΔFV1) + (PV2 · ΔFV2) + ... + (PVn · ΔFVn). (11) Using kon,total and koff,total, we compute kΔF using Eq. 4, as im- plemented in Data S1. (Scheme 1) (Scheme 2) The EVAP model was also used to predict the magnitude of GxTX-594 fluorescence changes on cell surfaces. In theory, the For voltage steps from −80 to +40 mV, Eq. 11 is: (cid:4) F Finit (cid:2) (P40mV · ΔF40mV) + (P−80mV · ΔF−80mV). We predicted EVAP kinetic responses as kΔF (cid:2) (PV1 · kΔF,V1) + (PV2 · kΔF,V2) + ... + (PVn · kΔF,Vn), (12) where PVn is as in Eq. 11 and kΔF,n is kΔF at that particular voltage. For voltage steps from −80 to +40 mV, Eq. 12 is: kΔF (cid:2) (P40mV · kΔF40mV) + (P−80mV · kΔF−80mV). Online supplemental material Fig. S1 pertains to the synthesis of GxTX-594. It shows a model of GxTX-594, and provides HPLC chromatograms and MALDI-TOF mass spectrometry profiles of Ser13Cys GxTX and GxTX-594. Fig. S2 shows GxTX-594 selectively labeling Kv2 proteins at the cell surfaces. Fig. S3 shows that GxTX-594 labeling of surface membranes requires Kv2 proteins. Fig. S4 demonstrates that extracellular access can impact GxTX-594 labeling kinetics. Fig. S5 demonstrates that variation in bath temperature does not account for variability of GxTX-594 kinetics. Fig. S6 is an ex- tended image gallery of GxTX-594 labeling CA1 hippocampal pyramidal neurons transfected with Kv2.1 GFP. Video 1 is a time-lapse image sequence of GxTX-594 fluorescence on a voltage-clamped CA1 hippocampal pyramidal neuron while it is depolarized from −70 to 0 mV. Data S1 is a spreadsheet that sets up calculations to generate EVAP model predictions. Results GxTX-594 retains bioactivity for Kv2.1 after chemoselective modification To monitor activation of Kv2 proteins in tissue slices, we syn- thesized an EVAP compatible with two-photon imaging. We previously presented an EVAP that was a synthetic derivative of GxTX conjugated to a DyLight 550 fluorophore (GxTX-550; Tilley et al., 2014). DyLight 550 has poor two-photon excitation properties, and for this study, it was replaced with Alexa Fluor Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 7 of 24 Figure 1. GxTX-594 modulates Kv2.1 conductance. (A) Representative Kv2.1-CHO current response under whole-cell voltage clamp. Cells were given 100-ms, 5-mV increment voltage steps ranging from −80 mV (blue) to +120 mV (red) and then stepped to 0 mV to record tail currents. The holding potential was −100 mV. (B) Kv2.1 currents from the same cell 5 min after the addition of 100 nM GxTX-594. Scale bars are the same for A and B. (C) Normalized conductance–voltage relationships from Kv2.1 tail currents before application of GxTX-594 (n = 13). Different symbols correspond to individual cells, and the green corresponds to cell in A. (D) Normalized conductance—voltage relationships in 100 nM GxTX-594 (n = 11). (E) Mean midpoint of each of four independent voltage sensors in the fourth-power Boltzmann fit (V1/2) before (−31 ± 6 mV SD) and after (+27 ± 10 mV SD) 100 nM GxTX-594. ***, P < 0.0001 by Mann–Whitney U test. (F) Mean e0 associated with Boltzmann fit (z) before (1.5 ± 0.3 e0 SD) and after (1.0 ± 0.4 e0 SD) 100 nM GxTX-594. ***, P = 0.0007 by Mann–Whitney U test. (G) Mean midpoint of conductance change in the fourth-power Boltzmann fit (Vmid) before (−2 ± 6 mV SD) and after (+73 ± 13 mV SD) 100 nM GxTX-594. ***, P < 0.0001 by Mann–Whitney U test. 594, a persulfonated Texas Red analogue with a large two- photon excitation cross section and ample spectral separation from GFP, making it well suited for multiplexed, two-photon excitation imaging experiments (Zito et al., 2004). We refer to this EVAP variant as GxTX-594. We performed electrophysiological analyses to determine whether GxTX-594 retains the ability to allosterically modulate Kv2.1 (Fig. 1). GxTX is a partial inverse agonist of Kv2.1, which lowers channel open probability by stabilizing voltage sensors in a resting conformation. Consequently, more positive intracellular voltage is required to activate voltage sensors and achieve the same open probability as without GxTX (Tilley et al., 2019). Previously, we estimated that a Kv2.1 voltage sensor with GxTX bound is 5,400-fold more stable in its resting conformation and requires more positive intracellular voltage to become activated (Tilley et al., 2019). To characterize the efficacy of GxTX-594 in allosterically modulating Kv2.1 gating, we voltage clamped Kv2.1- CHO cells and measured K+ currents in GxTX-594. We analyzed the Kv2.1 conductance–voltage (GV) relation by fitting with a fourth power Boltzmann function. The voltage at which the con- ductance of the fitted function is 50% of maximum, Vmid, was +73 ± 13 mV for 100 nM GxTX-594 (Fig. 1 G). For comparison, the Vmid of 100 nM GxTX was +67 ± 6 mV (Tilley et al., 2019). This shift of the GV indicates that GxTX-594 retains an efficacy similar to GxTX. Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 8 of 24 Table 1. Parameters used for calculations to generate Scheme 1: EVAP allosteric expansion Parameter kon,resting koff,resting kon,activated koff,activated V1/2,unlabeled V1/2,labeled z Value 0.30 μM−1 s−1 0.0081 s−1 0.21 μM−1s−1 0.39 s−1 −32 mV 41 mV 1.5 e0 GxTX-594 labels Kv2 proteins To determine the concentration range where GxTX-594 effec- tively labels Kv2.1 protein, we performed a concentration–effect experiment. As the Kd of our previously presented EVAP, GxTX- 550, was 30.0 ± 3.9 nM at a holding potential of −100 mV (Tilley et al., 2014), we presumed that 1,000 nM would be a sufficient upper bound for concentration–effect experiments with GxTX- 594 (Fig. 2, A and B). Cell surface fluorescence intensity was analyzed immediately after washout of 1, 10, 100, and 1,000 nM GxTX-594 to eliminate fluorescence from toxin in solution. This fluorescence labeling concentration–effect relation was fit with a Langmuir binding isotherm, resulting in a Kd of 26.9 ± 8.3 nM (Fig. 2 C). Due to the incomplete equilibration of labeling at 1 and 10 nM, our measure is expected to overestimate the Kd. The near saturation of the fluorescence concentration–effect relation with 100 nM GxTX-594, and the rate of fluorescence equilibration (Fig. 2 D) suggested that incubation with 100 nM GxTX-594 for 5 min results in near-maximal labeling on the glass-adhered surface of Kv2.1-CHO cells. We assessed whether GxTX-594 fluorescence on CHO-K1 cells is due to the presence of Kv2 proteins. CHO-K1 cells were transfected with either a rat Kv2.1-GFP or Kv2.2-GFP construct, each of which yielded delayed rectifier K+ currents (Fig. S2 A). 2 d after transfection, Kv2.1-GFP or Kv2.2-GFP fluorescence displayed distinct subcellular regions of high density at the glass-adhered surface (Fig. 3 A), which we refer to as Kv2 clusters, a term used to refer to similar high-density regions in neurons and other mammalian cell lines (Kirmiz et al., 2018b). Correlation coefficients indicated a high degree of subcellular colocalization of both Kv2 proteins with GxTX-594 (Fig. 3 B), quantitating the observation that GxTX-594 is not evenly dis- tributed throughout the membrane but is localized to Kv2 clusters. We did not detect a significant difference between the ratios of GxTX-594 to Kv2.1-GFP or Kv2.2-GFP fluorescence (Fig. 3 C; P = 0.74 by Mann–Whitney U test), consistent with the lack of discrimination of GxTX between Kv2.1 and Kv2.2 (Herrington et al., 2006). GxTX accesses the membrane- embedded voltage sensors of Kv2 proteins by partitioning into the outer leaflet of the plasma membrane bilayer (Milescu et al., 2009; Gupta et al., 2015), and a GxTX derivative labeled with a fluorophore that brightens in less polar environments is de- tectable in the membrane of CHO cells without Kv2 proteins (Fletcher-Taylor et al., 2020). However, we failed to find any GxTX-594 labeling of CHO cells in the absence of Kv2 proteins (Fig. S3). GxTX-594 selectively labels Kv2 proteins An important consideration for determination of whether GxTX-594 could reveal conformational changes of endogenous Kv2 proteins is whether the EVAP is selective for Kv2 proteins. Electrophysiological studies have concluded that the native GxTX peptide is selective for Kv2 channels, with some off-target Figure 2. Concentration–effect characteristics of GxTX-594 labeling. (A) Fluorescence from confluent Kv2.1-CHO cells incubated in indicated concen- trations of GxTX-594 for 15 min then washed out before imaging. Imaging plane was near the glass-adhered cell surface. Fluorescence shown corresponds to GxTX-594 (magenta). Scale bar, 20 μm. (B) Time-lapse fluorescence intensity from the concentration–effect experiment shown in A. Fluorescence was not background subtracted. Fmax is intensity while cells were incubated in 1,000 nM GxTX-594. (C) Relative fluorescence intensity of GxTX-594 that remains on cells immediately after washout of indicated concentrations of GxTX-594 from the extracellular solution. Symbols correspond to each of three experiments. Black line is the fit of a Langmuir isotherm for Kd = 26.9 nM ± 8.3. (D) Labeling or unlabeling kinetics for GxTX-594 at indicated concentrations. Symbols correspond to the same experiments as panel C. kΔF values obtained from monoexponential fits (Eq. 1). Error bars represent the SD of kΔF fitting. Black line is fit of the kΔF –concentration relation with Eq. 4, kon = 6.372 × 10−5 ± 0.049 × 10−5 nM−1 s−1; koff = 5.929 × 10−4 ± 0.043 × 10−4 s−1. Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 9 of 24 Figure 3. GxTX-594 colocalizes with Kv2-GFP. (A) Fluorescence from CHO cells transfected with Kv2.1-GFP (top) or Kv2.2-GFP (bottom) and labeled with GxTX-594. Optical sections were near the glass-adhered cell surface. Cells were incubated with 100 nM GxTX-594 for 5 min and rinsed before imaging. Fluorescence shown corresponds to emission of GFP (left), GxTX-594 (middle), or as an overlay of GFP and GxTX-594 (right). Scale bars, 20 μm. (B) Pearson correlation coefficients for GxTX-594 colocalization with Kv2.1-GFP or Kv2.2-GFP. (C) Ratio of fluorescence intensity of the same cells in B when excited at 594 nm versus 488 nm from cells expressing Kv2.1-GFP or Kv2.2-GFP. In B and C, pixel intensities were background subtracted before analyses by subtracting the average fluorescence of a background ROI that did not contain cells from the ROI containing the cell. Circles represent measurements from individual cells, and each transfection group contains data from two separate applications of GxTX-594. Bars represent the mean. No significant difference was detected between Kv2.1-GFP or Kv2.2-GFP by Mann–Whitney U test (P = 0.74). modulation of A-type Kv4 channels (Herrington et al., 2006; Liu and Bean, 2014; Speca et al., 2014). However, electrophysiological testing cannot determine whether ligands bind unless they also alter currents (Sack et al., 2013). Furthermore, structural differ- ences between wild-type GxTX and the GxTX-594 variant could potentially alter selectivity among channel protein subtypes. To test whether GxTX-594 binds other voltage-gated K+ channel subtypes, we quantified surface labeling and analyzed colocalization of GxTX-594 with a selection of GFP-tagged voltage-gated K+ channel subtypes (Fig. 4 A). The ratio of GxTX-594 fluorescence to each GFP-tagged K+ channel subtype was not distinguishable from zero for Kv4.2, Kv1.5, or BK channels (Fig. 4 B), indicating minimal, if any, binding. Fur- thermore, no colocalization was apparent between Kv4.2, Kv1.5, or BK channels and the residual GxTX-594 fluorescence (Fig. 4 C). An additional set of experiments conducted under different microscopy conditions and without the auxiliary subunits of Kv4.2 or Kv1.5 also gave no indication of GxTX-594 labeling (Fig. S2). While we cannot be certain that GxTX-594 does not bind any of the >80 known mammalian proteins containing voltage sensor domains, the lack of labeling of the related voltage-gated K+ channels indicates that GxTX-594 does not promiscuously label voltage sensors. The relationship between GxTX-594 cell-surface fluorescence and Kv2.1 voltage activation To understand the relationship between channel gating and GxTX-594 fluorescence, we determined how fluorescence intensity on cells expressing Kv2.1 responds to changes in membrane voltage. We found that GxTX-594 equilibrated more quickly on the sides of cells than on their glass-adhered surface, presumably due to restricted access to the extracellular space by the glass (Fig. S4). Due to this observation, we chose an imaging plane where fluorescence from GxTX-594 resembled an annulus (Fig. 6 A). This imaging plane varied between cells but was >1 μm above the glass. We developed a labeling protocol in which Kv2.1-CHO cells were incubated for 5 min in a bath solution (CEG) containing 100 nM GxTX-594, which was then diluted with extracellular solution to 9 nM (Fig. 5). Once GxTX-594 fluorescence intensity stabilized at the cell membrane (at least 9 min; Fig. 5), cells were voltage clamped in whole-cell mode. We measured the fluorescence response of GxTX-594 when the membrane voltage of Kv2.1-CHO cells was stepped from a holding potential of −80 mV to more positive voltages that ranged from −40 mV to +80 mV (Fig. 6 A). ROIs corresponding to the cell surface were manually identified and average fluo- rescence intensity quantified from time-lapse sequences. The voltage-dependent reduction in fluorescence equilibrates to a value above background. Even when cells were given a +80-mV depolarizing voltage stimulus, some GxTX-594 fluorescence re- mained (Fig. 6 B). Most of the residual fluorescence appeared to be localized to the cell surface membrane (Fig. 6 A) and varied between cells (Fig. 6 C). As such surface labeling was not pre- sent on CHO cells in the absence of Kv2 proteins (Fig. S3), we consider this residual fluorescence to originate from voltage- insensitive Kv2 proteins. Voltage-insensitive fluorescence Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 10 of 24 Figure 4. GxTX-594 selectively labels Kv2 proteins. (A) Fluorescence from live CHO cells transfected with Kv2.1-GFP, Kv2.2-GFP, Kv4.2- GFP + KChIP2, Kv1.5-GFP + Kvβ2, or BK-GFP (indicated by row) and labeled with GxTX-594. Airy disk imaging was from a plane above the glass-adhered surface. Cells were incubated with 100 nM GxTX-594 and 5 μg/ml WGA-405 then rinsed before imaging. Fluorescence corresponds to emission of GFP (column 1), GxTX-594 (col- umn 2), WGA-405 (column 3), or an overlay (column 4). Scale bars, 20 μm. (B) Intensity of GxTX-594 labeling for different K+ channel-GFP types. Fluorescence intensity resulting from 594-nm excitation of GxTX-594 is divided by fluorescence intensity resulting from 488-nm excitation of GFP. This value was normalized to the average 594:488 ratio from GxTX-594 and Kv2.1-GFP. Circles indicate measurements from individual cells. Only cells with obvious GFP expression were analyzed. For analysis, ROIs were drawn around the cell membrane indicated by WGA-405 fluorescence. Pixel in- tensities were background subtracted be- fore analyses by subtracting the average fluorescence of a background ROI that did not contain cells from the ROI containing the cell; this occasionally resulted in ROIs with negative intensity. Kv2.1, n = 16; Kv2.2, n = 10; Kv4.2, n = 13; Kv1.5, n = 13; and BK, n = 10; n indicates the number of individual cells analyzed in a single dish during a single application of GxTX-594 with the indicated K+ channel-GFP type. Bars represent the mean. Significant differences were observed be- tween 594:488 ratio for Kv2.1 or Kv2.2 and Kv1.5, Kv4.2, or BK by Mann–Whitney U test (P < 0.0001). The P value to determine signifi- cance is adjusted for multiple comparisons us- ing the Bonferroni method, where P < 0.0033 is considered significant, with the caveat that data points under each condition are technical repli- cates. (C) Pearson correlation coefficients between GxTX-594 and GFP. Same cells as B. Significant differences were observed between correlation coefficients for Kv2.1 or Kv2.2 and Kv1.5, Kv4.2, or BK by Mann-Whitney U test (P < 0.0001). could potentially be from Kv2.1–GxTX-594 complexes with im- mobilized voltage sensors or internalized Kv2.1–GxTX-594 com- plexes that remain just under the cell surface (Deutsch et al., 2012; Weigel et al., 2012; Fox et al., 2013a; Weigel et al., 2013). To compare voltage response properties between cells, we used a normalization procedure to analyze only the voltage- sensitive fraction of the fluorescence from each cell, which we defined as the fluorescence that changed between −80 mV and Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 11 of 24 Figure 5. GxTX-594 fluorescence equilibration after dilution from 100 nM to 9 nM. (A) CHO cells transfected with Kv2.1-GFP (left) in 100 nM GxTX-594 (middle) and after dilution to 9 nM GxTX-594 (right). Regions darker than the background bath solution are CHO cells not transfected with Kv2.1-GFP (ar- rowhead). Scale bar, 20 μm. (B) Fluorescence excited at 594 nm during dilution from 100 nM GxTX-594 to 9 nM GxTX-594. Timing of dilution is shown above the graph. Fluorescence was normalized to the mean 594 fluorescence before dilution to 9 nM. Data are from the cell labeled with an arrow in A. 100 nM and 9 nM images in A are from 0 and 12 min, respectively. (C) Rate of GxTX-594 fluorescence decay after dilution. Error bars are SDs of kΔF fit. n = 5 cells. +80 mV. The initial fluorescence at a holding potential of −80 mV was normalized to 100% F/Finit, and residual fluores- cence after a +80-mV step was normalized to 0% F/Finit (Fig. 6 D). To characterize the voltage dependence of the Kv2.1–GxTX-594 interaction, fluorescence–voltage (FV) responses were fit with a Boltzmann distribution (Eq. 2). This fit had a half maximal voltage midpoint (V1/2) of −27 mV and a steepness (z) of 1.4 e0 (Fig. 6 D, bottom panel, black line). This is similar to voltage sensor movement in Kv2.1-CHO cells without any GxTX present: V1/2 = −26 mV, z = 1.6 e0 (Tilley et al., 2019). These results suggest that at 9 nM GxTX-594, the FV appears to be a good surrogate for the gating current–voltage (QV) response of unlabeled Kv2 channels. To determine the voltage dependence of the kinetics of GxTX-594 labeling and unlabeling, we compared kΔF at varying step potentials. We quantified kΔF by fitting the average fluo- rescence from voltage-clamped cells with a monoexponential function (Eq. 1). In response to voltage steps from a holding potential of −80 mV to more positive potentials, kΔF increased progressively as step potential was increased above −40 mV and appeared to begin to saturate at higher voltages (Fig. 6 E). Upon return to −80 mV, kΔF was similar to −40 mV. While the kΔF did not clearly display saturation at positive voltages that would justify fitting with a Boltzmann function, a model of GxTX-594 dynamics, which we develop later in this study, indicated that Boltzmann fitting could yield physical insight (Fig. 6 E, bottom panel, black line). We noted that the degree of variability in kΔF measurements became greater at more positive potentials (Fig. 6 E, top and bottom panels). At −80 mV, there was a twofold range in kΔF values and a ninefold range at +80 mV. The relatively low variation in kΔF at −80 mV suggests that despite variance in fluo- rescence intensity after rebinding (Fig. S4 C), kΔF from fits of the upward relaxation at −80 mV are relatively consistent. The average kΔF equilibration at 10 nM GxTX-594 in concentration–effect ex- periments was comparable to Kv2.1-CHO cells incubated in 9 nM GxTX-594 and voltage clamped at −80 mV (0.0011 s−1 and 0.0014 s−1, respectively; Fig. S4 E and Fig. 2 D). This suggests that the Kv2 voltage sensors in the unclamped cells for concentration– effect experiments are in the same early resting conformation as voltage-clamped cells at −80 mV. Additionally, we determined that only a small fraction of the up to ninefold variability in kΔF at positive voltages could be attributed to temperature fluctuations (Fig. S5). Possible reasons for cell-to-cell variability at more positive voltages are discussed further in Limitations. The relation between voltage sensor activation and GxTX-594 dynamics can be recapitulated by rate theory modeling To enable translation of the intensity of fluorescence from GxTX-594 on a cell surface into a measure of Kv2 conformational change, we developed an EVAP model, a series of equations derived from rate theory that relate cell labeling to voltage sensor activation. The framework of the EVAP model is gener- alizable to fluorescent molecular probes that report conforma- tional changes by a change in binding affinity. In the EVAP model, the proportion of labeled versus unlabeled Kv2 in a membrane is determined by the proportion of voltage sensors in resting versus activated conformations. The model assumes that the innate voltage sensitivity of the Kv2 subunit is solely Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 12 of 24 responsible for controlling voltage dependence. EVAP labeling is voltage dependent because the binding and unbinding rates are different for resting and activated conformations of voltage sensors. Voltage activation of Kv2 channels involves many conformational changes (Scholle et al., 2004; Jara-Oseguera et al., 2011; Tilley et al., 2019). However, models that presume independent activation of a voltage sensor in each of the four Kv2.1 subunits accurately predict many aspects of voltage acti- vation and voltage sensor toxin binding (Lee et al., 2003; Tilley et al., 2019). For simplicity, we model Kv2 proteins as having only resting and activated conformations that are independent in each voltage sensor and developed a rate theory model con- sisting of four interconnected states (Scheme 1). When voltage sensors change from resting to activated con- formations, the binding rate of the GxTX-594 EVAP decreases, and the unbinding rate increases. When the membrane voltage is held constant for sufficient time, the proportions of labeled and unlabeled proteins reach an equilibrium. EVAP labeling requires seconds to equilibrate (Fig. 6), whereas Kv2 channel gating equilibrates in milliseconds (Tilley et al., 2019), three orders of magnitude more quickly. These distinct time scales of equilibration suggest an approximation to model the reversible EVAP labeling response: voltage sensor conformations achieve equilibrium quickly such that only their distribution at equi- librium is expected to greatly impact the kinetics of labeling and unlabeling, allowing Scheme 1 to collapse into Scheme 2, which depicts the structure of the EVAP model used for calculations. We constrained the EVAP model with measurements of GxTX-594 binding kinetics and GxTX impacts on Kv2.1 gating 9 nM GxTX-594. Color progression for pseudocoloring of fluorescence in- tensity is shown in vertical bar on right. Middle column in each row indicates voltage step taken from a holding potential of −80 mV. Times listed at top of each column correspond to time axis in panel B. Scale bar, 10 μm. (B) GxTX- 594 fluorescence during steps to indicated voltages. Smooth lines are mono- exponential fits (Eq. 1): −40 mV kΔF = 2.15 × 10−2 ± 0.22 × 10−2 s−1; 0 mV kΔF = 1.279 × 10−1 ± 0.023 × 10−1 s−1; 40 mV kΔF = 2.492 × 10−1 ± 0.062 × 10−1 s−1; and 80 mV kΔF = 4.20 × 10−1 ± 0.11 × 10−1 s−1. ROIs were hand-drawn around the apparent cell surface membrane based on GxTX-594 fluorescence. 0% was set by subtraction of background, which was the average intensity of a region that did not contain cells over the time course of the voltage proto- col. For each trace, 100% was set from the initial fluorescence intensity at −80 mV before the subsequent voltage step. Raw initial fluorescence values before normalization were within 10% of one another. (C) Fluorescence intensity remaining at the end of 50-s steps to +80 mV. Each circle repre- sents one cell. Background subtraction as in B. (D) Voltage dependence of fluorescence intensity at the end of 50-s steps. For each cell, 100% was set from the initial fluorescence intensity at −80 mV before the first step to another voltage. Cells did not always recover to initial fluorescence intensity during the −80-mV holding period between voltage steps. Top: Circle col- oring indicates data from the same cell, and lines connect points from the same cell. Gray circles represent data shown in B. Bottom: Black bars rep- resent the mean F/Finit at each voltage, and error bars represent the SEM. Black line is the fit of a first-order Boltzmann equation (Eq. 2): V1/2 = −27.4 ± 2.5 mV, z = 1.38 ± 0.13 e0. Green line is the prediction from the EVAP model at 9 nM GxTX. (E) Voltage dependence of fluorescence intensity kinetics (kΔF). Top: Circle coloring is the same as D. Bottom: Black bars represent the average kΔF at each voltage, and error bars represent the SEM. Black line is a first-order Boltzmann equation fit to the kΔF–voltage relation: V1/2 = +38 ± 15 mV, z = 1.43 ± 0.35 e0. Green line is the prediction from the EVAP model at 9 nM GxTX. Figure 6. GxTX-594 labeling responds to transmembrane voltage. (A) Fluorescence from an optical section of a voltage-clamped Kv2.1-CHO cell in Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 13 of 24 (see EVAP model and Data S1). We tested the viability of model predictions by comparison with GxTX-594 labeling measure- ments. The EVAP model predicts that the FV for GxTX-594 la- beling will conform to a Boltzmann distribution. The FV prediction in 9 nM GxTX-594 had a V1/2 and z that differ by only −4 mV and 0.06 e0, respectively, from the Boltzmann fit of ex- perimental data (Fig. 6 D, bottom panel, black and green lines). The EVAP model also predicts that the kΔF-V will conform to a Boltzmann distribution. The kΔF-V prediction differs by only 3 mV and 0.05 e0 from the Boltzmann fit of experimental data (Fig. 6 E, bottom panel, blue line). However, this fit was poorly constrained as we failed to obtain sufficient data at voltages above +80 mV where kΔF-V is predicted to saturate. In our at- tempts, the durations required at more positive voltages irre- versibly increased membrane leak. The V1/2 and z values from the GxTX-594 FV and kΔF-V were not used as constraints of the EVAP model, and the similarity between the predictions and empirical findings seemed remarkable enough to warrant fur- ther exploration of the EVAP model predictions. We used the EVAP model to investigate general principles of the relation between voltage sensor activation and reversible labeling. The EVAP model predicts that as GxTX-594 concen- tration decreases, the change in labeling (ΔF/ΔFmax) approaches the probability that unlabeled voltage sensors are resting (Fig. 7 A). This prediction explains the similarity between the FV in 9 nM GxTX-594 (V1/2 = −27 ± 3 mV, z = 1.4 ± 0.1 e0; Fig. 6 D) and the QV of Kv2.1 (V1/2 = −26 ± 1 mV, z = 1.6 ± 0.1 e0; Tilley et al., 2019). As the concentration of EVAP is increased, the FV shifts to more positive voltages such that the fractional change in fluorescence intensity is always less than the fraction of unlabeled voltage sensors that are active. As EVAP concentration increases and approaches the activated state Kd (1,790 nM), voltage sensor activation becomes less effective at dissociating the EVAP due to binding to activated voltage sensors (Fig. 7 B). The model pre- dicts that at any concentration, this simple interpretation will be valid: A decrease in EVAP surface fluorescence indicates acti- vation of unlabeled voltage sensors. The EVAP model also yields a simple interpretation of la- beling kinetics, it predicts that as GxTX-594 concentration de- creases, the rate of fluorescence change kΔF approaches the probability that labeled channels are active (Fig. 7 C). This pre- diction explains the similarity between the kΔF-V in 9 nM GxTX- 594 (V1/2 = 38 ± 15 mV, z = 1.4 ± 0.4 e0; Fig. 6 D) and the QV of Kv2.1 in saturating GxTX (V1/2 = 47 ± 1 mV, z = 1.6 ± 0.1 e0; Tilley et al., 2019). At low concentrations, the dependence of kΔF on the conformation of channels bound to GxTX-594 is due to the rate of unbinding dominating kΔF, with the rate of unbinding being solely determined by the conformation of channels bound to GxTX-594. Repetitive action potential–like stimuli amplify the GxTX-594 response Kv2 currents impact repetitive action potential firing (Du et al., 2000; Liu and Bean, 2014), making repetitive action potentials the voltage waveforms that are, arguably, most relevant to Kv2 function. However, action potentials occur on the millisecond time scale, orders of magnitude faster than the GxTX-594 response, which integrates Kv2 conformations occurring over many seconds. Kv2 channels have slow deactivation kinetics (Liu and Bean, 2014; Tilley et al., 2019), and high-frequency firing could prevent Kv2 proteins from fully deactivating be- fore a next action potential is triggered, creating a kinetic trap that progressively accumulates activated voltage sensors. This behavior of Kv2 proteins indicates that high-frequency firing could evoke a more robust fluorescence signal than the EVAP model predicts, as the model assumes continuous equilibrium of voltage sensors and cannot kinetically trap activated con- formations. To test this hypothesis, we crudely mimicked action potentials with trains of 2-ms voltage steps from −80 to +40 mV and observed the changes in fluorescence on GxTX-594–labeled Kv2.1-CHO cells (Fig. 8, A and B). To assess frequency response, step frequency was varied from 0.02 to 200 Hz. GxTX-594 un- labeling and kΔF increased with stimulus frequency (Fig. 8, C and D). We compared these fluorescence responses to those pre- dicted by the EVAP model. When the stimulus frequency was <50 Hz, the EVAP model did a reasonable job of predicting fluorescence change. At and >50 Hz, fluorescence decreased by more than the EVAP model predicted, even without accounting for a voltage-insensitive fraction of Kv2.1 proteins (Fig. 8 C, bottom panel, green line). As discussed above, this divergence from the equilibrium-based EVAP model is expected at frequencies where voltage steps are shorter than Kv2 equilibration times. The time constant of ac- tivating gating current decay from Kv2.1-CHO cells was 1.3 ms at +40 mV (Tilley et al., 2019), which means that the majority of voltage sensors are effectively activated during the 2-ms +40 mV steps. In contrast, the time constant of deactivating gating cur- rent decay from Kv2.1-CHO cells was 22 ms at −80 mV (Tilley et al., 2019), which means that Kv2.1 is expected to become ki- netically trapped in activated conformations when stimuli to +40 mV from −80 mV are ∼50 Hz or faster. Thus, the amplified EVAP response appears consistent with voltage sensors failing to deactivate before the next stimulus, leading to an accumulation of activated voltage sensors and a more dramatic fluorescence response than predicted by the EVAP model. Overall, these dy- namics indicate that the magnitude of the change in GxTX-594 fluorescence intensity will be amplified during repetitive action potentials, a regimen of electrophysiological signaling where Kv2 currents are critical. In contrast, the kinetics of the GxTX-594 response did not appear to deviate from EVAP model predictions at high fre- quencies (Fig. 8 D). As kΔF responds to the dynamics of Kv2 proteins bound by GxTX-594, it could be that the faster deacti- vation rate of voltage sensors bound by GxTX (Tilley et al., 2019) prevents the bound channels from being kinetically trapped. GxTX-594 labels brain slices transfected with Kv2.1-GFP To determine whether expression of Kv2 proteins embedded in tissue can be imaged with GxTX-594, we overexpressed Kv2.1- GFP in rat brain slices and examined CA1 pyramidal neurons of the hippocampus. We chose CA1 neurons for several reasons: They express Kv2 channels at a density typical of central neurons, the physiology of these neurons has been intensively studied, and their electrical properties are relatively homogeneous Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 14 of 24 Figure 7. Relationship of GxTX-594 labeling to probability of Kv2 voltage sensor activa- tion. (A) EVAP model predictions of concentra- tion and voltage dependence of cell surface fluorescence intensity at different concentrations of EVAP in solution. The bottom axis represents membrane voltage. The left axis represents the predicted fluorescence relative to when all voltage sensors are at rest (Finit) and does not include EVAP signal that is insensitive to voltage. The dashed line corresponds to the right axis and represents the probability that voltage sensors of unlabeled Kv2.1 are in their resting confor- mation. (B) EVAP model prediction of the cell surface fluorescence when cells are given a +40- mV depolarization relative to when all voltage sensors are at rest (Finit) at increasing concen- trations of EVAP. The orange dashed line repre- sents Kd of resting voltage sensors. The blue dashed line represents Kd of activated voltage sensors. (C) EVAP model predictions of con- centration and voltage dependence of kΔF. Col- ors correspond to A, except the dashed line is the probability that voltage sensors bound by GxTX-594 are an activated conformation. (Misonou et al., 2005). Organotypic hippocampal slice cultures prepared from postnatal day 5–7 rats were sparsely transfected with Kv2.1-GFP, resulting in a subset of neurons displaying green fluorescence. When imaged 2–4 d after transfection, GFP fluorescence was observed in the plasma membrane sur- rounding neuronal cell bodies and proximal dendrites (Fig. S6, A and B). Six days or more after transfection, Kv2.1-GFP fluo- rescence organized into clusters on the surface of the cell soma and proximal processes (Fig. 9 A and Fig. S6 C), a pattern consistent with a prior report of endogenous Kv2.1 in CA1 neurons (Misonou et al., 2005). After identifying a neuron expressing Kv2.1-GFP, solution flow into the imaging chamber was stopped, and GxTX-594 was added to the static bath solu- tion to a final concentration of 100 nM. After 5 min of incu- bation, solution flow was restarted, leading to washout of excess GxTX-594 from the imaging chamber. After washout, GxTX-594 fluorescence remained colocalized with Kv2.1-GFP (Fig. 9 and Fig. S6), indicating that GxTX-594 is able to per- meate through dense neural tissue and bind to Kv2 proteins on neuronal surfaces. Pearson correlation coefficients confirmed the colocalization of GxTX-594 with Kv2.1-GFP in multiple slices (Fig. 9 C). In most images of Kv2.1-GFP–expressing neu- rons, GxTX-594 also labeled puncta on neighboring neurons that did not express Kv2.1-GFP but at intensities that were roughly an order of magnitude dimmer (Fig. 9 B, white arrow). The clustered GxTX-594 fluorescence patterns on the cell body and proximal processes of CA1 neurons were strikingly similar to reported patterns of anti-Kv2 immunofluorescence patterns and are consistent with GxTX-594 labeling of endogenous Kv2 proteins in CA1 neurons. While we cannot exclude the possibility that CA1 neurons have a subset of Kv2 proteins on their surface that is not labeled by GxTX-594, we saw no indication of Kv2.1- GFP on neuronal surfaces that are not labeled by GxTX-594. While we observed GxTX-594 fluorescence that morpholog- ically resembles endogenous Kv2 protein localizations, we also found that GxTX-594 occasionally labels structures not consistent with Kv2 proteins (Fig. S6, bottom panel, arrows). This non-Kv2 labeling was most prevalent at the surface of the hippocampal slices and progressively decreased as the imaging plane was moved deeper into the tissue (data not shown). Our interpretation of this phenomenon is that GxTX-594 can accumulate in the dead tissue and debris that is present at the surface of a hippocampal section after it is cut. For this reason, we analyzed only GxTX-594 fluorescence with subcellular localizations consistent with Kv2 channels. GxTX-594 labeling in brain slices responds to neuronal depolarization To test whether reversible GxTX-594 labeling of neurons in brain slices is consistent with binding to endogenous Kv2 volt- age sensors, we determined whether GxTX-594 labeling re- sponds to voltage changes in tissue. First, we looked for Kv2-like patterns on CA1 pyramidal neurons in untransfected brain slices bathed in 100 nM GxTX-594. With two-photon excitation, Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 15 of 24 Figure 8. High-frequency repetitive stimuli amplify the GxTX- 594 response. (A) Fluorescence intensity from Kv2.1-CHO cells incubated in 9 nM GxTX-594. Arrow indicates voltage-clamped cell. Fluorescence at holding potential of −80 mV (left), after 50 s of 200-Hz stimulus (middle), and 100 s after the cell is returned to holding potential of −80 mV (right). Stimulus was a 2-ms step to +40 mV. Note that in each panel, the unclamped cell (left cell in each panel) does not show a change in fluorescence. Scale bar, 10 μm. (B) Representative trace with GxTX-594 unlabeling at 200 Hz. Red line, monoexponential fit (Eq. 1): kΔF = 0.2327 ± 0.0099 × 10−1 s−1. 0% F/Finit was set by subtraction of the average intensity of a region that did not contain cells. 100% was set by the initial fluo- rescence intensity at −80 mV. (C) Fluorescence–stimulus frequency relation. Points indicate F/Finit from individual cells. Top: Point coloring indicates data from the same cell. Bottom: Black bars represent the average F/Finit at each voltage, and error bars rep- resent the SEM. Green line is the prediction of the EVAP model at a concentration of 9 nM. (D) kΔF–stimulus frequency relation. Plotted as in C. optical sections are thinner than the neuronal cell bodies, and GxTX-594 fluorescence appeared as puncta circumscribed by dark intracellular spaces (Fig. 10 A). This was similar to the patterns of fluorescence in Kv2.1-GFP–transfected slices (Fig. 9 B) and consistent with the punctate expression pattern of Kv2.1 in CA1 pyramidal neurons seen in fixed brain slices (Misonou et al., 2005). We tested whether the punctate fluorescence was voltage sensitive using voltage clamp. To ensure voltage clamp of the neuronal cell body, slices were bathed in tetrodotoxin to block Na+ channels, and Cs+ was included in the patch pipette solution to block K+ currents. In each experiment, whole-cell configu- ration was achieved with a GxTX-594–labeled neuron, and holding potential was set to −70 mV. At this point, time-lapse imaging of a two-photon excitation optical section was initiated. Depolarization to 0 mV resulted in loss of fluorescence from a subset of puncta at the perimeter of the voltage-clamped neuron (Fig. 10 B, red arrows; and Video 1). Other fluorescent puncta appeared unaltered by the 0-mV step (Fig. 10 B, white arrows). These puncta could represent Kv2 proteins on a neighboring cell, off-target labeling by GxTX-594, or voltage-insensitive Kv2 proteins, possibly due to near-surface internalization. To assess whether the fluorescence changes were due to the voltage change, we compared fluorescence of the apparent cell mem- brane region with regions more distal from the voltage-clamped cell body. An ROI 30-pixels (1.2-μm) wide containing the ap- parent membrane of the cell body was compared with other regions within each image (Fig. 10 C). The region containing the membrane of the cell body lost fluorescence during the 0-mV step (Fig. 10 D, ROI 1), while neither of the regions more distal Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 16 of 24 Figure 9. GxTX-594 labels CA1 hippocampal pyramidal neurons transfected with Kv2.1-GFP. (A) Two-photon excitation images of fluorescence from the soma and proximal dendrites of a rat CA1 hippocampal pyramidal neuron in a brain slice 6 d after transfection with Kv2.1-GFP (left), labeled with 100 nM GxTX- 594 (middle), and the overlay (right). Image represents a Z-projection of 20 optical sections. Scale bar, 10 μm. (B) A single optical section of the two-photon excitation image shown in A. GxTX-594 labels both Kv2.1-GFP puncta from a transfected cell and apparent endogenous Kv2 proteins from an untransfected cell in the same cultured slice (right, arrow). Scale bar, 10 μm. (C) Pearson correlation coefficients from CA1 hippocampal neurons 2, 4, and 6 d after transfection with Kv2.1-GFP. Each circle represents a different neuron. Bars are arithmetic means. (ROI 2, ROI 3) or the intracellular region (ROI 0) showed a similar change. In three hippocampal slices from separate rats, we found a significant decrease in fluorescence in ROI 1 com- pared with ROI 2 and ROI 3 when neurons were given a voltage step to 0 mV (Fig. 10 E; ANOVA P < 0.001; Tukey’s post hoc test P < 0.001). The kinetics of fluorescence response of the voltage- clamped membrane region were similar between slices (Fig. 10, F and K). To determine whether the fluorescence response to depo- larization was driven by the Kv2-like puncta on the cell mem- brane, the fluorescence along a path containing the apparent cell membrane was selected by drawing a path connecting fluores- cent puncta surrounding the dark cell body region and averaging fluorescence in a region 10-pixels (0.4-μm) wide centered on this path (Fig. 10 G, yellow line). The fluorescence intensity along the path of the ROI revealed distinct peaks corresponding to puncta (Fig. 10 H, red trace). After stepping the neuron to 0 mV, the intensity of fluorescence of a subset of puncta lessened (Fig. 10 H, black trace, peaks 1, 3, 4, 5, and 8). While the data are noisy, it is clear that even more reduction is observed in indi- vidual fluorescence puncta in brain slice (Fig. 10 I, blue line). This suggests that the voltage sensors in these functionally se- lected puncta are extensively activated. The EVAP model pre- dicts only a 55% reduction of fluorescence from CHO cells stepped to 0 mV in 100 nM GxTX. The greater response of the endogenous puncta could be due to voltage sensors activating at more negative voltages in neurons than CHO cells. However, the GxTX-594 concentration within tissue may be more dilute than in the bath solution, and consequently, the sensitive fraction calculated from the EVAP model is a lower bound. Other puncta maintained or increased their brightness after depolarization (Fig. 10 H, black trace, peaks 2, 6, and 7), but it was unclear whether these voltage-insensitive puncta correspond to Kv2 proteins on the surface of the same voltage-clamped neuron. When the kinetics of fluorescence intensity decay of individual voltage-sensitive puncta were fit with Eq. 1, kΔF values were similar between puncta (Fig. 10 I), consistent with these spa- tially separated puncta all being on the surface of the voltage- clamped neuron. The kΔF changes measured in these puncta were similar to those predicted by the Fig. 7 model (Fig. 10 I, blue line), although the variability of these measurements was sub- stantial. To address whether the fluorescence change at the cell membrane was driven by decreases in regions of punctate fluo- rescence, the punctate and nonpunctate fluorescence intensity changes were analyzed separately. The regions with fluores- cence intensities above average for the path (Fig. 10 H, dotted line) were binned as one group. The above-average group, by definition, contained all punctate fluorescence. When compar- ing the fluorescence before and during the 0-mV step, the re- gions that were initially of below-average fluorescence Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 17 of 24 Figure 10. GxTX-594 puncta on hippocampal CA1 neurons are sensitive to voltage stimulus. (A) Two-photon excitation optical section from CA1 py- ramidal neurons in a cultured hippocampal slice after incubation with 100 nM GxTX-594. Scale bar, 5 μm. (B) 594 fluorescence from a single two-photon excitation optical section before and during depolarization of a whole-cell patch-clamped neuron. Text labels of holding potential (−70 mV or 0 mV) indicate approximate position of the patch-clamp pipette. Red arrows indicate stimulus-sensitive puncta; white arrows indicate stimulus-insensitive puncta. Left panel is the average fluorescence of the three frames at −70 mV before depolarization, while the right panel is the average fluorescence of three frames after holding potential was stepped to 0 mV. Scale bar, 5 μm. Video 1 contains time-lapse images from this experiment. (C) ROIs used in analysis for D–F. Same slice as B. ROI 1 contains the apparent plasma membrane of the cell body of the patch-clamped neuron; it was generated by drawing a path tracing the apparent plasma membrane and then expanding to an ROI containing 15 pixels on either side of the path (1.2 μm total width). ROI 2 contains the area 15–45 pixels outside the membrane path (1.2 μm total width). RO1 3 contains the area >45 pixels outside the membrane path. ROI 0 contains the area >15 pixels inside the membrane path. Scale bar, 5 μm. (D) Fluorescence from each ROI shown in C. Squares represent ROI 0, circles represent ROI 1, up triangles represent ROI 2, and down triangles represent ROI 3. Background was defined as the mean fluorescence of ROI 0 during the experiment. Finit was defined as the mean fluorescence of ROI 1 during the first six frames, after subtraction of background. Dotted lines represent the average fluorescence of the first six frames of each ROI. The voltage protocol is shown above the graph. (E) Change in fluorescence during a 0-mV step for different ROIs in three hippocampal slices from three separate rats. ROIs Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 18 of 24 for each slice were defined by methods described in C. Circles represent mean fluorescence from six frames during 0-mV stimulus (F), normalized to mean fluorescence from the same ROI in six frames before stimulus (Finit). Circle color is consistent between ROIs for each hippocampal slice; red circles are from the slice shown in D. Black bars are arithmetic mean ± SE from three hippocampal slices. A statistical difference between ROIs was detected by ANOVA (P = 0.0003) followed by the Tukey’s post hoc test, where ROI 1 versus ROI 2 (***, P < 0.001), ROI 1 versus ROI 3 (***, P < 0.001), and ROI 2 versus ROI 3 (n.s., P = 0.98). (F) Kinetics of fluorescence change from ROI 1 during a 0-mV step in three hippocampal slices. ROI 1 for each slice was defined by the methods described in C. Lines are monoexponential fits (Eq. 1): kΔF = 6.8 × 10−2 ± 2.6 × 10−2 s−1 (yellow), 6.8 × 10−2 ± 4.6 × 10−2 s−1 (red), and 4.9 × 10−2 ± 2.10−2 s−1 (green). The voltage protocol is shown above the graph. Colors of individual circles indicate the same slices as E. (G) ROI used in analysis for H–J. Same image as C. The shaded ROI was generated by drawing a path tracing the apparent plasma membrane and then expanding to an ROI containing 5 pixels on either side of the path (0.4 μm total width). Numbers indicate puncta that appear as peaks in H. Scale bar, 5 μm. (H) A plot of the fluorescence intensity along the ROI shown in G before (red trace) and during (black trace) 0-mV stimulus. Numbers above peaks correspond to puncta labeled in G. Red trace: Mean fluorescence intensity during the three frames immediately before the stimulus, normalized to mean intensity of entire ROI (black dotted line), plotted against distance along path. Black trace: Mean fluorescence intensity during three frames at 0 mV, normalized by the same Finit value as the red trace. (I) Kinetics of fluorescence change of individual puncta from G. Puncta intensity are average intensity of points extending to half maximum of each peak in H. Asterisks indicate mean fluorescence intensity of puncta 1 (pink), 4 (red), and 5 (dark red). Lines are monoexponential fits (Eq. 1): kΔF = 7.10−2 ± 2.9 × 10−2 s−1 (pink), 6.7 × 10−2 ± 2.5 × 10−2 s−1 (red), and 1.2 × 10−1 ± 5.6 × 10−2 s−1 (dark red). Fits to other puncta had SDs larger than kΔF values and were excluded. Finit was defined as the mean background- subtracted fluorescence of the puncta during the six frames before stimuli. The voltage protocol is displayed above the graph. Blue line is prediction from Scheme 1. (J) Comparison of fluorescence change of puncta (above average) and interpuncta (below average) regions in response to 0-mV stimulus. The regions before stimulus shown by the red line of H that had F/Finit ≥1 (above average) were binned separately from regions with F/Finit <1. The mean fluorescence of each region during 0-mV stimulus (H, black trace) was compared with the fluorescence before stimulus (H, red trace). Circles indicate values from three independent hippocampal slices; colors indicate same slices as E and F. A weak statistical difference in fluorescence was detected between interpuncta and puncta regions by Student’s t test (*, P = 0.046). Black bars are arithmetic mean ± SE. (K) Rate of fluorescence change of GxTX-594 after a 0-mV stimulus from puncta (as in I), ROI 1 (as in F), or Kv2.1-CHO cells in 100 nM GxTX-594. Kv2.1-CHO cells were imaged at the same temperature as neurons (30°C) using neuronal intracellular solution. CHO CE solution was used for Kv2.1-CHO cell experiments. Kv2.1-CHO measurements were made by Airy disk confocal imaging. Black bars are mean ± SEM from data shown. Blue dotted line is kΔF = 6.31 × 10−2 s−1 prediction of Scheme 1. No statistical difference was detected between groups by ANOVA (P = 0.11). maintained the same intensity (103 ± 8%); regions that were initially of above-average fluorescence decreased in intensity (70 ± 8%; Fig. 10 J). This suggests that the detectable unlabeling along the cell membrane originated in the Kv2-like puncta, not the regions of lower fluorescence intensity between them. To determine whether the dynamics of the voltage-dependent responses of GxTX-594 fluorescence on neurons in brain slices were consistent with reversible labeling of Kv2.1, we performed experiments with Kv2.1-expressing CHO cells under similar conditions as brain slice: 100 nM GxTX-594, 30°C, Cs+-containing patch pipette solution. The rate of fluorescence changes in Kv2.1- CHO cells was similar to neurons in brain slices and consistent with the kΔF of 0.06 s−1 predicted by the Fig. 7 model (Fig. 10 K). However, the data underlying of kΔF measurements were noisy, limiting our ability to detect differences. Discussion The molecular targeting, conformation selectivity, and spatial precision of fluorescence from GxTX-594 enable identification of where in tissue the conformational status of Kv2 voltage sensors becomes altered. However, the utility of GxTX-594 as an EVAP is limited by several factors, including emission intensity, variability between experiments, and inhibition of Kv2 proteins. We discuss the potential utility and limitations of the EVAP mechanism underlying GxTX-594. Unique capabilities of GxTX-594 The Kv2 EVAP presented here is the only imaging method we are aware of for measuring voltage-sensitive conformational changes of a specific, endogenous protein. As GxTX binding selectively stabilizes the fully resting conformation of Kv2.1 voltage sensors (Tilley et al., 2019), reversible GxTX-594 labeling is expected to bind with highest affinity specifically to the fully resting conformation of the Kv2 voltage sensor in which the first gating charge of the Kv2 S4 segment is in the gating charge transfer center (Tao et al., 2010). Images of GxTX-594 fluores- cence reveal this conformation’s occurrence with subcellular spatial resolution. Importantly, the EVAP model we developed allows deconvolution of the behavior of unlabeled Kv2 proteins. This enables the subcellular locations where Kv2 voltage sensing occurs to be seen for the first time. Electrophysiological approaches can detect the voltage- sensitive K+ conductance of Kv2 channels. However, the ma- jority of Kv2 proteins on cell surface membranes do not function as channels and are nonconducting (Benndorf et al., 1994; Malin and Nerbonne, 2002; O’Connell et al., 2010), and Kv2 proteins dynamically regulate cellular physiology by nonconducting functions (Antonucci et al., 2001; Singer-Lahat et al., 2007; Feinshreiber et al., 2010; Dai et al., 2012; Fox et al., 2015; Johnson et al., 2018; Kirmiz et al., 2018a, 2018b, 2019; Vierra et al., 2019). The Kv2 EVAP reports on the conformation of Kv2 voltage sensors independently from ion conductance, enabling the study of voltage sensor involvement in Kv2’s nonconducting physio- logical functions. We present this EVAP as a prototype molecular probe for imaging voltage sensing by endogenous proteins in tissue with molecular specificity. Here, during our initial testing of GxTX-594, we observed that the majority of Kv2 protein detected at discrete individual clusters was voltage sensitive. While it may not be surprising to find that voltage-gated ion channel proteins are voltage sensi- tive, the voltage sensors of surface-expressed proteins can be immobilized. For example, gating charge of the L-type Ca2+ channel Cav1.2 is immobilized until it is bound by an intracel- lular protein (Turner et al., 2020). Kv2.1 channel function is extensively regulated by neurons. In rat CA1 neurons, the clustered Kv2 channels are proposed to be nonconducting (Fox et al., 2013b). Our results show that clustered Kv2 proteins in rat Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 19 of 24 CA1 neurons remain voltage sensitive. Interestingly, when Kv2.1 is expressed in CHO cells, a fraction of the GxTX-594 fluorescence is voltage insensitive (Fig. 6 C). This observation is consistent with voltage sensor immobilization of some surface-expressed Kv2.1 protein, although it could be due to intracellular Kv2.1–GxTX-594 proteins that appear to be at the cell surface. We used images of GxTX-594 fluorescence to measure the coupling between endogenous Kv2 proteins and membrane potential at specific, subcellular anatomical locations. Similarly, GxTX-594 imaging should detect changes of voltage sensor status when Kv2 proteins become engaged or disengaged from nonconducting functions, such as formation of plasma membrane–endoplasmic reticulum junctions (Antonucci et al., 2001; Fox et al., 2015; Kirmiz et al., 2018a), regulation of exo- cytosis (Singer-Lahat et al., 2007; Feinshreiber et al., 2010), regulation of insulin secretion (Dai et al., 2012), interaction with kinases, phosphatases, and SUMOylases (Misonou et al., 2004; Park et al., 2006; Dai et al., 2009; Cerda and Trimmer, 2011; McCord and Aizenman, 2013), formation of specialized subcellular calcium signaling domains (Vierra et al., 2019), and interactions with astrocytic end feet (Du et al., 1998). GxTX-594 could potentially reveal conformational changes in organs throughout the body where Kv2 proteins are expressed, which include muscle, thymus, spleen, kidney, adrenal gland, pan- creas, lung, and reproductive organs (Bocksteins, 2016). Limitations of GxTX-594 There are important limitations to the GxTX-594 approach and of the underlying EVAP mechanism generally. We discuss sev- eral limitations that are worth considering in the design of any studies with GxTX-594. GxTX-594 labeling is slower than channel gating The kinetics of reversible GxTX-594 labeling are limited to measuring changes in Kv2 activity on a time scale of tens of seconds. While the temporal resolution of GxTX-594 is com- patible with live imaging and electrophysiology experiments, labeling kinetics do not provide sufficient time resolution to distinguish fast electrical signaling events. The response time of GxTX-594 is far slower than the kinetics of Kv2 conformational change, limiting measurements to the probability, averaged over time, that voltage sensors are resting or active. It is worth noting that the probability of a conformation’s occurrence can be a valuable measure and is the ultimate quantitation of many biophysical studies of ion channels (e.g., open probability, steady-state conductance, and gating charge–voltage relation). GxTX-594 dynamics are altered in confined extracellular spaces The kinetics of GxTX-594 dynamics varied within different re- gions of the same CHO cell (Fig. S4). The location dependence of kΔF was more pronounced during GxTX-594 labeling at −80 mV than unlabeling at +40 mV, and such a difference can be ex- plained by the distinct voltage-dependent affinities of Kv2.1 for GxTX-594. We suspect that the more extreme location depen- dence at −80 mV is due to a high density of Kv2.1 binding sites in the restricted extracellular space between the cell mem- brane and glass coverslip, such that GxTX-594 is depleted from solution by binding Kv2.1 before reaching the center of the cell. After unbinding at +40 mV, Kv2.1 proteins are in ac- tivated, low-affinity conformations, which are unlikely to re- bind GxTX-594 and, thus, do not slow their diffusion across the cell surface. The space extracellular to Kv2.1 channel clusters of hippocampal and cortical interneurons is restricted by as- trocytic end feet, which create an extracellular cleft only a few nanometers wide (Du et al., 1998). Such restricted spaces may slow the kinetics of labeling in the hippocampal slices (Fig. 10). GxTX-594 dynamics are variable between CHO cells The variability of GxTX-594 response rates and amplitudes be- tween CHO cells limited the precision of results. Some of this variability is expected from technical imprecisions: Fits of kΔF where the final value of the relaxation was poorly determined by the data, small changes due to variations of room temperature, photobleaching, and other potential sources. However, we no- ticed that results were more consistent between stimuli of the same cell, and the variability was greatest between cells (Fig. 6, D and E; and Fig. 8, C and D). We suspect that cell-to-cell differ- ences in Kv2.1 conformational equilibria are responsible for much of the variability in GxTX-594 response. The Kv2.1 conductance–voltage relation is regulated by many cellular pathways, including kinases, phosphatases, and SUMOylases (Misonou et al., 2004; Park et al., 2006; Dai et al., 2009; Cerda and Trimmer, 2011; McCord and Aizenman, 2013). Large cell-to- cell variation in Kv2.1 conductance–voltage and gating charge– voltage relations have been reported in CHO cells by our group and others (McCrossan et al., 2009; Tilley et al., 2014; Kang et al., 2019; Tilley et al., 2019). In this study, when we predicted the voltage sensor V1/2 of Kv2.1 from electrophysiology, we observed a 6.4-mV SD with a range of 19 mV, and this variance appeared to be exacerbated by GxTX-594 having a 9.7-mV SD and range of 36 mV (Fig. 1 E). As EVAP dynamics and the GV are both determined by voltage sensor activation, variability in the GxTX- 594 response is expected. The hypothesis that cell-to-cell varia- tion in fluorescence dynamics is due to the inherent variability of Kv2.1 voltage sensor activation could be more definitively tested by identifying whether a correlation exists between the V1/2 of the QV and fluorescence–voltage relationship from individual cells labeled with GxTX-594. While we have not attempted this, the structure of the variance in GxTX-594 fluorescence–voltage relationships is informative. The fluorescence–voltage relation- ships compiled from many cells become more variable near the midpoint of relevant voltage sensor movements. The response amplitude F/Finit (Fig. 6 D) is determined by unlabeled voltage sensor activation and appears most variable near the V1/2 of the unlabeled QV relation (−32 mV; Tilley et al., 2019). The response kinetics kΔF are determined by activation of voltage sensors, which have GxTX-594 bound and appear to become increasingly variable at voltages higher than −20 mV (Fig. 6 E, bottom panel). Despite this variability, the V1/2 and z from the Boltzmann fit of –voltage relationship from many cells were remarkably the kΔF close to the QV of the GxTX–Kv2.1 complex, with a V1/2 and z that differ by 3.3 mV and 0.07 e0, respectively (Tilley et al., 2019). Another possibility is that variability in surface membrane composition undergirds the variability of the GxTX-594 response. Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 20 of 24 GxTX affinities are influenced by the dynamically changing com- plement of lipids in the plasma membrane lipid composition. Lipids bind to the voltage-sensing domain near the GxTX binding site (Milescu et al., 2009; Gupta et al., 2015). Sphingomyelinase D treatment, which alters the membrane composition by converting sphingomyelin to ceramide-1-phosphate, has been shown to en- hance GxTX affinity for Kv2.1 fourfold (Milescu et al., 2009). Voltage activation of Kv2.1 is also affected by sphingomyelinase treatments. While variation in lipid composition is expected to cause variation in GxTX-594 dynamics, we do not know the degree to which the lipid composition varies between the Kv2.1-CHO cells in our studies. Fluorescence intensity Optical noise limits interpretation of EVAP imaging. While the GxTX-594 signal from CA1 neurons was sufficient to identify voltage sensing of endogenous Kv2 protein, fluorescence signal- to-noise issues limit interpretation with the EVAP model. This signal-to-noise ratio is influenced by the density of EVAP binding sites, the fluorescence intensity from each binding site, characteristics of the imaging system, and background fluores- cence from unbound EVAP or other sources. As the concentra- tion of fluorophore is lowered, background fluorescence from unbound EVAP will decrease, and the percentage of labeled binding sites will decrease. However, ΔF after voltage sensor activation remains similar, as does kΔF (Fig. 7). Due to this dy- namic, the minimum concentration of EVAP that is detectable is expected to be determined by the background fluorescence and the density of EVAP binding sites. Here, we see labeling of Kv2.1- CHO cells with 1 nM GxTX-594 (Fig. 2), and previously, we found a robust fluorescence–voltage response in 1 nM GxTX-550 (Tilley et al., 2014). in each case, <10% of resting voltage sensors are bound by the EVAPs. In theory, fluorescence measurements from a single EVAP molecule immobilized by binding to a single- voltage sensor could be informative, as fluorescence from a single binding site is eliminated altogether after unbinding and diffusion away. If the EVAP imaging method does not have single-molecule resolution, the density of EVAP binding sites can limit the signal to noise. CA1 hippocampal neurons express Kv2 proteins at a density typical of central neurons (Misonou et al., 2004; Vacher et al., 2008; Speca et al., 2014), and we expect that GxTX-594 imaging will have similar signal-to-noise characteristics in most brain regions. Improved signal to noise would be expected from such cells that express higher densities of Kv2 proteins, such as neurons of the subiculum, or the inner segment of photo- receptors (Maletic-Savatic et al., 1995; Gayet-Primo et al., 2018). Kv2 proteins are also expressed by many other cell types throughout the body (Bocksteins, 2016), where GxTX-594–labeling techniques may reveal Kv2 activity, if protein densities are sufficient. GxTX-594 inhibits Kv2 proteins GxTX-based probes inhibit the Kv2 proteins they label by sta- bilizing the resting conformation of Kv2 voltage sensors. The Kv2.1–GxTX-594 complex does not open to conduct K+ ions in the physiological voltage range (Fig. 1). Thus, GxTX-594 depletes the population of Kv2 proteins responding normally to physio- logical stimuli, which could alter Kv2 signaling. The concentra- tion of an EVAP can be lowered such that only an inconsequential minority of proteins are bound, with the trade-off being dim- mer fluorescence. With a related probe, we explored the impact of decreasing concentration on fluorescence response of a GxTX-based EVAP and saw substantial fluorescence responses to voltage while inhibiting only ∼10% of Kv2.1 current (Tilley et al., 2014). Here, we demonstrate that lower concentration and physiological stimuli are not always required for scientif- ically meaningful implementation of an EVAP. Even when GxTX-594 inhibits most Kv2 proteins, the behavior of unla- beled Kv2 proteins can be calculated using the EVAP model we have developed. Of course, the electrical feedback within cells will be altered by such protocols. The EVAP model is oversimplified Another limitation of the analysis developed here is that the model of Kv2 voltage sensor conformational change is an over- simplification. The gating dynamics of Kv2 channels are more complex than our model (Islas and Sigworth, 1999; Scholle et al., 2004; Jara-Oseguera et al., 2011; Tilley et al., 2019). Under some conditions, the assumption of voltage sensor independence will limit the model’s predictive power. With a related tarantula toxin, Hanatoxin, the concentration dependence for inhibition of Kv2.1 charge movement was consistent with independent binding to each voltage sensor inhibiting ∼25% of the total gating charge movement (Lee et al., 2003). This result suggests that the simplifying assumption of voltage sensor inde- pendence is reasonable. Additionally, the model of GxTX- 594–reversible labeling developed here assumes that voltage sensors are in continuous equilibrium. These deviations from equilibrium likely explain the deviation of the model from the data in response to high-frequency voltage steps (Fig. 8, C and D). This model could be further tested with a Kv2 mutant, which shifts channel gating without interfering with GxTX-594 binding. Conformation-selective probes reveal conformational changes of endogenous proteins Measurements of dynamic reversible labeling by a conformation- selective probe such as GxTX-594 can enable deduction of how unlabeled proteins behave. This is perhaps counterintuitive be- cause GxTX inhibits voltage sensor movement of the Kv2 protein it binds, and thus, only bound proteins generate optical signals. This approach is analogous to calcium imaging experiments, which have been spectacularly informative about physiological calcium signaling (Yang and Yuste, 2017), despite the fact that no optical signals originate from the physiologically relevant free Ca2+ but only from Ca2+ that is chelated by a dye. In all such experi- ments, fluorescence from Ca2+-bound dyes is deconvolved using the statistical thermodynamics of Ca2+ binding to calculate free Ca2+ (Adams, 2010). Similarly, GxTX-based probes dynamically bind to unlabeled Kv2 proteins, and the binding rate depends on the probability that unlabeled voltage sensors are in a resting conformation (Fig. 7). Thus, the conformations of unlabeled Kv2 proteins influence the dynamics of labeling with GxTX-based Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 21 of 24 probes. Consequently, the dynamics of labeling reveal the conformations of unlabeled Kv2 proteins. Deployment of GxTX-594 to report conformational changes of endogenous proteins demonstrates that conformation-selective ligands can be used to image occurrence of the conformations they bind to. The same principles of action apply to any conformation-selective labeling reagent, suggesting that probes for conformational changes of many different proteins could be developed. Probes could conceivably be developed from the many other voltage sensor toxins or other gating modifiers that act by a similar mechanism as GxTX yet target the voltage sensors of different ion channel proteins (McDonough et al., 1997; Sack et al., 2004; Catterall et al., 2007; Swartz, 2007; Schmalhofer et al., 2008; Peretz et al., 2010; McCormack et al., 2013; Ahuja et al., 2015; Dockendorff et al., 2018; Zhang et al., 2018). Conformation-selective binders have been engineered for a variety of other proteins, and methods to quantify con- formational changes from their fluorescence are needed. For example, fluorescently labeled conformation-selective binders have revealed that endocytosed GPCRs continue to remain in a physiologically activated conformation (Irannejad et al., 2013; Tsvetanova et al., 2015; Eichel and von Zastrow, 2018). A means to determine the conformational equilibria of GPCRs from fluorescence images has not yet been developed. We suggest that the statistical thermodynamic framework developed here could provide a starting point for more quantitative interpre- tation of other conformation-selective molecular probes. Acknowledgments Christopher J. Lingle served as editor. We thank Jim Trimmer (University of California, Davis) for numerous discussions and constructive critical reading of an early version of the manuscript. We thank Georgeann Sack (Afferent LLC) for critical reading, editing, and feedback. This research was supported by National Institutes of Health grants R01NS096317 (to J.T. Sack and B.E. Cohen), U01NS090581 (to J.T. Sack), R21EY026449 (to J.T. Sack), R01NS062736 (to K. Zito), U01NS103571 (to K. Zito), T32GM007377 (to R.J. Sepela), University of California, Davis New Research Initiative award (to J.T. Sack) and American Heart Association grant 17POST33670698 (to P. Thapa). GxTX variants were synthesized at the Molecular Foundry, supported by the Director, Office of Science, Office of Basic Energy Sciences, Division of Materials Sciences and Engineering, U.S. Department of Energy under contract DE-AC02-05CH11231. The authors declare no competing financial interests. Author contributions: P. Thapa: Conceptualization, formal analysis, investigation, methodology, visualization, writing- original draft, writing-reviewing and editing. R. Stewart: Conceptualization, formal analysis, investigation, methodology, visualization, writing-original draft, writing-reviewing and editing. R.J. Sepela: Conceptualization, formal analysis, inves- tigation, methodology, visualization, writing-original draft, writing-reviewing and editing. O. Vivas: Investigation, meth- odology, writing-reviewing and editing. L.K. Parajuli: Investi- gation, methodology, writing-reviewing and editing. M. Lillya: Conceptualization, formal analysis, investigation, methodology, visualization, writing-reviewing and editing. S. Fletcher-Taylor: Investigation, methodology, writing-reviewing and editing. B.E. Cohen: Conceptualization, funding acquisition, project adminis- tration, supervision, writing-reviewing and editing. K. Zito: Conceptualization, funding acquisition, investigation, metho- dology, project administration, supervision, writing-original draft, writing-reviewing and editing. J.T. Sack: Conceptualization, formal analysis, funding acquisition, investigation, methodology, project administration, supervision, visualization, writing-original draft, writing-reviewing and editing. Submitted: 3 January 2021 Accepted: 3 September 2021 References Adams, S.R. 2010. How calcium indicators work. Cold Spring Harb. Protoc. 2010:top70. https://doi.org/10.1101/pdb.top70 Aggarwal, S.K., and R. MacKinnon. 1996. Contribution of the S4 segment to gating charge in the Shaker K+ channel. Neuron. 16:1169–1177. https://doi .org/10.1016/S0896-6273(00)80143-9 Ahuja, S., S. Mukund, L. Deng, K. Khakh, E. Chang, H. Ho, S. Shriver, C. Young, S. Lin, J.P. Johnson Jr., et al. 2015. Structural basis of Nav1.7 inhibition by an isoform-selective small-molecule antagonist. Science. 350:aac5464. https://doi.org/10.1126/science.aac5464 An, W.F., M.R. Bowlby, M. Betty, J. Cao, H.P. Ling, G. Mendoza, J.W. Hinson, K.I. Mattsson, B.W. Strassle, J.S. Trimmer, and K.J. Rhodes. 2000. Modulation of A-type potassium channels by a family of calcium sen- sors. Nature. 403:553–556. https://doi.org/10.1038/35000592 Antonucci, D.E., S.T. Lim, S. Vassanelli, and J.S. Trimmer. 2001. Dynamic localization and clustering of dendritic Kv2.1 voltage-dependent po- tassium channels in developing hippocampal neurons. Neuroscience. 108:69–81. https://doi.org/10.1016/S0306-4522(01)00476-6 Armstrong, C.M., and F. Bezanilla. 1973. Currents related to movement of the gating particles of the sodium channels. Nature. 242:459–461. https:// doi.org/10.1038/242459a0 Benndorf, K., R. Koopmann, C. Lorra, and O. Pongs. 1994. Gating and conduc- tance properties of a human delayed rectifier K+ channel expressed in frog oocytes. J. Physiol. 477:1–14. https://doi.org/10.1113/jphysiol.1994.sp020166 Bezanilla, F. 2008. How membrane proteins sense voltage. Nat. Rev. Mol. Cell Biol. 9:323–332. https://doi.org/10.1038/nrm2376 Bezanilla, F. 2018. Gating currents. J. Gen. Physiol. 150:911–932. https://doi .org/10.1085/jgp.201812090 Bocksteins, E. 2016. Kv5, Kv6, Kv8, and Kv9 subunits: No simple silent by- standers. J. Gen. Physiol. 147:105–125. https://doi.org/10.1085/jgp .201511507 Bolte, S., and F.P. Cordelières. 2006. A guided tour into subcellular colocal- ization analysis in light microscopy. J. Microsc. 224:213–232. https://doi .org/10.1111/j.1365-2818.2006.01706.x Catterall, W.A., S. Cestèle, V. Yarov-Yarovoy, F.H. Yu, K. Konoki, and T. Scheuer. 2007. Voltage-gated ion channels and gating modifier toxins. Toxicon. 49:124–141. https://doi.org/10.1016/j.toxicon.2006.09.022 Cerda, O., and J.S. Trimmer. 2011. Activity-dependent phosphorylation of J. Biol. Chem. 286: neuronal Kv2.1 potassium channels by CDK5. 28738–28748. https://doi.org/10.1074/jbc.M111.251942 Dai, X.Q., J. Kolic, P. Marchi, S. Sipione, and P.E. Macdonald. 2009. SU- MOylation regulates Kv2.1 and modulates pancreatic beta-cell excita- bility. J. Cell Sci. 122:775–779. https://doi.org/10.1242/jcs.036632 Dai, X.Q., J.E. Manning Fox, D. Chikvashvili, M. Casimir, G. Plummer, C. Hajmrle, A.F. Spigelman, T. Kin, D. Singer-Lahat, Y. Kang, et al. 2012. The voltage-dependent potassium channel subunit Kv2.1 regulates in- sulin secretion from rodent and human islets independently of its electrical function. Diabetologia. 55:1709–1720. https://doi.org/10.1007/ s00125-012-2512-6 Deutsch, E., A.V. Weigel, E.J. Akin, P. Fox, G. Hansen, C.J. Haberkorn, R. Loftus, D. Krapf, and M.M. Tamkun. 2012. Kv2.1 cell surface clusters are insertion platforms for ion channel delivery to the plasma membrane. Mol. Biol. Cell. 23:2917–2929. https://doi.org/10.1091/mbc.e12-01-0047 Dockendorff, C., D.M. Gandhi, I.H. Kimball, K.S. Eum, R. Rusinova, H.I. In- gólfsson, R. Kapoor, T. Peyear, M.W. Dodge, S.F. Martin, et al. 2018. Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 22 of 24 Synthetic analogues of the snail toxin 6-bromo-2-mercaptotryptamine Dimer (BrMT) reveal that lipid bilayer perturbation does not underlie its modulation of voltage-gated potassium channels. Biochemistry. 57: 2733–2743. https://doi.org/10.1021/acs.biochem.8b00292 Du, J., J.H. Tao-Cheng, P. Zerfas, and C.J. McBain. 1998. The K+ channel, Kv2.1, is apposed to astrocytic processes and is associated with inhibitory postsynaptic membranes in hippocampal and cortical principal neurons and inhibitory interneurons. Neuroscience. 84:37–48. https://doi.org/10 .1016/S0306-4522(97)00519-8 Du, J., L.L. Haak, E. Phillips-Tansey, J.T. Russell, and C.J. McBain. 2000. Frequency-dependent regulation of rat hippocampal somato-dendritic excitability by the K+ channel subunit Kv2.1. J. Physiol. 522:19–31. https://doi.org/10.1111/j.1469-7793.2000.t01-2-00019.xm Eichel, K., and M. von Zastrow. 2018. Subcellular organization of GPCR sig- naling. Trends Pharmacol. Sci. 39:200–208. https://doi.org/10.1016/j.tips .2017.11.009 Feinshreiber, L., D. Singer-Lahat, R. Friedrich, U. Matti, A. Sheinin, O. Yizhar, R. Nachman, D. Chikvashvili, J. Rettig, U. Ashery, and I. Lotan. 2010. Non-conducting function of the Kv2.1 channel enables it to recruit vesicles for release in neuroendocrine and nerve cells. J. Cell Sci. 123: 1940–1947. https://doi.org/10.1242/jcs.063719 Fletcher-Taylor, S., P. Thapa, R.J. Sepela, R. Kaakati, V. Yarov-Yarovoy, J.T. Sack, and B.E. Cohen. 2020. Distinguishing potassium channel resting state conformations in live cells with environment-sensitive fluores- 11:2316–2326. https://doi.org/10.1021/ cence. ACS Chem. Neurosci. acschemneuro.0c00276 Fox, P.D., C.J. Haberkorn, A.V. Weigel, J.L. Higgins, E.J. Akin, M.J. Kennedy, D. Krapf, and M.M. Tamkun. 2013a. Plasma membrane domains enriched in cortical endoplasmic reticulum function as membrane protein traf- ficking hubs. Mol. Biol. Cell. 24:2703–2713. https://doi.org/10.1091/mbc .e12-12-0895 Fox, P.D., R.J. Loftus, and M.M. Tamkun. 2013b. Regulation of Kv2.1 K+ conductance by cell surface channel density. J. Neurosci. 33:1259–1270. https://doi.org/10.1523/JNEUROSCI.3008-12.2013 Fox, P.D., C.J. Haberkorn, E.J. Akin, P.J. Seel, D. Krapf, and M.M. Tamkun. 2015. Induction of stable ER-plasma-membrane junctions by Kv2.1 po- tassium channels. J. Cell Sci. 128:2096–2105. https://doi.org/10.1242/jcs .166009 Frech, G.C., A.M. VanDongen, G. Schuster, A.M. Brown, and R.H. Joho. 1989. A novel potassium channel with delayed rectifier properties isolated from rat brain by expression cloning. Nature. 340:642–645. https://doi .org/10.1038/340642a0 Gayet-Primo, J., D.B. Yaeger, R.A. Khanjian, and T. Puthussery. 2018. Het- eromeric KV2/KV8.2 channels mediate delayed rectifier potassium currents in primate photoreceptors. J. Neurosci. 38:3414–3427. https:// doi.org/10.1523/JNEUROSCI.2440-17.2018 Gordon, E., T.K. Roepke, and G.W. Abbott. 2006. Endogenous KCNE subunits govern Kv2.1 K+ channel activation kinetics in Xenopus oocyte studies. Biophys. J. 90:1223–1231. https://doi.org/10.1529/biophysj.105.072504 Gupta, K., M. Zamanian, C. Bae, M. Milescu, D. Krepkiy, D.C. Tilley, J.T. Sack, V. Yarov-Yarovoy, J.I. Kim, and K.J. Swartz. 2015. Tarantula toxins use common surfaces for interacting with Kv and ASIC ion channels. eLife. 4:e06774. https://doi.org/10.7554/eLife.06774 Herrington, J., Y.P. Zhou, R.M. Bugianesi, P.M. Dulski, Y. Feng, V.A. Warren, M.M. Smith, M.G. Kohler, V.M. Garsky, M. Sanchez, et al. 2006. Blockers of the delayed-rectifier potassium current in pancreatic β-cells enhance glucose-dependent insulin secretion. Diabetes. 55:1034–1042. https://doi.org/10.2337/diabetes.55.04.06.db05-0788 Irannejad, R., J.C. Tomshine, J.R. Tomshine, M. Chevalier, J.P. Mahoney, J. Steyaert, S.G. Rasmussen, R.K. Sunahara, H. El-Samad, B. Huang, and M. von Zastrow. 2013. Conformational biosensors reveal GPCR signalling from endosomes. Nature. 495:534–538. https://doi.org/10.1038/nature12000 Islas, L.D., and F.J. Sigworth. 1999. Voltage sensitivity and gating charge in Shaker and Shab family potassium channels. J. Gen. Physiol. 114:723–742. https://doi.org/10.1085/jgp.114.5.723 Jara-Oseguera, A., I.G. Ishida, G.E. Rangel-Yescas, N. Espinosa-Jalapa, J.A. P´erez-Guzm´an, D. El´ıas-Viñas, R. Le Lagadec, T. Rosenbaum, and L.D. Islas. 2011. Uncoupling charge movement from channel opening in voltage-gated potassium channels by ruthenium complexes. J. Biol. Chem. 286:16414–16425. https://doi.org/10.1074/jbc.M110.198010 Johnson, B., A.N. Leek, L. Sol´e, E.E. Maverick, T.P. Levine, and M.M. Tamkun. 2018. Kv2 potassium channels form endoplasmic reticu- lum/plasma membrane junctions via interaction with VAPA and VAPB. Proc. Natl. Acad. Sci. USA. 115:E7331–E7340. https://doi.org/10 .1073/pnas.1805757115 Kaczmarek, L.K. 2006. Non-conducting functions of voltage-gated ion channels. Nat. Rev. Neurosci. 7:761–771. https://doi.org/10.1038/nrn1988 Kang, S.K., C.G. Vanoye, S.N. Misra, D.M. Echevarria, J.D. Calhoun, J.B. O’Connor, K.L. Fabre, D. McKnight, L. Demmer, P. Goldenberg, et al. 2019. Spectrum of KV 2.1 dysfunction in KCNB1-associated neuro- developmental disorders. Ann. Neurol. 86:899–912. https://doi.org/10 .1002/ana.25607 Kirmiz, M., S. Palacio, P. Thapa, A.N. King, J.T. Sack, and J.S. Trimmer. 2018a. Remodeling neuronal ER-PM junctions is a conserved nonconducting function of Kv2 plasma membrane ion channels. Mol. Biol. Cell. 29: 2410–2432. https://doi.org/10.1091/mbc.E18-05-0337 Kirmiz, M., N.C. Vierra, S. Palacio, and J.S. Trimmer. 2018b. Identification of VAPA and VAPB as Kv2 channel-interacting proteins defining endo- plasmic reticulum-plasma membrane junctions in mammalian brain neurons. J. Neurosci. 38:7562–7584. https://doi.org/10.1523/JNEUROSCI .0893-18.2018 Kirmiz, M., T.E. Gillies, E.J. Dickson, and J.S. Trimmer. 2019. Neuronal ER- plasma membrane junctions organized by Kv2-VAP pairing recruit Nir proteins and affect phosphoinositide homeostasis. J. Biol. Chem. 294: 17735–17757. https://doi.org/10.1074/jbc.RA119.007635 Lee, H.C., J.M. Wang, and K.J. Swartz. 2003. Interaction between extracel- lular Hanatoxin and the resting conformation of the voltage-sensor paddle in Kv channels. Neuron. 40:527–536. https://doi.org/10.1016/ S0896-6273(03)00636-6 Legant, W.R., L. Shao, J.B. Grimm, T.A. Brown, D.E. Milkie, B.B. Avants, L.D. Lavis, and E. Betzig. 2016. High-density three-dimensional localization microscopy across large volumes. Nat. Methods. 13:359–365. https://doi .org/10.1038/nmeth.3797 Lewis, G.N. 1925. A new principle of equilibrium. Proc. Natl. Acad. Sci. USA. 11: 179–183. https://doi.org/10.1073/pnas.11.3.179 Li, H., W. Guo, H. Xu, R. Hood, A.T. Benedict, and J.M. Nerbonne. 2001. Functional expression of a GFP-tagged Kv1.5 alpha-subunit in mouse ventricle. Am. J. Physiol. Heart Circ. Physiol. 281:H1955–H1967. https://doi .org/10.1152/ajpheart.2001.281.5.H1955 Lin, M.Z., and M.J. Schnitzer. 2016. Genetically encoded indicators of neu- ronal activity. Nat. Neurosci. 19:1142–1153. https://doi.org/10.1038/nn .4359 Liu, P.W., and B.P. Bean. 2014. Kv2 channel regulation of action potential repolarization and firing patterns in superior cervical ganglion neurons and hippocampal CA1 pyramidal neurons. J. Neurosci. 34:4991–5002. https://doi.org/10.1523/JNEUROSCI.1925-13.2014 Long, S.B., E.B. Campbell, and R. Mackinnon. 2005. Voltage sensor of Kv1.2: structural basis of electromechanical coupling. Science. 309:903–908. https://doi.org/10.1126/science.1116270 Long, S.B., X. Tao, E.B. Campbell, and R. MacKinnon. 2007. Atomic structure of a voltage-dependent K+ channel in a lipid membrane-like environ- ment. Nature. 450:376–382. https://doi.org/10.1038/nature06265 MacDonald, P.E., A.M. Salapatek, and M.B. Wheeler. 2003. Temperature and redox state dependence of native Kv2.1 currents in rat pancreatic beta- J. Physiol. 546:647–653. https://doi.org/10.1113/jphysiol.2002 cells. .035709 Maletic-Savatic, M., N.J. Lenn, and J.S. Trimmer. 1995. Differential spatio- temporal expression of K+ channel polypeptides in rat hippocampal neurons developing in situ and in vitro. J. Neurosci. 15:3840–3851. https://doi.org/10.1523/JNEUROSCI.15-05-03840.1995 Malin, S.A., and J.M. Nerbonne. 2002. Delayed rectifier K+ currents, IK, are encoded by Kv2 alpha-subunits and regulate tonic firing in mammalian sympathetic neurons. J. Neurosci. 22:10094–10105. https://doi.org/10 .1523/JNEUROSCI.22-23-10094.2002 McCord, M.C., and E. Aizenman. 2013. Convergent Ca2+ and Zn2+ signaling regulates apoptotic Kv2.1 K+ currents. Proc. Natl. Acad. Sci. USA. 110: 13988–13993. https://doi.org/10.1073/pnas.1306238110 McCormack, K., S. Santos, M.L. Chapman, D.S. Krafte, B.E. Marron, C.W. West, M.J. Krambis, B.M. Antonio, S.G. Zellmer, D. Printzenhoff, et al. 2013. Voltage sensor interaction site for selective small molecule in- hibitors of voltage-gated sodium channels. Proc. Natl. Acad. Sci. USA. 110: E2724–E2732. https://doi.org/10.1073/pnas.1220844110 McCrossan, Z.A., T.K. Roepke, A. Lewis, G. Panaghie, and G.W. Abbott. 2009. Regulation of the Kv2.1 potassium channel by MinK and MiRP1. J. Membr. Biol. 228:1–14. https://doi.org/10.1007/s00232-009-9154-8 McDonough, S.I., I.M. Mintz, and B.P. Bean. 1997. Alteration of P-type cal- cium channel gating by the spider toxin omega-Aga-IVA. Biophys. J. 72: 2117–2128. https://doi.org/10.1016/S0006-3495(97)78854-4 Milescu, M., F. Bosmans, S. Lee, A.A. Alabi, J.I. Kim, and K.J. Swartz. 2009. Interactions between lipids and voltage sensor paddles detected with Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 23 of 24 tarantula toxins. Nat. Struct. Mol. Biol. 16:1080–1085. https://doi.org/10 .1038/nsmb.1679 Misonou, H., D.P. Mohapatra, E.W. Park, V. Leung, D. Zhen, K. Misonou, A.E. Anderson, and J.S. Trimmer. 2004. Regulation of ion channel localiza- tion and phosphorylation by neuronal activity. Nat. Neurosci. 7:711–718. https://doi.org/10.1038/nn1260 Misonou, H., D.P. Mohapatra, M. Menegola, and J.S. Trimmer. 2005. Calcium- and metabolic state-dependent modulation of the voltage-dependent Kv2.1 channel regulates neuronal excitability in response to ischemia. J. Neurosci. 25:11184–11193. https://doi.org/10.1523/JNEUROSCI.3370-05.2005 Murakoshi, H., G. Shi, R.H. Scannevin, and J.S. Trimmer. 1997. Phosphoryl- ation of the Kv2.1 K+ channel alters voltage-dependent activation. Mol. Pharmacol. 52:821–828. https://doi.org/10.1124/mol.52.5.821 O’Connell, K.M., R. Loftus, and M.M. Tamkun. 2010. Localization-dependent activity of the Kv2.1 delayed-rectifier K+ channel. Proc. Natl. Acad. Sci. USA. 107:12351–12356. https://doi.org/10.1073/pnas.1003028107 Opitz-Araya, X., and A. Barria. 2011. Organotypic hippocampal slice cultures. J. Vis. Exp. (48):2462. Park, K.S., D.P. Mohapatra, H. Misonou, and J.S. Trimmer. 2006. Graded regulation of the Kv2.1 potassium channel by variable phosphorylation. Science. 313:976–979. https://doi.org/10.1126/science.1124254 Peltola, M.A., J. Kuja-Panula, S.E. Lauri, T. Taira, and H. Rauvala. 2011. AMIGO is an auxiliary subunit of the Kv2.1 potassium channel. EMBO Rep. 12:1293–1299. https://doi.org/10.1038/embor.2011.204 Peretz, A., L. Pell, Y. Gofman, Y. Haitin, L. Shamgar, E. Patrich, P. Kornilov, O. Gourgy-Hacohen, N. Ben-Tal, and B. Attali. 2010. Targeting the voltage sensor of Kv7.2 voltage-gated K+ channels with a new gating-modifier. Proc. Natl. Acad. Sci. USA. 107:15637–15642. https://doi.org/10.1073/pnas .0911294107 Plant, L.D., E.J. Dowdell, I.S. Dementieva, J.D. Marks, and S.A. Goldstein. 2011. SUMO modification of cell surface Kv2.1 potassium channels regulates the activity of rat hippocampal neurons. J. Gen. Physiol. 137:441–454. https://doi.org/10.1085/jgp.201110604 Pologruto, T.A., B.L. Sabatini, and K. Svoboda. 2003. ScanImage: flexible software for operating laser scanning microscopes. Biomed. Eng. Online. 2:13. https://doi.org/10.1186/1475-925X-2-13 Ramu, Y., Y. Xu, and Z. Lu. 2006. Enzymatic activation of voltage-gated potassium channels. Nature. 442:696–699. https://doi.org/10.1038/ nature04880 Sack, J.T., R.W. Aldrich, and W.F. Gilly. 2004. A gastropod toxin selectively slows early transitions in the Shaker K channel’s activation pathway. J. Gen. Physiol. 123:685–696. https://doi.org/10.1085/jgp.200409047 Sack, J.T., N. Stephanopoulos, D.C. Austin, M.B. Francis, and J.S. Trimmer. 2013. Antibody-guided photoablation of voltage-gated potassium cur- rents. J. Gen. Physiol. 142:315–324. https://doi.org/10.1085/jgp.201311023 Schmalhofer, W.A., J. Calhoun, R. Burrows, T. Bailey, M.G. Kohler, A.B. Weinglass, G.J. Kaczorowski, M.L. Garcia, M. Koltzenburg, and B.T. Priest. 2008. ProTx-II, a selective inhibitor of NaV1.7 sodium channels, blocks action potential propagation in nociceptors. Mol. Pharmacol. 74: 1476–1484. https://doi.org/10.1124/mol.108.047670 Schneider, M.F., and W.K. Chandler. 1973. Voltage dependent charge move- ment of skeletal muscle: a possible step in excitation-contraction cou- pling. Nature. 242:244–246. https://doi.org/10.1038/242244a0 Schneider, C.A., W.S. Rasband, and K.W. Eliceiri. 2012. NIH Image to ImageJ: 25 years of image analysis. Nat. Methods. 9:671–675. https://doi.org/10 .1038/nmeth.2089 Scholle, A., S. Dugarmaa, T. Zimmer, M. Leonhardt, R. Koopmann, B. Enge- land, O. Pongs, and K. Benndorf. 2004. Rate-limiting reactions deter- mining different activation kinetics of Kv1.2 and Kv2.1 channels. J. Membr. Biol. 198:103–112. https://doi.org/10.1007/s00232-004-0664-0 Seoh, S.A., D. Sigg, D.M. Papazian, and F. Bezanilla. 1996. Voltage-sensing residues in the S2 and S4 segments of the Shaker K+ channel. Neuron. 16: 1159–1167. https://doi.org/10.1016/S0896-6273(00)80142-7 Shibata, R., H. Misonou, C.R. Campomanes, A.E. Anderson, L.A. Schrader, L.C. Doliveira, K.I. Carroll, J.D. Sweatt, K.J. Rhodes, and J.S. Trimmer. 2003. A fundamental role for KChIPs in determining the molecular properties and trafficking of Kv4.2 potassium channels. J. Biol. Chem. 278:36445–36454. https://doi.org/10.1074/jbc.M306142200 Singer-Lahat, D., A. Sheinin, D. Chikvashvili, S. Tsuk, D. Greitzer, R. Frie- drich, L. Feinshreiber, U. Ashery, M. Benveniste, E.S. Levitan, and I. Lotan. 2007. K+ channel facilitation of exocytosis by dynamic interac- tion with syntaxin. J. Neurosci. 27:1651–1658. https://doi.org/10.1523/ JNEUROSCI.4006-06.2007 Speca, D.J., G. Ogata, D. Mandikian, H.I. Bishop, S.W. Wiler, K. Eum, H.J. Wenzel, E.T. Doisy, L. Matt, K.L. Campi, et al. 2014. Deletion of the Kv2.1 delayed rectifier potassium channel leads to neuronal and behavioral hyperexcitability. Genes Brain Behav. 13:394–408. https://doi.org/10 .1111/gbb.12120 Stewart, R., B.E. Cohen, and J.T. Sack. 2021. Fluorescent toxins as ion channel activity sensors. Methods Enzymol. 653:295–318. https://doi.org/10.1016/ bs.mie.2021.02.014 Stoppini, L., P.A. Buchs, and D. Muller. 1991. A simple method for organotypic cultures of nervous tissue. J. Neurosci. Methods. 37:173–182. https://doi .org/10.1016/0165-0270(91)90128-M Swartz, K.J. 2007. Tarantula toxins interacting with voltage sensors in po- tassium channels. Toxicon. 49:213–230. https://doi.org/10.1016/j.toxicon .2006.09.024 Tanabe, T., K.G. Beam, J.A. Powell, and S. Numa. 1988. Restoration of excitation-contraction coupling and slow calcium current in dysgenic muscle by dihydropyridine receptor complementary DNA. Nature. 336: 134–139. https://doi.org/10.1038/336134a0 Tao, X., A. Lee, W. Limapichat, D.A. Dougherty, and R. MacKinnon. 2010. A gating charge transfer center in voltage sensors. Science. 328:67–73. https://doi.org/10.1126/science.1185954 Tilley, D.C., K.S. Eum, S. Fletcher-Taylor, D.C. Austin, C. Dupr´e, L.A. Patrón, R.L. Garcia, K. Lam, V. Yarov-Yarovoy, B.E. Cohen, and J.T. Sack. 2014. Chemoselective tarantula toxins report voltage activation of wild-type ion channels in live cells. Proc. Natl. Acad. Sci. USA. 111:E4789–E4796. https://doi.org/10.1073/pnas.1406876111 Tilley, D.C., J.M. Angueyra, K.S. Eum, H. Kim, L.H. Chao, A.W. Peng, and J.T. Sack. 2019. The tarantula toxin GxTx detains K+ channel gating charges in their resting conformation. J. Gen. Physiol. 151:292–315. https://doi .org/10.1085/jgp.201812213 Trapani, J.G., and S.J. Korn. 2003. Control of ion channel expression for patch clamp recordings using an inducible expression system in mammalian cell lines. BMC Neurosci. 4:15. https://doi.org/10.1186/1471-2202-4-15 Tsvetanova, N.G., R. Irannejad, and M. von Zastrow. 2015. G protein-coupled receptor (GPCR) signaling via heterotrimeric G proteins from endosomes. J. Biol. Chem. 290:6689–6696. https://doi.org/10.1074/jbc.R114.617951 Turner, M., D.E. Anderson, P. Bartels, M. Nieves-Cintron, A.M. Coleman, P.B. Henderson, K.N.M. Man, P.Y. Tseng, V. Yarov-Yarovoy, D.M. Bers, et al. 2020. α-Actinin-1 promotes activity of the L-type Ca2+ channel Cav 1.2. EMBO J. 39:e102622. https://doi.org/10.15252/embj.2020106171 Vacher, H., D.P. Mohapatra, and J.S. Trimmer. 2008. Localization and tar- geting of voltage-dependent ion channels in mammalian central neu- rons. Physiol. Rev. 88:1407–1447. https://doi.org/10.1152/physrev.00002 .2008 Vierra, N.C., M. Kirmiz, D. van der List, L.F. Santana, and J.S. Trimmer. 2019. Kv2.1 mediates spatial and functional coupling of L-type calcium channels and ryanodine receptors in mammalian neurons. eLife. 8: e49953. https://doi.org/10.7554/eLife.49953 Weigel, A.V., P.D. Fox, E.J. Akin, K.H. Ecklund, M.M. Tamkun, and D. Krapf. 2012. Size of cell-surface Kv2.1 domains is governed by growth fluctua- tions. Biophys. J. 103:1727–1734. https://doi.org/10.1016/j.bpj.2012.09.013 Weigel, A.V., M.M. Tamkun, and D. Krapf. 2013. Quantifying the dynamic in- teractions between a clathrin-coated pit and cargo molecules. Proc. Natl. Acad. Sci. USA. 110:E4591–E4600. https://doi.org/10.1073/pnas.1315202110 Woods, G., and K. Zito. 2008. Preparation of gene gun bullets and biolistic transfection of neurons in slice culture. J. Vis. Exp. 12:675. https://doi .org/10.3791/675 Xu, H., T. Li, A. Rohou, C.P. Arthur, F. Tzakoniati, E. Wong, A. Estevez, C. Kugel, Y. Franke, J. Chen, et al. 2019. Structural basis of Nav1.7 inhi- bition by a gating-modifier spider toxin. Cell. 176:1238–1239. https://doi .org/10.1016/j.cell.2019.01.047 Yang, W., and R. Yuste. 2017. In vivo imaging of neural activity. Nat. Methods. 14:349–359. https://doi.org/10.1038/nmeth.4230 Zagotta, W.N., T. Hoshi, J. Dittman, and R.W. Aldrich. 1994. Shaker potassium channel gating. II: Transitions in the activation pathway. J. Gen. Physiol. 103:279–319. https://doi.org/10.1085/jgp.103.2.279 Zhang, G., S. Zheng, H. Liu, and P.R. Chen. 2015. Illuminating biological processes through site-specific protein labeling. Chem. Soc. Rev. 44: 3405–3417. https://doi.org/10.1039/C4CS00393D Zhang, A.H., G. Sharma, E.A.B. Undheim, X. Jia, and M. Mobli. 2018. A complicated complex: Ion channels, voltage sensing, cell membranes and peptide inhibitors. Neurosci. Lett. 679:35–47. https://doi.org/10 .1016/j.neulet.2018.04.030 Zito, K., G. Knott, G.M. Shepherd, S. Shenolikar, and K. Svoboda. 2004. In- duction of spine growth and synapse formation by regulation of the spine actin cytoskeleton. Neuron. 44:321–334. https://doi.org/10.1016/j .neuron.2004.09.022 Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 24 of 24 Supplemental material Figure S1. Synthesis of GxTX-594. (A) Molecular model of GxTX-594. Scale bar, 10 ˚A. Backbone of GxTX peptide depicted with ribbon. Cys13(maleimide- Alexa 594) depicted with CPK coloring. (B) HPLC chromatogram of Ser13Cys GxTX. Gradient described in –GxTX synthesis. Ser13Cys GxTX eluted at 12.6 min, peak 1, which corresponds to 33% ACN. MALDI-TOF mass spectrometry profile of peak 1 (inset). (C) HPLC chromatogram of GxTX-594 conjugation reaction between Alexa Fluor 594-maleimide and Ser13Cys GxTX. Peak 1 is Ser13Cys GxTX (retention time, 12.8 min; 33% ACN); peak 2 is a minor product from conjugation; and peak 3 is GxTX-594, the major product from conjugation (retention time, 16.4 min; 35% ACN). The fractions corresponding to peak 3 were combined. (D) HPLC chromatogram of the combined peak 3 fractions from C, a GxTX-594 preparation used in this study. 2 μl of 13.1 μM GxTX-594 diluted in 200 μl of 0.1% TFA was injected. MALDI-TOF mass spectrometry profile of the combined peak 3 fractions (inset). a.m.u., atomic mass unit; Rel. Int., relative intensity. Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 S1 Figure S2. GxTX-594 selectively labels Kv2 proteins on cell surfaces. GxTX-594 labeling was assessed in CHO cells expressing Kv2.1-GFP, Kv2.2-GFP, Kv4.2-GFP, Kv1.5-GFP, and BK-GFP. Each of these channel subtypes were assessed for voltage-dependent outward currents to identify cell surface expression of the K+ channels. CHO cells were not cotransfected with the β subunits for Kv4.2, KChIP2, or Kv1.5, Kvβ2, to assess whether these β subunits interfere with GxTX-594 binding. (A) Exemplar whole-cell voltage clamp recordings of CHO cells expressing Kv2.1-GFP, Kv2.2-GFP, Kv4.2-GFP, Kv1.5-GFP, or BK-GFP. Re- cordings shown are representative responses to 100-ms steps from −100 mV to −40, 0, and +40 mV. (B) Confocal imaging of fluorescence from live CHO cells transfected with Kv2.1-GFP, Kv2.2-GFP, Kv4.2-GFP, Kv1.5-GFP, or BK-GFP (indicated by row) and labeled with GxTX-594. Confocal imaging plane was >1 μm above the glass-adhered surface. Cells were incubated with 100 nM GxTX-594 and 5 μg/ml WGA-405 and rinsed before imaging. Fluorescence shown cor- responds to emission of GFP (column 1); Alexa Fluor 594 (column 2); WGA-405 (column 3); or an overlay of GFP, Alexa Fluor 594, and WGA-405 (column 4). Scale bars, 20 μm. (C) Ratio of fluorescence intensity resulting from excitation of GxTX-594 at 594 nm and GFP at 488 nm. Analysis methods as in Fig. 4 B. Kv2.1, n = 11; Kv2.2, n = 9; Kv4.2, n = 13; Kv1.5, n = 13; and BK, n = 12; n indicates the number of individual cells analyzed in a single dish during a single application of GxTX-594 with the indicated K+ channel-GFP type. Bars represent the mean. Each circle corresponds to a cell. Significant differences were observed between GxTX:GFP ratio for Kv2.1 or Kv2.2 and Kv1.5, Kv4.2, or BK by Mann–Whitney U test (P < 0.0001). The P value to determine significance is adjusted for multiple comparisons using the Bonferroni method, where P < 0.0033 is considered significant, with the caveat that data points under each condition are technical replicates. (D) Pearson correlation coefficients between GxTX-594 and GFP. Same cells as C. Analysis methods as in Fig. 4 C. Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 S2 Figure S3. GxTX-594 labeling of surface membranes requires Kv2 protein. GxTX-594 partitioning into the membrane was assessed with fluorescence from nontransfected CHO cells and cells transfected with Kv2-GFP proteins. (A) Fluorescence from live CHO cells transfected with Kv2.1-GFP (top row) or Kv2.2-GFP (bottom row) and labeled with GxTX-594. Airy disk imaging was at a plane above the glass-adhered surface. Cells were incubated with 100 nM GxTX-594 and 5 μg/ml WGA-405 then diluted to 9 nM GxTX-594 and 0.45 μg/ml WGA-405 before imaging. Scale bars, 20 μm. Fluorescence shown was excited at 488 nm (column 1), 594 nm (column 2), or 405 nm (column 3). Scale bars, 20 μm. (B) Fluorescence intensity from Kv2.1-GFP transfected cells with excitation of GxTX-594 at 594 nm versus GFP at 488 nm. Fluorescence from WGA-405 was used as a mask to manually draw ROIs on cells. Each point represents one cell. Cells with obvious GFP fluorescence are green points, cells without are black points. Mean background fluorescence from a region that did not contain cells is indicated by dashed lines. Red line represents a linear fit of cells with obvious GFP fluorescence. Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 S3 Figure S4. Extracellular access can impact GxTX-594 labeling kinetics. (A) Time-lapse Airy disk images of the glass-adhered surface of a voltage-clamped Kv2.1-CHO cell in 9 nM GxTX-594. Time index is in the upper left corner of each panel, and membrane potential is indicated in the upper right corner. Color progression for pseudocoloring of fluorescence intensity is shown in vertical bar on right. Scale bar, 10 μm. (B) Airy disk image of the glass-adhered surface of a voltage-clamped Kv2.1-CHO cell in 9 nM GxTX-594. Gray lines indicate boundaries of ROIs. ROIs 1, 2, and 3 are concentric circles, each with a respective diameter of 1.8, 4.9, and 9.1 μm. ROI 4 was hand-drawn to contain the apparent cell surface. In all cells analyzed, ROIs 1–3 were concentric circles of the same sizes, while ROI 4 varied based on cell shape. Scale bar, 10 μm. (C) Representative traces of GxTX-594 intensity response to voltage changes. Red lines are monoexponential fits (Eq. 1): 40-mV step ROI 1, kΔF = 4.29 × 10−2 ± 0.26 × 10−2 s−1; ROI 2, kΔF = 4.39 × 10−2 ± 0.16 × 10−2 s−1; ROI 3, kΔF = 5.65 × 10−2 ± 0.15 × 10−2 s−1; and ROI 4, kΔF = 9.69 × 10−2 ± 0.33 × 10−2 s−1. −80-mV step ROI 1, kΔF = 4.27 × 10−4 ± 0.11 × 10−4 s−1; ROI 2, kΔF = 7.999 × 10−4 ± 0.053 × 10−4 s−1; ROI 3, kΔF = 2.7556 × 10−3 ± 0.0074 × 10−3 s−1; and ROI 4, kΔF = 5.46 × 10−3 ± 0.13 × 10−3 s−1. Background for subtraction was the average intensity of a region that did not contain cells over the time course of the voltage protocol. Each trace was normalized to initial fluorescence intensity before the application of the voltage stimulus. (D) kΔF at +40 mV from individual cells. (E) kΔF at −80 mV from individual cells. Circle coloring in D and E indicates data from the same cell. Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 S4 Figure S5. Variation in temperature does not account for cell-to-cell variability of GxTX-594 kinetics. Temperature dependence of GxTX-594 labeling was assessed by holding the cell bath solution at either 27°C or 37°C and stepping the membrane voltage from −80 mV to 0 mV for a measurement of kΔF at both temperatures. (A) Representative traces of GxTX-594 fluorescence intensity response to voltage changes at 27°C (black) and 37°C (gray). Smooth lines are fits of a monoexponential function (Eq. 1): 27°C, kΔF = 2.47 × 10−2 ± 0.39 × 10−2 s−1; 37°C, kΔF = 7.54 × 10−2 ± 0.54 × 10−2 s−1. Background subtraction was performed as in Fig. 6 B. (B) kΔF at 27°C and 37°C. The rate of fluorescence change was significantly faster at higher temperatures (Mann–Whitney P = 0.0005). From geometric means (bars), a Q10 of 3.8 was calculated between 27°C and 37°C. Each circle represents one cell, n = 7 both groups. ***, P < 0.0001. Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 S5 Figure S6. GxTX-594 labels CA1 hippocampal pyramidal neurons transfected with Kv2.1-GFP. Two-photon excitation images of rat CA1 hippocampal pyramidal neurons in brain slices as in Fig. 9 A. Kv2.1-GFP (left), GxTX-594 (middle), and overlay (right). Scale bars, 10 μm in all panels. (A) Pyramidal neurons 2 d after transfection with Kv2.1-GFP. (B) Pyramidal neurons 4 d after transfection with Kv2.1-GFP. (C) Pyramidal neurons 6 d after transfection with Kv2.1- GFP. Video 1. Time-lapse image sequence of GxTX-594 fluorescence on a voltage-clamped CA1 hippocampal pyramidal neuron while it is depolarized from −70 to 0 mV. Frame rate, 0.1 fps. Playback speed, 7 fps. Data S1 is provided online as a separate Excel file and shows a spreadsheet containing model calculations that can be used to generate model prediction. Thapa et al. Imaging conformational change of endogenous Kv2 Journal of General Physiology https://doi.org/10.1085/jgp.202012858 S6
null
10.1007_s13280-020-01405-w.pdf
null
null
Ambio 2021, 50:586–600 https://doi.org/10.1007/s13280-020-01405-w R E S E A R C H A R T I C L E Ecosystem service lens reveals diverse community values of small-scale fisheries Kara E. Pellowe , Heather M. Leslie Received: 3 March 2020 / Revised: 9 August 2020 / Accepted: 29 September 2020 / Published online: 3 November 2020 to of ocean benefits Abstract The coastal provides communities around the world, however, the depth and interactions with marine complexity people’s represented in many marine ecosystems are not well management initiatives. Many fisheries are managed to maximize provisioning value, which is readily quantified, while ignoring cultural values. An ecosystem services includes both provisioning and cultural approach that services will enable managers to better account for the diverse values marine fisheries provide to coastal communities. In this study, we assess community values related to a top fished species, the Mexican chocolate clam, Megapitaria squalida, in Loreto, Baja California Sur, Mexico. We conducted an exploratory analysis based on 42 household surveys, and found that community members perceive multiple provisioning and cultural benefits from the clam, including community economic, historical, and identity values. Despite reporting infrequent harvest and consumption of clams, participants perceive the species as an important part of community identity, highlighting the role of Mexican chocolate clams as a cultural keystone species in the Loreto region. Fisheries management that recognizes the full range of ecosystem services a species contributes to coastal communities will be better equipped to sustain these diverse values into the future. Keywords Community value (cid:2) Cultural ecosystem services (cid:2) Cultural keystone species (cid:2) Ecosystem services (cid:2) Gulf of california (cid:2) Small-scale fisheries Electronic supplementary material The online version of this article (https://doi.org/10.1007/s13280-020-01405-w) contains sup- plementary material, which is available to authorized users. INTRODUCTION The ocean provides many benefits to coastal communities, income, recreational opportunities, and including food, aesthetic values (Halpern et al. 2012; Loomis and Paterson 2014), yet interactions the depth and complexity of between people and marine ecosystems are not well understood (Villasante et al. 2013). Management of fish- eries and decisions related to governance of marine ecosystems reflect society’s values, priorities, and desires for ecosystems to produce certain benefits. These decisions are complicated by multiple and sometimes contradictory goals, with priority often given to values that can be readily quantified in economic terms (Loomis and Paterson 2014). A holistic understanding of the values produced by marine ecosystems is necessary, if management is to accurately reflect diverse values and balance trade-offs between alternate priorities. Full consideration of the values associated with ecosystem services will better enable resource managers to address the needs and perspectives of different stakeholders (Chan et al. 2012b). Ecosystem services are the benefits that an ecosystem provides to people (Millennium Ecosystem Assessment 2005). The Millennium Ecosystem Assessment (2005) outlined four categories of ecosystem services: supporting—those services that make it possible for ecosystems to continue providing the other three types of services (e.g., primary production); provisioning— products obtained from ecosystems (e.g., food); regulat- ing—benefits produced through ecological processes (e.g., water purification); and cultural—nonmaterial benefits of ecosystems (e.g., recreation and sense of place). The ecosystem services approach is a useful tool for under- standing the connections between humans and ecosystems that goes beyond quantifiable outcomes such as income and 123 (cid:2) The Author(s) 2020 www.kva.se/en Ambio 2021, 50:586–600 587 food provision to include cultural and social values (Chan et al. 2012b). Earlier work on ecosystem services involved the integration of biophysical and economic perspectives to assess the value of biophysical processes in economic terms (Daily et al. 2000; Turner and Daily 2008). Eco- nomic approaches have been useful in advancing under- standing of human–nature relationships and facilitating integration of ecosystem-related values into decision- making (Turner and Daily 2008). However, economic approaches fail to encompass dimensions of value that cannot be quantified in economic terms, including many cultural and non-use values (Chan et al. 2011, 2012b). Resource management that is focused on a limited set of ecosystem services may lead to unexpected regime shifts and sudden losses of other ecosystem services (Gordon et al. 2008; Bennett et al. 2009). A growing body of lit- erature highlights the importance of considering and assessing cultural ecosystem services, in addition to pro- visioning services (Martı´n-Lo´pez et al. 2012, 2013; Her- na´ndez-Morcillo et al. 2013; Oteros-Rozas et al. 2014; Dickinson and Hobbs 2017). Cataloguing the complete suite of values marine ecosystems produce is a crucial step in managing in a way that both protects crucial benefits and better attends to trade-offs among the diverse values and priorities of coastal communities (Loomis and Paterson 2014). Provisioning services, such as clean water, food, and income, are essential for providing the basic necessities of life, maintaining security, and protecting human health (Millennium Ecosystem Assessment 2005). As a country follows a development trajectory, human dependence on provisioning services tends to decrease, while dependence on cultural ecosystem services increases (Guo et al. 2010). Unlike provisioning services, which may be replaced by technical innovation or trade as they are degraded, cultural services are not as readily replaced (Millennium Ecosystem Assessment 2005). Cultural ecosystem services are more likely to be co-produced through the interactions between people and their environment, resulting in a tight coupling between the cultural benefits of ecosystems and people’s held values and preferences (Russell et al. 2013; Dickinson and Hobbs 2017). Cultural services are also reflective of people’s environmental decision-making (Martı´n-Lo´pez et al. 2013), and can improve human health and well-being through personal and community connections to natural systems (Russell et al. 2013). Thus, it is critically important to account for and assess both provisioning and cultural values if marine management is to preserve both the basic necessities of life provided by fisheries, as well as socio- cultural value that connects people and the sea. In fisheries, resource exploitation by humans can sig- nificantly affect system structure and functioning, and the long-term sustainability of human–resource impact interactions (Basurto et al. 2013; Partelow and Boda 2015). Fisheries provide many valuable benefits to coastal com- munities, yet their sustainability is threatened by overex- ploitation, pollution, and environmental variability, among other stressors (Be´ne´ 2006; Halpern et al. 2012). Tradi- tional fisheries management focuses primarily on fisheries yield as a product of ecological processes and driver of economic benefits, and has come a long way in acknowl- edging and understanding the heterogeneity of ecological systems. However, a parallel understanding of variety within social systems is often missing (St. Martin et al. 2007). Given the deep and complex ways in which people interact with marine ecosystems (Villasante et al. 2013), particularly through fisheries, the focus of traditional fish- eries management is too narrow and overlooks the many other ways in which people interact with and derive ben- efits from marine species and ecosystems. Meeting the challenge of fisheries management requires moving beyond assessments solely of environmental variables and species interactions to develop a better understanding of socio- cultural values and local knowledge of coastal communi- ties and fishers (St. Martin et al. 2007; Johnson 2018; Smith and Basurto 2019). For small-scale fisheries, our very definitions, typically centered on technology and harvest, ignore the sociocul- tural characteristics of these fisheries that set them apart from other types of fishing (Smith and Basurto 2019). An ecosystem services approach can illuminate important connections between people and nature and help untangle complex interactions shaping small-scale fishery systems. On the Gulf of California coast of Baja California Sur, Mexico, the Town of Loreto relies on fishing and tourism to support the local economy. These activities are primarily focused on the marine park that the town hosts, Loreto Bay National Park. The national park is home to many species, including the Mexican chocolate clam, Megapitaria squalida. The clam is one of the top species harvested by biomass in Loreto (Pellowe and Leslie 2017), and is a local culinary specialty with a rich history of use. As is the case for many fished species, fisheries management of Mexican chocolate clams in Loreto Bay National Park focuses on the maximization of fisheries and economic yield. How- ever, based on the importance of Mexican chocolate clams to local (Pellowe and Leslie 2019), we hypothesize that the relationship between people and Mexican chocolate clams in the Loreto region is more multi-dimensional than is currently captured by fisheries management. livelihoods This exploratory study presents a novel approach for assessing community values of a single fished species. Using household surveys, this study elicits data on the suite of ecosystem services provided by Mexican chocolate clams to households in this region, using a set of values (cid:2) The Author(s) 2020 www.kva.se/en 123 588 Ambio 2021, 50:586–600 adapted from previous studies of ecosystem services (Rolston and Coufal 1991; Reed and Brown 2003; Mil- lennium Ecosystem Assessment 2005; Raymond and Brown 2006). In addition to assessing the range of provi- sioning and cultural values that Mexican chocolate clams provide to households in Loreto, we also assess community perceptions of change related to the clams, since percep- tions of change shape people’s environmental decision- making and can help to illuminate priorities for manage- ment (Gobster et al. 2007). Finally, we explore how fishery management might better account for trade-offs among varied community values and priorities. MATERIALS AND METHODS Location of study The Town of Loreto, Baja California Sur, Mexico, lies along the sea between the Sierra de la Giganta Mountains and the Gulf of California (Fig. 1). Loreto is home to roughly 19 000 people, and the town’s economy depends on fisheries and tourism centered around the marine park it hosts (Instituto Nacional de Estadı´stica y Geografı´a, INEGI 2017). Loreto Bay National Park (LBNP) is one of the largest marine protected areas in Mexico with an area of 2065 km2. The park contains varied marine and estuarine habitat types, including rocky reefs, seagrass beds, man- groves, and sandy habitats, and hosts a variety of permitted including SCUBA diving, snorkeling, whale activities, watching, wildlife viewing, kayaking, and commercial and sport fishing of select species (Comisio´n Nacional de A´ reas Naturales Protegidas 2019). The waters of LBNP are home to 800 marine species, including the Mexican chocolate clam, M. squalida (Fig. 2). Mexican chocolate clams are soft-sediment burrowers that inhabit sandy-bottom habitat from the intertidal to depths of 160 m (Keen 1971). In Loreto Bay, Mexican chocolate clams are an important source of food and income for local fishing communities; they are among the top 5 species harvested by total bio- mass, and among the top 10 by total value (Pellowe and Leslie 2017). Mexican chocolate clams are in demand year-round, sometimes despite seasonal harvest bans. The clams are a long-standing culinary tradition in the region, headline the menu of local restaurants, and are the focus of an annual gastronomic festival held on Loreto’s waterfront. The clam also serves as a symbol of community pride and connection to the sea; murals around Loreto Bay depict smiling clams reminding locals to fish responsibly. For many families in the region, Mexican chocolate clams provide supplemen- tary food and income in times of limited resources, and serve as a safeguard against scarcity. Surveys From February to May 2019, we carried out 48 surveys with residents of Loreto, Baja California Sur, Mexico to explore community perspectives on a range of ecosystem services. Prior to survey administration, questions were carefully reviewed, translated, and pretested with local volunteers to ensure the validity and clarity of questions in both English and Spanish (Groves et al. 2011). Surveys with less than 25% of questions completed (12 or fewer questions answered out of 48 total questions) were removed from the sample. Forty-two surveys were inclu- ded in subsequent analyses. The participant population included adult community members (at least 18 years of age) of any occupation, residing in Loreto, Baja California Sur, Mexico at least 6 months of the year. Since Loreto has a large community of non-Mexican expat residents and Mexican nationals who are not originally from Loreto, the participant population included Mexican nationals origi- nally from Loreto (Loretanos), other Mexican nationals who reside in Loreto, and nationals of other countries who reside in Loreto. Survey participants were recruited via purposive sampling of contacts established during previous fieldwork in the region. Sampling excluded residents with known economic dependence on the fishery (e.g., fishers), but included residents of Loreto thought to value Mexican chocolate clams based on their interest in our previous research. Participants were asked to answer survey ques- tions from the perspective of their entire household, even if they themselves were not heads of household. Due to variable literacy rates in the region, participants had the option of taking the survey themselves or having survey questions read aloud to them and their responses recorded by the researcher. Of 48 total surveys adminis- tered, 38 participants elected to take the survey themselves, and 10 elected to have the survey administered to them. Participants who took the survey themselves were less likely to complete it (32 of 38 surveys completed), as compared to participants who elected to have the survey administered to them by the researcher (10 out of 10 sur- veys completed). Informed consent was obtained from all participants prior to survey administration. Surveys were conducted in both Spanish and English. The survey instrument was written in both languages, allowing par- ticipants to read and respond in their preferred language. For surveys administered by the researcher, participants had the option to choose their preferred language for questions and responses. Each survey took approximately 10–20 min to complete. All procedures performed in this study were in accordance with the Ethical Standards of the Institutional Review Board (University of Maine IRB Permit # 2018-07-01). 123 (cid:2) The Author(s) 2020 www.kva.se/en Ambio 2021, 50:586–600 589 Fig. 1 Map of Loreto Bay National Park, Baja California Sur, Mexico Surveys were anonymous and collected information on the socioeconomic characteristics of households, how fre- quently members of their household harvest, buy, sell, and consume Mexican chocolate clams, and changes they have observed in the availability, market demand, quantity, time (survey quality, price, and size of clams over (cid:2) The Author(s) 2020 www.kva.se/en 123 590 Ambio 2021, 50:586–600 Fig. 2 Mexican chocolate clams. Photo by K. E. Pellowe instrument available as Supplementary Material). Partici- pants were then asked, using a three-item Likert scale (Likert 1932), to indicate whether they agreed, disagreed, or neither agreed nor disagreed with a set of statements (Table 1), each relating to an ecosystem service they and their household may receive from Mexican chocolate clams. Participants could also elect not to answer any questions of their choosing. Surveys were designed to elicit both use and non-use values. Selection of the services assessed in this study resulted from the compilation and adaptation of lists of multiple provisioning and cultural ecosystem services identified in diverse ecosystems (Rol- ston and Coufal 1991; Reed and Brown 2003; Millennium Ecosystem Assessment 2005; Raymond and Brown 2006). The final list of services consisted of values appropriate for assessment for individual species, and included general (household level), general (community level), life-sustain- ing (household level), life-sustaining (ecological), eco- nomic (household level), economic (community level), tourism, subsistence, scientific/learning, recreation, aes- thetic, future use, historic, cultural, identity, community identity, existence, and intrinsic values (see individual 123 (cid:2) The Author(s) 2020 www.kva.se/en Ambio 2021, 50:586–600 591 Table 1 Value statements used to identify participants’ identification of ecosystem service values. Participants’ indication that they agreed with each statement (as opposed to having disagreed or said that they neither agreed nor disagreed) indicated their belief that Mexican chocolate clams provide the associated ecosystem service value. Intrinsic value was reverse-coded, where disagreement with the associated statement was taken as indication that the participant believed that Mexican chocolate clams have intrinsic value Ecosystem service value assessed Value statement General Chocolate clams are important to me and my family Chocolate clams are important to my community Life sustaining Chocolate clams help sustain me and my family Economic Tourism Subsistence Scientific/learning Recreation Aesthetic Future use Historic Cultural Individual identity Community identity Existence Intrinsic Chocolate clams help sustain other animals in Loreto Bay Chocolate clams provide income to my household Chocolate clams are important to the local economy Tourists spend money on chocolate clams when they visit Loreto Chocolate clams are a tourist attraction of Loreto Chocolate clams provide some of my family’s basic needs Chocolate clams are important for scientists to study Chocolate clams should be protected so that people can learn about them Chocolate clams are important for recreation, including exercise and fun It is fun or relaxing to look for or harvest chocolate clams Chocolate clams are beautiful Chocolate clams contribute to the unique beauty of Loreto Chocolate clams should be conserved for future generations Chocolate clams should be conserved because I or my family might want to harvest them in the future Chocolate clams are important because of their history in this area Chocolate clams are important to the culture of this area Chocolate clams are an important part of who I am as an individual Chocolate clams are an important part of what it means to be a Loretano or to live in this area Even when I don’t use chocolate clams, I like to know they are there Chocolate clams have value primarily because they provide benefits to people (reverse-coded) Table 1 for full ecosystem service values). list of statements used to determine We defined general value at the household level to be the overall importance of the clam to the participant’s household, while general value at the community level was the overall importance of the clam to the community. Life- sustaining value at the household level was considered to be the clams’ provision of life-sustaining benefits to the participant’s household, including food, income, or secu- rity, and life-sustaining value at the ecological scale was the clams’ role in sustaining other species or contributing to the broader coastal ecosystem. We defined economic value as the provision of income to the participant’s household, or to the broader community. Tourism value was defined as income generated from tourist activities (e.g., patronizing local restaurants to consume clams), or increased tourism as a result of the presence of Mexican chocolate clams in the region. Subsistence value was considered to be the provision of the participant’s basic needs, including food and/or income. Scientific/learning value was considered the potential for learning generated by the existence of the species, and the possibility for the advancement of science through studies of the species. Recreational value was defined as the potential for fun, relaxation, or enjoyment from harvesting or searching for clams. Aesthetic value was considered to be the beauty of the clam itself or its contribution to the overall beauty of the region. Future use value was defined as the ability of the participant or their household to harvest clams in the future, or the knowledge that future generations within the broader community would be able to harvest clams. His- toric value was considered to be the importance of the clam to regional history, and cultural value as the contribution of the clam to regional culture and practice. We considered individual identity value to be the importance of the clam in constructing individual worldview and sense of self. Community identity value was considered to be the con- tribution of the clam to a shared sense of what it means to be a member of the Loreto community. Existence value was considered to be the satisfaction of knowing that the clam exists in Loreto Bay National Park, and intrinsic value was the belief that Mexican chocolate clams have inherent value, outside of human interaction. (cid:2) The Author(s) 2020 www.kva.se/en 123 592 Ambio 2021, 50:586–600 The ecosystem services of tourism, scientific/learning, recreation, and aesthetic values were assessed each with two survey questions, and an average was taken from the two responses to determine whether participants identified these values from Mexican chocolate clams. Additionally, we assessed the following values both at the individual and the community level through two separate questions: gen- eral, economic, future use, and identity. For open-ended survey questions, including questions on the nature of changes observed, and participants’ perspectives on why changes had occurred, responses were coded into cate- gories. These categories emerged from analysis of partic- ipant responses by the researcher who conducted the surveys. Responses that were cited by two or more par- ticipants were considered response categories. We analyzed separately the responses of three partici- pant groups: Mexican nationals originally from Loreto; Mexican nationals not originally from Loreto; and foreign nationals. Maps and figures were created using R statistical software (R Core Team 2020) and the R packages ggplot2, ggspatial, rnaturalearth, and wesanderson (Wickham 2016; South 2017; Karthik and Wickham 2018; Dunnington 2020). RESULTS English (48%). Overall, 19% of survey participants were Mexican nationals originally from Loreto or Loretanos. Participants varied in the average length of time they had lived in Loreto, their mean monthly household income, mean household size, and reported frequency of use of Mexican chocolate clams (Table 2). None of the participants of any group reported clam- ming as a source of household income, despite Loretano participants reporting selling chocolate clams 7.4 times per year on average. 67% of Loretano participants responded that they had harvested Mexican chocolate clams at some point in the past. 50% of other Mexican participants and 25% of foreign participants reported harvesting Mexican chocolate clams at some point in the past. The participants originally from Loreto who indicated that they regularly harvest or used to regularly harvest Mexican chocolate clams had 13.3 years of harvest experience, on average, with a range of 5 to 20 years of experience. Loretano participants also had, on average, 34.6 years of experience buying Mexican chocolate clams, with a range of experi- ence from 1 to 82 years. Mexican participants not origi- nally from Loreto reported an average of 4.1 years of harvest experience and 15.7 years of buying experience, while foreign participants reported 8.7 years of harvest experience and 8.4 years of buying experience, on average. Perceptions of change Participant demographics and use behavior Of 42 survey participants whose responses were included in the final analyses, 52% were Mexican nationals (of which, 40% were originally from Loreto) and 48% were nationals of other countries, including the United States, Canada, Germany, Australia, Chile, Switzerland, and the United Kingdom. These numbers also correspond to the number of surveys conducted in Spanish (52%) and 83% of Loretanos surveyed, 93% of Mexican participants not originally from Loreto, and 50% of foreign participants said they had noticed at least one change over time in terms of market demand, quantity, quality, size, price, and/or availability of the species. Observations of change varied by participant group and type of change (Fig. 3). Differing levels of observations of change may have been due, in part, to varying lengths of time spent in the region among Table 2 Demographic characteristics and reported use behavior by participant group Demographic characteristics and use behavior Participant group Mexican nationals from Loreto Mexican nationals from elsewhere Foreign nationals n Mean time in Loreto (years) Mean monthly household income (US Dollars) Mean household size (number of people) Reported harvest of clams (times per year) Reported purchase of clams (times per year) Reported sales of chocolate clams (times per year) Reported consumption of chocolate clams (times per 8 42 654 3.9 4.0 10.9 7.4 7.6 14 17 917 2.4 14.5 19.7 0.0 20.8 20 8 3924 2.0 0.4 17.7 0.0 19.1 year) 123 (cid:2) The Author(s) 2020 www.kva.se/en Ambio 2021, 50:586–600 593 Fig. 3 Survey participants have observed changes in Mexican chocolate clams, including in market demand, availability, price, quantity and/or quality, and size of individual clams. Percentages of survey participants who have observed these changes vary by type of change and participant group. Participant groups include Mexican nationals originally from Loreto (n = 8), Mexican nationals not originally from Loreto (n = 14), and foreign nationals (n = 20). The most highly cited changes were in market demand and availability of clams, followed by price, quantity and/or quality, and size of clams Fig. 4 Survey participants who provided qualitative descriptions of changes observed in Mexican chocolate clams largely agreed on the directionality of change. Participants who provided information on the nature of changes observed included Mexican nationals originally from Loreto (n = 8), Mexican nationals not originally from Loreto (n = 14), and foreign nationals (n = 20) the three participant groups. Participants largely agreed on the directionality of changes (Fig. 4), and reported that demand for and price of clams had increased over time, while the availability, quantity and/or quality, and average size had decreased. Despite the fact that most participants had noticed qualitative changes related to the market demand, quantity, quality, size, price, and/or availability of Mexican choco- late clams, none of the participants in any group said that the changes they had observed had directly affected their household. When asked whether they had any thoughts on why these changes had occurred, participant responses fell into four main categories, in the order of most to least (cid:2) The Author(s) 2020 www.kva.se/en 123 594 Ambio 2021, 50:586–600 Table 3 Participant perspectives on why changes that have occurred fell into four primary categories, in order of most to least cited: fisheries management, overfishing, increased demand, and environmental change Do you have any thoughts on why these changes have occurred? Response category Times cited Illustrative quote(s) Fisheries management Overfishing Increased demand Environmental change 9 9 4 3 ‘‘It’s because of poor management of the clam’’, ‘‘It’s because of the cooperatives that use a compressor to harvest’’ ‘‘The uncontrolled exploitation’’ ‘‘It’s a tourist town, and this is the dish that represents our town’’; ‘‘There is more consumption now’’; ‘‘Supply and demand- there are more people in Loreto now’’ ‘‘The temperature—sometimes it’s too warm’’ cited: fisheries management, overfishing, demand, and environmental change (Table 3). increased Ecosystem service values All but one ecosystem service value assessed was reported to be provided by chocolate clams to survey participants: this was personal economic value (assessed with the statement, ‘‘Chocolate clams provide income to my household’’). This is consistent with the lack of reported income from clamming among those surveyed. Relatedly, none of the participants originally from Loreto, 7% of participants from elsewhere in Mexico, and none of the foreign participants reported that their household receives life-sustaining value from Mexican chocolate clams (assessed with the statement, ‘‘Chocolate clams help sus- tain me and my family’’), and 33% of Loretano partici- pants, 14% of Mexican participants not originally from Loreto, and none of the foreign participants reported receiving subsistence value (assessed with the statement, ‘‘Chocolate clams provide some of my family’s basic needs’’). However, several participants noted that while their household does not receive life-sustaining or subsis- tence value from Mexican chocolate clams, many other households in the community do. In fact, all participants originally from Loreto, all Mexican participants not origi- nally from Loreto, and 90% of foreign participants agreed that Mexican chocolate clams are important to the com- munity of Loreto (assessed with the statement, ‘‘Chocolate clams are important to my community’’). Participants also agreed that chocolate clams help to shape the community identity of Loreto; 100% of Loretano participants, 79% of Mexican participants not originally from Loreto, and 60% of foreign participants agreed with the statement, ‘‘Cho- colate clams are an important part of what it means to be a Loretano or to live in this area.’’ Perhaps unsurprisingly, more participants originally from Loreto than participants from elsewhere felt that the clam also played a role in shaping their individual identity; 50% of Loretano partic- ipants agreed with the statement, ‘‘Chocolate clams are an important part of who I am as an individual,’’ as compared to 7% of Mexican participants not originally from Loreto, and none of foreign participants. While participants surveyed reported that their house- receive economic value from Mexican holds do not (0% agreement with the statement, chocolate clams ‘‘Chocolate clams provide income to my household’’ across all three participant groups), nearly all agreed that the clams provide economic value to the community (100% of Loretano participants, 86% of Mexican participants not originally from Loreto, and 100% of foreign participants agreed with the statement, ‘‘Chocolate clams are important to the local economy’’). Additionally, nearly all partici- pants agreed that the clam contributes to local tourism (100% Loretano, 93% other Mexican participants, and 83% foreign participant agreement with the two statements, ‘‘Tourists spend money on chocolate clams when they visit Loreto’’ and ‘‘Chocolate clams are a tourist attraction of Loreto’’). Additional ecosystem services with high levels of agreement among survey participants include cultural value (100%, 100%, and 95% agreement among the three groups, respectively, with the statement, ‘‘Chocolate clams are important to the culture of this area’’), historic value in the region (100%, 93%, and 75% agreement among the three groups, respectively, with the statement, ‘‘Chocolate clams are important because of their history in this area’’), existence value (100%, 100%, and 80% agreement among the three groups, respectively, agreement with the state- ment ‘‘Even when I don’t use chocolate clams, I like to know they are there’’), and future community use value (100%, 86%, and 85% agreement among the three groups, respectively, agreement with the statement, ‘‘Chocolate clams should be conserved for future generations’’). A full report of values assessed and responses for each participant group can be found in Fig. 5. 123 (cid:2) The Author(s) 2020 www.kva.se/en Ambio 2021, 50:586–600 595 Fig. 5 Ecosystem services with the highest levels of agreement among participants across participant groups include general community value, economic community value, cultural value, and tourism value. Ecosystem services with the lowest levels of agreement among participants across groups include life-sustaining household, economic household, and subsistence values. For values with two corresponding statements in surveys (tourism, scientific and learning, recreation, and aesthetic values), response percentage represents the average response for the two statements. For intrinsic value, which was reverse-coded, responses have been reversed for ease of comparison with other values. Participant groups include Mexican nationals originally from Loreto (n = 8), Mexican nationals from elsewhere (n = 14), and foreign nationals (n = 20) (cid:2) The Author(s) 2020 www.kva.se/en 123 596 DISCUSSION Mexican chocolate clams provide a host of ecosystem services to households in the Loreto region that include both provisioning and cultural services. As bivalves, the clams also provide regulating services in the form of water filtration (Millennium Ecosystem Assessment 2005). The multitude of ecosystem services provided by the clams are not explicitly recognized in fisheries management; current management focuses on ecological and economic factors. We find that in addition to the provisioning services pro- duced by the fishery, households in the Loreto region derive many cultural ecosystem services from Mexican chocolate clams. Community members agree that in addi- tion to economic value generated by the Mexican chocolate clam fishery, this species also contributes to tourism, sci- entific/learning, recreation, aesthetic, historic, cultural, community identity, and existence values. This finding is consistent with other ecosystem service valuation studies that have found that high percentages of local stakeholders recognize their local ecosystems’ capacity to produce diverse ecosystem services including social and cultural values (Martı´n-Lo´pez et al. 2012; Oteros-Rozas et al. 2014). Community members report receiving many types of ecosystem services from the species, which supports our hypothesis that Mexican chocolate clams provide a diver- sity of both provisioning and cultural values to the com- munity of Loreto. None of the participants in the survey reported relying on income from clam harvest, yet nearly half of all participants indicated that they have collected Mexican chocolate clams at some point in the past, and a third said that they collect clams at least once per year. This indicates that residents of Loreto who are not fishers also participate in the harvest of Mexican chocolate clams, and that the fishery itself is much more heterogeneous than accounted for by current fisheries management. This find- ing is supported by the previous work demonstrating that multiple fisher types harvest Mexican chocolate clams in Loreto, and that marginalized fisher groups are excluded from fisheries management processes (Pellowe and Leslie 2019). Fisheries management decisions have consequences not only for fishers directly engaged in resource extraction, but also for the broader coastal community. In communities like Loreto, where relatively few individuals engage in regular harvest of the Mexican chocolate clam as com- mercial fishers (Pellowe and Leslie 2019), the values pro- vided by the species to the broader coastal community are diverse and significant. Accounting for diverse ecosystem services and community perspectives in management requires first, identifying the values and aims of the com- munity, and then, creating management that accounts for trade-offs and conflicts among multiple priorities (Loomis Ambio 2021, 50:586–600 and Paterson 2014). Fisheries management in Baja Cali- fornia Sur is improving in its ability to integrate the heterogeneity of ecological systems into policies, but the sociocultural richness of fisheries systems and coastal communities remains largely unaccounted for (see for example, Finkbeiner and Basurto 2015; Leslie et al. 2015). An ecosystem service assessment like what we present here can help inform ecosystem-based management that better incorporates and McLeod 2005). sociocultural (Rosenberg richness Cultural ecosystem services underpin stakeholders’ values and preferences (Russell et al. 2013). However, translating ecosystem service assessments into policy has many challenges, including reconciling the legitimacy of diverse knowledge types, and finding pathways to turn such knowledge into action (Posner et al. 2016). The purposive inclusion of cultural ecosystem services in these broader assessments is one way to ensure that the sociocultural richness of human–nature interactions as well as the knowledge and values of diverse stakeholders are incor- porated into management (Chan et al. 2012a; Loomis and Paterson 2014; Scholte et al. 2015). Previous work assessing diverse ecosystem services for management has largely focused on terrestrial environments (e.g., Martı´n- Lo´pez et al. 2012; Oteros-Rozas et al. 2014; Dickinson and Hobbs 2017), but there is growing interest in the utility of such approaches for integrating diverse values into the management of marine systems (Rees et al. 2010; Klain and Chan 2012; Loomis and Paterson 2014; Gregr et al. 2020). Stakeholders’ perceptions of change also provide valu- able information about changes in the delivery of benefits that can help to identify management priorities (Martı´n- Lo´pez et al. 2012, 2013; Oteros-Rozas et al. 2014). In this study, perceptions of change provide important insight into how community members may make decisions regarding the clams and resulting marine conservation outcomes, since stakeholder perceptions of ecological conditions underpin environmental behavior (Gobster et al. 2007). Stakeholders’ perceptions of change have been important to understand temporal shifts in other marine populations and ecosystems in the Gulf of California (Sala et al. 2004; Sa´enz-Arroyo et al. 2005a, b, 2006). A study of fisher perceptions of trends in the abundance of the Gulf grouper (Mycteroperca jordani) revealed dramatic declines in abundance that occurred prior to the collection of fisheries data in the Gulf of California, and were thus unaccounted for in fisheries management (Sa´enz-Arroyo et al. 2005b). Alongside fisheries statistics and surveys, fishers’ obser- vations of change over time have also revealed shifts in the species composition of coastal ecosystems of the Gulf of California, from mostly large, long-lived species in higher trophic levels to mostly small, short-lived species in lower 123 (cid:2) The Author(s) 2020 www.kva.se/en Ambio 2021, 50:586–600 597 trophic levels (Sala et al. 2004). Perceptions of change in marine environments are particularly valuable where long- term monitoring data are not available, as they contribute critical information for setting appropriate management targets (Sala et al. 2004). reduced availability, Participants in this study reported changes in Mexican chocolate clams over time in the form of increased market demand, higher prices, reduced quantity and quality, and smaller size. Changes were reported at higher rates by Mexican nationals than foreign nationals surveyed, perhaps because the Mexican nationals surveyed had lived in Loreto longer and had more years of experience harvesting and buying clams. Observed changes in demand, price, availability, quantity, quality, and size of clams affect the delivery of ecosystem services and reveal potential priorities for management. Survey participants proposed several possible causes of observed changes increased including fisheries management, overfishing, demand for chocolate clams, and environmental change. Although survey participants were predominantly non- fishers, the nature of their observations of change and attributed causes of change echo those reported by har- vesters of the Mexican chocolate clam in previous studies (Pellowe and Leslie 2019). Harvesters reported declines in Mexican chocolate clam populations over time, which they attributed to increased fishing effort resulting from changes in fisheries management (Pellowe and Leslie 2019). Stakeholders’ observations of change provide information on potential shifts in clam populations and the ecosystem services they generate that is critical for effective design and implementation of management strategies. Such stud- ies are particularly important in data-limited fisheries, like the Mexican chocolate clam fishery (Pellowe and Leslie 2020), where long-term abundance data may not be available. While survey participants did not feel acutely impacted by the changes they had observed, they believed other households in their community were affected. Similarly, community members we surveyed acknowledged the importance of the services provided by Mexican chocolate clams to the broader community of Loreto, even if they themselves did not feel that they received every service. Survey participants were more likely to report the delivery of both provisioning and cultural ecosystem services at the community level, especially for the values of general importance, life-sustaining value, economic value, future use value, and identity value, than they were to report the delivery of the same services at the individual or household level. Community members in Loreto recognize the com- munity value of the Mexican chocolate clam and the impacts of change on the delivery of ecosystem services at the community level. Of the values assessed in this study, the most important ecosystem services that the Mexican chocolate clam pro- vides to the community of Loreto include economic, tourism, future use, cultural, and existence values. Many locals recall childhood memories of collecting Mexican chocolate clams during family trips to the beach, learning to dig for clams in the sand with their toes, or holding their breath to grab a clam from the ocean floor (Pellowe unpublished data). Survey participants originally from Loreto were more likely to agree that the clam contributes to their individual identity than participants from else- where. However, most participants surveyed, regardless of their place of origin, agreed that the clam is an important it mean to be a member of the Loreto part of what community. Considering the wide recognition of cultural ecosystem services provided to Loreto households, and the clam’s contribution to local identity, the Mexican chocolate clam may be considered a cultural keystone species. Cultural keystone species are ‘‘culturally salient species that shape in a major way the cultural identity of a people’’ (Garibaldi and Turner 2004, p. 4). Such species are defined by the key role they play in defining cultural identity and are char- acterized by their high cultural significance. Cultural key- stone species are also marked by their provision of important ecosystem services, particularly cultural value (Butler et al. 2012). The concept of the cultural keystone species highlights the importance of communities’ rela- tionship to place, and the conservation status of these species may be a starting point for identifying management priorities (Garibaldi and Turner 2004). In the Torres Strait Islands in Australia, two cultural keystone species, turtles and dugongs, were catalysts for a shift towards adaptive co- management, which involves the formal sharing of power between local stakeholders and regional fisheries man- agers, and the formal local ecological knowledge into resource governance (Butler et al. 2012). the value of cultural keystone Their findings highlight species as catalysts for the integration of local knowledge into marine resource governance to enhance fisheries pol- icy and protect the future delivery of ecosystem services (Butler et al. 2012). integration of In Loreto, embracing Mexican chocolate clams as a cultural keystone species may facilitate greater community participation in marine resource management decisions that is reflective of the heterogeneity among those involved in the fishery, both directly and indirectly. It may also result in the integration of local ecological knowledge into future policy decisions including accounting for community and fisher observations of change and investigating possible causes of change in order to identify management priori- ties. Managing for Mexican chocolate clams’ diverse val- include protecting habitat, regulating water ues might (cid:2) The Author(s) 2020 www.kva.se/en 123 598 Ambio 2021, 50:586–600 quality, and privileging low-impact fishing practices to safeguard the future delivery of both provisioning and cultural ecosystem services. These practices would serve not only to conserve Mexican chocolate clams and the benefits they provide to Loreto households, but would also benefit many other marine species in Loreto’s nearshore waters including fish, rays, octopus, and other molluscs. While this study provides important insights about community members’ perceptions of change in the clam fishery and the provisioning and cultural ecosystem ser- vices that Mexican clams provide to households in the Loreto region, a larger sample size of survey participants would be needed to generalize our findings to the broader population of Loreto residents. A future study with a larger number of participants, systematically recruited to ensure representativeness of the socioeconomic makeup of Loreto households, could confirm whether our findings apply more broadly to the general population. Our participant pool consisted of many households of middle and upper socioeconomic status owing to the fact that nearly half of the participants surveyed were expats and nationals of the United States, Canada, and the European Union. The skewed socioeconomic characteristic of the participant pool in this study is a result of the purposive sampling method used to recruit participants. Future work should include surveys conducted with a more representative and wider participant pool in order to verify whether our findings hold true for Loreto residents more broadly. To investigate further the clam’s role in shaping community the identity in Loreto as a cultural keystone species, inclusion of more participants originally from Loreto should be prioritized in future work. The participants in this study did not rely on clams as a source of income, sustenance, or other basic needs, and we anticipate that the inclusion of more low-income households would lead to higher reporting of these services. Additionally, future work should include an expansion of the range of responses to value statements, in order to facilitate comparisons in the strength of participant response to different values. The three-item Likert scale employed in this study to assess survey participants’ agreement or disagreement with ecosystem service value statements could be expanded to a Likert scale that includes a greater range of degrees of agreement and disagreement. This would produce a richer understanding of participants’ experience of diverse values, as well as the relative importance of provisioning and cultural ecosystem services. The social and cultural values of species and ecosystems shape human–nature interactions, yet are often overlooked in decision-making and design of marine management (Chan et al. 2011). If such values are not explicitly understood and accounted for, they are likely to be poorly represented in natural resource policy (Klain and Chan 2012). Assessing these values and incorporating them into management creates robust policies that protect the future provision of valuable ecosystem services. Managing for a narrow set of ecosystem services may not only ignore other important values that a species or ecosystem provides to human communities, but can also reduce the fishery’s capacity to cope with future disturbance (Gordon et al. 2008; Bennett et al. 2009). Understanding the full suite of ecosystem services provided by fished species is a critical step in designing resource management that protects cru- cial benefits, while considering trade-offs among the diverse values and priorities of coastal communities. Acknowledgements We thank Yong Chen, Carla Guenther, Joshua Stoll, and Bridie McGreavy for their feedback on earlier conceptu- alizations of this paper. We thank Linda Ramirez and Eduardo Murillo for their help reviewing, translating, and pretesting survey questionnaires. We also thank Alfredo Baeza for logistical assistance in the field, and the community members of the Loreto region for their participation in this study. We thank Melissa Britsch for her helpful feedback, which helped to improve the clarity of the paper. We also thank two anonymous reviewers for their thoughtful and constructive feedback, which significantly improved the paper. Funding was pro- vided by the US National Science Foundation (Grant Number DEB 1632648 to HL). Funding Open access funding provided by Stockholm University. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons. org/licenses/by/4.0/. REFERENCES Basurto, X., S. Gelcich, and E. Ostrom. 2013. The social–ecological system framework as a knowledge classificatory system for benthic small-scale fisheries. Global Environmental Change 23: 1366–1380. https://doi.org/10.1016/j.gloenvcha.2013.08.001. Be´ne´, C. 2006. Small-scale fisheries: Assessing their contribution to rural livelihoods in developing countries. FAO Fisheries Circu- lar No. 1008: 57. Bennett, E.M., G.D. Peterson, and L.J. Gordon. 2009. Understanding relationships among multiple ecosystem services. Ecology Letters 12: 1394–1404. https://doi.org/10.1111/j.1461-0248. 2009.01387.x. Butler, J.R.A., A. Tawake, T. Skewes, L. Tawake, and V. McGrath. 2012. Integrating traditional ecological knowledge and fisheries management in the Torres Strait, Australia: The catalytic role of 123 (cid:2) The Author(s) 2020 www.kva.se/en Ambio 2021, 50:586–600 599 turtles and dugong as cultural keystone species. Ecology and Society 17: 34. https://doi.org/10.5751/ES-05165-170434. Chan, K.M.A., J. Goldstein, T. Satterfield, N. Hannahs, K. Kikiloi, R. Naidoo, N. Vadeboncoeur, and U. Woodside. 2011. Cultural services and non-use values. In Natural capital: Theory and practice of mapping ecosystem services, ed. P. Kareiva, H. Tallis, T.H. Ricketts, G.C. Daily, and S. Polasky, 206–228. Oxford: Oxford University Press. Chan, K.M.A., T. Satterfield, and J. Goldstein. 2012a. Rethinking ecosystem services to better address and navigate cultural values. Ecological Economics 74: 8–18. https://doi.org/10.1016/j. ecolecon.2011.11.011. Chan, K.M.A., A.D. Guerry, P. Balvanera, and S. Klain. 2012b. in ecosystem services? A engagement. BioScience 62: Where are cultural and social framework for 744–756. constructive Comisio´n Nacional de A´ reas Naturales Protegidas. 2019. Parque Nacional Bahı´a de Loreto. Gobierno de Me´xico. Daily, G.C., T. So¨derqvist, A. Aniyar, K. Arrow, P. Dasgupta, P.R. Ehrlich, C. Folke, A. Jansson, et al. 2000. The value of nature and the nature of value. Science 289: 395–396. Dickinson, D.C., and R.J. Hobbs. 2017. Cultural ecosystem services: Characteristics, challenges and lessons for urban green space research. Ecosystem Services 25: 179–194. https://doi.org/10. 1016/j.ecoser.2017.04.014. Dunnington, D. 2020. ggspatial: Spatial data framework for ggplot2. R package version 1.1.4. Finkbeiner, E.M., and X. Basurto. 2015. Re-defining co-management to facilitate small-scale fisheries reform: An illustration from northwest Mexico. Marine Policy 51: 433–441. https://doi.org/ 10.1016/j.marpol.2014.10.010. Garibaldi, A., and N. Turner. 2004. Cultural keystone species: Implications for ecological conservation and restoration. Ecol- ogy and Society 9: 1. https://doi.org/10.1146/annurev-pharmtox- 061008-103038. Gobster, P.H., J.I. Nassauer, T.C. Daniel, and G. Fry. 2007. The shared landscape: What does aesthetics have to do with ecology? Landscape Ecology 22: 959–972. Gordon, L.J., G.D. Peterson, and E.M. Bennett. 2008. Agricultural modifications of hydrological flows create ecological surprises. Trends in Ecology and Evolution 23: 211–219. Gregr, E.J., V. Christensen, L. Nichol, R.G. Martone, R.W. Markel, J.C. Watson, C.D.G. Harley, E.A. Pakhomov, et al. 2020. Cascading social–ecological costs and benefits triggered by a recovering keystone predator. Science 368: 1243–1247. https:// doi.org/10.1126/science.aay5342. Groves, R.M., F.J. Fowler, M.P. Couper, J.M. Lepkowski, E. Singer, and R. Tourangeau. 2011. Survey methodology, 2nd ed. Hobo- ken: Wiley. Guo, Z., L. Zhang, and Y. Li. 2010. Increased dependence of humans on ecosystem services and biodiversity. PLoS ONE 5: e13113. Halpern, B.S., C. Longo, D. Hardy, K.L. McLeod, J.F. Samhouri, S.K. Katona, K. Kleisner, S.E. Lester, et al. 2012. An index to assess the health and benefits of the global ocean. Nature 488: 615–620. https://doi.org/10.1038/nature11397. Herna´ndez-Morcillo, M., T. Plieninger, and C. Bieling. 2013. An empirical review of cultural ecosystem service indicators. Ecological Indicators 29: 434–444. https://doi.org/10.1016/j. ecolind.2013.01.013. Instituto Nacional de Estadı´stica y Geografı´a, INEGI. 2017. Anuario Estadı´stico y Geogra´fico de Baja California Sur 2017. https:// doi.org/10.7283/T5MW2F2D. Johnson, D. 2018. The values of small-scale fisheries. In Social wellbeing and the values of small-scale fisheries, ed. D.S. Johnson, T.G. Acott, N. Stacey, and J. Urquhart, 1–21. Amsterdam: Springer. Karthik, R., and H. Wickham. 2018. wesanderson: A Wes Anderson Palette Generator. R package version 0.3.6. Keen, A.M. 1971. Sea shells of tropical West America: Marine mollusks from Baja California to Peru. Stanford: Stanford University Press. Klain, S., and K.M.A. Chan. 2012. Navigating coastal values: Participatory mapping of ecosystem services for spatial plan- ning. Ecological Economics 82: 104–113. Leslie, H.M., X. Basurto, M. Nenadovic, L. Sievanen, K.C. Cavanaugh, J.J. Cota-Nieto, B.E. Erisman, E. Finkbeiner, et al. 2015. Operationalizing the social–ecological systems framework to assess sustainability. Proceedings of the National Academy of the United States of America 112: 5979–5984. Sciences of https://doi.org/10.1073/pnas.1414640112. Likert, R. 1932. A technique for the measurement of attitudes. Archives of Psychology 140: 1–55. Loomis, D.K., and S.K. Paterson. 2014. The human dimensions of coastal ecosystem services: Managing for social values. Ecolog- ical Indicators 44: 6–10. https://doi.org/10.1016/j.ecolind.2013. 09.035. Martı´n-Lo´pez, B., I. Iniesta-Arandia, M. Garcı´a-Llorente, I. Palomo, I. Casado-Arzuaga, D.G. Del Amo, E. Go´mez-Baggethun, E. Oteros-Rozas, et al. 2012. Uncovering ecosystem service ONE 7: PLoS bundles e38970. https://doi.org/10.1371/journal.pone.0038970. preferences. through social Martı´n-Lo´pez, B., E. Go´mez-Baggethun, and M. Garcı´a-Llorente. 2013. Trade-offs across value domains in ecosystem services assessment. Ecological Indicators 37: 220–228. Millennium Ecosystem Assessment. 2005. Millennium Ecosystem Assessment: Ecosystems and human well-being. World Health. https://doi.org/10.1196/annals.1439.003. Oteros-Rozas, E., B. Martı´n-Lo´pez, J.A. Gonza´lez, T. Plieninger, C.A. Lo´pez, and C. Montes. 2014. Socio-cultural valuation of ecosystem services in a transhumance social–ecological net- work. Regional Environmental Change 14: 1269–1289. https:// doi.org/10.1007/s10113-013-0571-y. Partelow, S., and C. Boda. 2015. A modified diagnostic social– ecological system framework for lobster fisheries: Case imple- mentation and sustainability assessment in Southern California. Ocean and Coastal Management 114: 204–217. https://doi.org/ 10.1016/j.ocecoaman.2015.06.022. Pellowe, K.E., and H.M. Leslie. 2017. Seasonal variability shapes resilience of small-scale fisheries in Baja California Sur, Mexico. PLoS ONE 12: 1–15. https://doi.org/10.1371/journal.pone. 0182200. Pellowe, K.E., and H.M. Leslie. 2019. Heterogeneity among clam harvesters in northwest Mexico shapes individual adaptive capacity. Ecology and Society 24: 25. https://doi.org/10.5751/ ES-11297-240425. Pellowe, K.E., and H.M. Leslie. 2020. Size-selective fishing leads to trade-offs between fishery productivity and reproductive capac- ity. Ecosphere 11: e03071. https://doi.org/10.1002/ecs2.3071. Posner, S.M., E. McKenzie, and T.H. Ricketts. 2016. Policy impacts of ecosystem services knowledge. Proceedings of the National Academy of Sciences of USA 113: 1760–1765. https://doi.org/10. 1073/pnas.1502452113. R Core Team. 2020. R: A language and environment for statistical computing. Vienna: R Foundation for Statistical Computing. Raymond, C.M., and G. Brown. 2006. A method for assessing protected area allocations using a typology of landscape values. Journal of Environmental Planning and Management 49: 797–812. Reed, P., and G. Brown. 2003. Values suitability analysis: a methodology for identifying and integrating public perceptions of forest ecosystem values in national forest planning. Journal of Environmental Planning and Management 46: 643–658. (cid:2) The Author(s) 2020 www.kva.se/en 123 600 Ambio 2021, 50:586–600 Rees, S.E., L.D. Rodwell, M.J. Attrill, M.C. Austen, and S.C. Mangi. 2010. The value of marine biodiversity to the leisure and recreation industry and its application to marine spatial planning. Marine Policy 34: 868–875. https://doi.org/10.1016/j.marpol. 2010.01.009. Rolston, H., and J. Coufal. 1991. A forest ethic and multivalue forest management. Journal of Forestry 89: 35–40. Rosenberg, A.A., and K.L. McLeod. 2005. Implementing ecosystem- based approaches to management the conservation of ecosystem services. Marine Ecology Progress Series 300: 270–274. for Russell, R., A.D. Guerry, P. Balvanera, R.K. Gould, X. Basurto, K.M.A. Chan, S. Klain, J. Levine, et al. 2013. Humans and nature: How knowing and experiencing nature affect well-being. Annual Review of Environment and Resources 38: 473–502. Sa´enz-Arroyo, A., C.M. Roberts, J. Torre, M. Carin˜o-Olvera, and R.R. Enrı´quez-Andrade. 2005a. Rapidly shifting environmental baselines among fishers of the Gulf of California. Proceedings of the Royal Society B: Biological Sciences 272: 1957–1962. https://doi.org/10.1098/rspb.2005.3175. Sa´enz-Arroyo, A., C. Roberts, J. Torres, and M. Carin˜o-Olvera. 2005b. Using fishers’ anecdotes, naturalists observations and grey literature to reassess marine species at risk: The case of the Gulf grouper in the Gulf of California, Mexico. Fish and Fisheries 6: 121–133. Sa´enz-Arroyo, A., C.M. Roberts, J. Torre, M. Carin˜o-Olvera, and J.P. Hawkins. 2006. The value of evidence about past abundance: Marine fauna of the Gulf of California through the eyes of 16th to 19th century travellers. Fish and Fisheries 7: 128–146. https:// doi.org/10.1111/j.1467-2979.2006.00214.x. Sala, E., O. Aburto-Oropeza, M. Reza, G. Paredes, and L.G. Lo´pez- Lemus. 2004. Fishing down coastal food webs in the Gulf of California. Fisheries 29: 19–25. https://doi.org/10.1577/1548- 8446(2004)29[19:FDCFWI]2.0.CO;2. Scholte, S.S.K., A.J.A. van Teeffelen, and P.H. Verburg. 2015. Integrating socio-cultural perspectives into ecosystem service valuation: A review of concepts and methods. Ecological Economics 114: 67–78. https://doi.org/10.1016/j.ecolecon.2015. 03.007. Smith, H., and X. Basurto. 2019. Defining small-scale fisheries and examining the role of science in shaping perceptions of who and what counts: A systematic review. Frontiers in Marine Science 6: 236. https://doi.org/10.3389/fmars.2019.00236. South, A. 2017. rnaturalearth: World Map Data from Natural Earth. R package version 0.1.0. St. Martin, K., B.J. McCay, G.D. Murray, T.R. Johnson, and B. Oles. 2007K. Communities, knowledge and fisheries of the future. International Journal of Global Environmental Issues 7: 221–239. Turner, R.K., and G.C. Daily. 2008. The ecosystem services framework and natural capital conservation. Environmental and Resource Economics 39: 25–35. Villasante, S., G. Macho, M. Antelo, D. Rodrı´guez-Gonza´lez, and M.J. Kaiser. 2013. Resilience and challenges of marine social– ecological systems under complex and interconnected drivers. Ambio 42: 905–909. https://doi.org/10.1007/s13280-013-0450-2. Wickham, H. 2016. ggplot2: Elegant graphics for data analysis. New York: Springer. https://ggplot2.tidyverse.org. Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. AUTHOR BIOGRAPHIES Kara E. Pellowe (&) is a Postdoctoral Fellow at the Stockholm Resilience Centre. Her research interests include sustainability sci- ence, marine conservation, and the dynamics and resilience of marine social–ecological systems. Address: Stockholm Resilience Centre, Stockholm University, Kra¨f- triket 2B, 106 91 Stockholm, Sweden. Address: Darling Marine Center, University of Maine, 193 Clarks Cove Road, Walpole, ME 04573, USA. e-mail: kara.pellowe@su.se Heather M. Leslie is Director of the Darling Marine Center and Associate Professor at the University of Maine. Her research interests include marine conservation science, and the ecology, policy, and management of coastal marine ecosystems. Address: Darling Marine Center, University of Maine, 193 Clarks Cove Road, Walpole, ME 04573, USA. Address: School of Marine Sciences, University of Maine, Orono, MA 04469, USA. e-mail: heather.leslie@maine.edu 123 (cid:2) The Author(s) 2020 www.kva.se/en
null
10.1088_1361-6595_ad03bd.pdf
Data availability statement The data cannot be made publicly available upon publication because no suitable repository exists for hosting data in this field of study. The data that support the findings of this study are available upon reasonable request from the authors.
Data availability statement The data cannot be made publicly available upon publication because no suitable repository exists for hosting data in this field of study. The data that support the findings of this study are available upon reasonable request from the authors.
Plasma Sources Sci. Technol. 32 (2023) 105015 (12pp) Plasma Sources Science and Technology https://doi.org/10.1088/1361-6595/ad03bd The role of recombination in constriction of a positive column of a glow discharge in inert gases A V Siasko ∗, V Yu Karasev and Yu B Golubovskii Faculty of Physics, St Petersburg State University, Ulianovskaia ul. 3, St Petersburg 198504, Russia E-mail: aleksei.siasko@gmail.com Received 30 August 2023, revised 5 October 2023 Accepted for publication 16 October 2023 Published 26 October 2023 Abstract The work is aimed at the experimental determination of the role of volume recombination in a positive column of a DC discharge in helium during the transition from a diffuse homogeneous state to a constricted stratified regime under real discharge conditions of high electron and gas ≈ 2 eV, Tg > 1000 K). The investigation is based on the probe measurements temperatures (Te of the wall current determined by the flux of charged particles towards the boundary of a cylindrical discharge tube. The experiments were carried out in helium, neon, and argon in the range of reduced pressures 1.2–300 Torr·cm. In heavy inert gases, the transition to a constricted regime is determined by the active loss of charged particles in the volume. In contrast to neon and argon, experiments in helium demonstrated that the role of volume recombination is insignificant during the transition to a constricted regime. The rate of volume losses in helium in real conditions is very low compared to neon and argon. The obtained results allow one to calculate the volume recombination rate by comparing the experimentally measured wall currents with the corresponding numerical calculations within the collision-radiative model. Keywords: glow discharge, positive column, constriction, optical constriction, plasma instabilities, recombination, probe diagnostics 1. Introduction The phenomenon of constriction—compression of plasma glow into a narrow filament near the discharge axis, which is often accompanied by the development of striations, has been described in numerous works beginning with the book of Stark [1]. The modern state of the problem of discharge constriction is described in classical experimental and the- oretical studies [2–21]. At present, the problem of plasma filamentation at intermediate and high pressures has not lost its relevance and there are several modern both exper- imental and theoretical works dedicated to plasma instabil- ities in sources of various types. For example, the paper [22] describes the development of striation-type ionization ∗ Author to whom any correspondence should be addressed. instability based on fluid simulation of a dielectric barrier dis- charge (DBD) discharge in argon at atmospheric pressure. The plasma parameters mimic the natural discharge condi- tions under which discharge stratification was experiment- ally observed in [23]. The development of instability is inter- preted as a disturbance of the spatial distribution of elec- trons along the discharge channel due to repeated stepwise ionization processes through metastable levels and ionization of excimers. Simultaneous constriction and stratification of a short discharge with an electrode spacing of a few millimeters were experimentally studied in argon and in helium with an admixture of nitrogen at pressures of hundreds of torr in [24, 25]. In [26], the constriction observed in a microwave discharge in molecular nitrogen at sub-atmospheric pressure was interpreted as an ionization-thermal instability due to a mixing of molecular states of nitrogen during rapid heating of the gas. A combined experimental and numerical study 1361-6595/23/105015+12$33.00 Printed in the UK 1 © 2023 IOP Publishing Ltd Plasma Sources Sci. Technol. 32 (2023) 105015 A V Siasko et al Figure 1. (a) Schottky linear diffusion theory (r∗ = R radiation compression or optical constriction. (c) Non-linear diffusion—recombination theory (r∗ < R formation of a plasma filament. ∗ = R/2.4), diffuse discharge. (b) Non-linear diffusion theory (r∗ < R; R ∗ < R/2.4), current constriction, ∗ ≈ R/2.4), −5–10 of microwave CO2 discharge [27] demonstrates the transition from a homogeneous state to a constricted state with an increase in pressure from 60 to 300 mbar due to a thermal- ionization mechanism at gas temperatures reaching 6500 K, −4. Along and the ionization degrees of the order of 10 with high pressures, the work on studying the instabilities is being carried out in low-pressure discharges. The development of S, P, and R ionization waves at low currents is described using a hybrid kinetic-fluid model in paper [28] and a fully kin- etic model in paper [29]. The simulation presented in the work [30] predicts the development of striations in the dusty plasma of the PK-4 system operating in micro-gravity on-board of the International Space Station. The obtained numerical results are confirmed by the experiments on the ground-based PK-4 rep- lica systems. To date, a certain point of view has been formed on the mechanism of discharge constriction. The initial equation for describing the compression of the electron density is the ioniz- ation balance equation. In the simplest case, under conditions of intermediate pressures and currents (tens to hundreds of torr and milliamp), ionization in each elementary plasma volume is balanced by the diffusion of charged particles and the volume recombination. Assuming two-body recombination with the rate αn2, for cylindrical geometry this equation can be written as: I (r) + ∇Da ∇n (r) − αn2 (r) = 0. (1) I is the number of ionization acts per unit volume per unit time, Da is the ambipolar diffusion coefficient, α is the coefficient of two-body recombination, and r is the radial coordinate. If for some reason the ionization sources are compressed towards the discharge axis and located in a region with a characteristic size ∗ , where I = 0, one r can assume Da/R is the distance over which charged particles diffuse during the recombina- tion time 1/αn. If the recombination coefficient α has a large ∗ < R may take place. R is the dis- value, then the relation R , the so-called recombination charge tube radius. The size R length, determines the radius of the current flow or the radius , then outside the ionization zone r > r ∗2 ∼ αn or R αn . Here R ∗2 ∼ Da ∗ ∗ ∗ ∗ ∗ charged particles dis- of the plasma filament since beyond R appear due to recombination. In the opposite case, when the ∗ > R. This recombination rate is low, it may turn out that R means that charged particles diffuse up to the wall, ionization is balanced by the ambipolar diffusion, and the recombination . In this case, a length coincides with the diffusion radius R diffuse discharge mode is established. Within the framework of the linear diffusion Schottky theory [31], a Bessel distribu- tion of the ionization rate and electron density is formed in the radial direction, which is described by the zero-order Bessel function J0(2.4 · r/R). This corresponds to the diffusion radius ∗ = R/2.4 and to the same characteristic radius of the ioniz- R ∗ = R/2.4. In the non-linear diffusion theory, ation sources r when the ionization rate depends exponentially on the elec- tron density, the size of the excitation zone can be noticeably ∗ < R). In the absence of smaller than the radius of the tube (r recombination, the ambipolar diffusion broadens the electron density profile. Thus, the excitation and the ionization zones are compressed, but the electron density profile only slightly differs from the Bessel one. This phenomenon, the so-called optical constriction, is discussed in detail in sections 2 and 5.2. More detailed discussions of discharge formation modes are described in the review [32]. Figure 1 illustrates all three cases described above. Figure 1(a)—formation of a diffuse discharge in the absence of recombination when the ionization is balanced by the ambi- ∗ = R/2.4). Figure 1(b) describes the polar diffusion (r compression of the ionization zone and a discharge glow with a smooth distribution of the electron density along the radius in the absence of recombination. In this case, a line radiation fil- ament is formed with a smooth decay of the bremsstrahlung ∗ ≈ continuum radiation- the optical constriction (r R/2.4). The characteristic radii of compression zones are ∗ ≈ R/10 [32]. Figure 1(c) of the order of R shows the case of the formation of a constricted current chan- nel, when the ionization zone is compressed, and the loss of charged particles occurs in the volume due to active recom- ∗ < R/2.4). In this case, charged particles are bination (r carried out of the ionization zone by the ambipolar diffusion and then disappear in the volume due on a small distance r ∗ ≈ R/5 and r ∗ < R, R ∗ = R ∗ < R ∗ 2 Plasma Sources Sci. Technol. 32 (2023) 105015 A V Siasko et al Figure 2. (a) Cross-sections of elastic electron-atom collisions depending on the electron energy in helium, argon, and neon. (b) Excitation frequencies of the lower metastable states depending on the ionization degree for characteristic electric fields under conditions of transition from a diffuse state to the constricted regime: E/N = 0.56 Td in argon and neon and E/N = 4.5 Td in helium. to recombination. In this mode, the characteristic values of the diffusion zone and ionization zone correspond to values of the ∗ ≈ R/50 [32], which is a small part ∗ ≈ R/40 and r order of R of the radius of the discharge tube. The main mechanisms of constriction of ionization sources can be associated with kinetic effects [33–37], as well as with the inhomogeneous heating of the gas (thermal mechanism) [37–48]. In the first case, one considers the depletion of the electron distribution function by fast electrons in the direction from the axis toward the wall. At the center of the tube, the electron density is high, the electron–electron collisions are efficient, and the distribution function is close to the Maxwell function. With the distance from the axis, the electron dens- ity decreases, the intensity of electron–electron collisions also decreases, and the distribution function becomes depleted by fast electrons capable of ionizing atoms. As a result, the ion- ization rate depends exponentially on the plasma ionization degree, which, in turn, decreases in the radial direction and leads to the constriction of ionization sources towards the dis- charge axis. The degree of compression of the electron density depends on the shape of the cross-section of elastic electron- atom collisions. This effect is especially pronounced in argon and other heavy gases—krypton and xenon. In these gases, the cross-section of elastic collisions increases with the increas- ing electron energy. For this reason, the distribution function in these gases differs greatly from the Maxwell one. It is lead- ing to a formation of a sharp non-linear dependence on the ionization degree, which results in a formation of a very thin zone of ionization and excitation in these gases. In neon, the elastic collisions cross-section is approximately constant and the non-linearity is less pronounced. The excitation and ioniz- ation zone is wider than in heavy gases. For helium, the elastic collision cross-section in the region exceeding 4 eV decreases with increasing velocity. The decrease in the cross-section contributes to the Maxwellization of the distribution func- tion, the non-linearity is very weakly expressed. This effect in helium, in contrast to other inert gases, plays a weak role in the constriction mechanism. Figure 2 compares the cross sections of elastic electron-atom collisions (a) in argon, neon, and helium and the non-linearities of the excitation frequency (b) caused by the competition of electron–atom and electron– electron collisions for these gases at fixed electric field corres- ponding to real discharge conditions under the transition from a diffuse state to the constricted one. The second possible mechanism of constriction of ioniza- tion sources is associated with inhomogeneous heating of the gas, which leads to a decrease in the reduced electric field E/N in the radial direction. Due to inhomogeneous heating, the temperature of the neutral gas decreases from the center toward the wall, the density of neutrals N inversely increases, and the reduced field E/N decreases. Accordingly, the electron temperature and the ionization rate decrease, which depends exponentially on the electron temperature. In works [49, 50], basing on a detailed experimental and theoretical study, the mechanisms leading to constriction of a discharge in neon and argon were investigated. It is shown that for these gases the main cause of constriction is the kin- etic mechanism arising from the depletion of the distribution function by fast electrons. The inhomogeneous heating of the neutral gas is of secondary importance. These works show the effect of each of these mechanisms on the current–voltage characteristics, on the radial size of the current filament and the size of line and bremsstrahlung radiation zones, on a hysteresis during the transition of a discharge from diffuse to constricted state and vice versa, on the appearance of ionization waves during this transition, etc. Such a subtle effect as an influence of the transport of resonance radiation on a formation of mac- roscopic characteristics of diffuse and constricted discharges was considered [20]. For such inert gases as neon and argon, an adequate theoretical description of the phenomena observed in the experiment is associated with available precise data on the cross-sections of elementary processes, on the probabilit- ies of radiative transitions, and, mainly, with the reliable data on the values and temperature dependences of the recombin- ation coefficients. The latter fact is of the greatest import- ance since the volume recombination leads to the compression 3 Plasma Sources Sci. Technol. 32 (2023) 105015 A V Siasko et al of the current channel and the formation of a thin plasma filament (figure 1(c)). distributions of line and bremsstrahlung radiation in these gases. Latest experimental study [51] was carried out to reveal the main mechanism of constriction in helium. It was shown that in helium, in contrast to neon and argon, optical constriction is observed (figure 1(b)) - with increasing current, an abrupt compression of the ionization zone occurs while the electron density smoothly decreases along the radius of the discharge tube. Experiments have shown that the main mechanism of constriction in helium is associated with the inhomogeneous heating of the gas. In this gas, the role of volume recombin- ation losses of charged particles remains unclear. On the one hand, the absence of the formation of a current filament upon constriction in helium indicates an insignificant role of the volume loss of charged particles due to recombination pro- cesses compared to the transport of particles due to ambipolar diffusion. On the other hand, the rate of dissociative recom- bination in helium, which is the main channel for particle loss at high pressures, depends on the density of the excited vibrational states of the helium molecular ion. As the gas temperature grows, a strong non-linear increase in the density of molecules in highly excited states can be observed. The literature data on the rate of dissociative recombination for helium are contradictory. The spread of experimental data is two orders of magnitude [52–57]. The most detailed study of recombination processes in plasma afterglow is presented in [55]. According to the data of this work, the rate of dissociative recombination is very low, and the temperature dependence remains indefinite—α < 5 · 10 −10(Te/293)−(1±1) cm3 s −1. The present work aims to conduct an experimental study to reveal the role of recombination processes in the construc- tion of a discharge in helium at high electron and gas tem- ≈ 2 eV, Tg > 1000 K [51]) at the wide range of perature (Te the reduced gas pressures from 2 to 100 Torr·cm and reduced −1. For this purpose, discharge currents from 8 to 140 mA cm it is proposed to measure the ion flux towards the tube wall using the wall probes. The flux of ions reaching the wall due to ambipolar diffusion is the difference between the number of ions born through the ionization and disappeared as a result of recombination in the plasma volume. This makes it pos- sible to estimate the integral rate of recombination losses over the volume. To reveal additional differences between constric- tion in helium and heavier inert gases, similar experiments were carried out in argon, where the role of volume losses and the rate constant of dissociative recombination are well studied. To complete the picture of the observed phenom- ena, some results will be generalized to the investigations in neon. 2. Brief phenomenology of discharge constriction in inert gases Despite the constricted positive column in helium being visu- ally similar to the constricted discharges in argon and neon (figure 3), there is a fundamental difference in the radial 4 In neon and helium, a bright luminous filament of line radi- ation is surrounded by a weaker glow of the bremsstrahlung continuum. In argon, the filament is so thin that the glow of lines and continuum is visually indistinguishable. In addition, due to thermal effects, the plasma column in argon emerges due to large temperature gradients and convective heat fluxes. These phenomenon was studied in detail in [50]. The radial distribution of line radiation approximately describes the ionization zone. The radial distribution of the bremsstrahlung continuum describes the zone where the elec- tron density is located and, accordingly, the current flow zone. Figure 4 shows the radial distributions of line radiation and bremsstrahlung continuum under conditions of diffuse and constricted regimes in argon (a and c) and helium (b and d) [51]. As can be seen from figure 4, for argon in the constricted regime, there is a strong compression of both line (figure 4(a)) and bremsstrahlung radiation (figure 4(c)) and a pronounced plasma filament is formed. In contrast, in helium only line radi- ation (figure 4(b)) is compressed, while the bremsstrahlung continuum (figure 4(d)) smoothly decreases from the axis to the wall in both regimes. In constricted mode in helium cur- rent flows through the entire cross-section as in the diffuse mode. This phenomenon has been called the optical constric- tion (figure 1(b)). The radial profile of the continuum describes the distri- bution of electrons over the radius of the discharge tube up to temperature inhomogeneity. As will be shown below- equation (6), the slope of the continuum near the tube radius describes the flux of charged particles toward the wall. From figure 4(c) it can be seen that for argon the slope of the con- tinuum sharply decreases during the transition to the con- striction regime. This indicates that in the constricted regime in argon charged particles almost do not reach the wall. In helium, the slope of the continuum almost coincides in both diffuse and constricted regimes, which indicates that the volume losses are low. Charged particles reach the wall in the ambipolar diffusion regime. It was shown in [51] that the formation of the constricted discharge in helium is strongly influenced by heat exchange conditions. When the tube walls are thermostated at room temperature, constriction is not observed in the investigated range of pressures and currents. In the regime of free heat exchange with the environment, an abrupt constriction of line radiation is observed, while the bremsstrahlung continuum does not experience noticeable compression. The transition is accompanied by a hysteresis, which is especially clearly seen in the current–voltage characteristics. By heating the walls of the discharge tube, it is possible to induce constric- tion at currents less than the critical ones in the free heat exchange regime. When the walls of the discharge tube are cooled, the opposite effect is observed—the radiation constric- tion disappears. In other inert gases, similar effects are not observed. Plasma Sources Sci. Technol. 32 (2023) 105015 A V Siasko et al Figure 3. Photographs of constricted discharges in argon at pR = 200 Torr·cm and i/R = 30 mA cm i/R = 47 mA cm −1 (b), and helium at pR = 200 Torr·cm and i/R = 100 mA cm −1 (c). −1 (a), neon at pR = 90 Torr·cm and Figure 4. Normalized radial profiles of spectral lines (a), (b) and bremsstrahlung continuum (c), (d) intensity in the diffuse (black lines) and constricted (red lines) states. Argon at pR = 96 Torr·cm. (a) - λ = 696.54 nm (c)- λ = 508. Helium at pR = 200 Torr·cm. (b) - λ = 587.56 nm, (d)- λ = 540 nm. 3. Experiment on the measurement of the wall currents To measure the ion fluxes toward the wall, discharge tubes with a radius of 1.2, 2.4, and 2.6 cm were used in the exper- iment. The tubes with a radius of 1.2 and 2.6 cm were made of molybdenum glass for studies in helium at low pressures and in argon. To work in helium at high pressures, a quartz tube with a radius of 2.4 cm was used. The use of a quartz tube is due to extremely high heating of the tube walls up to tem- peratures reaching the softening temperature of molybdenum glass. Using a forevacuum and diffusion pumps, the tubes were −6 Torr, which was registered pumped out to a pressure of 10 by the VIT-2 vacuum gauge, and filled with an inert gas up to the required pressure measured by a vacuum gauge which was calibrated by an oil pressure gauge. The discharge tubes were connected to two high-voltage DC power supply (figure 5) which had an output with positive and negative potentials, respectively. A ballast resistance was connected in series with the tube to establish a required discharge current. The exper- iments were carried out after multiple tube training by high currents and gas changes. The degree of purification was con- trolled by the emission spectrum. For reliable measurement of the wall currents, 5 mm diameter molybdenum and nickel probes were soldered into the tubes. The location and design of the probes are shown in figure 5. Some probes were addi- tionally equipped with a shielding ring (position 3 in figure 5) first proposed in [58]. To register the propagation of ionization waves in the form of striations, a photomultiplier (PMT) was placed across the discharge tube. The signal from the PMT was registered by an oscilloscope and then transferred to a PC. The electrical circuit to measure the probe characteristics is shown in figure 6. The value of the ion current towards the wall was obtained by linear extrapolation of the ion part of the probe characteristic to the potential of the isolated probe. Voltage from a stabilized power source, regulated by a poten- tiometer, was applied to the probe. The current from the probe was measured with a micro ammeter. Probes with a shield- ing ring were used to control the error in the measured wall 5 Plasma Sources Sci. Technol. 32 (2023) 105015 A V Siasko et al Figure 5. Scheme of a discharge tube with probes for measurement of the wall currents. 1 - molybdenum probe, 2 - nickel probe, 3 - molybdenum probe with a shielding ring. Figure 7. Probe characteristics in argon at pR = 100 Torr·cm and i/R = 4 mA cm 3- molybdenum probe with a shielding ring. −1. 1- molybdenum probe, 2- nickel probe, Figure 6. Electrical circuit for measurement of the probe characteristics. currents associated with edge effects. An example of probe characteristics registered in argon on a probe of nickel, molyb- denum with and without a shielding ring is shown in figure 7. For clarity, the characteristic of the molybdenum probe with a shielding ring is shifted along the abscissa axis by −20 V. It follows from this figure that, under the experimental condi- tions, the characteristics for nickel and molybdenum probes coincide and give the same value of the wall currents. The characteristic of a probe with a shielding ring has a differ- ent slope due to edge effects. With an increase in the voltage applied to the probe which is negative relative to the potential of the isolated probe, the ion current should reach saturation, while in a real experiment it constantly increases. This is due to an increase in the surface of the layer from which ions are attracted due to edge effects. Edge effects do not appear if a voltage equal to the voltage applied to the probe is applied to the shielding ring and the current flowing only from the probe is measured. Despite that the slope of the ion parts of the 6 Plasma Sources Sci. Technol. 32 (2023) 105015 A V Siasko et al probe characteristics is different for probes with and without the shielding ring, the wall current at the floating probe poten- tial almost coincides within the measurement errors (figure 7). This allows one to neglect the edge effects and also indicates that the thickness of the space charge layer is small compared to the transverse dimensions of the flat probe. It should be noted that the measured wall currents are not only the currents of charged particles moving from the plasma volume to the walls. In addition to electrons and ions, meta- stable atoms with a potential energy of the order of 11.5 eV for argon and 20 eV for helium diffuse towards the wall, as well as resonance photons with energies of the same orders of magnitude propagating in the plasma volume. These particles hitting the probe surface make it emit electrons which will lead to a distortion of the ion part of the probe characteristic and an increase in the measured wall current. Under the investigated conditions, the flux of secondary electrons is determined by the quantum yield of probe materials [59], as well as by the density of metastable and resonance states. When construct- ing a theory that would allow comparing the theoretical wall currents with the results of the experimental study, it would be necessary to take into account the contribution of the second- ary electron flux in the total value of the particle flux density. For correct measurement of the wall currents in the con- stricted state regime, four probes were soldered in each molyb- denum glass tube in one cross-section (figure 5). In the case of a small displacement of a filament from the axis of the dis- charge tube, the wall currents were recalculated according to the empirically determined equation: (cid:19) (cid:18) it = im , (2) 2 ′ R R where it is the true value of the current towards the probe, im is the measured value of the wall current at a small displacement ′ of the plasma filament from the axis of the discharge tube, R is the distance from the probe to the center of the plasma fila- ment. This allowed measuring the wall currents in the constric- ted state at small displacements of the filament from the axis of the tube, which is most important for measurements in argon, where the filament can experience Archimedes’ buoyant force [50]. Measurements of the wall current along the length of the discharge tube showed that the characteristics are uniform along the longitudinal direction. The value of the measured wall current provides informa- tion on the difference between the ionization and the recom- bination rates in the volume. The ionization balance equation for cylindrical geometry has the form: I (r) − Γ (r) = 1 r d dr rDa dn dr , with the corresponding boundary conditions: (cid:12) (cid:12) (cid:12) (cid:12) = 0, n| r=R = 0. dn dr r=0 (3) (4) 7 Multiplying both sides of the equation (3) by rdr and integrat- ing from 0 to R, one can obtain an expression for the particle flux towards the wall and the wall current jw, accordingly: ˆ R 0 I (r) rdr − ˆ R 0 Γ (r) rdr = −Da dn dr (cid:12) (cid:12) (cid:12) (cid:12) , r=R (cid:12) (cid:12) (cid:12) (cid:12) r=R . (5) (6) jw = −eDa dn dr It can be seen that the wall current can be experimentally determined either directly from the probe characteristic or using the bremsstrahlung continuum profile and measuring the slope of the restored electron density profile near the wall (figures 4(c) and (d)). 4. Results. Measurement of the wall currents at low pressures in the absence of constriction Measurements in argon, neon, and helium were first performed at low pressures from 1 to 36 Torr·cm, when the ionization balance of charged particles is determined by the creation of particles in the plasma volume and their loss on the tube walls due to ambipolar diffusion in the absence of volume recom- bination. Figure 8 shows the values of the wall current density jw depending on the reduced discharge current i/R for differ- ent reduced pressures pR. In helium, the measurements were carried out in tubes with a radius of 1.2 cm (figure 8(a)) and 2.4 cm (figure 8(b)). The results for argon (figure 8(c)) and neon (figure 8(d)) are given for the tube radius of 1.2 cm. In all gases, in a diffuse regime when charged particles drift towards the discharge tube wall, the wall current increases monotonously with an increase in the discharge current. A slight deviation from linear dependences can be associated with a decrease in the electric field with an increase in the dis- charge current. It can be seen from the figure that the highest values of the wall currents at equal reduced pressures and dis- charge currents are observed in helium. The smallest wall cur- rents are observed in argon. This picture corresponds to the difference in ion mobilities and electron temperatures for these gases, which determine the ambipolar diffusion coefficient. Characteristics presented in figures 8(a) and (b) allow checking the applicability of the similarity rules. Under con- ditions when the drift of charged particles to the wall is main- tained in the diffusion mode, it was found that for tubes of different radii the value of the wall current multiplied by the radius jwR remains constant for the same values of the reduced discharge current i/R and reduced pressure pR (figure 9). It can be seen that the similarity rules are fulfilled quite well. The similarity rule can also be derived from the equation (6) if taking into account the dependence of the ambipolar dif- fusion coefficient on pressure (Da = Da0/p) and introducing the dimensionless coordinates x = r/R and y = n/n0. In these variables, Da0 is the coefficient of ambipolar diffusion at a pressure of 1 Torr and n0 is the electron density at the axis of Plasma Sources Sci. Technol. 32 (2023) 105015 A V Siasko et al Figure 8. Dependence of the wall current density jw on the reduced discharge current i/R at low pressures in helium at R = 1.2 cm (a) and R = 2.4 cm (b), in argon at R = 1.2 cm (c), and in neon at R = 1.2 cm (d). The above similarity rule is valid when the volume processes in the ionization balance are two-particle. Three-particle pro- cesses, for example, the conversion of atomic ions into molecular ones followed by dissociative recombination, as well as chemoionization processes, resonance radiation trap- ping, etc can lead to a deviation from the similarity rules. At low pressures, as shown in figure 9, deviation from the simil- arity rules is not observed. 5. Results. Measurement of the wall currents at high pressures in the presence of constriction 5.1. Argon Since, as mentioned above, the process of dissociative recom- bination in argon has been studied sufficiently well, and the rate constant is known, it seems appropriate to begin the investigation of the effect of recombination processes on the discharge constriction from this gas. The switch of the diffuse regime to the constricted one occurs if the gas pressure and the discharge current exceed the critical values. The experiments in the constricted discharge in argon were carried out at reduced pressures of 100, 200, and 300 Torr·cm in tubes with a radius of 1.2 and 2.6 cm. In the constricted dis- charge in argon, the electron density profile is strongly com- pressed, leading to a noticeable decrease in the value of dn r=R dr in the equation (6) compared to the diffuse discharge regime. This should lead to a decrease in the wall currents during (cid:12) (cid:12) Figure 9. Dependence of the reduced wall current density jwR on the reduced discharge current i/R at low pressures in helium. Black dots- R = 1.2 cm, red dots- R = 2.4 cm. the discharge tube. Then the expression for the reduced wall current jwR will be written in terms of the reduced similarity parameters pR and n0R which is proportional to i/R: jwR = − eDa pR n0R (cid:12) (cid:12) (cid:12) (cid:12) dy dx . x=1 (7) 8 Plasma Sources Sci. Technol. 32 (2023) 105015 A V Siasko et al Figure 10. Dependence of the wall current density jw on the reduced discharge current i/R at high pressures in diffuse and constricted regimes in argon at R = 1.2 cm and R = 2.6 cm at pR = 100 Torr·cm (a), pR = 200 Torr·cm (b), and pR = 300 Torr·cm (c). Figure 11. Dependence of the reduced wall current density jwR on the reduced discharge current i/R at high pressures in diffuse and constricted regimes in argon. Black dots- R = 1.2 cm, red dots- R = 2.6 cm. the discharge constriction. Figure 10 shows the dependencies of the wall currents on the reduced discharge current for the mentioned pressures and discharge tube radii. It can be seen from the figures that at low discharge currents in the region of the diffuse regime, a smooth increase in the wall current is observed with an increase in the discharge current, which is similar to diffuse discharges at low pressures (figure 8). When the critical values of the discharge current are exceeded, an abrupt decrease in the flux of charged particles towards the wall is observed. This jump is due to a rapid increase in the rate of recombination processes and the loss of charged particles in the volume. Thus, constriction in argon develops by the scenario presented in figure 1(c). After the transition from the diffuse to the constricted state, the wall current changes only slightly. For reduced pressures of 100 and 200 Torr·cm, the similar- ity rules were verified. On figure 11 the results are presented in the similarity parameters. It can be seen that the similarity rules are satisfied not only for low pressures (figure 9), but also for high ones. 5.2. Helium Constriction of the ionization zone in helium can be observed starting from reduced pressures of about 50 Torr·cm. Due to very high gas temperatures at such pressures, all measure- ments in a constricted discharge in helium were carried out in a quartz tube of the radius R = 2.4 cm. Figure 12 demonstrates the dependence of the wall current density on the reduced discharge current in helium at pressures of 53 and 100 Torr·cm (a) and the current–voltage character- istic at a pressure of 100 Torr·cm (b). It can be seen that the probe current increases linearly over the entire range of meas- ured discharge currents and decreases inversely to the pres- sure, despite the presence of an abrupt transition to optical constriction, which is clearly observed in the current–voltage characteristic. This result is drastically different from the phe- nomena observed in argon, when constriction was accom- panied by an abrupt decrease in the wall current (figures 10 and 11). Hence it follows that, upon the transition to optical constriction, the regime of loss of charged particles is determ- ined by the ambipolar diffusion, and the radial profile should not experience noticeable compression. This correlates with the spectral distributions of the bremsstrahlung continuum shown in figure 4(d). Thus, the low volume recombination rates do not play a decisive role in the formation of a con- stricted discharge in helium. This corresponds to the values of volume recombination given in [55]. The constriction phe- nomenon in the case of helium, unlike argon, develops by the scenario presented in figure 1(b). 9 Plasma Sources Sci. Technol. 32 (2023) 105015 A V Siasko et al Figure 12. (a) Dependence of the reduced wall current density jwR on the reduced discharge current i/R at high pressures in diffuse and constricted regimes in helium at pR = 53 Torr·cm and pR = 100 Torr·cm. (b) Volt-ampere characteristic at pR = 100 Torr·cm. Figure 13. Time dependence of radiation intensity on the axis of a discharge tube in helium at pR = 100 Torr·cm. Homogeneous glow in a diffuse discharge at i/R = 8.3 mA cm i/R = 83.3 mA cm −1 (a). Development of ionization waves in a constricted discharge at −1 and i/R = 16.7 mA cm −1 (c). −1 (b) and i/R = 125 mA cm Figure 14. Time dependence of radiation intensity on the axis of a discharge tube in neon at pR = 96 Torr·cm. Homogeneous glow in a diffuse discharge at i/R = 48 mA cm −1 (a). Development of ionization waves in a constricted discharge at i/R = 51 mA cm −1 (b). The discharge constriction in helium, as in other inert gases, is accompanied by the simultaneous development of ionization instability in the form of striations. Figure 13 shows the res- ult of registration of radiation across the tube using a PMT for currents corresponding to diffuse (a) and constricted (b and c) discharges at a pressure of 100 Torr·cm. For comparison, figure 14 shows similar temporal characteristics of radiation in neon in diffuse and constricted states near the critical value 10 Plasma Sources Sci. Technol. 32 (2023) 105015 A V Siasko et al Figure 15. Illustration of the extrapolation of the diffusion theory to the region of a constricted discharge to find the volume loss rate in the case of argon (a) and helium (b). of the discharge current. Despite the differences in constric- tion mechanisms in these gases noted above, it is clear that in neon and helium the phenomenon of stratification proceeds in the identical way. In the diffuse state, the glow is homogen- eous. The constricted state is accompanied by the appearance of striations. Despite that the role of recombination in the ionization balance in helium is negligible, nevertheless, it can contrib- ute to the development of ionization instability. This question requires the construction of a collision-radiative model and the study of the stationary solution for stability with respect to small perturbations of plasma parameters. 6. Conclusion A complex of experimental measurements of the wall currents in argon, neon, and helium has been carried out in a wide range of pressures and discharge currents in diffuse and constricted regimes of a DC discharge. It is shown that the measurements of the wall currents can serve as a method for estimating the role of volume recombination in the ionization balance. The performed measurements have shown that the similar- ity rules take place for the values of the reduced wall current jwR depending on the reduced discharge current i/R and the reduced pressure p/R. The similarity rules are fulfilled both at low pressures and at high pressures during the transition to the constricted regime. In diffuse discharge, a monotonic increase of the wall cur- rent is observed in all gases with an increase in the discharge current. With increasing pressure, the wall current decreases inversely. In heavy gases, when the critical values of the dis- charge current and pressure are exceeded, the wall current undergoes an abrupt decrease and then remains approxim- ately constant with the after increase in the discharge current. In contrast, in helium, such jumps are not observed during the transition to the constricted regime. The wall current in helium increases proportionally to the discharge current des- pite the jump-like compression of the line radiation and the voltage drop in the current–voltage characteristic. This fact confirms the insignificant role of the volume recombination in the ionization balance in helium. Despite the difference in the constriction mechanisms, in all investigated gases, simultaneously with the compression of the discharge one can observe the development of ionization instability in the form of striations. The whole range of observed phenomena in neon and argon can be described quite well by the theory [20]. A quantitat- ive description of the phenomena occurring in helium requires the development of a detailed collision-radiative model. This model should predict an abrupt compression of line radiation with a smooth distribution of the electron density upon the transition to the constricted state. An analysis of the station- ary solution for stability with respect to small perturbations of the plasma parameters should describe the development of the ionization instability. The presented results of the measured wall currents can serve to determine the volume recombination rate. For this purpose, based on a detailed collision-radiative model, in the first approximation it is necessary to calculate the plasma para- meters without account of the recombination losses in the framework of diffusion theory. The obtained solution makes it possible to calculate the flux of charged particles towards the wall of the discharge tube. It can be expected that at low pressures and currents when recombination losses can be neg- lected, the wall current in the diffusion theory should coincide with the experimental results. Extrapolation of the diffusion theory to the region of high pressures and discharge currents should give a difference between the measured and calculated wall currents. By extrapolation, one means the calculation of plasma parameters on the basis of a detailed collision-radiative model in the absence of recombination. From this difference, one can estimate the volume-averaged value of recombination losses- equation (5). In the next approximation, one can intro- duce volume losses in the collision-radiative model and, by the iterative method, achieve the convergence of the result. This method will give the dependence of the volume-averaged recombination rate on the current and pressure, i.e. depend- ing on the electron temperature, gas temperature, and elec- tron density in real plasma conditions. The proposed method for calculating the recombination loss rate from the measured wall currents is qualitatively illustrated in figure 15 on the example of argon (a) at a pressure pR = 100 Torr·cm and radius 11 Plasma Sources Sci. Technol. 32 (2023) 105015 A V Siasko et al R = 2.6 cm and helium (b) at pR = 100 Torr·cm and radius R = 2.4 cm. In the subsequent work, is planned to construct a it collision-radiative model for helium and implement the pro- posed method for determining the rate of recombination losses and their temperature dependences for the quantitative inter- pretation of the phenomena observed in helium. Data availability statement The data cannot be made publicly available upon publication because no suitable repository exists for hosting data in this field of study. The data that support the findings of this study are available upon reasonable request from the authors. Acknowledgment Authors gratefully acknowledge RSF Grant No. 22-12-00002. ORCID iDs A V Siasko  https://orcid.org/0000-0002-3546-9541 V Yu Karasev  https://orcid.org/0000-0003-2584-0068 Yu B Golubovskii  https://orcid.org/0000-0001-7757-0616 References [21] Ridenti M A, de Amorim J, Dal Pino A, Guerra V and Petrov G M 2018 Phys. Rev. E 97 13201 [22] Jovanovi´c A P, Hoder T, Höft H, Loffhagen D and Becker M M 2023 Plasma Sources Sci. Technol. 32 055011 [23] Hoder T C V, Loffhagen D, Wilke C, Grosch H, Schäfer J, Weltmann K-D and Brandenburg R 2011 Phys. Rev. E 84 046404 [24] Zhu H, Yao W and Li Z 2020 Plasma Process. Polym. 17 1900271 [25] Zhu H, Huang Q, Wu Y, Li Y and Ren K 2022 Plasma Sci. Technol. 24 055406 [26] Kelly S, van de Steeg A, Hughes A, van Rooij G and Bogaerts A 2021 Plasma Sources Sci. Technol. 30 055005 [27] Viegas P, Vialetto L, Wolf A J, Peeters F J J, Groen P W C, Righart T W H, Bongers W A, van de Sanden M C M and Diomede P 2020 Plasma Sources Sci. Technol. 29 105014 [28] Kolobov V I and Arslanbekov R R 2022 Phys. Rev. E 106 065206 [29] Boeuf J P 2022 Phys. Plasmas 29 022105 [30] Hartmann P, Rosenberg M, Juhasz Z, Matthews L S, Sanford D L, Vermillion K, Carmona-Reyes J and Hyde T W 2020 Plasma Sources Sci. Technol. 29 115014 [31] Schottky W 1924 Phys. Z. 25 635–40 [32] Golubovskii Y B, Nekuchaev V O, Gorchakov S and Uhrlandt D 2011 Plasma Sources Sci. Technol. 20 53002 [33] Kagan Y and Lyagushchenko R 1964 Zh. Tekhn. Fiz. 34 1873 [34] Golubovskii Y B, Kagan Y M and Lyagushchenko R I 1966 Opt. Spectrosc. 21 295 [35] Wojaczek K 1966 Beitr. Plasma Phys. 6 211–25 [36] Smits R and Prins M 1979 Physica B+C 96 262–85 [37] Eletskii A 1982 Khim. Plazmy 9 151–78 [38] Massey J T 1965 J. Appl. Phys. 36 373–80 [39] Ellington H I 1969 J. Phys. D: Appl. Phys. 2 65–69 [40] Baranov V Y and Ul’yanov K 1969 Sov. Phys. Tech. Phys. 14 176 [1] Stark J 1902 Die Elektrizitat in Gasen (J. A. Barth) [2] Massey J T and Cannon S M 1965 J. Appl. Phys. 36 361 [3] Venzke D, Hayess E and Wojaczek K 1966 Contrib. Plasma [41] Rakhimov A T and Ulinich F 1969 Contraction of cylindrical gas discharge Technical Report Moscow State University [42] Mouwen C 1971 Investigation of the constricted positive Phys. 6 365–75 [4] Pfau S and Rutscher A 1968 Beitr. Plasma Phys. 8 73–84 [5] Pfau S, Rutscher A and Wojaczek K 1969 Contrib. Plasma Phys. 9 333–58 [6] Heymann P 1969 Beitr. Plasma Phys. 9 491–8 [7] Mouwen C and Claassens J 1970 Phys. Lett. A 31 123–4 [8] Venzke D 1975 Beitr. Plasma Phys. 15 35–45 [9] Smits R and Prins M 1975 Physica B+C 80 571–84 [10] Venzke D 1978 Beitr. Plasma Phys. 18 65–78 [11] Golubovskij J B and Nekuchajev V O 1986 Beitr. Plasma Phys. 26 67–80 [12] Petrov G M and Ferreira C M 1999 Phys. Rev. E 59 3571–82 [13] Ionikh Y Z, Meshchanov A V, Petrov F B, Dyatko N A and Napartovich A P 2008 Plasma Phys. Rep. 34 867–78 [14] Dyatko N A, Ionikh Y Z, Kochetov I V, Marinov D L, Meshchanov A V, Napartovich A P, Petrov F B and Starostin S A 2008 J. Phys. D: Appl. Phys. 41 055204 column in neon PhD Thesis Department of Applied Physics proefschrift [43] Uliyanov K N 1973 Sov. Phys. Tech. Phys. 18 360 [44] Wasserstrom E and Crispin Y 1982 J. Appl. Phys. 53 5565–77 [45] Lynch R H 1967 J. Appl. Phys. 38 3965–8 [46] Jaeger E F, Oster L and Phelps A V 1976 Phys. Fluids 19 819–30 [47] R˚uˇzicˇka T and Rohlena K 1976 Czech. J. Phys. B 26 282–93 [48] Hatori S and Shioda S 1976 J. Phys. Soc. Japan 40 1449–55 [49] Golubovskii Y B, Siasko A V, Kalanov D V and Nekuchaev V O 2018 Plasma Sources Sci. Technol. 27 85009 [50] Golubovskii Y B, Siasko A V and Nekuchaev V O 2019 Plasma Sources Sci. Technol. 28 045007 [51] Golubovskii Y B, Siasko A V and Nekuchaev V O 2020 Plasma Sources Sci. Technol. 29 065020 [52] Berlande J, Cheret M, Deloche R, Gonfalone A and Manus C [15] Shkurenkov I A, Mankelevich Y A and Rakhimova T V 2008 1970 Phys. Rev. A 1 887–96 Plasma Phys. Rep. 34 780–93 [16] Shkurenkov I A, Mankelevich Y A and Rakhimova T V 2009 [53] Johnson A W and Gerardo J B 1972 Phys. Rev. A 5 1410–8 [54] Boulmer J, Davy P, Delpech J F and Gauthier J C 1973 Phys. Phys. Rev. E 79 46406 Rev. Lett. 30 199–202 [17] Gnybida M, Loffhagen D and Uhrlandt D 2009 IEEE Trans. [55] Deloche R, Monchicourt P, Cheret M and Lambert F 1976 Plasma Sci. 37 1208–18 Phys. Rev. A 13 1140–76 [18] Ionikh Y Z, Dyatko N A, Meshchanov A V, Napartovich A P and Petrov F B 2012 Plasma Sources Sci. Technol. 21 55008 [56] Ivanov V and Skoblo Y E 1988 Opt. Spectrosc. 65 445–7 [57] Carata L, Orel A E and Suzor-Weiner A 1999 Phys. Rev. A [19] Carbone E A D, Hübner S, Palomares J M and van der Mullen J J A M 2012 J. Phys. D: Appl. Phys. 45 345203 [20] Golubovskii Y B, Kalanov D and Maiorov V A 2017 Phys. Rev. E 96 23206 59 2804–12 [58] Schneider W H 1956 Acta Phys. Austriaca 10 54 [59] Kulikov S, Mishchenko E, Nikitin V and Startsev G 1965 J. Appl. Spectrosc. 3 1–4 12
null
10.1265_ehpm.22-00245.pdf
Availability of data and material The datasets generated and/or analyzed during the current study are not publicly available because the study involves human participants with a nondisclosure provision of individual data stated in the written informed consent in order to prevent compromise of study participants’ privacy but are available from the corresponding author upon reasonable request.
Availability of data and material The datasets generated and/or analyzed during the current study are not publicly available because the study involves human participants with a nondisclosure provision of individual data stated in the written informed consent in order to prevent compromise of study participants' privacy but are available from the corresponding author upon reasonable request.
Environmental Health and Preventive Medicine RESEARCH ARTICLE Environmental Health and Preventive Medicine (2023) 28:22 https://doi.org/10.1265/ehpm.22-00245 Cross-sectional associations between early mobile device usage and problematic behaviors among school-aged children in the Hokkaido Study on Environment and Children’s Health Chihiro Miyashita1, Keiko Yamazaki1, Naomi Tamura1, Atsuko Ikeda-Araki1,2, Satoshi Suyama3, Takashi Hikage4, Manabu Omiya5, Masahiro Mizuta6 and Reiko Kishi1* *Correspondence: rkishi@med.hokudai.ac.jp 1Center for Environmental and Health Sciences, Hokkaido University, Sapporo, Japan. 2Faculty of Health Sciences, Hokkaido University, Sapporo, Japan. 3Funded Research Division of Child and Adolescent Psychiatry, Hokkaido University Hospital, Sapporo, Japan. 4Graduate School, Faculty of Information Science and Technology, Hokkaido University, Sapporo, Japan. 5Information Initiative Center, Hokkaido University, Sapporo, Japan. 6Center for Training Professors in Statistics, The Institute of Statistical Mathematics, Tokyo, Japan. Abstract Background: Concerns have been raised about the adverse health impacts of mobile device usage. The objective of this cross-sectional study was to examine the association between a child’s age at the first use of a mobile device and the duration of use as well as asso- ciated behavioral problems among school-aged children. Methods: This study focused on children aged 7–17 years participating in the Hokkaido Study on Environment and Children’s Health. Between October 2020 and October 2021, the participants (n = 3,021) completed a mobile device use-related questionnaire and the strengths and difficulties questionnaire (SDQ). According to the SDQ score (normal or borderline/high), the outcome variable was behavioral problems. The independent variable was child’s age at first use of a mobile device and the duration of use. Covariates included the child’s age at the time of survey, sex, sleep problems, internet addiction, health-related quality of life, and history of developmental concerns assessed at health checkups. Logistic regression analysis was performed for all children; the analysis was stratified based on the elementary, junior high, and senior high school levels. Results: According to the SDQ, children who were younger at their first use of a mobile device and used a mobile device for a longer duration represented more problematic behaviors. This association was more pronounced among elementary school children. Moreover, subscale SDQ analysis showed that hyperactivity, and peer and emotional problems among elementary school children, emotional problems among junior high school children, and conduct problems among senior high school children were related to early and long usage of mobile devices. Conclusions: Elementary school children are more sensitive to mobile device usage than older children, and early use of mobile devices may exacerbate emotional instability and oppositional behaviors in teenagers. Longitudinal follow-up studies are needed to clarify whether these problems disappear with age. Keywords: Hokkaido Study on Environment and Children’s Health, Mobile devices, Children, Behavioral problems 1 Introduction The usage of mobile devices, including smartphones and tablets, has rapidly increased across social classes. In 2018, the smartphone penetration rate was approximately 75% and 45% in developed and developing countries, re- spectively [24]. Globally, the age at which a child first uses a mobile device is constantly tipping toward earlier ages. Some parents allow their children to use mobile devices at early ages for entertainment purposes. Concerns have been raised about the negative early health impacts of regular contact with mobile devices, especially in relation to neu- robehavioral developmental delays and imbalances in healthy activities in children [1, 26, 28]. The World Health Organization and some developed countries, including the United States of America, Canada, and Australia, have recommended that parents should avoid giving screen- based devices to infants younger than 18 months and re- strict screen time to <1 hour daily for preschool children aged 2–5 years [2, 3, 23, 30, 31]. While these recommendations are based on previous studies that mainly targeted traditional devices such as © The Author(s) 2023. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data. Environmental Health and Preventive Medicine (2023) 28:22 2 of 11 the television, there is limited scientific evidence regarding the associations between the early use of contemporary mobile devices and development in children. Unstructured play with the hands and body and practical social commu- nication are important for the development of the central nervous system in early infancy [2, 3]. For establishing healthy behaviors, daytime activities, nighttime sleep, and routine mealtimes are essential. Early mobile device usage can interfere with comprehensive neurodevelopment and engagement in healthy activities, both of which foster lan- guage and cognitive development and social skills in in- fancy. According to a Korean study, children aged 1–3 years who spent longer times with touch screens displayed increased emotional problems and depression and anxiety symptoms [15]. Another study reported that the regular use of screen-based media among infants at 4 months was associated with poor performance on self-regulation tests but not cognitive flexibility or working memory tests at assessed at 14 months [17]. In Japan, the internet penetration rate among school- aged children was 93.2% in 2019. The penetration rate rapidly increased from 12.5% to 49.5%, for smartphones and from 15.3% to 41.0%, for tablets, among elementary school students, between 2014 and 2019. More than 50% of children have contact with the internet in their early life, with 4.7% of those aged 0 years and 50.2% of those aged 3 years using the internet [18]. A Japanese study suggested that regular and frequent use of mobile devices among first grade elementary school children was associated with in- creased emotional and behavioral problems [9]. However, this study did not assess school children aged >7 years. The association between mobile device usage in early life and developmental effects was not evaluated. Therefore, the current study aimed to assess the association between a child’s age at first use of a mobile device and the duration of use and the behavioral problems among elementary, junior high, and senior high school students. 2 Methods 2.1 Subjects This study focused on children aged 7–17 years between 2003 and 2012; the children were followed up until Oc- tober 2020 in a prospective birth cohort study of the Hokkaido Study on Environment and Children’s Health [11–13]. We mailed the strengths and difficulties question- naire (SDQ) and a questionnaire regarding mobile device use and lifestyle to 5,221 parent–child pairs between Oc- tober 2020 and January 2021 based on a random selection. A total of 3,364 responses were received by October 2021 (response rate = 64.4%). From the questionnaire, we as- sessed the associations between exposure to mobile de- vices (the child’s age at first use of a mobile device and duration of use), the outcomes (child behavioral problems as determined by the SDQ), and potential covariates, in- cluding the child’s sleep problems, internet addiction, health-related quality of life, and history of developmental concerns assessed at health checkups (Fig. 1). The child- ren’s age at survey were recorded based on the response date on the questionnaire. The child’s date of birth and sex were obtained from medical records. Additional informa- tion, including the parents’ age and educational and house- hold incomes, were obtained from the baseline question- naire during maternal pregnancy [11, 13]. Of the 3,364 questionnaires returned, we excluded two pre-school chil- dren and 341 participants with missing response data. A total of 3,021 children, including 1,433 elementary school, 1,121 junior high school, and 467 senior high school chil- dren, were finally included in the study (Table 1 and Fig. 1). 2.2 Ethics statement The institutional ethics board for epidemiological studies at Hokkaido University Graduate School of Medicine and Hokkaido University Center for Environmental and Health Sciences approved the study protocol (approval number 19 - 118). Informed consent was obtained from all study participants before enrollment. 2.3 Exposure assessment We used 4-type exposure factors to monitor child mobile device usage (Table 2); the first exposure factor was the child’s age at the first use of a mobile device, according to parents’ answer to the following question: “At what age did your child first use a mobile device? For example, you showed movies to your child on mobile devices, or your child used a mobile device by themselves?” The second exposure factor was the duration of mobile device usage, meaning usable years, calculated based on the child’s age at the first use of a mobile device and that at the time of Participants were born from 2003 to 2012, and have been followed up until October 2020 in the Hokkaido Study on Environment and Children’s Health (n = 13,899) Random selection of participants to receive mobile device usage and child health questionnaire (n = 5,221) Participants returned mobile device usage and child health questionnaire (n = 3,364) Excluded 2 participants who were pre-school children Excluded 341 participants who had missing data Study participants (n = 3,021) Fig. 1 Flow chart of study participants Environmental Health and Preventive Medicine (2023) 28:22 3 of 11 Table 1 Characteristics of participants (children and their parents). All children Categories Number (%) Mean « SD or Median (IQR) School type Elementary school Number (%) Mean « SD or Median (IQR) Junior high school Number (%) Mean « SD or Median (IQR) Senior high school Number (%) Mean « SD or Median (IQR) p 3021 12.4 « 2.4 1433 10.2 « 1.2 1121 13.7 « 0.9 467 15.8 « 0.5 <0.001 Child Age at survey Sex Siblings History of developmental concerns Boy Girl No Yes No Yes 1499 (49.6) 1522 (50.4) 401 (16.0) 2100 (84.0) 2726 (90.2) 295 (9.8) Personal mobile device use and restriction 564 (18.7) No Having personal mobile devices 2457 (81.3) Yes Restricted use of mobile devices on weekdays No Yes No Yes Restricted use of mobile devices on holidays Child health quality Health-related quality of life Sleep problems Internet addiction 1310 (43.4) 1709 (56.6) 1456 (48.2) 1565 (51.8) 3021 3021 3021 721 (50.3) 712 (49.7) 192 (17.2) 926 (82.8) 1271 (88.7) 162 (11.3) 434 (30.3) 999 (69.7) 389 (27.2) 1042 (72.8) 472 (32.9) 961 (67.1) 550 (49.1) 571 (50.9) 151 (15.4) 830 (84.6) 1028 (91.7) 93 (8.3) 129 (11.5) 992 (88.5) 540 (48.2) 581 (51.8) 589 (52.5) 532 (47.5) 228 (48.8) 239 (51.2) 58 (14.4) 344 (85.6) 427 (91.4) 40 (8.6) 1 (0.2) 466 (99.8) 381 (81.6) 86 (18.4) 395 (84.6) 72 (15.4) 0.766 0.342 0.025 <0.001 <0.001 <0.001 43.0 (38.0, 47.0) 1433 44.0 (40.0, 48.0) 1121 43.0 (37.0, 47.0) 467 41.0 (35.0, 46.0) <0.001 24.0 (22.0, 26.0) 21.0 (16.0, 26.0) 1433 1433 24.0 (22.0, 26.0) 19.0 (15.0, 24.0) 1121 1121 23.0 (22.0, 26.0) 21.0 (17.0, 26.0) 467 467 23.0 (21.0, 25.0) <0.001 23.0 (19.0, 27.0) <0.001 Mother Age at time of survey Educational level 3021 44.1 « 5.1 1433 42.1 « 4.9 1121 45.3 « 4.6 467 47.2 « 4.2 <0.001 ¯9 10–12 13–15 >16 80 (2.6) 1152 (38.1) 1376 (45.5) 413 (13.7) 37 (2.6) 553 (38.6) 630 (44.0) 213 (14.9) 32 (2.9) 421 (37.6) 522 (46.6) 146 (13.0) 11 (2.4) 178 (38.1) 224 (48.0) 54 (11.6) 0.478 Father Age at time of survey Educational level ¯9 10–12 13–15 >16 <3.0 3.0–4.9 5.0–7.9 ²8 Annual household income (million Japanese Yen) 2984 45.7 « 5.9 1419 43.7 « 5.7 1103 46.9 « 5.4 462 48.9 « 5.2 <0.001 167 (5.6) 1099 (36.8) 785 (26.3) 938 (31.4) 510 (19.0) 1230 (45.8) 722 (26.9) 226 (8.4) 82 (5.8) 508 (35.6) 385 (27.0) 451 (31.6) 252 (19.5) 594 (46.0) 347 (26.9) 98 (7.6) 56 (5.1) 409 (37.1) 292 (26.5) 346 (31.4) 194 (19.6) 446 (45.1) 263 (26.6) 87 (8.8) 29 (6.3) 182 (39.6) 108 (23.5) 141 (30.7) 64 (15.7) 190 (46.7) 112 (27.5) 41 (10.1) 0.637 0.481 SD, standard deviation. SDQ, Strength and Difficulties Questionnaire. IQR, interquartile range is the 75th and 25th percentile p value by one-way ANOVA, »2 test, and Kruskal-Wallis test, which were used to compare data among elementary, junior high, and senior high school children. survey. The third exposure factor was the child’s age at which they were given personal mobile devices, according to answers by parents to the following question: “At what age did your child first receive a personal mobile device such as a cell phone or tablet?” The fourth exposure factor was the duration of personal mobile device usage, indicat- ing that holding years were calculated using the age at which a child had a personal mobile device and that at the time of survey. 2.4 Assessment outcome We used the SDQ of the common methods for assessing behavioral and mental health problems among children and adolescents aged 4–17 in questionnaires, which were completed by the children’s parents or teachers [7]. The SDQ consists of 25 items, each rated as being not true (0), somewhat true (1), or certainly true (2). The items are divided into five subscales covering conduct problems, hyperactivity, emotional symptoms, peer problems, and Environmental Health and Preventive Medicine (2023) 28:22 4 of 11 Table 2 Child’s age at first use of a mobile device and the duration of use. Age at first use of a mobile device Duration of mobile device usage Age at first having a personal mobile device Duration for having a personal mobile device All children Median (IQR) 7.0 (5.0, 10.0) 5.0 (3.0, 7.0) 12.0 (8.0, 13.0) 2.0 (1.0, 3.0) School type Elementary school children Median (IQR) 6.0 (4.0, 8.0) 4.0 (2.0, 6.0) 8.0 (7.0, 10.0) 2.0 (1.0, 3.0) Junior high school children Median (IQR) 9.0 (6.0, 11.0) 5.0 (3.0, 8.0) 12.0 (11.0, 13.0) 2.0 (1.0, 3.0) Senior high school children Median (IQR) 10.0 (7.0, 12.0) 6.0 (3.0, 9.0) 14.0 (12.0, 15.0) 2.0 (1.0, 4.0) p <0.001 <0.001 <0.001 0.49 IQR, interquartile range. p value by Kruskal-Wallis test, which were used to compare median among elementary, junior high, and senior high school children. prosocial behavior [7]. Summing up the scores on the four subscales, i.e., excluding prosocial behavior, gives the SDQ total difficulties score (TDS), which can range from 0 to 40. The TDS from the Japanese version of the SDQ has a cut-off score of 12/13 for normal/borderline and 15/ 16 for borderline/high [16]. In this study, according to the parents’ answers, the children were divided into two groups—normal or borderline/high—based on the TDS and five subscales. The cut-off score for the five subscales of conduct problems, hyperactivity, emotional symptoms, peer problems, and prosocial behavior were 3/4, 5/6, 3/4, 3/4, and 6/5 for the normal and borderline/high groups, respectively [8]. 2.5 Covariates Based on previous studies, we used several covariates, in- cluding the children’s age at the time of survey, sex, and history of developmental concerns at health checkups, health-related quality of life, sleep problems, internet ad- diction, school type, and interaction between a child’s mo- bile device usage and school type, all of which had poten- tial confounding effects on a child’s mobile device usage and child behavioral problems based on the associations noted in this study (Table 4 and Supplemental Table 2 and 3) [9, 16, 19]. In fact, we assessed the generic health- related quality of life for children using the KIDSCREEN- 10 questionnaire [21]; this questionnaire consists of 10 items, including the physical, psychological, and social di- mensions of wellbeing [25]. We also assessed the tendency toward internet dependence using a modified Internet Ad- diction Test, which comprised 11 items [29, 32]. These questionnaires were completed by the children. We as- sessed sleep problems in the children based on 19 items from the short version of the sleep questionnaire for chil- dren [22], which was answered by their parents. Moreover, we used covariates to assess the interactions between a child’s mobile device usage and school type. The children included in this cross-sectional study had a wide birth-year period of 10 years, and their mobile device penetration rate had rapidly changed in the meantime [18]. The children’s behavioral problems not only changed as they developed but were also related to schooling from elementary to high school. We conducted school-specific analyses because we believed that the school type affected both the exposure and outcomes. Meanwhile, we did not use covariates of paren- tal household income, educational levels, and the presence of siblings as these factors were not associated with a child’s mobile device usage in this study (Table 4). This indicates that the association between mobile device usage and social class could be weakening. 2.6 Statistical analysis Simple associations between the parents’ and children’s characteristics and the child’s age at first use of a mobile device and duration of use as well as the children’s behav- ioral problems were assessed using one-way ANOVA, the »2 test, and the Kruskal–Wallis test. A logistic regression analysis was performed for all children, stratified by ele- mentary, junior high, and senior high school; the outcome (TDS and sub-analyses of child behavioral problems by SDQ) was considered the dependent variable, whereas ex- posure (child’s age at first use of a mobile device and the duration of use) was considered the independent variable. The covariates included child age at time of survey, sex, sleep problems, internet addiction, health-related quality of life, history of developmental concerns assessed at health checkups, and school type. The logistic regression analysis among children stratified by school type was adjusted for the same variables, except for school type. p < 0.05 was considered statistically significant. All statistical analyses were performed using SPSS software for Windows (ver- sion 21.0J; IBM, Armonk, NY, USA). 3 Results A total of 3,021 children, including 1,433 elementary school, 1,121 junior high school, and 467 senior high school children, were involved in this study. The children’s and parents’ characteristics and the differences among school types are shown in Table 1. The children’s and parents’ ages at the time of survey, the child’s internet addi- ction score, and the rate of having personal mobile devices increased as the children progressed from elementary to senior high school. The rate of restricted mobile device usage on weekdays and holidays and the health-related quality of life scores decreased as the children progressed from elementary to senior high school. Moreover, the his- tory of developmental concerns assessed at health check ups and sleep problem scores differed among elementary, junior high, and senior high school children (Table 1). Environmental Health and Preventive Medicine (2023) 28:22 5 of 11 Table 3 Number (%) of child behavioral problems based on the TDS and subscales according to SDQ. TDS Categories Normal Borderline/High Overall Number (%) 2586 (85.6) 435 (14.4) School type Elementary school Number (%) 1205 (84.1) 228 (15.9) Junior high school Number (%) 972 (86.7) 149 (13.3) Senior high school Number (%) 409 (87.6) 58 (12.4) p 0.072 Conduct problems Normal Borderline/High 2727 (90.3) 294 (9.7) 1260 (87.9) 173 (12.1) Hyperactivity/inattention Normal Borderline/High 2727 (90.3) 294 (9.7) 1257 (87.7) 176 (12.3) Emotional problems Normal Borderline/High 2598 (86.0) 423 (14.0) 1228 (85.7) 205 (14.3) Peer problems Normal Borderline/High 2594 (85.9) 427 (14.1) 1255 (87.6) 178 (12.4) 1032 (92.1) 89 (7.9) 1036 (92.4) 85 (7.6) 964 (86.0) 157 (14.0) 949 (84.7) 172 (15.3) 435 (93.1) 32 (6.9) 434 (92.9) 33 (7.1) 406 (86.9) 61 (13.1) 390 (83.5) 77 (16.5) <0.001 <0.001 0.798 0.031 Prosocial behavior 1011 (70.7) 418 (29.3) TDS, Total difficulties score. SDQ, Strength and Difficulties Questionnaire. P value by »2 test, which was used to compare the data among elementary, junior high, and senior high school children. Normal Borderline/High 2006 (66.6) 1005 (33.4) 715 (64.0) 403 (36.0) 280 (60.3) 184 (39.7) <0.001 The median age at first use of a mobile device and the duration of use was 7.0 and 5.0 years overall, 6.0 and 4.0 among elementary school children, 9.0 and 5.0 among junior high school children, and 10.0 and 6.0 among senior high school children, respectively (Table 2). The distribu- tion of child age at first use of a mobile device is shown in Supplemental Table 1. The number of borderline/high (cases) according to TDS was 435 (14.4%) overall, 228 (15.9%) among elementary school children, 149 (13.3%) among junior high school children, and 58 (12.4%) among senior high school children (Table 3). The five subscales of childhood behavioral problems according to SDQ are shown in Table 3. The associations between the age at which a child first used a mobile device and basic partic- ipant information are shown in Table 4, both unstratified and stratified by school type. Among all children, we ob- served positive associations between the child’s age at first use of a mobile device and the child’s and parent’s age at the time of survey. Negative associations were observed between the health-related quality of life score and sleep problems score. The age of the children at first use of a mobile device differed by their sex, history of develop- mental concerns assessed at health checkups, possession of personal mobile devices, and mobile device restriction on weekdays and holidays. When the children were strati- fied by school type, their age at first use of a mobile device was associated with the child’s and parent’s age at the time of survey, their sex, their history of developmental con- cerns assessed at health checkups, their health-related quality of life, their sleep problems, and their internet ad- diction, among at least one school type (Table 4). The associations between child behavioral problems (TDS) and the characteristics of participants among all children and among those stratified by school type are shown in Supplemental Tables 2 and 3. Among all chil- dren, the prevalence of child behavioral problems differed by the child’s and mother’s age at the time of survey, their sex, presence of siblings, their history of developmental concerns assessed by medical checkup, mobile device re- strictions on weekdays and holidays, their health-related quality of life, their sleep problems, their internet addic- tion, and their maternal educational levels (Supplemental Tables 2 and 3). When stratified by school type, the prev- alence of the children’s behavioral problems differed by the children’s age at the time of survey, their sex, presence of siblings, their history of developmental concerns as- sessed at health checkups, mobile device restrictions on weekdays and holidays, their health-related quality of life, their sleep problems, and their internet addiction, among at least one school type (Supplemental Tables 2 and 3). According to the logistic regression analysis among all children, compared to the normal group, the adjusted odds ratios of the borderline/high group significantly decreased with increasing child age at first use of a mobile device (95% CI = 0.85 [0.77, 0.93]); by contrast, it significantly increased with increasing duration for use of a mobile device (95% CI = 1.20 [1.08, 1.33]) (Table 5 and Fig. 2). In the logistic regression analysis stratified by school type, compared to the normal group, the adjusted odds ratios of child behavioral problems for the borderline/high group significantly decreased with increasing child’s age at first use of a mobile device (95% CI = 0.87 [0.81, 0.93]) and significantly increased with increasing duration for use of a mobile device (95% CI = 1.15 [1.08, 1.23]) among ele- mentary school children. However, the adjusted odds ra- tios for the child behavioral problems were not signifi- cantly associated with the child’s age and duration of mobile device use among junior and senior high school Environmental Health and Preventive Medicine (2023) 28:22 6 of 11 Table 4 Associations between child’s age at first use of a mobile device and characteristics of participants. Categories All children Median (IQR) r School type Elementary school children Median (IQR) r Junior high school children Median (IQR) r High school children r Median (IQR) Child Age at time of survey Sex Siblings History of developmental concerns Boy Girl No Yes No Yes 6.0 (5.0, 10.0)** 7.0 (5.0, 10.0) 7.0 (5.0, 10.0) 7.0 (5.0, 10.0) 7.0 (5.0, 10.0)** 6.0 (5.0, 9.0) Personal mobile device use and restriction Having personal mobile devices No Yes 6.0 (4.0, 8.0)** 7.0 (5.0, 10.0) Restricted use of mobile devices on weekdays Restricted use of mobile devices on holidays Child health quality Health-related quality of life Sleep problems Internet addiction Mother Age at time of survey Educational level Father Age at time of survey Educational level Annual household income (million Japanese Yen) No Yes No Yes 8.0 (5.0, 10.0)** 7.0 (5.0, 9.0) 7.5 (5.0, 10.0)** 7.0 (5.0, 9.0) ¯9 10–12 13–15 >16 ¯9 10–12 13–15 >16 <3.0 3.0–4.9 5.0–7.9 ²8 7.0 (5.0, 10.0) 7.0 (5.0, 10.0) 7.0 (5.0, 10.0) 7.0 (5.0, 10.0) 7.0 (5.0, 10.0) 7.0 (5.0, 10.0) 7.0 (5.0, 10.0) 7.0 (5.0, 10.0) 7.0 (5.0, 10.0) 7.0 (5.0, 10.0) 7.0 (5.0, 10.0) 7.0 (5.0, 10.0) 0.466** 0.272** 0.131** ¹0.031 5.0 (3.0, 7.0)** 6.0 (4.0, 8.0) 6.0 (4.0, 8.0) 6.0 (4.0, 8.0) 6.0 (4.0, 8.0) 5.5 (3.0, 7.0) 6.0 (4.0, 7.0) 6.0 (3.0, 8.0) 6.0 (4.0, 8.0) 6.0 (4.0, 8.0) 6.0 (4.0, 8.0) 6.0 (4.0, 8.0) 8.0 (5.0, 10.0)** 10.0 (6.0, 12.0) 8.0 (6.0, 11.0) 9.0 (6.0, 11.0) 9.0 (6.0, 11.0)* 7.0 (6.0, 10.0) 8.0 (6.0, 10.0) 9.0 (6.0, 11.0) 9.0 (6.0, 11.0) 9.0 (6.0, 11.0) 9.0 (6.0, 11.0) 9.0 (6.0, 11.0) 10.0 (6.0, 12.0)** 10.0 (7.0, 12.0) 10.5 (6.0, 12.3) 10.0 (7.0, 12.0) 10.0 (7.0, 12.0) 10.0 (6.0, 12.0) 15.0 10.0 (7.0, 12.0) 10.0 (6.0, 12.0) 10.0 (7.0, 13.0) 10.0 (6.0, 12.0) 10.0 (7.3, 13.0) ¹0.097** ¹0.118** 0.002 0.013 ¹0.054* ¹0.118** ¹0.072* ¹0.048 ¹0.072* ¹0.080 ¹0.113* ¹0.024 0.251** 0.157** 0.061* 0.030 5.0 (4.0, 8.0) 6.0 (4.0, 7.0) 6.0 (4.0, 8.0) 6.0 (4.0, 8.0) 8.0 (5.0, 10.0) 8.0 (6.0, 11.0) 9.0 (6.0, 11.0) 9.0 (6.0, 11.0) 10.0 (6.0, 12.0) 10.0 (6.0, 12.0) 10.0 (7.0, 12.0) 10.5 (7.8, 14.0) 0.251** 0.144** 0.112** 0.082 5.0 (3.8, 7.0) 6.0 (3.0, 7.0) 6.0 (4.0, 8.0) 6.0 (4.0, 8.0) 6.0 (4.0, 8.0) 6.0 (4.0, 7.0) 6.0 (4.0, 8.0) 6.0 (3.8, 8.0) 10.0 (6.3, 12.0) 9.0 (6.0, 11.0) 8.0 (6.0, 10.0) 9.0 (6.0, 11.0) 9.0 (6.0, 11.0) 9.0 (6.0, 11.0) 9.0 (6.0, 11.0) 9.0 (6.0, 11.0) 10.0 (7.0, 12.0) 10.0 (6.0, 12.0) 10.0 (6.0, 12.0) 10.0 (7.0, 12.0) 10.0 (6.0, 12.0) 10.0 (6.0, 12.0) 10.0 (7.0, 13.0) 10.0 (6.5, 12.0) IQR, interquartile range is the 75th and 25th percentile. r, Spearman’s rank correlation coefficient. *P < 0.05, **P < 0.01 by one-way ANOVA, »2 test, and Kruskal-Wallis test, which were used to compare data among all children, and among those stratified by school type children (Table 5 and Fig. 2). The effects of the interaction between exposure (age at first use of a mobile device and the duration) and the school type were statistically signifi- cant (Table 5, and Supplemental Table 4 and 8). The adjusted odds ratios of the five subscales—conduct problems, hyperactivity, emotional symptoms, peer prob- lems, and prosocial behavior—according to the logistic regression analysis results for all children and those strati- fied by the school type are shown in Table 6 and Fig. 2 (for the borderline/high and normal groups). Among all the children and among elementary school children only, the adjusted odds ratios for hyperactivity and peer problems significantly decreased with increasing child’s age at first use of a mobile device and decreasing child’s duration for the use of a mobile device. Among all the children and among those in elementary and junior high school, the adjusted odds ratios for emotional symptoms significantly decreased with increasing age at first use of a mobile de- vice and decreasing duration for use of a mobile device. Among all the children in senior high school, the adjusted Environmental Health and Preventive Medicine (2023) 28:22 7 of 11 Table 5 OR for child behavioral problems (TDS) according to child’s mobile device usage. Exposure All children OR (95% CI)a P for interaction Elementary school Junior high school Senior high school OR (95% CI)b OR (95% CI)b OR (95% CI)b Age at first use of a mobile device 0.85 (0.77, 0.93)** 0.009 1.20 (1.08, 1.33)** 0.005 Duration for use of a mobile device Age at first having personal mobile devices 0.776 0.94 (0.79, 1.11) 0.988 Duration for having personal mobile devices 1.10 (0.90, 1.34) TDS; total difficulties score. a (ALL); The logistic regression analysis models were adjusted for child age at time of survey, sex, history of developmental concerns, health-related quality of life, sleep problems, internet addiction, school type, and interaction between exposure and school type. b (Stratified by school type); The logistic regression analysis models were adjusted for child age at time of survey, sex, history of developmental concerns, health-related quality of life, sleep problems, and internet addiction. *P < 0.05, **P < 0.01 P for interaction; each exposure (child’s age at their first use of a mobile device and duration of use) and school type. 0.87 (0.81, 0.93)** 1.15 (1.08, 1.23)** 0.89 (0.77, 1.02) 1.13 (0.98, 1.30) 1.02 (0.96, 1.09) 0.98 (0.92, 1.04) 0.95 (0.85, 1.05) 1.06 (0.95, 1.17) 0.99 (0.91, 1.08) 1.01 (0.93, 1.10) 0.90 (0.80, 1.02) 1.11 (0.98, 1.25) odds ratios for conduct problems significantly decreased with increasing age at first use of personal mobile devices and decreasing duration for having personal mobile de- vices (Table 6). In the supplemental logistic regression analysis stratified by the children’s sex, boys who were younger at their first use of a mobile device and used such devices for longer durations were found to be hyperactive, have emotional instabilities, and experience peer problems in elementary school but displayed opposite behaviors in senior high school. Girls who were younger at their first use of mobile devices and used such devices for longer durations were found to have emotional instabilities in junior high school but displayed opposite behaviors in senior high school (Supplemental Tables 4–7). 4 Discussion In this cross-sectional study, children who were younger at their first use of a mobile device and used such devices for longer durations represented more problematic behaviors according to the SDQ. Moreover, when stratified by school type, the above associations remained statistically signifi- cant for elementary school children, but not for junior high school and older children (Table 5). Our results suggest that elementary school children are more sensitive to mo- bile device usage than junior high and senior high school children because they are in the early stages of socializa- tion and their behavioral and mental development is rap- idly growing. Four exposure factors were determined in this study: the child’s age at first use of a mobile device and the duration of use, and the child’s age at their first owning of a personal mobile device and the duration of owning. Given the rapid spread of mobile devices in recent years, elementary school children have started using such devices earlier than high school children did. By contrast, elementary school children used mobile devices for shorter periods than high school children. However, elementary school children represented more problematic behaviors, suggesting that starting to use mobile devices at an early age causes negative effects on developmental immature neural behaviors. Through a cross-sectional and longitudi- nal survey, the Danish National Birth Cohort has reported negative prenatal and postnatal effects of cell phone use on emotional and behavioral difficulties in children aged 7 and 11 [4, 5, 27]; our study corroborated the results of this study, showing that early exposure to mobile devices can cause developmental impacts on school-aged children. Using SDQ subscales, our study assessed problematic behaviors among school children aged 7–17 years. Ele- mentary school children who were younger at their first use of a mobile device and used these devices for longer durations had increased hyperactivity–inattention and dis- played peer and emotional problematic behaviors. More- over, junior high school children who were younger at their first use of a mobile device and used such devices for longer durations represented more emotional problem- atic behaviors. One possible reason for these differences in the relationships with the school type is that the contents of mobile device use may differ by the school type. Differ- ent effects with several content types were evaluated as screen time exposure, which has been noted as a risk factor for sensory development impacts, emotional and behavior- al problems, sleep disturbances, and internet addiction among school-aged children [6, 9]. A Swedish study tar- geting teenagers described that using mobile devices for social networking services is associated with increased communication skills as a positive effect; however, it is also associated with increased anxiety as a negative effect [10]. The results of our study corroborated those of the Swedish study, which stated that early exposure to mobile devices could exacerbate emotional instabilities in teen- agers. While the duration of personal mobile device usage did not differ by the children’s school type (Table 2), se- nior high school children who were younger and had a personal mobile device for a longer period represented more conducted problematic behaviors. This suggests a correlation between having a mobile device and exhibiting oppositional and defiant behaviors among high-school teenagers. When stratified by sex, only boys exhibited an association between mobile device usage and hyperactivity and peer problems. Moreover, associations between mo- bile device usage and emotional problems were observed Environmental Health and Preventive Medicine (2023) 28:22 8 of 11 Fig. 2 Behavioral problems in children and child age at first use of a mobile device OR: odds ratio. CI; confidence interval. TDS; total each difficulties score. The OR (95% CI) for children of borderline/high group, who was compared to children of normal group based on total each TDS and the five subscales—conduct problems, hyperactivity, emotional symptoms, peer problems, and prosocial behavior were calculated by the logistic regression analysis, which was adjusted for child age at time of survey, sex, history of developmental concerns, health-related quality of life, sleep problems, internet addiction, school type, and interaction between exposure and school type among all children. The logistic regression analysis among children stratified by school type was adjusted for the same variables excluding school type and interaction between exposure and school type. *P < 0.05, **P < 0.01 Environmental Health and Preventive Medicine (2023) 28:22 9 of 11 Table 6 OR of subscale of SDQ according to child’s mobile device usage. Exposure All children OR (95% CI)a Conduct problems Age at first use of a mobile device 0.97 (0.87, 1.08) Duration of mobile device use 1.02 (0.91, 1.14) Age at first having personal mobile devices 0.95 (0.78, 1.15) Duration of having personal mobile devices 1.07 (0.85, 1.33) Elementary school P for interaction OR (95% CI)b Junior high school Senior high school OR (95% CI)b OR (95% CI)b 0.543 0.741 0.696 0.831 0.99 (0.93, 1.06) 1.01 (0.94, 1.08) 0.91 (0.78, 1.06) 1.11 (0.95, 1.30) 1.01 (0.94, 1.08) 0.99 (0.93, 1.07) 0.95 (0.85, 1.08) 1.05 (0.93, 1.19) 1.01 (0.91, 1.12) 0.99 (0.89, 1.10) 0.87 (0.76, 0.99)* 1.15 (1.01, 1.32)* Hyperactivity/inattention Age at first use of a mobile device 0.89 (0.79, 0.99)* 0.173 0.264 Duration of mobile device use 1.12 (0.99, 1.25) Age at first having personal mobile devices 0.91 (0.74, 1.12) 0.829 0.875 Duration of having personal mobile devices 1.09 (0.86, 1.39) Age at first use of a mobile device Duration of mobile device use Age at first having personal mobile devices 1.01 (0.85, 1.19) Duration of having personal mobile devices 1.03 (0.85, 1.26) Emotional problems 0.91 (0.83, 1.00)* 0.541 1.15 (1.04, 1.26)** 0.129 0.652 0.996 Peer problems Age at first use of a mobile device 0.93 (0.85, 1.02) Duration of mobile device use 1.09 (0.99, 1.20) Age at first having personal mobile devices 0.90 (0.77, 1.06) Duration of having personal mobile devices 1.16 (0.96, 1.40) 0.490 0.350 0.191 0.130 0.93 (0.87, 1.00)* 1.08 (1.00, 1.15)* 0.94 (0.80, 1.11) 1.06 (0.90, 1.25) 0.96 (0.89, 1.04) 1.04 (0.96, 1.12) 0.91 (0.80, 1.04) 1.09 (0.96, 1.24) 1.01 (0.90, 1.13) 0.99 (0.88, 1.11) 0.95 (0.81, 1.11) 1.05 (0.90, 1.23) 0.92 (0.87, 0.98)* 1.08 (1.02, 1.15)* 0.97 (0.84, 1.11) 1.03 (0.90, 1.18) 0.92 (0.87, 0.97)** 1.09 (1.03, 1.15)** 0.98 (0.88, 1.08) 1.03 (0.92, 1.14) 0.99 (0.91, 1.08) 1.01 (0.93, 1.10) 0.97 (0.86, 1.10) 1.03 (0.91, 1.16) 0.93 (0.87, 0.99)* 1.07 (1.01, 1.15)* 0.93 (0.81, 1.07) 1.07 (0.93, 1.23) 0.97 (0.92, 1.02) 1.03 (0.98, 1.09) 0.96 (0.88, 1.06) 1.04 (0.94, 1.14) 1.00 (0.92, 1.07) 1.00 (0.93, 1.08) 1.13 (0.99, 1.28) 0.89 (0.78, 1.01) Prosocial behavior Age at first use of a mobile device 0.98 (0.91, 1.05) Duration of mobile device use 1.01 (0.95, 1.09) Age at first having personal mobile devices 1.05 (0.93, 1.19) Duration of having personal mobile devices 0.90 (0.77, 1.04) SDQ; The strengths and difficulties questionnaire. a (ALL); The logistic regression analysis models were adjusted for child age at time of survey, sex, history of developmental concerns, health-related quality of life, sleep problems, internet addiction, school type, and interaction between exposure and school type. b (Stratified by school type); The logistic regression analysis models were adjusted for child age at time of survey, sex, history of developmental concerns, health-related quality of life, sleep problems, and internet addiction. P for interaction; each exposure (child’s age at their first use of a mobile device and duration of use) and school type. *P < 0.05, **P < 0.01 0.98 (0.94, 1.03) 1.02 (0.97, 1.07) 1.05 (0.95, 1.17) 0.95 (0.85, 1.06) 1.01 (0.97, 1.05) 0.99 (0.95, 1.03) 1.04 (0.96, 1.12) 0.96 (0.89, 1.04) 1.00 (0.95, 1.06) 1.00 (0.95, 1.06) 0.99 (0.91, 1.07) 1.01 (0.93, 1.10) 0.513 0.690 0.606 0.235 in elementary school boys and junior high school girls (Supplemental Tables 4–7). The above results may be related to differences in developmental properties and tim- ing based on the children’s sex [16, 19]. The adverse effects of having personal mobile devices are inconclusive in this cross-sectional analysis, and a longitudinal evalua- tion throughout adolescence is needed in future studies. Parental supervision of their children’s use of mobile devices is recommended. However, the results of this study remained unchanged even after adjusting the paren- tal usage restrictions for weekdays and holidays (Supple- mental Tables 8 and 9); taken together, our study demon- strates the importance of delaying the use of mobile de- vices rather than imposing parental restrictions. In our study, 15.3% of the children aged ¯3 years were found to have used mobile devices according to their parents, which is lower than the rate of 50.2% for children who used the internet according to a 2019 Japanese survey [18]. This difference may be attributed to this study targeting a wide birth-year period of 10 years and excluding the use of internet with a wired connection (TV or personal com- puter). In our study, 69.7% of the elementary school-aged children had personal mobile devices, higher than the rates of 49.5% and 41.0% for smartphones and tablets, respec- tively, in 2019 of Japanese’s survey [18]. This difference may be attributed to the fact that this study included mo- bile game devices and other devices as well. In addition, during the survey period in 2020, the penetration of mobile devices among the school-aged children in Hokkaido pre- fecture increased because the administration started to pro- vide mobile devices to a part of school children for online classes. 4.1 Strengths This study was conducted from 2020 to 2021 and allowed us to examine recent associations between mobile device usage and behavioral problems in children aged 7–17 years. This study used four exposure factors, including the age at their first use of a mobile device and the duration of usage before or after owing a personal mobile device. Environmental Health and Preventive Medicine (2023) 28:22 10 of 11 This facilitated the evaluation of both early exposure to and having mobile devices. Previous studies have reported that the early use of mobile devices disrupts healthy activ- ities and increases the risk of sleep problems and internet addiction and lowers the health-related quality of life [2, 3, 31]. Children with developmental disorders often exhibit symptoms such as insomnia and anxiety [1]. These factors may be mutually related with mobile device usage and behavioral problems in children. Our results were obtained after adjusting for mutually related potential confounders, including sleep problems, internet addiction, health-related quality of life, and developmental concerns. 4.2 Limitations This cross-sectional study could not establish a clear causal direction. In fact, a previous study reported that children with developmental disorders are inclined toward heavy use of mobile devices owing to their weak self-regulation and the presence of restricted interests and repetitive behav- iors [14]. Also, parents of inattentive and hyperactive chil- dren are more likely to use mobile devices when calming down their children [20]. Furthermore, in this study, the parents retrospectively provided the age at which their chil- dren first used a mobile device and the duration of use; hence, recall bias may be possible. The differing results by school type may be associated with differences in the sample size, the rate of having personal mobile devices, and the varying content used among school types. 5 Conclusions Our findings suggest that elementary school children are more sensitive to mobile device usage than older children. Children who are younger at their first use of a mobile device and use such devices for longer durations may be prone to emotional instabilities as teenagers. Children who are younger and have had a personal mobile device for longer may show oppositional behaviors as teenagers. However, longitudinal follow-up studies are needed to clarify whether these problems disappear with age. Supplementary information The online version contains supplementary material available at https://doi.org/ 10.1265/ehpm.22-00245. Additional file 1: Supplemental Table 1. Distribution of child’s age at first use of a mobile device. Supplemental Table 2. Associations between child behavioral problems (TDS) and characteristics of participants (continuous variable). Supplemental Table 3. Associations between child behavioral problems (TDS) and characteristics of participants (categorical variable). Supplemental Table 4. OR for child behavioral problems (TDS) according to the child’s mobile device usage (boys). Supplemental Table 5. OR for subscales of SDQ according to children’s mobile device usage (boys). Sup- plemental Table 6. OR for child behavioral problems (TDS) according to child’s mobile device usage (girls). Supplemental Table 7. OR of subscale of SDQ according to child’s mobile device usage (girls). Supplemental Ta- ble 8. OR for child behavioral problems (TDS) according to child’s mobile device usage (Additional adjustment). Supplemental Table 9. OR of sub- scale of SDQ according to child’s mobile device usage (Additional adjust- ment). Declaration Ethics approval and consent to participate The institutional ethical board for human gene and genome studies at Hokkaido University Center for Environmental and Health Sciences (reference no. 139, August 30, 2022) and Hokkaido University Graduate School of Medicine (May 31, 2003) approved the study protocol. Written informed consent was obtained from all participants at the time of enrollment. Consent for publication Not applicable. Availability of data and material The datasets generated and/or analyzed during the current study are not publicly available because the study involves human participants with a nondisclosure provision of individual data stated in the written informed consent in order to prevent compromise of study participants’ privacy but are available from the corresponding author upon reasonable request. Competing interests The authors declare no conflict of interest. Funding This work was supported by the Grant-in-Aid for Health Science Research from the Japanese Ministry of Internal Affairs and Communications (JPMI10001). Authors’ contributions Reiko Kishi designed the study and developed the methodology. Chihiro Miyashita, Keiko Yamazaki, Naomi Tamura, and Atsuko Ikeda-Araki, collected the data and performed the analyses. Chihiro Miyashita, Keiko Yamazaki, and Naomi Tamura drafted the manuscript. Reiko Kishi, Satoshi Suyama, Takashi Hikage, Manabu Omiya, and Masahiro Mizuta provided critical revision of the manuscript. All authors take full responsibility for the content of this paper. All authors read and approved the final manuscript. Acknowledgements We would like to express our appreciation to all of the study participants of the Hokkaido Study on Environment and Children’ Health. We also express our profound gratitude to all personnel in the hospitals and clinics that collaborated including Sapporo Toho Hospital, Keiai Hospital, Endo Kikyo with the study, Maternity Clinic, Shiroishi Hospital, Memuro Municipal Hospital, Aoba Ladies Clinic, Obihiro-Kyokai Hospital, Akiyama Memorial Hospital, Sapporo Medical University Hospital, Hokkaido University Hospital, Kitami Red Cross Hospital, Hoyukai Sapporo Hospital, Gorinbashi Hospital, Hashimoto Clinic, Asahikawa Medical College Hospital, Hakodate Central General Hospital, Ohji General Hospital, Nakashibetsu Municipal Hospital, Sapporo Tokushukai Hospital, Asahi- kawa Red Cross Hospital, Wakkanai City Hospital, Kushiro Rosai Hospital, Sapporo-Kosei General Hospital, Shibetsu City General Hospital, Nikko Memorial Hospital, Sapporo City General Hospital, Kohnan Hospital, Hakodate City Hospital, Hokkaido Monbetsu Hospital, Tenshi Hospital, Hakodate Goryoukaku Hospital, Nakamura Hospital, Kin-ikyo Sapporo Hospital, Kitami Lady’s Clinic, Engaru-Kosei General Hospital, Kushiro Red Cross Hospital, Nayoro City General Hospital, and Obihiro-Kosei General Hospital. Received: 11 October 2022, Accepted: 3 March 2023 Published online: 12 April 2023 References 1. Byun YH, Ha M, Kwon HJ, Hong YC, Leem JH, Sakong J, et al. Mobile phone use, blood lead levels, and attention deficit hyperactivity symptoms in children: a longitudinal study. PLoS One. 2013;8:e59742. https://doi.org/ 10.1371/journal.pone.0059742. 2. Strasburger VC, Hogan MJ, Mulligan DA, Ameenuddin N, Christakis DA, Cross C, et al. Children, adolescents, and the media. Pediatrics. 2013;132: 958–61. https://doi.org/10.1542/peds.2013-2656. 3. Hill D, Ameenuddin N, Chassiakos R, Cross C, Hutchinson J, Levine A, et al. Media and young minds. Pediatrics. 2016;138:e20162591. https://doi. org/10.1542/peds.2016-2591. Environmental Health and Preventive Medicine (2023) 28:22 11 of 11 4. Divan HA, Kheifets L, Obel C, Olsen J. Cell phone use and behavioural problems in young children. J Epidemiol Community Health. 2012;66: 524–9. https://doi.org/10.1136/jech.2010.115402. 5. Divan HA, Kheifets L, Obel C, Olsen J. Prenatal and postnatal exposure to cell phone use and behavioral problems in children. J Epidemiol. 2008;19: S94–SS5. 6. Eirich R, McArthur BA, Anhorn C, McGuinness C, Christakis DA, Madigan S. Association of screen time with internalizing and externalizing behavior problems in children 12 years or younger: A systematic review and meta- analysis. JAMA Psychiatry. 2022;79:393–405. https://doi.org/10.1001/ jamapsychiatry.2022.0155. 7. Goodman R. The strengths and difficulties questionnaire: a research note. J Child Psychol Psychiatry. 1997;38:581–6. https://doi.org/10.1111/j.1469- 7610.1997.tb01545.x. 8. Goodman R. Psychometric properties of the strengths and difficulties questionnaire. J Am Acad Child Adolesc Psychiatry. 2001;40:1337–45. https://doi.org/10.1097/00004583-200111000-00015. 9. Hosokawa R, Katsura T. Association between mobile technology use and in early elementary school age. PLoS One. 2018;13: child adjustment e0199959. https://doi.org/10.1371/journal.pone.0199959. 10. Jenkins RH, Shen C, Dumontheil I, Thomas MSC, Elliott P, Röösli M, et al. Social networking site use in young adolescents: association with health- related quality of life and behavioural difficulties. Comput Human Behav. 2020;109. https://doi.org/10.1016/j.chb.2020.106320. 11. Kishi R, Araki A, Minatoya M, Hanaoka T, Miyashita C, Itoh S, et al. The Hokkaido Birth Cohort Study on Environment and Children’s Health: cohort profile-updated 2017. Environ Health Prev Med. 2017;22:46. https://doi.org/ 10.1186/s12199-017-0654-3. 12. Kishi R, Ikeda-Araki A, Miyashita C, Itoh S, Kobayashi S, Ait Bamai Y, et al. Hokkaido birth cohort study on environment and children’s health: cohort profile 2021. Environ Health Prev Med. 2021;26:59. https://doi.org/10.1186/ s12199-021-00980-y. 13. Kishi R, Kobayashi S, Ikeno T, Araki A, Miyashita C, Itoh S, et al. Ten years of progress in the Hokkaido birth cohort study on environment and children’s health: cohort profile—updated 2013. Environ Health Prev Med. 2013;18:429–50. https://doi.org/10.1007/s12199-013-0357-3. 14. Lawrence AC, Narayan MS, Choe DE. Association of Young Children’s Use of Mobile devices with their self-regulation. JAMA Pediatr. 2020;174:793–5. https://doi.org/10.1001/jamapediatrics.2020.0129. 15. Lin HP, Chen KL, Chou W, Yuan KS, Yen SY, Chen YS, et al. Prolonged touch screen device usage is associated with emotional and behavioral problems, but not language delay, in toddlers. Infant Behav Dev. 2020;58. https://doi.org/10.1016/j.infbeh.2020.101424. 16. Matsuishi T, Nagano M, Araki Y, Tanaka Y, Iwasaki M, Yamashita Y, et al. Scale properties of the Japanese version of the strengths and difficulties questionnaire (SDQ): a study of infant and school children in community samples. Brain Dev. 2008;30:410–5. https://doi.org/10.1016/j.braindev. 2007.12.003. 17. McHarg G, Ribner AD, Devine RT, Hughes C, Blair C, Hughes C. Infant screen exposure links to toddlers’ inhibition, but not other EF constructs: A propensity score study. Infancy. 2020;25:205–22. https://doi.org/10.1111/ infa.12325. 18. Japanese Cabinet Office; 2020. Survey on internet usage environment for young people in Japan. https://www8.cao.go.jp/youth/kankyou/internet_ torikumi/tyousa/r01/net-jittai/pdf/sokuhou.pdf. Accessed Jan 21, 2023 (in Japanese). 19. Moriwaki A, Kamio Y. Normative data and psychometric properties of the strengths and difficulties questionnaire among Japanese school-aged children. Child Adolesc Psychiatry Ment Health. 2014;8:1. https://doi.org/ 10.1186/1753-2000-8-1. 20. Muñoz-Silva A, Lago-Urbano R, Sanchez-Garcia M, Carmona-Márquez J. Child/adolescent’s ADHD and parenting stress: the mediating role of family impact and conduct problems. Front Psychol. 2017;8:2252. https://doi.org/ 10.3389/fpsyg.2017.02252. 21. Nezu S, Iwasaka H, Saeki K, Obayashi K, Ishizuka R, Goma H, et al. Reliability and validity of Japanese versions of KIDSCREEN-27 and KIDSCREEN-10 questionnaires. Environ Health Prev Med. 2016;21:154– 63. https://doi.org/10.1007/s12199-016-0510-x. 22. Okada M, Kitamura S, Iwadare Y, Tachimori H, Kamei Y, Higuchi S, et al. Reliability and validity of a brief sleep questionnaire for children in Japan. J Physiol Anthropol. 2017;36:35. https://doi.org/10.1186/s40101-017-0151-9. 23. Okely AD, Ghersi D, Hesketh KD, Santos R, Loughran SP, Cliff DP, et al. A collaborative approach to adopting/adapting guidelines—the Australian 24- hour Movement Guidelines for the early years (birth to 5 years): an integration of physical activity, sedentary behavior, and sleep. BMC Public Health. 2017;17:869. https://doi.org/10.1186/s12889-017-4867-6. 24. Poushter J, Bishop C, Chwe H. Social media use continues to rise in developing countries but plateaus across developed ones; 2018. https:// assets.pewresearch.org/wp-content/uploads/sites/2/2018/06/15135408/ Pew-Research-Center_Global-Tech-Social-Media-Use_2018.06.19.pdf. Pew Research Center. 25. Ravens-Sieberer U, Auquier P, Erhart M, Gosch A, Rajmil L, Bruil J, et al. The KIDSCREEN-27 quality of life measure for children and adolescents: psychometric results from a cross-cultural survey in 13 European countries. Qual Life Res. 2007;16:1347–56. https://doi.org/10.1007/s11136-007- 9240-2. 26. Roser K, Schoeni A, Röösli M. Mobile phone use, behavioural problems and concentration capacity in adolescents: A prospective study. Int J Hyg Environ Health. 2016;219:759–69. https://doi.org/10.1016/j.ijheh.2016.08. 007. 27. Sudan M, Olsen J, Arah OA, Obel C, Kheifets L. Prospective cohort analysis of cellphone use and emotional and behavioural difficulties in children. J Epidemiol Community Health. 2016;70:1207–13. https://doi.org/ 10.1136/jech-2016-207419. 28. Thomas S, Benke G, Dimitriadis C, Inyang I, Sim MR, Wolfe R, et al. Use of mobile phones and changes in cognitive function in adolescents. Occup Environ Med. 2010;67:861–6. https://doi.org/10.1136/oem.2009.054080. 29. Tobe H, Takeuchi K, Hori M. The relationship between the tendency toward internet dependence and mental health and the psychosocial problems of students. Jpn J Sch Health. 2010;52:125–34. 30. Tremblay MS, Chaput JP, Adamo KB, Aubert S, Barnes JD, Choquette L, et al. Canadian 24-hour movement guidelines for the early years (0–4 years): an integration of physical activity, sedentary behaviour, and sleep. BMC Public Health. 2017;17:874. https://doi.org/10.1186/s12889-017- 4859-6. 31. World Health Organization. Guidelines on physical activity, sedentary behaviour and sleep for children under 5 years of age. Geneva: World Health Organization; 2019. 32. Young KS. Caught in the net: how to recognize the sign of Internet addic- tion and a winning strategy for recovery. New York: John Wiley & Sons. (Odashima Y, Mainichi Publishing Co.; 1998, translated in Japanese); 1996.
null
10.1088_1402-4896_ad0bb9.pdf
Data availability statement The data cannot be made publicly available upon publication because they are owned by a third party and the terms of use prevent public distribution. The data that support the findings of this study are available upon reasonable request from the authors.
Data availability statement The data cannot be made publicly available upon publication because they are owned by a third party and the terms of use prevent public distribution. The data that support the findings of this study are available upon reasonable request from the authors.
Phys. Scr. 98 (2023) 125956 https://doi.org/10.1088/1402-4896/ad0bb9 RECEIVED 26 July 2023 REVISED 27 October 2023 ACCEPTED FOR PUBLICATION 10 November 2023 PUBLISHED 22 November 2023 PAPER Samarium-modified NBT ceramics: a comprehensive exploration of cumulative effects Jyothi Neeli1, Nitchal Kiran Jaladi1 Srinivasa Rao Kurapati2 1 Department of Physics, School of Applied Sciences and Humanities, Vignan’s Foundation for Science Technology and Research, , Nagamani Sangula1, Vijaya Lakshmi Garlapati1 and Vadlamudi, Guntur-522 213, A.P. India 2 Department of Physics, P.B.N college, Nidubrolu, 522124, A.P, India E-mail: kiran.nischal@gmail.com Keywords: rietveld analysis, band gap, color coordinates, vickers hardness, coefficient of friction and wear, VSM study Abstract In the present report, ceramic specimens of sodium bismuth titanate [Na0.5(Bi1-xSmx)0.5TiO3] were prepared through solid-state reaction method with variations in the dopant concentrations specifically, x = 0.0, 0.1, 0.3, 0.5. The structural, optical, mechanical, and magnetic properties of lead- free NBT ceramics were investigated. The rhombohedral phase with space group R3c was confirmed in all prepared ceramic samples using X-ray diffraction patterns and Rietveld analysis. SEM micrographs and Energy Dispersive x-ray spectroscopy (EDAX) assess the morphology, grain size, overall structure, and stoichiometry of the developed compounds. FTIR spectroscopy was used for characterizing and identifying the functional groups. UV–vis spectroscopy revealed that band gap values decreased as dopant concentration increased, confirming the use of NBT-based perovskite as a photoactive material. PL spectra at room temperature exhibited reddish-orange emission. Colour coordinates and CCT values are in the range of 3483 K to 5912 K. At a concentration of x = 0.3, the materials displayed a high Vickers hardness of 8.20 GPa and exhibited minimal wear with low frictional coefficient values. Ferromagnetic behaviour at room temperature (RTFM) was detected in Sm-modified ceramic samples, as confirmed by the VSM study. The cumulative effect impact of the rare earth dopant cation at the Bi-site of NBT was widespread and demonstrated significant potential for use in optoelectronic devices. 1. Introduction The remarkable advancement in science and technology, which has greatly enhanced our quality of life and made it intricately connected with the evolution of essential materials, has led to a concerted effort to create materials that meet our needs. In pursuit of this goal, extensive research has been conducted to uncover the intricate connection between the structural attributes of materials and their inherent properties. Over the past fifty years, lead-based ferroelectric (FE) and piezoelectric materials have been the subject of intense study due to their wide variety of potential uses [1, 2]. Several classes of materials are explored as potentially attractive alternates to lead-based materials, such as KNN, BST, BNT, alkaline niobates and non-perovskite BLSF structures in order to develop environmental friendly lead-free materials as regulated by the European Union in 2004 [3]. However, researchers have yet to fully comprehend the structure–property correlations of NBT since Smolenskii et al discovered it in 1960. Na0.5Bi0.5TiO3, a lead-free perovskite-type ceramic with the general formula ABO3 and rhombohedral symmetry consisting of Na at the B-site, where oxygen is octahedrally linked to the B site and the A site is 12-fold coordinated with oxygen has received much attention [4]. It has an ambient temperature rhombohedral structure with a space group of R 3c. At 250 °C, this phase changes into a tetragonal phase; at 520 °C, it becomes a cubic phase [5–8]. It exhibits residual polarization of 38 μC/cm2 and possesses a high curie temperature of 320 °C [9]. Enhanced characteristics of NBT with appropriate doping at A-site were illustrated by He et al, and Bi-site doping was preferred to obtain superior at the A-site, and Ti4+ , Bi3+ + © 2023 IOP Publishing Ltd Phys. Scr. 98 (2023) 125956 J Neeli et al performance of NBT [10–12]. Rare-earth (Gd, Yb, Eu, Dy, Nd, and Er) doped NBT systems have previously been developed by researchers to investigate their optical and magnetic properties. The addition of rare earth ions to the A site of the NBT sample changes the number of available cations. It improves the polarization of the electric charge by increasing the oxygen vacancies in the ceramic matrix [13]. From the addition of the Gd3+ ion at the Bi-site of NBT (x = 0.00–0.02), the paramagnetic behaviour in NBT was distinguished from the diamagnetic behaviour [14]. E L T França et al [15] proposed the large dielectric constant and the Yb-doped (x = 0.005–0.020) NBT ceramics exhibit reduced dielectric losses, which makes them suitable for use in technological applications. Santosh Beharaa et al introduced a Eu3+ doped NBT (x = 0.0–0.08) that offers possibilities for exploiting concentration-dependent amphotericity in luminescence device applications. Further, proposed Dy3+ NBT (x = 0.00–0.14) amphoteric and site-dependent luminescence variance, which exhibits a strong correlation between structure–property relationships in material chemistry and holds promise for use in pc-WLEDs and optoelectronic devices [16, 17]. Kumara Raja Kandula et al, examined the effects of Nd3+ multifunctional properties of Na0.5[Bi1-xNdx]0.5TiO3 (NBT) [18]. When an erbium ion is present in the A-site, Na0.5[Bi1-xErx]0.5TiO3 improves the optical and ferroelectric properties [19]. T.Wei et al, introduced Sm3+ doped NBT (x = 0.0–0.16); it may act as a potentially multifunctional optical-electro-material [20]. S. Lenka et al, have examined the effect of the Sm3+ ion at the Bi-site of NBT, which is suitable for high-temperature applications [21]. Based on the previous literature for lead-free piezoelectric systems other than doped BT [22, 23], there is still a lack of knowledge of defect chemistry and the studies were limited to structural, optical, dielectric, and piezoelectric properties with low concentration rare earth doping at Bi-site of NBT. The novelty of this work is to study a cumulative effect of high-concentration rare earth doping at the Bi-site of NBT to unravel the structural, optical, magnetic and mechanical properties and hence to examine their viability as optoelectronic devices, magnetic memory materials and wear-resistant tribo-materials. The multifunctional attributes of Sm-modified NBT materials are relevant for electronic, automotive, aerospace, and medical industries owing to the stability, reliability, and flexibility in meeting specific performance criteria by fine-tuning the concentration of Sm dopant to derive high performance. ion doping on the doped 2. Experimental procedure 2.1. Synthesis Polycrystalline ceramic samples of Na0.5Bi0.5TiO3 (NBT) and Sm-modified Na0.5(Bi1-xSmx)0.5TiO3 with x = 0.0, 0.1, 0.3, and 0.5 (abbreviated as NBT, NBS1T, NBS3T, and NBS5T) ceramic samples from solid-state reaction method. Stoichiometric amounts of powders of oxides Bi2O3 (99.0%), TiO2 (99.0%), and Sm2O3 (99.9%) and carbonates Na2CO3 (99.0%), manufactured by high-media were utilized to prepare. The powders were triturated for 8 h in an agate mortar using methanol as a mixing medium. The resulting powders were double calcined for 3 h at 850 °C to increase their homogeneity. The pellets with a 10 mm diameter and 1 mm thickness were prepared by using hydraulically pressing polyvinyl alcohol (PVA) mixed calcined powders. The pellets are then subjected to sintering for 3 h at 1150 °C in the the air with a heating and cooling rates of 5° per minute. 2.2. Characterization The translucent design of the samples were analyzed utilizing an XRD machine RigakuMiniflex 300/600 with CuKα radiation (λ = 1.541 Å) with a step size of 2°/min, considering 2θ in the range of 0°−90°. Rietveld refinement of XRD data has been conducted using FullProf software [24]. The three-dimensional crystal structures were set up using VESTA programming. Scanning electron microscopy (SEM) (vega3tescan) and Energy dispersive spectroscopy (EDAX) are used to confirm the phase purity and stoichiometry of the prepared compassions.YLS-QC-WQP-004 was used for the FTIR spectroscopic analysis of the powders in KBr medium, −1. The absorption spectrum was acquired through diffuse reflectance covering a spectral range of 4000–400 cm measurements on the prepared ceramic materials using a UV–vis spectrometer (Analytik Jena, SPECORD 210 PLUS). Horiba JobinYvon Fluorolog-3–21 was employed to capture photoluminescence spectra at room temperature. Ferromagnetism at room temperature (RTFM) was evaluated using a MicroSense-20130523-01 vibrating sample magnetometer (VSM). Mechanical characteristics, including Vicker’s hardness (HV), wear coefficient (K), coefficient of friction (μ), and specific wear Rate and Specific wear energy, were assessed with the Tribometer-201. 2 Phys. Scr. 98 (2023) 125956 J Neeli et al Figure 1. (a) XRD patterns of NBT, NBS1T, NBS3T, NBS5T,and (b) Enlarged XRD patterns in 2θ range from 32° to 34° for NBT, NBS1T, NBS3T, NBS5T ceramic samples. Table 1. The crystallite size, and Lattice strain, Tolerance factor values for NBT, NBS1T, NBS3T, and NBS5T ceramics. S.No Sample 1 2 3 4 NBT NBS1T NBS3T NBS5T Crystallite size (nm) 30.97 35.16 23.83 17.15 Lattice strain 0.0034 0.0031 0.0038 0.0053 Tolerance factor 0.9716 0.9695 0.9655 0.9616 3. Results and discussions 3.1. Structural analysis 3.1.1. X-ray diffraction analysis (XRD) XRD patterns of sodium bismuth titanate (NBT) and Sm-modified NBT systems are shown in figure 1. Employed Rietveld refinement method, examined all the prepared ceramic samples and verified the presence of rhombohedral structure with R 3c space group, confirming the successful diffusion of samarium into NBT. In samarium-modified NBT systems, an increase in the dopant concentration resulted in a noticeable shift in high- intensity diffraction peaks, which can be attributed to the ionic radii of the Sm3+ ion. Considering the ionic radii of the samarium (1.24 Å), bismuth (1.32 Å), and sodium (1.39 Å), it is evident that Sm3+ ions are readily able to substitute for Bi3+ formula allowed us to establish the average crystal size, which clearly signifies the diminishing trend in crystallite size. The crystallization percentage in 2θ range of 32 to 34° reiterates that there is a decrease with increasing in the concentration of samarium doping [26]. The lattice strain in all the studied materials was computed and observed to be increasing. The higher doping content has been linked to the decrease in crystallite size as presented in table 1. The Goldschmidt tolerance factor (tg) used to examine the structural stability of the compounds is expressed as [27] ions within a pristine NBT system [25]. The application of the Debye–Scherrer and Na1+ - + ) [( g t x 1 = Ra 1 ( r 2 B Where x is the dopant concentration, Ra1 is basic elements(Na, Bi) in A-site average ionic radii, Ra2 is the dopant element(Sm) in A-site ionic radii, rB is ionic radii of the B-site atoms, and ro is the ionic radius of oxygen. Goldschmidt tolerance factor (tg) values are listed in table 1. Incorporating Sm3+ place of larger ionic radii of Bi3+ provides evidence of reduced lattice parameters and a diminished unit cell volume in the examined samples. The ions caused the unit cells of Sm-doped NBT systems to contract [28]. Table 2 ions with smaller ionic radii in Ra 2 ) r o x + r o ] + ( ) 1 3 Phys. Scr. 98 (2023) 125956 J Neeli et al Table 2. Details of lattice parameters, cell volume, atomic position, occupancy, and goodness parameter values for NBT and samarium modified NBT ceramic samples. Compound NBT NBS1T NBS3T NBS5T Lattice Parameter a (Å) c (Å) Cell Volume(Å) Atomic Positions Na X Y Z Occupancy Bi X Y Z Occupancy Sm X Y Z Occupancy Ti X Y Z Occupancy O X Y Z Occupancy χ2 5.49287; 13.42146 350.695 5.48570 ± 0.00103 13.4180 ± 0.00480 349.587 ± 0.156 5.462 ± 0.00061 13.4545 ± 0.00192 347.680 ± 0.074 5.44375 ± 0.00068 13.4148 ± 0.00226 344.280 ± 0.00226 0.00000 0.00000 0.26311 0.143 0.00000 0.00000 0.23311 0.499 — — — — 0.00000 0.00000 0.01403 0.985 0.10863 0.33800 0.09333 1.510 0.926 0.00000 0.00000 0.26069 0.099 0.00000 0.00000 0.26069 0.449 0.00000 0.00000 0.26069 0.050 0.00000 0.00000 0.01164 0.975 0.0803 0.33741 0.06656 1.569 0.728 0.00000 0.00000 0.26338 0.165 0.00000 0.00000 0.26338 0.350 0.00000 0.00000 0.26338 0.149 0.00000 0.00000 0.01188 0.979 0.10899 0.33763 0.09459 1.216 0.970 0.00000 0.00000 0.26473 0.215 0.00000 0.00000 0.26473 0.250 0.00000 0.00000 0.26473 0.249 0.00000 0.00000 0.00978 0.974 0.10867 0.33810 0.07089 1.279 0.829 Full-Prof method was used for conducting the Rietveld refinement to analyze the crystal structure of each sample. The samples’ Rietveld refinement patterns are shown in figure 2. A nominal difference was noticed between the observed and calculated data and Bragg’s reflections of all the prepared ceramic samples were the same. The goodness parameter was observed to be less than 1, indicating a good fit of the observed and calculated values. Table 2 displays the refined lattice parameters, atomic positions, and goodness of fit χ2, while table 3 presents the instrumental parameters, residual factor Rp, and weighted residual factor Rw. VESTA software was used for the visualization of the crystal structure of pure NBT, Sm3+ ion-modified composition. Figure 3 represents the three-dimensional crystal structures of the NBT, NBS1T, NBS3T, and NBS5T. The Ti occupancy was nearly the same in all compositions. Average bond lengths were observed for all compositions, and the average bond length of A-site and B-site atoms to oxygen was observed to decrease with dopant concentration. Bond angles (O-Ti-O) variation was observed for all compositions.The Ti4+ occupy an octahedral position; Ti-O distances exhibit short and long lengths that result in deformed octahedra; the average value of Ti4+ -O is around 1.96 Å. These octahedra are stretched in three dimensions and alternately joined. The analysis of various inter-atomic distances in table 4 reveals that Na/Bi/Sm atoms form (Na/Bi/Sm)O12 polyhedra, and the calculated average length of the Na/Bi/Sm–O bond is approximately 2.65 Å. cations 3.1.2. Density and microstructural studies The density of the prepared ceramic samples was determined by using the Archimedes water displacement equipment (model: TTB15) [29] and identified to possess high density (95%) with induced porosity, shown in table 5. In order to gain insight into the materials’s mechanical properties performed shrinkage measurements [30] on the pellets after sintering and noticed that the percentage density was to the tune of induced porosity. The studied samples’ SEM micrographs and corresponding histogram distribution are shown in figure 4. The average grain size of each sample has been calculated using ImageJ software. The determined average grain size of the NBT, NBS1T, NBS3T, and NBS5T compositions was 2.65 μm, 4.86 μm, 5.27 μm, and 4.13μm, 4 Phys. Scr. 98 (2023) 125956 J Neeli et al Figure 2. Rietveld refinement XRD patterns of NBT and samarium-modified NBT ceramic samples. Table 3. Details of Rietveld refinement parameters NBT, NBS1T, NBS3T, and NBS5T ceramic compositions. Compound Wavelength (Å) Step scan increment 2θ range (°) Program Caglioti parameters Pseudo-Voigt function PV = ɳ L + (1 − ɳ) G Space group RF RB Rp Rw Rexp NBT NBS1T NBS3T NBS5T 1.541862 0.02 20–80 FULLPROF U = 0.091985 V = −0.000528 W = 0.008432 ɳ = 0.63315 R −3 c 14.1 13.6 12.0 16.6 17.26 1.541862 0.02 20–80 FULLPROF U = 0.013862 V = −0.009936 W = 0.006103 ɳ = 1.83336 R −3 c 12.5 11.5 11.4 15.1 1.68 1.541862 0.02 20–80 FULLPROF U = 0.016799 V = −0.009915 W = 0.005638 ɳ = 3.40818 R −3 c 13.5 12.1 13 16.4 16.66 1.541862 0.02 20–80 FULLPROF U = 0.004133 V = −0.007618 W = 0.006255 ɳ = 0.0001 R −3 c 12.6 12.7 11.9 15.1 16.61 respectively. The increase in grain size can be attributed to the sintering mechanism, in which the surface free energy of particles decreases due to the solid–vapour interface energy and solid-state interface energy contributions along with the diffusion of rare earth ions in the NBT host matrix and the dopant’s ionic radii and atomic mass [31, 32]. Among various rare earth elements, samarium is considered as one of the excellent dopants due to its optical capacity [33]. In general, there is a strong correlation between changes in the structure and morphology of the material and its optical properties [34]. This prompted the use of Sm3+ luminophore in the NBT matrix. A Photoluminescence on an Sm3+ ion-modified NBT system was deciphered by the intensity correlation between the studied optical properties and structural parameters (Rietveld analysis). Energy Dispersive X-Ray Spectroscopy (EDAX): figure 5(a)–(d) depicts the Energy Dispersive x-ray spectra ions as a (EDAX) and atomic, weight percentages of each constituent atom for all compositions, confirming the purity and stoichiometry. The XRD results confirmed the presence of only the rhombohedral perovskite phase, and this corroborates the fact that sufficient dissolution of samarium into the host matrix has occurred. 5 Phys. Scr. 98 (2023) 125956 J Neeli et al Table 4. Selected inter-atomic distances (Å) and O-Ti-O angles for NBT, NBS1T, NBS3T and NBS5T. Bond length Na/Bi-O Na/Bi-O Na/Bi-O Na/Bi-O Na/Bi-O Na/Bi-O <Na/Bi-O> Na/Bi/Sm -O Na/Bi/Sm -O Na/Bi/Sm -O Na/Bi/Sm -O Na/Bi/Sm -O Na/Bi/Sm -O 2.53134 2.71532 2.41529 2.43476 2.80849 3.00696 2.662466 2.665 2.8442 2.3566 2.3839 2.6902 3.0824 <Na/Bi/Sm –O> 2.6373 Na/Bi/Sm -O Na/Bi/Sm -O Na/Bi/Sm -O Na/Bi/Sm -O Na/Bi/Sm -O Na/Bi/Sm -O 2.4048 2.4273 2.7260 2.5119 3.0257 2.7956 <Na/Bi/Sm –O> 2.64855 Na/Bi/Sm -O Na/Bi/Sm -O Na/Bi/Sm -O Na/Bi/Sm -O Na/Bi/Sm -O Na/Bi/Sm -O 2.45 2.74 3.07 2.77 2.3935 2.421 NBT Ti-O Ti-O Ti-O Ti-O <Ti-O> NBS1T Ti-O Ti-O Ti-O Ti-O <Ti-O> NBS3T Ti-O Ti-O Ti-O Ti-O <Ti-O> NBS5T Ti-O Ti-O Ti-O Ti-O 1.98225 2.818436 1.78548 1.95351 1.97715 1.8062 2.0013 1.9545 2.2009 1.94275 2.17027 1.95084 1.97392 1.78487 1.97166 1.96 2.12 1.823 1.96 <Na/Bi/Sm –O> 2.6405 <Ti-O> 1.96575 Bond angles O-Ti-O O-Ti-O O-Ti-O O-Ti-O O-Ti-O O-Ti-O O-Ti-O O-Ti-O O-Ti-O O-Ti-O O-Ti-O O-Ti-O O-Ti-O O-Ti-O O-Ti-O O-Ti-O O-Ti-O O-Ti-O O-Ti-O O-Ti-O O-Ti-O O-Ti-O O-Ti-O O-Ti-O O-Ti-O O-Ti-O O-Ti-O O-Ti-O O-Ti-O 176.6785 93.2166 89.2448 87.3889 88.8680 88.430 104.505 78.709 93.895 88.579 86.950 86.575 88.960 89.242 88.154 87.813 89.584 89.833 81.169 101.019 91.332 88.9 101.3 80.5 86.45 89.21 88.3 87.9 89.59 −1, ∼1460 cm 3.1.3. Fourier infrared spectroscopy (FTIR) The FT-IR spectra of Na0.5(Bi1-xSmx)0.5TiO3 at room temperature with x values of 0.0, 0.1, 0.3, and 0.5 are −1. The vibrational bands are observed at wave depicted in figure 6 with spectral range from 400 to 4000 cm −1 and ∼3477 cm −1, ∼2923 cm −1, ∼1649 cm −1, ∼834 cm numbers ∼636 cm The metal-oxide band observed close to 636 cm −1. The stretching vibration vibrations of TiO6 octahedra may be responsible for the absorption band at 834 cm of BO6 octahedra in the perovskite structure for B–O bonds along the c-axis is represented by a weak absorption band at 1460 cm extending vibrational methods of the caught water molecule. The shift and broadening of vibrational bands with dopant concentration correspond to the change in lattice parameters attributed to crystal growth [36, 37]. −1 is typical for perovskite materials [35]. The Ti–O–Ti −1 [35]. The band at 2923 cm −1 are aligned with the O-H −1 and around 3477 cm −1 for NBT sample. −1, 1649 cm 3.2. Optical studies 3.2.1. Ultraviolet-visible (UV–vis) reflectance spectroscopy(DRS) The ultraviolet–visible absorption spectra of all the prepared samples within the 200–800 nm wavelength range as shown in figure 7 illustrate the interband photon energy absorption leads to the optical conduction of electrons. As the dopant concentration increase, it was observed that the wavelength at which maximum absorption occurred consistently shifted towards a longer wavelengths. Using the Tauc equation, the optical band gaps (Eg) of Sm3+ ion-doped NBT materials have been determined [38]. 6 Phys. Scr. 98 (2023) 125956 J Neeli et al Figure 3. VESTA software developed NBT and Sm-modified NBT structures from Rietveld refinement. Figure 4. SEM micrographs and corresponding histograms for NBT, NBS1T, NBS3T, and NBS5T. a n = n E ( h h – n ) g ( ) 2 Where the absorption coefficient(α), Planck’s constant(h), incident photon frequency(ν), and a material- dependent constant (A) are present, the nature of the electronic transitions is revealed by the power index (n) value. Direct and indirect interband transitions are expressed by the numbers n = 1/2 and 2, respectively. In the case of NBT perovskite materials, there is a direct charge shift from the valence band’s higher edge to the conduction band’s lower edge. In addition, the theoretical research by Zeng et al on basic NBT perovskite material indicated that the O-2p state is at the top of the valence band while the Ti-3d and Bi-6s states are at the bottom of the conduction band for direct interband transitions [39]. Figure 8 exhibits linear fits for direct band gap (n = ½) for all the ceramic samples produced elucidating the correlation between (αhv)2 versus hv plots. The direct band gap values of the NBT and samarium-modified ceramics were determined to fall within the descending range of 3.01 to 3.12 eV from that of NBT suggesting that the samples exhibit semiconducting properties, shown in table 6. A similar trend was observed by the earlier researchers [14, 40, 41]. This decrease in 7 Phys. Scr. 98 (2023) 125956 J Neeli et al Figure 5. EDAX spectra and corresponding atomic and weight percentages for NBT and Samarium modified NBT ceramics. Figure 6. FTIR spectra of all the prepared samples. 8 Table 5. Grain size, Relative density, and porosity values for NBT and Sm-modified NBT ceramic samples. 9 S. No Composition Experimental Density (g/cm3) Theoretical Density (g/cm3) Percentage of Shrinkage diameter (%) Grain size (μm) Relative Density (%) Porosity (%) 1 2 3 4 NBT NBS1T NBS3T NBS5T 6.158 5.217 5.529 5.273 6.580 5.497 5.826 5.572 14.1 15.8 14.4 15.3 2.65 4.86 5.27 4.13 95 95 95 95 0.049 0.051 0.051 0.054 P h y s . S c r . 9 8 ( 2 0 2 3 ) 1 2 5 9 5 6 J N e e l i e t a l Phys. Scr. 98 (2023) 125956 J Neeli et al Figure 7. Diffuse reflectance spectra (DRS) for NBSmxT with x = 0.0, 0.1, 0.3, and 0.5. Figure 8. Linear fitting for the direct band gap values of NBSmxT with x = 0.0, 0.1, 0.3, and 0.5. band gap value may have been caused by the formation of new subbands within the band gap of NBT material. Additionally, it may be linked to the microstructural changes brought on by variations in dopant concentration. Among the studied ceramic compositions, the NBS3T had the lowest optical bang gap (Eg = 3.01 eV), and it might be a promising perovskite photoactive material [42]. 3.2.2. Photoluminescence study In order to investigate the photoluminescence (PL) characteristics of Na0.5Bi0.5TiO3 and Na0.5(Bi1-xSmx)0.5TiO3 (where x = 0.0, 0.1, 0.3, and 0.5), the samples were subjected to excitation at a wavelength of 407 nm. The resulting photo emissions were then examined within the range of 550 nm to 675 nm, as depicted in figure 9(a). The results showed that the photoluminescence intensity was proportional to Sm concentration initially and then reached the highest value at x = 0.3 (critical limit). Once the critical concentration of Sm is surpassed the 10 Phys. Scr. 98 (2023) 125956 J Neeli et al Figure 9. (a) Photoluminescence emission spectra for Na0.5(Bi1-xSmx)0.5TiO3 at x = 0.0, 0.1, 0.3, 0.5 (b) CIE chromaticity diagram. Table 6. Direct energy band gap values for NBT and Sm-modified NBT ceramics. S. No Composition Direct band gap Eg (eV) 1 2 3 4 NBT NBS1T NBS3T NBS5T 3.12 3.09 3.01 3.07 ions reduces, leading to mutual ion interactions that induce non-radiative transitions. distance between Sm3+ This phenomenon subsequently diminished the fluorescence intensity. The decrease in photoluminescence intensity after the addition of excessive amounts of Sm can be attributed to the concentration quenching effect [43, 44]. The Blasse equation (3) can be used to determine the critical separation between two Sm3+ ions and is written as [45] =R 3 V p X N 4 C ⎡ ⎣⎢ 1 3 ( ) ⎤ ⎦⎥ ( ) 3 Here, N is the number of dopant-ready sites in the unit cell, and V is the unit cell volume at the critical doping concentration (Xc). The computed critical distance values for NBS1T, NBS3T, and NBS5T are 4.53 Å, 3.13 Å, and 2.63 Å, respectively. 3.2.3. CIE diagram Chromatic diagram based on Commission Internationale de l’Eclairage (CIE) (1931) [46], along with correlated colour temperature (CCT) values, for Na0.5(Bi1-xSmx)0.5TiO3 with x= 0.1, 0.3 and 0.5 are presented in figure 9(b). This diagram depicts the colour purity of the luminophores when excited at 407 nm. These are significant from the perspective of the material performance on colour luminescent emission in applications like LEDs that take place in the real world. The McCamy empirical relation (4) is used to calculate the CCT values, from the perspective of material performance on colour luminescent emission in practical applications like LEDs [47] these are significant. ( CCT x, y ) = - 449n3 + 3525n2 - 6823.3n + 5520.33 ( ) 4 , x e y e Where = - x n is the tangent of the angle between the y-axis, and this line is the reciprocal of the slope. (xe, n - y ye) = (0.3320, 0.1858) are the epicenter of the convergence at the point on the chromaticity diagram, and (x, y) is the colour coordinates of the sample. The colour purity of the dopant systems’ emitted colour is determined by using equation (5) [48] 11 Phys. Scr. 98 (2023) 125956 J Neeli et al Figure 10. M-H loops for NBT and Sm-modified NBT ceramics at room temperature. Table 7. Magnetization values for NBT and Samarium-modified NBT ceramic compositions at Room Temperature. S.No Composition Ms (memu /g) Mr (memu /g) 1 2 3 4 NBT NBS1T NBS3T NBS5T 0.0140 9.596 34.18 6.413 0.00036 0.703 1.49 0.337 S Hc(Oe) 0.07 0.49 0.43 0.42 0.00 0.50 96.31 116.23 Magnetic moment (μb) Anisotropy constant A(erg/cm3) 0.0005 0.0004 0.0012 0.0002 2 0.000 0.004 3.429 0.776 Color purity = ( x - 2 ) + (( y - x i x 2 ) y i y i 2 ´ 100% - d Where (xi, yi) is the illuminant point and (xd, yd) is the colour coordinates of a dominant wavelength. y d - + x ) ( ) ( i From figure 9(b), the Commission Internationale de l’Eclairage (CIE) colour coordinates for the NBSmxT systems (x = 0.1, 0.3, and 0.5) are (0.324, 0.329), (0.383. 0.326), (0.333. 0.301) respectively. The CCT values of Sm-modified NBT systems (x = 0.1, 0.3, and 0.5) are reported as 5912.12 K, 3483.53 K, and 5458.43 K, respectively. These systems are well within the reddish-orange region. Photometric parameters like CIE colour coordinates, CCT, and colour purity were different in all the studied compositions. The phosphors exhibit strong reddish-orange emission with acceptable CCT, colour purity, and reddish-orange chromaticity. Ceramic composition at x = 0.3 concentration displayed high-intensity reddish-orange emission based on the colour coordinates, CCT, and colour purity values. ( ) 5 3.3. Magnetic study The magnetic measurements of the prepared [Na0.5(Bi1-xSmx)0.5TiO3] ceramic samples with compositions x=0.0, 0.1, 0.3, and 0.5 were recorded at room temperature under the magnetic field of −20 kOe „H „ 20 kOe and are presented in figure10 along with NBT, which is with a diamagnetic attribute, the prepared compositions exhibited typical ferromagnetic (RTFM) behavior [49]. The equation (6), relates the magnetic moment (μb) with Bohr magneton [50]. m b = ´M M 5585 S ⎤ ⎦ ⎡ ⎣ 12 ( ) 6 Phys. Scr. 98 (2023) 125956 J Neeli et al Figure 11. (a) Vickers hardness as a function of concentration (b) Wear coefficient, coefficient of friction, and Specific wear energy as a function of concentration. Table 8. Vickers Hardness, Coefficient of Wear, Coefficient of Friction values for NBSmxT (x = 0.0, 0.1, 0.3 & 0.5) ceramics. S. No Composition 1 2 3 4 NBT NBS1T NBS3T NBS5T Vickers Hardness (HV) (GPa) Coefficient of Wear (K) (mm3/ N.m) Coefficient of friction (μ) SWE (* 104 J g −1) 5.59 6.78 8.20 5.86 –6 –6 0.68 × 10 0.78 × 10 –6 1.21×10 0.53 × 10 –6 0.582 0. 558 0. 641 0.545 14 76 93 58 Here, M is the molecular weight of the sample, Ms is saturation magnetization, and 5585 is the magnetic factor. The equation (7) was used to calculate the anisotropy constant [51] Anisotropy constant A = H C ´M S 0.96 ( ) 7 Where Ms is the saturation magnetization, Hc is the coercivity, and 0.96 is a constant. Table 7 presents the values of the reduced magnetization (S=Mr/Ms), magnetic moment (μb), and the anisotropy constant (A), as well as the saturation magnetization (Ms), remnant magnetization (Mr), and coercivity (Hc). The lattice distortion, indirect spin exchange interactions, the bond angle and length alteration impact ferromagnetic properties [52]. The reduced magnetization which is also known as the squareness factor, with values in between 0 and 1 is used in memory devices. All the prepared ceramic samples were identified to exhibit a soft magnetic nature with a multi-domain structure as the squareness factor values are less than 0.5 [53]. 3.4. Mechanical studies 3.4.1. Vickers microhardness Vickers hardness test was developed in 1924 by Smith and Sandland. Hardness is one of the most important properties of a ceramic. Vickers hardness values were calculated using equation (8) for NBT and Sm-modified NBT ceramic samples, the obtained results are noted in table 8. The input parameters applied are load 10 N and time 10 s. Vickers hardness as a function of concentration was shown in figure11(a) H = F 0.189 d ´ 2 Gpa ( ) 8 Where H is the hardness, F is the load in the indentation, and d is the average length of the diagonal line in the indentation. The NBS3T ceramic (8.20 GPa) has a hardness much higher than NBT (5.59 GPa) ceramics, and this may be due to submicron grains, which provide additional barriers to the movement of lattice dislocations in adjacent grains and, hence the increased number of grain boundaries [54–57]. 13 Phys. Scr. 98 (2023) 125956 J Neeli et al 3.4.2. Specific Wear Rate (SWR) The pins and ball-on-circle machines have been used to concentrate on the wear component in research, going from full-scale lab tests and downscale testing. Of the above techniques, the trunnion-on-circle machine is the easiest, most economical, and most efficient research centre test for concentrating on wear components [58]. Specific wear rate (SWR) has been described as the loss of material volume per unit of load and sliding distance during the wear of two bodies by G. Suresh et al [59]. Pin-on-disc equipment was used to measure the Coefficient −1 and applied load 10 N of of Wear (K) and Coefficient of friction (μ) with input values sliding velocity 1.224 m s the prepared ceramic samples. It is calculated by using equation (9). SWR = ∆ W ´ ´ L ( r Sd ) ( ) 9 Where ΔW = Sample weight before wear test - Sample weight after wear test, ρ = density of the sample, L= Applied Load, Sd= Sliding distance –6 All prepared ceramic samples with mild wear were found to have a specific wear rate range of less than 10 mm3/(N m), and this demonstrates that the present materials are favourable to be utilized as wear-safe tribo- materials when contrasted with Polytetrafluoroethylene (PTFE) [60, 61]. 3.4.3. Specific wear energy (SWE) It is the ratio between the frictional energy consumed at the interface and the mass lost due to wear. The Specific Wear Energy (SWE) is derived from the coefficient of friction and wear rate to determine the tribological characteristics of a material. It is said that composites with a high specific wear energy have a high specific wear rate [59]. SWE can be calculated using the following equation (10) ) Specific wear energy SWE ( = v m ´ ´ ´ w D m t ( ) 10 Where ‘v’ is the velocity in m/s, ‘w’ is the applied load in N, ‘μ’ is the coefficient of friction, ‘t’ is time in seconds, and ‘Δm’ is the weight loss in grams. The greater SWE value may be ascribed to the high specific wear rate and coefficient of friction listed in table 8. Specific wear energy values rise with the increase in Sm dopant concentration up to x = 0.3 after which they start to decrease. The decrease can be attributed to a reduction in the coefficient of friction. The variation of SWR, SWE, and Coefficient of friction with dopant concentration in NBT are shown in figure 11(b). 3.4.4. Coefficient of friction (μ) Discovering a low friction coefficient for specific conditions, such as temperature, time, applied load, sliding cycles on the sample, etc, is critical. This would prevent wear, save money on maintenance, and use less energy, all of which would significantly extend the life of the product. The pin-on-disk tests were performed with a 50 mm-diameter track and a normal load of 10 N at a sliding speed of 1.224 m/s for 300 s. At room temperature, the concentrated creative materials’ friction coefficients have produced results in the range of 0.5–0.6; these outcomes are coordinated with past researchers [59, 62]. The investigated ceramic materials might be a better choice for properties like self-lubrication. 4. Conclusions Sodium bismuth titanate [Na0.5(Bi1-xSmx)0.5TiO3] ceramic samples of varying concentrations with x = 0.0, 0.1, 0.3, & 0.5 were prepared from the solid-state reaction method. The rhombohedral phase with space group R3c was confirmed in all prepared ceramic samples using x-ray diffraction patterns and obtained lattice parameters utilizing Rietveld analysis. As the dopant concentration increased, the particle size of the samples decreased, and the lattice strain ion doping increased the granularity value, indicating that the Ostwald ripening concurrently increased. Sm3+ mechanism causes the particles to become coarse during sintering, as received from SEM micrographs. The EDAX spectra for various compositions are in line with stoichiometry, afforming the presence of all the constituent elements. The vibration bands shift and expand with an increase in dopant concentration, indicating lattice deformation. Based on the findings from UV–vis spectroscopy it is evident that as the doping concentration increased the band gap values decreased, thereby highlighting the potential utility of NBT-based perovskites as photoactive materials. At a concentration of x = 0.3, the luminophore demonstrates a pronounced reddish-orange emission characterized by excellent reddish-orange chromaticity, CCT, and colour purity. The squareness factors of the produced compositions vary from 0.07 to 0.49, rendering them compelling 14 Phys. Scr. 98 (2023) 125956 J Neeli et al materials for applications in memory device. The ceramic material at x=0.3 concentration, the Vickers hardness was very high (8.20 GPa) and exhibited mild wear with relatively low frictional coefficient values. Acknowledgments The authors would like to thank Vignan’s Foundation for Science Technology and Research (VFSTR) deemed to be University, for extending the CoExAMMPC facility for basic characterization of our research samples. Data availability statement The data cannot be made publicly available upon publication because they are owned by a third party and the terms of use prevent public distribution. The data that support the findings of this study are available upon reasonable request from the authors. ORCID iDs Nitchal Kiran Jaladi https://orcid.org/0000-0002-6585-6067 References [1] Haertling G H 1999 J. Am. Ceram. Soc. 82 797–818 [2] Tressler J F, Alkoy S and Newnham R E 1998 J. Electroceram. 2 257–72 [3] Saito Y, Takao H, Tani T, Nonoyama T, Takatori K, Homma T and Nakamura M 2004 Nature 432 84–7 [4] Smolenskii G A 1961 Soviet Physics-Solid State 2 2651–4 [5] Lu Y, López C A, Wang J, Alonso J A and Sun C 2018 J. Alloys Compd. 752 213–9 [6] Bradha M, Hussain S, Chakravarty S, Amarendra G and Ashok A 2015 J. Alloys Compd. 626 245–51 [7] Reichmann K, Feteira A and Li M 2015 Materials 8 8467–95 [8] Jones G O and Thomas P A 2002 Acta Crystallogr., Sect. B: Struct. Sci 58 168–78 [9] Damjanovic D, Klein N, Li J and Porokhonskyy V 2010 Functional Materials Letters 3 5–13 [10] He X and Mo Y 2015 Phys. Chem. Chem. Phys. 17 18035–44 [11] Bai Y, Zheng G P and Shi S Q 2011 Mater. Res. Bull. 46 1866–9 [12] Gou Q, Wu J, Li A, Wu B, Xiao D and Zhu J 2012 J. Alloys Compd. 521 4–7 [13] Benyoussef M, Zaari H, Belhadi J, El Amraoui Y, Ez-Zahraouy H, Lahmar A and El Marssi M 2022 J. Rare Earths 40 473–81 [14] Franco Jr A, Banerjee P and Romanholo P L 2018 J. Alloys Compd. 764 122–7 [15] França E L T, Romanholo P V V, Simões S S, Falcão E H L, Franco Jr A and Machado F L A 2021 J. Alloys Compd. 873 159845 [16] Behara S, Krishna R H, Muralidhar M, Murakami M, Irfan M, Najma S and Thomas T 2019 Materialia 7 100426 [17] Behara S, Ikeda K and Thomas T 2021 Ceram. Int. 47 12870–8 [18] Kandula K R, Asthana S and Raavi S S K 2018 RSC Adv. 8 15282–9 [19] Zannen M, Dietze M, Khemakhem H, Kabadou A and Es-Souni M 2014 Ceram. Int. 40 13461–9 [20] Wei T, Sun F C, Zhao C Z, Li C P, Yang M and Wang Y Q 2013 Ceram. Int. 39 9823–8 [21] Lenka S, Badapanda T, Nayak P, Sarangi S and Anwar S 2021 Ceram. Int. 47 5477–86 [22] Blömker M, Erdem E, Li S, Weber S, Klein A, Rödel J and Frömling T 2016 J. Am. Ceram. Soc. 99 543–50 [23] Smyth D M, Harmer M P and Peng P 1989 J. Am. Ceram. Soc. 72 2276–8 [24] Rodriguez-Carvajal J 1990 Toulouse, France 127 [25] Dinh T H, Lee H Y, Yoon C H, Malik R A, Kong Y M, Lee J S and Tran V D N 2013 J. Korean Phys. Soc. 62 1004–8 [26] Ganbavle V V, Kim J H and Rajpure K Y 2015 J. Electron. Mater. 44 874–85 [27] Rajesh R, John Ethilton S, Ramachandran K, Giridharan N V, Ramesh Kumar K and Vadla S S 2018 Int. J. Mod. Phys. B 32 1850277 [28] Lei N, Zhu M, Yang P, Wang L, Wang L, Hou Y and Yan H 2011 J. Appl. Phys. 109 054102 [29] Shelby J E 2015 (Royal society of chemistry) [30] Sangula N, Jaladi N K, Bhimavarapu S B R, Bhuvanagiri N R, Jaladi A K and Konapala S R 2021 ECS J. Solid State Sci. Technol. 10 041002 [31] Kumar A, Kumar P and Dhaliwal A S 2021 Advances in Applied Ceramics 120 307–18 [32] Roy R and Dutta A 2023 J. Rare Earths [33] Lenka S, Badapanda T, Ghosh S P, Richhariya T, Sarangi S and Tripathy S N 2023 Materials Today Communications 36 106738 [34] Prusty R K, Kuruva P, Ramamurty U and Thomas T 2013 Solid State Commun. 173 38–41 [35] Devi C S, Kumar G S and Prasad G 2013 Materials Science and Engineering: B 178 283–92 [36] Roy R and Dutta A 2020 Solid State Sci. 102 106174 [37] Darwish A G A, Badr Y, El Shaarawy M, Shash N M H and Battisha I K 2010 J. Alloys Compd. 489 451–5 [38] Tauc J 1968 Mater. Res. Bull. 3 37–46 [39] Zeng M, Or S W and Chan H L W 2010 J. Appl. Phys. 107 043513 [40] Jeshurun A, Mohammad I, Behara S and Bogala M R 2021 Materials Today Chemistry 20 100476 [41] Pradhan S K and De S K 2018 Ceram. Int. 44 15181–91 [42] Swain S, Ojha B and Mohanty S 2016 J. Mater. Sci., Mater. Electron. 27 8693–700 [43] Kuo T W, Huang C H and Chen T M 2010 Opt. Express 18 A231–6 [44] Naik R C, Karanjikar N P and Razvi M A N 1992 J. Lumin. 54 139–44 [45] Blasse G 1969 Philips Res. Rep. 24 131 [46] Smith T and Guild J 1931 Trans. Opt. Soc. 33 73 [47] McCamy C S 1992 Color Res. Appl. 17 142–4 15 Phys. Scr. 98 (2023) 125956 J Neeli et al [48] Singh D K, Mondal K and Manam J 2017 Ceram. Int. 43 13602–11 [49] Rawat M and Yadav K L 2015 Smart Mater. Struct. 24 045041 [50] Tahir M, Imran Ahmad S, Javid Ali M and Shahid K 2018 J. Mater. Sci., Mater. Electron. 29 5110–5 [51] Hooda A, Sanghi S, Agarwal A and Dahiya R 2015 J. Magn. Magn. Mater. 387 46–52 [52] Reetu R, Agarwal A, Sanghi S, Ashima A and Ahlawat N 2013 J. Appl. Phys. 113 023908 [53] Fonsecaa S G C and Neivab L S 2018 Mater. Res. 21 20170861 [54] Hoffmann M J, Hammer M, Endriss A and Lupascu D C 2001 Acta Mater. 49 1301–10 [55] Hall E O 1951 Proc. Phys. Soc. London, Sect. B 64 747 [56] Petch N J 1953 J. Iron Steel Inst. 174 25–8 [57] Yip S 1998 Nature 391 532–3 [58] Faccoli M, Petrogalli C and Ghidini A 2017 Tribol. Lett. 65 1–7 [59] Suresh G, Vasu V and Rao M V 2018 Silicon 10 2043–53 [60] Adachi K, Kato K and Chen N 1997 Wear 203–204291–303 [61] Şahin Y, Erdil E and Şahin H 2015 Cogent Eng 2 1000510 [62] Vijaya Kumar C and Sreenivasulu M 2021 J. Mater. Sci., Mater. Electron. 32 20499–509 16
null
10.1016_j.isci.2020.100959.pdf
DATA AND CODE AVAILABILITY RNA-seq data have been deposited in NCBI’s Gene Expression Omnibus (GEO). The accession number for the RNA-seq data reported in this paper is GEO: GSE145495.
DATA AND CODE AVAILABILITY RNA-seq data have been deposited in NCBI's Gene Expression Omnibus (GEO). The accession number for the RNA-seq data reported in this paper is GEO: GSE145495.
Article Loss of Asb2 Impairs Cardiomyocyte Differentiation and Leads to Congenital Double Outlet Right Ventricle Substrate adapter polyubiquitinated Substrate E3 Cullin5 Asb2 E2 X Heart Failure Embryonic Lethality DORV Flna Smad2 E7.5 FHF AHF Cardiac Crescent E9.0-9.5 OFT LV RV Looped Heart Negative regulation Positive regulation Abir Yamak, Dongjian Hu, Nikhil Mittal, ..., Christel Moog- Lutz, Patrick T. Ellinor, Ibrahim J. Domian fyamak@mgh.harvard.edu (A.Y.) idomian@mgh.harvard.edu (I.J.D.) HIGHLIGHTS Flna removal partially rescues embryonic lethality of Asb2-heart- specific knockout AHF-Asb2 knockouts harboring one Flna allele have double outlet right ventricle Asb2-Flna regulate TGFb-Smad2 signaling in the heart Conserved role of Asb2 in heart morphogenesis between mice and humans DATA AND CODE AVAILABILITY GSE145495 Yamak et al., iScience 23, 100959 March 27, 2020 ª 2020 The Author(s). https://doi.org/10.1016/ j.isci.2020.100959 Article Loss of Asb2 Impairs Cardiomyocyte Differentiation and Leads to Congenital Double Outlet Right Ventricle Abir Yamak,1,2,3,9,* Dongjian Hu,2,4 Nikhil Mittal,1,2 Jan W. Buikema,2,5 Sheraz Ditta,2,6 Pierre G. Lutz,7 Christel Moog-Lutz,7 Patrick T. Ellinor,1,2,3 and Ibrahim J. Domian1,2,8,* SUMMARY Defining the pathways that control cardiac development facilitates understanding the pathogenesis of congenital heart disease. Herein, we identify enrichment of a Cullin5 Ub ligase key subunit, Asb2, in myocardial progenitors and differentiated cardiomyocytes. Using two conditional murine knockouts, Nkx+/Cre.Asb2fl/fl and AHF-Cre.Asb2fl/fl, and tissue clarifying technique, we reveal Asb2 requirement for embryonic survival and complete heart looping. Deletion of Asb2 results in upregu- lation of its target Filamin A (Flna), and concurrent Flna deletion partially rescues embryonic lethality. Conditional AHF-Cre.Asb2 knockouts harboring one Flna allele have double outlet right ventricle (DORV), which is rescued by biallelic Flna excision. Transcriptomic and immunofluorescence analyses identify Tgfb/Smad as downstream targets of Asb2/Flna. Finally, using CRISPR/Cas9 genome editing, we demonstrate Asb2 requirement for human cardiomyocyte differentiation suggesting a conserved mechanism between mice and humans. Collectively, our study provides deeper mechanistic under- standing of the role of the ubiquitin proteasome system in cardiac development and suggests a pre- viously unidentified murine model for DORV. INTRODUCTION Congenital heart diseases (CHDs) are prenatal defects that affect the heart’s structure and/or function and are the leading cause of infant mortality under 1 year of age. Approximately 1%–2% of human babies are born with cardiac malformations that pose as major risk factors for adult cardiovascular problems (Bruneau, 2008; Nemer, 2008). The heart, the first functional organ in the developing embryo, starts to form early on during development, before the end of gastrulation. The first and second heart fields (FHF and SHF, respectively) as well as the proepicardial organ and the cardiac neural crest are the major contributors to the forming heart (Martinsen and Lohr, 2015). The FHF gives rise primarily to the left ventricle and most of the atria; the SHF contributes to the right ventricle, outflow tract, and parts of the atria (Srivastava, 2006; Yamak and Nemer, 2015). Induction of the cardiac fate and the proper morphogenesis of the verte- brate heart are controlled by a well-characterized and highly conserved combinatorial network of transcrip- tion factors and signaling molecules that act together to orchestrate the embryonic development of the four-chambered mammalian heart and the subsequent post-natal maturation. Of important note, the adult heart has minimal intrinsic regenerative capacity (Mercola et al., 2011). As a result, significant stressors on the heart can result in loss of viable or functional myocardial tissue and ultimately heart failure. This renders cardiovascular disease a leading cause of death worldwide and highlights an unmet clinical need for novel approaches for heart regeneration. One major approach is the use of stem cells that can be induced to give rise to the different cell types that constitute the heart. Understanding the cellular processes and signaling pathways that govern in vivo heart formation and maturation is necessary for the generation of functional mature cardiac tissue for clinical and preclinical applications (Hu et al., 2018). Targeted protein degradation by the ubiquitin proteasome system (UPS) is important for the regulation of cellular physiology and is required for normal organ formation (Glickman and Ciechanover, 2002). The UPS consists of three enzymes: Ubiquitin (Ub) activating enzyme, E1, which transfers activated Ub to the Ub conjugating enzyme, E2. This then interacts with the E3 Ub ligase that covalently links the Ub or Ub chain to a lysine residue in the substrate thus targeting it for degradation by the proteasome. The E3 Ub ligase is responsible for substrate specificity (Jung et al., 2009). Recent evidence points to a role of the UPS in heart disease, particularly in myocardial remodeling, familial cardiomyopathies, chronic heart failure, and 1Harvard Medical School, Boston, MA 02115, USA 2Cardiovascular Research Center, Massachusetts General Hospital, 185 Cambridge Street, CPZN3200, Boston, MA 02114, USA 3Cardiovascular Disease Initiative, Broad Institute of MIT and Harvard, Cambridge, MA 02142, USA 4Department of Biomedical Engineering, Boston University, Boston, MA 02215, USA 5University Medical Center Utrecht, 3584 CX Utrecht, Netherlands 6Department of Pharmaceutical Sciences, Utrecht University, 3512 JE Utrecht, Netherlands 7Institut de Pharmacologie et de Biologie Structurale, IPBS, Universite´ de Toulouse, CNRS, UPS, Toulouse, France 8Harvard Stem Cell Institute, Cambridge, MA 02138, USA 9Lead Contact *Correspondence: fyamak@mgh.harvard.edu (A.Y.), idomian@mgh.harvard.edu (I.J.D.) https://doi.org/10.1016/j.isci. 2020.100959 iScience 23, 100959, March 27, 2020 ª 2020 The Author(s). This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). 1 B D A C E Figure 1. Asb2 Is Expressed in the Developing and Adult Heart and Undergoes Isoform Switching during Differentiation (A) qPCR analysis of embryonic cardiomyocytes reveals predominant Asb2a expression in the R-G+ and the R+G+ populations. R-G+: Mef2c-.Nkx2-5+; R+G+: Mef2c+.Nkx2-5+; R+G-: Mef2c+.Nkx2-5-; NEG: Mef2c-.Nkx2-5-. (B) qPCR analysis of Asb2a and Asb2b on RNA from murine hearts of different embryonic stages as well as neonates and postnatal day 8–9. Note that the a isoform is equally expressed at all stages, whereas the b isoform expression increases with development. (C) Western blot analysis on whole tissue extracts from embryonic and adult heart, spleen, and skeletal muscle using Asb2-specific antibody. Note that Asb2 corresponding band in the embryonic heart co-migrates with that in the spleen where only the a isoform is expressed, whereas that in the adult heart co-migrates with that in the skeletal muscle that is known to express on the b isoform. These data are consistent with the qPCR data in (B). 2 iScience 23, 100959, March 27, 2020 Figure 1. Continued (D) In situ hybridization on E9.5 mouse embryo showing robust Asb2 expression in the LV, RV, and OFT and to a lesser extent the IFT. LV, left ventricle; RV, right ventricle; OFT, outflow tract; IFT, inflow tract. (E) Immunohistochemistry on E10.5 and E12.5 mouse embryos using Asb2-specific antibody (green) and Troponin T (red). DAPI (blue) marks nuclei. Note Asb2 expression colocalizes with Troponin T in the myocardium (arrows) and no expression is seen in the endocardial cells (arrow heads). Scale bar is equivalent to 250 mm in the first three heart images left to right at E10.5 (top) and E12.5 (bottom), 25 mm in the E10.5 heart image top far right, and 50 mm in the E12.5 heart image bottom far right as indicated in the figure. ischemia-reperfusion injury (Pagan et al., 2013). Pharmacological inhibition of the proteasome is a new and promising means for cardioprotection (Pagan et al., 2013). Paradoxically, enhancing UPS activity has in some cases also provided protection against heart disease (Bulteau et al., 2001; Li et al., 2011; Powell et al., 2007) highlighting the importance of defining the role of the UPS as a therapeutic target in cardiac disease. In addition to its role as a protein quality control, the UPS has also been shown to regulate the turnover of sarcomere proteins, including the myofibrillar proteins myosin, actin, and troponin. Examples of this include the E3 Ub ligases MuRF1 (muscle-specific RING finger 1, targeting troponin I) and F box pro- tein Fbx122 (targeting a-actinin and filamin C) (Kedar et al., 2004; Spaich et al., 2012). E3 ligases have also been shown to regulate important signaling pathways in the heart, such as the JNK (c-Jun N terminal kinase) (Laine and Ronai, 2005), calcineurin (Fan et al., 2008), and the VEGF (vascular endothelial growth factor) signaling pathways (Murdaca et al., 2004). This important role of the UPS in the heart and its poten- tial for a therapeutic target in cardiac disease brings about a need to understand its specific function in heart development and disease. Using our previously described transgenic reporter system (Domian et al., 2009), we identified Asb2 (ankyrin repeat-containing protein with a suppressor of cytokine signaling box [SOCS box] 2) as being enriched in FHF and SHF cardiac progenitors. Asb2, which encodes specificity subunit of Cullin 5 RING E3 Ub ligase, exists in two isoforms: Asb2a and Asb2b in mouse, corresponding to variants 2 and 1 in humans, respectively (Bello et al., 2009). It has previously been shown to regulate differ- entiation of myeloid leukemia cells and skeletal myogenesis through proteasomal degradation of filamin proteins (Bello et al., 2009; Guibal et al., 2002; Heuze´ et al., 2008). Filamins (Flna, Flnb, and Flnc in mice) are actin-binding proteins important for the stabilization of the actin-cytoskeleton (van der Flier and Son- nenberg, 2001). Flnc is the only isoform expressed in the heart muscle where it is required for normal contractility (Fujita et al., 2012). Flna expression in the heart is restricted to endocardial and mesenchymal cells of the cardiac cushions during development and to valve leaflets in the adult heart (Norris et al., 2010). FLNA and FLNC mutations have been linked to cardiac defects in humans (de Wit et al., 2011, 2009; Kyndt et al., 2007; Valde´ s-Mas et al., 2014). In zebrafish, an additional Asb2 target TCF3 was very recently identi- fied where TCF3 was negatively regulated by Asb2 during cardiogenesis (Fukuda et al., 2017). Asb2 down- regulation was also shown to be a mediator of follistatin-induced muscle hypertrophy and SMAD2/3 regu- lation of skeletal muscle mass in young adults in mice. The repression of Asb2 was, however, ameliorated in aging mice, some of which also displayed increasing Asb2 baseline levels (Davey et al., 2016). Asb2 over- expression was also shown to drive skeletal muscle atrophy in mice (Davey et al., 2016). A recent study also showed that Asb2 knockout is embryonic lethal and that Asb2a targets Flna for proteasomal degradation during early cardiomyocyte differentiation (Me´ tais et al., 2018). The embryonic lethality of Asb2 mutants was shown to be primarily due to heart defects (Me´ tais et al., 2018). Herein, we show that Asb2 knockout in the FHF and SHF are both embryonic lethal by E10.5 and E12.5, respectively. Using tissue clearing combined with immunofluorescence technique, we show that Asb2 mutant hearts have incomplete looping. Moreover, Asb2 regulates cardiac morphogenesis partly through Flna turnover, and we hereby propose a model where Asb2-Flna controls TGFb-SMAD signaling to drive early cardiac formation. Additionally, Asb2 lethality in the anterior heart field (AHF) is partially rescued by Flna removal from these hearts. We also show that Asb2 ablation in the AHF leads to double outlet right ventricle (DORV), which is corrected upon further deletion of Flna from these hearts. Finally, we reveal that Asb2 role in cardiomyocyte differentiation is conserved in human cardiomyocytes as well. Collectively, our results shed light on the UPS regulation of heart development and its role as a cardio-therapeutic target and provide evidence for the first time for the role of the UPS in the rare congenital heart defect, DORV. RESULTS Asb2 Is Highly Enriched in the Embryonic Heart We have previously characterized a transgenic reporter system for the isolation of three distinct mouse cardiac progenitor cells from developing embryos: FHF population, marked by Nkx2.5+.Mef2c- expression, and two SHF population subsets: Nkx2.5-.Mef2c+ and Nkx2.5+.Mef2c+ (Domian et al., 2009). Genome-wide iScience 23, 100959, March 27, 2020 3 E A B C D Figure 2. Asb2 Is Essential for Early Cardiac Development (A) Nkx2-5+/Cre.Asb2 E9.5 and E11.5 knockout (KO) embryos (fl/fl) versus wild-type littermates (Wt). Note the resorbing KO embryo at E11.5. Scale bar is equivalent to 0.4 mm for E9.5 and 0.5 mm for E11.5 as indicated. (B) AHF-Cre.Asb2 E10.5 and E12.5 knockout (KO) embryos (fl/fl) versus wild-type littermates (Wt). Note the resorbing KO embryo at E12.5. Scale bar is equivalent to 0.5 mm for E10.5 and E12.5 as indicated in the figure. (C) 3D reconstruction of CUBIC-cleared, Troponin-T-stained E9.5 whole control and Asb2 mutant embryos showing both ventral and dorsal views. Note the bulging in the right ventricle of the control heart that is lacking in the mutant (indicated by the red arrow heads). Scale bar is equivalent to 200 mm as indicated. 4 iScience 23, 100959, March 27, 2020 Figure 2. Continued (D) Measurement of the heart tube of control and Asb2 mutant hearts. Note the statistically significant shorter heart tubes of the mutants. N = 5 per group. Data are represented as mean G SEM. * = p < 0.005. Unpaired t test was used using GraphPad Prism; p < 0.05 is considered statistically significant. (E) Heatmap analysis of a subset of cardiac looping differentially expressed genes in RNA-seq data from control (Nkx2-5+/Cre.Asb2fl/+) versus Nkx2-5+/Cre.Asb2 knockout E9.5 murine hearts. N = 3 in each group (each sample is in itself a combination of three to four hearts to account for heterogeneity among different litters). transcriptional profiling and real-time PCR (qPCR) reveal Asb2 transcripts enrichment in the three popula- tions (Figure 1A) (Domian et al., 2009). To investigate the temporal expression of Asb2 in the developing heart, we performed qPCR analysis on RNA from mouse hearts at different stages of embryonic develop- ment. Our data show that Asb2a is expressed similarly throughout heart development, whereas Asb2b expression increases with development (Figure 1B). This is further confirmed by western blot analysis, which shows that the Asb2 band in the embryonic heart co-migrates with that in the spleen (which expresses Asb2a [Spinner et al., 2015]), whereas the Asb2 band in the adult heart co-migrates with that in the skeletal muscle (known to express Asb2b [Bello et al., 2009]) (Figure 1C). To further investigate in vivo spatial cardiac expres- sion of Abs2, we performed in situ hybridization on E9.5 embryos. Our data reveal robust expression of Asb2 transcripts predominantly in the left (LV) and right ventricles (RV) and to a lower extent in inflow (IFT) and outflow tracts (OFT) (Figure 1D). Furthermore, immunostaining of E10.5 and E11.5 (Figure 1E, upper and lower panels, respectively) embryonic sections using Asb2-specific antibody shows that, in the heart, Asb2 expression (green) is restricted to the myocardium overlapping with cardiac Troponin T (red). White arrows in the zoomed merged image at E10.5 (right panel) indicate overlap of Asb2 and Troponin T in the myocardial layer, but no expression is seen in the endocardial layer indicated by arrow heads. Asb2 Is Required for Early Cardiac Formation To investigate the role of Asb2 during cardiac development, we generated two conditional knockout lines (KO): Nkx2-5+/Cre (a mouse line with the Cre recombinase knocked into the Nkx2-5 locus) and AHF-Cre (a mouse line with a transgene placing Cre under the transcriptional control of the AHF enhancer of the Mef2c gene). These mouse lines allow for the targeted removal of (Lombardi et al., 2009) Asb2 from the whole heart and the SHF, respectively (Lombardi et al., 2009). The floxed alleles are in common region and inactivate both Asb2 isoforms. Both conditional KOs have pericardial edema and are embryonic lethal: Nkx2-5+/Cre.Asb2fl/fl mice die at E10.5–11 and AHF-Cre.Asb2fl/fl die at E11.5–12 (Figures 2A and 2B, respec- tively). AHF-Cre.Asb2fl/fl mice analyzed at E10.5 also have shorter OFT compared with their control litter- mates (Figure S1D). For Nkx2-5+/Cre.Asb2fl/fl, mice were analyzed at E8.5 (3 litters), E9.5 (23 litters), E10.5 (3 litters), and E11.5 (2 litters); for AHF-Cre.Asb2fl/fl, mice were analyzed at E9.5 (3 litters), E10.5 (4 litters), and E12.5 (2 litters). Each litter consists of 8–11 embryos in total. All embryos were genotyped. Figure S1A shows the reduced level of Asb2 in the heterozygotes (Nkx2-5+/Cre.Asb2fl/+) and the complete loss of Asb2 in the knockouts (Nkx2-5+/Cre.Asb2fl/fl). In order to perform a phenotypic analysis of the Nkx2-5+/Cre-Asb2fl/fl mutant embryos, we used state-of- the-art tissue clearing technique CUBIC combined with immunostaining. CUBIC can effectively clear mice embryos and embryonic hearts while preserving immunolabels (Kolesova´ et al., 2016; Tainaka et al., 2014). Nkx2-5+/Cre-Asb2fl/fl and control littermates e9.5 mice embryos were cleared with CUBIC and stained for Troponin T to mark cardiomyocytes as well as DAPI for nuclei. Confocal microscopy with optical sectioning followed by 3D-reconstruction allowed the precise visualization of the developing hearts without disruption of underlying anatomy. During cardiac morphogenesis, the straight heart tube undergoes sequential looping steps to get to the fully looped heart. The fully looped heart acquires a he- lical shape in mice that is also referred to as the mature S-loop in chicks (Le Garrec et al., 2017; Ma¨ nner, 2009). In Le Garrec et al. paper, they used computer modeling to simulate the biological process of mouse cardiac looping, incorporating in their model the left-right asymmetry and mechanical constraints seen in the looping heart. Their findings suggest that the lack of any of these parameters would lead to a C-shaped heart loop rather than the helical structure. In the chick, the heart is first transformed into a C-shaped heart, a process known as dextral looping. The C-loop is then converted into an immature S-loop that then trans- forms into a mature S-looped heart where the ventricular segments are curved outward to generate the left and right chambers (Ma¨ nner, 2009). As shown in Figure 2C, the mutant embryos do not form the full helical structure seen in the control littermates. Instead, they have partially looped hearts that resemble the C-shaped hearts in the chick (Ma¨ nner, 2009). Moreover, measurement of the heart tube length in mutant versus control hearts reveals statistically significant shorter tubes in the mutant hearts (Figure 2D). Video S1 is a z stack of stained control and mutant e9.5 embryos showing the incomplete looping in the mutant embryo. Figure S1C represents four images from the z stack at different depth in the embryo. Four to five iScience 23, 100959, March 27, 2020 5 A B C 6 iScience 23, 100959, March 27, 2020 Figure 3. Asb2 Targets Flna for Proteasomal Degradation in the Developing Heart and Asb2-Mutant Hearts Have an Altered Gene Expression Profile (A) Immunohistochemistry on E9.5 Abs2 heterozygote (Nkx2-5+/Cre.Asb2fl/+, middle pane) and mutant hearts (Nkx2-5+/Cre.Asb2fl/fl, lower panel) as well as Wt controls (top panel) using Flna (red) and Troponin-T (green)-specific antibodies. Note that FlnA expression is restricted to the endocardial layer (white arrow heads) in the Wt heart, whereas it is abnormally expressed in the myocardial layer in the Asb2-mutant hearts co-localizing with Troponin-T expression there (white arrows). Moreover, some cardiomyocytes in the outflow tract of the Asb2-heterozygous hearts also express Flna (yellow arrows) suggesting a dose- dependent regulation. Scale bar is equivalent to 250 mm in the first column (left), 100 mm in the second, third, and fourth columns, and 25 mm in the fifth (far right) column as indicated in the figure. (B) Heatmap analysis of RNA-seq data from control (Group1: Nkx2-5+/Cre.Asb2fl/+), Asb2 mutant (Group2: Nkx2-5+/Cre.Asb2fl/fl), Flna mutant (Group3: Nkx2-5+/Cre.Flnafl/y), and Asb2-Flna double mutant (Group4: Nkx2-5+/Cre.Asb2fl/fl.Flnafl/y) E9.5 murine hearts. Note the high level of differentially expressed genes in the Asb2-mutant and Asb2-Flna double mutant versus the control groups. A small subset of genes (indicated by arrows) that are perturbed in the Asb2-mutant hearts are restored to normal in the Asb2-Flna double mutants. N = 3 in each group (each sample is in itself a combination of three to four hearts to account for heterogeneity among different litters). (C) Heatmap analysis of a subset of genes from the RNA-seq data in (B) that are part of the Tgfb/Smad signaling pathway. Note that the Foxa genes expression levels (indicated with a yellow line) that are downstream of the Tgfb/Smad are restored to normal in the Asb2-Flna double mutants versus the Asb2-mutant hearts. embryos were analyzed for each condition. The efficiency of the CUBIC/immunostaining technique on e9.5 mouse embryo is evidenced by the clearly visible striations of the cardiac muscle fibers (Figure S1B). In order to identify Asb2 downstream targets in the heart, RNA sequencing (RNA-seq) analysis was per- formed on Nkx2-5+/Cre.Asb2fl/fl and control littermates (Figure 3B) (Figure S2C shows reduced levels of Asb2 transcripts in the Nkx2-5+/Cre.Asb2fl/fl knockout compared with the Nkx2-5+/Cre.Asb2fl/+ heterozygote control). The gene expression profile was greatly altered in the Asb2 mutant hearts compared with their control littermates (Group 2 versus Group 1) (Figure 3B). Of note, a number of genes that are mis-expressed in the Asb2 cardiac mutant hearts have been previously linked to abnormal cardiac looping in mice (Figure 2E) (Azhar et al., 2003; Bardot et al., 2017; Chen et al., 1997; Le Garrec et al., 2017; Mine et al., 2008; Ribeiro et al., 2007; Vincentz et al., 2011). Ingenuity Pathway Analysis also shows that ‘‘cardiovascular system devel- opment and function’’ as well as ‘‘cardiovascular disease’’ are among the top pathways altered in the Asb2 mutant hearts (Table S1, yellow highlights). Table S2 is an upstream analysis with the ones with a positive activation Z score > 1.5 highlighted in yellow. This list shows the pathways whose downstream targets are altered (upregulated or downregulated) in our knockouts versus controls. Targets with a positive Z score suggest upregulation pathways in the Asb2 mutant hearts. Asb2 Controls Cardiac Morphogenesis Partly through Regulating Filamin A Since Asb2 targets filamin proteins for degradation (Me´ tais et al., 2018) and Flna perturbations lead to car- diac defects and embryonic lethality (Feng et al., 2006), we investigated cardiac Flna expression in the Nkx2-5+/Cre.Asb2fl/fl. Flna expression in the control heart (Figure 3A, top panel) is restricted to endocardial and pericardial layers (red staining, white arrow heads). In the knockout embryos (Figure 3A, third panel), Flna’s expression domain is abnormally expanded to include the myocardial layer (white arrows), co-local- izing with Troponin T expression (green for Troponin and yellow for the co-localization). Moreover, in Nkx2-5+/Cre.Asb2fl/+ heterozygous hearts (Figure 3A, second panel), Flna is abnormally expressed in some cardiomyocytes of the OFT myocardium (yellow arrows) suggesting that Asb2 regulation of Flna turn- over is dose dependent. We then hypothesized that, if Asb2 cardiac mutant phenotype is due to overexpression of Flna, then concurrently deleting Flna along with Asb2 should suppress the Asb2 phenotype. (Please note that Flna is an x-linked gene so a knockout is denoted by fl/fl for female or fl/y for male, whereas a heterozygous is denoted by fl/x or fl/+.) To examine this hypothesis, we developed Nkx2-5+/Cre.Asb2fl/fl.Flnafl/y double mutants. Removal of Flna from the hearts of Nkx2-5+/Cre.Asb2fl/fl did not rescue lethality (Figure S2A). Approximately 16 litters were analyzed at E9.5 and 3 litters at E10.5. As expected, Nkx2-5+/Cre.Asb2fl/fl.Flnafl/fl double knockouts no longer harbor ectopic Flna expression in the myocardium (Figure S2B) as was previously seen with the Nkx2-5+/Cre.Asb2fl/fl single knockouts (Figure 3A). Instead, the double knockouts have normal endocardial expression of Flna similar to their control litter- mates (Figure S2B). RNA-seq analysis on e9.5 hearts from these mice show that their gene expression pro- file is closely related to the Nkx2-5+/Cre.Asb2fl/fl group (Group 2 versus Group 4) (Figure S2C shows reduced levels of Asb2 transcripts in the Nkx2-5+/Cre.Asb2fl/fl.Flnafl/y double knockout compared with the Nkx2-5+/Cre.Asb2fl/+ heterozygote control). However, some genes whose expression was altered in the Nkx2-5+/Cre.Asb2fl/fl group are restored to normal in the Nkx2-5+/Cre.Asb2fl/fl.Flnafl/y hearts (indicated by arrows and shown in Table S3). These results suggest that Flna concurrent deletion can restore the normal expression level of a subset of genes in the Asb2 mutant hearts. Among these genes are the iScience 23, 100959, March 27, 2020 7 A B C 8 iScience 23, 100959, March 27, 2020 Figure 4. Tgfb/Smad Signaling Activity Is Downstream Asb2-Flna in the Developing Heart (A) Schematic representation of Asb2-Flna-Smad2 interaction network using MetaCore Clarivate Analytics software. Note that Asb2 ubiquitinates and negatively regulates Flna, whereas Flna binds directly to and positively regulates Smad2. (B) Immunohistochemistry on Nkx2-5+/Cre.Asb2 mutant (middle panel) and Nkx2-5+/Cre.Asb2-Flna double mutant (last panel) murine hearts as well as wild- type controls (top panel) using pSmad2-specific antibody (green) and Troponin T (red). Note the nuclear localization of pSmad2 as a sign of Tgfb/Smad2 cycle activation. Examples of positive (purple arrowhead) and negative (yellow arrowheads) nuclei are indicated in the Wt sample (red box). DAPI marks all nuclei (AV, atrioventricular canal; V, primitive ventricle; OFT, outflow tract; Myo, myocytes; Endo, endocardial cells). Myocardial cells are marked by white arrows; endocardial cells are marked by white arrowheads. Scale bar is equivalent to 75 mm in the first two columns from the left and 25 mm in the third, fourth, and fifth columns as indicated in the figure. (C) Quantification of the immunostaining in (B) of percentage of pSmad2-positive nuclei in cardiomyocytes ((AV+V) Myo and OFT Myo) as well as endocardial cells (AV endo). Note the increased level of pSmad2-positive nuclei in Asb2-mutant myocytes that are restored to normal in the Asb2-FlnA double mutants. This regulation is not seen in the endocardial cells that do not express Asb2 (Figure 1E). n = 7 for Wt; n = 4 for Nkx2-5+/Cre.Asb2fl/fl; n = 3 for Nkx2-5+/Cre.Asb2fl/fl.Flnafl/fl. Data are represented as mean G SEM. * = p < 0.05 Control versus Nkx2-5+/Cre.Asb2fl/fl; # = p < 0.05 Nkx2-5+/Cre.Asb2fl/fl versus Nkx2-5+/Cre.Asb2fl/fl.Flnafl/fl; NS = not significant. Two-way ANOVA was used for analysis using GraphPad Prism. p < 0.05 is considered statistically significant. Foxa genes, which are downstream of the Tgfb/Smad signaling (Figure 3C, yellow line) (Tang et al., 2011). Other genes in the Tgfb/Smad pathway are also altered in both Asb2-mutant and Asb2.Flna double mutant hearts (Figure 3C). Figure S2D is a qPCR analysis confirming some of these altered genes. Tgfbr1 and InhA (which encodes a member of the Tgfb superfamily) are also among the positively regulated targets in the upstream analysis of the RNA-seq data of Asb2-mutant hearts versus control (Table S2). Both genes are no longer positively regulated in the upstream analysis of the list of genes corrected in the Asb2-Flna double mutant hearts (Table S3, yellow highlights). These data prompted further analysis of the Asb2/Flna regula- tion of TGFb/Smad signaling in the heart of these mice. Asb2 Regulates TGFb/Smad Signaling through Regulating Filamin A Protein TGFb signaling is initiated upon ligand-stimulated activation of serine/threonine receptor kinases that in turn lead to phosphorylation and activation of Smad proteins. Activated Smads interact with common signaling transducer Smad4, translocate to the nucleus, and activate downstream targets (Shi and Massague´ , 2003). Flna directly associates with Smad2 and Smad2 phosphorylation, and TGFb/Smad2 signaling is impaired in Fln-null human melanoma cells (Sasaki et al., 2001; Zhou et al., 2011). Moreover, FLNA mutations were linked to x-linked myxomatous valvular dystrophy, a multivalve degeneration disor- der, and disrupted TGFb/Smad2/3 signaling was implicated in the disease pathogenesis (Geirsson et al., 2012; Norris et al., 2010). Using the ‘‘Build Network’’ module in MetaCore Clarivate Analytics software, we investigated the Asb2-Flna-Smad2 interaction. As shown in Figure 4A, Asb2 negatively regulates Flna through ubiquitination and Flna positively regulates Smad2 through direct binding. Asb2, Flna, and Smad2 are shown in red for visualization. In order to investigate further Asb2/Flna regulation of TGFb/ Smad2 signaling in cardiac development, we immunostained E9.5 Asb2-mutant hearts with antisera directed against pSmad2 (Figure 4B) and then quantified the pSmad2-positive nuclei. Figure 4C shows significant increase in the percentage of pSmad2-positive nuclei in the Nkx2-5+/Cre.Asb2fl/fl myocytes (pre- viously shown to have overexpression of Flna [Figure 3A]) compared with their littermate controls. This increase was not seen in the endocardial cells where Flna expression is normal (Figure 3A). Interestingly, pSmad2 levels were restored to normal in the Nkx2-5+/Cre.Asb2fl/fl.Flnafl/fl (double mutant) myocytes further confirming that Asb2 regulates pSmad2 in the heart through the regulated turnover of Flna. Flna Removal from AHF-Cre.Asb2 Mutant Hearts Partially Rescues Embryonic Lethality To examine Flna expression in the AHF-Cre.Asb2fl/fl hearts (where Asb2 is knocked out in the RV and OFT only), Flna immunostaining was performed. As shown in Figure 5A, Flna expression (red) is restricted to the endocardial and epicardial layers in the control hearts (top panel, white arrow heads), whereas it is aber- rantly expressed in the myocardial layer of the OFT and RV only (red staining lower panel, white arrows), co-localizing with TroponinT expression (yellow staining lower panel) there. Flna expression was normal in the myocardial layer of the primitive left ventricle (PV) that harbors normal Asb2 expression and acts as an internal control in these mice. We then sought to examine the effect of further knocking out Flna from the AHF-Cre.Asb2-mutant hearts. To do this, we crossed Asb2fl/fl.Flnafl/fl with AHF-Cre.Asb2fl/+ mice. Our results show that AHF-Cre.Asb2fl/fl.Flnafl/y are born with the expected Mendelian ratios (Figure 5B); however, newborn pups die between P0.5 and P1.5. These results show that Flna deletion partially rescues Asb2 lethality. The AHF-Cre.Asb2fl/fl.Flnafl/+ also survive to birth albeit at a lower percent- age from what is expected by Mendelian ratios; these mice also die right after birth at P0.5. iScience 23, 100959, March 27, 2020 9 Troponin FlnA OFT DAPI Merge Ε9.5 250μm Ε9.5 100μm 100μm 100μm 25μm PV Troponin FlnA OFT PV DAPI Merge Ε9.5 250μm Ε9.5 100μm 100μm 100μm 25μm + / l f 2 b s A . e r C - F H A l f / l f 2 b s A . e r C - F H A A B AHF-Cre.Asb2fl/+ X Asb2fl/fl Expected Observed at P0.5 AHFCreAsb2fl/fl AHFCreAsb2fl/+ Control 25% 25% 50% 0/29 (0%) 13/29 (44.8%) 16/29 (55.1%) AHF-Cre.Asb2fl/+ X Asb2fl/fl.FlnAfl/fl e r C F H A Asb2fl/fl.FlnAfl/+ Asb2fl/fl.FlnAfl/y Asb2fl/+.FlnAfl/+ Asb2fl/+.FlnAfl/y Expected Observed at P0.5 1/31 (3.2%) 4/31 (12.9%) 6/31 (19.3%) 6/31 (19.3%) 12.5% 12.5% 12.5% 12.5% Control 50% 14/31 (45.2%) Dead at P0.5-1.5 Figure 5. Flna Removal from AHFCre.Asb2 Mutant Hearts Partially Rescues Their Lethality (A) Immunohistochemistry on E9.5 AHF-Cre.Abs2 mutant hearts (AHF-Cre.Asb2fl/fl, lower panel) and littermate controls (AHF-Cre.Asb2fl/+, top panel) using Flna (red)- and Troponin-T (green)-specific antibodies. Note overexpression of Flna in the OFT (white arrows) of the AHF-Cre.Asb2 mutant hearts but not the primitive left ventricle (PV) that harbors normal Asb2 levels thus serving as an internal control. Flna expression in the control hearts is restricted to the endocardial layer (white arrow heads). Scale bar is equivalent to 250 mm in the first column (left); 100 mm in the second, third, and fourth columns; 25 mm in the fifth column (far right) as indicated. (B) Table showing the survival of AHF-Cre.Asb2 mutant (top) and AHF-Cre.Asb2-Flna double mutant (bottom) mice. Note that no AHF-Cre.Abs2 mutant mice are observed at P0.5. However, the AHF-Cre.Asb2-Flna double mutant (Asb2fl/fl.Flnafl/y) mice are born at the expected Mendelian ratios. AHF-Cre.Asb2- mutant mice harboring one copy of Flna (Asb2fl/fl.Flnafl/+) are also born yet at lower percentage than what is expected by Mendelian genetics. These mice die, however, right after birth. P0.5, postnatal day 0.5. Asb2 Removal from the Anterior Heart Field Leads to Double Outlet Right Ventricle) in Mice To determine the cardiac defects of the AHF-Cre.Asb2fl/fl.Flnafl/+, we examined these mice at e16.5–e17.5 after the completion of cardiac morphogenesis but prior to the perinatal mortality associated with this genotype. Five litters were analyzed. Figure 6A shows the survival rate of these mice at E16.5. Gross examination of these hearts revealed that both the aorta and the pulmonary artery originate in the RV (Figure 6B, middle panel, yellow circle). In contrast, both the control hearts (Figure 6B, left panel) and those with AHF-Cre.Asb2-Flna double mutant (Fig- ure 6B, right panel) were grossly normal with the pulmonary artery originating in the RV and the aorta originating in 10 iScience 23, 100959, March 27, 2020 C * A B D Figure 6. AHF-Asb2 Mutant Hearts Have Double Outlet Right Ventricle and Ventricular Septal Defect (A) Table showing the survival of AHF-Cre.Asb2-Flna double mutant mice at E16.5. (B) E16.5 whole hearts of AHF-Cre.Asb2-mutant embryos with one copy of Flna (AHF-Cre.Asb2fl/fl.Flnafl/x), AHF- Cre.Asb2.Flna double mutants (AHF-Cre.Asb2fl/fl.Flnafl/fl), and wild-type control. Note that both the pulmonary artery (PA) and the aorta (Ao) are open in the right ventricle (RV) of the AHF-Cre.Asb2fl/fl.Flnafl/x hearts (yellow circle). Scale bar is equivalent to 0.02 mm as indicated. (C) Masson trichrome staining of E16.5 heart sections of control (Wt) (top), AHF-Cre.Asb2fl/fl.Flnafl/x (middle), and AHF- Cre.Asb2fl/fl.Flnafl/fl (bottom) embryos. Note that both the pulmonary artery and the aorta open in the right ventricle of the AHF-Cre.Asb2fl/fl.Flnafl/x hearts (yellow circle, middle panel) but not the Wt or the AHF-Cre.Asb2fl/fl.Flnafl/fl hearts. The AHF-Cre.Asb2fl/fl.Flnafl/x also have a VSD indicated by asterisk (middle panel, right). Ao, aorta; PA, pulmonary artery; RV, right ventricle; LV, left ventricle; IVS, interventricular septum. Scale bar is equivalent to 250 mm. (D) Number of E16.5 hearts with DORV in Wt, AHF-Cre.Asb2fl/fl.Flnafl/x, and AHF-Cre.Asb2fl/fl.Flnafl/fl embryos. Note that 5/5 Asb2fl/fl.Flnafl/x have DORV accompanied by a VSD suggesting 100% disease penetrance in these mice. the LV. Serial sections of mutant and control hearts (Figure 6C) further confirm that the AHF-Cre.Asb2fl/fl.Flnafl/x mice have DORV (Figure 6C middle panel, yellow oval). This is also accompanied by a ventricular septal defect (Figure 6C middle panel right, indicated by asterisk), a feature commonly associated with DORV in patients with congenital heart disease (Obler et al., 2008). As shown in Figure 6D, the DORV phenotype appeared to be fully penetrant in the AHF-Cre.Asb2fl/fl.Flnafl/x hearts. Notably, the DORV phenotype is corrected in the AHF-Cre.Asb2-Flna double mutant hearts (AHF-Cre.Asb2fl/fl.Flnafl/y, Figure 6C lower panel). Asb2 Is Required for Human Embryonic Stem Cell-Derived Cardiomyocyte Differentiation To further investigate if the requirement for Asb2 for cardiac development is conserved during human cardiomyo- (hESC)-derived cardiomyocyte in vitro cyte differentiation, we turned to human embryonic stem cell differentiation. Both ASB2 variants 1 (Asb2b in mice) and 2 (Asb2a in mice) are expressed at different stages of cardiomyocyte differentiation (Figure 7A, top and bottom graphs, respectively). Using CRISPR/Cas9 genome ed- iting technology, we then generated ASB2-null hESCs. The guides were designed in exon 2 (targeting variant 1 specifically) or exon 4 (targeting variants 1 and 2) (Figure S3A). Four wild-type (Wt) (received the CRISPR/Cas9 con- structs but failed to generate an in/del) and four knockout (KO) lines were generated. The genotype of all lines was confirmed by sequencing (refer to Transparent Methods), and the knockouts were confirmed by qPCR (Figure 7B, right panel). Wt clones were able to differentiate into beating cardiomyocytes, whereas all four KO lines failed to do so (Video S2, top panels for Wt clones and bottom panels for KO clones). Calcium cycling was also impaired in the Asb2-null derived hESCs (Video S3, left panel for Wt and right panel for KO, and Figure S3B). Two Wt and two KO lines were used for further investigation. qPCR analysis on RNA from cardiomyocytes derived from these cells iScience 23, 100959, March 27, 2020 11 A C B D Figure 7. ASB2 Is Essential for Human Embryonic Stem Cell (hESC)-derived Cardiomyocytes Differentiation (A) qPCR analysis of RNA from hESC-derived cardiomyocytes at different stages of differentiation. Note that both ASB2 variants are expressed at the different stages. N = 4 for D0, n = 5 for all other stages. Data are represented as mean G SEM. (B) qPCR analysis of RNA extracted from two different wild-type (Wt) clones and two ASB2 mutant (KO) clones d7 (left) and d15 (right) differentiated hESCs. Note reduced Troponin T (differentiation marker) transcripts in the mutants at d15 and no difference in MESP1 and NKX2-5 (cardiac progenitor markers) levels at d7. N = 3 per sample for d7 and N = 4 per sample for d14. Data are represented as mean G SEM. * = p < 0.05. Two-way ANOVA was used for analysis using GraphPad Prism. p < 0.05 is considered statistically significant. (C and D) Immunostaining of d8 (C) and d15 (D) differentiated Wt and ASB2 mutant cells (KO). Note reduced Troponin T (red)-positive mutant cells at d15 but no difference in NKX2-5 (green) levels between Wt and mutant cells at both stages. DAPI (blue) marks all nuclei. Scale bar is equivalent to 8 mm in (C) and 10 mm in (D) as indicated. shows that cardiac Troponin T transcript levels (TNNT2, marker of cardiomyocyte differentiation) are greatly reduced in the KO lines at d15 of differentiation (Figure 7B, right). On the other hand, both NKX2-5 and MESP1 (markers of cardiac progenitors) are normally expressed at d7 of differentiation (Figure 7B, left). This was further confirmed at the protein level by immunostaining that shows great reduction in cardiac Troponin T (red) expression at d15 (Figure 7D) and normal NKX2-5 levels (green) at days 15 and 8 (Figures 7C and 7D, respec- tively). These data suggest that Asb2-null hES cells can commit to the cardiac lineage but arrest in differentiation prior to the generation of functional cardiomyocytes. We then examined if ASB2 regulation of the TGFb/SMAD signaling seen in mice hearts is conserved in the human cells. As discussed above, upon TGFb/Smad activation, the signaling transducer Smad4 is 12 iScience 23, 100959, March 27, 2020 A B C Figure 8. ASB2 Is an Upstream Regulator of TGFb/SMAD Pathway in hESC-derived Cardiomyocytes (A) Immunostaining of d15 Wt and ASB2 mutant (KO) hESC-derived cardiomyocytes using SMAD4-specific antibody (green). Note reduced level of nuclear but not total mean fluorescence intensity of SMAD4-positive cells in the KOs (quantification graphs on the right). DAPI (blue) marks all nuclei. Scale bar is equivalent to 75 mm in the first column and 25 mm in the second and third columns as indicated. *: p < 0.005 significant versus Wt. Unpaired t test was used for analysis using GraphPad Prism. (B and C) (B) Western blot analysis of Wt and ASB2 mutant (KO) hESC-derived cardiomyocytes using SMAD4, SMAD2, and pSMAD2 antibodies. Note increase of SMAD4 and pSMAD2 in the mutant clones. Data are representative of three separate experiments (C) Quantification of the western blot analysis in (B). Data are average of quantification from three separate experiments. *: significant versus Wt1 and Wt2. p < 0.05 is considered statistically significant. One-way ANOVA was used for analysis using GraphPad Prism. translocated to the nucleus to activate downstream targets. Figure 8A shows an increase in nuclear SMAD4 (green) in the ASB2-null hES-derived cardiomyocytes. The nuclear versus total SMAD4 was quantified (Fig- ure 8A, right graph) showing that nuclear SMAD4 signal is doubled in the ASB2-null cells, whereas total Smad4 levels remain the same. Western blot analysis on total protein extracts from these cells also confirms significant increase in both SMAD4 and pSMAD2 protein levels (Figures 8B and 8C). This further confirms that the TGFb/SMAD signaling pathway is activated in Asb2-null cardiomyocytes. DISCUSSION In this study, we provide strong evidence for the role of Asb2 in controlling heart morphogenesis partly through its regulation of the actin-binding protein, Filamin A (Flna), and Tgfb/Smad signaling. We further show that this regulation is part of the DORV disease pathogenesis. Using CUBIC clearing technique combined with immunofluorescence and confocal microscopy, we show that the Asb2-mutant hearts have shorter heart tubes and do not form the fully looped helical structure. RNA-seq analysis also reveals that a number of genes that have been linked to cardiac looping defects iScience 23, 100959, March 27, 2020 13 are altered in the Asb2-mutant hearts. Recent morphological analysis of Asb2 null embryos suggested that cardiac looping in the total body null is largely intact (Me´ tais et al., 2018). To examine this more carefully, we exploited recent advances in tissue clearing coupled to optical sectioning and 3D reconstruction. This anal- ysis of the intact embryos, however, allows us to refine these findings and to examine the Asb2 mutant hearts more thoroughly and at a slightly later point in development. Although the hearts do start to loop, they arrest early on before making it to the helical fully looped heart. Measurement of the heart tube length reveals shorter heart tubes in the mutant hearts, which could explain the inability of the heart to fully form the helical structure. These data further reveal the important role of Asb2 regulation of cardi- omyocyte differentiation on the normal growth of the heart tube. We show that the CUBIC technique combined with immunofluorescence/confocal microscopy has distinct advantages over traditional morphological analysis for the phenotypic analysis of mouse embryos and allows for the detection of subtle phenotypes and morphological abnormalities. Our data further reveal that Filamin A (Flna) is aberrantly overexpressed in the Asb2-mutant cardiomyo- cytes that normally do not express Flna protein. This is consistent with the data that Metais et al. reported. We also show that this regulation is dose dependent. We further show that Asb2-Flna regulate Tgfb-Smad signaling. Nuclear pSmad2 is overexpressed in the Asb2-mutant hearts consistent with the upregulation of this signaling pathways. Its levels are restored to normal in the Asb2.Flna double mutants further showing that Asb2 regulates SMAD signaling through the Flna pathway. RNA-seq analysis also reveals that regula- tion of the Tgfb-Smad pathway in the Asb2-mutant hearts and the Foxa genes, which are downstream effectors of the Tgfb/Smad signaling (Tang et al., 2011), is in fact restored to normal in the Asb2-Flna dou- ble mutants. Flna has been previously shown to associate with Smad2 signaling (Sasaki et al., 2001). Moreover, Tgfb/Smad2/3 signaling is impaired in the multivalve degeneration disorder, X-linked myxoma- tous valvular dystrophy, in which FLNA mutations were reported (Geirsson et al., 2012; Norris et al., 2010). Our data provide further evidence for regulation of the Tgfb/Smad cycle by Flna and show that Asb2 is upstream of this regulatory pathway in the developing heart. Using human embryonic stem cell (hESC)-derived cardiomyocytes, we further show that the Asb2 role in embryonic heart differentiation is conserved in humans. Although ASB2-null hESCs are able to form cardiac progenitor cells (marked by expression of MESP1 and NKX2-5), they have an impaired ability to differen- tiate into beating TNNT+ cardiomyocytes. It is important to note here that the difference between Troponin T levels in the Asb2-null hESCs and the Asb2 mutants in vivo could be due to the total knockout in the cells that is more severe than the conditional in vivo knockout. Additionally, the cell system lacks signaling coming from the endocardium, which could also explain this difference. These results demon- strate that, in human PSCs differentiating in vitro, ASB2-mediated targeted degradation is required for the differentiation from NKX2-5+ progenitors to beating TNNT+ cardiomyocytes and that deletion of ASB2 results in a differentiation arrest at the progenitor stage. Moreover, the finding that these cells have increased levels of SMAD4 and pSMAD2, markers of TGFb/SMAD pathway activation, provides further evidence that ASB2 is an upstream regulator of the TGFb/SMAD pathway during the differentiation of human cardiomyocytes. These considerations become increasingly important given the potential of pluripotent stem cell-derived CMs to serve as a renewable cell source for cardiac regeneration in the injured heart. Given that Flna is a direct target of Asb2 that is aberrantly upregulated in Asb2-mutant cardiomyocytes, we then investigated whether Asb2 cardiac mutant embryonic lethality can be rescued by the concurrent deletion of Flna. Accordingly, we generated AHF-Cre.Asb2fl/fl.Flnafl/y double mutants. Our data show that, as opposed to the AHF-Cre.Asb2fl/fl single mutants that die by E11.5, the AHF-Cre.Asb2fl/fl.Flnafl/y double mutants are born with the expected Mendelian ratios (Figure 5B) but die shortly after birth. This suggests partial rescue of lethality seen in the AHF-Cre.Asb2fl/fl mutant hearts. Of note, we also generated Nkx2-5+/Cre.Asb2fl/fl.Flnafl/Y double mutants that did not rescue the Asb2 lethality (Figure S2A), suggesting a greater Asb2 dependency or a more complex phenotype in these mice. These findings are not surprising owing to earlier and broader expression domain of the Nxk2-5Cre compared with the AHF-Cre line. Moreover, AHF-Asb2-mutant mice with one Flna allele (AHF-Cre.Asb2fl/fl.Flnafl/x) sometimes also survive to P0, albeit at significantly lower than expected ratios (Fig- ure 5B). Not only do these results suggest a dose dependency of Asb2 and its target Flna but they also allow us to identify DORV associated with ventricular septal defect (VSD) as a penetrant cardiac phenotype. More inter- estingly, this phenotype was rescued when Flna was abrogated by the concurrent deletion of Flna showing that both Asb2 and Flna play a functional role in the pathogenesis of DORV. 14 iScience 23, 100959, March 27, 2020 During cardiac development, the heart first forms as a primitive heart tube that then elongates and starts to loop by addition of cells from the anterior, posterior, and second heart field at both the venous and arterial poles. At the onset of looping, left-right asymmetry in the heart becomes morphologically evident and any defects in this process can lead to complex congenital heart problems, including DORV and VSDs (Rams- dell, 2005). The heart is the first organ to break the left-right symmetry in the developing embryo, and it has been shown that the actin-cytoskeleton is fundamental for laterality and modulation associated with heart looping. It was shown to provide the built-in mechanism required for cells to acquire left-right asymmetry (Linask and Vanauker, 2007; Tee et al., 2015). Abnormalities in the control of construction of the cytoskel- eton has been previously shown to result in looping defects and ultimately lead to congenital heart prob- lems (Langdon et al., 2012; Linask and Vanauker, 2007). Our data and the data from Metais et al. provide solid evidence for the Asb2-Flna regulation of the actin cytoskeleton during heart morphogenesis (Me´ tais et al., 2018). Our data extend this regulation to show that it is important for normal heart tube and OFT development and, if perturbed, leads to DORV and VSD in the developing mammalian heart. Additionally, we suggest a mechanism where Asb2 downregulation leads to abnormal overexpression of Flna that ulti- mately leads to increased activity of the Tgfb/Smad2 signaling in the myocardium thus causing growth/ elongation defects and DORV in the mammalian heart. Indeed, prior reports have implicated the Tgfb superfamily and Smad2/3 in left-right asymmetry, and Tgfb2 mutant mice have been shown to develop DORV and die right after birth (Azhar et al., 2003; Sanford et al., 1997; Whitman and Mercola, 2001). In humans, a missense mutation in Flna (c.5290G>A (p.A1764T) has been reported in a patient with DORV (de Wit et al., 2011). Since missense mutations can result in both loss and gain of function, future studies will be required to determine the effect of this mutation on Flna expression and function. Thus, our data demonstrate a link between targeted protein turnover and the development of DORV and highlights the potential of the ASB2/FLNA axis as a diagnostic, prognostic, and/or therapeutic target for patients with DORV. Limitations of the Study Although our data show that the role of Asb2 in heart morphogenesis is conserved between mice and human, a limitation is that the in vivo murine system is a conditional knockout compared with the total knockout in the human cells system. Additionally, as we know, a cross talk between the endocardium and the myocardium occurs during heart morphogenesis, and this again is lacking in our human cell system. METHODS All methods can be found in the accompanying Transparent Methods supplemental file. DATA AND CODE AVAILABILITY RNA-seq data have been deposited in NCBI’s Gene Expression Omnibus (GEO). The accession number for the RNA-seq data reported in this paper is GEO: GSE145495. SUPPLEMENTAL INFORMATION Supplemental Information can be found online at https://doi.org/10.1016/j.isci.2020.100959. ACKNOWLEDGMENTS We thank all members of the Domian Lab for valuable insight and suggestions. We also thank the his- tology core at Dana Farber Cancer Institute/Harvard Medical School and the NextGen Sequencing core at Massachusetts General Hospital for technical support. This work was supported by the American Heart Association (17GRNT33630170), the Centre National de la Recherche Scientifique, and the Univer- sity of Toulouse. A.Y. is a recipient of the Fund for Medical Discovery (FMD) Award from the Massachu- setts General Hospital/Harvard Medical School. AUTHOR CONTRIBUTIONS Conceptualization, A.Y. and I.J.D.; Methodology, A.Y. and I.J.D.; Investigation A.Y., D.H., N.M., J.W.B., and S.D.; Writing – Original Draft, A.Y.; Writing – Review & Editing, A.Y., P.G.L., C.M.-L, P.T.L., and I.J.D.; Visu- alization, A.Y.; Supervision, A.Y. and I.J.D.; Funding Acquisition, A.Y. and I.J.D. iScience 23, 100959, March 27, 2020 15 DECLARATION OF INTERESTS The authors declare no competing interests. Received: August 16, 2019 Revised: December 17, 2019 Accepted: February 26, 2020 Published: March 27, 2020 REFERENCES Azhar, M., Schultz, J.E.J., Grupp, I., Dorn, G.W., Meneton, P., Molin, D.G.M., Gittenberger-de Groot, A.C., and Doetschman, T. (2003). Transforming growth factor beta in cardiovascular development and function. Cytokine Growth Factor Rev. 14, 391–407. Bardot, E., Calderon, D., Santoriello, F., Han, S., Cheung, K., Jadhav, B., Burtscher, I., Artap, S., Jain, R., Epstein, J., et al. (2017). Foxa2 identifies a cardiac progenitor population with ventricular differentiation potential. Nat. Commun. 8, 14428. Bello, N.F., Lamsoul, I., Heuze´ , M.L., Me´ tais, A., Moreaux, G., Calderwood, D.A., Duprez, D., Moog-Lutz, C., and Lutz, P.G. (2009). The E3 ubiquitin ligase specificity subunit ASB2beta is a novel regulator of muscle differentiation that targets filamin B to proteasomal degradation. Cell Death Differ. 16, 921–932. Bruneau, B.G. (2008). The developmental genetics of congenital heart disease. Nature 451, 943–948. Bulteau, A.L., Lundberg, K.C., Humphries, K.M., Sadek, H.A., Szweda, P.A., Friguet, B., and Szweda, L.I. (2001). Oxidative modification and inactivation of the proteasome during coronary occlusion/reperfusion. J. Biol. Chem. 276, 30057–30063. Chen, J.N., van Eeden, F.J., Warren, K.S., Chin, A., Nu¨ sslein-Volhard, C., Haffter, P., and Fishman, M.C. (1997). Left-right pattern of cardiac BMP4 may drive asymmetry of the heart in zebrafish. Development 124, 4373–4382. Davey, J.R., Watt, K.I., Parker, B.L., Chaudhuri, R., Ryall, J.G., Cunningham, L., Qian, H., Sartorelli, V., Sandri, M., Chamberlain, J., et al. (2016). Integrated expression analysis of muscle hypertrophy identifies Asb2 as a negative regulator of muscle mass. JCI Insight 1, https:// doi.org/10.1172/jci.insight.85477. de Wit, M.C.Y., de Coo, I.F.M., Lequin, M.H., Halley, D.J.J., Roos-Hesselink, J.W., and Mancini, G.M.S. (2011). Combined cardiological and neurological abnormalities due to filamin A gene mutation. Clin. Res. Cardiol. 100, 45–50. de Wit, M.C.Y., Kros, J.M., Halley, D.J.J., de Coo, I.F.M., Verdijk, R., Jacobs, B.C., and Mancini, G.M.S. (2009). Filamin A mutation, a common cause for periventricular heterotopia, aneurysms and cardiac defects. J. Neurol. Neurosurg. Psychiatry 80, 426–428. Domian, I.J., Chiravuri, M., van der Meer, P., Feinberg, A.W., Shi, X., Shao, Y., Wu, S.M., Parker, K.K., and Chien, K.R. (2009). Generation of functional ventricular heart muscle from mouse ventricular progenitor cells. Science 326, 426–429. 16 iScience 23, 100959, March 27, 2020 Fan, Y., Xie, P., Zhang, T., Zhang, H., Gu, D., She, M., and Li, H. (2008). Regulation of the stability and transcriptional activity of NFATc4 by ubiquitination. FEBS Lett. 582, 4008–4014. Feng, Y., Chen, M.H., Moskowitz, I.P., Mendonza, A.M., Vidali, L., Nakamura, F., Kwiatkowski, D.J., and Walsh, C.A. (2006). Filamin A (FLNA) is required for cell-cell contact in vascular development and cardiac morphogenesis. Proc. Natl. Acad. Sci. U S A 103, 19836–19841. Fujita, M., Mitsuhashi, H., Isogai, S., Nakata, T., Kawakami, A., Nonaka, I., Noguchi, S., Hayashi, Y.K., Nishino, I., and Kudo, A. (2012). Filamin C plays an essential role in the maintenance of the structural integrity of cardiac and skeletal muscles, revealed by the medaka mutant zacro. Dev. Biol. 361, 79–89. Fukuda, R., Gunawan, F., Beisaw, A., Jimenez- Amilburu, V., Maischein, H.-M., Kostin, S., Kawakami, K., and Stainier, D.Y.R. (2017). Proteolysis regulates cardiomyocyte maturation and tissue integration. Nat. Commun. 8, 14495. Geirsson, A., Singh, M., Ali, R., Abbas, H., Li, W., Sanchez, J.A., Hashim, S., and Tellides, G. (2012). Modulation of transforming growth factor-b signaling and extracellular matrix production in myxomatous mitral valves by angiotensin II receptor blockers. Circulation 126, S189–S197. Glickman, M.H., and Ciechanover, A. (2002). The ubiquitin-proteasome proteolytic pathway: destruction for the sake of construction. Physiol. Rev. 82, 373–428. Guibal, F.C., Moog-Lutz, C., Smolewski, P., Di Gioia, Y., Darzynkiewicz, Z., Lutz, P.G., and Cayre, Y.E. (2002). ASB-2 inhibits growth and promotes commitment in myeloid leukemia cells. J. Biol. Chem. 277, 218–224. Heuze´ , M.L., Lamsoul, I., Baldassarre, M., Lad, Y., Le´ veˆ que, S., Razinia, Z., Moog-Lutz, C., Calderwood, D.A., and Lutz, P.G. (2008). ASB2 targets filamins A and B to proteasomal degradation. Blood 112, 5130–5140. Hu, D., Linders, A., Yamak, A., Correia, C., Kijlstra, J.D., Garakani, A., Xiao, L., Milan, D.J., van der Meer, P., Serra, M., et al. (2018). Metabolic maturation of human pluripotent stem cell- derived cardiomyocytes by inhibition of HIF1a and LDHA. Circ. Res. 123, 1066–1079. Jung, T., Catalgol, B., and Grune, T. (2009). The proteasomal system. Mol. Aspects Med. 30, 191–296. Kedar, V., McDonough, H., Arya, R., Li, H.-H., Rockman, H.A., and Patterson, C. (2004). Muscle- specific RING finger 1 is a bona fide ubiquitin ligase that degrades cardiac troponin I. Proc. Natl. Acad. Sci. U S A 101, 18135–18140. (cid:1) Capek, M., Radochova´ , B., Jana´ (cid:1)cek, Kolesova´ , H., J., and Sedmera, D. (2016). Comparison of different tissue clearing methods and 3D imaging techniques for visualization of GFP-expressing mouse embryos and embryonic hearts. Histochem. Cell Biol. 146, 141–152. Kyndt, F., Gueffet, J.-P., Probst, V., Jaafar, P., Legendre, A., Le Bouffant, F., Toquet, C., Roy, E., McGregor, L., Lynch, S.A., et al. (2007). Mutations in the gene encoding filamin A as a cause for familial cardiac valvular dystrophy. Circulation 115, 40–49. Laine, A., and Ronai, Z. (2005). Ubiquitin chains in the ladder of MAPK signaling. Sci. STKE 2005, re5. Langdon, Y., Tandon, P., Paden, E., Duddy, J., Taylor, J.M., and Conlon, F.L. (2012). SHP-2 acts via ROCK to regulate the cardiac actin cytoskeleton. Development 139, 948–957. Le Garrec, J.-F., Domı´nguez, J.N., Desgrange, A., Ivanovitch, K.D., Raphae¨ l, E., Bangham, J.A., Torres, M., Coen, E., Mohun, T.J., and Meilhac, S.M. (2017). A predictive model of asymmetric morphogenesis from 3D reconstructions of mouse heart looping dynamics. Elife 6, https:// doi.org/10.7554/eLife.28951. Li, J., Horak, K.M., Su, H., Sanbe, A., Robbins, J., and Wang, X. (2011). Enhancement of proteasomal function protects against cardiac proteinopathy and ischemia/reperfusion injury in mice. J. Clin. Invest. 121, 3689–3700. Linask, K.K., and Vanauker, M. (2007). A role for the cytoskeleton in heart looping. ScientificWorldJournal 7, 280–298. Lombardi, R., Dong, J., Rodriguez, G., Bell, A., Leung, T.K., Schwartz, R.J., Willerson, J.T., Brugada, R., and Marian, A.J. (2009). Genetic fate mapping identifies second heart field progenitor cells as a source of adipocytes in arrhythmogenic right ventricular cardiomyopathy. Circ. Res. 104, 1076–1084. Ma¨ nner, J. (2009). The anatomy of cardiac looping: a step towards the understanding of the morphogenesis of several forms of congenital cardiac malformations. Clin. Anat. 22, 21–35. Martinsen, B.J., and Lohr, J.L. (2015). Cardiac development. In Handbook of Cardiac Anatomy, Physiology, and Devices, P.A. Laizzo, ed. (Springer), pp. 23–33. Mercola, M., Ruiz-Lozano, P., and Schneider, M.D. (2011). Cardiac muscle regeneration: lessons from development. Genes Dev. 25, 299–309. Me´ tais, A., Lamsoul, I., Melet, A., Uttenweiler- Joseph, S., Poincloux, R., Stefanovic, S., Valie` re, A., Gonzalez de Peredo, A., Stella, A., Burlet- Schiltz, O., et al. (2018). Asb2a-filamin A axis is essential for actin cytoskeleton remodeling during heart development. Circ. Res. 122, e34–e48. Mine, N., Anderson, R.M., and Klingensmith, J. (2008). BMP antagonism is required in both the node and lateral plate mesoderm for mammalian left-right axis establishment. Development 135, 2425–2434. Murdaca, J., Treins, C., Monthoue¨ l-Kartmann, M.-N., Pontier-Bres, R., Kumar, S., Van Obberghen, E., and Giorgetti-Peraldi, S. (2004). Grb10 prevents Nedd4-mediated vascular endothelial growth factor receptor-2 degradation. J. Biol. Chem. 279, 26754–26761. Nemer, M. (2008). Genetic insights into normal and abnormal heart development. Cardiovasc. Pathol. 17, 48–54. Norris, R.A., Moreno-Rodriguez, R., Wessels, A., Merot, J., Bruneval, P., Chester, A.H., Yacoub, M.H., Hage` ge, A., Slaugenhaupt, S.A., Aikawa, E., et al. (2010). Expression of the familial cardiac valvular dystrophy gene, filamin-A, during heart morphogenesis. Dev. Dyn. 239, 2118–2127. Obler, D., Juraszek, A.L., Smoot, L.B., and Natowicz, M.R. (2008). Double outlet right ventricle: aetiologies and associations. J. Med. Genet. 45, 481–497. Pagan, J., Seto, T., Pagano, M., and Cittadini, A. (2013). Role of the ubiquitin proteasome system in the heart. Circ. Res. 112, 1046–1058. Powell, S.R., Davies, K.J.A., and Divald, A. (2007). Optimal determination of heart tissue 26S- proteasome activity requires maximal stimulating ATP concentrations. J. Mol. Cell. Cardiol. 42, 265–269. Ramsdell, A.F. (2005). Left-right asymmetry and congenital cardiac defects: getting to the heart of the matter in vertebrate left-right axis determination. Dev. Biol. 288, 1–20. single-cell resolution by tissue decolorization. Cell 159, 911–924. Ribeiro, I., Kawakami, Y., Bu¨ scher, D., Raya, A., Rodrı´guez-Leo´ n, J., Morita, M., Rodrı´guez Esteban, C., and Izpisu´ a Belmonte, J.C. (2007). Tbx2 and Tbx3 regulate the dynamics of cell proliferation during heart remodeling. PLoS One 2, e398. Sanford, L.P., Ormsby, I., Gittenberger-de Groot, A.C., Sariola, H., Friedman, R., Boivin, G.P., Cardell, E.L., and Doetschman, T. (1997). TGFbeta2 knockout mice have multiple developmental defects that are non-overlapping with other TGFbeta knockout phenotypes. Development 124, 2659–2670. Sasaki, A., Masuda, Y., Ohta, Y., Ikeda, K., and Watanabe, K. (2001). Filamin associates with Smads and regulates transforming growth factor- beta signaling. J. Biol. Chem. 276, 17871–17877. Shi, Y., and Massague´ , J. (2003). Mechanisms of TGF-beta signaling from cell membrane to the nucleus. Cell 113, 685–700. Spaich, S., Will, R.D., Just, S., Spaich, S., Kuhn, C., Frank, D., Berger, I.M., Wiemann, S., Korn, B., Koegl, M., et al. (2012). F-box and leucine-rich repeat protein 22 is a cardiac-enriched F-box protein that regulates sarcomeric protein turnover and is essential for maintenance of contractile function in vivo. Circ. Res. 111, 1504–1516. Spinner, C.A., Uttenweiler-Joseph, S., Metais, A., Stella, A., Burlet-Schiltz, O., Moog-Lutz, C., Lamsoul, I., and Lutz, P.G. (2015). Substrates of the ASB2a E3 ubiquitin ligase in dendritic cells. Sci. Rep. 5, 16269. Srivastava, D. (2006). Making or breaking the heart: from lineage determination to morphogenesis. Cell 126, 1037–1048. Tainaka, K., Kubota, S.I., Suyama, T.Q., Susaki, E.A., Perrin, D., Ukai-Tadenuma, M., Ukai, H., and Ueda, H.R. (2014). Whole-body imaging with Tang, Y., Shu, G., Yuan, X., Jing, N., and Song, J. (2011). FOXA2 functions as a suppressor of tumor metastasis by inhibition of epithelial-to- mesenchymal transition in human lung cancers. Cell Res. 21, 316–326. Tee, Y.H., Shemesh, T., Thiagarajan, V., Hariadi, R.F., Anderson, K.L., Page, C., Volkmann, N., Hanein, D., Sivaramakrishnan, S., Kozlov, M.M., and Bershadsky, A.D. (2015). Cellular chirality arising from the self-organization of the actin cytoskeleton. Nat. Cell Biol. 17, 445–457. Valde´ s-Mas, R., Gutie´ rrez-Ferna´ ndez, A., Go´ mez, J., Coto, E., Astudillo, A., Puente, D.A., Reguero, J.R., A´ lvarez, V., Morı´s, C., Leo´ n, D., et al. (2014). Mutations in filamin C cause a new form of familial hypertrophic cardiomyopathy. Nat. Commun. 5, 5326. van der Flier, A., and Sonnenberg, A. (2001). Structural and functional aspects of filamins. Biochim. Biophys. Acta 1538, 99–117. Vincentz, J.W., Barnes, R.M., and Firulli, A.B. (2011). Hand factors as regulators of cardiac morphogenesis and implications for congenital heart defects. Birth Defects Res. A Clin. Mol. Teratol. 91, 485–494. Whitman, M., and Mercola, M. (2001). TGF-beta superfamily signaling and left-right asymmetry. Sci. STKE 2001, re1. Yamak, A., and Nemer, M. (2015). Role of embryonic and differentiated cells in cardiac development. In Biomaterials for Cardiac Regeneration, E.J. Suuronen and M. Ruel, eds. (Springer), pp. 37–70. Zhou, A.-X., Toylu, A., Nallapalli, R.K., Nilsson, G., Atabey, N., Heldin, C.-H., Bore´ n, J., Bergo, M.O., and Akyu¨ rek, L.M. (2011). Filamin a mediates HGF/c-MET signaling in tumor cell migration. Int. J. Cancer 128, 839–846. iScience 23, 100959, March 27, 2020 17 iScience, Volume 23 Supplemental Information Loss of Asb2 Impairs Cardiomyocyte Differentiation and Leads to Congenital Double Outlet Right Ventricle Abir Yamak, Dongjian Hu, Nikhil Mittal, Jan W. Buikema, Sheraz Ditta, Pierre G. Lutz, Christel Moog-Lutz, Patrick T. Ellinor, and Ibrahim J. Domian Supplementary Data Transparent Methods Animals. All animal experimentations were carried out in accordance with institutional guidelines for animal care. Experiments were approved by the Massachusetts General Hospital’s Subcommittee on Research Animal Care (SRAC), which serves as the Institutional Animal Care and Use Committee (IACUC) as required by the Public Health Service (PHS) Policy on Humane Welfare Regulations. The program and facilities have been fully accredited by the American Association for the Accreditation of Laboratory Animal Care (AAALAC) since July 30, 1993. The institutional assurance number with the Office for Protection from Research Risks at the N.I.H. is DI6-00361. All mice lines were kept on a C57BL/6 background. Approximately, 20 AHF-Cre, 20 Nkx2-5+/Cre, 100 Asb2fl/fl and 100 Asb2fl/fl.Flnafl/fl mice were used. To isolate embryos from pregnant females, cervical dislocation was used for euthanasia which is required for embryo collection in mice. Sex of the embryos was not an influence in this study due to the very early developmental stage. Embryos were analyzed at E8.5, E9.5, E10.5, and E11.5 as indicated in the results section where applicable. For the double outlet right ventricle analysis, hearts of E16.5 embryos were used. Generation of Asb2 and Flna knockout embryos. Asb2fl/fl or Asb2fl/fl.Flnafl/fl females were mated with Nkx2-5+/Cre or AHF-Cre male mice and plugs were checked on a daily basis. The day a plug is seen is considered embryonic day e0.5. Asb2fl/fl, Flnafl/fl, Nkx2-5+/Cre and AHF-Cre mice are previously described (Lamsoul et al., 2013; Lombardi et al., 2009; Pinto et al., 2014). Mice genotypes (adult and embryos) were determine by PCR genotyping. Genotyping oligos used are: Flna flox: 5’ TCT TCC TCT TTC AGC TGG 3’and 5’ ACA ACT GCT GCT CCA GAG 3’; Asb2 flox: 5’ CAGTGTCTGCTCTGAGGTCTCTC 3’ and 5’ CAATCTCTCCCTGGTAGAAACAGTTTGG 3’; Nkx2-5 Cre: 5’ GATTAGCTTAAGCGGAGCTGGGTGTCC 3’ and 5’ GCCGCATAACCAGTGAAACAGCATTGC 3’; AHF-Cre: 5’ CCAGGCAAAGGCAAGAATAA 3’ and 5’ ATGTTTAGCTGGCCCAAATG 3’. Immunohistochemistry. Immunofluorescence was done as previously described (Domian et al., 2009). Tissues were permeabilized with 0.3% Triton and antigen retrieval was done using citrate buffer. Tissues were blocked with goat or donkey serum and primary antibodies were incubated overnight at 4oC. Secondary antibodies linked to appropriate alexa fluor were incubated for 1 hour at room temperature. Excess antibodies were washed with Phosphate buffer saline with 0.2% tween-20. Tissues were mounted with prolong gold anti-fade mounting media. Antibodies used were: Asb2 (Abcam, ab13710); Filamin a (Abcam, ab76289); Nkx2-5 (Invitrogen, PA5-49431); pSmad2(Millipore, AB3849); Troponin T (Thermo- Scientific, MA5-12960; SMAD4 (Proteintech, 10231-1-AP). Masson Trichrome Staining was done on paraffin heart sections using the American Mastertech Scientific kit (Item No. KTMTR) according to the manufacturer’s protocol. Paraffin sections were deparaffinized with 3 rounds of xylene followed by rehydration with serial dilutions of ethanol baths prior to staining. Outflow tract measurements were done on 2D images using ImageJ. The landmarks used for measurement are as shown in supplementary figure 1D. CUBIC clearing and Immunostaining. Embryos were immersed in CUBIC-1 solution (25% urea, 15% TritonX-100, 25% N,N,N,N-tetrakis(2-hydroxypropyl)ethyl-enediamine) at 37oC with gentle shaking till efficiently cleared (2-5 days depending on developmental stage). Following clearing, embryos were washed thoroughly with PBS and stained with Troponin T and/or Filamin A antibodies for 4-5 days (at 4oC), washed with PBS and then incubated with the corresponding secondary antibodies coupled to Alexa Fluor 488 or 546 for additional 3-4 days (at 4oC). DAPI was added to CUBIC-1 solution and the following PBS washes to mark nuclei. Following staining, embryos were then cleared with CUBIC-2 solution (50% sucrose, 25% urea and 10% 2,2’,2’-nitrilotriethanol) for 1-2 days at 37oC with gentle shaking and then immediately transferred to immersion oil and imaged with laser confocal microscopy (Leica TCS SP8). 90-120 z-stacks were taken for each embryo that were then used to generate the 3D reconstructions using either the Leica software or image J. The 3D images were then further analyzed for phenotypic defects. At least 5 embryos were analyzed for each condition. The clearing/staining technique was adapted from the established protocol by Kolesova et al (Kolesová et al., 2016). Heart tube measurements were done on 3D images using ImageJ. The landmarks used for measuring the tube’s length are as described in Le Garrec et al paper (Le Garrec et al., 2017). RNA Extraction and qPCR. RNA extraction was done using the Qiagen RNeasy Micro Kit (Cat No. 74004) according to the manufacturer’s protocol. qPCR analysis was done using the Applied Biosystems PowerUp SYBR Green Master mix (Cat No. A25742) according to manufacturer’s protocol. Oligos used were: msAsb2a: 5’ GCTCTGTTTCACTCTGGCTCT 3’ and 5’ CTTCAGCACGGGGTCCATAG 3’; msAsb2b: 5’ AACCACCAGCCAGGACATTT 3’ and 5’ ACTTCTGCATGACCCCTTGG 3’; huASB2V1: 5’ ATTGGGCAGGAGGAGTACAG 3’ and 5’ AACTCTCAGGAGGTGCAGT 3’; huASB2V2: 5’ ATGACCCGCTTCTCCTATGC 3’ and 5’ CGAACTCTCAGGAGGTGCAG 3’. huTNNT2: 5’ ACTTGGAGGCAGAGAAGTTCG 3’ and 5’ CCCGGTGACTTTAGCCTTCC 3’; huNKX2-5: 5’ CGCACAGCTCTTTCTTTTCGG 3’ and 5’ CGCCTTCTATCCACGTGCC 3’; huMESP1: 5’ CTTTTTGGCCTCAGCACCTTC 3’ and 5’ AGTGTCTAGCCCTATGGGTCC 3’. RNA Sequencing. RNA was extracted from e9.5 embryo hearts using the Qiagen RNeasy Micro Kit (Cat No. 74004) and sent to the MGH Next Generation sequencing core. The libraries were sequenced using illumina HiSeq platform. Splice-aware alignment program STAR was used to map the sample sequencing reads to the Mus musculus mm10 reference genome. Gene expression counts were calculated using HTSeq based on current Ensembl annotation for mm10. The R package “edgeR” was then employed to make differential gene expression calls. Pathway analyses were done using “MetaCore-Clarivate” and “Ingenuity Pathway Analysis-Qiagen” softwares. Human Pluripotent Stem Cell Culture and Differentiation. HUES9 hESC line (NIH Human Embryonic Stem Cell Registry Number 0022, generated by HSCI iPS Core at Harvard University) was used in generating CRISPR KO cell line. hESC culture, differentiation and dissociation protocols were based on previously published works (Hu et al., 2018). Briefly, hESCs were cultured in Essential 8 Medium (Thermo Fisher Scientific, MA) in Matrigel (BD Biosciences) coated cell culture plates. hESCs were differentiated in RPMI GlutaMAX (Thermo Fisher Scientific, MA) plus Gem21 NeuroPlex Serum-Free Supplement without insulin (Gemini Bio Products, CA) for the first 5 days. Small molecules CHIR99021 (STEMCELL Technologies, Vancouver, Canada) and IWP-4 (STEMCELL Technologies, Vancouver, Canada) were added on day 1 and 3, respectively. Differentiation media was then switched to RPMI GlutaMAX plus Gem21 NeuroPlex Serum-Free Supplement from day 7 to 10. Differentiating hESCs then underwent glucose starvation for 6 days, which resulted in highly pure populations of beating CMs. hESC-CMs were re-plated onto Matrigel coated PDMS plates for confocal imaging. Imaging procedure and analysis were done based on previously published methods (Kijlstra et al., 2015). Briefly, Fluo-4, AM (Thermo Fisher Scientific, MA) calcium indicator were incubated with hESC-CMs prior to imaging. Movies of CMs at randomly selected regions were acquired in both DIC and GFP channels at 50 frames per second for 10 seconds. Calcium transients were analyzed using ImageJ software. In vitro differentiation of the SHF-dsRed/Nkx2.5-eGFP cells was done as previously described (Domian et al., 2009). Generation of ASB2-null hESCs. We used CRISPR/Cas9 genome editing technology to generate the ASB2-null hESCs according to the described protocol (Ran et al., 2013). Guide RNAs (gRNAs) specific for hASB2 variant 1 (equivalent to Asb2β in mouse) and those common for both variants 1 & 2 (mouse Asb2β & α respectively) were designed using CRISPR design online tool, cloned into CRISPR/Cas9-GFP plasmid backbone (pSpCas9 from Addgene) and sequenced. Plasmids with the efficient gRNAs were delivered by electroporation to hESCs. Single cell CRISPR clone selection, expansion and sequencing protocols were adapted from Peters et al (Peters et al., 2008). Following FACS selection, GFP+ hESCs were plated sparsely onto Matrigel coated dishes for growing single cell clones. After 10 days, individual clones were picked, plated into 96-well plates, and sequenced. Four clones harboring ASB2 gene locus modification along with four wild type (Wt) clones were expanded and differentiated into CMs for further analysis. 6 guides were tested individually (sequences below). Guides 1 and 6 were successful in inducing the knockouts. 1: used were: Guide 5' CACCGGTTGGTACATGCAGACGCGG 5' Guides AAACCCGCGTCTGCATGTACCAACC 3’; Guide 2: 5’ CACCGGTCCGCTAGGCTCTGCTCGA and 5’ AAACTCGAGCAGAGCCTAGCGGACC 3’; Guide 3: 5' CACCGGGCCCCTTGTCTTGTCCGCT 3’ and 5’ AAACAGCGGACAAGACAAGGGGCCC 3’; Guide 4: 5' CACCGGCCCGGGCCGGCGAACTCTC 3’ and 5' AAACGAGAGTTCGCCGGCCCGGGCC 3’; Guide 5: 5' CACCGCTCCTGAGAGTTCGCCGGCC 3’ and 5' AAACGGCCGGCGAACTCTCAGGAGC 3’; Guide 6: 5' CACCGCTGCACGAGGCCGCATACTA 3’ and 5' AAACTAGTATGCGGCCTCGTGCAGC 3’ and 3’ Western blot analysis. Total protein extracts were prepared using RIPA buffer. Proteins were run on 10% TGX pre-cast gels from biorad and transferred to PVDF membranes using Trans-blot turbo transfer kit (Biorad). Membranes were blocked with 5% non-fat milk or BSA (in case of pSmad2) and primary antibodies were incubated overnight at 4oC. Secondary antibodies linked to HRP (horseradish peroxidase) were incubated for 1 hour at room temperature and signal was revealed using super signal west femto or pico ECL substrates (Thermo-scientific). Antibodies used were Smad2 (5339, Cell Signaling), pSmad2 (3108, cell signaling) and Smad4 (ab40759, ABCAM). Western blots were then quantified using the Image Lab software. Statistical Analysis: Standard t-test was used for the QPCR analysis and the heart tube measurements. One-way ANOVA was used for the western blot quantification as well as the percentages of Smad4 and pSmad2 positive cells where. p<0.05 is considered statistically significant. Data and Software availability: RNAseq data have been deposited in NCBI’s Gene Expression Omnibus (GEO) and can be accessed through GEO Series accession number GSE145495. B. DAPI Troponin T Control Nkx2-5+/Cre.Asb2fl/+ Nkx2-5+/Cre.Asb2fl/fl 25μm + / fl 2 b s A . e r C / + 5 - 2 x k N fl / fl 2 b s A . e r C / + 5 - 2 x k N l o r t n o C H D P A G / 2 b s A l o r t n o C o t e v i t a e r l 1 0.8 0.6 0.4 0.2 0 Asb2 GAPDH E9.5 100μm A. C. 75 50 37 25 l o r t n o C t n a t u M E9.5 100μm D. Control AHF-Cre.Asb2fl/fl DAPI Troponin OFT OFT Ε10.5 100μm 100μm ) m μ ( h t g n e l T F O 600 400 200 0 * Control AHF-Cre.Asb2fl/fl Supplementary figure 1, related to figure 2. A. Western Blot analysis on hearts of e9.5 Nkx2- 5+/Cre.Asb2fl/+, Nkx2-5+/Cre.Asb2fl/fl, and their control littermates using Asb2 antibody and GAPDH for loading control. Notice reduced Asb2 protein levels in the heterozygous mice (fl/+) and the complete loss of Asb2 in the knockout mice (fl/fl) (quantification analysis on the rights) (5-6 hearts were uses per condition). B & C. CUBIC/Immunofluorescence of E9.5 mice embryos. B. High magnification showing the cardiac myocardial region of an E9.5 mouse embryo cleared with CUBIC and immuno-stained for Troponin T (green). Blue marls DAPI. Note the visible striations (yellow arrows). Scale bar is equivalent to 25µm. C. Serial sections of Control (top) and Mutant (bottom) E9.5 cleared/stained mice embryos, showing the heart region. Troponin T (green) was used to mark the myocardium. Note the bulging in the Control heart (right arrow) which is missing in the Mutant. D. Immunohistochemistry on E10.5 AHF-Cre.Asb2 hearts (AHF- Cre.Asb2fl/fl, right panel) and littermate control (left panel) using Troponin-T (gree)-specific antibody. Note the shorter outflow tract (OFT) of the AHF-Cre.Asb2fl/fl heart. Scale bar is equivalent to 100µm. 4 Control and 3 knockout hearts were analyzed. *: p<0.005 significant vs. control. Unpaired t-test was used for analysis using Graphpad Prism. Supplementary Figure 1, related for figure 2. A. Western Blot analysis on hearts of e9.5 Nkx+/Cre.Asb2fl/+, Nkx+/Cre.Asb2fl/fl and their control littermates using Asb2 antibody and GAPDH for loading control. Notice reduced Asb2 protein levels in the heterozygous mice (fl/+) and the complete loss of Asb2 in the knock out mice (fl/fl) (quantification analysis on the right) (5-6 hearts were used per condition). B. & C. CUBIC/Immunofluorescence in e9.5 mice embryos. B&C. CUBIC/Immunofluorescence in e9.5 mice embryos. B. High magnification showing the cardiac myocardial region of an E9.5 mouse embryo cleared with CUBIC and immuno-stained for TroponinT (green). Blue marks DAPI. Note the visible striations (yellow arrows). Scale bar is equivalent to 25μm. C. Serial sections of Control (top) and Mutant (bottom) E9.5 cleared/stained mice embryos, showing the heart region. TroponinT (green) was used to mark the myocar- dium. Note the bulging in the Control heart (right arrow) which is missing in the Mutant. D. Immunohistochem- istry on E10.5 AHF-Cre.Abs2 hearts (AHF-Cre.Asb2fl/fl, right panel) and littermate control (left panel) using Troponin-T (green)-specific antibody. Note the shorter outflow tract (OFT) of the AHF-Cre.Asb2fl/fl heart. Scale bar is equivalent to 100μm. 4 Control and 3 knockout hearts were analyzed. *: p<0.005 significant vs control. Unpaired t-test was used for analysis using Graphpad prism. A.A.A.A. B. Control Nkx2-5+/CreAsb2fl/fl.Flnafl/+ Nkx2-5+/CreAsb2fl/fl.Flnafl/y Troponin FlnA DAPI E9.5 0.2mm 0.2mm 0.2mm E10.5 0.4mm 0.4mm 0.4mm C. D. l o r t n o C fl / fl a n F l . fl / fl 2 b s A / . e r C + 5 - 2 x k N Ε9.5 Troponin 75μm 75μm 75μm FlnA DAPI Ε9.5 75μm 75μm 75μm + / fl 2 b s A / . e r C + 5 - 2 x k N o t e v i t a e R l Asb2 * * 1.5 1.0 0.5 0.0 Nkx2-5+/Cre.Asb2fl/+ Nkx2-5+/Cre.Asb2fl/fl Nkx2-5+/Cre.Asb2fl/fl.Flnafl/fl s l e v e l A N R m e v i t a l e R 6.0 4.5 3.0 3.0 2.5 2.0 1.5 1.0 0.5 0.0 * # @ * # * * # * Shh Hand2 Foxa2 Foxa3 *: significant vs Nkx2-5+/Cre.Asb2fl/+ Nkx2-5+/Cre.Asb2fl/+ Nkx2-5+/Cre.Asb2fl/fl #: significant vs Nkx2-5+/Cre.Asb2fl/+.Flnafl/y Nkx2-5+/Cre.Asb2fl/+.Flnafl/y @: significant vs Nkx2-5+/Cre.Asb2fl/fl.Flnafl/y Nkx2-5+/Cre.Asb2fl/fl.Flnafl/y Supplementary Figure 2, related to figure 3. A. Nkx2-5+/Cre.Asb2.Flna E9.5 and E10.5 embryos. Note Supplementary Figure 2, related to figure 3. A. Nkx2-5+/Cre.Asb2.Flna E9.5 and E10.5 embryos. Note the the smaller Nkx2-5+/Cre.Asb2fl/fl.Flnafl/+ and Nkx2-5+/Cre.Asb2fl/fl.Flnafl/y at E9.5 and E10.5. Nkx2- smaller Nkx2-5+/CreAsb2fl/fl.Flnafl/+ and Nkx2-5+/CreAsb2fl/fl.Flnafl/y at E9.5 and E10.5. Nkx2-5+/CreAsb2fl/fl.Flnafl/+ and Nkx2- 5+/Cre.Asb2fl/fl.Flnafl/+ and Nkx2-5+/Cre.Asb2fl/fl.Flnafl/y often presented with pericardial edema at both stages. 5+/CreAsb2fl/+.Flnafl/y often presented with pericardial edema at both stages. 16 litters were analyzed at E9.5 and 16 litters were analyzed at E9.5 and 3 litters at E10.5. Scale bar is equivalent to 0.2mm at E9.5 and 0.4mm 3 litters at E10.5. Scale bar is equivalent to 0.2mm in E9.5 embryos and 0.4mm in E10.5 embryos as at E10.5 embryos as indicated. B. Immunofluorescence on Nkx2-5+/Cre.Asb2fl/fl.Flnafl/y double knockouts and indicated. B. Immunofluorescence on Nkx2-5+/Cre.Asb2fl/fl.Flnafl/fl double knockouts and controls using Flna controls using Flna (red) and Troponin T (green) antibodies. Note absence of Flna expression in the (red) and Troponin T (green) antibodies. Note absence of Flna expression in the myocardium of the double myocardium of the double knockouts as opposed to its expression in the myocardium of the single knockouts as opposed to its expression in the myocardium of the single knockouts in figure 3A. Scale bar is knockouts in figure 3A. Scale bar is equivalent to 75µm. C. Asb2 transcipt levels from RNAseq data showing equivalent to 75μm. C. Asb2 transcript levels from RNAseq data showing reduced Asb2 levels in the single reduced Asb2 levels in the single and double knockouts compared to the Asb2 heterozygote control. *: and double knockouts compared to the Asb2 heterozygote control. * p<0.05. One-way ANOVA was used for p<0.05. One-way ANOVA was used for analysis using Graphpad Prism. D. QPCR analysis of hearts from analysis using Graphpad prism. D. QPCR analysis of hearts from Nkx2-5+/Cre.Asb2fl/+, Nkx2-5+/Cre.Asb2fl/fl, Nkx2-5+/Cre.Asb2fl/+, Nkx2-5+/Cre.Asb2fl/fl, Nkx2-5+/Cre.Asb2fl/+.Flnafl/y, Nkx2-5+/Cre.Asb2fl/fl.Flnafl/y E9.5 mice. Nkx2-5+/Cre.Asb2fl/+.Flnafl/y, and Nkx2-5+/Cre.Asb2fl/fl.Flnafl/y E9.5 mice. n=5-6 for Nkx2-5+/Cre.Asb2fl/+; n=5 for N=5-6 for Nkx2-5+/Cre.Asb2fl/+; n=5 for Nkx2-5+/Cre.Asb2fl/fl; n=5-6 for Nkx2-5+/Cre.Asb2fl/+.Flnafl/y; n=3 for Nkx2- Nkx2-5+/Cre.Asb2fl/fl; n=5-6 for Nkx2-5+/Cre.Asb2fl/+.Flnafl/y; n=3 for Nkx2-5+/Cre.Asb2fl/fl.Flnafl/y (each sample was a 5+/Cre.Asb2fl/fl.Flnafl/y (each sample is a combination of 2-3 hears to account for littermate variability). The combination of 2-3 hearts to account for littermate variability). The selected genes are among those identified selected genes are among those identified in RNAseq analysis in fugures 2 and 3. P<0.05 is considered in RNAseq analysis in figures 2 and 3. p<0.05 is considered statistically significant. T-test was used for anly- statistically significant. T-test was used for analysis using Graphpad Prism. sis using Graphpad Prism. A. 1 e d u G i Variant 1 1 2 3a 5’UTR 6 e d u G i 4 5 6 7 8 9 10 3’UTR Variant 2 3b 4 5 6 7 8 9 10 B. 3.0 2.5 0 F F / 2.0 1.5 1.0 0 5’UTR WT KO 3’UTR 2 4 6 Time (s) 8 10 Supplementary Figure 3, related to figure 7. A. Schematic representation of the two Asb2 isoforms Supplementary Figure 3, related to figure 7. A. Schematic representation of the two ASB2 isoforms showing the location of the guides used for CRISPR/Cas9 genome editing. B. Representative calcium showing the location of the guides used for CRISPR/Cas9 genome editing. B. Representative calcium transients of hiPSC-CMs (WT: blue, KO: red). transients of hiPSC-CMs (WT: blue, KO: red).
null
10.7554_elife.85878.pdf
Data availability All data generated or analyzed in this study are included in the manuscript and supporting files. Source data files have been provided for Figure 1b, Figure 1c, Figure 1f, Figure 1g, Figure 1- figure supplement 2a- c, Figure 2b, Figure 2g, Figure 2- figure supplement 1a- f, Figure 3a, Figure 3c, Figure 3e, Figure 3h, Figure 3 supplement 1b- c, Figure 4a, Figure 4b, Figure 4e, Figure 4f, Figure 4h, Figure 4i, Figure 4- figure supplement 1a- b, Figure 4- figure supplement 2a, Figure 5b.
Data availability All data generated or analyzed in this study are included in the manuscript and supporting files. Source data files have been provided for Figure 1b , Figure 1c, Figure 1f, Figure 1g, Figure 1-figure supplement 2a-c, Figure 2b, Figure 2g, Figure 2-figure supplement 1a-f, Figure 3a, Figure 3c, Figure 3e, Figure 3h, Figure 3 supplement 1b-c, Figure 4a, Figure 4b, Figure 4e, Figure 4f, Figure 4h, Figure 4i, Figure 4-figure supplement 1a-b, Figure 4-figure supplement 2a, Figure 5b.
RESEARCH ARTICLE Two RNA- binding proteins mediate the sorting of miR223 from mitochondria into exosomes Liang Ma, Jasleen Singh, Randy Schekman* Department of Molecular and Cell Biology, Howard Hughes Medical Institute, University of California, Berkeley, United States Abstract Fusion of multivesicular bodies (MVBs) with the plasma membrane results in the secre- tion of intraluminal vesicles (ILVs), or exosomes. The sorting of one exosomal cargo RNA, miR223, is facilitated by the RNA- binding protein, YBX1 (Shurtleff et al., 2016). We found that miR223 specif- ically binds a ‘cold shock’ domain (CSD) of YBX1 through a 5’ proximal sequence motif UCAGU that may represent a binding site or structural feature required for sorting. Prior to sorting into exosomes, most of the cytoplasmic miR223 resides in mitochondria. An RNA- binding protein local- ized to the mitochondrial matrix, YBAP1, appears to serve as a negative regulator of miR223 enrich- ment into exosomes. miR223 levels decreased in the mitochondria and increased in exosomes after loss of YBAP1. We observed YBX1 shuttle between mitochondria and endosomes in live cells. YBX1 also partitions into P body granules in the cytoplasm (Liu et al., 2021). We propose a model in which miR223 and likely other miRNAs are stored in mitochondria and are then mobilized by YBX1 to cyto- plasmic phase condensate granules for capture into invaginations in the endosome that give rise to exosomes. Editor's evaluation This important study presents a novel mechanism of miRNA223 sorting into exosomes involving its storage within mitochondria, specifically by a mitochondrially localized protein YBAP1. The evidence supporting the findings is convincing and opens avenues for future studies on molecular mecha- nisms. This paper is a valuable addition to the cellular sorting of miRNA involving interplay with and between the organelles, interesting for miRNAs researchers, as well as cell biologists. Introduction Extracellular vesicles (EVs) bud from the plasma membrane or are secreted when multivesicular bodies (MVB) fuse with the plasma membrane to release a population of vesicles called exosomes. EVs and their cargos are highly dependent on their membrane source. Microvesicles released by budding from the plasma membrane are a heterogeneous population of EVs ranging in size from 30 nm to 1000 nm (Cocucci et al., 2009). Exosomes are 30 nm to 150 nm in size and originate as vesicles invaginated into the interior of an MVB to form intraluminal vesicles (ILVs; Harding et al., 1983). Many RNAs are selectively sorted into EVs, especially small RNAs. Several studies have indicated that RNA binding proteins (RNPs) may be involved in the enrichment of RNAs into EVs (Mukherjee et al., 2016; Santangelo et al., 2016; Teng et al., 2017; Villarroya- Beltri et al., 2013). However, many of these studies used sedimentation at  ~100,000  g to collect EVs, which may also collect RNP particles not enclosed within membranes which complicates the interpretation of these data. To address this question, we previously developed buoyant density- based methods to separate EVs *For correspondence: schekman@berkeley.edu Competing interest: The authors declare that no competing interests exist. Funding: See page 20 Received: 30 December 2022 Preprinted: 11 January 2023 Accepted: 24 July 2023 Published: 25 July 2023 Reviewing Editor: Agnieszka Chacinska, IMol Polish Academy of Sciences, Poland Copyright Ma et al. This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited. Ma et al. eLife 2023;12:e85878. DOI: https://doi.org/10.7554/eLife.85878 1 of 23 Research article from non- vesicular aggregates and found that EVs form two distinct populations of high and low buoyant density (Shurtleff et al., 2016; Temoche- Diaz et al., 2020). We found that some miRNAs are selectively enriched in a high buoyant density vesicle fraction characterized by an enrichment in the exosomal marker protein CD63, whereas the low buoyant density EVs are fairly non- selective in the capture of miRNAs (Temoche- Diaz et  al., 2019). We developed a cell- free reaction to identify YBX1 as required for miR223 sorting into exosomes and demonstrated that it plays an important role in the enrichment of miR223 into exosomes in HEK293T cells (Shurtleff et al., 2016). We subse- quently found that phase separated YBX1 condensates selectively recruit miR223 in vitro and sort it into exosomes in cells (Liu et al., 2021). In this study, we report that YBX1 directly and specifically binds miR223 by its ‘cold shock’ domain (CSD). We have identified a sequence motif, UCAGU, that facilitates the sorting of miR223 into exosomes. We also found a significant fraction of cytoplasmic miR223 localized within mitochondria, tightly associated with the mitochondrial envelope and that a mitochondrial RNA- binding protein, YBAP1, may control the transfer of miR223 from mitochondria to exosomes. Results YBX1 directly and specifically binds miR223 We previously documented that YBX1 facilitates miR223 sorting into exosomes (Shurtleff et  al., 2016) and that exosomal miR223 is decreased in YBX1 knockout cells (Liu et al., 2021). We reexam- ined the enrichment and confirmed that exosomal miR223 was decreased in exosomes purified from YBX1 KO cells (Figure 1a). We used a Nanosight particle tracking device to quantify buoyant density purified vesicles and found that knockout of YBX1 did not affect exosome secretion (Figure 1—figure supplement 1). Whereas the importance of YBX1 for miR223 sorting has been established, the mechanism of their interaction was not known. To examine the direct interaction of YBX1 and miR223, we used an electrophoretic mobility shift assay (EMSA) with purified recombinant YBX1, expressed in insect cells (Figure 1—figure supplement 2a), and chemically synthetic miR223 and miR190, a cytoplasmic miRNA that is not enriched in exosomes. Purified YBX1 was titrated and incubated with 5’ fluorescently labeled miR223 at 30 ℃ for 30 min. miR223- YBX1 complexes were separated by electrophoresis and detected by in- gel fluorescence. The EMSA data showed that YBX1 directly and specifically bound to miR223, but ~140 fold less well with miR190 (Figure 1b–c). The measured Kd for YBX1:miR223 was 4.2 nM (Figure 1d). YBX1 has three major domains including an N- terminal alanine/proline- rich (A/P) domain, a central cold shock domain (CSD) and a C- terminal domain (CTD) (Figure 1e). To explore which specific domain of YBX1 binds miR223, we constructed a series of fragments: the A/P domain, CSD and CTD. The YBX1 fragments were expressed in and purified from insect cells (Figure 1—figure supplement 2b). EMSA data showed that the A/P domain and CTD had little or no affinity for miR223, whereas the CSD domain bound miR223 but with an affinity much reduced compared to full length YBX1 (Figure 1f). We then constructed two combined fragments of the A/P and CSD and CSD and CTD domains (Figure 1—figure supplement 2c). The EMSA data showed that the A/P domain was dispensable, whereas binding of miR223 to CSD plus CTD was comparable to full- length YBX1 (Figure 1g). YBX1- F85A in the CSD domain was reported to block the YBX1- specific binding of mRNA (Lyons et al., 2016). Purified YBX1- F85A protein failed to bind miR223 (Figure 1g, Figure 1—figure supple- ment 2c). These data suggest that YBX1 directly and specifically binds miR223 via the CSD. The CTD of YBX1 did not appear to bind miR223 but may somehow facilitate a higher affinity interaction of the CSD with miR223. A binding or structural motif on miR223 that promotes interaction with YBX1 and enrichment into exosomes We next sought to determine the miR223 sequence motif responsible for interaction with YBX1 and enrichment into exosomes. We used an EMSA competition assay with a series of miR223 mutants. Purified YBX1 and 5’ fluorescently labeled miR223 were incubated with miR223 mutant constructs titrated in a range from 1 nM to 1 μM. miR223 variants in a binding domain should not compete for interaction of YBX1 with 5’ fluorescently tagged miR223 whereas variations in sequences irrelevant to Ma et al. eLife 2023;12:e85878. DOI: https://doi.org/10.7554/eLife.85878 2 of 23 Cell Biology Research article a. e g n a h c d o F l / ) T W O K 1 X B Y ( 8 4 2 1 0.5 0.25 d. miR223-3p miR190a-5p Cell Exosome 100 80 60 40 20 ) % ( d n u o B n o i t c a r F 0 10 0 M μ 1 [miRNA] = 1nM Kd of miR223 = 4.2nM Kd of miR190 = 574.5nM 10 1 10 2 YBX1 [ nM ] 10 3 c. M μ 0 YBX1 M μ 1 Bound miR190 Free miR190 miR223 miR190 b. M μ 0 YBX1 Bound miR223 Free miR223 YBX1 N 1 51 129 A/P CSD CTD 324 C M μ μ 0 0 YBX1(1-51) M μ 1 M μ μ 0 0 YBX1(52-129) M μ 1 YBX1(130-324) M μ μ 0 0 M μ 1 1 51 A/P 52 129 CSD 130 324 CTD YBX1(1-129) M μ μ 0 0 M μ 1 M μ μ 0 0 YBX1(52-324) M μ 1 M μ μ 0 0 YBX1(F85A) M μ 1 e. f. Bound miR223 Free miR223 g. Bound miR223 Free miR223 1 129 A/P CSD 52 324 1 51 324 CSD CTD A/P CSD CTD F85A Figure 1. YBX1 directly and specifically binds miR223. (a) RT- qPCR analysis of fold change of miR- 223 and miR- 190 in cells and purified exosomes from 293 T WT cells and YBX1 knockout cells. Data are plotted from three independent experiments, each independent experiment with triplicate qPCR reactions; error bars represent standard deviations. (b–c) EMSA assays using 1 nM 5’ fluorescently labeled miR223 or miR190 and purified YBX1. Figure 1 continued on next page Ma et al. eLife 2023;12:e85878. DOI: https://doi.org/10.7554/eLife.85878 3 of 23 Cell Biology Research article Figure 1 continued Purified YBX1 was titrated from 500pM to 1 μM. In gel fluorescence was detected. Quantification of (d) shows the calculated Kd. (e) Schematic diagrams of the different domains of YBX1. (f) EMSA assay using 1 nM 5’ fluorescently labeled miR223 and purified YBX1 truncations. [YBX1(1–51) or YBX1(52–129) or YBX1(130–324).] (g) EMSA assay using 1 nM 5’ fluorescently labeled miR223 and purified YBX1 truncations [YBX1(1–129) or YBX1(52–324)] or YBX1(F85A) mutant. The online version of this article includes the following source data and figure supplement(s) for figure 1: Source data 1. Uncropped gel images corresponding to Figure 1. Figure supplement 1. Knockout of YBX1 did not change exosome secretion. Figure supplement 2. Purified YBX1 full length protein and different truncations and mutation. Figure supplement 2—source data 1. Uncropped gel images corresponding to Figure 1—figure supplement 2. interaction would compete. Using this EMSA competition assay to screen the miR223 binding motif, we found that the competitive binding of miR223mut (3- 6) and miR223mut (4- 7) were decreased (Figure  2—figure supplement 1). This suggested that the sequence UCAGU was critical for inter- action with YBX1. To test this directly, we employed a variant sequence, termed miR223mut, where the UCAGU was substituted with AGACA. As a positive control, we employed a variant of miR190, miR190sort, where the sequence AUAUG was substituted with UCAGU (Figure  2a). EMSA data showed a~27- fold reduced YBX1 interaction of with miR223mut, whereas the affinity of miR190sort with YBX1 was increased ~eightfold compared to wt sequences. To test whether this motif is critical for miR223 enrichment into exosomes, we purified exosomes from 293T cells transiently transfected to overexpress one of the four miRNA constructs (Figure 2d). RT- qPCR data showed that the level of miR223 in exosomes was ~fourfold dependent on the puta- tive exosomal sorting motif (Figure  3e) and the enrichment of miR190sort into exosomes was increased ~fivefold compared to miR190 WT (Figure 2f). In previous work, we developed a cell- free reaction to test the biochemical requirement for YBX1 in the sorting of miR223 into vesicles formed with membranes and cytosol isolated from broken HEK293 cells (Shurtleff et  al., 2016). In this work, we showed that the sorting of miR223 and of a CD63- luciferase fusion protein into an enclosed membrane were coincidentally inhibited by GW4869 an inhibitor of neutral sphingomyelinase (NS2) known to interfere with exosome biogenesis and secre- tion. On this basis, we concluded that the cell- free reaction recapitulated the sorting event leading to the packaging of miR223 into exosomes. We refined this assay to measure the incorporation of 32P- 5’ end- labeled wt and mutant miR223 into vesicles formed in vitro. Isolated membranes and cytosol were incubated with 32P- labled wt or mutant miR223 at 30 °C for 20 min, after which RNase I was added to digest any unpackaged miRNA. Controls including 1% Triton X- 100 were used to measure background RNase resistant radiolabel. Samples were resolved on a gel for visual and quantitative evaluation of membrane sequestered RNA (Figure 2g and h). The results suggested that the UCAGU motif is critical for miR223 packaging into vesicles in the cell- free reaction. Taken together the results in Figure  2 show that the miR223 sequence UCAGA promotes the binding of YBX1 in order to sort the miRNA into vesicles formed in cells and in a cell- free reaction. We suggest this sorting facilitates the export of miR223 in exosomes secreted from HEK293 cells. Mitochondria contribute to miR223 enrichment into exosomes In a recent study, we showed that YBX1 is sorted into P- bodies in cells and that these biomolecular condensates may initiate the sorting of miR223 into vesicles budding into the interior of endosomes (Liu et al., 2021; Shurtleff et al., 2016). Mitochondria represent another apparent intracellular loca- tion of miR223 (Wang et al., 2020). We used cell fractionation of homogenates of HEK293 cells to evaluate the subcellular distribution of endogenous miR223. Fractionation was evaluated by immu- noblot using marker proteins characteristic of various cell organelles (Figure  3a). Analysis of RNA extracted from isolated membranous organelles confirmed that miR223 but not miR190 was signifi- cantly enriched in mitochondria but not in ER or cytosol (Figure 3b, Figure 3—figure supplement 1a; Wang et al., 2020). Ma et al. eLife 2023;12:e85878. DOI: https://doi.org/10.7554/eLife.85878 4 of 23 Cell Biology Research article a. miR223-3p UGUCAGUUUGUCAAAUACCCCA miR223mut UGAGACAUUGUCAAAUACCCCA miR190-5p UGAUAUGUUUGAUAUAUUAGGU miR190sort UGUCAGUUUUGAUAUAUUAGGU c. 100 ) % ( d n u o B n o i t c a r F 75 50 25 b. M μ 0 YBX1 M μ 1 M μ 0 YBX1 0 0.1 M μ 1 1 M μ 0 10 100 YBX1 [nM] YBX1 miR223 miR190sort miR190a miR223mut [miRNA] = 1nM Kd of miR223mut = 112.4nM Kd of miR190sort = 77.5nM 1000 10000 M μ 1 M μ 0 YBX1 M μ 1 d. miR223 Transfect plasmid which express miRNAs or mutants miR223mut miR190 miR190sort medium 1,500 g Supernatant100,000 g 10% 40% 150,000 g Exosomes 10,000 g 120,000 g 60% Sucrose cushion EV 60% e. f. p=0.0013 1.00 0.27 1.2 1.0 0.8 0.6 0.4 0.2 0.0 g. Temp (℃) RNase I Membranes Cytosol Triton(1%) miR223 3030 30 30 4 4 _ + + + + + _ + + + + + + _ + + + + _ _ _ _ + _ miR223mut 30 + + _ + + _ _ _ _ 3030 304 + + + + + + + + + _ + 4 _ + + _ l e g n a h C d o F s A N R m i l a m o s o x E ) l l e c o t e v i t a e r ( l miR223 miR223mut p=0.0194 5.33 h. 1.00 miR190 miR190sort 8 6 4 2 0 i 3 2 2 R m d e t c e t o r p % 25 20 15 10 5 0 C M ) l l e c o t e v i t a e r ( l l e g n a h C d o F s A N R m i l a m o s o x E miR223 miR223mut degree) C+M+Triton C+M C+M(4 Figure 2. miR223 sequence motif UCAGU binds YBX1. (a) RNA oligonucleotides corresponding to miR223, miR190 and versions with mutated sorting motif (miR223mut) or mutation to introduce the sorting motif (miR190sort). (b) EMSA assays using 1 nM 5’ fluorescently labeled miR223 WT or miR223mut or miR190 WT or miR190sort and purified YBX1. Purified YBX1 was titrated from 500pM to 1 µM. In gel fluorescence was detected. (c) Binding affinity curves as calculated by EMSA data from (b) (d) Schematic shows exosome purification with buoyant density flotation in a sucrose Figure 2 continued on next page Ma et al. eLife 2023;12:e85878. DOI: https://doi.org/10.7554/eLife.85878 5 of 23 Cell Biology Research article Figure 2 continued step gradient from 293T cells overexpressing miR223 WT or mutant or miR190 WT or miR190sort. (e) RT- qPCR analysis of relative abundance of miR223 or miR223mut detected in exosomes compared to cellular level in 293T cells overexpressing miR223 WT or miR223mut. Data are plotted from three independent experiments and error bars represent standard deviations. (f) RT- qPCR analysis of relative abundance of miR190 or miR190sort detected in exosomes compared to cellular level in 293T cells overexpressing miR190 WT or miR190sort. Data are plotted from three independent experiments and error bars represent standard derivations. (g) In vitro packaging assay using 32P 5’end- labeled miR223 and miR223mut. Cell- free packaging of miR223 and miR223mut measured as protected radioactive signal from 32P labeled miR223 and miR223mut. Reactions with or without membrane, cytosol, and 1% Triton X- 100, and incubated at 4 or 30 °C are indicated. For the samples containing only cytosol plus membrane at 4 °C, only one- third of the samples were loaded. Each sample was supplemented with 300 mM urea to reduce the background signal. (h) Data quantification showed protected fraction of miR223 and miR223mut as calculated from in vitro packaging data shown in (g). The online version of this article includes the following source data and figure supplement(s) for figure 2: Source data 1. Uncropped gel images corresponding to Figure 2. Figure supplement 1. Screening of exosomal sorting motif of miR223. Figure supplement 1—source data 1. Uncropped gel images corresponding to Figure 2—figure supplement 1. To determine the localization of miR223 on or within mitochondria, we prepared mitoplasts using digitonin to strip away the mitochondrial outer membrane followed by fractionation on a Percoll density gradient. Immunoblots of the enriched mitochondria and isolated mitoplasts showed that the outer membrane, marked by Tom20, was largely removed with retention of the inner membrane marker Tim23 (Figure 3c). RNA was extracted from the purified mitoplast and RT- qPCR data indicated that miR223 was enriched along with mRNA for COX1, but not with nuclear U6 snRNA (Figure 3d). As an independent means to assess the localization of cytoplasmic miR223, we used immunopre- cipitation to purify mitochondria. Isolated mitochondria were then converted to mitoplasts by osmotic shock and treated with proteinase K and RNase. Immunoblots of the immunoprecipitated mitochon- dria and isolated mitoplasts showed that the outer membrane, marked by Tom20, and intermembrane space, marked by AIF, were largely removed with retention of the mitochondrial matrix marker citrate synthase (Figure 3e). RNA was extracted from the immunoprecipitated mitochondria and mitoplasts and RT- qPCR data indicated that miR223 was enriched along with mRNA for COX1, but not with miR190 or nuclear U6 snRNA (Figure 3f). We also used immunoprecipitated mitochondria (Figure  3—figure supplement 1b) and either Triton X- 100 to solubilize the membrane or freeze- thaw to allow the matrix and envelope fractions to be separated by centrifugation. Mitochondrial membrane proteins, such as Tom20 and COX IV, were solubilized and retained in the supernatant fraction (Figure  3—figure supplement 1c). The freeze- thaw regimen released citrate synthase to a supernatant fraction whereas Tom20 and COX IV sedimented in the pellet fraction. RNA was extracted from the detergent supernatant and pellet fractions where we found similar distributions of COX1 and miR223, neither of which were as readily solubilized as the inner and outer membrane proteins (Figure 3—figure supplement 1d). RT- qPCR quantification of fractions from the freeze- thaw regimen showed that both COX1 mRNA and miR223 remained largely associated with the sedimentable membrane fraction (Figure 3—figure supplement 1e). We conclude that miR223 is enclosed within mitochondria, possibly in association with the inner membrane. We sought a test of the role of mitochondria in the secretion of miR223 in exosomes. For this purpose, we generated cells depleted of mitochondria (Correia- Melo et  al., 2017). U- 2 OS cells expressing GFP- parkin were treated with CCCP for 48 hr, conditions that cause mitochondria to be removed by mitophagy. We confirmed mitochondrial depletion after CCCP treatment by RT- qPCR of mitochondrial COX1 mRNA (Figure  3g) and immunoblot of the mitochondrial inner membrane marker Tim23 (Figure 3h). We then compared the levels of both miR223 and miR190 from GFP- parkin expressing U- 2 OS cells with and without CCCP treatment. RT- qPCR data showed that miR223, but not miR190, increased threefold in cells treated with CCCP (Figure  3j). To test the possibility that miR223 accumulated in cells as a result of a failure of mobilization into exosomes, we compared the miR223 levels in exosomes purified from untreated and CCCP treated cells (Figure  3j). Although exosome secretion, as measured with a CD63- luciferase marker, did not change after CCCP treatment (Figure 3—figure supplement 2a), we found that CCCP treatment lowered the amount of miR223 in EVs fourfold (Figure 3j). The increase in cellular at the expense of exosomal miR223 may reflect a critical role for mitochondria in the mobilization of this RNA to exosomes. Ma et al. eLife 2023;12:e85878. DOI: https://doi.org/10.7554/eLife.85878 6 of 23 Cell Biology c. d. l l e C T M P M ⍺-Tim23 ⍺-Tom20 ⍺-Tubulin ⍺-PDI t n e m h c i r n E A N R 10 8 6 4 2 0 COX1 miR223 U6 snRNA cell mitoplast cell mitoplast cell mitoplast COX1 miR223 miR190 U6 snRNA IP-MT Cell IP-MP IP-MT Cell IP-MP IP-MT IP-MP Cell IP-MT Cell IP-MP + CCCP Exosome Free Medium + Uridine CCCP removed exosome Cell Exosome Research article a. l l e C o t y C o t i M R E b. t n e m h c i r n E 3 2 2 R m i ⍺-Tim23 ⍺-Calnexin ⍺-Tubulin ) l l e c o t e v i t a e r ( l 40 30 20 10 0 Cyto ER Mito * f. t n e m h c i r n E A N R 20 15 10 5 0 ⍺-Citrate Synthase ⍺-LAMP1 ⍺-Tom20 ⍺-AIF ⍺-GRP78 ⍺-GAPDH h. P=0.0004 i. ⍺-Tim23 ⍺-GAPDH Parkin-GFP e. g. l l e c n i A N R m 1 X O C f o % 120 100 80 60 40 20 0 NC 24uM CCCP j. e g n a h c d o F l ) / C N P C C C ( 8 4 2 1 0.5 0.25 0.125 miR223 miR190 Figure 3. Mitochondria contribute to miR223 enrichment into exosomes. (a) Immunoblot analysis of protein markers for different subcellular fractions isolated from 293T cells. (b) RT- qPCR analysis of miR223 fold changes of different subcellular fractions isolated from 293T cells relative to cell lysate. (c) Immunoblot analysis of protein markers for mitoplasts purified from 293T cells by Percoll gradient fractionation (MT: mitochondria; MP: mitoplast). (d) RT- analysis of COX1 mRNA, miR223 and U6 snRNA fold changes for mitoplasts purified from 293T cells relative to cell lysate. (e) Immunoblot analysis Figure 3 continued on next page Ma et al. eLife 2023;12:e85878. DOI: https://doi.org/10.7554/eLife.85878 7 of 23 Cell Biology Research article Figure 3 continued of protein markers for immunoprecipitated mitochondria and osmotic shock generated mitoplasts. Mitochondria were purified from a 293T 3xHA- EGFP- OMP25 overexpressing cell line using anti- HA magnetic beads. Mitoplasts were purified following mitochondrial immunoprecipitation by osmotic shock, proteinase K and RNase treatment (IP- MT: immunoprecipitated mitochondria; IP- MP: immunoprecipitated mitoplasts). (f) RT- analysis of COX1 mRNA, miR223, miR190 and U6 snRNA fold changes for immunoprecipitaed mitochondria and mitoplasts purified from the 293T 3xHA- EGFP- OMP25 overexpressing cell line. Data are plotted from three independent experiments and error bars represent standard deviations. (g) RT- qPCR analysis of mitochondrial mRNA COX1 in U2OS cells expressing GFP- Parkin treated with or without CCCP. Data are plotted from three independent experiments and error bars represent standard deviations. (h) Immunoblot analysis of mitochondrial marker Tim23 in U2OS cells expressing GFP- Parkin treated with or without CCCP. (i) Schematic of exosome purification from mitochondria depleted GFP- Parkin expressing U2OS cells. (j) RT- qPCR analysis of fold change of miR- 223 and miR- 190 in cells and purified exosomes from U2OS cells expressing GFP- Parkin which were treated with or without CCCP. Data are plotted from three independent experiments and error bars represent standard deviations. The online version of this article includes the following source data and figure supplement(s) for figure 3: Source data 1. Uncropped immunoblot images corresponding to Figure 3. Figure supplement 1. miR223, but not miR190, enriched in mitochondria. Figure supplement 1—source data 1. Uncropped immunoblot images corresponding to Figure 3—figure supplement 1. Figure supplement 2. Mitochondrial depletion did not change exosome secretion. YBAP1 binds miR223 in the mitochondria and in vitro In the course of purifying a tagged version of YBX1 from 293T cells, we observed another protein that copurified and found that it corresponded to YBAP1 (Figure 4a and b). Such a complex of YBX1 and YBAP1 has previously been reported (Matsumoto et al., 2005). We confirmed that purified YBX1 and YBAP1 bind each other by coexpression and affinity purification from insect cells (Figure 4—figure supplement 1b). YBAP1 is a mitochondrial matrix protein with a standard N- terminal transit peptide sequence (Muta et al., 1997). We confirmed this mitochondrial localization in U- 2 OS cells expressing Tom22- mCherry transiently transfected with a YBAP1- GFP construct (Figure 4c–d). We also showed that YBAP1 is localized within mitochondria by performing a proteinase K protection assay on purified mitochondria. Mitochondria were isolated from non- transfected cells and exposed to proteinase K in the presence or absence of Triton X- 100 and the degradation of YBAP1 was evaluated by immuno- blot. YBAP1 was resistant to proteinase K digestion as was the mitochondrial inner membrane marker Tim23. Both were degraded by proteinase treatment in the presence of Triton X- 100 (Figure  4e), consistent with the localization of YBAP1 within mitochondria. To test whether YBAP1 was bound to miR223 in mitochondria, we used YBAP1 immunoprecipita- tion with mitochondria purified by fractionation on a Percoll density gradient (Figure 4f). RT qPCR data showed that mitochondrial miR223 was immunoprecipitated by YBAP1 antibody but not by a control antibody (Figure 4g). To determine whether the YBAP1 and miR223 interaction was direct, we used the EMSA assay and found that purified YBAP1 bound miR223, but not miR190. The YBAP1 interaction with miR223 was not dependent on the RNA sequence motif responsible for YBX1 binding (Figure 4—figure supplement 2a–b). Taken together, these data suggest that YBAP1 binds miR223 in mitochondria and in vitro. YBAP1 may control the transit of miR223 from mitochondria to exosomes To investigate the function of YBAP1 in the transit of miR223 into exosomes, we generated a 293T YBAP1 KO cell line and compared the level of miR223 enrichment in exosomes and mitochondria isolated from WT and mutant cells (Figure 5a and b). Although knockout of YBAP1 did not change exosome secretion (Figure  5—figure supplement 1a), RT- qPCR analysis showed that miR223 decreased twofold in mitochondria but increased eightfold in exosomes purified from mutant and WT cells, respectively (Figure 5c). This apparent inverse relationship is consistent with a role for YBAP1 protein in the retention of miR223 in mitochondria. YBX1 puncta shuttled from mitochondria to endosomes In previous work we reported the localization of YBX1 to P- bodies and suggested this may repre- sent an intermediate stage in the concentrative sorting of miRNAs for secretion in exosomes (Liu et al., 2021). In other earlier work, P- bodies were seen in association with mitochondria (Huang et al., Ma et al. eLife 2023;12:e85878. DOI: https://doi.org/10.7554/eLife.85878 8 of 23 Cell Biology Research article a. b. d. e u a v l YBX1 Copurified Band ⍺-YBAP1 c. y a r G d e z i l a m r o N 1.5 1.0 0.5 0.0 0 Tom22 YBAP1 100 50 Distance (pixels) 150 Tom22-mC e. e. Mitochondria Protease K Triton X-100 + - - + + - + + + 10um YBAP1-GFP f. Input IP ⍺-Tim23 ⍺-Tom20 ⍺-YBAP1 ⍺-YBAP1 ⍺-Tom20 h. Bound miR223 Free miR223 i. M μ μ 0 0 YBAP1 M μ 1 M μ μ 0 0 YBAP1 M μ 1 Bound miR190 Free miR190 merge g. 3 2 2 R m i f o t u p n I % 60 40 20 0 j. 100 ) % ( d n o B n o i t c a r F 75 50 25 P<0.0001 mIgGIP YBAP1IP miR223-3p miR190a-5p 0 0.1 1 10 100 YBAP1 [nM] 1000 10000 [miRNA] = 1nM Kd of miR223 = 173.2nM Figure 4. YBAP1 directly and specifically binds miR223. (a) Strep II- YBX1 was overexpressed in HEK293T cells. Coomassie blue detection of unknown band copurified with YBX1 from 293T cells. (b) Immunoblot identified unknown band was YBAP1. (c) Tom22- mCherry expressing U2OS was transfected with a YBAP1- GFP- expressing plasmid, cultured for 12 hr and observed by confocal microscopy. Scale bar, 10 μm. (d) Quantification of the fluorescence intensity of the different channels indicated by the solid white line of (c). (e) YBAP1 resides in mitochondria. Proteinase K protection assay for YBAP1 Figure 4 continued on next page Ma et al. eLife 2023;12:e85878. DOI: https://doi.org/10.7554/eLife.85878 9 of 23 Cell Biology Research article Figure 4 continued using purified mitochondria from 293T cells. Samples were treated with or without proteinase K (10 μg/ml) and or Triton X- 100 (0.5%). Immunoblots for Tim23, Tom20, and YBAP1 are shown. (f) Mitochondria were purified for immunoprecipitation with YBAP1 antibody. Immunoblot detection of YBAP1 and Tom20. (g) RT- qPCR analysis of miR223 fold changes of YBAP1 IP samples. Data are plotted from three independent experiments and error bars represent standard deviations. (h–i) EMSA assays using 1 nM 5’ fluorescently labeled miR223 or miR190. Purified YBAP1 was titrated from 500pM to 1 μM. In gel fluorescence was detected. Quantification of (j) shown the calculated Kd. The online version of this article includes the following source data and figure supplement(s) for figure 4: Source data 1. Uncropped immunoblot and gel images corresponding to Figure 4. Figure supplement 1. YBX1 and YBAP1 copurify as a complex from transfected SF9 cells. Figure supplement 1—source data 1. Uncropped gel images corresponding to Figure 4—figure supplement 1. Figure supplement 2. YBAP1 does not share the same miR223- binding motif as YBX1. Figure supplement 2—source data 1. Uncropped gel images corresponding to Figure 4—figure supplement 2. 2011). To explore this possibility, we visualized endogenous YBX1 and YBAP1 by IF and observed YBX1 puncta colocalized with mitochondria (Figure  6a and b). In order to detect the proximity of endosomes to this point of contact between YBX1 puncta and mitochondria, we used U- 2 OS cells transfected with Rab5(Q79L)- mCherry, which we employed previously to enlarge and detect the inter- nalization of YBX1 into endosomes (Liu et al., 2021). We then used three color visualization of the U- 2 OS cells also transfected with YFP- YBX1 and mito- BFP. Time- lapse imaging showed YBX1 puncta in close proximity to mitochondria or endosomes, followed quickly by transfer between them (Figure 6c and d). Taken together, these data suggest a mechanism whereby miR223 stored in mitochondria, possibly sequestered by YBAP1, may be captured in a tighter interaction with YBX1 in P- bodies and delivered to endosomes for sorting and secretion in exosomes. Discussion Selected miRNAs are sorted, some with very high fidelity, into invaginations in the endosome that give rise to exosomes secreted from cultured human cells and likely from many if not all cells in metazoan organisms. The means by which these miRNAs are sorted and the possible extracellular functions they serve is a subject of interest in normal and disease physiology. Here we report the role of the RNA- binding protein YBX1 and a sorting or structural signal on one target RNA, miR223, and the indirect path miR223 may take from storage in mitochondria into exosomes. We have identified a sequence motif on miR223, UCAGU, responsible for high- affinity interaction with YBX and for sorting into vesicles formed in a cell- free reaction as well as for secretion in exosomes by HEK293 cells. Previously we performed this in vitro packaging assay in the presence of an inhibitor (GW4869) of neutral sphingomyelinase (NS2). This inhibitor has been shown to reduce the secretion of exosomes and exosome- associated miRNAs in other studies (Li et al., 2013; Trajkovic et al., 2008; Yuyama et  al., 2012). In our cell- free assay, GW4869 inhibited the protection of CD63- luciferase and miR- 223 at concentrations known to inhibit the activity of NS2 in partially purified enzyme frac- tions (Shurtleff et al., 2016). We concluded that our cell- free reaction provides a model that mimics aspects of exosome biogenesis. The YBX1 protein has three distinct domains, one of which, the cold- shock domain (CSD) appears to be the principal site for RNA binding, including at least one critical residue, F85, required for binding miR223 as well as other RNAs (Lyons et al., 2016). The C- terminal domain (CTD) includes an intrinsically disordered domain (IDR) that promotes the formation of a liquid- liquid phase separation likely responsible for the organization of YBX1 in P- bodies (Liu et al., 2021). This domain does not itself interact with RNA, but it appears to facilitate the folding or stabilization of the CSD to promote high affinity binding to miR223. In other work using a similar approach, we identified two separate sorting signals, a 5’UGGA and a 3’UUU, on miR122 to which the RNA- binding protein La binds en route to secretion in exosomes by the breast cancer cell line MDA- MB- 231 (Temoche- Diaz et al., 2019). Other distinct sorting signals and their cognate RNA- binding proteins have been documented in different cell lines. miRNAs with a GGAG sorting motif recognized by a sumolyated form of hnRNPA2B1 was shown to be enriched in exosomes (Villarroya- Beltri et al., 2013). Another sequence, AAUGC, was found to be enriched in Ma et al. eLife 2023;12:e85878. DOI: https://doi.org/10.7554/eLife.85878 10 of 23 Cell Biology Research article a. Mitochondria Cell medium RNA Extraction and RT-qPCR WT or YBAP1-KO Exosome b. c. e g n a h c d o F l / ) T W O K 1 P A B Y ( 16 8 4 2 1 0.5 0.25 ⍺-YBAP1 ⍺-Citrate Synthase ⍺-COX IV ⍺-Tim23 ⍺-AIF ⍺-Tom20 ⍺-YBX1 ⍺-Tubulin Cell Exosome mitochondria miR223 miR190 Figure 5. YBAP1 sequesters miR223 which is released and secreted in YBAP1 KO cells. (a) Schematic shows exosome and mitochondria purification from 293 T WT cells and YBAP1 knock out cells for RT- qPCR analysis. (b) Analysis of 293 T WT and CRISPR/Cas9 genome- edited cells by immunoblot for YBAP1, YBX1 and mitochondrial markers (c) RT- qPCR analysis of miR223 enrichment in mitochondria purified from 293 T WT cells and YBAP1 KO cells relative to cell lysate. Data are plotted from three independent experiments and error bars represent standard deviations. (d) RT- qPCR analysis of miR223 and miR190 fold change in cells, purified mitochondria and exosomes from 293 T WT cells and YBAP1 KO cells. Data are plotted from three independent experiments and error bars represent standard deviations. The online version of this article includes the following source data and figure supplement(s) for figure 5: Source data 1. Uncropped immunoblot images corresponding to Figure 5. Figure supplement 1. Knockout of YBAP1 did not change exosome secretion. exosomal miRNA and dependent on the RNA- binding protein FMR1 for miRNA secretion (Wozniak et al., 2020). Diverse cell lines and likely tissues appear to invoke distinct sorting signals decoded by different RNA- binding proteins. Many of the proteins may engage in biomolecular condensates such as P- bodies as a mechanism to sort RNAs for secretion (Liu et al., 2021). Ma et al. eLife 2023;12:e85878. DOI: https://doi.org/10.7554/eLife.85878 11 of 23 Cell Biology Research article a. b. d e h c a t t a a t c n u p 1 X B Y % a i r d n o h c o t i m e h t n o 80 60 40 20 0 d. s t n e v e e l t t u h s a t c n u p 1 X B Y l l e c r e p 40 30 20 10 0 YBX1 YBAP1 5μm c. YFP-YBX1 Rab5(Q79L)-mC mito-BFP 0s 3m48s 6m39s 2μm 9m30s 10m27s 28m28s Figure 6. YBX1 puncta relocalize from mitochondria to endosomes. (a) YBX1 puncta on the mitochondria. U2OS cells were stained with anti- YBX1 and anti- YBAP1 antibodies and observed by confocal microscopy. The right panel shows enlarged regions of interest from the left panel. Scale bar, 5 μm. (b) The statistics are of the percentage of YBX1 puncta detected in proximity to mitochondria. N=30 cells. (c) YBX1 puncta relocalize from mitochondria to endosomes. U2OS cells overexpressed YFP- YBX1, Rab5(Q79L)- mCherry and mito- BFP. Time- lapse images were acquired with a Zeiss LSM900 confocal microscope. Scale bar, 2 μm. (d) The statistics are of YBX1 puncta shuttle events per cell. The data was represented as violin plots. N=34 cells. Ma et al. eLife 2023;12:e85878. DOI: https://doi.org/10.7554/eLife.85878 12 of 23 Cell Biology Research article miR223 appears to be an example of a number of small nuclear- encoded RNAs localized to mito- chondria (Jeandard et al., 2019). In some cases these RNAs, such as tRNAs, serve an essential func- tion such as in mitochondrial protein synthesis, however, for miRNAs with no obvious mitochondrial genome target, the function remains unclear (Jeandard et  al., 2019). Nonetheless, others have documented the localization of these miRNAs enclosed within the mitochondrion and in the case of miR223, it appears to be tightly associated with the inner membrane. The exact organization and function of miR223 in this location remains to be investigated but in the context of exosomal secre- tion, the mitochondrial localization appears to serve as a reservoir. In relation to the mitochondrial localization of miR223, we found a mitochondrial RNA- binding protein, YBAP1, that copurified with a tagged form of YBX1 expressed in HEK293 cells. YBAP1 has previously been reported to interact with YBX1 and independently found associated with mitochon- dria where its localization is dependent on an N- terminal transit peptide sequence (Muta et  al., 1997). The association of mitochondrial YBAP1 and cytoplasmic YBX1 was reproduced by coexpres- sion of recombinant forms of the two proteins in baculovirus- infected SF9 cells (Figure  4—figure supplement 1b). YBAP1 binds miR223 selectively but with an affinity significantly below that of YBX1 (Figure 4h–j). Although YBX1 does not localize to the mitochondrion, the stable interaction of the complex may suggest a transient relationship, perhaps during the biogenesis of YBAP1 as it transits from the cytoplasm into the mitochondrion. A functional relationship between YBAP1 and YBX1 is suggested by the reduction in miR223 in mito- chondria and increase in secretion of miR223 in exosomes secreted from YBAP1 KO cells (Figure 5). In contrast, removal of mitochondria by treatment of cells with CCCP resulted in an increase in cyto- plasmic miR223 at the expense of secretion in exosomes (Figure 3). Although YBAP1 may facilitate the retention of miR223 within mitochondria, mitochondrial RNA import and export may serve an MVB Exosomes YBX1 YBAP1 miR223 Figure 7. Diagram representing a model of miR223 sorting from mitochondria into exosomes. Stages in the transfer of miR223 from mitochondria. Cytosolic miR223 is enriched in mitochondria where it may be sequestered by a weak interaction with YBAP1. Cytoplasmic YBX1 interacts more tightly with miR223 which may drive the removal of miR223 from mitochondria. YBX1 in RNA granules may accumulate miR223 removed from mitochondria. YBX1 puncta may give rise to small particles carrying miR223 for uptake into endosomes and secretion in exosomes. Ma et al. eLife 2023;12:e85878. DOI: https://doi.org/10.7554/eLife.85878 13 of 23 Cell Biology Research article independent role in the selective capture of miRNAs by YBX1 in cytoplasmic P- body condensates. YBX1 puncta appear to shuttle between mitochondria and endosomes at which point miR223 bound to YBX1 may be further sorted into invaginations budding into the interior of endosomes. The highly selective nature of miRNA sorting and secretion in exosomes suggests an important role in the trafficking of miRNAs between cells. Numerous studies have suggested a role for secreted miRNAs in recipient cells (Cha et  al., 2015; Mittelbrunn et  al., 2011; Pegtel et  al., 2010; Valadi et al., 2007). Nonetheless, as miRNAs ordinarily act stoichiometrically on target mRNAs, the extremely low abundance and copy number of miRNAs/vesicle is hard to reconcile with such a functional role of secreted miRNA (Chevillet et al., 2014; Shurtleff et al., 2017). Our observation that the bulk of cellular miR223 is held within mitochondria suggests an alternative role in some structural or regula- tory process, perhaps essential for mitochondrial homeostasis, controlled by the selective extraction of unwanted miRNA into RNA granules and further by secretion in exosomes (Figure 7). Key resources table Materials and methods Reagent type (species) or resource Designation Source or reference Identifiers Additional information Cell line (Spodoptera frugiperda) Sf9 Cell line (Homo sapiens) HEK 293T cells Cell line (Homo sapiens) HEK 293T- YBX1 KO cells Other Other Other Cell culture facility at UC Berkeley Cell culture facility at UC Berkeley Obtained by CRISPR- Cas9 in Schekman Lab Cell line (Homo sapiens) HEK 293T- YBAP1 KO This study Obtained by CRISPR- Cas9 in Schekman Lab Cell line (Homo sapiens) HEK 293T- 3xHA- EGFP- OMP25 This study Obtained by overexpression of pLJM1- 3XHA- EGFP- OMP25 in Schekman lab Cell line (Homo sapiens) U- 2OS cells Other Cell culture facility at UC Berkeley Cell line (Homo sapiens) U- 2OS Parkin- GFP cells This study Obtained by overexpression of Parkin- GFP in Schekman lab Recombinant DNA reagent pFastBac His6 MBP N10 TEV LIC cloning vector (4 C) Addgene RRID: Addgene_30116 N/A Recombinant DNA reagent Tom22- mCherry (plasmid) This study Gift of Dr Li Yu lab Recombinant DNA reagent His- MBP- YBX1 (plasmid) This study Recombinant DNA reagent His- MBP- YBX1(1–51) (plasmid) This study Recombinant DNA reagent His- MBP- YBX1(52–129) (plasmid) This study Recombinant DNA reagent His- MBP- YBX1(130–324) (plasmid) This study Recombinant DNA reagent His- MBP- YBX1(1–129) (plasmid) This study Recombinant DNA reagent His- MBP- YBX1(F85A) (plasmid) This study Recombinant DNA reagent His- MBP- YBAP1 (plasmid) This study To express YBX1 in insect cells. Plasmid maintained in Schekman lab To express YBX1(1–51) in insect cells. Plasmid maintained in Schekman lab To express YBX1(52–129) in insect cells. Plasmid maintained in Schekman lab To express YBX1(1–51) in insect cells. Plasmid maintained in Schekman lab To express YBX1(1–129) in insect cells. Plasmid maintained in Schekman lab To express YBX1(F85A) in insect cells. Plasmid maintained in Schekman lab To express YBAP1 in insect cells. Plasmid maintained in Schekman lab Recombinant DNA reagent Mito- BFP This study Gift of Dr. Samantha Lewis lab Recombinant DNA reagent mCherry- Rab5(Q79L) (plasmid) Addgene RRID: Addgene_35138 Recombinant DNA reagent pLJM1- 3XHA- EGFP- OMP25 This study Continued on next page To express 3xHA- EGFP- OMP25 in HEK293T cells. Plasmid maintained in Schekman lab Ma et al. eLife 2023;12:e85878. DOI: https://doi.org/10.7554/eLife.85878 14 of 23 Cell Biology Research article Continued Reagent type (species) or resource Designation Source or reference Identifiers Additional information Anti- YBX1 (Rabbit polyclonal) Anti- YBAP1 (Mouse monoclonal) Anti- YBAP1(Rabbit polyclonal) Abcam RRID: AB_1950384 WB 1:1000 Santa Cruz Thermo Fisher Scientific RRID: AB_10611471 WB 1:1000 RRID: AB_2638956 WB 1:1000 Anti- Tim23 (Mouse monoclonal) BD Biosciences RRID: AB_398754 WB 1:1000 Anti- Tom20 (Mouse monoclonal) Abcam RRID: AB_945896 WB 1:1000 Anti- Calnexin (Rabbit polyclonal) Abcam RRID: AB_2069006 WB 1:2000 Anti- HA (Rabbit monoclonal) Cell Signaling RRID: AB_1549585 WB 1:1000 Anti- COX IV (Rabbit Monoclonal) Cell signaling RRID: AB_2085424 WB 1:1000 Anti- Citrate Synthase (Rabbit monoclonal) Cell signaling RRID: AB_2665545 WB 1:1000 Anti- Rab5 (Rabbit monoclonal) Cell signaling RRID: AB_2300649 WB 1:1000 Anti- LAMP1 (Rabbit monoclonal) Cell signaling RRID: AB_2687579 WB 1:1000 Anti- GRP78 (Rabbit polyclonal) Abcam RRID: AB_2119834 WB 1:3000 Antibody Antibody Antibody Antibody Antibody Antibody Antibody Antibody Antibody Antibody Antibody Antibody Antibody Antibody Antibody Anti- GADPH (Rabbit monoclonal) Anti- alpha Tubulin (Mouse monoclonal) Anti- beta Actin (Mouse monoclonal) Cell signaling RRID: AB_561053 WB 1:5000 Abcam RRID: AB_2241126 WB 1:5000 Abcam LICOR NIH RRID: AB_449644 WB 1:5000 https://www.licor.com/bio/image-studio-lite/ RRID: SCR_002285 https://fiji.sc/ Software, algorithm Image Studio Lite Software, algorithm FIJI Software, algorithm Prism 9 GraphPad RRID: SCR_002798 https://www.graphpad.com Cell lines and cell culture All immortalized cell lines were obtained from the UC- Berkeley Cell Culture Facility and were confirmed by short tandem repeat (STR) profiling and tested negative for mycoplasma contamina- tion. HEK 293T cells were cultured in DMEM with 10% FBS(VWR), NEAA (Gibco, Cat No: 11140050) and 1  mM Sodium Pyruvate (Gibco, Cat No: 11360070). For exosome production, we seeded cells at 10~20% confluency in 150 mm tissue culture dishes (Fisher Scientific, Cat No: 12- 565- 100) containing 30  ml of exosome- free medium. Exosomes were collected from 80% confluent cells (~48 hr). Exosome purification Conditioned medium was harvested from 80% to 90%  confluent HEK 293T cultured cells. All procedures were performed at 4 ℃. Cells and large debris were removed by centrifugation in a Sorvall R6  +centrifuge at 1000xg for 15  min followed by 10,000xg for 15  min using a FIBERlite F14−6x500 y rotor. The supernatant fraction was then centrifuged onto a 60% sucrose cushion in a buffer with 10 mM HEPES (pH 7.4) and 0.85% w/v NaCl at ~100,000 x g (28,000 RPM) for 1.5 h in a SW32Ti rotor. The interface over the sucrose cushion was collected and pooled for an additional centrifugation onto a 2 ml 60% sucrose cushion at ~120,000 x g (31,500 RPM) for 15 h using an SW41Ti rotor. The first collected interface was measured by refractometry and adjusted a sucrose concentration not exceeding 21%. For bulk purification, the EVs collected from the interface over the sucrose cushion after the first SW41Ti centrifugation were mixed with 60% sucrose to a final volume of 10  ml (the concentration of sucrose  ~50%). One ml of 40% and 1  ml of 10% sucrose Ma et al. eLife 2023;12:e85878. DOI: https://doi.org/10.7554/eLife.85878 15 of 23 Cell Biology Research article were sequentially overlaid and the samples were centrifuged at ~150,000 x g (36,500 rpm) for 15 h in an SW41Ti rotor. The exosomes were located at the 10%/40% interface and collected for RNA extraction or immunoblot. Density gradient isolation of mitochondria and mitoplasts Mitochondria were isolated according to a well- established published protocol (Wang et al., 2020). HEK293T Cells were harvested at 80% confluency and were homogenized in 6  vol of HB buffer (225 mM mannitol, 25 mM sucrose, 0.5% BSA, 0.5 mM EGTA, 30 mM Tris–HCl, pH 7.4, and protease inhibitors) in a prechilled Dounce homogenizer (Kontes). The lysate was centrifuged and the postnu- clear supernatant was collected. Crude mitochondria were centrifuged at 6300 x g for 8 min, washed once with MRB buffer (250 mM mannitol, 0.5 mM EGTA, and 5 mM HEPES, pH 7.4), resuspended in 1 ml MRB buffer, laid over a 30% Percoll solution (9 ml) and centrifuged at 95,000 g for 45 min. The buoyant, purified fraction of mitochondria was collected for further analysis. For mitoplast purification, crude mitochondria were resuspended into 10 vol MRB buffer with 0.2 mg/ml digitonin and incubated on ice for 15 min. Digitonin- treated crude mitochondria were laid over a 30% Percoll solution (9 ml) and centrifuged at 95,000 g for 45 min. A buoyant, purified fraction of mitoplasts was collected for further analysis. YBAP1 immunoprecipitation from mitochondria After the Percoll gradient purification, the enriched mitochondria were diluted 2 x into MRB buffer and centrifuged at 12,000  g for 10  min. The mitochondrial pellet was lysed in 0.5  ml RIPA buffer (50 mM Tris- HCl, pH 7.5, 150 mM NaCl, 2 mM EDTA, 1% DDM, 1 mM PMSF) containing protease inhibitors (1  mM 4- aminobenzamidine dihydrochloride, 1  µg/ml antipain dihydrochloride, 1  µg/ml aprotinin, 1  µg/ml leupeptin, 1  µg/ml chymostatin, 1  mM phenylmethylsulphonyl fluoride, 50  µM N- tosyl- L- phenylalanine chloromethyl ketone and 1 µg/ml pepstatin) and RNase inhibitors, followed by centrifugation at 12,000  g for 10  min. Supernatant fractions were incubated with 10  µl washed protein A Dynabeads (ThermoFisher Scientific, Catalog number: 10001D) and 0.5 µg mouse mono- clonal IgG antibody and rotated at 4 ℃ for 1 h. A magnetic rack was used to remove protein A beads and the resulting supernatant fractions were incubated with 40 µl washed protein A Dynabeads and 4 µg YBAP1 antibody or mouse IgG antibody and rotated at 4 ℃ overnight. The beads were collected using a magnetic rack, washed 3 x with 1 ml of RIPA buffer, and collected for immunoblot and RNA extraction. Mitochondria immunoprecipitation Mito- IP was performed as previously described with slight modifications (Chen et  al., 2016). The mito- IP cell- line was grown to ~90% confluency in 15 cm dishes. All the subsequent steps of mito- IP were performed using ice- cold buffers either in a cold- room or on ice. Cells (2x107) were washed twice with 10  ml of PBS and then harvested in 10  ml of mito- IP buffer (10  mM KH2PO4, 137  mM KCl) containing protease inhibitors (1  mM 4- aminobenzamidine dihydrochloride, 1  µg/ml antipain dihydrochloride, 1  µg/ml aprotinin, 1  µg/ml leupeptin, 1  µg/ml chymostatin, 1  mM phenylmethyl- sulphonyl fluoride, 50 µM N- tosyl- L- phenylalanine chloromethyl ketone and 1 µg/ml pepstatin) and TCEP 0.5 mM. The final mito IP buffer also contained 6 ml of OptiPrep (Sigma) per 100 ml. Cells were collected at 700xg for 5 min and resuspended in 1 ml of mito- IP buffer per 15 cm plate and then lysed using 5–10 passes through a 22 G needle. A post- nuclear supernatant (PNS) fraction was obtained after centrifuging the lysate at 1500xg for 10  min to remove unbroken cells and nuclei. Whenever necessary, a fraction of PNS was saved for immunoblot analysis. The resulting PNS was incubated with 100 µl of anti- HA magnetic beads (Sigma) pre- equilibrated in the mito- IP buffer in 1.5 ml microcentri- fuge tubes and then gently rotated on a mixer for 15 min. The beads were collected using a magnetic rack and washed 3 x for 5 min with 1 ml of mito- IP buffer. For mitoplast purification by osmotic shock, the supernatant was discarded after the final wash of the mito- IP sample, and the beads were gently resuspended in 200 µl of hypotonic osmotic shock buffer (OSB) containing 20  mM HEPES at pH 7.4. The resuspended sample was incubated on ice for 30 min and then the beads were centrifuged at 15,000 g for 15 min to sediment mitochondria/ mitoplasts. Beads were then resuspended in 100 µl of KPBS and proteinase K was added to achieve a final concentration of 10 µg/ml and samples were incubated on a rotating mixer at 4 °C for 15 min. Ma et al. eLife 2023;12:e85878. DOI: https://doi.org/10.7554/eLife.85878 16 of 23 Cell Biology Research article Subsequently, PMSF was added to a final concentration of 1  mM, along with a protease inhibitor cocktail, and the sample was incubated on ice for 5 min. Next, 2.5 units of RNase ONE (Promega) was added to the sample, which was further incubated on a rotating mixer at room temperature for 15 min. For protein analysis, the sample was eluted directly in the SDS loading buffer. Alternatively, Trizol was added to the sample to stop the reaction for RNA purification. To assess their quality, we assayed the immunoprecipitated mitochondria for the following protein markers by immunoblotting using the rabbit primary antibodies- anti- HA (1:1000), anti- COX IV (1:1000), anti- TOM20 (1:1000), anti- Citrate Synthase (1:1000), anti- RAB5 (1:1000), anti- LAMP1 (1:2000), anti- GAPDH (1:5000), and anti- GRP78 (Abcam) (1:3000). All the above antibodies were sourced from Cell Signaling Technology, unless stated otherwise. As a negative control, non- transduced HEK293T cells were used in these experiments to assess the non- specific capture of the marker proteins. Mitochondrial fractionation One mito- IP was performed per sample as described above. To ensure an even distribution of mito- chondria across the samples, we pooled washed beads from all the IPs and equally distributed aliquots for subsequent treatments. Mitochondria were lysed in a 50 μl final volume using either 1% vol/vol Triton X- 100 (final concentration) or by three sequential rounds of freeze/thaw using liquid- nitrogen, as indicated. Urea was added to a final concentration of 3 M. After a 10 min incubation on ice, samples were centrifuged at 15000xg for 15 min and supernatant and pellet fractions were collected as indi- cated. The total fractionated mitochondria were analyzed by immunoblotting for various mitochon- drial markers. To analyze the specific RNA content of total or fractionated mitochondria, we extracted RNA using Trizol (Invitrogen) as per manufacturer’s recommendations followed by q- PCR. In vitro packaging of miR223 and miR223mut Preparation of membranes and cytosol The membrane and cytosol fractions were prepared from HEK293T cells as previously described with slight modifications (Shurtleff et al., 2016; Temoche- Diaz et al., 2020). All steps were carried out in either the cold- room or on ice using ice- cold buffers and pre- chilled equipment. Briefly, HEK293T cells (80% confluency) were washed twice with PBS and then harvested in the homoge- nization buffer (HB) (250 mM sorbitol, 20 mM HEPES- KOH pH 7.4) containing protease inhibitor cocktail (1  mM 4- aminobenzamidine dihydrochloride, 1  µg/ml antipain dihydrochloride, 1  µg/ml aprotinin, 1 µg/ml leupeptin, 1 µg/ml chymostatin, 1 mM phenylmethylsulphonyl fluoride, 50 µM N- tosyl- L- phenylalanine chloromethyl ketone and 1 µg/ml pepstatin). The cell pellet was obtained by centrifuging the cells at 500xg for 5 min. After discarding the supernatant, cells were weighed and resuspended in two volumes of HB followed by lysis with 5–10 passages through a 22  G needle. The lysate was centrifuged at 1500xg for 10  min to remove unbroken cells and nuclei to obtain a PNS which was then centrifuged at 20,000xg for 30 min to obtain a membrane frac- tion. The supernatant from above was centrifuged at 150,000xg for 30 min using a TLA- 55 rotor (Beckman Coulter ) and the resulting supernatant was used as the cytosol fraction (~6 mg protein/ ml). Membranes from the first 20,000xg sedimentation were resuspended in 1 ml of HB and centri- fuged again at 20,000xg for 30 min. The pellet fraction was resuspended in one volume of HB and rested on an ice block for a minimum of 10 min until the insoluble components and debris settled at the bottom of the tube. The finely resuspended material in the resulting supernatant fraction was then transferred to a new microcentrifuge tube (to avoid the settled debris) and was used as the membrane fraction. Preparation of radiolabeled miR223 and miR223mut substrates HPLC purified miR223 and miR223mut oligos were obtained from IDT. A stock solution of these oligos (1  μl of a 10  μM) was 5’-end- labeled using T4PNK (NEB) and 5  μl of ATP, [γ–32P]- 6000  Ci/mmol 10mCi/ml EasyTide (PerkinElmer BLU502Z250UC) as per manufacturer’s recommendations in a 50 μl reaction volume. T4PNK was heat- inactivated at 70 °C for 15 min. Unincorporated radionucleotides were removed by passing through PerformaTM spin columns (EdgeBio). The flow- through (radiola- beled substrate) was collected and stored at –20 °C until further use. Ma et al. eLife 2023;12:e85878. DOI: https://doi.org/10.7554/eLife.85878 17 of 23 Cell Biology Research article In vitro miR223 packaging reaction Wherever indicated, 10 μl of cytosol (~5 mg/ml), 17 μl of membranes, 2 μl of radiolabeled substrate, 9 μl of 5 x incorporation buffer (400 mM KCl, 100 mM CaCl2, 60 mM HEPES- NaOH, pH 7.4, 6 mM, MgOAc), 4.5 μl of 10 x ATP regeneration system (400 mM creatine phosphate, 2 mg/ml creatine phos- phokinase, 10 mM ATP, 20 mM HEPES pH 7.2, 250 mM sorbitol, 150 mM KOAc, 5 mM MgOAc), 1 μl of ATP (100 mM, Promega), 0.5 μl of GTP (100 mM, Promega), 1 μl of Ribolock (40 U/μl, Invitrogen) were mixed to setup a 45 μl in vitro packaging reaction. In samples without the cytosol or membranes, the final reaction volumes were adjusted to 45 μl using HB. The reactions were incubated at either 30 °C or on ice, as indicated, for 15 min. Following the incubation, the indicated samples were subjected to RNAse ONE(Promega) using 10 U of the enzyme in the presence of urea (300 mM final concentration) in a total reaction volume of 60 μl. Wherever indicated, TritonX- 100 was added to a final concentra- tion of 1%. The RNAse treatments were carried out for 20 min at 30 °C followed by RNA extraction using DirectZol (Zymo Research) kits as per manufacturer’s protocol. RNA was precipitated overnight at –20 °C by the addition of 3 vol of ethanol, 1/10th volume of 3 M sodium acetate (pH5.2) and 30 μg Glycoblue reagent (Invitrogen). Precipitated RNA was sedimented at 16,000xg for 30 min followed by washing with ice- cold 70% ethanol. The RNA pellet was resuspended in 2  X RNA loading dye (NEB) and heated for 5 min at 70 °C. RNA was separated using a 15% denaturing polyacrylamide gel, followed by gel drying using a vacuum gel dryer (Model 583, Biorad). Radioactive bands were visual- ized by phosphorimaging using a Kodak storage phosphor screen and the Pharos FX Plus Molecular Imager (Biorad). Immunoblots Cell lysates and other samples were prepared by adding 2% SDS and heated at 95 ℃ for 10 min. Protein was quantified using a BCA Protein Assay Kit (Thermo Fisher Scientific) and appropriate amounts were mixed with 5 x SDS loading buffer. Samples were heated at 95℃ for 10 min and sepa- rated on 4–20% acrylamide Tris- glycine gradient gels (Life Technologies). Proteins were transferred to PVDF membranes (EMD Millipore, Darmstadt, Germany) and the membrane was blocked with 5% fat- free milk powder in TBST and incubated for 1 h at room temperature or overnight at 4 °C with primary antibodies. Blots were then washed in three washes of TBST for 10 min each. Membranes were incubated with anti- rabbit or anti- mouse secondary antibodies (GE Healthcare Life Sciences, Pittsburgh, PA) for 1 hr at room temperature and rinsed in three washes of TBST for 10 min each. Blots were developed with ECL- 2 reagent (Thermo Fisher Scientific). Primary antibodies used in this study were as follows: anti- Tim23 (BD, 611222), Calnexin (Abcam, ab22595), Actin (Abcam, ab8224), Tubulin (Abcam, ab7291), YBAP1 (Santa Cruz Biotechnology, sc- 271200). Immunofluorescence Cells were cultured on 12 mm round coverslips (corning) and were fixed with 4% EM- grade parafor- maldehyde (Electron Microscopy Science, Hatfield, PA) in PBS pH7.4 for 10 min at room temperature. Cells were then washed 3 x with PBS for 10 min each, treated with permeabilizing buffer (10% FBS in PBS) containing 0.1% saponin for 20 min and treated in blocking buffer for 30 min. Subsequently, cells were incubated with primary antibodies in permeabilizing buffer for 1 hr at room temperature, washed 3 x with PBS for 10 min each and incubated with secondary antibodies in permeabilizing buffer for 1 hr at room temperature and finally washed 3 x with PBS for 10 min each. Cells were mounted on slides with Prolong Gold with DAPI (Thermo Fisher Scientific, P36931). Primary antibodies used in the immu- nofluorescence studies were as follows: anti- YBX1 (Abcam, ab12148), YBAP1 (Santa Cruz Biotech- nology, sc- 271200). Images were acquired with Zeiss LSM900 confocal microscope and analyzed with the Fiji software (http://fiji.sc/Fiji). Quantitative real-time PCR Cellular and EV RNAs were extracted using a mirVana miRNA isolation kit (Thermo Fisher Scien- tific, AM1560) or Direct- zol RNA Miniprep kits (Zymo Research). Taqman miRNA assays for miRNA detection were purchased from Life Technologies. Assay numbers were: hsa- miR- 223–3 p, 002295; hsa- mir- 190–5  p, 000489; U6 snRNA, 001973. Total RNAs were quantified using RNA bioanalyzer (Agilent). Taqman qPCR master mix with no Amperase UNG was obtained from Life Technologies for reverse transcription. For mRNA, RevertAid First Strand cDNA Synthesis Kit (Thermo Scientific, K1621) Ma et al. eLife 2023;12:e85878. DOI: https://doi.org/10.7554/eLife.85878 18 of 23 Cell Biology Research article was used for reverse transcription. COX1 qPCR primers: Forward- 5’- TCTC AGGC TACA CCCT AGAC CA-3’, Reverse- 5’- ATCG GGGT AGTC CGAG TAAC GT-3’. GAPDH qPCR primers: Forward- 5’- CTGA CTTC AACA GCGA CACC -3’, Reverse- 5’- TAGC CAAA TTCG TTGT CATA CC-3’. Quantitative real- time PCR was performed using an QuantStudio 5 Real- Time PCR System (Applied Biosystems). Protein purification Twin Strep tag hybrid YBX1 was expressed and the protein was isolated 48  hr after PEI- mediated transfection of 293T cells. Cells were resuspended in PBS and collected by centrifugation for 5 min at 600  g. Pellet fractions were resuspended in 35  ml lysis buffer (50  mM Tris- HCl (pH 8),150mM NaCl,1mM EDTA, 2  mM DTT, 1  mM PMSF and 1  x protease inhibitor cocktail). After sonication of the cell suspension the crude lysate was centrifuged for 60 min at 20,000 rpm at 4 °C. The resulting supernatant fraction was incubated with 2 ml Strep- Tactin Sepharose resin (IBA, 2- 1201- 010) for 1 h. Strep- Tactin Sepharose resin samples were transferred to columns (18 ml) and protein- bound beads were washed with 60 ml wash buffer (50 mM Tris- HCl (pH 8), 500 mM NaCl, 1 mM EDTA, 2 mM DTT) until no protein was eluted as monitored by the Bio- Rad protein assay (Bio- Rad, Catalog #5000006). Proteins were eluted with 10 ml elution buffer (50 mM Tris- HCl (PH = 8),150mM NaCl, 10 mM desthi- obiotin, 1  mM EDTA, 2  mM DTT) and concentrated using an Amicon Ultra Centrifugal Filter Unit (50 kDa, 4 ml) (Fisher Scientific, EMD Millipore). Proteins were further purified by gel filtration chroma- tography (Superdex- 200, GE Healthcare) with columns equilibrated in storage buffer (50 mM Tris- HCl 7.4, 500 mM KCl, 5% glycerol, 1 mM DTT). Peak fractions corresponding to the appropriate fusion protein were pooled, concentrated, and distributed in 10  µl aliquots in PCR tubes, flash- frozen in liquid nitrogen and stored at –80 °C. Protein concentration was determined by known concentrations of BSA assessed by Coomassie Blue staining. Tagged (6xHis) and maltose- binding protein hybrid genes were expressed in baculovirus- infected SF9 insect cells (Lemaitre et al., 2019). Insect cell cultures (1 l, 1x106 cells/ml) were harvested 48 h after viral infection and collected by centrifugation for 20 min at 2000 rpm. The pellet fractions were resuspended in 35  ml lysis buffer (50  mM Tris- HCl 7.4, 0.5  M KCl, 5% glycerol, 10  mM imidazole, 0.5 µl/ml Benzonase nuclease (Sigma, 70746–3), 1 mM DTT, 1 mM PMSF and 1 x protease inhibitor cocktail). Cells were lysed by sonication and the crude lysate was centrifuged for 60 min at 20,000 rpm at 4 °C. After centrifugation, the supernatant fraction was incubated with 2 ml Ni- NTA His- Pur resin (Thermo Fisher, PI88222) for 1  hr. Ni- NTA resin samples were transferred to columns (18  ml) and protein- bound beads were washed with 60 ml lysis buffer until no protein was eluted as monitored by the Bio- Rad protein assay (Bio- Rad, Catalog #5000006). Proteins were eluted with 10 ml elution buffer (50 mM Tris- HCl 7.4, 0.5 M KCl, 5% glycerol, 500 mM imidazole). The eluted sample was incubated with 2 ml amylose resin (New England Biolabs, E8021L) for 1 hr at 4 °C. Amylose resin samples were transferred to columns and protein- bound beads were washed with 60 ml lysis buffer until no protein was eluted as monitored by the Bio- Rad protein assay. Proteins were eluted with 10 ml elution buffer (50  mM Tris- HCl 7.4, 500  mM KCl, 5% glycerol, 50  mM maltose) and were concentrated using an Amicon Ultra Centrifugal Filter Unit (50 kDa, 4 ml) (Thermo Fisher Scientific, EMD Millipore). Proteins were further purified by gel filtration chromatography (Superdex- 200, GE Healthcare) with columns equilibrated in storage buffer (50 mM Tris- HCl 7.4, 500 mM KCl, 5% glycerol, 1 mM DTT). Peak frac- tions corresponding to the appropriate fusion protein were pooled, concentrated, and distributed in 10 µl aliquots in PCR tubes, flash- frozen in liquid nitrogen and stored at –80 °C. Protein concentration was determined by known concentrations of BSA based on Coomassie blue staining. CRISPR/Cas9 genome editing A pX330- based plasmid expressing Venus fluorescent protein (Shurtleff et  al., 2016) was used to clone the gRNAs targeting YBAP1. A CRISPR guide RNA targeting the first exon of the YBAP1 open reading frame was designed following the CRISPR design website (http://crispor.tefor.net/crispor.py): CGCT GCGT GCCC CGTG TGCT . Oligonucleotides encoding gRNAs were annealed and cloned into pX330- Venus as described (Cong et  al., 2013). HEK293T cells were transfected by Lipofectamine 2000 for 48 hr at low passage number, trypsinized and sorted for single, Venus positive cells in 96- well plates by a BD Influx cell sorter. YBAP1 knockout candidates were confirmed by immunoblot. HEK 293T YBX1 knockout cells were generously provided by Dr. Xiaoman Liu (Liu et al., 2021). Ma et al. eLife 2023;12:e85878. DOI: https://doi.org/10.7554/eLife.85878 19 of 23 Cell Biology Research article Electrophoretic mobility shift assay Fluorescently labeled RNAs (5’-IRD800CWN) for detecting free and protein- bound RNA were ordered from Integrated DNA Technologies (IDT, Coralville, IA). EMSA was performed as described with some modification (Rio, 2014). Briefly, 1  nM of IRD800CWN- labeled RNA was incubated with increasing amounts of purified proteins, ranging from 500 pM - 1  μM. Buffer E was used in this incubation (25 mM Tris pH8.0, 100 mM KCl, 1.5 mM MgCl2, 0.2 mM EGTA, 0.05% Nonidet P- 40, 1 mM DTT, 5% glycerol, 50 µg/ml heparin). Reactions were incubated at 30 ℃ for 30 min then chilled on ice for 10 min. Samples were mixed with 6 x loading buffer (60 mM KCl, 10 mM Tris pH 7,6, 50% glycerol, 0.03% (w/v) xylene cyanol). Mixtures (5  µl) were loaded onto a 6% native polyacrylamide gel and electrophoresed at 200 V for 45 min in a cold room. The fluorescence signal was detected using an Odyssey CLx Imaging System (LI- COR Biosciences, Lincoln, NE). The software of the Odyssey CLx Imaging System was used to quantify fluorescence. To calculate Kds, we fitted used Hill equations with quantified data points. CD63-Nluc exosome secretion assay The CD63- Nluc exosome secretion assay was carried out as described (Williams et al., 2023). Briefly, cells stably expressing CD63- Nluc were cultured in 24- well plates until reaching approximately 80% confluence. All subsequent procedures were performed at 4 °C. Conditioned medium (200 µl) was collected from the appropriate wells and transferred to microcentrifuge tubes. The tubes were subjected to centrifugation at 1000×g for 15 min to remove intact cells, followed by an additional centrifugation at 10,000×g for 15 min to eliminate cellular debris. Supernatant fractions (50 µl) were used for measuring CD63- Nluc exosome luminescence. Cells were kept on ice and washed once with cold PBS, and then lysed in 200 µl of PBS containing 1% TX- 100 and protease inhibitor cocktail. For the measurement of CD63- Nluc exosome secretion, a master mix was prepared by diluting the Extracellular NanoLuc Inhibitor at a 1:1000 ratio and the NanoBRET Nano- Glo Substrate at a 1:333 ratio in PBS (Promega, Madison, WI, USA). Aliquots of the Nluc substrate/inhibitor master mix (100 µl) were added to 50  µl of the supernatant fraction obtained from the medium- speed centrifugation. The mixture was briefly vortexed, and luminescence was measured using a Promega GlowMax 20/20 Luminometer (Promega, Madison, WI, USA). Following luminescence measurements, 1.5 µl of 10% TX- 100 was added to each reaction tube to achieve a final concentration of 0.1% TX- 100. Samples were vortexed briefly, and luminescence was measured again. For intracellular normalization, the lumi- nescence of 50  µl of cell lysate was measured using the Nano- Glo Luciferase Assay kit (Promega, Madison, WI, USA) following the manufacturer’s instructions. The exosome production index (EPI) for each sample was calculated using the formula: EPI = ([medium] - [medium +0.1% TX- 100])/cell lysate. Acknowledgements We thank Dr. Samantha Lewis for advice about localization to mitochondria and for sharing a plasmid; thanks Matthew J Shurtleff, David Melville, Shenjie Wu, Jordan Ngo, Congyan Zhang, Justin Williams, Morayma M Temoche- Diaz for suggestions and reading and editing the manuscript. We also thank staff at the UC Berkeley shared facilities, the Cell Culture Facility, the Flow Cytometry Facility and QB3- Berkeley (The California Institute for Quantitative Biosciences at UC Berkeley). LM and JS are supported as Research Associates of the HHMI. RS is an Investigator of the HHMI, a Senior fellow of the UC Berkeley Miller Institute of Science and Scientific Director of Aligning Science Across Parkin- son’s Disease. Additional information Funding Funder Howard Hughes Medical Institute Grant reference number Author Randy Schekman Ma et al. eLife 2023;12:e85878. DOI: https://doi.org/10.7554/eLife.85878 20 of 23 Cell Biology Research article Funder Grant reference number Author The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication. Author contributions Liang Ma, Conceptualization, Data curation, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing – original draft, Writing – review and editing; Jasleen Singh, Data curation, Formal analysis, Methodology, Writing – original draft, Writing – review and editing; Randy Schekman, Conceptualization, Resources, Supervision, Funding acquisition, Validation, Investigation, Method- ology, Writing – original draft, Project administration, Writing – review and editing Author ORCIDs Liang Ma Randy Schekman http://orcid.org/0000-0003-3227-5917 http://orcid.org/0000-0001-8615-6409 Decision letter and Author response Decision letter https://doi.org/10.7554/eLife.85878.sa1 Author response https://doi.org/10.7554/eLife.85878.sa2 Additional files Supplementary files • MDAR checklist Data availability All data generated or analyzed in this study are included in the manuscript and supporting files. Source data files have been provided for Figure 1b, Figure 1c, Figure 1f, Figure 1g, Figure 1- figure supplement 2a- c, Figure 2b, Figure 2g, Figure 2- figure supplement 1a- f, Figure 3a, Figure 3c, Figure 3e, Figure 3h, Figure 3 supplement 1b- c, Figure 4a, Figure 4b, Figure 4e, Figure 4f, Figure 4h, Figure 4i, Figure 4- figure supplement 1a- b, Figure 4- figure supplement 2a, Figure 5b. References Cha DJ, Franklin JL, Dou Y, Liu Q, Higginbotham JN, Demory Beckler M, Weaver AM, Vickers K, Prasad N, Levy S, Zhang B, Coffey RJ, Patton JG. 2015. KRAS- dependent sorting of miRNA to exosomes. eLife 4:e07197. DOI: https://doi.org/10.7554/eLife.07197, PMID: 26132860 Chen WW, Freinkman E, Wang T, Birsoy K, Sabatini DM. 2016. Absolute quantification of matrix metabolites reveals the dynamics of mitochondrial metabolism. Cell 166:1324–1337.. DOI: https://doi.org/10.1016/j.cell. 2016.07.040, PMID: 27565352 Chevillet JR, Kang Q, Ruf IK, Briggs HA, Vojtech LN, Hughes SM, Cheng HH, Arroyo JD, Meredith EK, Gallichotte EN, Pogosova- Agadjanyan EL, Morrissey C, Stirewalt DL, Hladik F, Yu EY, Higano CS, Tewari M. 2014. Quantitative and stoichiometric analysis of the microRNA content of exosomes. PNAS 111:14888–14893. DOI: https://doi.org/10.1073/pnas.1408301111, PMID: 25267620 Cocucci E, Racchetti G, Meldolesi J. 2009. Shedding microvesicles: artefacts no more. Trends in Cell Biology 19:43–51. DOI: https://doi.org/10.1016/j.tcb.2008.11.003, PMID: 19144520 Cong L, Ran FA, Cox D, Lin S, Barretto R, Habib N, Hsu PD, Wu X, Jiang W, Marraffini LA, Zhang F. 2013. Multiplex genome engineering using CRISPR/Cas systems. Science 339:819–823. DOI: https://doi.org/10. 1126/science.1231143, PMID: 23287718 Correia- Melo C, Ichim G, Tait SWG, Passos JF. 2017. Depletion of mitochondria in mammalian cells through enforced mitophagy. Nature Protocols 12:183–194. DOI: https://doi.org/10.1038/nprot.2016.159, PMID: 28005069 Harding C, Heuser J, Stahl P. 1983. Receptor- mediated endocytosis of transferrin and recycling of the transferrin receptor in rat reticulocytes. The Journal of Cell Biology 97:329–339. DOI: https://doi.org/10.1083/jcb.97.2. 329, PMID: 6309857 Huang L, Mollet S, Souquere S, Le Roy F, Ernoult- Lange M, Pierron G, Dautry F, Weil D. 2011. Mitochondria associate with P- bodies and modulate microRNA- mediated RNA interference. The Journal of Biological Chemistry 286:24219–24230. DOI: https://doi.org/10.1074/jbc.M111.240259, PMID: 21576251 Jeandard D, Smirnova A, Tarassov I, Barrey E, Smirnov A, Entelis N. 2019. Import of Non- Coding RNAs into Human Mitochondria: A. Critical Review and Emerging Approaches. Cells 8:286. DOI: https://doi.org/10.3390/ cells8030286 Ma et al. eLife 2023;12:e85878. DOI: https://doi.org/10.7554/eLife.85878 21 of 23 Cell Biology Research article Lemaitre RP, Bogdanova A, Borgonovo B, Woodruff JB, Drechsel DN. 2019. FlexiBAC: a versatile, open- source baculovirus vector system for protein expression, secretion, and proteolytic processing. BMC Biotechnology 19:20. DOI: https://doi.org/10.1186/s12896-019-0512-z, PMID: 30925874 Li J, Liu K, Liu Y, Xu Y, Zhang F, Yang H, Liu J, Pan T, Chen J, Wu M, Zhou X, Yuan Z. 2013. Exosomes mediate the cell- to- cell transmission of IFN-α-induced antiviral activity. Nature Immunology 14:793–803. DOI: https://doi. org/10.1038/ni.2647, PMID: 23832071 Liu XM, Ma L, Schekman R. 2021. Selective sorting of microRNAs into exosomes by phase- separated YBX1 condensates. eLife 10:e71982. DOI: https://doi.org/10.7554/eLife.71982, PMID: 34766549 Lyons SM, Achorn C, Kedersha NL, Anderson PJ, Ivanov P. 2016. YB- 1 regulates tiRNA- induced Stress Granule formation but not translational repression. Nucleic Acids Research 44:6949–6960. DOI: https://doi.org/10. 1093/nar/gkw418, PMID: 27174937 Matsumoto K, Tanaka KJ, Tsujimoto M. 2005. An acidic protein, YBAP1, mediates the release of YB- 1 from mRNA and relieves the translational repression activity of YB- 1. Molecular and Cellular Biology 25:1779–1792. DOI: https://doi.org/10.1128/MCB.25.5.1779-1792.2005, PMID: 15713634 Mittelbrunn M, Gutiérrez- Vázquez C, Villarroya- Beltri C, González S, Sánchez- Cabo F, González MÁ, Bernad A, Sánchez- Madrid F. 2011. Unidirectional transfer of microRNA- loaded exosomes from T cells to antigen- presenting cells. Nature Communications 2:282. DOI: https://doi.org/10.1038/ncomms1285, PMID: 21505438 Mukherjee K, Ghoshal B, Ghosh S, Chakrabarty Y, Shwetha S, Das S, Bhattacharyya SN. 2016. Reversible HuR- microRNA binding controls extracellular export of miR- 122 and augments stress response. EMBO Reports 17:1184–1203. DOI: https://doi.org/10.15252/embr.201541930, PMID: 27402548 Muta T, Kang D, Kitajima S, Fujiwara T, Hamasaki N. 1997. p32 protein, a splicing factor 2- associated protein, is localized in mitochondrial matrix and is functionally important in maintaining oxidative phosphorylation. The Journal of Biological Chemistry 272:24363–24370. DOI: https://doi.org/10.1074/jbc.272.39.24363, PMID: 9305894 Pegtel DM, Cosmopoulos K, Thorley- Lawson DA, van Eijndhoven MAJ, Hopmans ES, Lindenberg JL, de Gruijl TD, Würdinger T, Middeldorp JM. 2010. Functional delivery of viral miRNAs via exosomes. PNAS 107:6328–6333. DOI: https://doi.org/10.1073/pnas.0914843107, PMID: 20304794 Rio DC. 2014. Electrophoretic Mobility Shift Assays for RNA–Protein Complexes. Cold Spring Harbor Protocols 2014:435–440. DOI: https://doi.org/10.1101/pdb.prot080721 Santangelo L, Giurato G, Cicchini C, Montaldo C, Mancone C, Tarallo R, Battistelli C, Alonzi T, Weisz A, Tripodi M. 2016. The RNA- Binding Protein SYNCRIP Is a Component of the Hepatocyte Exosomal Machinery Controlling MicroRNA Sorting. Cell Reports 17:799–808. DOI: https://doi.org/10.1016/j.celrep.2016.09.031, PMID: 27732855 Shurtleff MJ, Temoche- Diaz MM, Karfilis KV, Ri S, Schekman R. 2016. Y- box protein 1 is required to sort microRNAs into exosomes in cells and in a cell- free reaction. eLife 5:e19276. DOI: https://doi.org/10.7554/ eLife.19276, PMID: 27559612 Shurtleff MJ, Yao J, Qin Y, Nottingham RM, Temoche- Diaz MM, Schekman R, Lambowitz AM. 2017. Broad role for YBX1 in defining the small noncoding RNA composition of exosomes. PNAS 114:E8987–E8995. DOI: https://doi.org/10.1073/pnas.1712108114, PMID: 29073095 Temoche- Diaz MM, Shurtleff MJ, Nottingham RM, Yao J, Fadadu RP, Lambowitz AM, Schekman R. 2019. Distinct mechanisms of microRNA sorting into cancer cell- derived extracellular vesicle subtypes. eLife 8:e47544. DOI: https://doi.org/10.7554/eLife.47544, PMID: 31436530 Temoche- Diaz MM, Shurtleff MJ, Schekman R. 2020. Buoyant density fractionation of small extracellular vesicle sub- populations derived from mammalian cells. Bio- Protocol 10:e3706. DOI: https://doi.org/10.21769/ BioProtoc.3706, PMID: 33659370 Teng Y, Ren Y, Hu X, Mu J, Samykutty A, Zhuang X, Deng Z, Kumar A, Zhang L, Merchant ML, Yan J, Miller DM, Zhang H- G. 2017. MVP- mediated exosomal sorting of miR- 193a promotes colon cancer progression. Nature Communications 8:14448. DOI: https://doi.org/10.1038/ncomms14448, PMID: 28211508 Trajkovic K, Hsu C, Chiantia S, Rajendran L, Wenzel D, Wieland F, Schwille P, Brügger B, Simons M. 2008. Ceramide triggers budding of exosome vesicles into multivesicular endosomes. Science 319:1244–1247. DOI: https://doi.org/10.1126/science.1153124, PMID: 18309083 Valadi H, Ekström K, Bossios A, Sjöstrand M, Lee JJ, Lötvall JO. 2007. Exosome- mediated transfer of mRNAs and microRNAs is a novel mechanism of genetic exchange between cells. Nature Cell Biology 9:654–659. DOI: https://doi.org/10.1038/ncb1596, PMID: 17486113 Villarroya- Beltri C, Gutiérrez- Vázquez C, Sánchez- Cabo F, Pérez- Hernández D, Vázquez J, Martin- Cofreces N, Martinez- Herrera DJ, Pascual- Montano A, Mittelbrunn M, Sánchez- Madrid F. 2013. Sumoylated hnRNPA2B1 controls the sorting of miRNAs into exosomes through binding to specific motifs. Nature Communications 4:2980. DOI: https://doi.org/10.1038/ncomms3980, PMID: 24356509 Wang WX, Prajapati P, Nelson PT, Springer JE. 2020. The mitochondria- associated er membranes are novel subcellular locations enriched for inflammatory- responsive microRNAs. Molecular Neurobiology 57:2996–3013. DOI: https://doi.org/10.1007/s12035-020-01937-y, PMID: 32451872 Williams JK, Ngo JM, Lehman IM, Schekman R. 2023. Annexin A6 mediates calcium- dependent exosome secretion during plasma membrane repair. eLife 12:e86556. DOI: https://doi.org/10.7554/eLife.86556, PMID: 37204294 Wozniak AL, Adams A, King KE, Dunn W, Christenson LK, Hung WT, Weinman SA. 2020. The RNA binding protein FMR1 controls selective exosomal miRNA cargo loading during inflammation. The Journal of Cell Biology 219:e201912074. DOI: https://doi.org/10.1083/jcb.201912074, PMID: 32970791 Ma et al. eLife 2023;12:e85878. DOI: https://doi.org/10.7554/eLife.85878 22 of 23 Cell Biology Research article Yuyama K, Sun H, Mitsutake S, Igarashi Y. 2012. Sphingolipid- modulated exosome secretion promotes clearance of amyloid-β by microglia. The Journal of Biological Chemistry 287:10977–10989. DOI: https://doi.org/10. 1074/jbc.M111.324616, PMID: 22303002 Ma et al. eLife 2023;12:e85878. DOI: https://doi.org/10.7554/eLife.85878 23 of 23 Cell Biology
null
10.1371_journal.pntd.0009424.pdf
Data Availability Statement: All relevant data are within the manuscript and its Supporting Information files.
null
RESEARCH ARTICLE The utilization of advance telemetry to investigate critical physiological parameters including electroencephalography in cynomolgus macaques following aerosol challenge with eastern equine encephalitis virus 1, Franco D. Rossi1, Michael V. Accardi2, Brandi L. Dorsey1, Thomas John C. TrefryID R. SpragueID E. KimmelID P. CardileID L. PittID 1, Suzanne E. Wollen-RobertsID 1, Pamela J. GlassID 4, Darci R. SmithID 1* 1, Joshua D. Shamblin1, Adrienne 1, Lynn J. Miller3, Crystal W. Burke1, Anthony 1, Sina Bavari1, Simon Authier2, William D. PrattID 1*, Farooq NasarID 1, Margaret 1 Virology Division, United States Army Medical Research Institute of Infectious Diseases, Frederick, Maryland, United States of America, 2 Charles River (formerly Citoxlab North America), Laval, Canada, 3 Veterinary Medicine Division, United States Army Medical Research Institute of Infectious Diseases, Frederick, Maryland, United States of America, 4 Division of Medicine, United States Army Medical Research Institute of Infectious Diseases, Frederick, Maryland, United States of America * margaret.l.pitt.civ@mail.mil (MLP); fanasar@icloud.com (FN) Abstract Most alphaviruses are mosquito-borne and can cause severe disease in humans and domesticated animals. In North America, eastern equine encephalitis virus (EEEV) is an important human pathogen with case fatality rates of 30–90%. Currently, there are no therapeutics or vaccines to treat and/or prevent human infection. One critical impediment in countermeasure development is the lack of insight into clinically relevant parameters in a susceptible animal model. This study examined the disease course of EEEV in a cynomolgus macaque model utilizing advanced telemetry technology to continuously and simultaneously measure temperature, respiration, activity, heart rate, blood pressure, electrocardiogram (ECG), and electroencephalography (EEG) following an aerosol chal- lenge at 7.0 log10 PFU. Following challenge, all parameters were rapidly and substantially altered with peak alterations from baseline ranged as follows: temperature (+3.0–4.2˚C), respiration rate (+56–128%), activity (-15-76% daytime and +5–22% nighttime), heart rate (+67–190%), systolic (+44–67%) and diastolic blood pressure (+45–80%). Cardiac abnor- malities comprised of alterations in QRS and PR duration, QTc Bazett, T wave morphology, amplitude of the QRS complex, and sinoatrial arrest. An unexpected finding of the study was the first documented evidence of a critical cardiac event as an immediate cause of euthanasia in one NHP. All brain waves were rapidly (*12–24 hpi) and profoundly altered with increases of up to 6,800% and severe diffuse slowing of all waves with decreases of ~99%. Lastly, all NHPs exhibited disruption of the circadian rhythm, sleep, and food/fluid a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Trefry JC, Rossi FD, Accardi MV, Dorsey BL, Sprague TR, Wollen-Roberts SE, et al. (2021) The utilization of advance telemetry to investigate critical physiological parameters including electroencephalography in cynomolgus macaques following aerosol challenge with eastern equine encephalitis virus. PLoS Negl Trop Dis 15(6): e0009424. https://doi.org/10.1371/journal. pntd.0009424 Editor: Alain Kohl, University of Glasgow, UNITED KINGDOM Received: December 4, 2020 Accepted: April 29, 2021 Published: June 17, 2021 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pntd.0009424 Copyright: This is an open access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication. PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 1 / 32 Data Availability Statement: All relevant data are within the manuscript and its Supporting Information files. Funding: This study was supported by a grant from Medical Countermeasure Systems-Joint Vaccine Acquisition Program [Grant #A5XA0A7444182001 (FN and MLP)]. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Utilization of advance telemetry in cynomolgus macaques infected with EEEV intake. Accordingly, all NHPs met the euthanasia criteria by ~106–140 hpi. This is the first of its kind study utilizing state of the art telemetry to investigate multiple clinical parameters rel- evant to human EEEV infection in a susceptible cynomolgus macaque model. The study provides critical insights into EEEV pathogenesis and the parameters identified will improve animal model development to facilitate rapid evaluation of vaccines and therapeutics. Author summary In North America, EEEV causes the most severe mosquito-borne disease in humans highlighted by fatal encephalitis and permeant debilitating neurological sequelae in survi- vors. The first confirmed human cases were reported more than 80 years ago and since then multiple sporadic outbreaks have occurred including one of the largest in 2019. Unfortunately, most human infections are diagnosed at the on-set of severe neurological symptoms and consequently a detailed disease course in humans is lacking. This gap in knowledge is a significant obstacle in the development of appropriate animal models to evaluate countermeasures. Here, we performed a cutting-edge study by utilizing a new telemetry technology to understand the course of EEEV infection in a susceptible macaque model by measuring multiple physiological parameters relevant to human dis- ease. Our study demonstrates that the infection rapidly produces considerable alterations in many critical parameters including the electrical activity of the heart and the brain lead- ing to severe disease. The study also highlights the extraordinary potential of new teleme- try technology to develop the next generation of animal models to comprehensively investigate pathogenesis as well as evaluate countermeasures to treat and/or prevent EEEV disease. Introduction The genus Alphavirus in the family Togaviridae is comprised of small, spherical, enveloped viruses with genomes consisting of a single stranded, positive-sense RNA ~11–12 kb in length. Alphaviruses comprise 31 recognized species classified into eleven complexes based on anti- genic and/or genetic similarities [1–5]. The two aquatic alphavirus complexes [Salmon pancre- atic disease virus (SPDV) and Southern elephant seal virus (SESV)] are not known to utilize arthropods in their transmission cycles, whereas all of the remaining complexes [Barmah For- est, Ndumu, Middelburg, Semliki Forest, Venezuelan (VEE), eastern (EEE), western equine encephalitis (WEE), Trocara, and Eilat], consist of arboviruses that almost exclusively utilize mosquitoes as vectors [6]. Mosquito-borne alphaviruses infect diverse vertebrate hosts includ- ing equids, birds, amphibians, reptiles, rodents, pigs, nonhuman primates (NHPs), and humans [6]. The ability to infect both mosquitoes and vertebrates enables the maintenance of alpha- viruses in natural endemic transmission cycles that occasionally spillover into the human pop- ulation and cause disease. Infections with Old World alphaviruses such as chikungunya, o’nyong-nyong, Sindbis, and Ross River are rarely fatal but disease is characterized by rash and debilitating arthralgia that can persist for months or years [6]. In contrast, New World alphaviruses such as eastern (EEEV), western (WEEV), and Venezuelan equine encephalitis virus (VEEV) can cause fatal encephalitis [6]. Of the New World alphaviruses, EEEV is of foremost importance in North America. EEEV is comprised of four lineages; one North American (NA) and three South American [7,8]. The PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 2 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV NA lineage is associated with severe human disease, and is endemic in the eastern United States and Canada, and the Gulf coast of the United States [6]. The main transmission cycle is between passerine birds and Culiseta melanura mosquitoes, with enzootic foci concentrated in the Mid-Atlantic, New England, Michigan, Wisconsin, and Florida [6,9]. This cycle can spill- over into humans and domesticated animals to cause severe disease with human and equid case-fatality rates of 30–90% and >90%, respectively [6,10]. Human survivors can suffer from debilitating and permanent long-term neurological sequelae at rates of 35–80% [6,10]. In addi- tion to natural infections, many properties of EEEV including ease of isolation from nature and amplification in tissue culture, high virus titers, virus stability, high infectivity and uni- form lethality via aerosol route in non-human primates are conducive to weaponization. These properties facilitated the development of EEEV as a potential biological weapon during the Cold War by the United States and the former Union of Soviet Socialist Republics (USSR) [11,12]. These traits have also led to the assignment of EEEV to the NIAID category B list and as a select agent. Currently, there are no licensed therapeutics and/or vaccines to treat or pre- vent EEEV infection and the U.S. population remains vulnerable to bioterror event/s or natu- ral disease outbreaks. In order to develop effective therapeutic and/or vaccine countermeasures, various rodent and NHP models have been utilized to recapitulate various aspects of human disease. Among these models, aerosol infection of the cynomolgus macaques can produce alterations in blood chemistry and hematology, febrile illness, viremia, neurological disease, and lethality [13–15]. However, the data in the model are limited and requires further examination to gain insights into EEEV clinical disease course and pathogenesis. In this study, we investigated disease course in the cynomolgus macaque model following an aerosol challenge with EEEV utilizing state of the art telemetry to measure clinical signs including temperature, activity, respiration, heart rate, blood pressure, electrocardiogram (ECG), and electroencephalography (EEG). Materials and methods Ethics statement This work was supported by an approved USAMRIID Institute Animal Care and Use Com- mittee (IACUC) animal research protocol. Research was conducted under an IACUC approved protocol in compliance with the Animal Welfare Act, PHS Policy, and other Federal statutes and regulations relating to animals and experiments involving animals. The facility where this research was conducted is accredited by the Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC International) and adheres to principles stated in the Guide for the Care and Use of Laboratory Animals, National Research Council, 2011 [16]. Virus and cells Eastern equine encephalitis virus isolate V105-00210 was obtained from internal USAMRIID collection. The virus (Vero-1) was received from the Centers for Disease Control and Preven- tion (CDC) Fort Collins, CO [17]. The virus stock was passed in Vero-76 cells (American Type Culture Collection, ATCC; Bethesda, MD) twice for the production of Master (Vero-2) and Working (Vero-3) virus stocks. The virus stock was deep sequenced to verify genomic sequence and to ensure purity. In addition, the stock was tested and determined to be negative for both endotoxin and mycoplasma. Vero-76 cells were propagated at 37˚C with 5% CO2 in Dulbecco’s Minimal Essential Medium (DMEM) (CellGro) containing 2% (v/v) fetal bovine serum (FBS) (Hyclone), sodium PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 3 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV pyruvate (1 mM) (CellGro), 1% (v/v) non-essential amino acids (CellGro), and 50 μg/mL gen- tamicin (Invitrogen). Nonhuman primate study design Four (2 males, 2 females) cynomolgus macaques (Macaca fascicularis) of Chinese origin ages 5–8 years and weighing *3–9 kg were obtained (Covance). All NHPs were prescreened and determined to be negative for Herpes B virus, simian T-lymphotropic virus 1, simian immuno- deficiency virus, simian retrovirus D 1/2/3, tuberculosis, Salmonella spp., Campylobacter spp., hypermucoviscous Klebsiella spp., and Shigella spp. NHPs were also screened for the presence of neutralizing antibodies to EEEV, VEEV IAB, and WEEV by plaque reduction neutralization test (PRNT80). Telemetry devices and data collection All NHPs were implanted with devices by Covance/DSI and 4-weeks post-implantation the NHPs were transported to USAMRIID. NHPs were implanted with a Data Sciences Interna- tional (DSI) PhysioTel Digital M11 and two DSI PhysioTel Digital M01 implants. Each M01 device was dedicated to each hemisphere of the brain to measure EEG activity. The devices were implanted at left and right scapula and the leads were placed intracranially. The M11 implant was utilized to measure temperature, activity, respiration and heart rates, blood pres- sure, and ECG. Following implantation, the NHPs completely recovered after 2–4 weeks. Implanted animals were placed in individual cages with a single DSI TRX-1 receiver per cage with additional TRX-1 receivers placed in the study room for redundancy. These receivers were connected via Cat 5e cable to communication link controllers. The digital data was routed to data acquisition computers, which captured and archived the digital data using the Notocord-hem Evolution software platform (Version 4.3, Notocord Inc., Newark, NJ). The telemetry devices were activated in the NHPs and pre-challenge baseline values for each parameter (temperature, activity, respiration, heart rate, blood pressure, ECG, and EEG) were obtained for five days. All physiological parameters were sampled at various rates; activity and temperature (1 Hz), blood pressure and ECG (500 Hz), and EEG (1,000 Hz). Respiration and heart rates were derived utilizing NOTOCORD software (NOTOCORD Systems, Instem Company, Le Pecq, France). EEG analysis was performed using NeuroScore software version 3.1 (Data Sciences International, St. Paul, Minnesota, USA). Establishing baseline for each physiological parameter To establish a baseline of each parameter for individual animals, telemetry data was collected continuously for five days prior to challenge. Data for each animal and parameter was utilized to generate a 0.5-hr interval by averaging up to 1,800 data points. Subsequently, a 48-point ref- erence baseline for a 24-hr period was generated by averaging time-matched five previously calculated baseline values. Baseline 0.5-hr averages and standard deviations (SD) were gener- ated. Analysis was also performed in 12-hr day/nighttime intervals. Daytime and nighttime are defined as 6 am to 6 pm and 6 pm to 6 am, respectively. To generate a 12-hr average, all raw data points with respective times were averaged to generate each day/nighttime values. All comparisons between pre- and post-challenge were time matched. Following determination of baseline values, NHPs were challenged with a target dose of 7.0 log10 PFU of EEEV via the aerosol route. The NHPs were observed for signs of clinical disease and data for each parameter (temperature, activity, respiration and heart rates, blood pressure, ECG, and EEG) was obtained. Time-matched comparisons were made between pre-challenge baseline and post-challenge values. PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 4 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV Aerosol challenge NHPs were exposed to the target inhalation dose of 7.0 log10 PFU of EEEV in the head-only Automated Bioaerosol Exposure System (ABES-II). The virus stock at 9.3 log10 PFU/mL was diluted to 9.0 log10 PFU/mL and utilized in the nebulizer. The inhalation challenge was gener- ated using a Collison Nebulizer to produce a highly respirable aerosol (flow rate 7.5±0.1 L/ minute). The system generated a target inhalation of 1 to 3 μm mass median aerodynamic diameters determined by TSI Aerodynamic Particle Sizer. Samples of the pre-spray suspension and inhalation collected from the exposure chamber using an all-glass impinger (AGI) during the challenge were analyzed by plaque assay to determine the inhaled PFU. The inhalation challenge dose for each NHP was calculated from the minute volume determined with a whole-body plethysmograph box using Buxco XA software. The total volume of inhaled dose was determined by the exposure time required to deliver the estimated inhaled dose. Individ- ual NHPs were challenged successively in the ABES-II. Post-exposure monitoring and score criteria NHP observations began five days prior to aerosol exposure to obtain baseline data. Following aerosol challenge all NHPs were monitored daily via continuous 24-hr remote monitoring to limit room entries. Clinical signs of disease were observed and a score for each NHP was deter- mined by evaluating three parameters; neurological score, temperature, and responsiveness score. The neurological score scale was as follows: 0 = normal; 1 = mild and infrequent trem- ors, 2 = hyperactivity, infrequent tremors, 3 = constant and repetitive tremors, and 10 = unre- sponsive. Temperature score scale was as follows; 0 = normal baseline, 1 = 1˚C above or below baseline, 2 = 2˚C above or below baseline, 3 = 3˚C above or below baseline, 10 = 4˚C above or below baseline. Responsive score scale was as follows; 0 = normal, 1 = mild unre- sponsiveness, 2 = moderate unresponsiveness, 3 = severe unresponsiveness, and 10 = unre- sponsive. NHPs with a total score �10 met the euthanasia criteria. Tissue preparation All tissues and fluids were collected at the time of euthanasia and frozen. For quantification of virus, frozen samples of brain (frontal cortex), olfactory bulb, cervical spinal cord, and heart were thawed, weighed, and suspended in 1X PBS to generate 10% (w:v) tissue suspensions using a Mixer Mill 300 (Retsch, Haan, Germany). Tissue homogenates were centrifuged at 5,000 x g for 5 mins and clarified supernatants were used immediately for plaque assay as described below. Cerebral spinal fluid (CSF), plasma, and serum samples were thawed and used immediately in plaque assays. Plaque assay ATCC Vero 76 cells were seeded overnight on 6-well tissue culture plates to achieve 90–95% confluence. Triplicate wells were infected with 0.1-ml aliquots from serial 10-fold dilutions in Hanks’ Balanced Saline Solution (HBSS) and virus was adsorbed for 1 hr at 37˚C, 5% CO2. After incubation, cells were overlaid with Eagle’s Basal Medium (BME) (Gibco A15950DK) containing 0.6% agarose supplemented with 10% heat-inactivated FBS, 2% Penicillin/Strepto- mycin (10,000 IU/mL and 10,000 μg/mL, respectively), and incubated for 24 hr at 37˚C, 5% CO2. A second agarose overlay, prepared as described above, containing 5% neutral red vital stain (Gibco 02-0066DG) was added to the wells and incubated for 18–24 hr for visualization of plaques. Plaques were counted and expressed in either plaque forming units (PFU) per mL (PFU/mL) or PFU/g of tissue. PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 5 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV Plaque reduction neutralization test (PRNT80) Serum samples were heat-inactivated at 56˚C for 30 mins. Samples were serially diluted 2-fold starting at 1:10 and were mixed with equal volumes of medium containing ~2,000 PFU/mL of virus and incubated at 37˚C, 5% CO2. Following incubation, six-well plates containing mono- layers of Vero-76 cells were infected with 100μL of virus-serum mixtures in triplicates and incu- bated at 37˚C, 5% CO2 for ~1 hr. Following incubation, a secondary agarose overlay containing 5% neutral red vital stain was added to the wells and incubated for ~24 hr for visualization of plaques. Plaques were counted, PRNT titers were calculated and expressed as the reciprocal of serum dilution yielding a >80% reduction (PRNT80) in the number of plaques. The limit of detection in PRNT80 assay is <1:20. All samples were analyzed three times in the assay. Statistics All comparisons between pre- and post-challenge were time matched. GraphPad Prism ver- sion 7.00 for Windows (GraphPad Software, La Jolla, California, USA) software was utilized for statistical analysis. Significant differences in each parameter were determined using one- way ANOVA followed by a Tukey Test. Results EEEV challenge study design, survival, and detection of infectious virus in tissues at terminal time point Four macaques (2 males and 2 females) were implanted with telemetry devices to continuously and simultaneously monitor physiological parameters for the duration of the study. Baseline for each parameter in each NHP was determined by obtaining data for five daytime and night- time cycles. Following establishment of baseline, the NHPs were challenged via the aerosol route with EEEV V105 strain at a target dose of 7.0 log10 PFU (Fig 1). The NHPs received a virus dose ranging between 6.4–6.8 log10 PFU (Fig 2A). Following challenge, the NHPs were observed for signs of clinical disease and each NHP was assigned a score comprising of alter- ations in temperature, responsiveness, and neurological signs. NHPs with a score of ten or higher met the euthanasia criteria. All NHPs exhibited signs of clinical disease by ~48–72 hours post-infection (hpi) (S1 Fig). NHP #1 and #2 exhibited rapid increase in scores and met the euthanasia criteria by ~106–120 hpi (Figs 2B and S1). The two remaining NHPs displayed a similar initial rise in scores, followed by a transient decline and a rapid progression to severe disease at ~140 hpi (Figs 2B and S1). Various NHP tissues were collected at the time of euthanasia and EEEV was quantitated via plaque assay. Virus was titrated from samples including serum, plasma, brain (frontal cortex), olfactory bulb, cervical spinal cord, and heart of each NHP (Fig 3). Cerebrospinal fluid (CSF) was collected and titrated for NHPs #1 and #2 (Fig 3). Infectious virus was detected in serum and plasma of only NHP #1 with titers of ~3.5 and ~3.2 log10 PFU, respectively (Fig 3). In con- trast, EEEV was detected in brain and olfactory bulb in all four NHPs with titers ranging from ~4.1 to ~7.9 log10 PFU (Fig 3). The cervical spinal cord of NHP #2 and #4 had virus titers of ~7.6 and ~4.1 log10 PFU, respectively (Fig 3). Virus was detected in the heart tissue of only one NHP (NHP #3) with a titer of ~4.0 log10 PFU (Fig 3). Lastly, the virus was present in CSF of NHP #1 and #2 with titers of ~5.9 and ~5.5 log10 PFU, respectively (Fig 3). Alteration of animal behavior The NHPs were observed were continuously monitoring with limited disruptions due to human activity. This provided a rare opportunity to study the impact of EEEV infection on PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 6 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV Fig 1. Experimental design of the NHP study. All parameters were continuously and simultaneously measured throughout the study. Baseline was established for each parameter for an individual NHP by measuring 5 daytime (6 am-6 pm) and nighttime (6 pm– 6 am). All comparisons of post- challenge data was time-matched to pre-challenge baseline. https://doi.org/10.1371/journal.pntd.0009424.g001 animal behavior and determine the onset of neurological signs. The baseline behavior was determined for five days prior to challenge for each individual NHP for day and night times. Daytime and nighttime were defined as 6 am to 6pm and 6pm to 6 am, respectively. Alteration of animal behavior were evaluated by observing three parameters; sleep, activity, and food/ fluid consumption. Alteration of the circadian rhythm could be detected as early as 24 hpi in NHP #4 and ~42–54 hpi in the remaining three NHPs (S1 Table). The disruption of circadian rhythm was characterized by a decrease in sleep at night with a concomitant increase in night- time activity (S1 Table). The following daytime period was characterized by a decrease in day- time activity with short periods of sleep (S1 Table). However, following the onset of clinical signs, all NHPs exhibited a rapid progression to no or minimal sleep for the remainder of the study (S1 Table). Similar to sleep, food/fluid consumption also decreased between ~43–90 hpi in three of the four NHPs with minimal food/fluid consumption ~15–36 hrs prior to euthana- sia. Lastly, the onset of overt seizures was observed within last ~7 hrs of the study in three of the four NHPs (S1 Table). Neutralizing antibody response The presence of neutralizing antibodies was measured via PRNT80 at days -7, 0, and terminal time points (S2 Table). None of the NHPs had detectable neutralizing antibody titers prior to or at the time of challenge and were assigned the limit of detection of the assay (<1:20) (S2 Table). At the time of euthanasia, NHPs #1 and #2 did not have any detectable neutralizing PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 7 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV Fig 2. Administered aerosol EEEV dose (A) and survival of the NHPs (B). https://doi.org/10.1371/journal.pntd.0009424.g002 antibody response (S2 Table). In contrast, both NHP #3 and #4 had PRNT80 titers of 1:40 and 1:80, respectively (S2 Table). Temperature The baseline temperature was determined for each NHP for five days and data was analyzed in 0.5- (Fig 4A) and 12-hr (Fig 4B) intervals. All post-challenge comparisons were time-matched to pre-challenge baseline values. The average temperature prior to challenge ranged from ~36.3–38˚C (Fig 4A). Following challenge, an increase in temperature was observed within ~26–56 hpi in all four NHPs (Fig 4A, S3 Table). The onset of fever (>1.5˚C above baseline) was at ~53–61 hpi and it remained considerably elevated for duration of ~47–53 hrs (Fig 4A, S3 Table). The peak temperature ranged from 40.1–41˚C, and peak magnitude of fever ranged from ~3.0–4.2˚C and ~2.2–3.8˚C for 0.5- and 12-hr interval analyses, respectively (Fig 4A and 4B, S3 Table). In both analyses, NHPs displayed the highest elevation in temperature at night time (Fig 4A and 4B). Three of the NHPs exhibited hyperpyrexia (>3.0˚C above baseline) for a duration of ~11–22 hrs (Fig 4A and 4B). Following the sustained fever, a decline in tempera- ture was observed in the last ~10–24 hrs prior to euthanasia in all four NHPs (Fig 4A). NHPs #3 and 4 exhibited a rapid decline in temperature ~4–6 hrs prior to euthanasia with NHP #3 displaying a decline of 3.3˚C (Fig 4A). Respiration rate Similar to temperature, respiration rate was analyzed in 0.5- (Fig 5A) and 12-hr (Fig 5B) inter- vals. The baseline respiration rate for each NHP ranged from ~15–32 breaths per minute (bpm) (Fig 5A). Following challenge, an increase in respiration rate was observed starting at PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 8 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV Fig 3. Quantitation of infectious EEEV in NHP tissues collected at the time of euthanasia. The time of tissue collection for each NHP is provided in the x-axis. Limit of detection of plaque assay is 1.0 log10 PFU/mL or 1.0 log10 PFU/g and are indicated by a dashed line. https://doi.org/10.1371/journal.pntd.0009424.g003 24 hpi in all NHPs (Fig 5A and 5B). However, in contrast to temperature, the respiration rate displayed intermittent increase/decrease throughout first 96 hpi (Fig 5A). Sustained increase was observed at 100–112 hpi in all four NHPs, however the duration and magnitude of the increase differed considerably (Fig 5A). In NHPs #1 and #2, the sustained increase was observed ~3–8 hrs prior to euthanasia with peak respiration rate of 31–38 bpm, an increase of ~56–71% or ~17–22% in 0.5- and 12-hr interval analysis, respectively (Fig 5A and 5B). In con- trast, NHPs #3 and #4 experienced considerably higher peak respiration rate of ~49–50 bpm for a duration of ~32–40 hrs (Fig 5A and 5B). At peak, the percent increase in respiration rate in 0.5- and 12-hr interval analysis ranged from ~105–128% and ~85–95%, respectively (Fig 5A and 5B). Activity The activity of each NHP was analyzed in 6- (Fig 6A) and 12-hr (Fig 6B) intervals for both day- time and nighttime periods. Daytime activity differed markedly in the four NHPs. The average daytime activity for NHP #1 and #3 ranged from ~803–1704 units/6hrs, whereas the range for remaining two NHPs was ~410–505 units/6hrs (Fig 6A). Alteration in daytime activity could be observed within ~36–102 hpi with considerable decline in all NHPs at ~12–36 hrs prior to euthanasia (Fig 6A). The highest magnitude of decline was in NHP #1 and #3 with values of ~754–1303 units/6hrs, a decline of ~69–76% (Fig 6A). A 12-hr daytime interval analysis showed a sustained decline in activity ranging from ~64–73% in both NHPs (Fig 6B). NHP #2 PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 9 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV Fig 4. Body temperature of NHPs pre- and post-EEEV challenge. Data analysis is shown in 0.5- (A) and 12-hr (B) daytime/nighttime intervals. All NHPs were continuously monitored pre- and post-challenge. Pre-challenge baseline temperature was measured for five day/night cycles and a 0.5-hr interval baseline average was calculated by averaging raw data of five time-matched day or night time intervals. Forty-eight 0.5-hr interval averages were used to construct a baseline temperature for a day/night cycle and is shown as a grey line (A). Post-challenge values within �3 standard deviations (SD) are indicated with ( ). Hyperpyrexia is indicated by ( Temperature change in a 12-hr interval is shown in grey (daytime) and in black (nighttime) (B). All raw data from either 12-hr daytime or nighttime intervals were averaged and normalized. The last data point for NHP #3 and #4 is an average of 8 hrs (B). ), and >3 SD below baseline are indicated with ( ), >3 SD above baseline are indicated with ( ) (A). https://doi.org/10.1371/journal.pntd.0009424.g004 and #4 exhibited a decline in daytime activity ranging ~62–118 units/6hrs, a decline of ~15– 23% and a sustained decline of ~11–20% in the 12-hr interval analysis, respectively (Fig 6B). The baseline nighttime activity of all four NHPs was comparable, ranging from 333–383 units/6hrs (Fig 6A). Similar to daytime activity, a substantial increase in nighttime activity was observed as early as 24 hpi in NHP #4 and at 6–48 hrs prior to euthanasia in the other three NHPs (Fig 6A). Both 6- and 12-hr analysis showed an increase of ~5–22% and ~3–14%, respectively (Fig 6A and 6B). The considerable increase in nighttime activity was further cor- roborated via continuous monitoring (S1 Table). PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 10 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV Heart rate The baseline heart rate in NHPs ranged from ~60–150 bpm (Fig 7A). Following challenge, intermittent alterations were observed within ~24 hpi in all four NHPs (Fig 7A). Elevation in nighttime heart rate was observed within ~56 hpi followed by a return to normal daytime base- line values (Fig 7A). Sustained elevated heart rate was observed ~79–105 hpi that peaked at ~24–42 hrs prior to euthanasia with values of ~185–243 bpm (Fig 7A). The peak elevations from baseline ranged from ~67–190% and ~60–134% in 0.5- and 12-hr interval analysis, respectively (Fig 7A and 7B). Fig 5. Respiration rate of NHPs pre- and post-EEEV challenge. Data analysis is shown in 0.5- (A) and 12-hr (B) daytime/nighttime intervals. All NHPs were continuously monitored pre- and post-challenge. Pre-challenge baseline respiration rate was measured for five day/night cycles and a 0.5-hr interval baseline average was calculated by averaging raw data of five time-matched day or night time intervals. Forty-eight 0.5-hr interval averages were used to construct a baseline respiration rate for a day/night cycle and is shown as a grey line (A).Post-challenge values within �3 standard deviations (SD) are indicated with ( interval is shown in grey (daytime) and in black (nighttime) (B). All raw data from either 12-hr daytime or nighttime intervals were averaged and normalized. The last data point for NHP #3 and 4 is an average of 8 hrs (B). p-values �0.03 are indicated with �. bpm = breaths per minute. ), and >3 SD below baseline are indicated with ( ), >3 SD above baseline are indicated with ( ). Percent change from baseline in 12-hr https://doi.org/10.1371/journal.pntd.0009424.g005 PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 11 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV Fig 6. Measurement of activity in NHPs pre- and post-EEEV challenge. Data analysis is shown in 6- (A) and 12-hr (B) daytime/nighttime intervals. All NHPs were continuously monitored pre- and post-challenge. Pre-challenge baseline activity was measured for five day/night cycles and a 6-hr interval baseline average was calculated by averaging raw data of five time-matched day or night time intervals. Four 6-hr interval averages were used to construct a baseline activity for a day/night cycle and is shown as a grey line (A). Post-challenge values within �3 standard deviations (SD) are indicated with ( ), >3 SD above baseline are indicated with ( ). Percent change from baseline in 12-hr interval is shown in grey (daytime) and in black (nighttime) (B). All raw data from either 12-hr daytime or nighttime intervals were averaged and normalized. The last data point for NHP #3 and 4 is an average of 8 hrs (B). p-values �0.041 are indicated with �. ), and >3 SD below baseline are indicated with ( https://doi.org/10.1371/journal.pntd.0009424.g006 Blood pressure The baseline systolic and diastolic blood pressure values ranged from ~90–116 and ~58–87 mm Hg, respectively (Figs 8A and 9A). Following challenge, both systolic and diastolic pres- sure intermittently increased within 24 hpi followed by a sustained elevation at ~54–76 hpi for PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 12 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV Fig 7. Heart rate of NHPs pre- and post-EEEV challenge. Data analysis is shown in 0.5- (A) and 12-hr (B) daytime/nighttime intervals. All NHPs were continuously monitored pre- and post-challenge. Pre-challenge baseline heart rate was measured for five day/night cycles and a 0.5-hr interval baseline average was calculated by averaging raw data of five time-matched day or night time intervals. Forty-eight 0.5-hr interval averages were used to construct a baseline heart rate for a day/night cycle and is shown as a grey line (A).Post-challenge values within �3 standard deviations (SD) are indicated with ( baseline are indicated with ( and in black (nighttime) (B). All raw data from either 12-hr daytime or nighttime intervals were averaged and normalized. The last data point for NHP #3 and 4 is an average of 8 hrs (B). p-values �0.035 are indicated with �. bpm = beats per minute. ), >3 SD above ). Percent change from baseline in 12-hr interval is shown in grey (daytime) ), and >3 SD below baseline are indicated with ( https://doi.org/10.1371/journal.pntd.0009424.g007 a duration of ~48–80 hrs (Figs 8A and 9A). At peak, both systolic and diastolic pressures ran- ged from ~153–180 and ~102–128 mm Hg, respectively (Figs 8A and 9A). The peak systolic pressure values in 0.5-hr and 12-hr intervals represent percent increases of ~44–67% and ~35– 57%, respectively (Fig 8A and 8B). Similarly, peak diastolic pressure values in 0.5-hr and 12-hr analysis ranged from ~45–80% and ~38–65%, respectively (Fig 9A and 9B). PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 13 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV Fig 8. Systolic blood pressure of NHPs pre- and post-EEEV challenge. Data analysis is shown in 0.5- (A) and 12-hr (B) daytime/nighttime intervals. All NHPs were continuously monitored pre- and post-challenge. Pre-challenge baseline systolic blood pressure was measured for five day/night cycles and a 0.5-hr interval baseline average was calculated by averaging raw data of five time-matched day or night time intervals. Forty-eight 0.5-hr interval averages were used to construct a baseline systolic blood pressure for a day/night cycle and is shown as a grey line (A). Post-challenge values within �3 standard deviations (SD) are indicated with ( ). Percent change from baseline in 12-hr interval is shown in grey (daytime) and in black (nighttime) (B). All raw data from either 12-hr daytime or nighttime intervals were averaged and normalized. The last data point for NHP #3 and 4 is an average of 8 hrs (B). p-values �0.028 are indicated with �. ), and >3 SD below baseline are indicated with ( ), >3 SD above baseline are indicated with ( https://doi.org/10.1371/journal.pntd.0009424.g008 ECG Intermittent changes in ECG were observed in all four NHPs following challenge as early as 6–12 hpi (S2 and S3 Figs). As expected, the sustained elevation in heart rate at ~48–72 hpi led to reduction in QTc Bazett, RR, PR, and QRS duration (S2 and S3 Figs). However, 24 hrs prior to euthanasia multiple abnormalities were observed. An increase in QRS duration was observed for ~19 hrs with peak increase of ~11 msec in NHP #1 (S2 Fig). An increase in QTc Bazett was observed in three of the four NHPs (#1, 2, and 4) for duration of ~4–12 hrs with PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 14 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV Fig 9. Diastolic blood pressure of NHPs pre- and post-EEEV challenge. Data analysis is shown in 0.5- (A) and 12-hr (B) daytime/nighttime intervals. All NHPs were continuously monitored pre- and post-challenge. Pre-challenge baseline diastolic blood pressure was measured for five day/night cycles and a 0.5-hr interval baseline average was calculated by averaging raw data of five time-matched day or night time intervals. Forty-eight 0.5-hr interval averages were used to construct a baseline diastolic blood pressure for a day/night cycle and is shown as a grey line (A). Post-challenge values within �3 standard deviations (SD) are indicated with ( ). Percent change from baseline in 12-hr interval is shown in grey (daytime) and in black (nighttime) (B). All raw data from either 12-hr daytime or nighttime intervals were averaged and normalized. The last data point for NHP #3 and 4 is an average of 8 hrs (B). p-values �0.023 are indicated with �. ), and >3 SD below baseline are indicated with ( ), >3 SD above baseline are indicated with ( https://doi.org/10.1371/journal.pntd.0009424.g009 peak increase ranging from ~22–44 msec (S2 Fig). The ECG of NHPs #1, 3, and 4 displayed either intermittent or sustained elongation of PR duration (S3 Fig). NHP #4 displayed an elon- gation in PR duration for ~6 hrs with a peak increase of ~24 msec prior to euthanasia (S3 Fig). Alteration of T wave morphology and a decline (~1.5 mV) in the magnitude of QRS complex were both observed in NHP #2 (Fig 10). Lastly, both NHPs #1 and 4 displayed sinoatrial arrest in the last 24-hrs prior to euthanasia (Fig 10). PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 15 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV Fig 10. ECG abnormalities in EEEV infected NHPs 24 hrs prior to euthanasia. Representative ~5 second intervals of pre- and post-EEEV challenge are shown. https://doi.org/10.1371/journal.pntd.0009424.g010 Quantitative EEG The EEG data for each NHP pre- and post-challenge are shown as heat maps (Figs 11 and 12). Six night and five day 12-hr intervals were used to generate a pre-challenge heat map for each NHP. In all four NHPs, the baseline heat map displayed similar patterns with the majority of values between -50 to +50%. An increase in gamma waves were observed during the daytime (generally around anticipated feeding periods and increased human activity due to arrival of NHPs in biocontainment and one day prior to challenge) in all four NHPs during the pre-chal- lenge period attributed to electromyographic artifacts. However, post- challenge distinct pat- terns could be observed rapidly in all NHPs. NHP #1 displayed an increase in the individual frequencies comprising of the beta and gamma power bands at nighttime within ~15 hr post-challenge with maximal increases of ~1,300% and ~5,200%, respectively (Fig 11). The next ~6–10 hrs were characterized by PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 16 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV Fig 11. Pre- and post-EEEV challenge quantitative electroencephalography (qEEG) heat maps of NHPs #1 and #2. The top and bottom x-axes display brain waves [(delta (δ), theta (θ), alpha (α), sigma (σ), and gamma (γ)] and frequency in hertz (Hz), respectively. The left and right y-axes display time (hr) and 12-hr day/night time intervals, respectively. 12-hr nighttime is indicated by (▐ ) and daytime by a gap. Pre-EEEV challenge baseline (left) and post-EEEV challenge (right) heat maps are shown. https://doi.org/10.1371/journal.pntd.0009424.g011 increases of up to ~800, 400, and 600% in the theta, alpha, and sigma bands, respectively, dur- ing the daytime while the NHP was awake. These increases were followed by a decline in all frequencies to near or below basal levels for ~6 hrs during the daytime (Fig 11). This pattern continued into the second night with a general increase of up to ~2,700% across all individual PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 17 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV frequencies within the gamma band and ~1,200% across all other frequencies (Fig 11). At 48 hpi, a prolonged decline of up to 93% was observed across all frequencies during the daytime (Fig 11). The third night was characterized by intermittent increases in high frequencies within the gamma band, however, a sustained increase of up to ~250% in all other brain waves was observed including low gamma frequencies (Fig 11). An increase of up to 6,800% in gamma frequencies was observed in the first ~3 hrs of the next daytime, followed by a sustained decline to below time-matched basal values (Fig 11). The decrease in gamma frequencies was accompanied by a simultaneous increase in theta, alpha, sigma, and low beta frequencies (in an awake NHP) with the alpha frequencies displaying the most pronounced increase with val- ues exceeding 1,000% (Fig 11). The fourth night displayed similar patterns as the previous night with sustained increase in alpha frequencies of ~600% (Fig 11). The last daytime for this NHP was characterized by a considerable and sustained decline in delta and gamma bands to at or below basal values, with a simultaneous increase in theta, alpha, sigma, and beta frequen- cies (Fig 11). The magnitude of the latter waves ranged from ~80–1,400%, with alpha band dis- playing the largest increases (Fig 11). NHP #2 displayed a similar pattern as NHP #1 with an early increase in the individual beta and gamma frequencies within ~12 hpi ranging from ~300 to 3,000% (Fig 11). This was fol- lowed by an elevation in delta, theta, alpha, and sigma frequencies during the daytime with increases of up to ~100 to 700% relative to time-matched baseline values (Fig 11). Concomi- tant to the rise in low frequencies, a decline to at or near baseline values was observed in both beta and gamma bands (Fig 11). This pattern continued into 24-hr period, however, the daytime displayed a sustained decline of both beta and gamma bands followed by an increase of up to ~750% in beta waves (Fig 11). The third night was characterized by increases in alpha, sigma, beta, and high gamma frequencies with increases of up to 2,100% in an awake NHP (Fig 11). The next daytime exhibited a prolonged decline in delta, theta, and gamma bands with a maximum reduction of nearly 99% from individual basal values (Fig 11). In contrast, sigma and alpha frequencies displayed increases of up to 660% (Fig 11). The fourth nighttime period exhibited a similar pattern to the previous nighttime. The next day- time was characterized by prolonged and considerable decline of all individual spectral fre- quencies to below baseline values (Fig 11). The last nighttime exhibited similar pattern to the two previous night times. Five hours prior to euthanasia was characterized by decline in all brain waves except delta and theta waves which had slight increases from ~42 to 370% in an awake NHP. NHP #3 predominately displayed an increase in gamma frequencies in the first three nights and two daytimes with values exceeding 1,000% as compared to the time-matched baseline val- ues (Fig 12). However, the third night also exhibited a rise in alpha, sigma, and beta frequen- cies not exceeding a 250% increase (Fig 12). The following daytime was predominately characterized by a prolonged and considerable decline in all individual frequencies to below time-matched baseline values. The subsequent evening and night were largely characterized by increases in alpha and gamma frequencies followed by an elevation in all frequencies except delta waves (Fig 12). The next daytime displayed a substantial reduction in all frequencies followed by a fifth nighttime with considerable increases in beta and gamma frequencies (Fig 12). The last day for NHP #3 was characterized by rise in theta, alpha, sigma, and beta waves in an awake NHP followed by a decline in all waves (Fig 12). The last nighttime showed a prolonged rise in all waves particularly in delta, theta, alpha, and sigma, in an awake NHP (Fig 12). NHP #4 displayed an increase in both beta and gamma frequencies 12 hpi that continued for most of the nighttime and daytime following challenge (Fig 12). Surprisingly, elevation in delta, theta, alpha, and sigma were also observed within 12 hpi and were sustained until PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 18 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV Fig 12. Pre- and post-EEEV challenge quantitative electroencephalography (qEEG) heat maps of NHPs #3 and #4. The top and bottom x-axes display brain waves [(delta (δ), theta (θ), alpha (α), sigma (σ), and gamma (γ)] and frequency in hertz (Hz), respectively. The left and right y-axes display time (hr) and 12-hr day/night time intervals, respectively. 12-hr nighttime is indicated by (▐ ) and daytime by a gap. Pre-EEEV challenge baseline (left) and post-EEEV challenge (right) heat maps are shown. https://doi.org/10.1371/journal.pntd.0009424.g012 108–120 hpi with peak increases up to 1,800% in an awake NHP (Fig 12). This sustained increase over baseline was followed by a precipitous decline to below time-matched basal levels in frequencies from all four power bands from 108–136 hpi (Fig 12). Several hours prior to euthanasia, an increase in both delta and theta frequencies was observed (Fig 12). PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 19 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV Fig 13. Timeline of events for NHP #1 3 hrs prior to euthanasia and profile of seizure events. Percent change in EEG waves with +/- standard deviations (error bars) are shown. https://doi.org/10.1371/journal.pntd.0009424.g013 Clinical parameters of nhp #1 three hours prior to a cardiac event NHP#1 had experienced considerable alterations of many important physiological parameters (temperature, respiration, heart rate, blood pressure, ECG, and EEG) ~24–80 hrs prior to euthanasia. In addition, there was substantial decline in sleep and food/fluid consumption ~40–52 hrs prior to euthanasia (S1 Table). During the last 3-hrs prior to euthanasia, NHP#1 experienced two seizures ~2 hrs apart (Fig 13). Both EEG seizure profiles showed involvement of all brain waves with increases ranging from ~99% to ~368% (Fig 13). The time frame of events occurred at the shift from daytime to nighttime, when the baseline values for each parameter would naturally decline in healthy animals (Fig 14). Following the onset of the first seizure, many of the parameters remained elevated and/or increased relative to baseline. The comparison of peak respiration and heart rates to baseline ranged from 31 vs. 18 bpm and 187 vs. 65 bpm, respectively (Fig 14A and 14B). Similar peak increases were also observed for sys- tolic and diastolic blood pressure relative to baseline with 180 vs.103 mmHg and 121 vs. 67 mmHg, respectively (Fig 14C and 14D). These elevations in parameters represent percent PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 20 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 21 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV Fig 14. Alteration in respiration, heart rate, and systolic and diastolic blood pressure of NHP #1 3 hrs prior to euthanasia. 0.5-hr interval analysis (left) and percent change from baseline (right). The average for each time-matched 0.5 hr interval in shown in grey. https://doi.org/10.1371/journal.pntd.0009424.g014 increase from baseline of 71% (respiration rate), 190% (heart rate), 55% (systolic blood pres- sure), and 80% (diastolic blood pressure) (Fig 14). Following the onset of two seizures, NHP #1 experienced a cardiac event characterized by non-sustained ventricular tachycardia, fol- lowed by sustained ventricular tachycardia, and finally ventricular fibrillation (Fig 15). Imme- diately following the cardiac event, the NHP was euthanized. Discussion The susceptibility of cynomolgus macaques to North American lineage of EEEV via the aerosol route has been explored previously with two isolates, FL91-4679 and FL-93939. Both isolates were obtained from mosquito pools; Aedes albopictus (FL91-4679) and Culiseta melanura (FL- 93939) [14,15]. Aerosol exposure to either isolate at 107 PFU produced uniform disease and NHPs met the euthanasia criteria within ~96–130 hpi [14,15]. In this study, we utilized an iso- late from the brain tissue of a fatal human case in Massachusetts in 2005 [16,18]. The data from our study are in agreement with two previous studies with all four NHPs meeting eutha- nasia criteria by ~106–140 hpi [14,15]. Taken together, all three studies demonstrate that low passage isolates of EEEV-NA lineage, irrespective of the isolation from mosquito and human hosts, are uniformly lethal at 107 PFU dose via the aerosol route. Two decades ago, Pratt and colleagues successfully utilized telemetry in biocontainment to demonstrate continuous monitoring of temperature in NHPs following VEEV challenge [19]. This success led to incorporation of temperature in NHP studies for category A and B patho- gens including Ebola, Marburg, VEEV, EEEV, and WEEV [14,15,19–29]. However, the tech- nology of telemetry implants has advanced considerably and is now able to monitor many important physiological parameters continuously and simultaneously in biocontainment. The current devices can measure physiological parameters at 1–1,000 Hz and produce enormous data sets ranging from 1,800 to 1,800,000 data points every 30 mins to describe a given param- eter. Accordingly, the technology has substantial potential to improve animal model develop- ment particularly for Risk Group 3 and 4 agents. This is the first of its kind study in biocontainment to investigate clinical disease course of EEEV by implanting multiple devices in a single NHP to continuously and simultaneously monitor temperature, activity, respira- tion, heart rate, blood pressure, ECG, and EEG. All physiological parameters were altered con- siderably post-challenge and all four NHPs met the euthanasia criteria rapidly. However, the onset and sustained duration of each parameter differed considerably. Surprisingly, EEG was the earliest parameter to change within ~12–36 hpi in all four NHPs, followed by temperature, blood pressure, and activity/clinical signs at ~48–72 hpi, and all others �72 hpi. In previous studies, only temperature and heart rate parameters have been explored [14,15]. However, the heart rate data was limited as minimal baseline day/night time data was provided and partial post-challenge data was reported [14,15]. In both previous studies, the onset of sustained fever occurred ~48 hpi with peak increase in temperature ranging from 1.8– 3.5˚C, followed by a rapid decline prior to euthanasia [14,15]. The onset of sustained elevated heart rate was between ~40–72 hpi with peak heart rate increase of ~40–130 bpm [14,15]. Our temperature and heart rate data are in agreement with both of the previous studies. Cynomolgus macaques are a prey species and to avoid the potential confounding effects of prey response elicited by cage-side human observations, we utilized 24-hr continuous remote monitoring to gain greater insight into alteration of macaque behavior and clinical PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 22 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 23 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV Fig 15. ECG profile of the cardiac event in NHP #1. Representative baseline and cardiac event ECG graphs ~5 seconds in duration are shown. https://doi.org/10.1371/journal.pntd.0009424.g015 manifestations post-EEEV infection. Disturbances in the circadian rhythm post-challenge were observed as early as 24 hpi. The NHPs were observed staying awake and alert at night time with a concomitant decrease in daytime activity accompanied by short periods of sleep. The alteration in sleep/wake cycle was also accompanied by decrease in fluid and food intake, followed by a rapid progression to minimal sleep, activity, and food/fluid consumption. The substantial alteration of these parameters is likely to considerably exacerbate the observed clin- ical signs in the terminal phase of infection. The weakness and lack of coordination observed in the last 24 hrs prior to euthanasia may be in part due to the lack of nutrients, electrolyte imbalances, and lack of sleep. Lastly, the continuous remote observation of the NHPs enabled the study to accurately measure the onset of neurological disease. Seizures were observed in three of the four NHPs, however, the on-set of seizures was very late in the disease, ~1–3 hrs prior to euthanasia. Taken together, these data demonstrate that remote monitoring of animals can substantially enhance the understanding of natural disease progression and should be incorporated in the objective assessment of clinical disease in future NHP studies. This study investigated ECG by measuring QTc Bazett, PR, RR, and QRS duration. These intervals are dependent on the heart rate. As the heart rate increases or decreases the intervals get shorter or longer, respectively [30]. Alterations in heart rate were observed as early as ~24 hpi and sustained increases were maintained through ~79–140 hpi. Consequently, the decrease in the intervals are consistent with an increase in heart rate. A recent study measured QT and RR intervals following EEEV infection via aerosol route [31]. A decrease in both intervals was observed following onset of severe disease [31]. Our data are in agreement with this study. However, there were ECG abnormalities detected within the last 24 hrs prior to euthanasia. The NHPs displayed ECG abnormalities consisting of alterations in QRS and PR duration, QTc Bazett, T wave morphology, amplitude of the QRS complex, and sinoatrial arrest. These abnormalities are indicative of electrical conductivity issues in the heart and are associated with ventricular arrhythmias [32]. In addition to these abnormalities, NHP #1 experienced a critical cardiac event leading immediate euthanasia. This is the first evidence of life-threaten- ing critical cardiac events as a consequence of EEEV infection. There are several potential explanations for the cardiac abnormalities. First, the abnormalities may be due to EEEV infec- tion of the myocardium and/or the pericardium [33–39]. EEEV infection of the myocardium and subsequent degeneration of the spontaneously contracting cardiac muscle tissue has been reported in equines, swine, and humans [40–42]. Second, the abnormalities could be due to host inflammatory responses resulting in myocarditis and/or pericarditis [33–39]. QRS and QT prolongation, ventricular arrhythmias, and T wave morphology changes have been reported for viral infections such as Coxsackievirus, HIV, influenza A, HSV, and adenovirus [32–39]. Third, the EEEV may target important autonomic control center such as the hypo- thalamus, thalamus, and medulla oblongata and thus interfere with electrical conductivity of the heart. Fourth, the electrolyte imbalance due to lack of food and fluid intake prior to meet- ing the euthanasia criteria may exacerbate any of the explanations outlined above. Many human EEEV infection cases are often misdiagnosed and/or diagnosed at the onset of severe symptoms, and accordingly a detailed clinical disease course is not available for human infection [40,43–57]. One important goal of animal model development is to gain insights into progression of brain abnormalities leading to encephalitis. To achieve this goal, we investigated EEG as a potential tool enabling continuous monitoring of the brain electrical activity following challenge. EEG has been utilized previously to monitor human EEEV, PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 24 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV WEEV, and VEEV infections [43–45,47,49–51,53,54,56,58–61]. The limited human EEG data shows diffuse slowing particularly of delta, theta, and alpha and gamma waves [43–45,47,49– 51,53,54,56,58–61]. Similar to human data, severe diffuse slowing was observed in all four NHPs post-EEEV challenge. The data in the present study is in agreement with previous human EEG data. However, in addition to diffuse slowing, other gross abnormalities were also observed. First, brain waves of all four NHPs displayed rapid and extreme fluctuations. In NHP #3, a profound decrease in all brain frequencies of up to nearly 99% was observed at ~94–108 hpi, followed by an increase of over 600, 650, 2,500 and 6,300% in alpha, sigma, beta, and gamma frequencies, respectively, for a duration of ~4–12 hrs. Similarly, NHP #4 displayed an increase of ~230 to 3,500% in delta, theta, alpha, and/or sigma frequencies at ~84–108 hpi followed by a near complete decline in all four waves for a duration of ~30 hr. Second, there was a profound decrease in gamma frequencies of nearly 99% during the daytime that contin- ued for up to ~10–15 hrs. Third, the presence of brain waves associated with sleep (delta, theta, and alpha waves) in awake NHPs was observed. All three brain abnormalities are a sign of sig- nificant brain injury and have been observed in human cases of viral encephalitis such as Japa- nese encephalitis virus and Herpes simplex virus [62–67]. Further studies are underway to characterize these profound alterations in the brain waves as well as the underlying mecha- nism/s responsible for these abnormalities. Traumatic brain injury (TBI) is defined as a non-degenerative and non-congenital insult to the brain that results in temporary or permanent impairment of cognitive and physical func- tions [68]. All four NHPs exhibited many signs associated with TBI such as disturbances in cir- cadian rhythm, food/fluid consumption, inability to initiate or maintain a normal sleep pattern, decrease overall activity, increased slow wave (delta, theta, and alpha) activity while awake, and neurological signs [68–72]. These data strongly suggest that EEEV infection via aerosol route can rapidly (~12–70 hpi) induce many signs of severe TBI. The potential mecha- nism/s that underlie such rapid induction of TBI signs require further exploration. The potential route of virus dissemination following aerosol infection has not been explored previously [14,15]. The infectious virus in the present study was either minimal or not detected in the periphery. In contrast, high level of infectious virus (>6.0 log10 PFU/g) was detected in the olfactory bulb and the central nervous system (CNS) in all animals at the time of euthana- sia. In addition, the physiological parameters measured in this study such as heart and respira- tion rates, blood pressure, temperature, sleep/wake cycle, and hunger/satiety are controlled by the autonomic nervous system (ANS). The rapid (~24–50 hpi) and considerable alteration of these functions suggest that EEEV infection via the aerosol route likely enables direct access to the ANS via the neuronal projections between the olfactory bulb and the critical control cen- ters such as the thalamus, hypothalamus, and the brainstem. Taken together these data suggest that following aerosol infection the virus spreads and infects the CNS and ANS producing injury to substantially disrupt animal behavior and the control of critical physiological param- eters to produce severe disease. However, this hypothesis requires further investigation to elu- cidate the potential mechanism. The clinical scores in of NHPs displayed two distinct patterns, a rapid rise (NHPs #1–2) or biphasic pattern with an initial rise followed by a decline (NHPs #3 and 4). The clinical score is comprised of neurological disease, temperature, and responsiveness. The initial rise in clinical scores of NHPs #3 and 4 were mainly due to alteration in both temperature and responsive- ness. However, the temperature declined between 90–110 hpi and the responsiveness improved to yield lower clinical scores. Nonetheless, both NHPs rapidly progressed to meet the euthanasia criteria (clinical score = 10). Surprisingly, the temperature rapidly declined to below baseline and both NHPs exhibited hypothermia ~1–3 hrs prior to euthanasia. These data suggest both NHPs likely were unable to regulate body temperature and support the PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 25 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV hypothesis of ANS dysregulation above. The data also highlight the rapid progression to termi- nal disease of EEEV infected NHPs. Neutralizing antibodies are generally considered to be a correlate of protection against alphavirus infection [73–78]. However, two NHPs in this study had neutralizing antibody at the time of euthanasia. One potential explanation is that EEEV infection and subsequent injury to the brain tissue is rapid and profound via the aerosol route, and by the time the neutralizing antibody response is generated it has minimal or no efficacy. This is supported by previous vaccine studies in cynomolgus macaques with similar rapid aerosol infection kinetics of EEEV [13,15]. In both studies, NHPs generated neutralizing antibody responses post-vaccination and yet met the euthanasia criteria following challenge [13,15]. Previous studies in cynomolgus macaques with lethal aerosol EEEV have focused mainly on five parameters (temperature, virus quantitation, hematological parameters, clinical dis- ease, and lethality) for NHP model development [13–15]. This study provides six additional parameters for countermeasure evaluation including activity, respiration and heart rates, blood pressure, ECG, and EEG in a lethal challenge model. For this first study, we utilized the most lethal of the three encephalitic alphaviruses, EEEV. Whether a similar alteration of these parameters following aerosol challenge in NHPs can be produced with WEEV and VEEV or a sub-lethal EEEV dose requires further investigation. The FDA approval of medical countermeasures for Risk Group 3 and 4 agents will rely on the U. S. Food and Drug Administration’s Animal Rule (21 CFR 601.90), which allows the utilization of animals in pivotal efficacy studies to support licensure in lieu of human efficacy studies. The mea- surement of clinical parameters via telemetry offers several advantages for animal model develop- ment in this regard. First, the technology enables measurements of important clinical parameters that are relevant to humans. Second, it enables identification of multiple parameters for rapid animal model development and countermeasure evaluation in event of a natural outbreak and/or bioterror event. The additional parameters may be particularly appealing for investigating and/or refining ani- mal model development for partially lethal or non-lethal agents such as WEEV and MERS. Further- more, the high sampling rate of the devices enables accurate and real-time (hourly/daily) evaluation of countermeasures. Third, it can identify potential side effects of a countermeasure prior to its utili- zation in human clinical trials. These are substantial advantages of utilization of advanced telemetry in NHP animal models for Risk Group 3 and 4 agents and require further investigation. In summary, we utilized state of the art telemetry to investigate important physiological parameters in the cynomolgus macaque model following EEEV aerosol challenge. All parame- ters measured, including 6 never-before examined including brain waves, were substantially and rapidly (~12 hpi) altered post-challenge. These alterations were the earliest documented signs of disease that were not readily observable without the use of physiological radio-telemetry devices, and add possible additional endpoints to future efficacy experiments for EEEV in an NHP model. This is the first detailed disease course of EEEV in an NHP model and the parame- ters identified will improve animal model development and countermeasure evaluation. Disclosure statement The views expressed in this article are those of the authors and do not reflect the official policy or position of the U.S. Department of Defense, or the Department of the Army. Supporting information S1 Fig. Clinical scores of NHPs post-EEEV challenge. Following aerosol challenge all NHPs were monitored daily and NHPs with a total score �10 met the euthanasia criteria. (TIF) PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 26 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV S2 Fig. QRS duration and QTc Bazett in NHPs pre- and post-EEEV challenge. Pre-chal- lenge baseline QRS duration and QTc Bazett are shown in grey (A). All NHPs were continu- ously monitored pre- and post-challenge. Pre-challenge baseline QRS duration and QTc Bazett were measured for five day/night cycles and a 0.5-hr interval baseline average was calcu- lated by averaging raw data of five time-matched day or night time intervals. Forty-eight 0.5-hr interval averages were used to construct baselines for QRS duration and QTc Bazett for a day/night cycle and are shown as a grey line (A). Post-challenge values within �3 standard deviations (SD) are indicated with ( ), and >3 SD below baseline are indicated with ( (TIF) ), >3 SD above baseline are indicated with ( ). S3 Fig. RR and PR duration in NHPs pre- and post-EEEV challenge. Pre-challenge baseline RR and PR duration are shown in grey (A). All NHPs were continuously monitored pre- and post-challenge. Pre-challenge baseline RR and PR duration were measured for five day/night cycles and a 0.5-hr interval baseline average was calculated by averaging raw data of five time- matched day or night time intervals. Forty-eight 0.5-hr interval averages were used to con- struct baselines for RR and PR duration for a day/night cycle and are shown as a grey line (A). Post-challenge values within �3 standard deviations (SD) are indicated with ( above baseline are indicated with ( (TIF) ), and >3 SD below baseline are indicated with ( ), >3 SD ). S1 Table. Alteration in NHP behavior and onset of neurological signs post-EEEV chal- lenge. NHP behavior comprised of food/fluid intake, sleep, activity, and onset of seizures were monitored pre- and post-EEEV challenge. # = modest decline, ## = moderate decline, and ### = severe decline. " = modest increase. (TIF) S2 Table. Neutralizing antibody response in terminal samples. Neutralizing antibody was measured via PRNT80 assay. The limit of detection in PRNT80 assay is indicated by italic font (<1:20). All samples were analyzed three times in the PRNT80 assay. (TIF) S3 Table. Summary of fever data in NHPs. Fever hours is calculated as the sum of the signifi- cant temperature elevations. ΔTmax = maximum change in temperature. (TIF) Author Contributions Conceptualization: William D. Pratt, Margaret L. Pitt, Farooq Nasar. Formal analysis: Franco D. Rossi, Michael V. Accardi, Simon Authier, William D. Pratt. Funding acquisition: Margaret L. Pitt, Farooq Nasar. Investigation: John C. Trefry, Franco D. Rossi, Michael V. Accardi, Brandi L. Dorsey, Thomas R. Sprague, Suzanne E. Wollen-Roberts, Joshua D. Shamblin, Adrienne E. Kimmel, Lynn J. Miller, Anthony P. Cardile, Darci R. Smith, Sina Bavari, Simon Authier, William D. Pratt, Farooq Nasar. Methodology: John C. Trefry, Franco D. Rossi, Michael V. Accardi, Simon Authier, William D. Pratt, Farooq Nasar. Project administration: John C. Trefry, Margaret L. Pitt, Farooq Nasar. PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 27 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV Resources: Pamela J. Glass, Crystal W. Burke. Supervision: Margaret L. Pitt, Farooq Nasar. Writing – original draft: John C. Trefry, Farooq Nasar. Writing – review & editing: John C. Trefry, Franco D. Rossi, Michael V. Accardi, Brandi L. Dorsey, Thomas R. Sprague, Suzanne E. Wollen-Roberts, Joshua D. Shamblin, Adrienne E. Kimmel, Pamela J. Glass, Lynn J. Miller, Crystal W. Burke, Anthony P. Cardile, Darci R. Smith, Sina Bavari, Simon Authier, William D. Pratt, Margaret L. Pitt, Farooq Nasar. References 1. Forrester NL, Palacios G, Tesh RB, Savji N, Guzman H, Sherman M, et al. Genome-scale phylogeny of the alphavirus genus suggests a marine origin. J Virol. 2012; 86(5):2729–38. https://doi.org/10.1128/ JVI.05591-11 PMID: 22190718; PubMed Central PMCID: PMC3302268. 2. La Linn M, Gardner J, Warrilow D, Darnell GA, McMahon CR, Field I, et al. Arbovirus of marine mam- mals: a new alphavirus isolated from the elephant seal louse, Lepidophthirus macrorhini. J Virol. 2001; 75(9):4103–9. https://doi.org/10.1128/JVI.75.9.4103-4109.2001 PMID: 11287559; PubMed Central PMCID: PMC114155. 3. Nasar F, Palacios G, Gorchakov RV, Guzman H, Da Rosa AP, Savji N, et al. Eilat virus, a unique alpha- virus with host range restricted to insects by RNA replication. Proc Natl Acad Sci U S A. 2012; 109 (36):14622–7. https://doi.org/10.1073/pnas.1204787109 PMID: 22908261; PubMed Central PMCID: PMC3437828. 4. Villoing S, Bearzotti M, Chilmonczyk S, Castric J, Bremont M. Rainbow trout sleeping disease virus is an atypical alphavirus. J Virol. 2000; 74(1):173–83. https://doi.org/10.1128/jvi.74.1.173-183.2000 PMID: 10590104; PubMed Central PMCID: PMC111526. 5. Weston JH, Welsh MD, McLoughlin MF, Todd D. Salmon pancreas disease virus, an alphavirus infect- ing farmed Atlantic salmon, Salmo salar L. Virology. 1999; 256(2):188–95. https://doi.org/10.1006/viro. 1999.9654 PMID: 10191183. 6. Fields BN, Knipe DM, Howley PM. Fields virology. 6th ed. Philadelphia: Wolters Kluwer Health/Lippin- cott Williams & Wilkins; 2013. 7. Arrigo NC, Adams AP, Weaver SC. Evolutionary patterns of eastern equine encephalitis virus in North versus South America suggest ecological differences and taxonomic revision. J Virol. 2010; 84 (2):1014–25. Epub 2009/11/06. https://doi.org/10.1128/JVI.01586-09 PMID: 19889755; PubMed Cen- tral PMCID: PMC2798374. 8. Chen R, Mukhopadhyay S, Merits A, Bolling B, Nasar F, Coffey LL, et al. ICTV Virus Taxonomy Profile: Togaviridae. J Gen Virol. 2018; 99(6):761–2. Epub 2018/05/11. https://doi.org/10.1099/jgv.0.001072 PMID: 29745869. 9. Scott TW, Weaver SC. Eastern equine encephalomyelitis virus: epidemiology and evolution of mosquito transmission. Adv Virus Res. 1989; 37:277–328. Epub 1989/01/01. https://doi.org/10.1016/s0065-3527 (08)60838-6 PMID: 2574935. 10. Lindsey NP, Staples JE, Fischer M. Eastern Equine Encephalitis Virus in the United States, 2003– 2016. Am J Trop Med Hyg. 2018; 98(5):1472–7. Epub 2018/03/21. https://doi.org/10.4269/ajtmh.17- 0927 PMID: 29557336; PubMed Central PMCID: PMC5953388. 11. Hawley RJ, Eitzen EM Jr., Biological weapons—a primer for microbiologists. Annu Rev Microbiol. 2001; 55:235–53. Epub 2001/09/07. https://doi.org/10.1146/annurev.micro.55.1.235 PMID: 11544355. 12. Ecker DJ, Sampath R, Willett P, Wyatt JR, Samant V, Massire C, et al. The Microbial Rosetta Stone Database: a compilation of global and emerging infectious microorganisms and bioterrorist threat agents. BMC Microbiol. 2005; 5:19. Epub 2005/04/27. https://doi.org/10.1186/1471-2180-5-19 PMID: 15850481; PubMed Central PMCID: PMC1127111. 13. Reed DS, Glass PJ, Bakken RR, Barth JF, Lind CM, da Silva L, et al. Combined alphavirus replicon par- ticle vaccine induces durable and cross-protective immune responses against equine encephalitis viruses. J Virol. 2014; 88(20):12077–86. https://doi.org/10.1128/JVI.01406-14 PMID: 25122801; PubMed Central PMCID: PMC4178741. 14. Reed DS, Lackemeyer MG, Garza NL, Norris S, Gamble S, Sullivan LJ, et al. Severe encephalitis in cynomolgus macaques exposed to aerosolized Eastern equine encephalitis virus. J Infect Dis. 2007; 196(3):441–50. https://doi.org/10.1086/519391 PMID: 17597459. 15. Roy CJ, Adams AP, Wang E, Leal G, Seymour RL, Sivasubramani SK, et al. A chimeric Sindbis-based vaccine protects cynomolgus macaques against a lethal aerosol challenge of eastern equine PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 28 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV encephalitis virus. Vaccine. 2013; 31(11):1464–70. https://doi.org/10.1016/j.vaccine.2013.01.014 PMID: 23333212; PubMed Central PMCID: PMC3581708. 16. National Research Council (U.S.). Committee for the Update of the Guide for the Care and Use of Labo- ratory Animals., Institute for Laboratory Animal Research (U.S.), National Academies Press (U.S.). Guide for the care and use of laboratory animals. 8th ed. Washington, D.C.: National Academies Press; 2011. xxv, 220 p. p. 17. Yu GY, Wiley MR, Kugelman JR, Ladner JT, Beitzel BF, Eccleston LT, et al. Complete coding sequences of eastern equine encephalitis virus and venezuelan equine encephalitis virus strains iso- lated from human cases. Genome Announc. 2015; 3(2). Epub 2015/04/25. https://doi.org/10.1128/ genomeA.00243-15 PMID: 25908124; PubMed Central PMCID: PMC4408325. 18. Centers for Disease C, Prevention. Eastern equine encephalitis—New Hampshire and Massachusetts, August-September 2005. MMWR Morb Mortal Wkly Rep. 2006; 55(25):697–700. Epub 2006/07/01. PMID: 16810146. 19. Pratt WD, Gibbs P, Pitt ML, Schmaljohn AL. Use of telemetry to assess vaccine-induced protection against parenteral and aerosol infections of Venezuelan equine encephalitis virus in non-human pri- mates. Vaccine. 1998; 16(9–10):1056–64. Epub 1998/07/31. https://doi.org/10.1016/s0264-410x(97) 00192-8 PMID: 9682359. 20. Reed DS, Larsen T, Sullivan LJ, Lind CM, Lackemeyer MG, Pratt WD, et al. Aerosol exposure to west- ern equine encephalitis virus causes fever and encephalitis in cynomolgus macaques. J Infect Dis. 2005; 192(7):1173–82. https://doi.org/10.1086/444397 PMID: 16136459. 21. Reed DS, Lind CM, Lackemeyer MG, Sullivan LJ, Pratt WD, Parker MD. Genetically engineered, live, attenuated vaccines protect nonhuman primates against aerosol challenge with a virulent IE strain of Venezuelan equine encephalitis virus. Vaccine. 2005; 23(24):3139–47. Epub 2005/04/20. https://doi. org/10.1016/j.vaccine.2004.12.023 PMID: 15837213. 22. Reed DS, Lind CM, Sullivan LJ, Pratt WD, Parker MD. Aerosol infection of cynomolgus macaques with enzootic strains of venezuelan equine encephalitis viruses. J Infect Dis. 2004; 189(6):1013–7. https:// doi.org/10.1086/382281 PMID: 14999604. 23. Ewers EC, Pratt WD, Twenhafel NA, Shamblin J, Donnelly G, Esham H, et al. Natural History of Aerosol Exposure with Marburg Virus in Rhesus Macaques. Viruses. 2016; 8(4):87. Epub 2016/04/05. https:// doi.org/10.3390/v8040087 PMID: 27043611; PubMed Central PMCID: PMC4848582. 24. Trefry JC, Wollen SE, Nasar F, Shamblin JD, Kern SJ, Bearss JJ, et al. Ebola Virus Infections in Nonhu- man Primates Are Temporally Influenced by Glycoprotein Poly-U Editing Site Populations in the Expo- sure Material. Viruses. 2015; 7(12):6739–54. Epub 2015/12/26. https://doi.org/10.3390/v7122969 PMID: 26703716; PubMed Central PMCID: PMC4690892. 25. Pratt WD, Davis NL, Johnston RE, Smith JF. Genetically engineered, live attenuated vaccines for Vene- zuelan equine encephalitis: testing in animal models. Vaccine. 2003; 21(25–26):3854–62. Epub 2003/ 08/19. https://doi.org/10.1016/s0264-410x(03)00328-1 PMID: 12922119. 26. Langsjoen RM, Haller SL, Roy CJ, Vinet-Oliphant H, Bergren NA, Erasmus JH, et al. Chikungunya Virus Strains Show Lineage-Specific Variations in Virulence and Cross-Protective Ability in Murine and Nonhuman Primate Models. MBio. 2018; 9(2). Epub 2018/03/08. https://doi.org/10.1128/mBio.02449- 17 PMID: 29511072; PubMed Central PMCID: PMC5844994. 27. Erasmus JH, Auguste AJ, Kaelber JT, Luo H, Rossi SL, Fenton K, et al. A chikungunya fever vaccine utilizing an insect-specific virus platform. Nat Med. 2017; 23(2):192–9. Epub 2016/12/20. https://doi.org/ 10.1038/nm.4253 PMID: 27991917; PubMed Central PMCID: PMC5296253. 28. Rossi SL, Russell-Lodrigue KE, Killeen SZ, Wang E, Leal G, Bergren NA, et al. IRES-Containing VEEV Vaccine Protects Cynomolgus Macaques from IE Venezuelan Equine Encephalitis Virus Aerosol Chal- lenge. PLoS Negl Trop Dis. 2015; 9(5):e0003797. Epub 2015/05/29. https://doi.org/10.1371/journal. pntd.0003797 PMID: 26020513; PubMed Central PMCID: PMC4447396. 29. Roy CJ, Adams AP, Wang E, Plante K, Gorchakov R, Seymour RL, et al. Chikungunya vaccine candi- date is highly attenuated and protects nonhuman primates against telemetrically monitored disease fol- lowing a single dose. J Infect Dis. 2014; 209(12):1891–9. Epub 2014/01/10. https://doi.org/10.1093/ infdis/jiu014 PMID: 24403555; PubMed Central PMCID: PMC4038141. 30. Becker DE. Fundamentals of electrocardiography interpretation. Anesth Prog. 2006; 53(2):53–63; quiz 4. Epub 2006/07/26. https://doi.org/10.2344/0003-3006(2006)53[53:FOEI]2.0.CO;2 PMID: 16863387; PubMed Central PMCID: PMC1614214. 31. Ma H, Lundy JD, O’Malley KJ, Klimstra WB, Hartman AL, Reed DS. Electrocardiography Abnormalities in Macaques after Infection with Encephalitic Alphaviruses. Pathogens. 2019; 8(4). Epub 2019/11/21. https://doi.org/10.3390/pathogens8040240 PMID: 31744158; PubMed Central PMCID: PMC6969904. PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 29 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV 32. Baksi AJ, Kanaganayagam GS, Prasad SK. Arrhythmias in viral myocarditis and pericarditis. Card Elec- trophysiol Clin. 2015; 7(2):269–81. Epub 2015/05/24. https://doi.org/10.1016/j.ccep.2015.03.009 PMID: 26002391. 33. 34. Lange RA, Hillis LD. Clinical practice. Acute pericarditis. N Engl J Med. 2004; 351(21):2195–202. Epub 2004/11/19. https://doi.org/10.1056/NEJMcp041997 PMID: 15548780. Fung G, Luo H, Qiu Y, Yang D, McManus B. Myocarditis. Circ Res. 2016; 118(3):496–514. Epub 2016/ 02/06. https://doi.org/10.1161/CIRCRESAHA.115.306573 PMID: 26846643. 35. Kindermann I, Barth C, Mahfoud F, Ukena C, Lenski M, Yilmaz A, et al. Update on myocarditis. J Am Coll Cardiol. 2012; 59(9):779–92. Epub 2012/03/01. https://doi.org/10.1016/j.jacc.2011.09.074 PMID: 22361396. 36. Cooper LT Jr. Myocarditis. N Engl J Med. 2009; 360(15):1526–38. Epub 2009/04/10. https://doi.org/10. 1056/NEJMra0800028 PMID: 19357408; PubMed Central PMCID: PMC5814110. 37. Dennert R, Crijns HJ, Heymans S. Acute viral myocarditis. Eur Heart J. 2008; 29(17):2073–82. Epub 2008/07/12. https://doi.org/10.1093/eurheartj/ehn296 PMID: 18617482; PubMed Central PMCID: PMC2519249. 38. 39. Imazio M, Spodick DH, Brucato A, Trinchero R, Markel G, Adler Y. Diagnostic issues in the clinical man- agement of pericarditis. Int J Clin Pract. 2010; 64(10):1384–92. Epub 2010/05/22. https://doi.org/10. 1111/j.1742-1241.2009.02178.x PMID: 20487049. Tingle LE, Molina D, Calvert CW. Acute pericarditis. Am Fam Physician. 2007; 76(10):1509–14. Epub 2007/12/07. PMID: 18052017. 40. Pouch SM, Katugaha SB, Shieh WJ, Annambhotla P, Walker WL, Basavaraju SV, et al. Transmission of Eastern Equine Encephalitis Virus from an Organ Donor to Three Transplant Recipients. Clin Infect Dis. 2018. Epub 2018/10/30. https://doi.org/10.1093/cid/ciy923 PMID: 30371754; PubMed Central PMCID: PMC6488434. 41. Del Piero F, Wilkins PA, Dubovi EJ, Biolatti B, Cantile C. Clinical, pathologic, immunohistochemical, and virologic findings of eastern equine encephalomyelitis in two horses. Vet Pathol. 2001; 38(4):451– 6. Epub 2001/07/27. https://doi.org/10.1354/vp.38-4-451 PMID: 11467481. 42. Elvinger F, Liggett AD, Tang KN, Harrison LR, Cole JR Jr, Baldwin CA, et al. Eastern equine encephalo- myelitis virus infection in swine. J Am Vet Med Assoc. 1994; 205(7):1014–6. Epub 1994/10/01. PMID: 7852154. 43. Ethier M, Rogg J. Eastern equine encephalitis: MRI findings in two patients. Med Health R I. 2012; 95 (7):227–9. Epub 2012/08/30. PMID: 22928238. 44. Harvala H, Bremner J, Kealey S, Weller B, McLellan S, Lloyd G, et al. Case report: Eastern equine encephalitis virus imported to the UK. J Med Virol. 2009; 81(2):305–8. Epub 2008/12/25. https://doi.org/ 10.1002/jmv.21379 PMID: 19107960. 45. Hirsch MS, DeMaria A Jr, Schaefer PW, Branda JA. Case records of the Massachusetts General Hospi- tal. Case 22–2008. A 52-year-old woman with fever and confusion. N Engl J Med. 2008; 359(3):294– 303. Epub 2008/07/19. https://doi.org/10.1056/NEJMcpc0804149 PMID: 18635435. 46. Lury KM, Castillo M. Eastern equine encephalitis: CT and MRI findings in one case. Emerg Radiol. 2004; 11(1):46–8. Epub 2004/08/17. https://doi.org/10.1007/s10140-004-0350-7 PMID: 15309665. 47. Deresiewicz RL, Thaler SJ, Hsu L, Zamani AA. Clinical and neuroradiographic manifestations of east- ern equine encephalitis. N Engl J Med. 1997; 336(26):1867–74. Epub 1997/06/26. https://doi.org/10. 1056/NEJM199706263362604 PMID: 9197215. 48. Wendell LC, Potter NS, Roth JL, Salloway SP, Thompson BB. Successful management of severe neu- roinvasive eastern equine encephalitis. Neurocrit Care. 2013; 19(1):111–5. Epub 2013/06/05. https:// doi.org/10.1007/s12028-013-9822-5 PMID: 23733173. 49. Morse RP, Bennish ML, Darras BT. Eastern equine encephalitis presenting with a focal brain lesion. Pediatr Neurol. 1992; 8(6):473–5. Epub 1992/11/01. https://doi.org/10.1016/0887-8994(92)90013-o PMID: 1476580. 50. Silverman MA, Misasi J, Smole S, Feldman HA, Cohen AB, Santagata S, et al. Eastern equine enceph- alitis in children, Massachusetts and New Hampshire,USA, 1970–2010. Emerg Infect Dis. 2013; 19 (2):194–201; quiz 352. Epub 2013/01/25. https://doi.org/10.3201/eid1902.120039 PMID: 23343480; PubMed Central PMCID: PMC3559032. 51. Shah KJ, Cherabuddi K. Case of eastern equine encephalitis presenting in winter. BMJ Case Rep. 2016;2016. Epub 2016/05/12. https://doi.org/10.1136/bcr-2016-215270 PMID: 27165999; PubMed Central PMCID: PMC4885365. 52. Mukerji SS, Lam AD, Wilson MR. Eastern Equine Encephalitis Treated With Intravenous Immunoglobu- lins. Neurohospitalist. 2016; 6(1):29–31. Epub 2016/01/08. https://doi.org/10.1177/1941874415578533 PMID: 26740855; PubMed Central PMCID: PMC4680893. PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 30 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV 53. Hrabak T, Yerkey MW, Callerame K, Graham K. Eastern equine encephalitis presenting as psychosis. J Miss State Med Assoc. 2002; 43(4):109–10. Epub 2002/05/07. PMID: 11989192. 54. Piliero PJ, Brody J, Zamani A, Deresiewicz RL. Eastern equine encephalitis presenting as focal neuror- adiographic abnormalities: case report and review. Clin Infect Dis. 1994; 18(6):985–8. Epub 1994/06/ 01. https://doi.org/10.1093/clinids/18.6.985 PMID: 8086564. 55. Cho JJ, Wong JK, Henkel J, DeJesus RO, Nazario-Lopez B. Acute Seroconversion of Eastern Equine Encephalitis Coinfection With California Serogroup Encephalitis Virus. Front Neurol. 2019; 10:242. Epub 2019/04/04. https://doi.org/10.3389/fneur.2019.00242 PMID: 30941092; PubMed Central PMCID: PMC6433933. 56. Berlin D, Gilani AI, Grewal AK, Fowkes M. Eastern equine encephalitis. Pract Neurol. 2017; 17(5):387– 91. Epub 2017/07/30. https://doi.org/10.1136/practneurol-2017-001659 PMID: 28754695. 57. Nickerson JP, Kannabiran S, Burbank HN. MRI findings in eastern equine encephalitis: the "parenthe- sis" sign. Clin Imaging. 2016; 40(2):222–3. Epub 2016/03/21. https://doi.org/10.1016/j.clinimag.2015. 10.012 PMID: 26995574. 58. Anderson BA. Focal neurologic signs in western equine encephalitis. Can Med Assoc J. 1984; 130 (8):1019–21. Epub 1984/04/15. PMID: 6704848; PubMed Central PMCID: PMC1876072. 59. Sundaram MB, Siemens P. Lateralized EEG abnormalities in western equine encephalitis. Can Med Assoc J. 1984; 131(3):186, 8. Epub 1984/08/01. PMID: 6744162; PubMed Central PMCID: PMC1483294. 60. Delfraro A, Burgueno A, Morel N, Gonzalez G, Garcia A, Morelli J, et al. Fatal human case of Western equine encephalitis, Uruguay. Emerg Infect Dis. 2011; 17(5):952–4. Epub 2011/05/03. https://doi.org/ 10.3201/eid1705.101068 PMID: 21529429; PubMed Central PMCID: PMC3321764. 61. Ehrenkranz NJ, Ventura AK. Venezuelan equine encephalitis virus infection in man. Annu Rev Med. 1974; 25:9–14. Epub 1974/01/01. https://doi.org/10.1146/annurev.me.25.020174.000301 PMID: 4824504. 62. Mutti C, Curti E, Ciliento R, Melpignano A, Florindo I, Zinno L, et al. Herpes Simplex Virus 1 encephalitis with normal cerebrospinal fluid after brain radiotherapy in a patient with glioblastoma. A case report and review of literature. Acta Biomed. 2019; 90(2):327–30. Epub 2019/05/28. https://doi.org/10.23750/abm. v90i2.8218 PMID: 31125013. 63. Kalita J, Misra UK, Mani VE, Bhoi SK. Can we differentiate between herpes simplex encephalitis and Japanese encephalitis? J Neurol Sci. 2016; 366:110–5. Epub 2016/06/12. https://doi.org/10.1016/j.jns. 2016.05.017 PMID: 27288787. 64. Grande-Martin A, Pardal-Fernandez JM, Garcia-Lopez FA. Utility of EEG findings in the management of a case of herpes simplex encephalitis. Neurologia. 2017; 32(3):193–5. Epub 2015/06/11. https://doi. org/10.1016/j.nrl.2015.04.001 PMID: 26059805. 65. Stelten BM, Venhovens J, van der Velden LB, Meulstee J, Verhagen WI. Syndrome of transient head- ache and neurological deficits with cerebrospinal fluid lymphocytosis (HaNDL): A case report with serial electroencephalography (EEG) recordings. Is there an association with human herpes virus type 7 (HHV-7) infection? Cephalalgia. 2016; 36(13):1296–301. Epub 2015/12/20. https://doi.org/10.1177/ 0333102415618616 PMID: 26682576. 66. Sutter R, Kaplan PW, Cervenka MC, Thakur KT, Asemota AO, Venkatesan A, et al. Electroencephalog- raphy for diagnosis and prognosis of acute encephalitis. Clin Neurophysiol. 2015; 126(8):1524–31. Epub 2014/12/06. https://doi.org/10.1016/j.clinph.2014.11.006 PMID: 25476700. 67. Kalita J, Misra UK, Pandey S, Dhole TN. A comparison of clinical and radiological findings in adults and children with Japanese encephalitis. Arch Neurol. 2003; 60(12):1760–4. Epub 2003/12/17. https://doi. org/10.1001/archneur.60.12.1760 PMID: 14676053. 68. Zetterberg H, Blennow K. Fluid biomarkers for mild traumatic brain injury and related conditions. Nat Rev Neurol. 2016; 12(10):563–74. Epub 2016/09/17. https://doi.org/10.1038/nrneurol.2016.127 PMID: 27632903. 69. Gan ZS, Stein SC, Swanson R, Guan S, Garcia L, Mehta D, et al. Blood Biomarkers for Traumatic Brain Injury: A Quantitative Assessment of Diagnostic and Prognostic Accuracy. Front Neurol. 2019; 10:446. Epub 2019/05/21. https://doi.org/10.3389/fneur.2019.00446 PMID: 31105646; PubMed Central PMCID: PMC6498532. 70. Ianof JN, Anghinah R. Traumatic brain injury: An EEG point of view. Dement Neuropsychol. 2017; 11 (1):3–5. Epub 2017/12/08. https://doi.org/10.1590/1980-57642016dn11-010002 PMID: 29213487; PubMed Central PMCID: PMC5619208. 71. Amyot F, Arciniegas DB, Brazaitis MP, Curley KC, Diaz-Arrastia R, Gandjbakhche A, et al. A Review of the Effectiveness of Neuroimaging Modalities for the Detection of Traumatic Brain Injury. J Neuro- trauma. 2015; 32(22):1693–721. Epub 2015/07/16. https://doi.org/10.1089/neu.2013.3306 PMID: 26176603; PubMed Central PMCID: PMC4651019. PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 31 / 32 PLOS NEGLECTED TROPICAL DISEASES Utilization of advance telemetry in cynomolgus macaques infected with EEEV 72. Sandsmark DK, Elliott JE, Lim MM. Sleep-Wake Disturbances After Traumatic Brain Injury: Synthesis of Human and Animal Studies. Sleep. 2017; 40(5). Epub 2017/03/23. https://doi.org/10.1093/sleep/ zsx044 PMID: 28329120; PubMed Central PMCID: PMC6251652. 73. Hunt AR, Frederickson S, Maruyama T, Roehrig JT, Blair CD. The first human epitope map of the alpha- viral E1 and E2 proteins reveals a new E2 epitope with significant virus neutralizing activity. PLoS Negl Trop Dis. 2010; 4(7):e739. Epub 2010/07/21. https://doi.org/10.1371/journal.pntd.0000739 PMID: 20644615; PubMed Central PMCID: PMC2903468. 74. 75. Zhang R, Hryc CF, Cong Y, Liu X, Jakana J, Gorchakov R, et al. 4.4 A cryo-EM structure of an envel- oped alphavirus Venezuelan equine encephalitis virus. EMBO J. 2011; 30(18):3854–63. Epub 2011/08/ 11. https://doi.org/10.1038/emboj.2011.261 PMID: 21829169; PubMed Central PMCID: PMC3173789. Johnson BJ, Brubaker JR, Roehrig JT, Trent DW. Variants of Venezuelan equine encephalitis virus that resist neutralization define a domain of the E2 glycoprotein. Virology. 1990; 177(2):676–83. Epub 1990/ 08/01. https://doi.org/10.1016/0042-6822(90)90533-w PMID: 1695412. 76. Voss JE, Vaney MC, Duquerroy S, Vonrhein C, Girard-Blanc C, Crublet E, et al. Glycoprotein organiza- tion of Chikungunya virus particles revealed by X-ray crystallography. Nature. 2010; 468(7324):709–12. Epub 2010/12/03. https://doi.org/10.1038/nature09555 PMID: 21124458. 77. Pittman PR, Liu CT, Cannon TL, Mangiafico JA, Gibbs PH. Immune interference after sequential alpha- virus vaccine vaccinations. Vaccine. 2009; 27(36):4879–82. Epub 2009/07/07. https://doi.org/10.1016/j. vaccine.2009.02.090 PMID: 19576665. 78. Pittman PR, Makuch RS, Mangiafico JA, Cannon TL, Gibbs PH, Peters CJ. Long-term duration of detectable neutralizing antibodies after administration of live-attenuated VEE vaccine and following booster vaccination with inactivated VEE vaccine. Vaccine. 1996; 14(4):337–43. Epub 1996/03/01. https://doi.org/10.1016/0264-410x(95)00168-z PMID: 8744562. PLOS Neglected Tropical Diseases | https://doi.org/10.1371/journal.pntd.0009424 June 17, 2021 32 / 32 PLOS NEGLECTED TROPICAL DISEASES
null
10.7554_elife.85521.pdf
Data availability Figure 3—source data 1, Figure 5—source data 1, Figure 6—source data 1, and Figure 7—source data 1 contain the numerical and statistical data used to generate the figures. The confocal imaging dataset is available at Brain Image Library under DOI https://doi.org/10.35077/g.933.
Data availability Figure 3 -source data 1, Figure 5 -source data 1, Figure 6 -source data 1, and Figure 7 -source data 1 contain the numerical and statistical data used to generate the figures. The confocal imaging dataset is available at Brain Image Library under DOI https://doi.org/10.35077/g.933 . The following dataset was generated:
RESEARCH ARTICLE Origin of wiring specificity in an olfactory map revealed by neuron type–specific, time- lapse imaging of dendrite targeting Kenneth Kin Lam Wong1, Tongchao Li1*†, Tian- Ming Fu2‡, Gaoxiang Liu3, Cheng Lyu1, Sayeh Kohani1, Qijing Xie1, David J Luginbuhl1, Srigokul Upadhyayula3,4,5, Eric Betzig2,3,6, Liqun Luo1* 1Department of Biology, Howard Hughes Medical Institute, Stanford University, Stanford, United States; 2Howard Hughes Medical Institute, Janelia Research Campus, Ashburn, United States; 3Advanced Bioimaging Center, Department of Molecular and Cell Biology, University of California, Berkeley, Berkeley, United States; 4Molecular Biophysics and Integrated Bioimaging Division, Lawrence Berkeley National Laboratory, Berkeley, United States; 5Chan Zuckerberg Biohub, San Francisco, United States; 6Departments of Molecular and Cell Biology and Physics, Howard Hughes Medical Institute, Helen Wills Neuroscience Institute, University of California, Berkeley, United States Abstract How does wiring specificity of neural maps emerge during development? Formation of the adult Drosophila olfactory glomerular map begins with the patterning of projection neuron (PN) dendrites at the early pupal stage. To better understand the origin of wiring specificity of this map, we created genetic tools to systematically characterize dendrite patterning across develop- ment at PN type–specific resolution. We find that PNs use lineage and birth order combinatorially to build the initial dendritic map. Specifically, birth order directs dendrite targeting in rotating and binary manners for PNs of the anterodorsal and lateral lineages, respectively. Two- photon– and adaptive optical lattice light- sheet microscope–based time- lapse imaging reveals that PN dendrites initiate active targeting with direction- dependent branch stabilization on the timescale of seconds. Moreover, PNs that are used in both the larval and adult olfactory circuits prune their larval- specific dendrites and re- extend new dendrites simultaneously to facilitate timely olfactory map organization. Our work highlights the power and necessity of type- specific neuronal access and time- lapse imaging in identifying wiring mechanisms that underlie complex patterns of func- tional neural maps. Editor's evaluation When a neuron is born it correlates with where it targets in the neuropil and this has been best demonstrated in the olfactory lobe of Drosophila. This important study uses sophisticated genetics and advanced live imaging to provide a compelling description of how neuronal dendrites explore the target field, eliminate excessive branches, and assort into the correct region during develop- ment. In the process, it develops valuable tools. The study brings us closer to a comprehensive understanding of how the birth order of a neuron translates to dendrite patterning within the Drosophila antennal lobe circuit. *For correspondence: ltongchao@outlook.com (TL); lluo@stanford.edu (LL) Present address: †Liangzhu Laboratory, MOE Frontier Science Center for Brain Science and Brain- machine Integration, State Key Laboratory of Brain- machine Intelligence, Zhejiang University, Hangzhou, China; ‡Department of Electrical and Computer Engineering, Princeton University, Princeton, United States Competing interest: The authors declare that no competing interests exist. Funding: See page 25 Received: 11 December 2022 Preprinted: 29 December 2022 Accepted: 27 March 2023 Published: 28 March 2023 Reviewing Editor: Sonia Sen, Tata Institute for Genetics and Society, India Copyright Wong et al. This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited. Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 1 of 33 Research article eLife digest The brain’s ability to sense, act and remember relies on the intricate network of connections between neurons. Organization of these connections into neural maps is critical for processing sensory information. For instance, different odors are represented by specific neurons in a part of the brain known as the olfactory bulb, allowing animals to distinguish between smells. Projection neurons in the olfactory bulb have extensions known as dendrites that receive signals from sensory neurons. Scientists have extensively used the olfactory map in adult fruit flies to study brain wiring because of the specific connections between their sensory and projection neurons. This has led to the discovery of similar wiring strategies in mammals. But how the olfactory map is formed during development is not fully understood. To investigate, Wong et al. built genetic tools to label specific types of olfactory projection neurons during the pupal stage of fruit fly development. This showed that a group of projection neurons directed their dendrites in a clockwise rotation pattern depending on the order in which they were born: the first- born neuron sent dendrites towards the top right of the antennal lobe (the fruit fly equivalent of the olfactory bulb), while the last- born sent dendrites towards the top left. Wong et al. also carried out high- resolution time- lapse imaging of live brains grown in the labora- tory to determine how dendrites make wiring decisions. This revealed that projection neurons send dendrites in all directions, but preferentially stabilize those that extend in the direction which the neurons eventually target. Also, live imaging showed neurons could remove old dendrites (used in the larvae) and build new ones (to be used in the adult) simultaneously, allowing them to quickly create new circuits. These experiments demonstrate the value of imaging specific types of neurons to understand the mechanisms that assemble neural maps in the developing brain. Further work could use the genetic tools created by Wong et al. to study how wiring decisions are determined in this and other neural maps by specific genes, potentially yielding insights into neurological disorders associated with wiring defects. Introduction Organization of neuronal connectivity into spatial maps occurs widely in the nervous systems across species (Luo and Flanagan, 2007; Cang and Feldheim, 2013; Luo, 2021). For example, in the retino- topic map of the visual system, nearby neurons in the input field project axons to nearby neurons in the target field (Cang and Feldheim, 2013). Such a continuous organization preserves spatial relation- ships in the visual world. Contrary to retinotopy, the olfactory glomerular map consists of discrete units called glomeruli in which input neurons connect with the cognate output neurons based on neuronal type rather than soma position (Mombaerts et al., 1996; Gao et al., 2000; Vosshall et al., 2000). This discrete map represents a given odor by the combinatorial activation of specific glomeruli. Whereas continuous maps are readily built using gradients of guidance cues (Cang and Feldheim, 2013), how glomeruli are placed at specific locations in discrete maps is less clear (Murthy, 2011). Understanding the developmental origins of these neural maps is fundamental for deciphering the logic of their func- tional organization through which information is properly represented and processed. The adult Drosophila olfactory map in the antennal lobe (the equivalent of the vertebrate olfactory bulb) has proven to be a powerful model for studying mechanisms of wiring specificity, thanks to the type- specific connections between the presynaptic olfactory receptor neurons (ORNs) and the cognate postsynaptic projection neurons (PNs). Molecules and mechanisms first identified in this circuit have been found to play similar roles in the wiring of the mammalian brain (e.g. Hong et al., 2012; Berns et al., 2018; Pederick et al., 2021). Assembly of the fly olfactory map begins with dendritic growth and patterning of PNs derived primarily from the anterodorsal (adPNs) and lateral (lPNs) lineages and born with an invariant birth order within each lineage (Jefferis et al., 2001; Jefferis et al., 2004; Marin et al., 2005; Yu et al., 2010; Lin et al., 2012; Figure 1A and B). This patterning creates a prototypic olfactory map, prior to ORN axon innervation, indicative of the PN- autonomous ability to target dendrites into specific regions. However, earlier studies could only unambiguously follow the development of one single PN type – DL1 PNs (Jefferis et al., 2004). It remains unclear to date how the prototypic olfactory map is organized and what cellular mechanisms PN dendrites use to achieve Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 2 of 33 Developmental Biology | Neuroscience Research article Figure 1. Organization and development of the adult olfactory circuit in Drosophila. (A, B) Timeline (A) and schematic illustration (B) of Drosophila olfactory circuit development. Green, red, and blue circles denote the birth of embryonic- born anterodorsal projection neuron (adPN), larval- born adPN, and larval- born lPN, respectively. At the onset of metamorphosis, the larval- specific olfactory circuit degenerates; larval olfactory receptor neurons (ORNs) die while embryonic- born adPNs prune their larval- specific processes and re- extend new processes into the adult- specific olfactory circuit. In the adult- specific olfactory circuit, projection neuron (PN) dendrites extend first and form a prototypic map. This is followed by an extension of ORN axons and synaptic partner matching between cognate PN dendrites and ORN axons to form a mature map. Solid and open arrowheads in A indicate onset of innervation for PN dendrites and ORN axons, respectively. (C) Overview of this study investigating the logic of dendritic patterning (C1; see Figures 3 and 4) as well as cellular mechanisms of dendrite targeting specificity (C2; see Figures 6 and 7) and re- wiring (C3; see Figure 8) that contribute to the developmental origin of the adult Drosophila olfactory map. (D) Staining of fixed brains at indicated stages showing dendrite development of adPNs (VT033006+ run+ ; labeled in yellow) and lPNs (VT033006+ run–; labeled in cyan). As run- FLP is expressed before 0 h APF in adPN but not lPN neuroblasts, we can use it to label adPNs and lPNs with two distinct colors using an intersectional reporter (see Materials and methods for the genotype). Yellow arrowheads in (D1) mark larval- and adult- specific dendrites of adPNs in larval- and adult- specific antennal lobes, respectively. Cyan arrowheads in (D3) denote specific targeting of lPN dendrites at the opposite ends of the dorsomedial- ventrolateral axis. (D1): N=12; (D2): N=7; (D3): N=17; (D4): N=10; (D5): N=12. Common notations in this study: Unless otherwise indicated, all images in this and subsequent figures are partial z projections of confocal stacks of representative images. N indicates the number of antennal lobes imaged. Antennal lobe neuropils are revealed by N- Cadherin (Ncad; in blue) staining. Adult- specific (developing) antennal lobe is outlined with a white solid line. Larval- specific antennal lobe is outlined with an orange line (dashed line used to denote the degeneration stage) and is distinguished from the developing antennal lobe by the more intense nc82 staining as shown in Figure 1—figure supplement 1 (nc82 channel not shown here). Asterisks (*) indicate PN cell bodies, which are outside the antennal lobe neuropil (and sometimes appear on top because of the z- projections). Arrowheads mark PN dendrites. Arrows mark PN axons projecting towards higher olfactory centers (see Figure 1—figure supplement 2 for PN axons at their targets in the mushroom body and lateral horn). h APF: hours after puparium formation; h ALH: hours after larval hatching. DL: dorsolateral; DM: dorsomedial; VM: ventromedial; VL: ventrolateral. Scale bar = 10 µm. The online version of this article includes the following figure supplement(s) for figure 1: Figure supplement 1. Visualization of larval- and adult- specific antennal lobes by co- staining of Ncad and nc82. Figure supplement 2. Projection neuron (PN) axon development across pupal stages. Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 3 of 33 Developmental Biology | Neuroscience Research article targeting specificity (Figure 1C1- 2). The initial map formation is further complicated by circuit remod- eling during which embryonic- born PNs used in both the larval and adult circuits reorganize their neurites (Marin et  al., 2005). How embryonic- born PNs coordinate remodeling with re- integration into the adult circuit is not known (Figure 1C3). Here, we set out to explore the origin of the olfactory map by performing a systematic and compar- ative study of PN dendrite development at type- specific resolution in vivo, and two- photon– and adaptive optical lattice light- sheet microscope–based time- lapse imaging of PN dendrites in early pupal brain explants. As our overarching goal is to understand how the wiring specificity between ORNs and PNs arises, we focus on PNs that project to single glomeruli. Neurons from the lateral lineage that innervate multiple glomeruli or project to other regions of the adult brain (Lin et  al., 2012) are not studied here. Our study uncovers wiring logic that directs PN dendrites to create an organized olfactory map, dendritic branch dynamics that lead to directional selectivity, and a novel re- wiring mechanism that facilitates timely olfactory map formation. These wiring strategies used in the initial map organization lay the foundation of precise synaptic connectivity between PNs and ORNs in the final glomerular map. Results Overview of Drosophila olfactory circuit development at a lineage- specific resolution We first described the development of the Drosophila olfactory circuit using pupal brains double- labeled for adPNs and lPNs (Figure 1D; see the genetic design in Figure 2). At the onset of metamor- phosis (0 hr after puparium formation; 0 hr APF), the adult- specific antennal lobe (also referred to as ‘developing antennal lobe’) remained relatively small, located dorsolateral and posterior to the larval- specific antennal lobe (also referred to as ‘degenerating antennal lobe’) (Figure 1D1). As PN dendrites continued to grow and innervate the developing antennal lobe, its size increased considerably (Figure 1D1–3). By 12 hr APF, PNs already appeared to be sorting their dendrites into specific regions to form a prototypic map, as revealed by the heterogeneous patterning of lPN dendrites (arrowheads in Figure  1D3). From 21 hr to 50 hr APF, dendrites of adPNs and lPNs gradually segregated and eventually formed intercalated but non- overlapping glomeruli (Figure 1D4–5). The development of the adult- specific antennal lobe partially overlapped with the degeneration of the larval- specific antennal lobe, as indicated by fragmentation of the larval- specific dendrites of embryonic- born PNs at 3  hr APF (Figure 1D2). This gross characterization at the resolution of two PN lineages was consistent with earlier studies (Jefferis et al., 2004; Marin et al., 2005). However, the resolution was not sufficiently high to answer the questions we raised in the Introduction (Figure 1C). Expanded genetic toolkit for type-specific labeling of PNs during early pupal development To reveal how PN dendrites initiate olfactory map formation at the high spatiotemporal resolution, we needed genetic access to specific PN types during early pupal development. From our recently deciphered single- cell PN transcriptomes (Xie et  al., 2021), we searched for genetic markers that are expressed strongly and persistently in single or a few PN types across pupal development. This transcriptome- instructed search led to the identification of CR45223 (in place of this non- coding gene, we used the adjacent CG14322 that exhibits nearly identical expression pattern), lov, and tsh (Figure 2A and B; Figure 2—figure supplement 1). Next, using CRISPR/Cas9, we generated knock- in transgenic QF2 expression driver lines in which T2A- QF2 (or T2A- FLP for intersection) was inserted immediately before the stop codon of the endog- enous gene (Figure  2—figure supplement 2). The self- cleaving peptide T2A allows QF2 to be expressed in the same pattern as the endogenous gene (Diao and White, 2012). With these new QF2 lines together with existing GAL4 lines that label additional PN types (Xie et al., 2019), we now have an expanded toolkit accessing PNs ranging from early- to late- born PNs, from adPN to lPN lineages, and from PNs with neighboring glomerular projections to those with distant projections in the adult antennal lobe (Figure 2C and D). As QF2/QUAS and GAL4/UAS expression systems operate orthogo- nally to each other (Potter et al., 2010; Riabinina et al., 2015), we crossed our QF2 lines with existing GAL4 lines for simultaneous labeling of distinct PN types in the same brain (see inset in Figure 2C). Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 4 of 33 Developmental Biology | Neuroscience Research article Figure 2. Expanded genetic toolkit for dual- color, type- specific labeling of projection neurons (PNs). (A) tSNE plot of PN single- cell transcriptomes, color- coded according to CR45223 expression level in [log2(CPM +1)], where CPM stands for transcript counts per million reads. Zoom- in of boxes in the tSNE plot (left) is shown on the right, and color- coded according to PN types and developmental stages. (B) Dot plot showing the expression of acj6, vvl, CR45223, CG14322, lov, and tsh in 0 hr APF PNs arranged according to their birth order and lineage (green: embryonic- born anterodorsal projection neuron (adPNs); red: larval- born adPNs; blue: larval- born lPNs). Unit of expression is [log2(CPM +1)] as in A. Data from panels A are B are from Xie et al., 2021. (C) Birth orders of adPNs and lPNs summarized by Lin et al., 2012; Yu et al., 2010 and genetic tools used to access them. Left: Accessible PN types are colored. Circles beneath the PN types denote QF2/GAL4 drivers used to access them. Asterisks beneath the PN types denote access by MARCM. Gray arrowhead marks neuroblast (NB) rest. Right: Genetic tools. Inset shows the combinatorial use of QF2/FLP and GAL4 (linked by dashed lines) for comparative analyses of dendrite development of two groups of PNs in the same animal. (D) Schematic of glomerular projections of QF2/ GAL4- accessible PNs in the adult antennal lobe. Indicated glomeruli are color- coded based on the genetic tools used to access them. See the color code in C. (E, F) Schematic of intersectional logic gates for dual- color labeling of PNs. See Figure 2—figure supplement 2 for newly generated FLP- out reporters. The online version of this article includes the following figure supplement(s) for figure 2: Figure supplement 1. Expression of projection neuron (PN) marker genes across development. Figure supplement 2. Generation of T2A- QF2/FLP transgenic flies by CRISPR/Cas9. Figure supplement 3. Design of single- and dual- color FLP- out reporters. This combinatorial use of driver lines permitted comparative analyses of the development of distinct PN types with minimal biological and technical variations (Supplementary file 1). To limit driver expression only in PNs, we applied intersectional logic gates (AND and NOT gates) using our newly generated conditional reporters genetically encoding either mGreenLantern, Halo tags, and/or SNAP tags (Kohl et al., 2014; Sutcliffe et al., 2017; Campbell et al., 2020; Figure 2E and F; Figure 2—figure supplement 3). These reporters can be broadly used in other systems. Finally, Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 5 of 33 Developmental Biology | Neuroscience Research article we used MARCM (Lee and Luo, 1999) to label PNs that remain inaccessible due to a lack of drivers (Figure 2C; discussed in Figure 3). Early larval-born adPN dendrites initially share similar targeting regions Using the new genetic tools, we first re- visited the dendrite development of DL1 PNs—the first larval- born adPN type—using pupal brains double- labeled for DL1 PNs (labeled by 71B05- GAL4) and adPNs (Figure  3A). Consistent with our previous study (Jefferis et  al., 2004), DL1 PNs already showed robust dendritic growth at the wandering third instar larval stage (Figure 3—figure supplement 1A). At 0 hr APF, DL1 PN dendrites extended radially outwards from the main process, reaching nearly the entire developing antennal lobe and often overshooting it (white arrowheads in Figure 3A1), likely surveying the surroundings. By 6 hr APF, most of the dendrites already occupied the dorsolateral (DL) corner of the antennal lobe (Figure 3A2). As the antennal lobe continued to grow, this dorsolateral positioning of the DL1 PN dendrites remained largely unchanged (Figure  3A3–6). From 21  hr APF onwards, the dendrites underwent progressive refinement: they were restricted into a smaller area by 30 hr APF (Figure 3A4–5), and eventually formed a compact, posterior glomerulus by 50 hr APF (Figure 3A6 showing a single z section). To assess whether other PN types follow the same developmental trajectory, we next examined CG14322+ PNs, which include DL1 PNs and DA3 PNs—the first and second larval- born adPN types, respectively. In the same brain, we also labeled with a different fluorophore DC2 PNs—the third larval- born adPN type (Figure  3B). The dendritic pattern of DL1/DA3 PNs appeared indistinguish- able from that of DL1 PNs from 0 hr to 12 hr APF (compare the yellow channel of Figure 3B1–3 with Figure 3A1–3), suggesting that DL1 and DA3 PN sent dendrites to the same region in the antennal lobe. We began to see differences in 21 hr APF pupal brains in which DL1/DA3 PN dendrites not only occupied the dorsolateral region but also spread ventrally (white arrowhead in Figure 3B4; compare with Figure  3A4). The more ventrally targeted dendrites likely belong to DA3 PNs. This suggests that ~21 hr APF marks the beginning of dendritic segregation of DL1 and DA3 PNs. By 30 h APF, DL1 and DA3 dendrites were clearly separable (Figure 3B5), which respectively formed more posteriorly and anteriorly targeted glomeruli at 50 hr APF (Figure 3B6; see single z sections in Figure 3—figure supplement 1C). Next, we focused on the third- born—DC2 PNs labeled by 91G04- GAL4 (Figure 3B). This GAL4 labeled additional embryonic- born adPNs from 0 hr to 6 hr APF, but the expression in these PNs diminished afterward. As embryonic- born adPNs do not have any dendrites in the developing antennal lobe at 0 hr APF (discussed in Figure 8), dendrites found in the antennal lobe should belong to the larval- born DC2 PNs. Like DL1/DA3 PNs, DC2 PNs initiated radial dendritic extension across the antennal lobe at 0 hr APF (Figure 3B1; Figure 3—figure supplement 1B). Notably, DL1/DA3 and DC2 PN dendrites exhibited substantial overlap from 0 hr to 12 hr APF and shared a similar targeting region at the dorsolateral corner from 6 hr to 12 hr APF (Figure 3B1–3). It was not until 21 hr APF that DL1, DA3, and DC2 dendrites began to segregate from each other along both medial- lateral and anterior- posterior axes (Figure 3B4–5). By 50 hr APF, the DC2 glomerulus was separated from DL1/DA3 glomeruli by intermediate glomeruli (Figure 3B6). In summary, dendrites of consecutively larval- born DL1, DA3, and DC2 adPNs (here collectively named ‘early larval- born adPNs’; see its definition in next section) develop in a similar fashion and share a similar targeting region at early pupal stages (0–12  hr APF). This is then followed by their segregation into distinct regions close to their adult glomerular positions during mid- pupal stages (21–50 hr APF). Larval-born adPNs with distant birth order send dendrites to distinct regions The analysis of early larval- born adPNs (Figure  3A and B) led us to hypothesize that larval- born adPNs might use their birth order to coordinate dendrite targeting during early pupal stages. If this were true, we would expect dendrites of larval- born adPNs with distant birth order to occupy distinct regions. To test this hypothesis, we compared dendrite- targeting regions of early larval- born adPNs with those of later- born adPNs. We first examined DC3/VA1d adPNs (referred to as ‘mid- early larval- born adPNs’) using Mz19- GAL4 (Figure 3C). This GAL4 is expressed in three PN types from 24 hr APF to adulthood: DC3 adPNs, VA1d Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 6 of 33 Developmental Biology | Neuroscience Research article Figure 3. Birth order–dependent spatial patterning of anterodorsal projection neuron (adPN) dendrites in the developing antennal lobe. (A) Confocal images of fixed brains at indicated stages showing dendrite development of adPNs (acj6+; labeled in green) and DL1 adPNs (71B05+; labeled in yellow). Right column of A1 shows a zoom- in of the dashed box. The labeling of acj6+ adPNs outlines the developing antennal lobe and is used in dual- color Figure 3 continued on next page Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 7 of 33 Developmental Biology | Neuroscience Research article Figure 3 continued AO- LLSM imaging later (see Figure 7A–C). White arrowheads in (A1) mark dendrites overshooting the antennal lobe. (A1): N=14; (A2): N=12; (A3): N=14; (A4): N=6; (A5): N=4; (A6): N=4. (B) Confocal images of fixed brains at indicated stages showing dendrite development of DL1/DA3 adPNs (CG14322+; labeled in yellow) and DC2 adPNs (91G04+; labeled in magenta). As 91G04- GAL4 labels some embryonic- born projection neurons (PNs) from 0 to 6 hr APF, their neurites are found in the larval- specific antennal lobe (B1, 2). Right column of (B1) shows a zoom- in of the dashed box. White arrowhead in (B4) denotes the more ventrally targeted DL1/DA3 dendrites. (B1): N=6; (B2): N=5; (B3): N=12; (B4): N=4; (B5): N=7; (B6): N=2. (C) Confocal images of fixed brains at indicated stages showing dendrite development of DC3/VA1d adPNs (Mz19+ acj6+; labeled in red) and DA1 lPNs (Mz19+ acj6–; labeled in cyan). (C1): N=14; (C2): N=6; (C3): N=4; (C4): N=10; (C5): N=10; (C6): N=6; (C7): N=4. (D) Confocal images of single- cell MARCM clones (in yellow) of DL1 PNs (D1–3), mid- late larval- born adPNs (D4–6), and late larval- born adPNs (D7–9) in 12 hr APF pupal brains, generated by heat shocks (hs) at indicated times. Three biological samples are shown for each of the indicated adPN cohorts. D1–3: N=5; D4–6: N=4; D7–9: N=8. (E) Summary of wiring logic of larval- born adPN dendrites to form an olfactory map in the 12 hr APF developing antennal lobe. See Figure 1 legend for common notations. The online version of this article includes the following video, source data, and figure supplement(s) for figure 3: Figure supplement 1. Dendrite development of early larval- born projection neurons (PNs). Figure supplement 2. MARCM- labeled single- cell projection neurons (PNs) of indicated lineages in adult brains. Figure supplement 3. Dendrite development of DL1, middle larval- born, and late larval- born projection neurons (PNs) at early stages. Figure supplement 3—source data 1. Source data for Figure 3—figure supplement 3F and G. Figure 3—video 1. 3D rendering of z stacks of indicated projection neurons (PNs) in 12 hr APF antennal lobe. https://elifesciences.org/articles/85521/figures#fig3video1 adPNs, and DA1 lPNs (Jefferis et al., 2004). To distinguish adPNs from lPNs, we previously adopted an FLP- out strategy labeling Mz19+ PNs with either GFP or RFP based on their lineages and studied dendrite segregation and refinement during mid- pupal stages (Li et al., 2021; Figure 3C4–7). However, the weak GAL4 expression before 24 hr APF prevented us from visualizing any dendrites at earlier stages. To overcome this, we incorporated Halo and SNAP chemical labeling (Kohl et al., 2014) in place of the immunofluorescence approach. This modification substantially extended the detection to developmental stages as early as 12 hr APF (Figure 3C1). We found that, from 12 hr to 21 hr APF, DC3/VA1d PN dendrites targeted the ventrolateral (VL) corner of the antennal lobe (Figure 3C1–4). Thus, early (DL1/DA3/DC2) and mid- early (DC3/VA1d) larval- born adPN dendrites occupy distinct regions at 12 hr APF. As we did not have reliable drivers to access other later- born PNs at early pupal stages, we turned to MARCM (Lee and Luo, 1999) to generate heat shock- induced single- cell clones of PNs born at different times (Figure 3—figure supplement 2). We used GH146- GAL4(IV), a PN driver that labels the majority of PN types, including later- born adPNs (Figure  3—figure supplement 2D–E), with a tight temporal control of heat shock and analyzed heat shock- induced animals that were among the first to form puparium to minimize the effects of unsynchronized development among individual animals (see Materials and methods for details). These optimizations permitted a systematic clonal analysis at higher PN type- specific resolution that correlates with birth time. Based on birth timing that corresponds to the heat shock time we applied to induce single- cell MARCM clones, we assigned larval- born adPNs to approximate temporal cohorts: (1) heat shock at 0–24 hr ALH (after larval hatching): first- born (DL1), (2) heat shock at 42–48 hr ALH: early- born (DL1, DA3, DC2, and D), (3) heat shock at 66–72 hr ALH: mid- late born (VM7v, VM7d, VM2, DM6, and VA1v), and (4) heat shock at 96–100 hr ALH: late- born (DM6, VA1v, DL2v, DL2d) (Figure 3E1). We assigned DC3/VA1d PNs labeled by Mz19- GAL4 to the mid- early cohort because they are born between the early and mid- late adPNs. We note that DM6 and VA1v PNs were assigned to both cohorts of mid- late and late- born adPNs, reflecting the nature of short birth timing differences and overlaps between adjacent cohorts. Using this strategy, we could also label lPNs born at different times and assigned them into approximate temporal cohorts (Figure 3—figure supplement 2F). Clonal analysis revealed that, at 12 hr APF, the first- born DL1 adPNs sent dendrites to the dorso- lateral corner of the antennal lobe as expected (Figure  3D1–3). By contrast, dendrites of mid- late larval- born adPNs occupied a large region on the medial/dorsomedial (M/DM) side (Figure  3D4–6). Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 8 of 33 Developmental Biology | Neuroscience Research article The dendritic arborization patterns of these PNs varied widely, most likely because they belonged to different PN types. Intriguingly, late larval- born adPN dendrites targeted the peripheral, dorsomedial (abbreviated as pDM) corner where the staining of the pan- neuropil marker N- Cadherin was relatively weak (Figure 3D7–9). The weak staining implies that this area is less populated by PN dendrites (the major constituent of the antennal lobe neuropil at this stage), possibly because (1) this area is not innervated by many PNs and/or (2) the dendrites of late- born PNs innervate later and remain less elaborate than earlier- born PNs (we will explore this later). Together, our data (Figure 3A–D) suggest that larval- born adPNs with adjacent birth order send dendrites to similar regions of the developing antennal lobe whereas those with distant birth order send dendrites to distinct regions (Figure 3E2,3). Notably, the birth order of the examined PNs does not specify dendrite targeting randomly (Figure 3E4). Rather, the stereotyped dendritic pattern in the prototypic map correlates with the birth order in an organized manner (rotating clockwise in the right hemisphere when viewed from the front; anti- clockwise in the left: early↔DL; mid- early↔VL; mid- late↔M/DM; late↔pDM). One can, therefore, infer at least the approximate birth order of a larval- born adPN based on its initial dendrite targeting, and vice versa. As the antennal lobe is a 3D structure, we also visualized PN dendrite targeting in the 12 hr APF map with 3D rendering generated from z stacks with rotation along the y- axis (Figure 3—video 1). We found that, along the short anterior- posterior axis (spanning about 20  µm), PN dendrites were located primarily on the periphery of the antennal lobe, whereas the center housed the axon bundle projecting out of the antennal lobe. Some dendrites could reach almost the entire depth, suggesting active exploration of the surroundings in many directions. While 3D projections provide rich details in depth and different viewing angles, we did not find an apparent relationship between birth order and dendrite targeting along the anterior- posterior axis, at least for the examined PN types at 12 hr APF. Thus, the approximate 2D projection (Figure  3E2–4) conveys the logic of dendrite patterning effectively. Dendrite targeting timing of larval-born adPN depends on birth order Having provided evidence for birth order–dependent spatial patterning of larval- born adPN dendrites, we next asked whether the timing of dendritic extension and targeting is also influenced by birth order. We noticed that the extent of dendritic innervation of 0 hr APF first- born DL1 adPNs resembled that of 6 hr APF mid- late born adPNs (compare Figure 3—figure supplement 3A1–4 with Figure 3— figure supplement 3B5–8). Such a resemblance was also seen between 0 hr APF mid- late and 6 hr APF late- born adPNs (compare Figure  3—figure supplement 3B1–4 with Figure  3—figure supplement 3C). Quantitative analyses of the exploring volume of dendrites and the number of terminal branches showed that, at 0 hr APF, DL1 PN dendrites were more elaborate than mid- late born PN dendrites (Figure 3—figure supplement 3F). By 6 hr APF, the mid- late born appeared to catch up, showing an extent of innervation comparable to DL1 PNs. We next examined when the dendrites reach their targeting regions. We found that whereas early larval- born adPNs (DL1, DA3, DC2) concentrated their dendrites to the dorsolateral corner by 6 hr APF (Figure 3B2; Figure 3—figure supplement 3A5–8), later- born PNs concentrated their dendrites to the medial/dorsomedial or peripheral dorsomedial side at 12 hr APF (Figure 3D4- 9; Figure 3—figure supplement 3B5- 8, C). Thus, our results suggest larval- born adPN dendrites innervate and pattern the antennal lobe using a ‘first born, first developed’ strategy. Contribution of lineage to early PN dendritic patterning Both lineage and birth order of PNs contributes to the eventual glomerular choice of their dendrites (Jefferis et al., 2001). What is the involvement of lineage in the prototypic map formation? Do lPN dendrites pattern the developing antennal lobe following similar rules as adPNs? To characterize lPN dendrite development at type–specific resolution, we used tsh- GAL4 to genetically access DA1/ DL3 lPNs, and MARCM clones of lPNs as a complementary approach (Figure 4). We focused on the dendritic patterns of tsh+ DA1/DL3 lPNs from 0 hr to 12 hr APF as tsh- GAL4 labeled additional PNs from 21 hr APF onwards (Figure 4A4–6; Figure 4—figure supplement 1B4–6; Figure 4—figure supple- ment 2; Figure 2—figure supplement 1). Examination of pupal brains double- labeled with DA1/DL3 lPNs (referred to as ‘middle larval- born lPNs’) and DL1/DA3 adPNs revealed that, like the early larval- born adPNs, dendritic growth of DA1/ Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 9 of 33 Developmental Biology | Neuroscience Research article Figure 4. Birth order–dependent spatial patterning of lPN dendrites in the developing antennal lobe. (A) Confocal images of fixed brains at indicated stages showing dendrite development of DL1/DA3 adPNs (CG14322+; labeled in yellow) and DA1/DL3 lPNs (tsh+; labeled in cyan). Right column of A1 shows a zoom- in of the dashed box. (A1): N=8; (A2): N=4; (A3): N=6; (A4): N=10; (A5): N=4; (A6): N=5. (B) MARCM clones (in cyan) of early (B1–3) and late (B4–6) larval- born lPNs in 12 hr APF pupal brains, generated by heat shocks (hs) at indicated times. In (B3), (B5), and (B6), single- cell clones of anterodorsal projection neuron (adPN) (yellow arrowheads) and lPN (cyan arrowheads) lineages were simultaneously labeled. Three biological samples are shown for each of the indicated lPN cohorts. B1–3: N=4; B4–6: N=6. (C) Summary of wiring logic of larval- born lPN dendrites to form an olfactory map in the 12 hr APF developing antennal lobe. (D) Summary of determination of dendrite targeting of larval- born PNs by lineage and birth order. See Figure 1 legend for common notations. The online version of this article includes the following video and figure supplement(s) for figure 4: Figure supplement 1. Dendrite development of DL1/DA3 and DA1/DL3 projection neurons (PNs). Figure supplement 2. Expression patterns of tsh in the developing antennal lobe during mid- pupal stages. Figure 4—video 1. 3D rendering of z stacks of indicated projection neurons (PNs) in 12 hr APF antennal lobe. https://elifesciences.org/articles/85521/figures#fig4video1 DL3 lPNs was evident by the wandering third instar larval stage (Figure 4—figure supplement 1A). At this stage, most DA1/DL3 lPN dendrites innervated the antennal lobe and intermingled with those of DL1/DA3 adPNs. From 0 hr to 12 hr APF, despite a high degree of overlap among those dendrites that explored the surroundings, DA1/DL3 lPN dendrites primarily targeted an area ventrolateral to those of DL1/DA3 adPNs (Figure 4A1–3; see 3D rendering in Figure 4—video 1). Such a spatial distinction was also observed between middle larval- born adPNs and lPNs in 0  hr and 6  hr APF pupal brains where occasionally single- cell clones from both lineages were simultaneously generated by MARCM Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 10 of 33 Developmental Biology | Neuroscience Research article (Figure 3—figure supplement 3D1–4, 7–10). Thus, at least some adPNs and lPNs sort their dendrites into distinct regions very early on regardless of birth timing. Next, we used MARCM to ask if lPNs born earlier and later than DA1/DL3 lPNs would send dendrites to regions different from that of DA1/DL3 lPNs. We found that dendrites of early- born lPNs primarily occupied the medial/dorsomedial side of the antennal lobe (Figure  4B1–3); we note that adPNs born at the same time sent dendrites to the dorsolateral side (see yellow arrowhead in Figure 4B3). Also, in contrast to the ventrolateral targeting of middle- born lPN dendrites, late- born lPNs sent dendrites to the dorsomedial corner (Figures 4B4–6). Like larval- born adPNs, late- born lPNs innervated the antennal lobe later than earlier- born lPNs (Figure 3—figure supplement 3D7–12–E, G). These data suggest that, at early pupal stages, lPN dendrites pattern the developing antennal lobe following similar rules as larval- born adPNs: adjacent birth order → similar dendrite targeting; distant birth order → distinct dendrite targeting; ‘first born, first developed.’ However, unlike the correlation of birth order and target positions in a rotational manner for adPNs (Figure 3E), the lPN dendritic map formation appears binary: early↔M/DM; middle↔VL; late↔DM (Figure  4C). Our type- specific characterization corroborated with the gross examination of the lPN dendrites as previously reported (Jefferis et al., 2004): at 12 hr APF, lPN dendrites mostly occupied the opposite corners along the dorsomedial- ventrolateral axis, leaving the middle of the axis largely devoid of lPN dendrites (arrow- heads in Figure 1D3). In summary, we propose that lineage and birth order of larval- born PNs contribute to their dendrite targeting in a combinatorial fashion (Figure 4D). The wiring logic of PN dendrites in the developing antennal lobe can, therefore, be represented by [lineage, birth order]=dendrite targeting; one can deduce the unknown if the other two are known. An explant system for time-lapse imaging of PN development at early pupal stages So far, we have identified wiring logic governing the initial dendritic map formation (Figures 3 and 4) by examining specifically labeled neuron types in the fixed brain at different developmental stages. To examine dendrite targeting at the higher spatiotemporal resolution, we established an early- pupal brain explant culture system based on previous protocols (Özel et  al., 2015; Rabinovich et  al., 2015; Li and Luo, 2021; Li et  al., 2021), and performed single- or dual- color time- lapse imaging with two- photon microscopy as well as adaptive optical lattice light- sheet microscopy (AO- LLSM) (Figure  5A–C). The following lines of evidence support that our explant system recapitulates key features of in vivo olfactory circuit development. First, during normal development, the morphology of the brain lobes changes from spherical at 0 hr APF to more elongated rectangular shapes at 15 hr APF (Rabinovich et al., 2015). After 22 hr ex vivo culture, the spherical hemispheres of brains dissected at 3 hr APF became more elongated, mimicking  ~15  hr APF in vivo brains characterized by the separation of the optic lobes from the central brain (Figure 5D). Second, dual- color, two- photon imaging of PNs every 20 min for 22 hr revealed that lPNs in 3 hr APF brains initially produced dynamic but transient dendritic protrusions in many directions, followed by extensive innervation into the antennal lobe (arrowheads in Figure 5E1–3; Figure 5—video 1). In higher brain centers, lPN axons clearly showed direction- specific outgrowth of collateral branches into the mushroom body calyx as well as forward extension into the lateral horn (arrows in Figure 5E3), thus resembling in vivo development (Figure 1—figure supplement 2). Third, larval- specific dendrites observed in 0 hr APF brains cultured for 12 hr ex vivo (orange arrow- head in Figure 5F4) were no longer seen in those cultured for 24 hr ex vivo (Figure 5F5), indicative of successful pruning and clearance of larval- specific dendrites. Also, the size of the developing antennal lobe in the brains cultured for 24  hr ex vivo increased considerably (Figure  5F5). These imply that olfactory circuit remodeling (degeneration of larval- specific processes and growth of adult- specific processes) proceeds normally, albeit at a slower rate (compare with Figure 5F1–3). Fourth, dendrites from genetically identified DL1 and DA1/DL3 PNs targeted to their stereotyped locations in the antennal lobe in 0 hr APF brains cultured for 24 hr ex vivo (Figure 5G), mimicking in vivo development (Figure 4A). Finally, the segregation of dendrites of PNs targeting to neighboring proto- glomeruli could be recapitulated in brains dissected at 24 hr APF and cultured for 8 hr (Figure 5—figure supplement 1; Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 11 of 33 Developmental Biology | Neuroscience Research article Figure 5. Establishment of an explant system for time- lapse imaging of olfactory map formation. (A) Schematic of the anatomical organization of the olfactory circuit in early pupal brain (0–3 hr APF). Green, red, and blue denote embryonic- born adPN, larval- born anterodorsal projection neuron (adPN), and larval- born lPN, respectively. MB: mushroom body; LH: lateral horn. (B) Schematic of explant culture system for early pupal brains. Wells created in the Sylgard plate from which brains were imbedded are shown in blue. (C) Schematic of explant culture and imaging system for early pupal brains. (D)  Top: Schematic of morphological changes of brain lobes from 0 hr to ~15 hr APF during normal development. Bottom: Morphologies of a brain explant dissected at 3 hr APF and cultured for 0 hr ex vivo and cultured for 22 hr ex vivo. (E) Two- photon time- lapse imaging of adPNs (VT033006+ run+ ; labeled in magenta) and lPNs (VT033006+ run–; labeled in green) in pupal brain dissected at 3 hr APF and cultured for 0–22 hr ex vivo. Arrowheads mark dynamic but transient dendritic protrusions of lPNs in E1, 2, and extensive dendritic innervation of lPNs in (E3). Arrows in (E3) mark axonal innervation of lPNs in the mushroom body calyx and lateral horn. N=3. (F) Confocal images of antennal lobes labeled by VT033006+ projection neurons (PNs) (in green) at 0 hr (F1), 6 hr (F2), and 12 hr (F3) APF in vivo. Confocal images of antennal lobes labeled by VT033006+ PNs in pupal brains were dissected at 0 hr APF and cultured for 12 hr (F4) and 24 hr (F5) ex vivo. (F1): N=6; (F2): N=5; (F3): N=6; (F4): N=8; (F5): N=8. (G) Dendrite targeting regions of DL1 PNs (71B05+; in yellow; G1) and DA1/DL3 PNs (tsh+; in cyan; G2) in the antennal lobes in pupal brains dissected at 0 hr APF and cultured for 24 hr ex vivo. Antennal lobes are revealed by N- Cadherin (Ncad; in blue) staining. (G1): N=5; (G2): N=6. See Figure 1 legend for common notations. The online version of this article includes the following video, source data, and figure supplement(s) for figure 5: Figure supplement 1. Dendritic segregation of DC3/VA1d adPNs and DA1 lPNs targeting neighboring proto- glomeruli. Figure supplement 1—source data 1. Source data for Figure 5—figure supplement 1C and D. Figure 5—video 1. Two- photon time- lapse imaging of projection neuron (PN) development. https://elifesciences.org/articles/85521/figures#fig5video1 Figure 5—video 2. Two- photon time- lapse imaging of projection neuron (PN) dendritic segregation. https://elifesciences.org/articles/85521/figures#fig5video2 Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 12 of 33 Developmental Biology | Neuroscience Research article Figure 5—video 2). Specifically, despite constant dynamic interactions among dendrites that explore the surroundings (arrowheads in Figure 5—figure supplement 1A2–4), DC3/VA1d and DA1 PNs exhib- ited a 1–2 µm increase in the distance between centers of the two dendritic masses and a substantial decrease in the overlap of their core targeting regions (Figure 5—figure supplement 1B–D). Taken together, these data support that the explant culture and imaging system established here reliably captures key neurodevelopmental events starting from early pupal stages. Single-cell, two-photon imaging reveals active dendrite targeting Our observation in fixed brains revealed that dendrites of DL1 adPNs transition from a uniform exten- sion in the antennal lobe at 0 hr APF to concentration at the dorsolateral corner of the antennal lobe at 6 hr APF (Figure 3A). To identify mechanisms of dendrite targeting specificity that could be missed in static developmental snapshots, we performed two- photon time- lapse imaging of single- cell MARCM clones of DL1 PNs in 3 hr APF brains (Figure 6; Figure 6—figure supplement 1; Figure 6—video 1). Although we did not have a counterstain outlining the antennal lobe, we could use the background signals to discern the orientation of DL1 PNs in the brain (Figure  6—figure supplement 1A). The final targeting regions relative to the antennal lobe revealed by post hoc fixation and immunostaining confirmed proper dendrite targeting (yellow arrowhead in Figure 6A10; Figure 6—figure supplement 1B–C). Using DL1 PN in Figure  6A (pseudo- colored in yellow; Figure  6—video 1) as an example, we observed that the PN initially extended dendrites in every direction (Figure  6A1–3), like what we observed in fixed tissues (Figure 3A1). The first sign of active targeting emerged at 2 hr 20 min ex vivo when DL1 PN began to generate long, albeit transient, dendritic protrusions in the dorsolateral direc- tion; these selective protrusions were more prominent at 3 hr ex vivo (arrowheads in Figure 6A4–6). The dorsolateral targeting continued to intensify, leading to the formation of a highly focal dendritic mass seen at 13 hr ex vivo (arrowhead in Figure 6A8). As the dendrites reached the dorsolateral corner and explored locally, the change in shape appeared less pronounced (Figure 6A9). To quantitatively characterize the active targeting process, we categorized the bulk dendritic masses emanating from the main process according to their targeting directions: DL, DM, VM, and VL (Figure 6B). During the initial phase, the percentage of dendritic volume in each direction varied from 10% to 40% (Figure 6C and D), indicative of active exploration with little targeting specificity. Despite these variations, the total amount of dendritic mass seen in the VM direction over the entire imaging time (area under the graph of Figure  6C) was the smallest across all samples examined (Figure 6E). The initial phase of exploration in every direction was followed by a ~4 hr transitional phase during which DL1 PNs predominantly extended dendrites in 2 of the 4 directions (Figure 6C; Figure 6—figure supplement 1D–E). One of the 2 directions was always DL whereas the other was either DM or VL but never VM. In the final phase, DL1 PN dendrites always preferred DL out of the two available directions. Lastly, we analyzed the bulk dendritic movements. We defined bulk extension and retraction events when dendrites respectively extended and retracted more than 2 μm between two consecutive time frames. The analyses showed a striking shift from frequent extension and retraction towards stabilization, reflecting the pre- and post- targeting dynamics, respectively (Figure 6F and G). Hence, long- term two- photon imaging of single- cell DL1 PNs revealed that dendrite targeting specificity increases over time via active targeting in a specific direction and stepwise elimination of unfavorable trajectory choices (see summary in Figure 7F1–3). AO-LLSM imaging suggests a cellular mechanism underlying dendrite targeting specificity To capture fast dynamics of single dendritic branches, we performed dual- color adaptive optical lattice sheet microscopy (AO- LLSM) imaging (Chen et al., 2014; Wang et al., 2014; Liu et al., 2018) of PNs every 30 s for 15 min, following a protocol we recently established (Li et  al., 2021; Li and Luo, 2021). We selected 3 hr, 6 hr, and 12 hr APF pupal brains double- labeled with DL1 PNs and bulk adPNs (Figure 7A–C; Figure 7—videos 1–3). The labeling of adPNs with GFP outlined PN cell bodies and the developing antennal lobe but not the degenerating one, presumably because the GFP in larval- specific dendrites was quickly quenched upon glial phagocytosis (Marin et al., 2005). In the 15 min imaging window, we observed four types of terminal branches regardless of neuronal types or developmental stages: (1) stable branch that existed throughout the entire imaging time, Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 13 of 33 Developmental Biology | Neuroscience Research article Figure 6. Two- photon time- lapse imaging reveals active dendrite targeting. (A) Two- photon time- lapse imaging of MARCM- labeled DL1 projection neuron (PN) (pseudo- colored in yellow) in a brain dissected at 3 hr APF and cultured for 21 hr ex vivo (A1–9). Arrowheads in A4–6 denote protrusions of dendritic branches towards the dorsolateral direction. After 21 hr culture, the explant was fixed and immuno- stained for N- Cadherin (Ncad; in blue) to outline the developing antennal lobe (A10). Yellow and cyan arrowheads indicate DL1 PN dendrites and processes of other GH146+ cells, respectively. (B) Neurite tracing of DL1 PN at the beginning of live imaging (3 hr APF + 0 hr ex vivo). Dendrites are categorized based on the directions to which they extend and color- coded accordingly. (C) Left: Quantification of the percentage of dendritic volume in indicated direction during the time- lapse imaging period reveals a transitional phase during which dendrites were found in only two out of the four directions. Right: Schematic of the initial, transitional, and final phases during the course of targeting. ‘½’ denotes the reduction of available trajectory directions by half. Timestamp 00:00 refers to HH:mm; H, hour; m, minute. See Figure 6—source data 1. (D) Quantification of the percentage of DL1 PN dendritic volume in an indicated direction in 3 hr APF cultured brains at the beginning (0 hr ex vivo) and at/near the end of imaging (18 hr ex vivo). DL1 PN sample size = 3. t- test; *p<0.05. Timestamp 00:00 refers to HH:mm; H, hour; m, minute. (E) Quantification of the percentage of the sum of DL1 PN dendritic volume in indicated directions throughout the entire imaging time. DL1 PN sample size = 3. (F) Bulk dendrite dynamics of DL1 PN in Figure 6A. Each row represents bulk dendritic dynamics in the indicated direction (color- coded as in Figure 6B) across the 21 hr imaging period. Each block represents a 20 min window. Bulk extension (in green) and retraction (in magenta) events are defined as dendrites extending and retracting more than 2 μm between two consecutive time windows. The first and last six consecutive windows refer to the initial and final phases of imaging. (G) Quantification of the number of bulk extension and retraction events in the dorsolateral direction during the initial and final phases of imaging. DL1 PN sample size = 3. t- test; *p<0.05. The online version of this article includes the following video, source data, and figure supplement(s) for figure 6: Source data 1. Source data for Figure 6C–G and Figure 6—figure supplement 1D and E. Figure supplement 1. Two- photon time- lapse imaging of DL1 projection neuron (PNs). Figure 6—video 1. Two- photon time- lapse imaging of DL1 projection neuron (PN) dendrites. https://elifesciences.org/articles/85521/figures#fig6video1 Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 14 of 33 Developmental Biology | Neuroscience Research article (2) transient branch that was produced and eliminated within the imaging window, (3) emerging branch that was produced after imaging began, and (4) retracting branch that was eliminated within the imaging period (Figure  7—figure supplement 1A). To examine if terminal branch dynamics exhibit any directional preference, we assigned the branches according to their targeting directions (Figure  7D). Extension and retraction events were defined when the speed exceeded 0.5 μm/min. Terminal branches were selected for analyses as branches closer to the main process were too dense to resolve. Figure 7D1- 3 showed the dynamics of ~15 randomly selected terminal branches in each direction from the representative 3 hr, 6 hr, and 12 hr APF DL1 PNs (Figure 7A–C). Quantitative analyses revealed that at 3  hr APF, DL1 PNs constantly produced, eliminated, extended, and retracted dendritic branches (Figure 7A, Figure 7D1, Figure 7—video 1). Even stable branches were not immobile. Rather, they spent comparable amounts of time extending and retracting at ~1.5 μm/min (Figure 7—figure supplement 1A1, 1B). Transient, emerging, and retracting branches had similar, but more variable speeds, ranging from 1 to 2.5 μm/min. Although there was no correla- tion between targeting direction and frequency/speed of extension/retraction, the number of stable branches in the VM direction was significantly lower than in other directions across all 3 hr DL1 PN samples examined (Figure 7E1). This suggests that even though dendritic branches were developed in every direction at the early stages, those branches in the VM direction were short- lived and might be eliminated by retraction. The direction- dependent stability/lifespan of dendritic branches on the timescale of seconds uncovered from AO- LLSM imaging explains why bulk dendrites in unfavorable trajectories failed to persist in long- term two- photon imaging. From 6 hr to 12 hr APF, DL1 PNs no longer manifested direction- specific branch de/stabiliza- tion (Figure 7B–C, Figure 7D2–3, Figure 7—videos 2–3). At the same developmental stage, stable branches in one direction appeared indistinguishable from those in other directions in terms of abun- dance, frequency, and speed (Figure 7D2–3, Figure 7—figure supplement 1C–D). This suggests that the entire dendritic mass tends to stay in equilibrium upon arrival at target regions. At 12 hr APF, the abundance of stable branches of DL1 PNs was the highest (Figure 7D–E1). Also, the stable branches of 12 hr APF DL1 PNs moved at a significantly lower speed (~1 μm/min) (Figure 7E2) and spent more time being stationary than those at 3 hr and 6 hr (Figure 7—figure supplement 1B–D). The reduced branch dynamics at 12 hr APF is consistent with observations from two- photon imaging showing fewer bulk extension/retraction events in the final phase of targeting (Figure 6F–G). Despite the slowdown, dendritic arborization was evident in terminal branches of 12  hr APF DL1 PNs (Figure  7—figure supplement 1E), suggesting that PN dendrites are transitioning from simple to complex branch architectures. Although it remains unclear if there is a causal relationship between reduced branch dynamics and increased structural complexity, we propose that both contribute to the sustentation of dendrite targeting specificity. In summary, AO- LLSM imaging reveals that PNs selectively stabilize branches in the direction towards the target and destabilize those in the opposite direction, providing a cellular basis of dendrite targeting specificity. Upon arrival at the target, the specificity is sustained through branch stabilization in a direction- independent manner (summarized in Figure 7F4–7). Embryonic-born PNs timely integrate into an adult olfactory circuit by simultaneous dendritic pruning and re-extension In earlier sections, we uncovered wiring logic of larval- born PN dendritic patterning and cellular mech- anisms of dendrite targeting specificity used to initiate olfactory map formation (Figures 3–7). In this final section, we focused on embryonic- born PNs, which participate in both larval and adult olfactory circuits by reorganizing their processes (Marin et al., 2005). Our previous study demonstrates that embryonic- born PNs prune their larval- specific dendrites during early metamorphosis (Marin et al., 2005; Figure  1D1–3). Here, we examined when and how embryonic- born PNs re- extend dendrites used in the adult olfactory circuit. It is known that γ neurons of Drosophila mushroom body (γ Kenyon cells) and sensory Class IV dendritic arborization (C4da) neurons prune their processes between 4 hr and 18 hr APF and show no signs of re- extension at 18 hr APF (Lee et al., 2000; Watts et al., 2003; Lee et al., 2009). Do embryonic- born adPNs follow a similar timeframe? We first examined developing brains double- labeled for embryonic- born DA4l/VA6/VA2 adPNs (collectively referred to as ‘lov+ PNs’) and early larval- born DC2 adPNs (Figure 8A; Figure 8—figure supplement 1). We found that, by 12 hr APF, Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 15 of 33 Developmental Biology | Neuroscience Research article Figure 7. AO- LLSM time- lapse imaging reveals cellular mechanisms of dendrite targeting specificity. (A–C) AO- LLSM imaging of DL1 projection neurons (PNs) (71B05+; labeled in yellow) and anterodorsal projection neurons (adPNs) (acj6+; labeled in blue) in cultured brains dissected at 3 hr (A), 6 hr (B), and 12 hr (C) APF. Zoom- in, single z- section images of (A1), (B1), and (C1) (outlined in dashed boxes) are shown in A2, B2 and C2, respectively. (D) Single dendritic branch dynamics of 3 hr (D1), 6 hr (D2), and 12 hr (D3) DL1 PNs shown in A–C. Terminal branches are analyzed and categorized based on the directions in which they extend. Their speeds are color- coded using purple- gray- green gradients (negative speeds, retraction; positive speeds, extension). Individual branches are also assigned into four categories: stable, transient, emerging, and retracting (color- coded on the right; see Figure 7— figure supplement 1A). Each block represents a 30s window. Each row represents individual branch dynamics across the 15 min imaging period. (E) Quantification of the abundance (in percentage) of DL1 PN stable branches in indicated direction at 3 hr, 6 hr, and 12 hr (E1). Average speed of DL1 PN stable branches in indicated direction at 3 hr, 6 hr, and 12 hr (E2). DL1 PN sample size: 3 hr=4; 6 hr=3; 12 hr=3. Error bars, SEM; t-test; One- way ANOVA; *p<0.05; n.s., p≥0.05. SEM, standard error of the mean; n.s., not significant. See Figure 7—source data 1. (F) Summary of mechanisms underlying the emergence of dendrite targeting specificity revealed by two- photon and AO- LLSM imaging of DL1 PN dendrites. The online version of this article includes the following video, source data, and figure supplement(s) for figure 7: Source data 1. Source data for Figure 7E. Figure supplement 1. Analyses of DL1 projection neuron (PN) dendritic branches captured by AO- LLSM imaging. Figure 7—video 1. AO- LLSM time- lapse imaging of 3 hr DL1 projection neuron (PN) dendrites. https://elifesciences.org/articles/85521/figures#fig7video1 Figure 7—video 2. AO- LLSM time- lapse imaging of 6 hr DL1 projection neuron (PN) dendrites. https://elifesciences.org/articles/85521/figures#fig7video2 Figure 7—video 3. AO- LLSM time- lapse imaging of 12 hr DL1 projection neuron (PN) dendrites. https://elifesciences.org/articles/85521/figures#fig7video3 Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 16 of 33 Developmental Biology | Neuroscience Research article lov+ PNs already sent adult- specific dendrites to a region ventromedial to DC2 PN dendrites (green arrowhead in Figure 8A3; see 3D rendering in Figure 8—video 1). This implies that lov+ PNs have already caught up with DC2 PNs on dendrite development at this stage, and the re- extension of lov+ PN dendrites must have happened even earlier. Indeed, we observed lov+ PN dendrites innervated the developing antennal lobe extensively at 6 hr APF (Figure 8A2). Such innervation was not observed at 0  hr APF (Figure  8A1). After 12  hr APF, the time course of lov+ PN dendrite development was comparable to that of DC2 PNs (Figure 8A4–6). To characterize dendritic re- extension at single- cell resolution, we developed a sparse, stochastic labeling strategy to label single lov+ PNs (Figure  8B). We found that lov+ PNs produced nascent branches from the main process dorsal to larval- specific dendrites as early as 3 hr APF (Figure 8C2–3; arrowheads in Figure 8C6–7). At 6 hr APF, when larval- specific dendrites were completely segregated from lov+ PNs, the robust extension of adult- specific dendrites was seen across the developing antennal lobe (Figure 8C4). These data indicate that lov+ PNs re- extend their adult- specific dendrites at a more dorsal location before the larval- specific dendrites are completely pruned. Do other embryonic- born PNs prune and re- extend their dendrites simultaneously? Like lov drivers, Mz612- GAL4 labels embryonic- born PNs, one of which is VA6 PN (Marin et al., 2005). In 3 hr APF brains co- labeled for Mz612+ and lov+ PNs, we could unambiguously access three single embryonic- born PN types: (1) lov+ Mz612– PN, (2) lov– Mz612+ PN, and (3) lov+ Mz612+PN (Figure 8—figure supplement 2A–B). Tracing of individual dendritic branches showed that all these PNs already re- ex- tended dendrites to varying extents prior to the separation of larval- specific dendrites from the rest of the processes (Figure 8—figure supplement 2C). Thus, concurrent pruning and re- extension apply to multiple embryonic- born PN types. To capture the remodeling at the higher temporal resolution, we performed two- photon time- lapse imaging of single embryonic- born PNs labeled by Split7- GAL4 (Figure 8D, Figure 8—video 2, Figure 8—figure supplement 3). This GAL4 labels one embryonic- born PN (either VA6 or VA2 PN) at early pupal stages but eight PN types at 24 hr APF (Xie et al., 2021). Initially (3 hr APF + 0 hr ex vivo), no adult- specific dendrites were detected in live Split7+ PNs (Figure 8D1). The following ~3 hr ex vivo saw thickening of the main process (arrowhead in Figure 8D3). From 4 hr ex vivo onwards, re- extension occurred in the presumed developing antennal lobe located dorsal to larval- specific dendrites (arrowheads in Figure 8D4–8; see traces in Figure 8D9). Live imaging of Split7+ PNs also revealed that fragmentation of larval- specific dendrites occurred at the distal ends (Figure 8—figure supplement 3B1–5), and the process leading to larval- specific dendrites gradually disappeared as pruning approached completion (Figure 8—figure supplement 3B6–10). These observations suggest that pruning of embryonic- born PN dendrites is not initiated by severing at the proximal end. Distal- to- proximal pruning, rather than in the reversed direction, further supports concurrent but spatially segregated pruning and re- extension processes. It has been shown that dendritic pruning of embryonic- born PNs requires ecdysone signaling in a cell- autonomous manner (Marin et al., 2005). We asked if the re- extension process also depends on ecdysone signaling. We expressed a dominant negative form of ecdysone receptor (EcR- DN) in most PNs (including lov+ PNs) and monitored the development of lov+ PN dendrites (Figure  8— figure supplement 4). We found that inhibition of ecdysone signaling by EcR- DN expression not only suppressed pruning, but also blocked re- extension. This is consistent with a previous study reporting the dual requirement of ecdysone signaling in the pruning and re- extension of Drosophila anterior paired lateral (APL) neurons, although, unlike embryonic- born PNs, APL neurons prune and re- extend processes sequentially (at 6 hr and 18  hr APF, respectively) (Mayseless et  al., 2018). We currently could not distinguish if the lack of re- extension is due to defective pruning, or if ecdysone signaling controls pruning and re- extension independently. Taken together, our data demonstrate that embryonic- born PNs prune and re- extend dendrites simultaneously at spatially distinct regions, and that both processes require ecdysone signaling (Figure 8E). Such a ‘multi- tasking’ ability explains how embryonic- born PNs can re- integrate into the adult olfactory circuit and engage in its prototypic map formation in a timely manner. Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 17 of 33 Developmental Biology | Neuroscience Research article Figure 8. Embryonic- born projection neurons (PNs) timely participate in olfactory map formation via simultaneous pruning and re- extension. (A) Confocal images of fixed brains at indicated stages showing dendrite development of lov+ PNs (embryonic- born; labeled in green) and 91G04+DC2 PNs (larval- born; labeled in magenta). As 91G04- GAL4 also labels some embryonic- born PNs from 0 to 6 hr APF, their processes are found in the larval- specific antennal lobe (A1, 2). Right columns of A1, 2 show a zoom- in of the dashed boxes. Green arrowhead in (A2) indicates robust dendrite re- extension of embryonic- born PNs across the developing antennal lobe at 6 hr APF. (A1): N=6; (A2): N=12; (A3): N=9; (A4): N=12; (A5): N=9; (A6): N=5. (B) Schematic of the sparse, stochastic, and dual- color labeling strategy. In this strategy, the same cell has one copy of UAS- responsive conditional reporter 1 and one copy of QUAS- responsive reporter 2, both of which are integrated into the same 86Fb genomic locus (i.e. UAS- FRT- stop- FRT- reporter1/QUAS- FRT- stop- FRT- reporter2). FLP expression yields cis and trans recombination of FRT sites in a stochastic manner. Upon GAL4 expression, reporter 1 is expressed in cells with cis recombination, whereas reporter 2 is expressed only when cis and trans recombination events co- occur. (C) Sparse labeling of lov+ PNs (labeled in green; single- cell lov+ PNs in gray) at indicated developmental stages. (C6) and (C7) are zoom- in images of the rectangular boxes in (C2) and (C3), respectively. Arrowheads indicate nascent, adult- specific dendrites. Larval- specific dendrites are outlined by dashed orange lines. Arrows indicate axons projecting towards high brain centers. (C1): N=6; (C2–3): N=6; (C4): N=4; (C5): N=4. (D) Two- photon time- lapse imaging of a single Figure 8 continued on next page Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 18 of 33 Developmental Biology | Neuroscience Research article Figure 8 continued embryonic- born PN (Split7+; pseudo- colored in yellow) in a brain dissected at 3 hr APF and cultured for 23 hr ex vivo. Arrowhead in (D3) denote the thickening of the main process. Arrowheads in D4, 5 denote dendritic protrusions dorsal to larval- specific dendrites. (D9) shows neurite tracing of the embryonic- born PN. Triangles in (D9) indicate the degenerating larval- specific dendrites. N=3. (E) Schematic summary of remodeling of embryonic- born PN dendrites. Following simultaneous pruning and re- extension, embryonic- born PNs timely integrate into an adult olfactory circuit and, together with larval- born PNs, participate in the prototypic map formation. The online version of this article includes the following video and figure supplement(s) for figure 8: Figure supplement 1. Dendrite development of lov+ embryonic- born projection neurons (PNs). Figure supplement 2. Dendrite re- extension of lov+ and Mz612+ embryonic- born projection neurons (PNs). Figure supplement 3. Two- photon time- lapse imaging of Split7+ projection neuron (PN) dendrites. Figure supplement 4. Dual requirement of ecdysone signaling in pruning and re- extension of embryonic- born projection neuron (PN) dendrites. Figure 8—video 1. 3D rendering of z stacks of indicated projection neurons (PNs) in 12 hr APF antennal lobe. https://elifesciences.org/articles/85521/figures#fig8video1 Figure 8—video 2. Two- photon time- lapse imaging of Split7+ projection neuron (PN) dendrites. https://elifesciences.org/articles/85521/figures#fig8video2 Discussion Wiring logic for the prototypic olfactory map Prior to this study, no apparent logic linking PN lineage, birth order, and adult glomerular position has been found. Our systematic analyses of dendritic patterning at the resolution of specific PN types across development identified wiring logic underlying the spatial organization of the prototypic olfac- tory map (Figures 3 and 4). We found that PNs of a given lineage and temporal cohort share similar dendrite targeting spec- ificity and timing. Notably, dendrites of adPNs and lPNs respectively pattern the antennal lobe in rotating and binary manners following birth order. Based on our new observations and previous find- ings, we discuss possible mechanisms that execute the wiring logic to form the initial map: (1) speci- fication of the initial dendrite targeting through combinatorial inputs from lineage and birth order, (2) PN dendrite- dendrite interactions, and (3) contribution of the degenerating larval- specific antennal lobe. The spatial distinctions of cell bodies (e.g. Figure 1D1), axons (e.g. Figure 1—figure supplement 2A), and dendrites (e.g. Figure 4A1) of adPNs and lPNs observed in 0 hr APF pupal brain suggest that lineage endows projection specificity very early on. Lineage- specific transcription factors have been identified to instruct PN neurite targeting (Komiyama et al., 2003; Komiyama and Luo, 2007; Li et al., 2017; Xie et al., 2022), which might explain the differences between the adPN and lPN dendritic maps. Nonetheless, lineage alone does not account for the characteristic dendritic patterns. Rather, dendrite targeting can be predicted using combinatorial inputs from lineage and birth order. This combinatorial strategy is also seen in neuronal fate diversification and wiring of the Drosophila optic lobe and ventral nerve cord (Erclik et  al., 2017; Mark et  al., 2021), suggesting that it is a general principle in wiring the fly brain and likely also used in vertebrates (Holguera and Desplan, 2018; Sen, 2023). Substantial advances have been made in understanding how temporal patterning arises for intra- lineage specification (Doe, 2017; Miyares and Lee, 2019). For instance, the embry- onic ventral nerve cord neuroblasts sequentially express a cascade of temporal transcription factors (TTFs) to specify temporal identity (Isshiki et al., 2001). Larval optic lobe neuroblasts also deploy the same strategy but use a completely different TTF cascade (Li et al., 2013). Earlier studies show Chinmo, a TTF, and RNA- binding proteins that regulate Chinmo translation, control neuronal cell fate of the adPN lineage (Zhu et al., 2006; Liu et al., 2015). Specifically, DL1 PNs mutant for Chinmo project dendrites to D glomerulus that is targeted by the fourth larval- born adPNs (Zhu et al., 2006), demonstrating temporal order specifies final glomerular targeting. However, whether approximate temporal cohorts of a given PN lineage we described arise from sequential expression of temporal factors, and how such factors translate into initial dendrite patterning remains a fertile ground for future studies. Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 19 of 33 Developmental Biology | Neuroscience Research article Our time- lapse imaging data reveals robust PN dendritic dynamics during the initial targeting process (Figures  5–8), suggesting that cellular interactions among PN dendrites contribute to the initial map formation. This appears to contrast with the PN- ORN map in the mature antennal lobe, which is highly stable; connection specificity remains largely unchanged upon genetic ablation of their synaptic partners (Berdnik et al., 2006). Future works using early- onset genetic drivers for specific PN types for ablation can be used to investigate interactions between different PN groups, such as adPNs and lPNs, in the construction of the initial PN dendrite map. Does the degenerating larval- specific antennal lobe contribute to the initial dendrite patterning of the developing adult- specific antennal lobe? Earlier studies found that the larval- specific ORN axons secrete semaphorins, Sema- 2a and Sema- 2b, which act as repulsive ligands for dendrites of Sema- 1a- expressing PNs (including DL1 PNs) (Komiyama et  al., 2007; Sweeney et  al., 2011). As the larval- specific lobe is located ventromedial to the adult- specific lobe, Sema- 2a/b and Sema- 1a form opposing gradients along the dorsolateral- ventromedial axis. When DL1 PNs (the first- born/ developed) begin to target their dendrites, this repulsive action could destabilize branches in the ventromedial direction and thus favor dorsolateral targeting. This provides a plausible explanation as to why the adPN rotation pattern begins at the dorsolateral position. It would be interesting to see if the pattern is perturbed upon ablation of larval- specific ORNs. Our new tools for labeling and genetic manipulation of distinct PN types (Figure 2) will now enable in- depth investigations into the potential cellular interactions and molecular mechanisms leading to the initial map organization. Wiring logic evolves as development proceeds After the initial map formation at 12  hr APF, dendrite positions in the antennal lobe could change substantially in the next 36 hr (for example, see DC2 PNs in Figure 3B4–6 and DA1 and VA1d/DC3 PNs in Figure 3C4–7). These changes occur when dendrites of PNs with neighboring birth order begin to segregate and when ORN axons begin to invade the antennal lobe. Accordingly, the ovoid- shaped antennal lobe turns into a globular shape (30–50 hr APF; Figure 3C6- 7). These PN- autonomous and non- autonomous changes likely mask the initial wiring logic, explaining why previous studies, which mostly focused on examining the final glomerular targets in adults (Jefferis et al., 2001), have missed the earlier organization. Interestingly, the process of PN dendritic segregation coincides with the peak of PN transcriptomic diversity at 24 hr APF (Li et al., 2017; Xie et al., 2021). Recent proteomics and genetic analyses have indicated that PN dendrite targeting is mediated by cell- surface proteins cooperating as a combinatorial code (Xie et al., 2022). The evolving wiring logic, which is consistent with the stepwise assembly of an olfactory circuit (Hong and Luo, 2014), suggests the combinatorial codes are not static. We propose that PNs use a numerically simpler code for initial dendrite targeting. Following the expansion of transcriptomic diversity, PNs acquire a more complex code mediating dendritic segregation of neighboring PNs and matching of PN dendrites and ORN axons. Functional characterization of differentially expressed genes between 12 hr and 24 hr APF PNs may provide molecular insights into how the degree of discreteness in the olfactory map arises. Although the initial wiring logic is not apparent in the final map, several lines of evidence suggest the final map depends on the initial map. First, as mentioned above, the change of the temporal identity of DL1 PNs affects glomerular targeting (Zhu et al., 2006). Second, loss of Sema- 1a in DL1 PNs occasionally causes mistargeting in areas outside of the antennal lobe, and dendrite mistargeting phenotype along the dorsolateral- ventromedial axis is persistent across development as well as in adulthood (Komiyama et  al., 2007). Our work thus demonstrates that identification of the wiring logic in the early stages should help us better resolve the architectures in complex neural circuits. Selective branch stabilization as a cellular mechanism for dendrite targeting Utilizing an early pupal brain explant culture system coupled with two- photon and AO- LLSM imaging (Figure 5), we presented the first time- lapse videos following dendrite development of a specific PN type – DL1 PNs (Figures 6 and 7). We found that DL1 PN dendrites initiate active targeting towards their dorsolateral target with direction- dependent branch stabilization. This directional selectivity provides a cellular basis for the emerging targeting specificity of PN dendrites at the beginning of olfactory map formation. Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 20 of 33 Developmental Biology | Neuroscience Research article Although selective branch stabilization as a mechanism to achieve axon targeting specificity has been described in neurons in the vertebrate and invertebrate systems (e.g. Yates et  al., 2001; Li et  al., 2021), our time- lapse imaging showed, for the first time to our knowledge, that selective branch stabilization is also used to achieve dendrite targeting specificity. Furthermore, AO- LLSM imaging revealed that selective stabilization and destabilization of dendritic branches occur on the timescale of seconds. As the rate of olfactory circuit development in the brain explants was slower than normal development (Figure 5F), we might have captured PN dendritic dynamics in slow motion. Using AO- LLSM for high spatiotemporal resolution imaging, we just begin to appreciate how fast PN dendrites are coordinating trajectory choices with branch stabilization to make the appropriate deci- sion. Having characterized the dendritic branch dynamics of the wild- type DL1 PNs, we have set the stage for future studies addressing how positional cues and the downstream signaling instruct wiring, and whether other PN types follow similar rules as DL1 PNs. Simultaneous pruning and re-extension as novel remodeling mechanism for neuronal remodeling Our data on embryonic- born adPN dendrite development reveals a novel mode of neuronal remod- eling during metamorphosis (Figure  8). In mushroom body γ neurons and body wall somatosen- sory neurons, two well- characterized systems, larval- specific neurites are first pruned, followed by re- extension of adult- specific processes (Watts et al., 2003; Williams and Truman, 2005; Yaniv and Schuldiner, 2016). However, embryonic- born adPNs prune larval- specific dendrites and re- extend adult- specific dendrites simultaneously but at spatially separated subcellular compartments. Such spatial segregation suggests that regional external cues could elicit compartmentalized downstream signals leading to opposite effects on the dendrites. Subcellular compartmentalization of signaling and cytoskeletal organization has been observed in diverse neuron types across species (Rolls et al., 2007; Kanamori et al., 2013; O’Hare et al., 2022). Why do embryonic- born adPNs ‘rush’ to re- extend dendrites? During normal development, it takes at least 18 hr for embryonic- born adPNs to produce and properly target dendrites (growth at 3–6 hr APF, initial targeting at 6–12  hr APF, and segregation at 21–30  hr APF). Given that the dendritic re- extension of embryonic- born PNs is ecdysone dependent (Figure 8—figure supplement 4), if the PNs did not re- extend dendrites at 3 hr APF, they would have to wait for the next ecdysone surge at ~20 hr APF (Thummel, 2001), which might be too late for their dendrites to engage in the proto- typic map formation. Thus, embryonic- born PNs develop a remodeling strategy that coordinates with the timing of systemic ecdysone release. By simultaneous pruning and re- extension, embryonic- born adPNs timely re- integrate into the adult prototypic map that readily serves as a target for subsequent ORN axon innervation. In conclusion, our study highlights the power and necessity of type- specific neuronal access and time- lapse imaging to identify wiring logic and mechanisms underlying the origin of an olfactory map. Applying similar approaches to other developing neural maps across species should broaden our understanding of the generic and specialized designs that give rise to functional maps with diverse architectures. Materials and methods Drosophila stocks and husbandry Flies were maintained on a standard cornmeal medium at 25 °C. Fly lines used in this study included GH146- FLP (Hong et  al., 2009), QUAS- FRT- stop- FRT- mCD8- GFP (Potter et  al., 2010), UAS- mCD8- GFP (Lee and Luo, 1999), UAS- mCD8- FRT- GFP- FRT- RFP (Stork et al., 2014), VT033006- GAL4 (Tirian and Dickson, 2017), Mz19- GAL4 (Jefferis et al., 2004), 91G04- GAL4 (Jenett et al., 2012), Mz612- GAL4 (Marin et al., 2005), 71B05- GAL4 (Jenett et al., 2012), Split7- GAL4 (Xie et al., 2021), QUAS- FLP (Potter et al., 2010), and UAS- EcR.B1-ΔC655.F645A (Cherbas et al., 2003). The following GAL4 lines were obtained from Bloomington Drosophila Stock Center (BDSC): tsh- GAL4 (BDSC #3040) and lov- GAL4 (BDSC #3737). The following two stocks were used for MARCM analyses: (1) UAS- mCD8- GFP, hs- FLP; FRTG13, tub- GAL80;; GH146- GAL4, and (2) FRTG13, UAS- mCD8- GFP (Lee and Luo, 1999). Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 21 of 33 Developmental Biology | Neuroscience Research article The following lines were generated in this study: UAS- FRT10- stop- FRT10- 3xHalo7- CAAX (on either II or III chromosome), UAS- FRT- myr- 4xSNAPf- FRT- 3xHalo7- CAAX (III), UAS- FRT- myr- mGreenLantern- FRT- 3xHalo7- CAAX (II), QUAS- FRT- stop- FRT- myr- 4xSNAPf (III), run- T2A- FLP (X), acj6- T2A- FLP (X), acj6- T2A- QF2 (X), CG14322- T2A- QF2 (III), and lov- T2A- QF2 (II). Drosophila genotypes tub- GAL80/FRTG13, UAS- mCD8- GFP;; supplement 1B: GH146- FLP/UAS- FRT10- stop- FRT10- 3xHalo7- CAAX; supplement 1A: GH146- FLP/UAS- FRT10- stop- FRT10- 3xHalo7- CAAX; Figure 1D, Figure 1—figure supplement 1, Figure 1—figure supplement 2: run- T2A- FLP/+; UAS- mCD8- FRT- GFP- FRT- RFP/+; VT033006- GAL4/+ Figure  3A: acj6- T2A- QF2/+; GH146- FLP, QUAS- FRT- stop- FRT- mCD8- GFP/UAS- FRT10- stop- FRT10- 3xHalo7- CAAX; 71B05- GAL4/+ Figure  3B, Figure  3—figure supplement 1C: GH146- FLP/UAS- FRT10- stop- FRT10- 3xHalo7- CAAX; 91G04- GAL4/CG14322- T2A- QF2, QUAS- FRT- stop- FRT- myr- 4xSNAPf Figure 3C: acj6- T2A- FLP/+; Mz19- GAL4; UAS- FRT- myr- 4xSNAPf- FRT- 3xHalo7- CAAX/+ Figure 3D, Figure 3—figure supplement 2, Figure 3—figure supplement 3: UAS- mCD8- GFP, hs- FLP/+; FRTG13, tub- GAL80/FRTG13, UAS- mCD8- GFP;; GH146- GAL4 (IV)/+ Figure  3—figure 71B05- GAL4/+ Figure  3—figure 91G04- GAL4/+ Figure 3—video 1: Please refer to Figure 3 for genotypes. Figure  4A, Figure  4—figure supplement 1: GH146- FLP, UAS- FRT10- stop- FRT10- 3xHalo7- CAAX/tsh- GAL4; CG14322- T2A- QF2, QUAS- FRT- stop- FRT- myr- 4xSNAPf/+ Figure  4B: UAS- mCD8- GFP, hs- FLP/+; FRTG13, GH146- GAL4 (IV)/+ Figure 8—figure supplement 2: acj6- T2A- FLP/+; tsh- GAL4, UAS- mCD8- FRT- GFP- FRT- RFP Figure 4—video 1: Please refer to Figure 4 for genotypes. Figure  5E, Figure  5—video 1: run- T2A- FLP/+; UAS- FRT- myr- mGreenLantern- FRT- 3xHalo7- CAAX/+; VT033006- GAL4/+ Figure 5F: UAS- mCD8- GFP/+; VT033006- GAL4/+ Figure 5G1: GH146- FLP/UAS- FRT10- stop- FRT10- 3xHalo7- CAAX; 71B05- GAL4/+ Figure 5G2: GH146- FLP/tsh- GAL4; UAS- FRT10- stop- FRT10- 3xHalo7- CAAX/+ Figure  5—figure supplement 1, Figure  5—video 2: acj6- T2A- FLP/+; Mz19- GAL4/ UAS- FRT- myr- mGreenLantern- FRT- 3xHalo7- CAAX Figure  6A, Figure  6—figure supplement 1, Figure  6—video 1: UAS- mCD8- GFP, hs- FLP/+; FRTG13, tub- GAL80/FRTG13, UAS- mCD8- GFP;; GH146- GAL4 (IV)/+ Figure  7A–C, Figure  7—figure supplement 1, Figure  7—videos 1–3: acj6- T2A- QF2/+; QUAS- FRT- stop- FRT- mCD8- GFP/UAS- FRT10- stop- FRT10- 3xHalo7- CAAX; GH146- FLP, 71B05- GAL4/+ Figure 8A, Figure 8—figure supplement 1: GH146- FLP, QUAS- FRT- stop- FRT- mCD8- GFP/lov- T2A- QF2; UAS- FRT10- stop- FRT10- 3xHalo7- CAAX/91G04- GAL4 8C: Figure QUAS- FRT- stop- FRT- myr- 4xSNAPf Figure  8D, Figure  8—figure supplement 3, Figure  8—video 2: UAS- mCD8- GFP/+; Split7- GAL4 (i.e. FlyLight SS01867: 72C11- p65ADZp; VT033006- ZpGDBD)/+ Figure  8—figure supplement 2: GH146- FLP, QUAS- FRT- stop- FRT- mCD8- GFP/lov- T2A- QF2, Mz612- GAL4; UAS- FRT10- stop- FRT10- 3xHalo7- CAAX/+ Figure  8—figure UAS- mCD8- FRT- GFP- FRT- RFP Figure  8—figure supplement 4B: lov- T2A- QF2, QUAS- FLP/UAS- EcR- DN; VT033006- GAL4/ UAS- mCD8- FRT- GFP- FRT- RFP Figure 8—video 1: Please refer to Figure 8 for genotypes. lov- T2A- QF2, QUAS- FLP/+; VT033006- GAL4/ UAS- FRT10- stop- FRT10- 3xHalo7- CAAX/ GH146- FLP/lov- GAL4; supplement 4A: Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 22 of 33 Developmental Biology | Neuroscience Research article MARCM clonal analyses MARCM clonal analyses have been previously described (Lee and Luo, 1999). Larvae of the genotype UAS- mCD8- GFP, hs- FLP/+; FRTG13, tub- GAL80/FRTG13, UAS- mCD8- GFP;; GH146- GAL4/+ were  heat shocked at 37 °C for 1 hr. To label the first- born DL1 PNs, heat shock was applied at 0–24 hr after larval hatching (ALH). MARCM clones of early, middle (mid- late for adPNs), and late larval- born PNs were generated by applying heat shocks at 42–48 hr, 66–72 hr, and 96–100 hr ALH, respectively. As larvae developed at different rates (Tennessen and Thummel, 2011), we reasoned that even if we could collect 0 hr–2 hr ALH larvae, their development might have varied by the time of heat shock. To minimize the effects of unsynchronized development, we selected those heat- shocked larvae that were among the first to form puparia and collected these white pupae in a ~3 hr window for the clonal analyses. Transcriptomic analyses Transcriptomic analyses have been described previously (Xie et al., 2021). tSNE plots and dot plots were generated in Python using PN single- cell RNA sequencing data and code available at https:// github.com/Qijing-Xie/FlyPN_development (Xie, 2021). Generation of T2A-QF2/FLP lines To generate a T2A- QF2/FLP donor vector for acj6 (we used the same strategy for run, CG14322 and lov), a ~2000  bp genomic sequence flanking the stop codon of acj6 was PCR amplified and introduced into pCR- Blunt II- TOPO (ThermoFisher Scientific #450245), forming pTOPO- acj6. To build pTopo- acj6- T2A- QF2, T2A- QF2 including loxP- flanked 3xP3- RFP was PCR amplified from pBPGUw- HACK- QF2 (Addgene #80276), followed by insertion into pTOPO- acj6 right before the stop codon of acj6 by DNA assembly (New England BioLabs #E2621S). To generate T2A- FLP, we PCR- amplified FLP from the genomic DNA of GH146- FLP strain. QF2 in pTopo- acj6- T2A- QF2 was then replaced by FLP through DNA assembly. Using CRISPR Optimal Target Finder (Gratz et al., 2014), we selected a 20 bp gRNA target sequence that flanked the stop codon and cloned it into pU6- BbsI- chiRNA (Addgene #45946). If the gRNA sequence did not flank the stop codon, silent mutations were introduced at the PAM site of the donor vector by site- directed mutagenesis. Donor and gRNA vectors were co- injected into Cas9 embryos in- house or through BestGene. Generation of FLP-out reporters To generate pUAS- FRT10- stop- FRT10- 3xHalo7- CAAX, FRT10- stop- FRT10 was PCR amplified from pUAS- FRT10- stop- FRT10- mCD8- GFP (Li et  al., 2021) and inserted into pUAS- 3xHalo7- CAAX (Addgene #87646) through NotI and DNA assembly. To generate pUAS- FRT- myr- 4xSNAPf- FRT- 3xHalo7- CAAX, we first PCR amplified myr- 4xSNAPf from pUAS- myr- 4xSNAPf (Addgene #87637) using FRT- containing primers. FRT- myr- 4xSNAPf- FRT was then introduced into pCR- Blunt II- TOPO, forming pTOPO- FRT- myr- 4xSNAPf- FRT. Using NotI- containing primers, FRT- myr- 4xSNAPf- FRT was PCR amplified and subcloned into pUAS- 3xHalo7- CAAX through NotI. To generate pUAS- FRT- myr- mGreenLantern- FRT- 3xHalo7- CAAX, we first PCR amplified mGreen- Lantern from pcDNA3.1- mGreenLantern (Addgene #161912). Using MluI and XbaI, we replaced 4xSNAPf in pUAS- myr- 4xSNAPf with mGreenLantern to build pUAS- myr- mGreenLantern. myr- mGreenLantern was PCR amplified with the introduction of FRT sequence, followed by insertion into pCR- Blunt II- TOPO. Using the NotI- containing primers, FRT- myr- mGreenLantern- FRT was PCR ampli- fied and subcloned into pUAS- 3xHalo7- CAAX through NotI. To generate pQUAS- FRT- stop- FRT- myr- 4xSNAPf, we first PCR amplified FRT- stop from pJFRC7- 20XUAS- FRT- stop- FRT- mCD8- GFP (Li et al., 2021) and inserted it into pTOPO- FRT- myr- 4xSNAPf- FRT through DNA assembly to form pTOPO- FRT- stop- FRT- myr- 4xSNAPf- FRT. Using NotI- containing forward and KpnI- containing reverse primers, FRT- stop- FRT- myr- 4xSNAPf was PCR amplified and subcloned into p10XQUAST. p10XQUAST was generated using p5XQUAS (Addgene #24349) and p10xQUAS- CsChrimson (Addgene #163629). attP24 and 86Fb landing sites were used for site- directed integration. Immunofluorescence staining and confocal imaging Fly brain dissection for immunostaining and live imaging has been described (Wu and Luo, 2006). Briefly, brains were dissected in phosphate- buffered saline (PBS) and fixed with 4% Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 23 of 33 Developmental Biology | Neuroscience Research article paraformaldehyde in PBS for 20 min on a nutator at room temperature. Fixed brains were washed with 0.1% Triton X- 100 in PBS (PBST) for 10  min twice. After blocking with 5% normal donkey serum in PBST for 1 hr at room temperature, the brains were incubated with primary antibodies overnight at 4  °C. After PBST wash, brains were incubated with secondary antibodies (1:1000; Jackson ImmunoResearch) in dark for 2  hr at room temperature. Washed and mounted brains were imaged with confocal laser scanning microscopy (ZEISS LSM 780; LSM 900 with Airyscan 2). Images were processed with ImageJ. Neurite tracing images were generated using Simple Neurite Tracer (SNT) (Arshadi et al., 2021). Primary antibodies used included chicken anti- GFP (1:1000; Aves Lab #GFP- 1020), rabbit anti- DsRed (1:500; TaKaRa #632496), rat anti- Cadherin DN (1:30; Developmental Studies Hybridoma Bank DSHB DN- Ex#8 supernatant), and mouse anti- Bruchpilot (1:30; DSHB nc82 supernatant). Chemical labeling Chemical labeling of Drosophila brains has been described (Kohl et al., 2014). Janelia Fluor (JF) Halo and SNAP ligands (stocks at 1 mM) were gifts from Dr. Luke Lavis (Grimm et al., 2017; Grimm et al., 2021). Fixed brains were washed with PBST for 5  min, followed by incubation with Halo and/or SNAP ligands (diluted in PBS) for 45  min at room temperature. Brains were then washed with PBST for 5  min, followed by blocking and immunostaining if necessary. For the co- incubation of Halo and SNAP ligands, JF503- cpSNAP (1:1000) and JF646- Halo (1:1000) were used. Alternatively, JFX650- SNAP (1:1000) and JFX554- Halo (1:10,000) were used. When only Halo ligands were needed, either JF646- Halo or JF635- Halo (1:1000) was used. For live brain imaging, dissected brains were incubated with Halo ligands diluted in culture media (described below) for 30 min at room temperature. For two- photon imaging, JF570- Halo was used at 1:5000. For AO- LLSM imaging, following JF646- Halo incubation at 1:1000, the brains were incubated with 1 µM Sulforhodamine 101 (Sigma) for 5 min at room temperature. The brains were then briefly washed with culture media before imaging. Brain explant culture setup and medium preparation Brain explant culture setup was modified based on Li et al., 2021; Li and Luo, 2021. A Sylgard plate with a thickness of ~2 millimeters was prepared by mixing base and curing agent at 10:1 ratio (DOW SYLGARD 184 Silicone Elastomer Kit). The mixture was poured into a 60 mm × 15 mm dish in which it was cured for two days at room temperature. Once cured, the plate was cut into small squares (~15 mm × ~15 mm). Indentations were created based on the size of an early pupal brain using a No.11 scalpel. Additional slits were made around the indentations for attaching imaginal discs which served as anchors to hold the brain position. A square Sylgard piece was then placed in a 60 mm × 15 mm dish or on a 25 mm round coverslip in preparation for two- photon/AO- LLSM imaging. Culture medium was prepared based on published methods (Rabinovich et al., 2015; Li and Luo, 2021; Li et al., 2021). The medium contained Schneider’s Drosophila Medium (ThermoFisher Scientific #21720001), 10% heat- inactivated Fetal Bovine Serum (ThermoFisher Scientific #16140071), 10 µg/mL human recombinant insulin (ThermoFisher Scientific #12585014; stock = 4 mg/mL), 1:100 Penicillin- Streptomycin (ThermoFisher Scientific #15140122). For 0 hr–6 hr APF brain culture, 0.5 mM ascorbic acid (Sigma #A4544; stock concentration = 50 mg/mL in water) was included. 20- hydroxyecdysone (Sigma #H5142; stock concentration = 1 mg/mL in ethanol) was used for 0 hr–6 hr and 12 hr brain explants at 20 µM and 2 µM, respectively. Culture medium was oxygenated for 20 min before use. Single- and dual-color imaging with two-photon microscopy Single- and dual- color imaging of PNs were performed at room temperature using a custom- built two- photon microscope (Prairie Technologies) with a Chameleon Ti:Sapphire laser (Coherent) and a 16 X water- immersion objective (0.8 NA; Nikon). Excitation wavelength was set at 920 nm for GFP imaging, and at 935 nm for co- imaging of mGreenLantern and JF570- Halo. z- stacks were obtained at 4 µm increments (10 µm increments for Figure 5—video 1). Images were acquired at a resolution of 1024 × 1024 pixel2 (512 × 512 for Figure 5—video 1), with a pixel dwell time of 6.8 µs and an optical zoom of 2.1, and at a frequency every 20 min for 8–23 hr. Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 24 of 33 Developmental Biology | Neuroscience Research article Dual-color imaging with AO-LLSM For AO- LLSM- based imaging, the excitation and detection objectives along with the 25 mm coverslip were immersed in ~40 mL of culture medium at room temperature. Explant brains held on Sylgard plate were excited simultaneously using 488 nm (for GFP) and 642 nm (for JF- 646) lasers operating with ~2–10 mW input power to the microscope (corresponding to ~10–50 µW at the back aperture of the excitation objective). An exposure time of 20–50 msec was used to balance imaging speed and signal- to- noise ratio (SNR). Dithered lattice light- sheet patterns with an inner/outer numerical aperture of 0.35/0.4 or 0.38/0.4 were used. The optical sections were collected by an axial step size of 250 nm in the detection objective coordinate, with a total of 81–201 steps (corresponding to a total axial scan range of 20–50 µm). Emission light from GFP and JF- 646 was separated by a dichromatic mirror (Di03- R561, Semrock, IDEX Health & Science, LLC, Rochester, NY) and captured by two Hamamatsu ORCA- Fusion sCMOS cameras simultaneously (Hamamatsu Photonics, Hamamatsu City, Japan). Prior to the acquisition of the time series data, the imaged volume was corrected for optical aberrations using a two- photon guide star- based adaptive optics method (Chen et al., 2014; Wang et al., 2014; Liu et al., 2018). Each imaged volume was deconvolved using Richardson- Lucy algorithm on HHMI Janelia Research Campus’ or Advanced Bioimaging Center’s computing cluster (https://github.com/ scopetools/cudadecon, Lambert et al., 2023; https://github.com/abcucberkeley/LLSM3DTools, Ruan and Upadhyayula, 2020) with experimentally measured point spread functions obtained from 100 or 200 nm fluorescent beads (Invitrogen FluoSpheres Carboxylate- Modified Microspheres, 505/515 nm, F8803, FF8811). The AO- LLSM was operated using a custom LabVIEW software (National Instruments, Woburn, MA). Statistics For data analyses, t- test and one- way ANOVA were used to determine p values as indicated in the figure legend for each graph, and graphs were generated using Excel. Exact p values were provided in source data files. Material and data availability All reagents generated in this study are available from the lead corresponding author upon request. Figure 3—figure supplement 3—source data 1, Figure 6—source data 1, and Figure 7—source data 1 contain the numerical and statistical data used to generate the figures. The confocal imaging dataset is available at Brain Image Library under DOI https://doi.org/10.35077/g.933. Acknowledgements We thank the Luo lab members for constructive feedback on the manuscript; Tzumin Lee for sharing equipment at Janelia Research Campus; Luke Lavis for sharing JF dyes. This work was supported by a grant from NIH (R01 DC005982 to LL). TL was supported by NIH 1K99DC01883001. GL and SU are funded by Philomathia Foundation. SU is funded by the Chan Zuckerberg Initiative Imaging Scientist program. SU is a Chan Zuckerberg Biohub Investigator. EB and LL are HHMI investigators. Additional information Funding Funder National Institutes of Health Philomathia Foundation Chan Zuckerberg Initiative National Institutes of Health Grant reference number Author R01 DC005982 Liqun Luo Gaoxiang Liu Srigokul Upadhyayula Srigokul Upadhyayula 1K99DC01883001 Tongchao Li Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 25 of 33 Developmental Biology | Neuroscience Research article Funder Grant reference number Author Howard Hughes Medical Institute Eric Betzig Liqun Luo The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication. Author contributions Kenneth Kin Lam Wong, Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing - original draft; Tongchao Li, Resources, Investigation, Methodology, Writing – review and editing; Tian- Ming Fu, Gaoxiang Liu, Resources, Data curation, Investigation, Meth- odology, Writing – review and editing; Cheng Lyu, Resources, Methodology, Writing – review and editing; Sayeh Kohani, Data curation, Investigation; Qijing Xie, Data curation, Investigation, Writing – review and editing; David J Luginbuhl, Resources, Data curation, Writing – review and editing; Srigokul Upadhyayula, Eric Betzig, Resources, Supervision, Writing – review and editing; Liqun Luo, Conceptualization, Supervision, Funding acquisition, Investigation, Methodology, Project administra- tion, Writing – review and editing Author ORCIDs Kenneth Kin Lam Wong Tian- Ming Fu Liqun Luo http://orcid.org/0000-0001-6265-0859 http://orcid.org/0000-0001-5467-9264 http://orcid.org/0000-0002-5597-4051 Decision letter and Author response Decision letter https://doi.org/10.7554/eLife.85521.sa1 Author response https://doi.org/10.7554/eLife.85521.sa2 Additional files Supplementary files • Supplementary file 1. Sample variability among individual brains. A supplemental table describing the biological and technical variations we observed among individual brain samples, and measures we took to minimize them, if possible. • MDAR checklist Data availability Figure 3—source data 1, Figure 5—source data 1, Figure 6—source data 1, and Figure 7—source data 1 contain the numerical and statistical data used to generate the figures. The confocal imaging dataset is available at Brain Image Library under DOI https://doi.org/10.35077/g.933. The following dataset was generated: Author(s) Wong KLK Year 2023 Dataset title Dataset URL Database and Identifier https:// doi. org/ 10. 35077/ g. 933 Brain Image Library, 10.35077/g.933 Origin of wiring specificity in an olfactory map revealed by neuron type- specific, time- lapse imaging of dendrite targeting: Confocal imaging of developing fly brain Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 26 of 33 Developmental Biology | Neuroscience Research article The following previously published dataset was used: Dataset title Dataset URL Database and Identifier Temporal evolution of single- cell transcriptomes of Drosophila olfactory projection neurons https://www. ncbi. nlm. nih. gov/ geo/ query/ acc. cgi? acc= GSE161228 NCBI Gene Expression Omnibus, GSE161228 Year 2021 Author(s) Xie Q, Brbic M, Horns F, Kolluru SS, Jones RC, Li J, Reddy AR, Xie A, Kohani S, Li Z, McLaughlin CN, Li T, Xu C, Vacek D, Luginbuhl DJ, Leskovec J, Quake SR, Luo L Li H References Arshadi C, Günther U, Eddison M, Harrington KIS, Ferreira TA. 2021. Snt: a unifying toolbox for quantification of neuronal anatomy. Nature Methods 18:374–377. DOI: https://doi.org/10.1038/s41592-021-01105-7, PMID: 33795878 Berdnik D, Chihara T, Couto A, Luo L. 2006. Wiring stability of the adult Drosophila olfactory circuit after lesion. The Journal of Neuroscience 26:3367–3376. DOI: https://doi.org/10.1523/JNEUROSCI.4941-05.2006, PMID: 16571743 Berns DS, DeNardo LA, Pederick DT, Luo L. 2018. Teneurin- 3 controls topographic circuit assembly in the hippocampus. Nature 554:328–333. DOI: https://doi.org/10.1038/nature25463, PMID: 29414938 Campbell BC, Nabel EM, Murdock MH, Lao- Peregrin C, Tsoulfas P, Blackmore MG, Lee FS, Liston C, Morishita H, Petsko GA. 2020. MGreenLantern: a bright monomeric fluorescent protein with rapid expression and cell filling properties for neuronal imaging. PNAS 117:30710–30721. DOI: https://doi.org/10.1073/pnas.2000942117, PMID: 33208539 Cang J, Feldheim DA. 2013. Developmental mechanisms of topographic MAP formation and alignment. Annual Review of Neuroscience 36:51–77. DOI: https://doi.org/10.1146/annurev-neuro-062012-170341, PMID: 23642132 Chen BC, Legant WR, Wang K, Shao L, Milkie DE, Davidson MW, Janetopoulos C, Wu XS, Hammer JA, Liu Z, English BP, Mimori- Kiyosue Y, Romero DP, Ritter AT, Lippincott- Schwartz J, Fritz- Laylin L, Mullins RD, Mitchell DM, Bembenek JN, Reymann AC, et al. 2014. Lattice light- sheet microscopy: imaging molecules to embryos at high spatiotemporal resolution. Science 346:1257998. DOI: https://doi.org/10.1126/science. 1257998, PMID: 25342811 Cherbas L, Hu X, Zhimulev I, Belyaeva E, Cherbas P. 2003. EcR isoforms in Drosophila: testing tissue- specific requirements by targeted blockade and rescue. Development 130:271–284. DOI: https://doi.org/10.1242/dev. 00205, PMID: 12466195 Diao F, White BH. 2012. A novel approach for directing transgene expression in Drosophila: T2A- gal4 in- frame fusion. Genetics 190:1139–1144. DOI: https://doi.org/10.1534/genetics.111.136291, PMID: 22209908 Doe CQ. 2017. Temporal patterning in the Drosophila CNS. Annual Review of Cell and Developmental Biology 33:219–240. DOI: https://doi.org/10.1146/annurev-cellbio-111315-125210, PMID: 28992439 Erclik T, Li X, Courgeon M, Bertet C, Chen Z, Baumert R, Ng J, Koo C, Arain U, Behnia R, del Valle Rodriguez A, Senderowicz L, Negre N, White KP, Desplan C. 2017. Integration of temporal and spatial patterning generates neural diversity. Nature 541:365–370. DOI: https://doi.org/10.1038/nature20794, PMID: 28077877 Gao Q, Yuan B, Chess A. 2000. Convergent projections of Drosophila olfactory neurons to specific glomeruli in the antennal lobe. Nature Neuroscience 3:780–785. DOI: https://doi.org/10.1038/77680, PMID: 10903570 Gratz SJ, Ukken FP, Rubinstein CD, Thiede G, Donohue LK, Cummings AM, O’Connor- Giles KM. 2014. Highly specific and efficient CRISPR/Cas9- catalyzed homology- directed repair in Drosophila. Genetics 196:961–971. DOI: https://doi.org/10.1534/genetics.113.160713, PMID: 24478335 Grimm JB, Muthusamy AK, Liang Y, Brown TA, Lemon WC, Patel R, Lu R, Macklin JJ, Keller PJ, Ji N, Lavis LD. 2017. A general method to fine- tune fluorophores for live- cell and in vivo imaging. Nature Methods 14:987– 994. DOI: https://doi.org/10.1038/nmeth.4403, PMID: 28869757 Grimm JB, Xie L, Casler JC, Patel R, Tkachuk AN, Falco N, Choi H, Lippincott- Schwartz J, Brown TA, Glick BS, Liu Z, Lavis LD. 2021. A general method to improve fluorophores using deuterated auxochromes. JACS Au 1:690–696. DOI: https://doi.org/10.1021/jacsau.1c00006, PMID: 34056637 Holguera I, Desplan C. 2018. Neuronal specification in space and time. Science 362:176–180. DOI: https://doi. org/10.1126/science.aas9435, PMID: 30309944 Hong W, Zhu H, Potter CJ, Barsh G, Kurusu M, Zinn K, Luo L. 2009. Leucine- Rich repeat transmembrane proteins instruct discrete dendrite targeting in an olfactory MAP. Nature Neuroscience 12:1542–1550. DOI: https://doi. org/10.1038/nn.2442, PMID: 19915565 Hong W, Mosca TJ, Luo L. 2012. Teneurins instruct synaptic partner matching in an olfactory MAP. Nature 484:201–207. DOI: https://doi.org/10.1038/nature10926, PMID: 22425994 Hong W, Luo L. 2014. Genetic control of wiring specificity in the fly olfactory system. Genetics 196:17–29. DOI: https://doi.org/10.1534/genetics.113.154336, PMID: 24395823 Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 27 of 33 Developmental Biology | Neuroscience Research article Isshiki T, Pearson B, Holbrook S, Doe CQ. 2001. Drosophila neuroblasts sequentially express transcription factors which specify the temporal identity of their neuronal progeny. Cell 106:511–521. DOI: https://doi.org/10.1016/ s0092-8674(01)00465-2, PMID: 11525736 Jefferis GSXE, Marin EC, Stocker RF, Luo L. 2001. Target neuron prespecification in the olfactory map of Drosophila. Nature 414:204–208. DOI: https://doi.org/10.1038/35102574, PMID: 11719930 Jefferis G, Vyas RM, Berdnik D, Ramaekers A, Stocker RF, Tanaka NK, Ito K, Luo L. 2004. Developmental origin of wiring specificity in the olfactory system of Drosophila. Development 131:117–130. DOI: https://doi.org/10. 1242/dev.00896, PMID: 14645123 Jenett A, Rubin GM, Ngo T- TB, Shepherd D, Murphy C, Dionne H, Pfeiffer BD, Cavallaro A, Hall D, Jeter J, Iyer N, Fetter D, Hausenfluck JH, Peng H, Trautman ET, Svirskas RR, Myers EW, Iwinski ZR, Aso Y, DePasquale GM, et al. 2012. A GAL4- driver line resource for Drosophila neurobiology. Cell Reports 2:991– 1001. DOI: https://doi.org/10.1016/j.celrep.2012.09.011, PMID: 23063364 Kanamori T, Kanai MI, Dairyo Y, Yasunaga K, Morikawa RK, Emoto K. 2013. Compartmentalized calcium transients trigger dendrite pruning in Drosophila sensory neurons. Science 340:1475–1478. DOI: https://doi. org/10.1126/science.1234879, PMID: 23722427 Kohl J, Ng J, Cachero S, Ciabatti E, Dolan MJ, Sutcliffe B, Tozer A, Ruehle S, Krueger D, Frechter S, Branco T, Tripodi M, Jefferis G. 2014. Ultrafast tissue staining with chemical tags. PNAS 111:E3805–E3814. DOI: https:// doi.org/10.1073/pnas.1411087111, PMID: 25157152 Komiyama T, Johnson WA, Luo L, Jefferis GSXE. 2003. From lineage to wiring specificity: POU domain transcription factors control precise connections of Drosophila olfactory projection neurons. Cell 112:157–167. DOI: https://doi.org/10.1016/s0092-8674(03)00030-8, PMID: 12553905 Komiyama T, Luo L. 2007. Intrinsic control of precise dendritic targeting by an ensemble of transcription factors. Current Biology 17:278–285. DOI: https://doi.org/10.1016/j.cub.2006.11.067, PMID: 17276922 Komiyama T, Sweeney LB, Schuldiner O, Garcia KC, Luo L. 2007. Graded expression of semaphorin- 1a cell- autonomously directs dendritic targeting of olfactory projection neurons. Cell 128:399–410. DOI: https://doi. org/10.1016/j.cell.2006.12.028, PMID: 17254975 Lambert L, Shao L, dmilkie. 2023. cudaDecon. GitHub. https://github.com/scopetools/cudadecon Lee T, Luo L. 1999. Mosaic analysis with a repressible cell marker for studies of gene function in neuronal morphogenesis. Neuron 22:451–461. DOI: https://doi.org/10.1016/s0896-6273(00)80701-1, PMID: 10197526 Lee T, Marticke S, Sung C, Robinow S, Luo L. 2000. Cell- Autonomous requirement of the USP/ecr- B ecdysone receptor for mushroom body neuronal remodeling in Drosophila. Neuron 28:807–818. DOI: https://doi.org/10. 1016/s0896-6273(00)00155-0, PMID: 11163268 Lee HH, Jan LY, Jan YN. 2009. Drosophila IKK- related kinase ik2 and katanin p60- like 1 regulate dendrite pruning of sensory neuron during metamorphosis. PNAS 106:6363–6368. DOI: https://doi.org/10.1073/pnas. 0902051106, PMID: 19329489 Li X, Erclik T, Bertet C, Chen Z, Voutev R, Venkatesh S, Morante J, Celik A, Desplan C. 2013. Temporal patterning of Drosophila medulla neuroblasts controls neural fates. Nature 498:456–462. DOI: https://doi.org/10.1038/ nature12319, PMID: 23783517 Li H, Horns F, Wu B, Xie Q, Li J, Li T, Luginbuhl DJ, Quake SR, Luo L. 2017. Classifying Drosophila olfactory projection neuron subtypes by single- cell RNA sequencing. Cell 171:1206–1220. DOI: https://doi.org/10.1016/ j.cell.2017.10.019, PMID: 29149607 Li T, Fu TM, Wong KKL, Li H, Xie Q, Luginbuhl DJ, Wagner MJ, Betzig E, Luo L. 2021. Cellular bases of olfactory circuit assembly revealed by systematic time- lapse imaging. Cell 184:5107–5121. DOI: https://doi.org/10.1016/ j.cell.2021.08.030, PMID: 34551316 Li T, Luo L. 2021. An explant system for time- lapse imaging studies of olfactory circuit assembly in Drosophila. Journal of Visualized Experiments 2021:176. DOI: https://doi.org/10.3791/62983, PMID: 34723938 Lin S, Kao CF, Yu HH, Huang Y, Lee T. 2012. Lineage analysis of Drosophila lateral antennal lobe neurons reveals notch- dependent binary temporal fate decisions. PLOS Biology 10:e1001425. DOI: https://doi.org/10.1371/ journal.pbio.1001425, PMID: 23185131 Liu Z, Yang CP, Sugino K, Fu CC, Liu LY, Yao X, Lee LP, Lee T. 2015. Opposing intrinsic temporal gradients guide neural stem cell production of varied neuronal fates. Science 350:317–320. DOI: https://doi.org/10.1126/ science.aad1886, PMID: 26472907 Liu TL, Upadhyayula S, Milkie DE, Singh V, Wang K, Swinburne IA, Mosaliganti KR, Collins ZM, Hiscock TW, Shea J, Kohrman AQ, Medwig TN, Dambournet D, Forster R, Cunniff B, Ruan Y, Yashiro H, Scholpp S, Meyerowitz EM, Hockemeyer D, et al. 2018. Observing the cell in its native state: imaging subcellular dynamics in multicellular organisms. Science 360:eaaq1392. DOI: https://doi.org/10.1126/science.aaq1392, PMID: 29674564 Luo L, Flanagan JG. 2007. Development of continuous and discrete neural maps. Neuron 56:284–300. DOI: https://doi.org/10.1016/j.neuron.2007.10.014, PMID: 17964246 Luo L. 2021. Architectures of neuronal circuits. Science 373:eabg7285. DOI: https://doi.org/10.1126/science. abg7285 Marin EC, Watts RJ, Tanaka NK, Ito K, Luo L. 2005. Developmentally programmed remodeling of the Drosophila olfactory circuit. Development 132:725–737. DOI: https://doi.org/10.1242/dev.01614, PMID: 15659487 Mark B, Lai SL, Zarin AA, Manning L, Pollington HQ, Litwin- Kumar A, Cardona A, Truman JW, Doe CQ. 2021. A developmental framework linking neurogenesis and circuit formation in the Drosophila CNS. eLife 10:e67510. DOI: https://doi.org/10.7554/eLife.67510, PMID: 33973523 Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 28 of 33 Developmental Biology | Neuroscience Research article Mayseless O, Berns DS, Yu XM, Riemensperger T, Fiala A, Schuldiner O. 2018. Developmental coordination during olfactory circuit remodeling in Drosophila. Neuron 99:1204–1215.. DOI: https://doi.org/10.1016/j. neuron.2018.07.050, PMID: 30146303 Miyares RL, Lee T. 2019. Temporal control of Drosophila central nervous system development. Current Opinion in Neurobiology 56:24–32. DOI: https://doi.org/10.1016/j.conb.2018.10.016, PMID: 30500514 Mombaerts P, Wang F, Dulac C, Chao SK, Nemes A, Mendelsohn M, Edmondson J, Axel R. 1996. Visualizing an olfactory sensory map. Cell 87:675–686. DOI: https://doi.org/10.1016/s0092-8674(00)81387-2, PMID: 8929536 Murthy VN. 2011. Olfactory maps in the brain. Annual Review of Neuroscience 34:233–258. DOI: https://doi. org/10.1146/annurev-neuro-061010-113738, PMID: 21692659 O’Hare JK, Gonzalez KC, Herrlinger SA, Hirabayashi Y, Hewitt VL, Blockus H, Szoboszlay M, Rolotti SV, Geiller TC, Negrean A, Chelur V, Polleux F, Losonczy A. 2022. Compartment- Specific tuning of dendritic feature selectivity by intracellular Ca2+ release. Science 375:eabm1670. DOI: https://doi.org/10.1126/science. abm1670, PMID: 35298275 Özel MN, Langen M, Hassan BA, Hiesinger PR. 2015. Filopodial dynamics and growth cone stabilization in Drosophila visual circuit development. eLife 4:e10721. DOI: https://doi.org/10.7554/eLife.10721, PMID: 26512889 Pederick DT, Lui JH, Gingrich EC, Xu C, Wagner MJ, Liu Y, He Z, Quake SR, Luo L. 2021. Reciprocal repulsions instruct the precise assembly of parallel hippocampal networks. Science 372:1068–1073. DOI: https://doi.org/ 10.1126/science.abg1774, PMID: 34083484 Potter CJ, Tasic B, Russler EV, Liang L, Luo L. 2010. The Q system: a repressible binary system for transgene expression, lineage tracing, and mosaic analysis. Cell 141:536–548. DOI: https://doi.org/10.1016/j.cell.2010.02. 025, PMID: 20434990 Rabinovich D, Mayseless O, Schuldiner O. 2015. Long term ex vivo culturing of Drosophila brain as a method to live image pupal brains: insights into the cellular mechanisms of neuronal remodeling. Frontiers in Cellular Neuroscience 9:327. DOI: https://doi.org/10.3389/fncel.2015.00327, PMID: 26379498 Riabinina O, Luginbuhl D, Marr E, Liu S, Wu MN, Luo L, Potter CJ. 2015. Improved and expanded Q- system reagents for genetic manipulations. Nature Methods 12:219–222, . DOI: https://doi.org/10.1038/nmeth.3250, PMID: 25581800 Rolls MM, Satoh D, Clyne PJ, Henner AL, Uemura T, Doe CQ. 2007. Polarity and intracellular compartmentalization of Drosophila neurons. Neural Development 2:7. DOI: https://doi.org/10.1186/1749- 8104-2-7, PMID: 17470283 Ruan X, Upadhyayula S. 2020. Llsm3Dtools: Tools for the analysis of 3D live images from lattice light- sheet microscopy (LLSM). d482a69. GitHub. https://github.com/abcucberkeley/LLSM3DTools Sen SQ. 2023. Generating neural diversity through spatial and temporal patterning. Seminars in Cell & Developmental Biology 142:54–66. DOI: https://doi.org/10.1016/j.semcdb.2022.06.002, PMID: 35738966 Stork T, Sheehan A, Tasdemir- Yilmaz OE, Freeman MR. 2014. Neuron- Glia interactions through the heartless FGF receptor signaling pathway mediate morphogenesis of Drosophila astrocytes. Neuron 83:388–403. DOI: https://doi.org/10.1016/j.neuron.2014.06.026, PMID: 25033182 Sutcliffe B, Ng J, Auer TO, Pasche M, Benton R, Jefferis GSXE, Cachero S. 2017. Second- Generation Drosophila chemical tags: sensitivity, versatility, and speed. Genetics 205:1399–1408. DOI: https://doi.org/10.1534/ genetics.116.199281, PMID: 28209589 Sweeney LB, Chou YH, Wu Z, Joo W, Komiyama T, Potter CJ, Kolodkin AL, Garcia KC, Luo L. 2011. Secreted semaphorins from degenerating larval Orn axons direct adult projection neuron dendrite targeting. Neuron 72:734–747. DOI: https://doi.org/10.1016/j.neuron.2011.09.026, PMID: 22153371 Tennessen JM, Thummel CS. 2011. Coordinating growth and maturation- insights from Drosophila. Current Biology 21:R750–R757. DOI: https://doi.org/10.1016/j.cub.2011.06.033, PMID: 21959165 Thummel CS. 2001. Molecular mechanisms of developmental timing in C. elegans and Drosophila. Developmental Cell 1:453–465. DOI: https://doi.org/10.1016/s1534-5807(01)00060-0, PMID: 11703937 Tirian L, Dickson BJ. 2017. The VT GAL4, LexA, and Split- GAL4 Driver Line Collections for Targeted Expression in the Drosophila Nervous System. bioRxiv. DOI: https://doi.org/10.1101/198648 Vosshall LB, Wong AM, Axel R. 2000. An olfactory sensory map in the fly brain. Cell 102:147–159. DOI: https:// doi.org/10.1016/s0092-8674(00)00021-0, PMID: 10943836 Wang K, Milkie DE, Saxena A, Engerer P, Misgeld T, Bronner ME, Mumm J, Betzig E. 2014. Rapid adaptive optical recovery of optimal resolution over large volumes. Nature Methods 11:625–628. DOI: https://doi.org/ 10.1038/nmeth.2925, PMID: 24727653 Watts RJ, Hoopfer ED, Luo L. 2003. Axon pruning during Drosophila metamorphosis: evidence for local degeneration and requirement of the ubiquitin- proteasome system. Neuron 38:871–885. DOI: https://doi.org/ 10.1016/s0896-6273(03)00295-2, PMID: 12818174 Williams DW, Truman JW. 2005. Cellular mechanisms of dendrite pruning in Drosophila: insights from in vivo time- lapse of remodeling dendritic arborizing sensory neurons. Development 132:3631–3642. DOI: https://doi. org/10.1242/dev.01928, PMID: 16033801 Wu JS, Luo L. 2006. A protocol for dissecting Drosophila melanogaster brains for live imaging or immunostaining. Nature Protocols 1:2110–2115. DOI: https://doi.org/10.1038/nprot.2006.336, PMID: 17487202 Xie Q, Wu B, Li J, Xu C, Li H, Luginbuhl DJ, Wang X, Ward A, Luo L. 2019. Transsynaptic fish- lips signaling prevents misconnections between nonsynaptic partner olfactory neurons. PNAS 116:16068–16073. DOI: https://doi.org/10.1073/pnas.1905832116, PMID: 31341080 Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 29 of 33 Developmental Biology | Neuroscience Research article Xie Q. 2021. Flypn_Development. bf76a37. GitHub. https://github.com/Qijing-Xie/FlyPN_development Xie Q, Brbic M, Horns F, Kolluru SS, Jones RC, Li J, Reddy AR, Xie A, Kohani S, Li Z, McLaughlin CN, Li T, Xu C, Vacek D, Luginbuhl DJ, Leskovec J, Quake SR, Luo L, Li H. 2021. Temporal evolution of single- cell transcriptomes of Drosophila olfactory projection neurons. eLife 10:e63450. DOI: https://doi.org/10.7554/ eLife.63450, PMID: 33427646 Xie Q, Li J, Li H, Udeshi ND, Svinkina T, Orlin D, Kohani S, Guajardo R, Mani DR, Xu C, Li T, Han S, Wei W, Shuster SA, Luginbuhl DJ, Quake SR, Murthy SE, Ting AY, Carr SA, Luo L. 2022. Transcription factor acj6 controls dendrite targeting via a combinatorial cell- surface code. Neuron 110:2299–2314.. DOI: https://doi. org/10.1016/j.neuron.2022.04.026, PMID: 35613619 Yaniv SP, Schuldiner O. 2016. A fly’s view of neuronal remodeling. Wiley Interdisciplinary Reviews. Developmental Biology 5:618–635. DOI: https://doi.org/10.1002/wdev.241, PMID: 27351747 Yates PA, Roskies AL, McLaughlin T, O’Leary DD. 2001. Topographic- specific axon branching controlled by ephrin- as is the critical event in retinotectal MAP development. The Journal of Neuroscience 21:8548–8563. DOI: https://doi.org/10.1523/JNEUROSCI.21-21-08548.2001, PMID: 11606643 Yu HH, Kao CF, He Y, Ding P, Kao JC, Lee T. 2010. A complete developmental sequence of A Drosophila neuronal lineage as revealed by twin- spot MARCM. PLOS Biology 8:e1000461. DOI: https://doi.org/10.1371/ journal.pbio.1000461, PMID: 20808769 Zhu S, Lin S, Kao CF, Awasaki T, Chiang AS, Lee T. 2006. Gradients of the Drosophila chinmo BTB- zinc finger protein govern neuronal temporal identity. Cell 127:409–422. DOI: https://doi.org/10.1016/j.cell.2006.08.045, PMID: 17055440 Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 30 of 33 Developmental Biology | Neuroscience Research article Appendix 1 Appendix 1—key resources table Reagent type (species) or resource Designation Source or reference Identifiers Additional information Genetic reagent (D. melanogaster) GH146- FLP DOI: 10.1038/nn.2442 Genetic reagent (D. melanogaster) QUAS- FRT- stop- FRT- mCD8- GFP DOI: 10.1016 /j. cell.2010.02.025 Genetic reagent (D. melanogaster) UAS- mCD8- GFP DOI: 10.1016 / s0896- 6273(00)80701–1 Genetic reagent (D. melanogaster) UAS- mCD8- FRT- GFP- FRT- RFP DOI: 10.1016 /j. neuron.2014.06.026 Genetic reagent (D. melanogaster) VT033006- GAL4 Genetic reagent (D. melanogaster) Mz19- GAL4 Genetic reagent (D. melanogaster) 91 G04- GAL4 Genetic reagent (D. melanogaster) Mz612- GAL4 Genetic reagent (D. melanogaster) 71B05- GAL4 Genetic reagent (D. melanogaster) Split7- GAL4 Genetic reagent (D. melanogaster) QUAS- FLP DOI: 10.1101/198648 DOI: 10.1242/dev.00896 DOI: 10.1016 /j. celrep.2012.09.011 DOI: 10.1242/dev.01614 DOI: 10.1016 /j. celrep.2012.09.011 DOI: 10.7554/eLife.63450 FlyLight:SS01867 DOI: 10.1016 /j. cell.2010.02.025 Genetic reagent (D. melanogaster) UAS- EcR.B1-ΔC655.F645A DOI: 10.1242/dev.00205 Genetic reagent (D. melanogaster) tsh- GAL4 Genetic reagent (D. melanogaster) lov- GAL4 Bloomington Drosophila Stock Center BDSC:3040 Bloomington Drosophila Stock Center BDSC:3737 Genetic reagent (D. melanogaster) UAS- mCD8- GFP, hs- FLP; FRTG13, tub- GAL80;; GH146- GAL4 DOI: 10.1016 / s0896- 6273(00)80701–1 Genetic reagent (D. melanogaster) FRTG13, UAS- mCD8- GFP DOI: 10.1016 / s0896- 6273(00)80701–1 Genetic reagent (D. melanogaster) UAS- FRT10- stop- FRT10- 3xHalo7- CAAX this paper Genetic reagent (D. melanogaster) UAS- FRT- myr- 4xSNAPf- FRT- 3xHalo7- CAAX this paper Genetic reagent (D. melanogaster) UAS- FRT- myr- mGreenLantern- FRT- 3xHalo7- CAAX this paper Genetic reagent (D. melanogaster) QUAS- FRT- stop- FRT- myr- 4xSNAPf this paper Genetic reagent (D. melanogaster) run- T2A- FLP Genetic reagent (D. melanogaster) acj6- T2A- FLP Genetic reagent (D. melanogaster) acj6- T2A- QF2 this paper this paper this paper Genetic reagent (D. melanogaster) CG14322- T2A- QF2 this paper Genetic reagent (D. melanogaster) lov- T2A- QF2 this paper Antibody chicken polyclonal anti- GFP Aves Lab Appendix 1 Continued on next page on either II or III chromosome; see Materials and methods on III chromosome; see Materials and methods on II chromosome; see Materials and methods on III chromosome; see Materials and methods on X chromosome; see Materials and methods on X chromosome; see Materials and methods on X chromosome; see Materials and methods on III chromosome; see Materials and methods on II chromosome; see Materials and methods RRID:AB_10000240; Aves Lab:GFP- 1020 (1:1000) Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 31 of 33 Developmental Biology | Neuroscience Research article Appendix 1 Continued Reagent type (species) or resource Designation Source or reference Identifiers Additional information Antibody rabbit polyclonal anti- DsRed TaKaRa RRID:AB_10013483; TaKaRa:632496 (1:500) Antibody rat monoclonal anti- Cadherin DN Developmental Studies Hybridoma Bank RRID:AB_528121; DSHB:DN- Ex#8 (1:30) Antibody mouse monoclonal anti- Bruchpilot Developmental Studies Hybridoma Bank RRID:AB_2314866; DSHB:nc82 supernatant (1:30) Recombinant DNA reagent Recombinant DNA reagent Recombinant DNA reagent Recombinant DNA reagent Recombinant DNA reagent pBPGUw- HACK- QF2 Addgene RRID:Addgene_80276 pU6- BbsI- chiRNA Addgene RRID:Addgene_45946 pUAS- 3xHalo7- CAAX Addgene RRID:Addgene_87646 pUAS- myr- 4xSNAPf Addgene RRID:Addgene_87637 pcDNA3.1- mGreenLantern Addgene RRID:Addgene_161912 Recombinant DNA reagent p5XQUAS Addgene RRID:Addgene_24349 Recombinant DNA reagent p10xQUAS- CsChrimson Addgene RRID:Addgene_163629 Recombinant DNA reagent pUAS- FRT10- stop- FRT10- 3xHalo7- CAAX this paper Recombinant DNA reagent pUAS- FRT- myr- 4xSNAPf- FRT- 3xHalo7- CAAX this paper Recombinant DNA reagent pUAS- FRT- myr- mGreenLantern- FRT- 3xHalo7- CAAX this paper Recombinant DNA reagent Recombinant DNA reagent pUAS- myr- mGreenLantern this paper pQUAS- FRT- stop- FRT- myr- 4xSNAPf this paper Chemical compound, drug SYLGARD 184 Silicone Elastomer Kit DOW Schneider’s Drosophila Medium ThermoFisher Scientific Fetal Bovine Serum ThermoFisher Scientific Human recombinant insulin ThermoFisher Scientific Penicillin- Streptomycin ThermoFisher Scientific backbone from pUAS- 3xHalo7- CAAX; see Materials and methods backbone from pUAS- 3xHalo7- CAAX; see Materials and methods backbone from pUAS- 3xHalo7- CAAX; see Materials and methods backbone from pUAS- myr- 4xSNAPf; see Materials and methods backbone from p5XQUAS; see Materials and methods DOW:2646340 ThermoFisher Scientific:21720001 ThermoFisher Scientific:16140071 ThermoFisher Scientific:12585014 ThermoFisher Scientific:15140122 used at 10% used at 10 µg/mL (1:100) Ascorbic acid Sigma Sigma:A4544 used at 50 mg/mL in water 20- hydroxyecdysone Sigma Sigma:H5142 used at 20 µM and 2 µM JF503- cpSNAP Chemical compound, drug JF646- Halo Chemical compound, drug Chemical compound, drug JFX650- SNAP JFX554- Halo Appendix 1 Continued on next page DOI: 10.1038/nmeth.4403; DOI: 10.1021/jacsau.1c00006 DOI: 10.1038/nmeth.4403; DOI: 10.1021/jacsau.1c00006 DOI: 10.1038/nmeth.4403; DOI: 10.1021/jacsau.1c00006 DOI: 10.1038/nmeth.4403; DOI: 10.1021/jacsau.1c00006 (1:1000); gift from Dr. Luke Lavis (1:1000); gift from Dr. Luke Lavis (1:1000); gift from Dr. Luke Lavis (1:10000); gift from Dr. Luke Lavis Chemical compound, drug Chemical compound, drug Chemical compound, drug Chemical compound, drug Chemical compound, drug Chemical compound, drug Chemical compound, drug Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 32 of 33 Developmental Biology | Neuroscience Research article Appendix 1 Continued Reagent type (species) or resource Designation Source or reference Identifiers Additional information Chemical compound, drug JF635- Halo Chemical compound, drug JF570- Halo DOI: 10.1038/nmeth.4403; DOI: 10.1021/jacsau.1c00006 DOI: 10.1038/nmeth.4403; DOI: 10.1021/jacsau.1c00006 (1:1000); gift from Dr. Luke Lavis (1:5000); gift from Dr. Luke Lavis Chemical compound, drug Sulforhodamine 101 Sigma Sigma:S7635 used at 1 µM Software, algorithm ZEN Carl Zeiss RRID:SCR_013672 Software, algorithm ImageJ National Institutes of Health RRID:SCR_003070 Software, algorithm Python Programming Language Python RRID:SCR_008394 http://www.python.org/ Wong et al. eLife 2023;12:e85521. DOI: https://doi.org/10.7554/eLife.85521 33 of 33 Developmental Biology | Neuroscience
null
10.1371_journal.pone.0258085.pdf
Data Availability Statement: Due to legal and participants confidentiality, data will only be available upon request. The data underlying the results presented in the study are available from Shenzhen Luohu Disease Prevention and Control Center via contacting Weihong Chen, director of Shenzhen Luohu Disease Prevention and Control Center, at 1433529760@qq.com.
Due to legal and participants confidentiality, data will only be available upon request. The data underlying the results presented in the study are available from Shenzhen Luohu
RESEARCH ARTICLE Level of engagement of recreational physical activity of urban villagers in Luohu, Shenzhen, China Lu ShiID 1*, Willie Leung2, Qingming Zheng3, Jie Wu3 1 Public Health, School of Social and Behavioral Health Science, College of Public Health and Human Sciences, Oregon State University, Corvallis, OR, United States of America, 2 Department of Health Sciences and Human Performance, College of Natural and Health Sciences, The University of Tampa, Tampa, FL, United States of America, 3 Shenzhen Luohu Disease Prevention and Control Center, Shenzhen, Guangdong, China * shil@oregonstate.edu Abstract Physical activity is important for health. However, there is a lack of literature related to the physical activity levels of adults living in urban villagers, which is a vulnerable population in China. The aim of this study is to compare the physical activity and sedentary behavior engagements between urban villagers and non-urban villagers using the 2019 Luohu Shen- zhen, China Community Diagnosis Questionnaire. A total of 1205 adults living in urban vil- lages and non-urban villages were included in the analysis. Unadjusted and multiple multivariate logistic regression were conducted for the dependent variable of engagement in recreational physical activity, frequency of recreational physical activity per week, and hours spent in sedentary behaviors per day. Descriptive analysis was conducted to identify the reasons for not engaging in physical activity among urban villagers and non-urban villagers. Across the included sample, 29.05% were urban villagers and 70.95% were non-urban vil- lagers. The results suggested that urban villagers are more likely to engage in physical activ- ity than non-urban villager (OR = 1.90, 95% CI [1.40, 2.59], p < 0.001). However, it was also found that urban village status had no significant association for frequency in engaging in physical activity and average hours spent in sedentary behaviors. Both urban villagers and non-urban villages indicated that lack of time, lack of safe and appropriate environment, and working in labor intensive occupations as some of the reasons for not engaging in physical activity. There is a need for tailed interventions and policies for promoting physical activity among urban villagers and non-urban villagers. Additional studies are needed to further our understanding of the physical activity behaviors among urban villagers in China. Introduction Benefits of physical activity a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Shi L, Leung W, Zheng Q, Wu J (2021) Level of engagement of recreational physical activity of urban villagers in Luohu, Shenzhen, China. PLoS ONE 16(10): e0258085. https://doi. org/10.1371/journal.pone.0258085 Editor: Francisco Javier Huertas-Delgado, La Inmaculada Teacher Training Centre (University of Granada), SPAIN Received: November 21, 2020 Accepted: September 20, 2021 Published: October 28, 2021 Copyright: © 2021 Shi et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: Due to legal and participants confidentiality, data will only be available upon request. The data underlying the results presented in the study are available from Shenzhen Luohu Disease Prevention and Control Center via contacting Weihong Chen, director of Shenzhen Luohu Disease Prevention and Control Center, at 1433529760@qq.com. Funding: The authors received no specific funding for this work. The benefit of engagement of physical activity is well documented [1, 2]. The numerous bene- fits included weight management, lower blood cholesterol levels and blood pressure, PLOS ONE | https://doi.org/10.1371/journal.pone.0258085 October 28, 2021 1 / 17 PLOS ONE Competing interests: The authors have declared that no competing interests exist. Urban villagers’ physical activity levels strengthening bones, muscles, and joint, and reducing risk of cardiovascular disease and cer- tain types of cancers [3]. In addition to physical health-related benefits, engagement in physical activity could lead to benefits of social and mental benefits. Regular engagement in physical activity is associated with reduced stress, improved mental health, emotional regulation, low- ered depression, increased social functioning, and increased sense of community [4]. Further, engagement of regular physical activity is related to reduce the risk of developing disabilities and maintenance of functional independences [5, 6]. Currently, physical inactivity is the fourth leading cause of mortality, according to the World Health Organization (WHO) [7]. WHO’s physical activity guidelines are 150 minutes of moderate physical activity or 75 minutes of vigorous physical activity per week or an equiva- lent combination of moderate- and vigorous-intensity activity for adults [7]. Individuals can perform various activities, such as leisure time physical activity, active transportation, and occupational activities to accumulate the minutes required to meet the guidelines. These guidelines apply to all individuals regardless of gender, race, ethnicity, or income levels. Physical activity levels of Chinese people Past literature had examined the physical activity levels individuals living in China [8, 9]. Using the data from the 2012 to 2015 China Hypertension Survey (CHS), it was found that 28.1% of Chinese adults were overweight and 5.2% were obese [10]. The results also found that regionals different of the prevalence of overweight and obesity different between Northern and Southern China with adults from Northern China more likely to be obese and overweight. According to a report published in the official Report on Cardiovascular Diseases in China 2017, 290 millions of Chinese adults are suffering from cardiovascular disease [11]. It was also found that China is facing a fast growing cardiovascular disease epidemic with a widening rural-urban disparities [12]. Similar physical activity trends found in Western countries were observed among Chinese adults as well. Trends such as male are more likely to engage in physical activity than female and older adults are less physical active than younger adults were found among individuals liv- ing in China [8, 9]. It was found that 66.3% of adults between the ages of 35 to 74 years were physically active according to the data from the International Collaborative Study of Cardio- vascular Disease in Asia from 2000–2001 [9]. Using accelerometers to measure physical activ- ity, it was found that Chinese adults in Shanghai spent 317 minutes per day in physical activity, while spent 509 minutes per day in sedentary behaviors [13]. It was reported that Chinese adults are more likely to report engaging in work-related or occupational physical activity (63.3%) than leisure time physical or recreational physical activity (24.5%) [9]. There were dis- parities between urban and rural residents with more rural residents (78.1%) spending time in physical activity than urban residents (21.8%) [9]. In addition to regional different, it was found that socioeconomic status (SES) impact physical activity levels among Chinese adults [14]. Using a community-based survey with 3567 adults living in Jiaxing, China, Chen et al. found that adults with lower SES are more likely to engage in household physical activity, adults with middle SES engages in higher intensity of occupational physical activity, and adults with higher SES levels were more likely to exercise but spent longer time in sedentary behav- iors [14]. The physical activity of subpopulation of Chinese adults had been well examined, especially for adults with different living area (rural vs. urban) and SES [14, 15]. However, there is a lack of literature examining the physical activity levels of urban villagers. Urban villagers refer to the individuals living in urban village. Urban village or chengzhongcun are typically low quality and high density with many closely packed apartment blocks of between 2 and 8 floors [16]. PLOS ONE | https://doi.org/10.1371/journal.pone.0258085 October 28, 2021 2 / 17 PLOS ONE Urban villagers’ physical activity levels Urban villages are transitional neighborhoods typically found in urban areas or cities with rapid economic growth [16, 17]. Urban villages can be described as narrow roads, face-to-face buildings, a thin strip of sky, and inner streets packed with shops, grocery stores and service outlets [16]. Many of these urban villages are associated with unsuitable land use, poor housing construction, severe infrastructure deficiencies, intensified social disorder, and deteriorated urban environment [18]. In addition, urban villages often have poor sanitary condition, where pipelines and drainage systems are poorly constructed and water flows over the ground along with garage [17]. Many urban villagers are individuals with low SES status due to financial situ- ation. These urban villagers could include rural-to-urban migrants workers with limited skill- sets and educations or individuals who recently graduated from colleges and universities. They are attracted to urban villages due to the cheap housing accommodation. Overall, these urban villagers aggregate in urban village in large cities, such as Guangzhou and Beijing with limited infrastructure and poor living environments due to affordable living accommodations. Due to the unique living situations of urban villages and limited healthcare resources [19], urban villagers’ physical activity need to be better examined [20]. Knowing physical activity- related information of urban villagers could better design and develop interventions targeting the needs of urban villagers in the community. Regular engagement in physical activity is asso- ciated with better health-related outcomes [21], considering urban villagers is more at risk for poor health outcomes due to poor living situation [22, 23]. Previous studies had examined the physical activity levels of youths and adolescents living in urban village [24, 25]. Therefore, to better understand the physical activity levels of adult urban villagers, the purpose of this study is to compare the physical activity and sedentary behaviors engagements between urban villag- ers and non-urban villagers using the 2019 Luohu Shenzhen, China Community Diagnosis Questionnaire. Materials and methods Design and sample This study is secondary data analysis using data from the 2019 Luohu Shenzhen, China Com- munity Diagnosis Questionnaire. The questionnaire is part of a community health diagnosis program funded by the Center for Disease Control and Prevention of Shenzhen. Due to the unique status of Shenzhen as the Special Economic Zones (SEZ), it attracted various Chinese citizens with different background to settle in the areas. This allows assessments of health- related behaviors on various groups of Chinese citizens (e.g., household registration status, migrants status, employments status, income levels, etc.) within the same survey and living within the same area. The goal of the survey is to grasp the main health problems existing in the residents of Luohu District, determine the causes of community health problems, and determine the priority needs of the public health services and factors affecting residents’ health. The survey also served as an evaluation of Shenzhen residents satisfaction on the vari- ous healthcare institutes available to them, such as community health centers. The survey con- sisted of seven parts: 1) family demographics, 2) family medical history, 3) adults healthcare needs and access to healthcare, 4) health and quality of life of adults over the ages of 60 years old, 5) health, healthcare and reproductive healthcare needs of married women under the ages of 50 years old, 6) healthcare needs and health of children, and 7) examination of blood pres- sure, height, weight, hip length, and waist length. Data collection of the survey was approved by the IRB at Shenzhen Luohu Disease Prevention and Control Center. Analysis of the survey data was approved by the IRB at Oregon State University. Participants of the survey were selected by multiple stages of random selection. First, seven communities were randomly selected in Dongmen community, Luohu district, Shenzhen, PLOS ONE | https://doi.org/10.1371/journal.pone.0258085 October 28, 2021 3 / 17 PLOS ONE Urban villagers’ physical activity levels Luohu as seen in Fig 1. Then 116 community grids were randomly selected from the seven selected communities in Dongmen community, Luohu district. Lastly, family household, serv- ing as survey unit, were randomly selected for interview based on the size of the community. All members of the household participated in the survey. Further, only individuals living in Shen- zhen for at least six months prior to the interview were included in the survey. The number of household participants in the survey is based on the size of the community. 200 households were randomly selected if the community sample size have more than two million individuals, 150 households for community sample size between one to two million, 100 households for community sample size between half of a million to one million, and 50 households for commu- nity less than half of a million. The random selection of communities was to identify individuals living in the various type of communities within the Shenzhen area. All data were collected between January and September of 2019. All data were collected through face-to-face interview. A total of 2122 participants were interviewed for the survey. However, only 2089 participants completed the survey with valid data. 1205 adults were included in the analysis. Across the sample, 54.52% of the participants were female and 45.48% of the participants were male. The average age of the participants were 38.8 years old. The average BMI were 22.88 Fig 1. Participants recruitment process. https://doi.org/10.1371/journal.pone.0258085.g001 PLOS ONE | https://doi.org/10.1371/journal.pone.0258085 October 28, 2021 4 / 17 PLOS ONE Urban villagers’ physical activity levels kg/cm2 with the hip-to-waist ratio of .89. A majority of the participants were employed (83.24%). Interestingly, 32.20% of the participants have no formal education and completed pri- mary education, which made up almost of one third of the sample. 28.54% of the participants completed middle school, 26.97% completed high school, and 12.28% completed professional school, college, and university. 76.02% of the participants were married or partnered and 23.98% were singled or not partnered. The sample consisted of more participants with non- Shenzhen hukou (62.41%). Across the sample, there were more participants without diagnosis of hypertension and diabetes. Only 6.56% and 1.91% of the participants reported having hyper- tension and diabetes, respectively. 76.43% of the participants reported they did not smoke. Measures The independent variable of the analysis is the living status of the participants. The variable is based on the location of the community grids participants resides in. Shenzhen used the com- munity grid system to identify local community [26]. Urban villages are typically located within one grid. Therefore, community grids serve as an indicator for urban villages. Partici- pants were classified as urban villagers if they live in an urban village and all other participants were classified as non-urban villagers. Physical activity and sedentary behavior-related variables from the survey component about health and quality of life of adults from 18 to 59 years old were selected for this analysis. A total of four variables were determined to be related to physical activity from the survey. The variables were engagement in recreational physical activity, frequency of recreational physical activity per week, hours spend in sedentary behaviors per day, and reasons for not engaging in physical activity. All variables were categorical variables. Engagement in recreational physical activity was based on the question of “Within the past six months, what types of recreational physical activity did you participated in?”. The respond options included: 1) did not participate in any activities, 2) machine equipment physical activity, 3) aerobic activity or aerobic dances, 4) swimming, 5) ambulatory activity (e.g., brisk walking, jogging, running, hiking), 6) ball- related sports (e.g. basketball, baseball, soccer, etc.), 7) sports or fitness competition, 8) martial arts, or 9) other. Participants were considered not to be engaged in physical activity when they responded with did not participate in any activities, else participants were classified as engaged in physical activity. Recreational physical activity is defined as physical activity that is done at leisure time. The variable of frequency of recreational physical activity were based on the ques- tion of “Within the past six months, how often do you exercise per week?” with the respond options of 1) 6 or more times per week, 2) 3 to 5 times per week, 3) 1 to 2 times per week, and 4) lesser than 1 time. The variable of hours spend in sedentary behaviors was based on the respond to the question of “In the past month, what is the average accumulated hours spend in sedentary activities (e.g., studying, working, watching TV, using computer, etc.)?”. The respond options included 1) lesser than 2 hours per day, 2) 2 to 4 hours per day, 3) 4 to 8 hours per day, 4) 8 to 12 hours per day, and 5) more than 12 hours per day. Reasons for not engaging in physical activity were only for participants who responded that they engaged in physical activity within the past six months. The survey item aims to identify how prevent them from engaging in physical activity throughout their routine. Participants were asked the reasons when they were unable to engage in physical activity weekly. Participants were able to select multiple options of 1) no recreational physical activity is needed due to labor intensive occupations, 2) no time to engage in physical activity, 3) there were no appropriate places and/ or environments for physical activity, 4) I feel healthy, I do not need physical activity, 5) do not want to engage in physical activity, 6) feeling ill, unable to participate in physical activity, and 7) other reasons. PLOS ONE | https://doi.org/10.1371/journal.pone.0258085 October 28, 2021 5 / 17 PLOS ONE Urban villagers’ physical activity levels The covariates included in the analysis were gender, age, employment status, education, marital status, household registration, body mass index (BMI), diagnosis of hypertension, diagnosis of diabetes, and smoking status. Gender was a binary variable consisted of male and female. Age was a continuous variable between 18 to 59 years old. Employment status was a binary variable of being employed or unemployed. Education was a categorical variable including professional college and university, high school, middle school, and primary school or no formal education. Marital status was a binary variable of either being married or single. Household registration or Hukou were based on participants self-reporting their registration of either Shenzhen Hukou or non-Shenzhen Hukou. BMI is a continuous variable between 15.02 to 36.11 kg/cm2, which was calculated based on the participants’ height and weight by the survey. Hip-to-waist was calculated based on the hip and waist of the participants by the survey. Diagnosis of hypertension, diagnosis of diabetes, and smoking status were all binary variables with yes and no. These covariates were selected due to their relationship with physical activity engagement. Data analyses Descriptive analysis was conducted for the independent variables, dependent variables, and the covariates. To determine the physical activity engagement between urban villagers and non-urban villagers, unadjusted and multiple multivariate logistic regression were conducted for the dependent variable of engagement in recreational physical activity, frequency of recrea- tional physical activity per week, hours spend in sedentary behaviors per day, and reasons for not engaging in physical activity. All analyses were conducted using STATA version 16 (Stata- Corp LLC., College Station, TX, USA). The alpha levels were set at .05. The study protocol was approved by the Oregon State University (IRB: IRB-2020-0509). Results Across the sample, 29.05% (n = 350) of the participants were urban villagers and 70.95% (n = 855) were non-urban villagers. Pearson’s chi square test found significant different between education levels, marital status, and household registration status between the urban villagers and non-urban villagers. There were more non-urban villagers with completed mid- dle school, high school, and professional school, college, and university (χ2 = 99.46, p < 0.001). There were more non-urban villagers who were either married or partnered than urban villag- ers (χ2 = 3.77, p = 0.05). Regrading to household registration or hukou, there were higher pro- portion of non-urban villagers with Shenzhen hukou and higher proportion of urban villagers with non-Shenzhen hukou (χ2 = 180.60, p < 0.001). Also, there were significant different in age found between the two groups with non-urban villagers had a higher average age. Non-sig- nificant differences were found between urban villagers and non-urban villagers among other covariates (e.g., gender, employment, diagnosis of hypertension, diagnosis of diabetes, smok- ing status, BMI, and hip-to-waist ratio). Engagement in recreational physical activity From the total sample size (n = 1205), 63.73% (n = 768) of participants reported not engage in any recreational physical activity while 36.27% (n = 474) reported engaged in recreational physical activity. A significant difference in proportion of engaging in recreational physical activity were found between urban and non-urban villagers (χ2 = 60.79, p < 0.001) with higher proportion of urban villagers (53.14%) reported engaging in recreational physical activity than non-urban villagers (29.36%) as shown in Table 1. The unadjusted logistic regression found that urban villagers were 2.73 (95% CI [2.11, 3.53], p < 0.001) times the odds of non-urban PLOS ONE | https://doi.org/10.1371/journal.pone.0258085 October 28, 2021 6 / 17 PLOS ONE Table 1. Characteristics of urban villagers and non-urban villagers engaging in recreation physical activity. Urban Villagers Non-Urban Villagers Total n Mean/Proportion n Mean/Proportion n Mean/Proportion χ2/ t P Urban villagers’ physical activity levels Engagement in recreational physical activity, % Yes No Frequency of recreational physical activity per week, % > 6 times 3–5 times 1–2 times < 1 time Average hours spend in sedentary behaviors per day, % > 12 hours 9–12 hours 5–8 hours 2–4 hours < 2 hours Gender, % Female Male Age, years Employment status, % Yes No Education levels, % College & university High school Middle school Primary school & none Marital Status Married/partnered Single Household registration (hukou), % Shenzhen hukou Non-Shenzhen hukou Body Mass Index, kg/m2 Hip-to-waist ratio, % Hypertension, % Yes No Diabetes, % Yes No Smoking status, % Yes No 186 164 35 57 62 10 15 46 99 86 104 183 167 350 302 48 46 108 122 74 253 97 29 321 350 350 19 331 8 342 94 256 53.14 46.86 21.34 34.76 37.80 6.10 4.29 13.14 28.29 24.57 29.71 52.29 47.71 37.75 86.29 13.71 13.14 30.86 34.86 21.14 72.29 27.71 8.29 91.71 22.99 89 5.43 94.57 2.29 97.71 26.86 73.14 251 604 154 201 216 34 43 113 221 261 217 474 381 855 701 154 342 236 203 74 663 192 424 431 855 855 60 795 15 840 190 665 29.36 70.64 25.45 33.22 35.70 5.62 5.03 13.22 25.85 30.53 25.38 55.44 44.56 39.24 81.99 18.01 40.00 27.60 23.74 8.65 77.54 22.46 49.59 50.41 22.83 89 7.02 92.98 1.75 98.25 22.22 77.78 437 768 189 358 278 44 58 159 320 347 321 657 548 1205 1003 202 148 325 344 388 916 286 453 752 1205 1205 79 1126 23 1182 284 921 36.27 63.73 21.75 41.20 31.99 5.06 4.81 13.20 26.56 28.80 26.64 54.52 45.48 38.8 83.24 16.76 12.28 26.97 28.55 32.20 76.02 23.98 37.59 62.41 22.87 89 6.56 93.44 1.91 98.09 81.66 18.34 60.79 <0.001� 1.19 0.76 5.65 0.23 1.00 0.32 2.21 0.03� 3.29 0.07 99.46 <0.001� 3.77 0.05 180.60 0 < .001� -0.70 0.41 0.48 0.68 1.02 0.31 0.37 0.54 2.96 0.09 Note. n, sample size; χ2, chi-square statistic comparing between Urban Villagers and non-Urban Villagers for categorical variables; t, t-statistic comparing between Urban Villagers and non-Urban Villagers for continuous variables, p, p-value associated with the statistic comparison test; �, p < 0.05. https://doi.org/10.1371/journal.pone.0258085.t001 PLOS ONE | https://doi.org/10.1371/journal.pone.0258085 October 28, 2021 7 / 17 PLOS ONE Urban villagers’ physical activity levels villagers in engaging in recreational physical activity as shown in Table 2. The results of the multivariate logistic regression found that Urban Villagers were 1.90 (95% CI [1.40, 2.57], p < 0.001) times the odds of non-urban villagers in engaging in recreational physical activity after controlling for covariates. The analysis also found the education levels, household regis- tration, and BMI are significant factors contributing to the results of the odds ratios between urban villagers and non-urban villagers in engaging in recreational physical activity. Frequency of recreational physical activity per week 21.34% of urban villagers reported engaging in recreational physical activity more than six times per week, in compared to 25.45% of non-urban villagers reported the same frequency. 34.76% of urban villagers and 33.72% of non- urban villagers reported engaging in recreational physical activity 3 to 5 time per week, 37.80% of urban villagers and 35.70% of non- urban villagers reported engaging recreational physical activity 1 to 2 times per week. And 6.10% of urban vil- lagers and 5.62% of non-urban villagers reported engaged in lesser than recreational physical activity per week. No significant different was found between the two groups regarding the fre- quency of engaging recreational in physical activity per week (χ2 = 1.19, p = 0.76). The odds ratio of the unadjusted logistic regression for each level of the frequency of engaging in recrea- tional physical activity per week with references of less than 1 time per week were 0.98 (95% CI [0.46, 2.09], p = 0.95) for 1 to 2 time per week, 0.96 (95% CI [0.45, 2.07], p = 0.93) for 3 to 5 times per week, and 0.77 (95% CI [0.35, 1.71], p = 0.95) for more than six times per week for urban villagers in engaging in recreational physical activity compared to non-urban villagers. The results of the multivariate logistic regress found that urban villagers status is not a significant factor in estimating the odds ratio of frequency in engaging recreational physical activity per week with the reference groups of lesser than 1 time per week as shown in Tables 3 and 4. Average hours spend in sedentary behaviors per day 4.29% of urban villagers and 5.03% non-urban villagers reported spending more than 12 hours per day in sedentary, which made up the smallest proportion of the participants in their respective group. 13.14% of urban villagers and 13.22% of non-urban villagers reported Table 2. Odd ratios of urban villagers and non-urban villagers in engaging in recreational physical activity. Urban villagers Non-urban villagers Engagement in recreational physical activity Unadjusted Modelb Adjusted Modelc OR 2.73� 1 (ref.) 95% CI 2.11, 3.53 OR 1.90� 1 (ref.) 95% CI 1.40, 2.57 Abbreviations: OR, odds ratio; CI, confidence interval. aBoldfaced numerals indicate p-value <0.05. bOdd ratio from logistic regression model were computed for the outcome variable of engagement in recreational physical activity (yes/no) with the exposure variable of living situation (urban village/non-urban village). cOdd ratio from multivariable logistic regression model were computed for the outcome variable of engagement in recreational physical activity (yes/no) with the exposure variable of living situation (urban village/non-urban village) adjusted for gender (male/female), age (continuous), employment status (yes/no), education levels (college & university, high school, middle school, primary school & none), marital status (married & partnered/single), household registration (hukou) (Shenzhen/non-Shenzhen), BMI (continuous), hip-to-waist ratio (continuous), hypertension (yes/no), diabetes (yes/no), and smoking status (yes/no). d Detail adjusted model outcome were showed in S1 Table. https://doi.org/10.1371/journal.pone.0258085.t002 PLOS ONE | https://doi.org/10.1371/journal.pone.0258085 October 28, 2021 8 / 17 PLOS ONE Urban villagers’ physical activity levels Table 3. Odd ratios of frequency of engaging in recreational physical activity per week between urban villagers and non-urban villagers. 1–2 times vs. < 1 time (ref.) OR 0.98 95% CI 0.46, 2.09 Urban villagers Non-urban villagers 1 (ref.) Unadjusted odd ratiosb 3–5 times vs. < 1 time (ref.) > 6 times vs. < 1 time (ref.) 1–2 times vs. < 1 time (ref.) Adjusted odd ratiosc 3–5 times vs. < 1 time (ref.) > 6 times vs. < 1 time (ref.) OR 0.96 1 (ref.) 95% CI 0.45, 2.07 95% CI 0.35, 1.71 OR 0.77 1 (ref.) 95% CI .44, 2.64 AOR 1.07 1 (ref.) AOR 0.98 1 (ref.) 95% CI .39, 2.43 AOR 0.83 1 (ref.) 95% CI .32, 2.15 Abbreviations: OR, odds ratio; CI, confidence interval. aBoldfaced numerals indicate p-value <0.05. bOdd ratio from logistic regression model were computed for the outcome variable of engagement in recreational physical activity (yes/no) with the exposure variable of living situation (urban village/non-urban village). cOdd ratio from multivariable logistic regression model were computed for the outcome variable of engagement in recreational physical activity (yes/no) with the exposure variable of living situation (urban village/non-urban village) adjusted for gender (male/female), age (continuous), employment status (yes/no), education levels (college & university, high school, middle school, primary school & none), marital status (married & partnered/single), household registration (hukou) (Shenzhen/non- Shenzhen), BMI (continuous), hip-to-waist ratio (continuous), hypertension (yes/no), diabetes (yes/no), and smoking status (yes/no). d Detail adjusted model outcome were showed in S2 Table. https://doi.org/10.1371/journal.pone.0258085.t003 spending 8 to 12 hours per day in sedentary behaviors. 28.29% of urban villagers and 25.85% of non-urban villagers reported spending 4 to 8 hours per day on sedentary behaviors, while 24.57% and 30.53% of urban villagers and non-urban villagers spend 2 to 4 hours per day on sedentary behaviors. For lowest amount of time spend in sedentary behaviors, 29.71% of urban villagers and 25.36% of non-urban villagers reported spending lesser than 2 hours on it. Non-significant different was found between the two groups regarded to the self-reported hours spend in sedentary hours (χ2 = 5.65, p = 0.23). From the unadjusted logistic regression with the reference group of spending less than 2 hours per day in sedentary behaviors and urban villagers, the odd ratios were 0.69 (95% CI [0.49, 0.96], p = 0.03) for 2 to 4 hours, 0.93 (95% CI [0.67, 1.30], p = .69) for 4 to 8 hours, and 0.85 (95% CI [0.56, 1.29], p = 0.44) for 8 to 12 hours. The results of the multivariate logistic regression found that urban villagers status is not a significant factor in estimating the hours spend in sedentary behaviors per day with the reference groups of lesser than 2 hours per day as shown in Table 4. However, across all levels Table 4. Odd ratios of average hours spend in sedentary behaviors per day between urban villagers and non-urban villagers. 2–4 hours vs. < 2 hours 5–8 hours vs. < 2 hours 9–12 hours vs. < 2 hours >12 hours vs. < 2 hours 2–4 hours vs. < 2 hours 5–8 hours vs. < 2 hours 9–12 hours vs. < 2 hours >12 hours vs. < 2 hours OR 0.69� 1 (ref.) Urban villagers Non-urban villagers 95% CI OR 95% CI .49, .96 0.93 .67, 1.30 OR 0.85 95% CI .56, 1.29 OR 0.73 95% CI .39, 1.37 OR 0.85 95% CI .58, 1.25 OR 1.18 95% CI .79, 1.75 OR 1.43 95% CI .86, 2.38 OR 0.06 95% CI 0, 2.07 1 (ref.) 1 (ref.) 1 (ref.) 1 (ref.) 1 (ref.) 1 (ref.) 1 (ref.) Abbreviations: OR, odds ratio; CI, confidence interval. aBoldfaced numerals indicate p-value <0.05. bOdd ratio from logistic regression model were computed for the outcome variable of engagement in recreational physical activity (yes/no) with the exposure variable of living situation (urban village/non-urban village). cOdd ratio from multivariable logistic regression model were computed for the outcome variable of engagement in recreational physical activity (yes/no) with the exposure variable of living situation (urban village/non-urban village) adjusted for gender (male/female), age (continuous), employment status (yes/no), education levels (college & university, high school, middle school, primary school & none), marital status (married & partnered/single), household registration (hukou) (Shenzhen/non- Shenzhen), BMI (continuous), hip-to-waist ratio (continuous), hypertension (yes/no), diabetes (yes/no), and smoking status (yes/no). d Detail adjusted model outcome were showed in S3 Table. https://doi.org/10.1371/journal.pone.0258085.t004 PLOS ONE | https://doi.org/10.1371/journal.pone.0258085 October 28, 2021 9 / 17 PLOS ONE Urban villagers’ physical activity levels Fig 2. Reasons for not engaging in physical activity among urban villagers and non-urban villagers who engage in physical activity. https://doi.org/10.1371/journal.pone.0258085.g002 of hours spend in sedentary behaviors, completing professional school, college, and university had a higher odd of spending more time in sedentary behaviors. Reasons for not engagement in recreational physical activity Among participants who engage in recreational physical activity, many indicated that no time to exercise as the main reason why they did not engage in physical activity (n = 273) as shown in Fig 2. The second top reasons participants selected as the reasons for not engaging in recrea- tion physical activity was no need to exercise due to labor intensive occupation (n = 91), follow by unwilling to exercise and no place to exercise (n = 69). Some participants also respond that they did not engage in recreation physical activity due to feeling healthy (n = 14) and no need to exercise and unable to engage in recreational physical activity due to illness (n = 5). When stratified by urban village status, lack of time is the most cited reason for not engag- ing in physical activity for both urban villagers (n = 107) and non-urban villagers (n = 166). There were more urban villagers (n = 58) compared to non-urban villagers expressed that they do not need to engage in physical activity due to occupations being labor intensive. There were more non-urban villagers (n = 51) expressed that they were unwilling to engage in physical activity than urban villagers (n = 18). Also, higher number of non-urban villagers (n = 28) reported not having appropriate places and/or environments for physical activity compared to urban villagers (n = 19). Discussion The purpose of this secondary data analysis is to determine and compare the prevalence of physical activity engagement among the special population of Chinese urban villagers and non-urban villagers. Both the unadjusted and adjusted logistic regression identified that urban PLOS ONE | https://doi.org/10.1371/journal.pone.0258085 October 28, 2021 10 / 17 PLOS ONE Urban villagers’ physical activity levels villagers are more likely to engage in recreational physical activity than their counterpart of non-urban villagers. No significant relationship was found between the frequency of engage- ment in recreational physical activity and urban village status. The multinomial logistic regres- sion also found no significant relationship between hours spend in sedentary behaviors and urban village status. Descriptive analysis shown that both urban villagers and non-urban vil- lagers shared reasons for not engaging in recreational physical activity, such as lack of time to exercise. However, more urban villagers indicated that their labor-intensive occupations are sufficient enough for physical activity. While more non- urban villagers indicated that they are more unwilling to exercise and there are no appropriate places and/or environments for recre- ational physical activity. While both urban villagers and non-urban villagers live in urban and well-developed area, the levels of engagement in recreational physical activity were different between the two groups. The results demonstrated that even within the same city, engagement in recreational physical activity could be different by social characteristics. Urban villagers, like non-urban vil- lagers, have access to different public physical activity facilities within the urban area. Physical activity facilities such as parks, sidewalks, and outside of the urban villages are facilities that urban villagers have access to. This is supported by the results that less urban villagers indi- cated that there are a lack of appropriate places and/or environments for recreational physical activity in compared to non-urban villagers. The ability of utilizing free public physical activity facilities increase the opportunities for urban villagers to engage in recreational physical activ- ity. Having these opportunities allow for urban villagers to obtain a healthier lifestyle of regu- larly engagement in recreational of physical activity. While it has been found that lower- income neighborhoods, such as urban villages, have less commercial physical activity-related facilities [27]. The results of this study was different from the study conducted by Ortiz-Her- na´ndez and Ramos-Iba´ñez [28], where they found that Mexican adults living in urban locali- ties and cities with low socio-economic status had a lower probability of engaging in physical activity. However, it is difficult to compare results across different countries as culture and environments are widely different between the countries. Therefore, it is not appropriate to compare the results between the studies. Studies conducted in the US [29] and in the Europe [30] found similar results of adults living in rural areas less likely to engage in physical activity and other psychosocial factors could influence physical activity behaviors. These highlight that there is a need of global effort to promote physical activity in various countries. Further, due to the unique situation of urban village in China, where the housing is surrounded by well-devel- oped buildings and infrastructures, urban villagers have easy access to these different infrastructures. Income status could potentially be one of the factors explaining the different proportion of urban villagers and non-urban villagers in engagement of recreational physical activity. Indi- viduals living in urban village are more likely to be individuals with lower economic status. Many of these individuals chose to reside in urban village due to the cheap accommodation [31, 32]. Further, many of these individuals might held lower wages and labor-intensive occu- pations. As evidence by the results of reasons for not engaging recreational physical activity, more urban villagers reported that their occupations are labor intensive enough that either they are too tired to engage in additional physical activity or they felt that they do not need to engage in additional physical activity. This aligned with previous study finding that more rural adults in China engage in work-related physical activity than urban adults [9]. In comparison to urban villagers, fewer participants in the non-urban village group reporting their occupa- tions are too physically demanding that they felt that engagement in recreational physical activity is not necessary. Non-urban villagers are more likely to held office-related occupations, therefore, it limits their ability to engage in physical activity. Past studies had demonstrated PLOS ONE | https://doi.org/10.1371/journal.pone.0258085 October 28, 2021 11 / 17 PLOS ONE Urban villagers’ physical activity levels that officer workers are more likely to engage in less physical activity and more sedentary behaviors [33]. Further, non-urban villagers might be more likely to have better technology access than urban villagers. Technology such as television and media are found to be associated with lower physical activity levels and high sedentary behavior [34, 35]. This might relate to the higher number of non-Urban Villagers reporting unwilling to engage in recreational physi- cal activity. It is also surprising to find that there are higher numbers of non-urban villagers indicating that the reason for not engaging in recreational physical activity was lack of appro- priate places and/or environments. Being consistent with previous research by Munter et al. [9] where Chinese urban adults are less likely to engage in physical activity than Chinese adults with lower economic status living in rural area. Based on the results of this study, more tailed intervention is needed for Chinese adults not living in urban villages. Even though urban villagers are more likely to be in poor health due to poor housing situation [36, 37], they are more likely to engage in recreational physical activity than non-urban villagers. While the two groups have large number of participants reporting lack of time to engage in recreational physical activity, different interventions should be devel- oped for the two groups. Due to differences in living situations, economic status, and occupa- tions, different reactions and responses to interventions might be different between urban villagers and non-urban villagers. When designing physical activity interventions, there is a need to consider demographic characteristics and socioeconomic factors. For urban villagers, tailed interventions are needed to target group of individuals that believe that physical activity performed during their job are sufficient enough for health. Multiple studies had demon- strated that leisure time physical activity and recreational physical activity are associated with better health quality of life [38–40]. Occupational-related physical activity is not considered to be recreation or leisure physical activity. Therefore, specific interventions are needed targeting urban villagers. Developing interventions in targeting these reasons and solving these barriers for non-urban villagers will be important step for increase the proportion of non-urban villag- ers in engaging in recreational physical activity. For example, Gu et al. [41] found change in physical activity among office workers after the implementation of a worksite intervention programs at 17 worksites in the urban city of Shanghai with pedometers for 100 days. The goal of using and developing physical activity interventions are to promote recreational physical activity levels among both urban villagers and non-urban villagers. Further research and studies are warrants in determine the physical activity levels among urban villagers and non-urban villagers. Study had done in the past to examine the physical activity levels of Chines adults [9, 13, 42], but there is a lack of empirical evidence on the physical activity levels of urban villagers. Using additional techniques, such as accelerometers, to collected more detailed data could increase our understanding of physi- cal activity levels of urban villagers. More detailed data such as minutes spend in each intensity of physical activity or number of steps taken each day can better represent the physical activity levels of urban villagers. It has been proposed that an intersectionality approach should be taken when measuring and discussing physical activity levels [43–45]. The interacting factors could provide more detail information on the physical activity of special population such as urban villagers. Often, urban villagers might be considered as individuals living in urban area. However, due to the unique situation of urban village, they are considered a special population living in the urban area. This study demonstrated that there is a need to examine the physical activity levels of special populations living in China. As shown in this study, the proportion of urban villagers and non-urban villagers engaging in recreational physical activity is different, so more research is needed. This data could further facilitate the development of physical activity intervention targeting urban villagers and non-urban villagers. Future researches should also focus on urban PLOS ONE | https://doi.org/10.1371/journal.pone.0258085 October 28, 2021 12 / 17 PLOS ONE Urban villagers’ physical activity levels villagers and non-urban villagers in meeting physical activity guidelines by the World Health Organization [21]. The current physical activity guidelines for adults over the ages of 18 years old is at least 150 minutes of moderate-to-vigorous physical activity or 75 min- utes of vigorous physical activity per week. Examining the prevalence of urban villagers and non-urban villagers in meeting these physical activity guidelines could increase our understanding of the physical activity behaviors and dose-response relationship between physical activity and health among these populations. It is important to note that there is a lack of national and regional physical activity guidelines in China [46]. Developing these physical activity guidelines could be beneficial for Chinese citizen as there is a guideline for them to follow. One interesting find of the analysis was that education might have an influence on physical activity-related outcomes among urban villagers and non-urban villagers. Based on the adjusted logistic regression model, in compare to no formal education and only completing primary education, other education levels (i.e., middle school, high school, professional college and university) are less likely to engage in physical activity. The analysis also found that higher education is associated with longer time spent in sedentary behaviors. The results align with previous study examining the decline of physical activity levels among Chinese adults [14]. The study found that the greater availability of higher educational institutions is strongly asso- ciated with the declines of physical activities based on data from the 1991–2006 China Health and Nutrition Surveys [47]. Individuals with higher education are more likely to have office- related positions. Officer workers are more likely to spend more time in sedentary behaviors [48]. In addition, it was found that Chinese adults who completed high school education are less likely to engage in occupational-related physical activity [9]. These results suggested that physical activity interventions are needed for individuals with higher education. To ensure that physical activity become a lifelong habit among Chinese adults, there is need to develop physical activity intervention targeting adults at various educational levels. For example, requiring physical education or physical activity classes for students in middle schools, high schools, and colleges and universities. Requirement of physical education in early childhood is positively associated with physical activity levels in adulthood [49]. Individuals who had taken a physical activity course while in colleges and universities report higher physical activity levels in adulthood compared to those that did not take a physical activity course [50]. Continuation promotion of physical activity through various different educational institutions could poten- tially increase physical activity levels of adults. Limitation To the authors’ knowledge, this is the first of the few studies that examined the physical activity levels of urban villagers in China. The strength of this study is including the special population of urban villagers. However, this study is not without its limitation. The data used in the analy- sis are based on self-reported data. There could be potential recall and social bias. These biases could lead to misclassification of data and results [51]. In addition to biases, there could be low generalizability of the results. Due to the data only included participants living in the Luohu, Shenzhen, China, the results might be only generalized to this particular populations living in Shenzhen. However, it is assumed that urban villagers across China shared the similar charac- teristics of lower economic status, migrant workers, labor intensive worker, poor living situa- tion, and lack of infrastructures. It is important to note that the survey did not utilized the International Physical Activity Questionary (IPAQ) in the surveillance system. This could lead to misunderstanding of questions by the participants. To limit misunderstanding, all data col- lected were in Chinese via face-to-face interview by trained personals. PLOS ONE | https://doi.org/10.1371/journal.pone.0258085 October 28, 2021 13 / 17 PLOS ONE Urban villagers’ physical activity levels Conclusion Overall, the proportion of urban villagers and non-urban villagers in engaging in recreational physical activity are different with urban villagers more likely to engage in recreational physical activity. While participants from both groups expressed that lack of time as a barrier in engag- ing in recreational physical activity, non-urban villagers are more likely to reported that they are unwilling to participate in recreational physical activity and lack appropriate place and/ environment for recreational physical activity. Urban villagers are more likely to reported that they do not engage in recreational physical activity due to work-related physical activity. Physi- cal activity interventions are needed to target these various barriers in preventing urban villag- ers and non-urban villagers in participating from recreational physical activity. Further research is warranted in order to better understanding the physical activity levels of the special population of urban villagers living in China. Supporting information S1 Table. Odd ratios of urban villagers and non-urban villagers in engaging in recreational physical activity: Adjusted model outcomes. (DOCX) S2 Table. Odd ratios of frequency of engaging in recreational physical activity per week between urban villagers and non-urban villagers: Adjusted model outcomes. (DOCX) S3 Table. Odd ratios of average hours spend in sedentary behaviors per day between urban villagers and non-urban villagers. (DOCX) S1 File. Questionnaire Chinese. (DOCX) S2 File. Questionnaire English. (DOCX) Author Contributions Conceptualization: Lu Shi, Willie Leung, Qingming Zheng, Jie Wu. Data curation: Lu Shi, Qingming Zheng, Jie Wu. Formal analysis: Lu Shi, Willie Leung. Investigation: Jie Wu. Methodology: Lu Shi, Willie Leung, Qingming Zheng, Jie Wu. Project administration: Qingming Zheng, Jie Wu. Resources: Qingming Zheng. Software: Lu Shi. Supervision: Qingming Zheng. Visualization: Lu Shi. Writing – original draft: Lu Shi, Willie Leung. Writing – review & editing: Lu Shi, Willie Leung. PLOS ONE | https://doi.org/10.1371/journal.pone.0258085 October 28, 2021 14 / 17 PLOS ONE Urban villagers’ physical activity levels References 1. Haskell WL, Blair SN, Hill JO. Physical activity: Health outcomes and importance for public health pol- icy. Preventive Medicine. 2009 Oct 1; 49(4):280–2. https://doi.org/10.1016/j.ypmed.2009.05.002 PMID: 19463850 2. Steinbeck KS. The importance of physical activity in the prevention of overweight and obesity in child- hood: a review and an opinion. Obesity Reviews. 2001; 2(2):117–30. https://doi.org/10.1046/j.1467- 789x.2001.00033.x PMID: 12119663 3. Warburton DER, Nicol CW, Bredin SSD. Health benefits of physical activity: the evidence. Canadian Medi- cal Association Journal. 2006 Mar 14; 174(6):801–9. https://doi.org/10.1503/cmaj.051351 PMID: 16534088 4. Eime RM, Young JA, Harvey JT, Charity MJ, Payne WR. A systematic review of the psychological and social benefits of participation in sport for adults: informing development of a conceptual model of health through sport. Int J Behav Nutr Phys Act. 2013 Dec 7; 10:135. https://doi.org/10.1186/1479-5868-10- 135 PMID: 24313992 5. Singh MAF. Exercise to prevent and treat functional disability. Clin Geriatr Med. 2002 Aug; 18(3):431– 62, vi–vii. https://doi.org/10.1016/s0749-0690(02)00016-2 PMID: 12424867 6. Tak E, Kuiper R, Chorus A, Hopman-Rock M. Prevention of onset and progression of basic ADL disabil- ity by physical activity in community dwelling older adults: A meta-analysis. Ageing Research Reviews. 2013 Jan 1; 12(1):329–38. https://doi.org/10.1016/j.arr.2012.10.001 PMID: 23063488 7. World Health Organization. WHO | Physical Inactivity: A Global Public Health Problem [Internet]. WHO. World Health Organization; 2020 [cited 2020 Sep 22]. Available from: https://www.who.int/ dietphysicalactivity/factsheet_inactivity/en/. 8. Li F. Physical activity and health in the presence of China’s economic growth: Meeting the public health challenges of the aging population. Journal of Sport and Health Science. 2016 Sep 1; 5(3):258–69. https://doi.org/10.1016/j.jshs.2016.06.004 PMID: 30356539 9. Muntner P, Gu D, Wildman RP, Chen J, Qan W, Whelton PK, et al. Prevalence of Physical Activity Among Chinese Adults: Results From the International Collaborative Study of Cardiovascular Disease in Asia. Am J Public Health. 2005 Sep 1; 95(9):1631–6. https://doi.org/10.2105/AJPH.2004.044743 PMID: 16051938 10. Zhang L, Wang Z, Wang X, Chen Z, Shao L, Tian Y, et al. Prevalence of overweight and obesity in China: Results from a cross-sectional study of 441 thousand adults, 2012–2015. Obesity Research & Clinical Practice. 2020 Mar 1; 14(2):119–26. 11. Natinal Center for Cardiovascular Disease. China. Report on Cardiovascular Diseases in China 2017. Beijing: Encyclopedia of China Publishing House; 2018. 12. Liu L. Rural-Urban Disparities in Cardiovascular Disease Mortality Among Middle-Age Men in China. Asia Pac J Public Health. 2020 Sep 11;1010539520956446. https://doi.org/10.1177/ 1010539520956446 PMID: 32917101 13. Peters TM, Moore SC, Xiang YB, Yang G, Shu XO, Ekelund U, et al. Accelerometer-Measured Physical Activity in Chinese Adults. American Journal of Preventive Medicine. 2010 Jun 1; 38(6):583–91. https:// doi.org/10.1016/j.amepre.2010.02.012 PMID: 20494234 14. Chen M, Wu Y, Narimatsu H, Li X, Wang C, Luo J, et al. Socioeconomic Status and Physical Activity in Chinese Adults: A Report from a Community-Based Survey in Jiaxing, China. PLOS ONE. 2015 Jul 15; 10(7):e0132918. https://doi.org/10.1371/journal.pone.0132918 PMID: 26177205 15. Shi Z, Lien N, Kumar BN, Holmboe-Ottesen G. Physical activity and associated socio-demographic fac- tors among school adolescents in Jiangsu Province, China. Preventive Medicine. 2006 Sep 1; 43 (3):218–21. https://doi.org/10.1016/j.ypmed.2006.04.017 PMID: 16762405 16. 17. 18. Liu Y, He S. Chinese Urban Villages as Marginalized Neighbourhoods under Rapid Urbanization. In: Wu F, Webster C, editors. Marginalization in Urban China: Comparative Perspectives [Internet]. Lon- don: Palgrave Macmillan UK; 2010 [cited 2020 Oct 2]. p. 177–200. (International Political Economy Series). Available from: https://doi.org/10.1057/9780230299122_10 Liu Y, He S, Wu F, Webster C. Urban villages under China’s rapid urbanization: Unregulated assets and transitional neighbourhoods. Habitat International. 2010 Apr 1; 34(2):135–44. Zhang L, Zhao SXB, Tian JP. Self-help in housing and chengzhongcun in China’s urbanization. Interna- tional Journal of Urban and Regional Research. 2003; 27(4):912–37. 19. Shi L, Patil VP, Leung W, Zheng Q. Willingness to use and satisfaction of primary care services among locals and migrants in Shenzhen, China. Health & Social Care in the Community [Internet]. [cited 2021 Jul 7];n/a(n/a). Available from: https://onlinelibrary.wiley.com/doi/abs/10.1111/hsc.13418 PMID: 33978287 20. Yang X. Research on the Optimization of Social Sports Participation Path in “Urban Villages.” In 2018 [cited 2020 Oct 23]. Available from: https://kns.cnki.net/KCMS/detail/detail.aspx?dbcode=IPFD&db PLOS ONE | https://doi.org/10.1371/journal.pone.0258085 October 28, 2021 15 / 17 PLOS ONE Urban villagers’ physical activity levels name=IPFDLAST2018&filename=LRCM201802002061&v=MTc1OTRiS0lGc1hLVC9JWTdHNEg5b k1yWTlGWnVzSkRSTkt1aGRobmo5OFRuanFxeGRFZU1PVUtyaWZaZVp1RmluZ1Vy. 21. World Health Organization. Global recommendations on physical activity for health. Geneva: World Health Organization; 2010. 22. Bork T, Kraas F, Xue D, Li Z. Urban environmental health challenges in China’s villages-in-the-city. Geographische Zeitschrift. 2011; 99(1):16–35. 23. Jiang J, Wang P. Health status in a transitional society: urban-rural disparities from a dynamic perspec- tive in China. Population Health Metrics. 2018 Dec 27; 16(1):22. https://doi.org/10.1186/s12963-018- 0179-z PMID: 30591053 24. Duan J, Cao D. Investigation on the Status Quo of Mass Sports Participation in Urban Villages——Tak- ing Wanbolin District of Taiyuan City as an Example [Internet]. North University of China; 2017 [cited 2020 Oct 23]. Available from: https://kns.cnki.net/KCMS/detail/detail.aspx?dbcode=CJFQ&dbname= CJFDLAST2017&filename=TYGJ201703008&v=MDMxOTk4ZVgxTHV4WVM3RGgxVDNxVHJXT TFGckNVUjdxZVorWnJGaUhuVnIvTE1UVE1aTEc0SDliTXJJOUZiSVI=. 25. Long H, Li Y, Yu H, Yang W, Wu M. Research on Youth Sports Participation in Urban Villages of Kun- ming City in the Process of Urbanization. Contemporary Sports Technology. 2019; 9(06):167–8. 26. Chen X. The Problems Research of Griddization Management in Longgang District of Shenzhen City [Internet] [Master]. Central China Normal University; 2015 [cited 2020 Oct 24]. Available from: https:// gb.oversea.cnki.net/kcms/detail/detail.aspx?recid=&FileName=1016038151.nh&DbName= CMFD201602&DbCode=CMFD. 27. Powell LM, Slater S, Chaloupka FJ, Harper D. Availability of Physical Activity–Related Facilities and Neighborhood Demographic and Socioeconomic Characteristics: A National Study. Am J Public Health. 2006 Sep; 96(9):1676–80. https://doi.org/10.2105/AJPH.2005.065573 PMID: 16873753 28. Ortiz-Herna´ ndez L, Ramos-Iba´ñez N. Sociodemographic factors associated with physical activity in Mexican adults. Public Health Nutrition. 2010 Jul; 13(7):1131–8. https://doi.org/10.1017/ S1368980010000261 PMID: 20196912 29. Wilcox S, Castro C, King AC, Housemann R, Brownson RC. Determinants of leisure time physical activ- ity in rural compared with urban older and ethnically diverse women in the United States. Journal of Epi- demiology & Community Health. 2000 Sep 1; 54(9):667–72. https://doi.org/10.1136/jech.54.9.667 PMID: 10942445 30. Moreno-Llamas A, Garcı´a-Mayor J, De la Cruz-Sa´nchez E. Urban-rural differences in trajectories of physical activity in Europe from 2002 to 2017. Health & Place. 2021 May 1; 69:102570. https://doi.org/ 10.1016/j.healthplace.2021.102570 PMID: 33873131 31. Keung Wong DF, Li CY, Song HX. Rural migrant workers in urban China: living a marginalised life: Rural migrant workers in urban China. International Journal of Social Welfare. 2007 Jan; 16(1):32–40. 32. Lu Z, Song S. Rural–urban migration and wage determination: The case of Tianjin, China. China Eco- nomic Review. 2006 Jan; 17(3):337–45. 33. Parry S, Straker L. The contribution of office work to sedentary behaviour associated risk. BMC Public Health. 2013 Apr 4; 13:296. https://doi.org/10.1186/1471-2458-13-296 PMID: 23557495 34. Harris JL, Bargh JA. The Relationship between Television Viewing and Unhealthy Eating: Implications for Children and Media Interventions. Health Commun. 2009 Oct; 24(7):660–73. https://doi.org/10. 1080/10410230903242267 PMID: 20183373 35. Keadle SK, Arem H, Moore SC, Sampson JN, Matthews CE. Impact of changes in television viewing time and physical activity on longevity: a prospective cohort study. International Journal of Behavioral Nutrition and Physical Activity. 2015 Dec 18; 12(1):156. https://doi.org/10.1186/s12966-015-0315-0 PMID: 26678502 36. Gao Y, Shahab S, Ahmadpoor N. Morphology of Urban Villages in China: A Case Study of Dayuan Vil- lage in Guangzhou. Urban Science. 2020 May 7; 4(2):23. 37. Wu F. Housing in Chinese Urban Villages: The Dwellers, Conditions and Tenancy Informality. Housing Studies. 2016 Oct 2; 31(7):852–70. 38. Tessier S, Vuillemin A, Bertrais S, Boini S, Le Bihan E, Oppert J-M, et al. Association between leisure- time physical activity and health-related quality of life changes over time. Preventive Medicine. 2007 Mar 1; 44(3):202–8. https://doi.org/10.1016/j.ypmed.2006.11.012 PMID: 17208289 39. Vuillemin A, Boini S, Bertrais S, Tessier S, Oppert J-M, Hercberg S, et al. Leisure time physical activity and health-related quality of life. Preventive Medicine. 2005 Aug 1; 41(2):562–9. https://doi.org/10. 1016/j.ypmed.2005.01.006 PMID: 15917053 40. Wendel-Vos GCW, Schuit AJ, Tijhuis MAR, Kromhout D. Leisure time physical activity and health- related quality of life: Cross-sectional and longitudinal associations. Qual Life Res. 2004 Apr 1; 13 (3):667–77. https://doi.org/10.1023/B:QURE.0000021313.51397.33 PMID: 15130029 PLOS ONE | https://doi.org/10.1371/journal.pone.0258085 October 28, 2021 16 / 17 PLOS ONE Urban villagers’ physical activity levels 41. Gu M, Wang Y, Shi Y, Yu J, Xu J, Jia Y, et al. Impact of a group-based intervention program on physical activity and health-related outcomes in worksite settings. BMC Public Health. 2020 Jun 15; 20(1):935. https://doi.org/10.1186/s12889-020-09036-2 PMID: 32539787 42. Zhu W, Chi A, Sun Y. Physical activity among older Chinese adults living in urban and rural areas: A review. Journal of Sport and Health Science. 2016 Sep 1; 5(3):281–6. https://doi.org/10.1016/j.jshs. 2016.07.004 PMID: 30356525 43. Abichahine H, Veenstra G. Inter-categorical intersectionality and leisure-based physical activity in Can- ada. Health Promot Int. 2017 Aug 1; 32(4):691–701. https://doi.org/10.1093/heapro/daw009 PMID: 26976822 44. Herrick SSC, Duncan LR. A Qualitative Exploration of LGBTQ+ and Intersecting Identities Within Physi- cal Activity Contexts. Journal of Sport and Exercise Psychology. 2018 Dec 1; 40(6):325–35. https://doi. org/10.1123/jsep.2018-0090 PMID: 30537884 45. Ray R. An Intersectional Analysis to Explaining a Lack of Physical Activity Among Middle Class Black Women. Sociology Compass. 2014; 8(6):780–91. 46. Xu J, Gao C. Physical activity guidelines for Chinese children and adolescents: The next essential step. J Sport Health Sci. 2018 Jan; 7(1):120–2. https://doi.org/10.1016/j.jshs.2017.07.001 PMID: 30356441 47. Ng SW, Norton EC, Popkin BM. Why have physical activity levels declined among Chinese adults? Findings from the 1991–2006 China health and nutrition surveys. Social Science & Medicine. 2009 Apr 1; 68(7):1305–14. https://doi.org/10.1016/j.socscimed.2009.01.035 PMID: 19232811 48. Clemes SA, O’Connell SE, Edwardson CL. Office Workers’ Objectively Measured Sedentary Behavior and Physical Activity During and Outside Working Hours. Journal of Occupational and Environmental Medicine. 2014 Mar; 56(3):298–303. https://doi.org/10.1097/JOM.0000000000000101 PMID: 24603203 49. Trudeau F, Laurencelle L, Shephard RJ. Tracking of physical activity from childhood to adulthood. Med Sci Sports Exerc. 2004 Nov; 36(11):1937–43. https://doi.org/10.1249/01.mss.0000145525.29140.3b PMID: 15514510 50. Sparling PB, Snow TK. Physical activity patterns in recent college alumni. Res Q Exerc Sport. 2002 Jun; 73(2):200–5. https://doi.org/10.1080/02701367.2002.10609009 PMID: 12092895 51. Althubaiti A. Information bias in health research: definition, pitfalls, and adjustment methods. J Multidis- cip Healthc. 2016 May 4; 9:211–7. https://doi.org/10.2147/JMDH.S104807 PMID: 27217764 PLOS ONE | https://doi.org/10.1371/journal.pone.0258085 October 28, 2021 17 / 17 PLOS ONE
null
10.1371_journal.ppat.1011871.pdf
Data Availability Statement: Raw and processed RNA-sequencing data can be accessed from the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) database under accession number GSE212205.
Raw and processed RNA-sequencing data can be accessed from the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) database under accession number GSE212205.
RESEARCH ARTICLE Exposure to Mycobacterium remodels alveolar macrophages and the early innate response to Mycobacterium tuberculosis infection Dat Mai1, Ana Jahn1, Tara Murray1, Michael Morikubo1, Pamelia N. Lim2,3, Maritza M. Cervantes2, Linh K. Pham2,4, Johannes Nemeth1¤, Kevin Urdahl1, Alan H. Diercks1, Alan Aderem1, Alissa C. RothchildID 2* 1 Center for Global Infectious Disease Research, Seattle Children’s Research Institute, Seattle, Washington, United States of America, 2 Department of Veterinary and Animal Sciences, University of Massachusetts Amherst, Amherst, Massachusetts, United States of America, 3 Molecular and Cellular Biology Graduate Program, University of Massachusetts Amherst, Amherst, Massachusetts, United States of America, 4 Animal Biotechnology and Biomedical Sciences Graduate Program, University of Massachusetts Amherst, Amherst, Massachusetts, United States of America ¤ Current address: University Hospital Zurich, University of Zurich, Division of Infectious Diseases and Hospital Epidemiology, Zu¨rich, Switzerland * arothchild@umass.edu Abstract Alveolar macrophages (AMs) play a critical role during Mycobacterium tuberculosis (Mtb) infec- tion as the first cells in the lung to encounter bacteria. We previously showed that AMs initially respond to Mtb in vivo by mounting a cell-protective, rather than pro-inflammatory response. However, the plasticity of the initial AM response was unknown. Here, we characterize how previous exposure to Mycobacterium, either through subcutaneous vaccination with Mycobac- terium bovis (scBCG) or through a contained Mtb infection (coMtb) that mimics aspects of con- comitant immunity, impacts the initial response by AMs. We find that both scBCG and coMtb accelerate early innate cell activation and recruitment and generate a stronger pro-inflamma- tory response to Mtb in vivo by AMs. Within the lung environment, AMs from scBCG vaccinated mice mount a robust interferon-associated response, while AMs from coMtb mice produce a broader inflammatory response that is not dominated by Interferon Stimulated Genes. Using scRNAseq, we identify changes to the frequency and phenotype of airway-resident macro- phages following Mycobacterium exposure, with enrichment for both interferon-associated and pro-inflammatory populations of AMs. In contrast, minimal changes were found for airway-resi- dent T cells and dendritic cells after exposures. Ex vivo stimulation of AMs with Pam3Cys, LPS and Mtb reveal that scBCG and coMtb exposures generate stronger interferon-associated responses to LPS and Mtb that are cell-intrinsic changes. However, AM profiles that were unique to each exposure modality following Mtb infection in vivo are dependent on the lung environment and do not emerge following ex vivo stimulation. Overall, our studies reveal signifi- cant and durable remodeling of AMs following exposure to Mycobacterium, with evidence for both AM-intrinsic changes and contributions from the altered lung microenvironments. Com- parisons between the scBCG and coMtb models highlight the plasticity of AMs in the airway and opportunities to target their function through vaccination or host-directed therapies. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Mai D, Jahn A, Murray T, Morikubo M, Lim PN, Cervantes MM, et al. (2024) Exposure to Mycobacterium remodels alveolar macrophages and the early innate response to Mycobacterium tuberculosis infection. PLoS Pathog 20(1): e1011871. https://doi.org/10.1371/journal. ppat.1011871 Editor: Padmini Salgame, New Jersey Medical School, UNITED STATES Received: August 3, 2023 Accepted: November 27, 2023 Published: January 18, 2024 Copyright: © 2024 Mai et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: Raw and processed RNA-sequencing data can be accessed from the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) database under accession number GSE212205. Funding: This work was supported by National Institute of Allergy and Infectious Disease of the National Institute of Health under Awards U19AI135976 (A.A.), R01AI032972 (A.A.), 75N93019C00070 (K.U., A.C.R., A.A.), and PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 1 / 28 PLOS PATHOGENS R21AI163809 (A.C.R.). J.N. was supported by the Swiss National Foundation under grant 310030_200407. P.L. was supported by National Research Service Award T32 GM135096 from the National Institutes of Health. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: I have read the journal’s policy and the authors of this manuscript have the following competing interests: J.N. received honoraria for presentations from Oxford Immunotec, Gilead and ViiV. Alveolar macrophage remodeling by Mycobacterium Author summary Tuberculosis, a disease caused by the bacteria Mycobacterium tuberculosis (Mtb), claims around 1.6 million lives each year, making it one of the leading causes of death worldwide by an infectious agent. Based on principles of conventional immunological memory, prior exposure to either Mtb or M. bovis BCG leads to antigen-specific long-lasting changes to the adaptive immune response that can be effective at protecting against subsequent chal- lenge. However, how these exposures may also impact the innate immune response is less understood. Alveolar macrophages are tissue-resident myeloid cells that play an impor- tant role during Mtb infection as innate immune sentinels in the lung and the first host cells to respond to infection. Here, we examined how prior Mycobacterium exposure, either through BCG vaccination or a model of contained Mtb infection, impacts the early innate response by alveolar macrophages. We find that prior exposure remodels the alveo- lar macrophage response to Mtb through both cell-intrinsic changes and signals that depend on the altered lung environment. These findings suggest that the early innate immune response could be targeted through vaccination or host-directed therapy and could complement existing strategies to enhance the host response to Mtb. Introduction Mycobacterium tuberculosis (Mtb), the causative agent of Tuberculosis (TB), claimed more than 1.6 million lives in 2021. For the first time since 2005, the number of TB deaths worldwide is increasing [1,2]. These trends highlight the urgent need for new vaccine and therapeutic strategies. Traditionally, vaccine design has focused on generating a rapid, robust, and effective adaptive immune response. However, recent studies suggest that the innate immune system can undergo long-term changes in the form of trained immunity [3], which affect the outcome of infection and could function as important components of an effective TB vaccine [4,5]. Ini- tial trained immunity studies focused on central trained immunity, long-term changes to hematopoietic stem cells that lead to functional changes in short-lived innate cell compart- ments (i.e., monocytes, NK cells, dendritic cells) [3]. More recent studies have examined innate training in tissue-resident macrophages and demonstrated that these cells are also affected by prior exposures. Tissue-resident macrophages can respond to remote injury and inflammation [6], undergo long-term changes [3], and display altered responses to bacteria after pulmonary viral infection [7–9]. Lung resident alveolar macrophages (AMs) are the first cells to become infected with inhaled Mtb and engage a cell-protective response, mediated by the transcription factor Nrf2, that impedes their ability to effectively control bacterial growth [10,11]. In this study, we exam- ined how prior mycobacterial exposure reprograms AMs and alters the overall innate response in the lung to aerosol challenge with Mtb. To evaluate the range of AM plasticity, we chose to compare the effects of subcutaneous BCG vaccination (scBCG) with those arising from a con- tained Mtb-infection (coMtb) model. BCG, a live-attenuated TB vaccine derived from M. bovis and typically given during infancy, provides protection against disseminated pediatric disease but has lower efficiency against adult pulmonary disease [12–14]. In addition to enhancement of Mtb-specific adaptive responses, based on shared antigens, BCG vaccination also leads to changes in hematopoiesis and epigenetic reprogramming of myeloid cells in the bone marrow [15], early monocyte recruitment and Mtb dissemination [16], and innate acti- vation of dendritic cells critical for T cell priming [17]. Intranasal BCG vaccination protects PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 2 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium against Streptococcus pneumoniae and induces long term activation of AMs [18]. A recent study has shown that one mechanism by which BCG vaccination can elicit innate training effects on AMs, separate from alterations to the monocyte population, is through changes to the gut microbiome and microbial metabolites [19]. BCG vaccination is also associated with trained immunity effects in humans [20–22], including well-described reductions in all-cause neonatal mortality and protection against bladder cancer [3,23]. The coMtb model is generated by intradermal inoculation with virulent Mtb into the ears of mice and leads to a contained but persistent lymph node Mtb infection [24,25]. The model replicates observations in both humans and non-human primates (NHPs) that prior exposure to Mtb infection provides protection against subsequent exposure, through a form of concomi- tant immunity [26,27]. In a previous study, we found that coMtb leads to protection against challenge with aerosol Mtb infection and protects mice against heterologous challenges, including infection with Listeria monocytogenes and expansion of B16 melanoma cells, results which suggest there is substantial remodeling of innate immune responses [25]. We found that AMs from coMtb mice mount a more inflammatory response to Mtb infection compared to AMs from control mice, and the enhancement in AM activation after infection, as measured by MHC II expression, was dependent on IFNγR signaling [25]. Here, we show that while both coMtb and scBCG protect against low dose Mtb aerosol challenge, they remodel the in vivo innate response in different ways. In AMs, scBCG elicits a very strong interferon response in AMs, while coMtb promotes a broader pro-inflammatory response that is less dominated by Interferon Stimulated Genes. Prior exposure to Mycobacte- rium also remodels the frequency and phenotype of AM subsets in the lung prior to aerosol challenge and leads to significant changes in the early dynamics of the overall innate response. While changes in the AM responses that are unique to each exposure (scBCG, coMtb) depend on the lung environment, stronger interferon-associated responses following both LPS and Mtb stimulation ex vivo reveal cell-intrinsic changes. Results Prior exposure to Mycobacterium accelerates activation and innate cell recruitment associated with Mtb control We first determined the earliest stage of infection when the immune response was altered by prior exposure to Mycobacterium. Mice were vaccinated with scBCG or treated with coMtb, rested for 8 weeks, and then challenged with low-dose H37Rv aerosol infection. We measured both the cellularity and activation of innate immune cells in the lung at 10, 12 and 14 days fol- lowing infection, the earliest timepoints when innate cells are known to be recruited [10,11,28]. We observed a significant increase in MHC II Median Fluorescence Intensity (MFI) as early as day 10 for AMs from coMtb mice and day 12 for AMs from scBCG mice compared to controls (Figs 1A and S1). There were no significant differences in MHC II expression prior to challenge on day 0 (Fig 1A). There were also significant increases in the numbers of monocyte-derived macrophages (MDM), neutrophils (PMN), dendritic cells, and Ly6C+ CD11b+ monocytes by day 10 in coMtb mice compared to controls, with further increases by days 12 and 14 (Figs 1B and S1). scBCG elicited similar increases in these popula- tions starting at day 10, but the increases were not as robust or rapid as those observed in coMtb. Significant differences between scBCG and coMtb groups were found at days 10, 12, and 14 in MDM, day 14 in PMN, days 12 and 14 in dendritic cells, and day 14 in Ly6C+ CD11b+ monocytes (Fig 1B). While there were not significant differences in AM cell number between the three conditions, there was a modest drop in viability for both AMs from scBCG and coMtb mice by day 14 (Fig 1C). PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 3 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium Fig 1. Prior exposure to Mycobacterium leads to faster activation and innate cell recruitment following aerosol Mtb challenge. Control, scBCG, and coMtb mice, 8 weeks following exposure, challenged with standard low-dose H37Rv. Lungs collected on day 10, 12, and 14 post-infection. A) AM MHC II MFI. B) Total numbers of MDMs, PMN, DC, and Ly6C+CD11b+ monocytes. C) AM viability (% Zombie Violet-). D) Total numbers of CD44+ CD4+ T PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 4 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium cells, ESAT6-tetramer+ CD4+ T cells, CD44+ CD8+ T cells, and TB10.4-tetramer+ CD8+ T cells. Mean +/- SEM, 5 mice per group, representative of 3 independent experiments. One-way ANOVA with Tukey post-test. * p< 0.05, **p< 0.01, ***p < 0.001. B, C) *, **, and *** scBCG or coMtb vs control; +, + + scBCG vs coMtb. https://doi.org/10.1371/journal.ppat.1011871.g001 In addition to early changes in innate cell activation and recruitment, we observed early recruitment of activated CD44+ CD4+ and CD8+ T cells in the lungs of both coMtb and scBCG mice starting at day 10 as well as TB antigen-specific T cells, ESAT6-tetramer+ CD4+ T cells and TB10.4-tetramer+ CD8+ T cells in coMtb mice starting at day 10 compared to con- trols and scBCG mice (Figs 1D and S1). The differences in the recruitment of ESAT6-tetra- mer+ CD4+ T cells between scBCG and coMtb were expected, as the ESAT6 antigen is expressed by H37Rv but not by BCG. We also evaluated whether these cell recruitment differences correlated with changes in bacterial burden. To compile CFU results from three independent experiments, each with slightly different bacterial growth (S2A Fig), we calculated a ΔCFU value that compared the bacterial burden of each sample to the average for the respective control based on timepoint, organ, and experiment. We found that both modalities generated a significant reduction in bacterial burden compared to controls in the lung, spleen, and lung-draining lymph node (LN) at day 14 and at day 28, as previously reported [16,25,29] (S2A–S2D Fig). At day 10, we observed no difference in lung bacterial burden in scBCG or coMtb mice compared to controls and a small increase in coMtb mice over scBCG. The majority of control mice had undetect- able bacteria in spleen and LN at this time. There was a significant reduction in bacterial bur- den in the lung by day 12 in coMtb but not scBCG mice and a significant reduction in CFU in the LN in both models compared to controls (S2B Fig). Our results demonstrate that prior Mycobacterium exposure leads to accelerated innate cell activation and recruitment, alongside an increase in activated T cells, within the first two weeks of infection, with coMtb generating a faster and more robust response compared to scBCG. These early immune changes are asso- ciated with reductions in bacterial load in the lung. Differences in bacterial burden in the LN and spleen suggest delays in bacterial dissemination, which first appear in the LN at day 12 and then in the spleen at day 14 (S2A Fig). Mycobacterium exposure alters the in vivo alveolar macrophage response to Mtb infection To examine the earliest response to Mtb, we measured the gene expression profiles of Mtb- infected AMs isolated by bronchoalveolar lavage and cell sorting, as previously described [10], 24 hours following aerosol challenge with high dose mEmerald-H37Rv (depositions: 4667, 4800) in scBCG-vaccinated mice and compared these measurements to previously generated profiles of AMs from control (unexposed) mice [10] and coMtb mice [25] (S1 Table). As pre- viously observed for the high dose infection, an average of 1.79% (range: 0.91–3.18%) of total isolated AMs were Mtb infected 24 hours after infection. Changes induced by Mtb infection were measured by comparing gene expression between Mtb-infected AMs and respective naïve AMs for each of the three groups (control, scBCG, coMtb). Principal Component Analy- sis on Mtb infection-induced changes showed that each of the three conditions led to distinct expression changes (Fig 2A) and the majority of up-regulated Differentially Expressed Genes (DEG) (fold change > 2, FDR < 0.05) were unique to each condition (control: 151 unique/257 total DEG, scBCG: 222/289, coMtb: 156/229) (Fig 2B). The divergence in the responses of Mtb-infected AMs from each of the 3 conditions was also reflected in the diversity in the Top 20 Canonical Pathways identified by Ingenuity Pathway Analysis (S3 Fig). PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 5 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium Fig 2. Mycobacterium exposure alters the alveolar macrophage transcriptional response to Mtb infection in vivo. Bulk RNA-seq profiles of Mtb-infected AMs 24 hours following high-dose mEmerald-H37Rv infection. Gene expression changes are compared to respective naïve samples: Mtb-inf control vs naïve control; Mtb-inf scBCG vs naïve scBCG; Mtb-inf coMtb vs naïve coMtb (controls- reported in Rothchild et al, 2019 [10]; coMtb- reported in Nemeth et al, 2020 [25]). A) Principal Component Analysis using DEG (|fold change| > 2, FDR< 0.05) in Mtb-infected AMs compared to respective naïve AMs (control, scBCG, or coMtb). B) Venn Diagram and Intersection plot of overlap in up-regulated DEG between the 3 conditions. C) Gene Set Enrichment Analysis of 50 Hallmark Pathways. Pathways shown have |NES| > 1.5 and FDR< 0.05 for at least one of the conditions. * FDR< 0.05, **FDR< 0.01, ***FDR< 0.001. D) Heatmap of 131 original in vivo DEG at 24 hours in Mtb-infected AM (left), Interferon Stimulated Genes, derived from macrophage response to IFNα (fold change >2, p- value < 0.01) Mostafavi et al, 2016 [30] (center-left), IL6 JAK STAT3 hallmark pathway (center-right) and selected coMtb signature genes (right, *FDR< 0.05, PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 6 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium FC> 2). E) Scatterplots depicting fold change (log2) for Mtb-infected AMs over naïve AMs for scBCG versus coMtb. Highlighted pathways: Nrf2-associated genes out of 131 original in vivo DEG (56 genes, purple), shared leading edge genes for scBCG Interferon Alpha Response and Interferon Gamma Response pathways (61 genes, orange), and leading edge genes for coMtb IL6 JAK STAT3 pathway (23 genes, green). Compiled from 4 independent experiments per condition for control, 2 independent experiments per condition for scBCG and coMtb. https://doi.org/10.1371/journal.ppat.1011871.g002 To identify trends between groups, we performed Gene Set Enrichment Analysis using a set of 50 Hallmark Pathways. As we’ve shown previously, Mtb-infected AMs from control mice at 24 hours had strong enrichment for “Xenobiotic Metabolism” and “Reactive Oxygen Species” pathways, indicative of the Nrf2-associated cell-protective response (Fig 2C). While these two pathways were not among the most enriched pathways in the exposed groups, Mtb-infected AMs from all groups upregulated genes associated with the 131 in vivo DEG that make up the cell-protective Nrf2-driven response at 24 hours [10] (Fig 2D). Expression profiles for Mtb- infected AMs from scBCG mice showed the strongest enrichment for “Interferon Alpha Response” and “Interferon Gamma Response” pathways, which contain many shared genes (Fig 2C). The strength of the interferon response was further highlighted by examining gene expression changes in a set of Interferon Stimulated Genes (ISGs) identified from macro- phages responding to IFNα (fold change > 2, p-value < 0.01) [30] (Fig 2D). Expression pro- files for Mtb-infected AMs from coMtb mice showed a weaker enrichment for interferon response pathways with fewer up-regulated ISGs compared to scBCG, and instead showed enrichment across a number of inflammatory pathways including “IL6 JAK STAT3 signaling” in comparison to the other groups (Fig 2C and 2D). A direct comparison between the gene expression patterns for AMs from scBCG versus coMtb mice could be visualized more readily by scatterplots highlighting either Nrf2-associated, Interferon Alpha and Gamma Response, or IL-6 JAK STAT3 pathway genes (Fig 2E). In summary, Mycobacterium exposures alter the initial in vivo response of AMs to Mtb infection 24 hours after challenge and remodel the AM response in distinct ways. AMs from scBCG vaccinated animals mount a strong interferon-associated response, while AMs from coMtb mice express a more diverse inflammatory profile consisting of both interferon-associ- ated genes as well as other pro-inflammatory genes, including those within the IL-6 JAK STAT3 pathway. Mycobacterium exposure modifies the baseline phenotype of alveolar macrophages in the airway Although scBCG and coMtb exposures alter the AM responses to Mtb infection in vivo, tran- scriptional effects are not widely evident prior to infection as measured by bulk RNA-sequenc- ing of naïve AMs from control, scBCG, or coMtb mice, including expression of innate receptors and adaptors (S4 Fig). However, we posited that remodeling effects were likely not homogenous across the entire AM population and that small heterogenous changes to baseline profiles might be detectable using a single cell approach. We therefore analyzed pooled BAL samples taken from 10 age- and sex-matched mice from each of the three conditions (control, scBCG, coMtb) eight weeks following Mycobacterium exposure by single cell RNA-sequencing (scRNAseq). Gross cellularity was unaffected by mycobacterial exposure as measured by flow cytometry analysis of common lineage markers with AMs being the dominant hematopoietic cell type (57.4–85.8% of CD45+ live cells), followed by lymphocytes (5.26–22.7% of CD45+ live cells) with smaller contributions from other innate cell populations (S5 Fig). Six samples, with an average of 2,709 cells per sample (range: 2,117–4,232), were analyzed together for a total of 17,788 genes detected. The most prominent expression cluster mapped to an AM profile, with smaller clusters mapping to T and B lymphocytes, dendritic cells, and PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 7 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium neutrophils (Fig 3A). All cells that mapped to a macrophage profile were extracted and reclus- tered into 11 macrophage subclusters (Fig 3B and 3C). All but two of the macrophage subclus- ters (clusters 6 and 8) expressed AM lineage markers (Siglecf, Mertk, Fcgr1 (CD64), Lyz2 (LysM), and Itgax (CD11c) and had low expression of Itgam (CD11b) (Fig 3D). Cluster 6 showed high Itgam and Lyz2 expression and lower Siglecf expression, likely representing a small monocyte-derived macrophage population in the airway, while cluster 8 displayed high Lyz2 expression, low expression for other AM markers, and expression of Sftpa1 and Wfdc2 (S2 Table), genes most commonly expressed by pulmonary epithelial cells, suggesting that this cluster represents a small population of epithelial cells, To interpret the various expression subclusters, we identified the genes that most distin- guished each cluster from the others (S6 Fig and S2 Table). As has been reported by other groups [31,32], a small proportion of the AMs in two clusters (Clusters 4, 9) had high expres- sion of cell cycle genes (i.e., Top2a, Mki67), indicative of cell proliferation (Fig 3E and S2 Table). Cluster 0 was the most abundant macrophage cluster with high expression of lipid metabolism genes (i.e., Abcg1, Fabp1) (Fig 3F and S2 Table). Cluster 2 was significantly increased in relative frequency for scBCG samples compared to coMtb (p = 0.032, One-way ANOVA with Tukey post-test) and associated with oxidative stress response genes (Hmox1, Gclm). Several Cluster 2 associated genes, Slc7a11, Hmox1, and Sqstm1 also had higher overall expression level in scBCG samples compared to either control or coMtb (Fig 3G and S2 Table). Cluster 7 was the only cluster with an increase in relative frequency trending for both scBCG and coMtb (p = 0.076, One-way ANOVA). Cells in this cluster had high expression of Interferon Stimulated Genes (Ifit1, Isg15) and within this cluster, cells from scBCG samples had higher expression of Axl and Ifi204 than cells from coMtb samples. (Fig 3H and S2 Table). Cluster 3 had significantly higher relative frequency for coMtb samples compared to control and scBCG samples (p = 0.021, 0.039, respectively, One-way ANOVA with Tukey post- test) and was distinguished by expression of macrophage-associated transcription factors (Cebpb, Zeb2, Bhlhe40) [33,34], mitochondrial oxidative phosphorylation (mt-Co1, mt-Cytb, mt-Nd2), chromatin remodeling (Ankrd11, Baz1a), and immune signaling including the CARD9 complex (Malt1, Bcl10, Prkcd) (Figs 3I and S7 and S2 Table). This expression profile closely matches a subcluster of AMs previously described by Pisu et al, as an “interstitial mac- rophage-like” AM population (labeled “AM_2”) that expanded in relative frequency in lung samples 3 weeks following low-dose H37Rv infection [31]. Relative expression level for Cebpb, Mt-Cyb, and Lars2 within Cluster 3 was higher for cells from coMtb samples compared to either control or scBCG samples. Interestingly, Cluster 2 (higher relative frequency in scBCG) and Cluster 3 (higher relative frequency in coMtb) represent divergent endpoints of a pseudotime plot generated by a trajec- tory inference analysis, regardless of whether the starting point is the most abundant cluster in the control group (Cluster 0) (Fig 3J, top) or the cluster of proliferating cells (Cluster 4) (Fig 3J, bottom). This result suggests that scBCG and coMtb may drive AM phenotypes in diver- gent directions and indicates that AM responses can be remodeled into more than one flavor, rather than only a binary “on/off” state. To further investigate whether a sub-cluster of AMs might be responsible for the increased enrichment for Interferon Alpha/Gamma Response pathways in the in vivo Mtb response in scBCG and coMtb mice, we scored each cluster based on the ISG gene module, previously used in Fig 2D. As expected, we observed that only Cluster 7 showed strong enrichment for ISGs, which trended up in frequency for both scBCG and coMtb samples (Fig 3K). To investigate potential reprogramming of non-AM macrophages, we examined Cluster 6, the macrophage cluster with low Siglecf and high Itgam expression that is consistent with a monocyte-derived macrophage population. We observed no statistically significant differences PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 8 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium Fig 3. Mycobacterium exposure modifies the alveolar macrophage phenotype in the airway pre-challenge. Single-cell RNA-sequencing of BAL samples from control, scBCG, and coMtb mice pre-aerosol challenge. A) Compiled scRNAseq data for all BAL samples, highlighted by major clusters, annotated based on closest Immgen sample match. B) Highlighting of the two clusters used for macrophage subcluster analysis. C) The 11 clusters generated by the macrophage subcluster analysis, separated by condition. D) Expression of major macrophage-specific markers: Siglecf, Mertk, Fcgr1, Lyz2, Itgam (CD11b), and Itgax (CD11c). E-I) Relative frequency of each macrophage subcluster by condition. (violin plots by cluster) Expression level of representative genes distinguished by that cluster compared to other clusters. One-way PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 9 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium ANOVA with Tukey post-test, * p< 0.05. (3-way violin plots by condition) Differentially expressed genes within Clusters 2, 7, and 3 between control vs scBCG vs coMtb samples. Wilcoxon Rank Sum Test, Bonferroni adjusted p-value. *adj-p< 0.05, **adj-p< 0.01, ***adj-p< 0.001. J) Pseudotime analysis (Monocle3) with starting node at the largest cluster in control, Cluster 0 (top) and at the cluster of proliferating cells, Cluster 4,9 (bottom). K) ISG Module Score by cluster. Module derived from macrophage response to IFNα (fold change> 2, p-value< 0.01) (Mostafavi et al, 2016) [30]. Data is compiled from two independent experiments (circle, triangle) with 3 conditions each for a total of 6 samples. https://doi.org/10.1371/journal.ppat.1011871.g003 in the relative size of this cluster between each of the three conditions (S8A Fig). However, there were a number of Differentially Expressed Genes (DEGs) between the groups, including decreases in expression of CD11b (Itgam) and Macrophage scavenger receptor (Msr1) for scBCG and coMtb macrophages compared to controls, increases in MHC-related genes (H2-Aa, Cd74) and iron-metabolism associated genes (Cd63, Fth1, Ftl1) for coMtb macro- phages compared to controls (S8B Fig). A previous study found IV BCG induced chromatin accessibility changes in AMs and IMs for some of these genes [31]. Additionally, we compared baseline changes to AMs following scBCG and coMtb expo- sures to AM changes following ivBCG vaccination (S9 Fig). Overall, we found that ivBCG exposure led to similar changes in AM populations to that of scBCG vaccination, with increased frequency of AMs clustering to “oxidative stress response” and “interferon stimu- lated genes (ISGs)” (S9C Fig). These baseline changes by scRNAseq mirror what is observed for the response of Mtb-infected AMs from ivBCG mice 24 hours after infection by bulk RNA- seq (S9A and S9B Fig). Profiles of Mtb-infected AMs from ivBCG vaccinated mice most closely match those of Mtb-infected AMs from scBCG vaccinated mice, with robust up-regula- tion of Interferon Stimulated Genes. These results demonstrate that both SC and IV BCG vac- cination lead to similar remodeling of AMs, with profiles distinct from that of coMtb exposure. In summary, scRNAseq analysis of macrophages isolated by BAL demonstrate that Myco- bacterium exposure leads to subtle changes in a small minority of AM subsets in the airway, including ones associated with interferon responses and an interstitial macrophage phenotype, while leaving the most abundant subsets of AMs unchanged in frequency or gene expression. We hypothesize that these small changes in baseline profiles may be sufficient to drive the more substantial changes observed in the AM Mtb response in vivo, as described in Fig 2. Mycobacterium exposure has minimal impact on T cell populations in the airway While AMs are the dominant immune cell type in the airway, other cell populations make up an average of 18.4% of the cells within the BAL in controls (range: 10.4–26.3%) and 31.3% in exposed groups (range: 14.0–48.8%). To examine how Mycobacterium exposure influenced other cells in the airway, we focused on T cells and dendritic cells (DCs) which have two of the highest relative frequencies after AMs (Fig 4A and 4B). T cells and DCs were each combined from two original clusters each. Neither population showed a statistically significant difference in relative frequency (Fig 4B). To examine qualitative changes in the T cell population in greater detail, we next reclustered the T cells, resulting in 7 T cell clusters. We manually anno- tated each of the clusters based on the most closely matched Immgen profiles and the expres- sion of key lineage specific markers (Figs 4A–4C and S10). We focused on the 5 most abundant T cell subclusters (Clusters 0–4). While we observed subtle shifts in the relative fre- quency of each group, none reached statistical significance. Cluster 0, the most abundant clus- ter, had an expression profile most consistent with γδ T cells, including expression of Cd3e with low to nil Cd4 and Cd8a and some expression of Zbtb16 (PLZF) and Tmem176a, an ion channel regulated by RORγt and reported to be expressed by lung γδ T cells [35,36] (Figs 4D– PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 10 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium Fig 4. Airway T cell and dendritic cell profiles following Mycobacterium exposure. Single-cell RNA-sequencing of BAL samples from control, scBCG, and coMtb mice pre-aerosol challenge. A) Compiled scRNAseq data for all BAL samples, with T cell and dendritic cell clusters highlighted. B) Relative frequency of T cells and DCs. C-F) T cell subcluster analysis. C) T cell subclusters compiled and split by condition. Annotations made following Immgen profile matches and manual marker inspection. D) Relative frequency of Clusters 0–4 for each condition. E) UMAP gene expression plot for general T cell markers. F) UMAP gene expression plot cluster-specific markers split by condition. G-J) Dendritic cell subcluster analysis. G) Dendritic cell subcluster, colored by each of 3 different clusters. H) Relative frequency of Clusters 0–2 for each condition. I) Violin plots for cluster-specific markers and genes of interest. J) Differentially expressed genes in Cluster 0 split by condition. *adj-p< 0.05, **adj-p< 0 .01, ***adj-p< 0.001, Wilcoxon Rank Sum Test, Bonferroni adjusted p-values. Data is compiled from two independent experiments with 3 conditions each for a total of 6 samples. https://doi.org/10.1371/journal.ppat.1011871.g004 4F and S10). Cluster 1 matched a profile for effector CD4+ T cells (Figs 4D–4F and S10), and Cluster 2 matched a profile for naïve CD8+ T cells (Figs 4D–4F and S10). Cluster 3 had a pro- file consistent with effector memory/resident memory CD8+ T cells (TEM/RM) (Figs 4D–4F and S10) and Cluster 4 had a profile consistent with NK cells. Overall, there were no PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 11 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium significant changes in the relative frequency of T cell or NK subclusters, despite detection of a number of different lymphocyte subsets in the airway. Mycobacterium exposure modifies the dendritic cell airway landscape Re-clustering of DCs yielded 2 major clusters (Cluster 0, 1) and 1 minor cluster (Cluster 2), which had a mixed phenotype with expression of genes from both major clusters (Fig 4G). Cells in Cluster 0 had high expression of Clec9a, Itgae (CD103), and MHC II genes (H2-Ab1, H2-DMa) consistent with an expression profile of lung CD103+ cDCs [37] (Fig 4I), while cells in Cluster 1 had higher expression of Batf3, Ccr7, and Fscn1. All three of the clusters had high Irf8 expression and lower expression of Xcr1, Irf4, and Itgam (CD11b) (Fig 4I). While the coMtb samples trended higher in relative frequency for Cluster 0 and low for Cluster 1, com- pared to the control or scBCG samples, these differences did not meet statistical significance (One-way ANOVA with Tukey post-test, p = 0.16, p = 0.11) (Fig 4H). This was likely due to the limit in statistical power with only 2 replicates. However, it was notable that for cells within Cluster 0, there was a significantly higher expression level for MHC II genes (H2-Aa, H2-DMb1, and Cd74) for coMtb cells compared to control or scBCG cells (Fig 4J). This sug- gests that coMtb might be able to elicit more mature or activated DCs in the airway. Overall, scRNAseq analysis shows that Mycobacterium exposure leads to minimal changes in T cell and dendritic cell populations in the airway, although we hypothesize that small changes in DC maturation/activation could have important impacts on adaptive immune priming dynamics after aerosol infection. Cell-intrinsic remodeling of alveolar macrophages following Mycobacterium exposure licenses an interferon response in vitro Analysis of the AM response to Mtb in vivo demonstrates that the very earliest immune response to Mtb is altered by previous Mycobacterium exposure. However, one limitation to this approach is the inability to discern whether changes to AMs are cell-intrinsic or dependent on the altered tissue environment, especially the presence of Mtb-specific T cells. Therefore, to determine whether Mycobacterium exposure induces cell-intrinsic changes to AMs, we iso- lated AMs from control, scBCG, and coMtb mice, stimulated them ex vivo with LPS, Pam3Cys, or H37Rv, and measured their transcriptional profiles 6 hours later (Fig 5A). First, PAMP-spe- cific trends were notable. AMs from coMtb and scBCG mice showed distinct responses com- pared to AMs from control mice following LPS and H37Rv stimulation, but only minimal changes following Pam3Cys stimulation(Fig 5B and S3 Table). No obvious changes in innate receptor or adaptor expression explain the PAMP-specific differences (S11 Fig). Second, as we have previously reported, Mtb-infected AMs did not strongly up-regulate Nrf2-associated genes ex vivo (Fig 5C). Third, when we examined the gene sets that distinguished the in vivo AM response between scBCG and coMtb mice, “Interferon Alpha/Gamma Response” and “IL6 JAK STAT3” (Fig 2E), we found that the differences between exposure modalities were diminished ex vivo, suggesting contribution of the lung environment to the quality of the response (Fig 5C). Using Gene Set Enrichment Analysis, we identified “Interferon Gamma Response”, “Interferon Alpha Response”, “TNFa signaling via NF-kB”, and”Inflammatory Response” pathways as the most strongly enriched for LPS and H37Rv responses from scBCG and coMtb AMs (Fig 5D). To assess whether the cell-intrinsic changes observed were long- lasting, we compared the responses of AMs at 8 or 23 weeks following scBCG vaccination by RT-qPCR. Increases in gene expression were as robust or even enhanced 23 weeks following exposure compared to 8 weeks, suggesting that exposure-induced changes to AMs are rela- tively long-lived (S12 Fig). To validate whether changes in gene expression were reflected at PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 12 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium Fig 5. Cell-intrinsic remodeling of alveolar macrophages following Mycobacterium exposure. A) AM isolation 8 weeks following scBCG or coMtb exposure. AMs were stimulated with Pam3Cys (10 ng/ml), LPS (10 ng/ml), and H37Rv (effective MOI ~2:1) for 6 hours (RNA-seq) or 20 hours (flow cytometry). B-D) Gene expression changes measured by bulk RNA-seq for stimulated AMs compared to respective unstimulated AMs (i.e., LPS-stim control AM vs unstim control AM; LPS-stim scBCG AM vs unstim scBCG AM; LPS-stim coMtb AM vs unstim coMtb AM). B) Scatterplots, log2 fold change gene expression for stimulated to unstimulated AMs for each condition (control, scBCG, coMtb). Differentially expressed genes (DEG) are highlighted for one or both conditions (|Fold change| > 2, FDR< 0.05 for Pam3Cys and LPS; |Fold change| > 2, FDR< 0.2 for H37Rv). C) Scatterplots, log2 fold change gene expression for H37Rv-stimulated to unstimulated scBCG versus coMtb AMs. Genes highlighted derived from gene sets in Fig 2E. Nrf2-associated genes (56 genes, purple), Interferon Alpha/ Gamma Response (61 genes, orange), and IL6 JAK STAT3 (23 genes, green). D) Gene Set Enrichment Analysis results for 50 HALLMARK pathways. Pathways shown have NES> 1.5 and FDR< 0.05 for at least one of the conditions. *FDR< 0.05, **FDR< 0.01, ***FDR< 0.001. E) Gating strategy and MHC II and TNF histograms for coMtb AMs, no stimulation versus LPS. F-G) MHC II and TNF MFI in control, scBCG, and coMtb AMs after 20 hours of LPS stimulation. *p< 0.05, **p< 0.01, ***p< 0.001, One-way ANOVA with Tukey post-test. https://doi.org/10.1371/journal.ppat.1011871.g005 the protein level, we sought to develop a flow cytometry-based assay to assess AM-specific responses. Primary AMs were stimulated with LPS for 20 hours and both MHC II and TNF expression were measured by flow cytometry (Fig 5E). We found that AM from coMtb mice PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 13 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium had significantly higher MHC II expression than controls and a similar pattern was seen for scBCG AM in 1 of 2 experiments (Fig 5F). AMs from coMtb mice also showed a significant increase in TNF expression in 1 of 2 experiments (Fig 5G). Because the “Interferon Alpha Response” and “Interferon Gamma Response” pathways were most highly enriched for the H37Rv stimulation following Mycobacterium exposure, we decided to further investigate the Interferon-associated response [30]. We specifically sought out a dataset that would identify ISGs specific to Mtb-infected macrophages. To generate an IFNγ-derived signature, we would need a macrophage-T cell co-culture system and to sort out the Mtb-infected macrophages, because murine macrophages do not produce IFNγ during Mtb infection in vitro. Therefore, we decided to examine an IFNα/β-derived signature from a data set of Mtb-infected IFNAR-/- bone marrow derived macrophages (BMDMs). We catego- rized the macrophage response to H37Rv stimulation as “IFN-dependent” or “IFN-indepen- dent” based on gene expression of WT versus IFNAR-/- BMDMs following H37Rv infection (see methods section) (S4 Table) [38]. Expression of IFN-dependent genes was minimally induced in control AMs but strongly up-regulated in AMs from Mycobacterium exposed mice, as measured by the GSEA normalized enrichment score (NES) (Fig 6A, left). In contrast, expression of IFN-independent genes was modestly upregulated in control AMs and only slightly altered by Mycobacterium exposure (Fig 6A, right). When we applied these two gene sets to the in vivo response profiles described in Fig 2 generated for Mtb-infected AMs follow- ing high dose infection with mEmerald-H37Rv, we observe that Mtb-infected AMs from scBCG mice up-regulate the IFN-dependent response in vivo, suggesting that the licensing of the IFN-dependent response plays a role in vivo following BCG vaccination (Fig 6B). The dif- ference between the in vitro and in vivo response for AMs from coMtb mice points to an addi- tional contribution of the lung environment. These results demonstrate that prior Mycobacterium exposure leads to cell-intrinsic changes in AMs that license an enhancement of IFN-dependent responses to Mtb that are retained in vitro, while qualitative differences in the response between scBCG and coMtb in vivo are dependent on signals from the lung environment. Discussion Here we describe remodeling of AMs, long-lived airway-resident innate cells, following two modalities of Mycobacterium exposure, scBCG vaccination and coMtb, a model of contained Mtb infection. AMs are the first cells to be productively infected in the lung following aerosol Mtb infection [10,11]. We previously showed that AMs initially respond to Mtb infection with a cell-protective, Nrf2-driven program that is detrimental to early host control [10], suggesting that the lack of a robust response by AMs prevents effective host control early on. In line with this model, others have shown that depletion of AMs or strategies that “bypass” AMs including directly injecting antigen-primed DCs or activating DCs accelerate the immune response and reduce bacterial burden [17,39,40]. However, how vaccination or prior exposures impact the initial response of AMs and whether there are therapeutic strategies that would enhance their initial response to infection have not been well studied [41]. Most studies examining the impacts of prior exposure to either Mtb or other species of mycobacteria, including BCG vaccination, have focused on the durable antigen-specific changes to the adaptive immune response. In contrast, we focused on changes to tissue-resi- dent innate cells and their responses at the earliest stages of infection (� 14 days). Along with early changes to the T cell response, both scBCG and coMtb accelerate innate cell activation and immune cell recruitment in the first 10–14 days following Mtb aerosol infection, and even the very initial AM response to Mtb, within the first 24 hours of infection, is remodeled PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 14 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium Fig 6. Mycobacterium exposure licenses an interferon-dependent response to H37Rv by alveolar macrophages. Gene expression changes measured by bulk RNA-seq for Mtb-infected AMs compared to respective unstimulated AMs (i.e., Mtb-inf control AM vs unstim control AM; Mtb-inf scBCG AM vs unstim scBCG AM; Mtb-inf coMtb AM vs unstim coMtb AM). A) Gene expression for control, scBCG, and coMtb AMs, 6 hour H37Rv infection ex vivo, log2 fold change (Mtb-infected/uninfected). IFN-dependent genes (339 total) and IFN-independent genes (352 total) based on WT vs IFNAR-/- BMDM bulk RNA-seq dataset (Olson et al, 2021) (see Methods section). B) Gene expression for control, scBCG, and coMtb AMs, 24 hour in vivo H37Rv infection, Mtb-infected sorted, log2 fold change (Mtb-infected/uninfected) for the same IFN-dependent and IFN-independent gene sets in (A). Grey bars indicate N.D. Normalized Enrichment Score (NES) calculated by GSEA for two data sets alongside Hallmark Pathways. +FDR< 0.05, ++FDR< 0.01, +++FDR< 0.001. https://doi.org/10.1371/journal.ppat.1011871.g006 following exposure to Mycobacterium. The durable changes observed fit with a number of recent studies which have uncovered either enhanced AM antimicrobial phenotypes [7–9] or impaired responses [42,43] following viral infection. Other studies have identified long-lasting changes to AMs following intranasal immunization of either adenoviral-based or inactivated whole cell vaccines [18,44,45]. We observe that the most robust cell-intrinsic changes to AM responses following scBCG or coMtb are found in IFN-dependent genes (Fig 6) and ISGs (Fig 2D), suggesting a critical role for interferon signaling in the changes to the early innate response in the lung during infection. Notably, this finding is not limited to the murine model. BAL from NHPs following IV, ID, or aerosol BCG vaccination similarly show AMs enriched for Interferon Gamma Response pathway genes [46]. AMs can respond to both Type I (IFNα/β) and Type II Interfer- ons (IFNγ) and it is not possible to distinguish between responses to IFNα/β and IFNγ based on transcriptional analysis alone. The presence of live bacteria within both scBCG and coMtb models limits system-wide perturbations, such as T cell depletion or anti-IFNγ blockade, which would reverse containment [24]. For this reason, we have not been able to directly test how interferon signals derived from scBCG or coMtb remodel AMs in a cell-autonomous manner, but we envision future studies to examine the specific effects of individual cytokines on AM remodeling. PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 15 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium Even though IFNAR-/- macrophages were used to generate the ISG signature identified in Fig 6, IFNγ is the more likely candidate to contribute to AM remodeling following Mycobacte- rium exposure. IFNγ is required for the generation of trained immunity in bone marrow- derived myeloid cells following IV BCG vaccination [15,47]. While IV and aerosol H37Rv infection was found to induce Type I IFNs and reduce myelopoiesis [47], we previously found that coMtb, in which Mtb is contained within the ear-draining lymph node, leads to low-level systemic cytokinemia, including IFNγ production. Using WT:Ifngr1-/- mixed bone marrow chimeras, we showed that IFNγ signaling was responsible for monocyte and AM activation fol- lowing establishment of coMtb [25]. Additionally, several reports have identified T cell-derived IFNγ as critical for altering AM function, although the immunological outcome varies sub- stantially based on the context. In one study, T cell-derived IFNγ following adenoviral infec- tion leads to AM activation, innate training and protection from S. pneumoniae [8], while in another study influenza-induced T cell-derived IFNγ leads to AM dysfunction and impaired clearance of S. pneumoniae [43]. A study of 88 SARS-CoV-2 patients identified AMs and T cell-derived IFNγ as part of a positive feedback loop in the airway [48]. In contrast, type I IFN signatures are associated with active TB or TB disease progression in both humans and non- human primates [49–51]. Host perturbations such as treatment with poly I:C or viral co-infec- tion that induce type I IFN lead to worsened disease [52,53], type I IFN has been shown to block production of IL-1β in myeloid cells during Mtb infection [54], and type I IFN drives mitochondrial stress and metabolic dysfunction in Mtb infected macrophages [38]. We note that the two modalities tested here consist of different mycobacterial species, dif- ferent doses, and different routes. We expect that all three of these factors likely contribute to the quality of AM remodeling. For example, they could be important for the location, timing, and amount of IFNγ that AMs are exposed to. While an in-depth examination of each of these factors is beyond the scope of this study, the side-by-side comparison of the two different exposure models, scBCG and coMtb, allows us to examine the plasticity of AM phenotypes and the impact of the local and/or systemic environments leading to different responses. It is notable that scBCG is quickly cleared from WT mice, while coMtb replication is sustained in the superficial cervical lymph node for up to a year or longer [25]. Protection from H37Rv challenge mediated by coMtb is abrogated following antibiotic treatment but not completely lost [25]. This suggests that there may be different contributions to AM remodeling from active bacterial replication and from long-term microenvironment changes following bacterial clearance, which will be addressed in follow-up studies. In particular, it is notable that the modality-specific signatures identified through in vivo transcriptional analysis disappeared following ex vivo isolation, along with the Nrf2 signature. The difference between in vivo and ex vivo signatures suggests a critical contribution of the altered lung microenvironments in AM remodeling, which deserves additional follow-up studies. One additional limitation of our approach is that the ex vivo samples were collected in bulk, in the absence of cell sorting, and so, unlike the in vivo studies, a very small number of bystander AMs were likely collected alongside the Mtb-infected AMs, which could have had minor impacts on the transcriptional signatures. The fact that AMs can be remodeled into more than one state suggests additional complexity in innate immune features that has not yet been fully explored. Heterogeneity in myeloid reprogramming is not limited to the murine model and has also been observed in human monocytes [55]. Several studies have recently described innate-adaptive interactions within the airway that are thought to impact infection dynamics [46,48]. We note that in these models we observe innate cell activation and recruitment occurring at the same time as T cell activation and recruitment, and that these events are likely promoting one another. We are particularly PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 16 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium intrigued by the changes in AM MHC class II expression that we observed in vivo during the first two weeks of infection (Fig 1A) and following ex vivo stimulation (Fig 5F). AMs are considered to be poor antigen presenters, relative to other myeloid subsets, yet the faster Mtb-specific T cells are recruited to the lung, the more likely it is that AMs will serve as pri- mary T cell targets [37,56–60]. Our results suggest that enhancement of AM antigen presenta- tion could be one innate mechanism that could be targeted to complement and synergize with the adaptive immune response during infection. Other potential mechanisms by which AM remodeling may contribute to enhanced bacterial control after Mtb aerosol challenge include enhanced phagocytic activity or increased direct antimycobacterial activity, as previously dem- onstrated by Jeyanathan et al [19]. Future studies are needed to further interrogate the contri- bution of these innate mechanisms. There are many other remaining questions. While we identify both cell-intrinsic changes and changes dependent on the lung environment, we do not yet know whether the cell-intrin- sic changes are retained long-term in the absence of environmental cues. We do not know the durability of the changes, both cell-intrinsic and environment-dependent, and whether they are mediated by epigenetic effects. Our longest experiment showed retention of cell-intrinsic changes to AMs after 23 weeks. In Nemeth et al, we showed that antibiotic treatment lessened the protection mediated by coMtb, suggesting that ongoing replication is a key part of host protection [25]. AM remodeling is retained 8 weeks or longer after the initial exposures, a timepoint when there is little to no detectable mycobacteria in the lung, ruling out a require- ment for local ongoing bacterial replication in AM remodeling, although systemic signals derived from remote bacterial replication may still play a role. We also performed several of these studies with intravenous BCG vaccination (ivBCG), which in the mouse model leads to more sustained bacterial replication than scBCG [61]. While we observed similar remodeling to AMs in the ivBCG model, these were not different in quality to those of scBCG vaccination, despite the major differences in bacterial replication and far greater T cell recruitment to the airway, suggesting that these changes are not required for AM remodeling (S9 Fig). There is still much unknown about the signals that drive reprogramming of tissue-resident innate cells. Ideally, vaccines would be designed to leverage these signals in order to promote the most effective interactions between innate and adaptive responses. Identifying the ways that AMs are reprogrammed by inflammatory signals and the effects of their changed pheno- types on the early stages of infection will help to improve future vaccines or host-directed therapies. Materials and methods Ethics statement Animal studies performed at Seattle Children’s Research Institute were performed in compli- ance with and approval by the Seattle Children’s Research Institute’s Institutional Animal Care and Use Committee. Animal studies performed at University of Massachusetts Amherst were performed in compliance with and approval by the University of Massachusetts Amherst’s Institutional Animal Care and Use Committee. All mice were housed and maintained in spe- cific pathogen-free conditions. Mice C57BL/6 mice were purchased from Jackson Laboratories (Bar Harbor, ME). 6–12 week old male and female mice were used for all experiments, except for RNA-sequencing, which used only female mice for uniformity. Mice infected with Mtb were housed in Biosafety Level 3 facilities in Animal Biohazard Containment Suites. PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 17 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium Mycobacteria exposure models: BCG immunization and establishment of coMtb BCG-Pasteur was cultured in Middlebrook 7H9 broth at 37˚C to an OD of 0.1–0.3. Bacteria was diluted in PBS and 1 x 106 CFU in 200 ml was injected subcutaneous (SC) or intravenous (IV). Intradermal infections to establish coMtb were performed as formerly described [24], with some modifications as detailed previously [25]. Briefly, 10,000 CFU of Mtb (H37Rv) in logarithmic phase growth were injected intradermally into the ear in 10 μL PBS using a 10 μL Hamilton Syringe, following anesthesia with ketamine/xylazine. M. tuberculosis aerosol infections and lung mononuclear cell isolation Aerosol infections were performed with wildtype H37Rv, including some transformed with an mEmerald reporter pMV261 plasmid, generously provided by Dr. Chris Sassetti and Christina Baer (University of Massachusetts Medical School, Worcester, MA). For both standard (~100 CFU) and high dose (1,473–4,800 CFU) infections, mice were enclosed in an aerosol infection chamber (Glas-Col) and frozen stocks of bacteria were thawed and placed inside the associated nebulizer. To determine the infectious dose, three mice in each infection were sacrificed one day later and lung homogenates were plated onto 7H10 plates for CFU enumeration. High dose challenge and sorting of Mtb-infected AM was performed 4 weeks following scBCG vac- cination and 2 weeks following coMtb vaccination as previously described [62]. All other anal- ysis was performed 8 weeks following Mycobacterium exposures. Lung single cell suspensions At each time point, lungs were removed, and single-cell suspensions of lung mononuclear cells were prepared by Liberase Blendzyme 3 (70 ug/ml, Roche) digestion containing DNaseI (30 μg/ml; Sigma-Aldrich) for 30 mins at 37˚C and mechanical disruption using a gentle- MACS dissociator (Miltenyi Biotec), followed by filtering through a 70 μM cell strainer. Cells were resuspended in FACS buffer (PBS, 1% FBS, and 0.1% NaN3) prior to staining for flow cytometry. For bacterial enumeration, lungs were processed in 0.05% Tween-80 in PBS using a gentleMACS dissociator (Miltenyi Biotec) and were plated onto 7H10 plates for CFU enumer- ation. ΔCFU (log) was calculated as follows: ΔCFU = log((sample CFU)/(average control CFU*). *For respective experiment, timepoint, and organ. A ΔCFU value of -1 corresponds to a 10-fold reduction in CFU for the sample, compared to the control. Similarly, a ΔCFU value of 1 corresponds to a 10-fold increase in CFU. Alveolar macrophage isolation AMs for cell sorting, bulk RNA-sequencing, single cell RNA-sequencing, and ex vivo stimula- tion were collected by bronchoalveolar lavage (BAL). BAL was performed by exposing the tra- chea of euthanized mice, puncturing the trachea with Vannas Micro Scissors (VWR) and injecting 1 mL PBS using a 20G-1” IV catheter (McKesson) connected to a 1 mL syringe. The PBS was flushed into the lung and then aspirated three times and the recovered fluid was placed in a 15mL tube on ice. The wash was repeated 3 additional times. Cells were filtered and spun down. For antibody staining, cells were suspended in FACS buffer. For cell culture, cells were plated at a density of 5 x 104 cells/well (96-well plate) in complete RPMI (RPMI plus FBS (10%, VWR), L-glutamine (2mM, Invitrogen), and Penicillin-Streptomycin (100 U/ml; Invitrogen) and allowed to adhere overnight in a 37˚C humidified incubator (5% CO2). Media with antibiotics were washed out prior to infection with Mtb. PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 18 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium Cell sorting and flow cytometry Fc receptors were blocked with anti-CD16/32 (2.4G2, BD Pharmingen). Cell viability was assessed using Zombie Violet dye (Biolegend). Cells were suspended in 1X PBS (pH 7.4) con- taining 0.01% NaN3 and 1% fetal bovine serum (i.e., FACS buffer). Surface staining, performed at 4 degrees for 20 minutes, included antibodies specific for murine: Siglec F (E50-2440, BD Pharmingen), CD11b (M1/70), CD64 (X54-5/7.1), CD45 (104), CD3 (17A2, eBiosciences), CD19 (1D3, eBiosciences), CD11c (N418), I-A/I-E (M5/114.15.2), Ly6G (1A8), Ly6C (HK1.4), TNF (MP6-XT22). For ICS, Brefeldin A was added for duration of LPS stimulation. Cyto-Fast Fix/Perm and Cyto-Fast Perm Wash reagents were used for intracellular staining. Reagents are from Biolegend unless otherwise noted. MHC class II tetramers ESAT-6 (I-A(b) 4–17, sequence: QQWNFAGIEAAASA) and MHC class I tetramers TB10.4 (H-2K(b) 4–11, sequence: IMYNYPAM) were obtained from the National Institutes of Health Tetramer Core Facility. Cell sorting was performed on a FACS Aria (BD Biosciences). Sorted cells were col- lected in complete media, spun down, resuspended in TRIzol, and frozen at -80˚C overnight prior to RNA isolation. Samples for flow cytometry were fixed in 2% paraformaldehyde solu- tion in PBS and analyzed using a LSRII flow cytometer (BD Biosciences) and FlowJo software (Tree Star, Inc.). Bulk RNA-sequencing and analysis All high dose infections and sorting for bulk RNA-sequencing of Mtb-infected AMs (control, scBCG, and coMtb) were performed in the ABSL-3 facility at Seattle Children’s Research Insti- tute. All infections used the same Mtb strain, mEmerald-H37Rv, and the TRIzol-based RNA isolation protocol was performed by the same individual (D.M.). RNA isolation was performed using TRIzol (Invitrogen), two sequential chloroform extractions, Glycoblue carrier (Thermo Fisher), isopropanol precipitation, and washes with 75% ethanol. RNA was quantified with the Bioanalyzer RNA 6000 Pico Kit (Agilent). cDNA libraries were constructed using the SMAR- Ter Stranded Total RNA-Seq Kit (v2)–Pico Input Mammalian (Clontech) following the manu- facturer’s instructions. Libraries were amplified and then sequenced on an Illumina NextSeq (2 x 76, paired-end (sorted BAL cells) or 2 x 151, paired-end (ex vivo stimulation samples)). Stranded paired-end reads were preprocessed: The first three nucleotides of R2 were removed as described in the SMARTer Stranded Total RNA-Seq Kit–Pico Input Mammalian User Man- ual (v2: 063017) and read ends consisting of more than 66% of the same nucleotide were removed). The remaining read pairs were aligned to the mouse genome (mm10) + Mtb H37Rv genome using the gsnap aligner [63] (v. 2018-07-04) allowing for novel splicing. Con- cordantly mapping read pairs (~20 million / sample) that aligned uniquely were assigned to exons using the iocond program and gene definitions from Ensembl Mus_Musculus GRCm38.78 coding and non-coding genes. Genes with low expression were filtered using the “filterByExpr” function in the edgeR package [64]. Differential expression was calculated using the “edgeR” package [64] from ioconductor.org. False discovery rate was computed with the Benjamini-Hochberg algorithm. Hierarchical clusterings were performed in R using ‘Tsclust’ and ‘hclust’ libraries. Heat map and scatterplot visualizations were generated in R using the ‘heatmap.2’ and ‘ggplot2’ libraries, respectively. Gene Set Enrichment Analysis (GSEA) Input data for GSEA consisted of lists, ranked by -log(p-value), comparing RNAseq expression measures of target samples and naïve controls including directionality of fold-change. Mouse orthologs of human Hallmark genes were defined using a list provided by Molecular Signa- tures Database (MsigDB) [65]. GSEA software was used to calculate enrichment of ranked lists PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 19 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium in each of the respective hallmark gene lists, as described previously [66]. A nominal p-value for each ES is calculated based on the null distribution of 1,000 random permutations. To cor- rect for multiple hypothesis testing, a normalized enrichment score (NES) is calculated that corrects the ES based on the null distribution. A false-discovery rate (FDR) is calculated for each NES. Leading edge subsets are defined as the genes in a particular gene set that are part of the ranked list at or before the running sum reaches its maximum value. Ingenuity Pathway Analysis (IPA) IPA (QIAGEN) was used to identify enriched pathways for differentially expressed genes between naïve and Mtb-infected AMs (cut-off values: FDR < 0.01, |FC| > 2). The top 20 canonical pathways with enrichment score p-value < 0.05 with greater than 10 gene members are reported. Single cell RNA-sequencing BAL from 10 mice per condition was pooled for each sample, with two independent replicates per condition. Samples were prepared for methanol fixation following protocol “CG000136 Rev. D” from 10X Genomics [67]. Briefly, samples were filtered with 70 μm filters and red blood cells were lysed with ACK lysis buffer. Samples were resuspended in 1 mL ice-cold DPBS using a wide-bore tip and transferred to a 1.5 mL low-bind Eppendorf tube. Samples were centrifuged at 700 × g for 5 minutes at 4˚C. Supernatant was carefully removed with a p1000 pipette, and the cell pellet was washed two more times with DPBS, counted, and resus- pended in 200 μL ice-cold DPBS/1 × 106 cells. 800 μL of ice-cold methanol was added drop- wise for a final concentration of 80% methanol. Samples were incubated at -20˚C for 30 min- utes and then stored at -80˚C for up to 6 weeks prior to rehydration. For rehydration, frozen samples were equilibrated to 4˚C, centrifuged at 1,000 × g for 10 minutes at 4˚C, and resus- pended in 50 μL of Wash-Resuspension Buffer (0.04% BSA + 1mM DTT + 0.2U/μL Protector RNAase Inhibitor in 3× SSC buffer) to achieve ~1,000 cells/μL (assuming 75% sample loss). Single cell RNA-sequencing analysis Libraries were prepared using the Next GEM Single Cell 30 Reagent Kits v3.1 (Dual Index) (10X Genomics) following the manufacturer’s instructions. Raw sequencing data were aligned to the mouse genome (mm10) and UMI counts determined using the Cell Ranger pipeline (10X Genomics). Data processing, integration, and analysis was performed with Seurat v.3 [68]. Droplets containing less than 200 detected genes, more than 4000 detected genes (doublet discrimination), or more than 5% mitochondrial were discarded. Genes expressed by less than 3 cells across all samples were removed. Unbiased annotation of clusters using the Immgen database [69] as a reference was performed with “SingleR” package [70]. Pseudotime analysis was performed using the “SeuratWrappers” and “Monocle3” R packages [71]. Data visualiza- tion was performed with the “Seurat”, “tidyverse”, “cowplot”, and “viridis” R packages. Alveolar macrophage Ex Vivo stimulation AMs were isolated by bronchoalveolar lavage and pooled from 5 mice per group. Cells were plated at a density of 5 x 104 cells/well (96-well plate) in complete RPMI (RPMI plus FBS (10%, VWR), L-glutamine (2mM, Invitrogen), and Penicillin-Streptomycin (100 U/ml; Invitrogen) and allowed to adhere overnight in a 37˚C humidified incubator (5% CO2). Media with antibi- otics and non-adherent cells were washed out prior to stimulation. AM were stimulated with LPS (LPS from Salmonella Minnesota, List Biologicals, #R595, 10 ng/ml), Pam3Cys PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 20 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium (Pam3CSK4, EMC Microcollections, GmbH, 10 ng/ml), or H37Rv (effective MOI ~2:1). H37Rv was prepared by culturing from frozen stock in 7H9 media at 37˚C for 48 hours to O. D. of 0.1–0.3. The final concentration was calculated based on strain titer and bacteria was added to macrophages for two hours. Cultures were then washed three times to remove extra- cellular bacteria. Cell cultures were washed once in PBS after 6 hours to remove dead cells and collected in TRIzol for RNA isolation via chloroform/isopropanol extraction or collected after 20 hours for flow cytometry and ICS. Filtering for IFN dependent and independent gene sets “IFN dependent” and “IFN independent” gene sets were generated from data from Olson et al [38], using the following filters starting from a total of 1,233 genes up-regulated in H37Rv- stimulated WT BMDM with average CPM >1, log2 fold change > 1 and FDR < 0.01: “IFN dependent” = H37Rv-stimulated IFNAR-/- BMDM: log2 fold change < 1 AND H37Rv-stimulated WT vs IFNAR-/-: log2 fold change > 2 = 339 genes “IFN independent” = H37Rv-stimulated IFNAR-/- BMDM: log2 fold change > 1, FDR < 0.01 AND H37Rv-stimulated WT vs IFNAR-/-: log2 fold change < 2 = 352 genes qRT-PCR Quantitative PCR reactions were carried out using TaqMan primer probes (ABI) and TaqMan Fast Universal PCR Master Mix (ThermoFisher) in a CFX384 Touch Real-Time PCR Detec- tion System (BioRad). Data were normalized by the level of Ef1a expression in individual samples. Statistical analyses RNA-sequencing was analyzed using the edgeR package from Bioconductor.org and the false discovery rate was computed using the Benjamini-Hochberg algorithm. All other data are pre- sented as mean ± SEM and analyzed by one-way ANOVA (95% confidence interval) with Tukey post-test (for comparison of multiple conditions). Statistical analysis and graphical representation of data was performed using either GraphPad Prism v6.0 software or R. PCA plots generated using “Prcomp” and “Biplot” packages. Venn diagrams and gene set intersec- tion analysis was performed using Intervene [72]. p-values, * p < 0.05, ** p < 0.01, *** p < 0.001. Supporting information S1 Fig. (related to Fig 1). Flow cytometry gating schemes. Gating strategies for myeloid (A) and T cell (B) analysis. (TIF) S2 Fig. (related to Fig 1). Mycobacterium exposure provides protection against standard low-dose H37Rv aerosol challenge. A) Lung, spleen, and lung-draining lymph node (LN) CFU in control mice at deposition, days 10, 12, 14, and 28. B-E) Summary plots of ΔCFU (log) in lung, spleen, and LN following low-dose infection with H37Rv at day 10 (B), day 12 (C), day 14 (D), and day 28 (E). *p < 0.05, **p < 0.01, ***p < 0.001. One-way ANOVA with Tukey post-test. Data compiled from 2–3 independent experiments per condition, with 5 mice per group for each experiment. (TIF) PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 21 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium S3 Fig. (related to Fig 2). Top 20 Canonical Pathways by Ingenuity Pathway Analysis for up-regulated genes by Mtb-infected alveolar macrophages. IPA analysis for Mtb-infected AMs from control, scBCG, and coMtb mice 24 hours following high dose mEmerald-H37Rv infection. Data representative of 3 independent experiments per condition. (TIF) S4 Fig. (related to Fig 2). Transcriptional changes to naive alveolar macrophages following Mycobacterium exposure by bulk RNA-sequencing. Bulk RNA-seq profiles of naive AMs (isolated alongside Mtb-infected AMs). Gene expression changes within naïve AMs are com- pared to AMs from control mice: naïve scBCG AM vs naïve control AM; naïve coMtb AM vs naïve control AM. A-B) Volcano plots depicting changes in baseline gene expression of naive AMs from scBCG (A) and coMtb(B) mice compared to naive AMs from control mice. Signifi- cantly changed genes (FDR < 0.05, |FC| > 2) highlighted and labeled. C) Gene expression for innate receptors and adaptors of interest, log2 fold change, unstimulated AMs from scBCG and coMtb mice compared to unstimulated AMs from control mice. * FDR < 0.01. Compiled from 2 independent experiments for each condition. (TIF) S5 Fig. (related to Fig 3). Flow analysis of BAL samples prepared for 10X single-cell RNA- sequencing. Percentage of each population (AM, lymphocytes, eosinophils, MDM, other CD45+) out of CD45+ ZV-. AM = Siglec F+ CD64+, Eosinophils = Siglec F+ CD64-, lymphocytes = CD3/CD19+, MDM = Siglec F- CD64+, other CD45+ = CD3- CD19- Siglec F- CD64-. Note: One of the two coMtb samples analyzed by flow cytometry did not have an accompanying 10X sample. The second coMtb 10X sample was processed separately without flow analysis. (TIF) S6 Fig. (related to Fig 3). Top 10 genes differentially expressed for each of 11 macrophage subclusters. Heatmap of genes that are most differentially expressed for each of 11 clusters with all other clusters. Genes filtered with log fold change threshold of > 0.25 and minimum percentage expression of 25% of cells. All genes but one (Gsto1) had an adjusted p-value of < 1.0x10-5. *Five genes (Fabp4, Fabp5, Stmn1, Mki67, Cbr2) met this criterion for more than one cluster, grouped with the more abundant cluster. Data is compiled from two inde- pendent experiments, 3 conditions each, for a total of 6 samples. (TIF) S7 Fig. (related to Fig 3). UMAP gene expression plots for genes associated with macro- phage subcluster 3 and found in AM_2 (Pisu et al) (31). Genes associated with mitochon- drial oxidative phosphorylation (mt-Co1, mt-Cytb, mt-Nd2), chromatin remodeling (Ankrd11, Baz1a), macrophage-associated transcription factors (Cebpb, Zeb2, Bhlhe40, Hif1a), and CARD9 signaling (Malt1, Bcl10). Data is compiled from two independent experiments with 3 conditions each, for a total of 6 samples. (TIF) S8 Fig. (related to Fig 3). Frequency and gene expression of Cluster 6 macrophages across exposure conditions. A) Single-cell RNA-sequencing from BAL samples from control, scBCG, and coMtb mice. Subcluster of macrophages with each cluster annotated. Relative fre- quency of Cluster 6 for each replicate. B) Differentially expressed genes within Cluster 6 between control vs scBCG vs coMtb samples. Wilcoxon Rank Sum Test, Bonferroni adjusted p-value, ***adj-p < 0.001. (TIF) PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 22 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium S9 Fig. (related to Fig 3). IV BCG vaccination leads to similar remodeling of alveolar mac- rophages as SC BCG vaccination. A-B) Bulk RNA-seq gene expression analysis between naive and Mtb-infected AMs 24 hours following high-dose mEmerald-H37Rv infection in mice previously exposed to scBCG, ivBCG, and coMtb, compared to controls. (controls- reported in Rothchild et al, 2019(10); CMTB- reported in Nemeth et al, 2020(25)). A) Principal Component Analysis using DEG (fold change > |2|, FDR < 0.05) in Mtb-infected AMs at 24 hours. B) Heatmap of 131 DEG at 24 hours in Mtb-infected AM (left), Interferon Stimulated Genes, derived from macrophage response to IFNα (fold change >2, p-value < 0.01) Mosta- favi et al, 2016 (30)(middle), IL6 JAK STAT3 hallmark pathway (right). C) Relative frequency of 3 key clusters for macrophage subset from control, scBCG, ivBCG, and coMtb scRNAseq BAL samples. (A-B) Compiled from 3+ independent experiments per condition for control, 2 independent experiments per condition for scBCG, ivBCG, and coMtb. (C) Data is compiled from two independent experiments (circle, triangle) with 3 conditions each for a total of 6 samples. (TIF) S10 Fig. (related to Fig 4). UMAP gene expression plots of cluster and lineage marker genes of interest for T cell subclusters. Data is compiled from two independent experiments with 3 conditions each for a total of 6 samples. (TIF) S11 Fig. (related to Fig 5). Gene expression of alveolar macrophages from ex vivo stimula- tions. A) Gene expression changes measured by bulk RNA-seq for stimulated AMs compared to respective unstimulated AMs (i.e., LPS-stim control AM vs unstim control AM; LPS-stim scBCG AM vs unstim scBCG AM; LPS-stim coMtb AM vs unstim coMtb AM). AMs were stimulated for 6 hours with Pam3Cys (10 ng/ml), LPS (10 ng/ml), or H37Rv (effective MOI ~2:1). Volcano plots depict fold change (log2) and P-value (-log10) for each stimulation condi- tion for each of the three groups (control scBCG, coMtb) compared to the respective unstimu- lated controls. DEG (p-value < 0.001; |fold change| > 2) highlighted and labeled, space permitting. B) Baseline gene expression for innate receptors and adaptors of interest from scBCG and coMtb AM compared to control AM, log2 fold change, unstim scBCG AM vs unstim control AM; unstim coMtb AM vs unstim control AM. Compiled from 3 independent experiments. (TIF) S12 Fig. (related to Fig 5). Cell-intrinsic changes in alveolar macrophage response is retained 23 weeks following vaccination. Gene expression of Mx1, Cxcl10, Il1b, Cxcl2, Irf7, and Il6 as measured by qPCR in AMs isolated by BAL from mice 8 and 23 weeks following scBCG vaccination and from age-matched controls, with and without LPS (10 ng/ml) stimula- tion. Data is representative of technical AM duplicates from a single experiment. (TIF) S1 Table. RNA-Sequencing data for alveolar macrophages 24 hours following high dose H37Rv-mEmerald challenge from scBCG mice. (XLSX) S2 Table. Top differentially expressed genes for individual clusters for macrophage, T cell, and dendritic cell sub-cluster analysis. (XLSX) S3 Table. RNA-Sequencing data for ex vivo stimulated alveolar macrophages. (XLSX) PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 23 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium S4 Table. IFN-independent and IFN-dependent genes based on WT and IFNAR-/- BMDM RNA-seq data. (XLSX) Acknowledgments We thank the Animal Care staff at Seattle Children’s Research Institute and University of Mas- sachusetts Amherst, Pamela Troisch and the Next Gen Sequencing core at the Institute for Sys- tems Biology. The authors acknowledge Research Scientific Computing at Seattle Children’s Research Institute for providing HPC resources that have contributed to the research results reported within this paper. We thank members of the Aderem, Urdahl, and Rothchild labs for helpful discussions. Author Contributions Conceptualization: Dat Mai, Johannes Nemeth, Kevin Urdahl, Alan H. Diercks, Alan Aderem, Alissa C. Rothchild. Data curation: Michael Morikubo, Alan H. Diercks, Alissa C. Rothchild. Formal analysis: Dat Mai, Michael Morikubo, Alan H. Diercks, Alissa C. Rothchild. Funding acquisition: Kevin Urdahl, Alan H. Diercks, Alan Aderem, Alissa C. Rothchild. Investigation: Dat Mai, Ana Jahn, Tara Murray, Pamelia N. Lim, Maritza M. Cervantes, Linh K. Pham, Alissa C. Rothchild. Supervision: Alan H. Diercks, Alan Aderem, Alissa C. Rothchild. Validation: Dat Mai, Alissa C. Rothchild. Writing – original draft: Alan H. Diercks, Alissa C. Rothchild. Writing – review & editing: Dat Mai, Pamelia N. Lim, Linh K. Pham, Alan H. Diercks, Alissa C. Rothchild. References 1. World Health Organization. Global tuberculosis report 2021. Geneva. Licence: CC BY-NC-SA 3.0 IGO. 2022. 2. Dheda K, Perumal T, Moultrie H, Perumal R, Esmail A, Scott AJ, et al. The intersecting pandemics of tuberculosis and COVID-19: population-level and patient-level impact, clinical presentation, and correc- tive interventions. Lancet Respir Med. 2022. https://doi.org/10.1016/S2213-2600(22)00092-3 PMID: 35338841 3. Netea MG, Dominguez-Andres J, Barreiro LB, Chavakis T, Divangahi M, Fuchs E, et al. Defining trained immunity and its role in health and disease. Nat Rev Immunol. 2020; 20(6):375–88. https://doi.org/10. 1038/s41577-020-0285-6 PMID: 32132681 4. Sherwood ER, Burelbach KR, McBride MA, Stothers CL, Owen AM, Hernandez A, et al. Innate Immune Memory and the Host Response to Infection. J Immunol. 2022; 208(4):785–92. https://doi.org/10.4049/ jimmunol.2101058 PMID: 35115374 5. Khader SA, Divangahi M, Hanekom W, Hill PC, Maeurer M, Makar KW, et al. Targeting innate immunity for tuberculosis vaccination. J Clin Invest. 2019; 129(9):3482–91. https://doi.org/10.1172/JCI128877 PMID: 31478909 6. Hoyer FF, Naxerova K, Schloss MJ, Hulsmans M, Nair AV, Dutta P, et al. Tissue-Specific Macrophage Responses to Remote Injury Impact the Outcome of Subsequent Local Immune Challenge. Immunity. 2019; 51(5):899–914 e7. https://doi.org/10.1016/j.immuni.2019.10.010 PMID: 31732166 PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 24 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium 7. Aegerter H, Kulikauskaite J, Crotta S, Patel H, Kelly G, Hessel EM, et al. Influenza-induced monocyte- derived alveolar macrophages confer prolonged antibacterial protection. Nat Immunol. 2020; 21 (2):145–57. https://doi.org/10.1038/s41590-019-0568-x PMID: 31932810 8. Yao Y, Jeyanathan M, Haddadi S, Barra NG, Vaseghi-Shanjani M, Damjanovic D, et al. Induction of Autonomous Memory Alveolar Macrophages Requires T Cell Help and Is Critical to Trained Immunity. Cell. 2018; 175(6):1634–50 e17. https://doi.org/10.1016/j.cell.2018.09.042 PMID: 30433869 9. Zhu B, Wu Y, Huang S, Zhang R, Son YM, Li C, et al. Uncoupling of macrophage inflammation from self-renewal modulates host recovery from respiratory viral infection. Immunity. 2021; 54(6):1200–18 e9. https://doi.org/10.1016/j.immuni.2021.04.001 PMID: 33951416 10. Rothchild AC, Olson GS, Nemeth J, Amon LM, Mai D, Gold ES, et al. Alveolar macrophages generate a noncanonical NRF2-driven transcriptional response to Mycobacterium tuberculosis in vivo. Sci Immu- nol. 2019; 4(37). https://doi.org/10.1126/sciimmunol.aaw6693 PMID: 31350281 11. Cohen SB, Gern BH, Delahaye JL, Adams KN, Plumlee CR, Winkler JK, et al. Alveolar Macrophages Provide an Early Mycobacterium tuberculosis Niche and Initiate Dissemination. Cell Host Microbe. 2018; 24(3):439–46 e4. https://doi.org/10.1016/j.chom.2018.08.001 PMID: 30146391 12. Mangtani P, Abubakar I, Ariti C, Beynon R, Pimpin L, Fine PE, et al. Protection by BCG vaccine against tuberculosis: a systematic review of randomized controlled trials. Clin Infect Dis. 2014; 58(4):470–80. https://doi.org/10.1093/cid/cit790 PMID: 24336911 13. 14. Trunz BB, Fine P, Dye C. Effect of BCG vaccination on childhood tuberculous meningitis and miliary tuberculosis worldwide: a meta-analysis and assessment of cost-effectiveness. Lancet. 2006; 367 (9517):1173–80. https://doi.org/10.1016/S0140-6736(06)68507-3 PMID: 16616560 Lange C, Aaby P, Behr MA, Donald PR, Kaufmann SHE, Netea MG, et al. 100 years of Mycobacterium bovis bacille Calmette-Guerin. Lancet Infect Dis. 2022; 22(1):e2–e12. 15. Kaufmann E, Sanz J, Dunn JL, Khan N, Mendonca LE, Pacis A, et al. BCG Educates Hematopoietic Stem Cells to Generate Protective Innate Immunity against Tuberculosis. Cell. 2018; 172(1–2):176–90 e19. https://doi.org/10.1016/j.cell.2017.12.031 PMID: 29328912 16. Delahaye JL, Gern BH, Cohen SB, Plumlee CR, Shafiani S, Gerner MY, et al. Cutting Edge: Bacillus Calmette-Guerin-Induced T Cells Shape Mycobacterium tuberculosis Infection before Reducing the Bacterial Burden. J Immunol. 2019; 203(4):807–12. 17. Das S, Marin ND, Esaulova E, Ahmed M, Swain A, Rosa BA, et al. Lung Epithelial Signaling Mediates Early Vaccine-Induced CD4(+) T Cell Activation and Mycobacterium tuberculosis Control. mBio. 2021; 12(4):e0146821. https://doi.org/10.1128/mBio.01468-21 PMID: 34253059 18. Mata E, Tarancon R, Guerrero C, Moreo E, Moreau F, Uranga S, et al. Pulmonary BCG induces lung- resident macrophage activation and confers long-term protection against tuberculosis. Sci Immunol. 2021; 6(63):eabc2934. https://doi.org/10.1126/sciimmunol.abc2934 PMID: 34559551 19. Jeyanathan M, Vaseghi-Shanjani M, Afkhami S, Grondin JA, Kang A, D’Agostino MR, et al. Parenteral BCG vaccine induces lung-resident memory macrophages and trained immunity via the gut-lung axis. Nat Immunol. 2022; 23(12):1687–702. https://doi.org/10.1038/s41590-022-01354-4 PMID: 36456739 20. Arts RJW, Moorlag S, Novakovic B, Li Y, Wang SY, Oosting M, et al. BCG Vaccination Protects against Experimental Viral Infection in Humans through the Induction of Cytokines Associated with Trained Immunity. Cell Host Microbe. 2018; 23(1):89–100 e5. https://doi.org/10.1016/j.chom.2017.12.010 PMID: 29324233 21. Kleinnijenhuis J, Quintin J, Preijers F, Joosten LA, Jacobs C, Xavier RJ, et al. BCG-induced trained immunity in NK cells: Role for non-specific protection to infection. Clin Immunol. 2014; 155(2):213–9. https://doi.org/10.1016/j.clim.2014.10.005 PMID: 25451159 22. Koeken V, van der Pasch ES, Leijte GP, Mourits VP, de Bree LCJ, Moorlag S, et al. The effect of BCG vaccination on alveolar macrophages obtained from induced sputum from healthy volunteers. Cytokine. 2020; 133:155135. https://doi.org/10.1016/j.cyto.2020.155135 PMID: 32534356 23. Soto JA, Galvez NMS, Andrade CA, Ramirez MA, Riedel CA, Kalergis AM, et al. BCG vaccination induces cross-protective immunity against pathogenic microorganisms. Trends Immunol. 2022; 43 (4):322–35. https://doi.org/10.1016/j.it.2021.12.006 PMID: 35074254 24. Kupz A, Zedler U, Staber M, Kaufmann SH. A Mouse Model of Latent Tuberculosis Infection to Study Intervention Strategies to Prevent Reactivation. PLoS One. 2016; 11(7):e0158849. https://doi.org/10. 1371/journal.pone.0158849 PMID: 27391012 25. Nemeth J, Olson GS, Rothchild AC, Jahn AN, Mai D, Duffy FJ, et al. Contained Mycobacterium tubercu- losis infection induces concomitant and heterologous protection. PLoS Pathog. 2020; 16(7):e1008655. https://doi.org/10.1371/journal.ppat.1008655 PMID: 32673357 PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 25 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium 26. Cadena AM, Hopkins FF, Maiello P, Carey AF, Wong EA, Martin CJ, et al. Concurrent infection with Mycobacterium tuberculosis confers robust protection against secondary infection in macaques. PLoS Pathog. 2018; 14(10):e1007305. https://doi.org/10.1371/journal.ppat.1007305 PMID: 30312351 27. Andrews JR, Noubary F, Walensky RP, Cerda R, Losina E, Horsburgh CR. Risk of progression to active tuberculosis following reinfection with Mycobacterium tuberculosis. Clin Infect Dis. 2012; 54(6):784–91. https://doi.org/10.1093/cid/cir951 PMID: 22267721 28. Wolf AJ, Linas B, Trevejo-Nunez GJ, Kincaid E, Tamura T, Takatsu K, et al. Mycobacterium tuberculo- sis infects dendritic cells with high frequency and impairs their function in vivo. J Immunol. 2007; 179 (4):2509–19. https://doi.org/10.4049/jimmunol.179.4.2509 PMID: 17675513 29. Mollenkopf HJ, Kursar M, Kaufmann SH. Immune response to postprimary tuberculosis in mice: Myco- bacterium tuberculosis and Miycobacterium bovis bacille Calmette-Guerin induce equal protection. J Infect Dis. 2004; 190(3):588–97. 30. Mostafavi S, Yoshida H, Moodley D, LeBoite H, Rothamel K, Raj T, et al. Parsing the Interferon Tran- scriptional Network and Its Disease Associations. Cell. 2016; 164(3):564–78. https://doi.org/10.1016/j. cell.2015.12.032 PMID: 26824662 31. Pisu D, Huang L, Narang V, Theriault M, Le-Bury G, Lee B, et al. Single cell analysis of M. tuberculosis phenotype and macrophage lineages in the infected lung. J Exp Med. 2021; 218(9). https://doi.org/10. 1084/jem.20210615 PMID: 34292313 32. Travaglini KJ, Nabhan AN, Penland L, Sinha R, Gillich A, Sit RV, et al. A molecular cell atlas of the human lung from single-cell RNA sequencing. Nature. 2020; 587(7835):619–25. https://doi.org/10. 1038/s41586-020-2922-4 PMID: 33208946 33. Scott CL, T’Jonck W, Martens L, Todorov H, Sichien D, Soen B, et al. The Transcription Factor ZEB2 Is Required to Maintain the Tissue-Specific Identities of Macrophages. Immunity. 2018; 49(2):312–25 e5. https://doi.org/10.1016/j.immuni.2018.07.004 PMID: 30076102 34. Cain DW, O’Koren EG, Kan MJ, Womble M, Sempowski GD, Hopper K, et al. Identification of a tissue- specific, C/EBPbeta-dependent pathway of differentiation for murine peritoneal macrophages. J Immu- nol. 2013; 191(9):4665–75. 35. Ciofani M, Madar A, Galan C, Sellars M, Mace K, Pauli F, et al. A validated regulatory network for Th17 cell specification. Cell. 2012; 151(2):289–303. https://doi.org/10.1016/j.cell.2012.09.016 PMID: 23021777 36. Edwards SC, Hedley A, Hoevenaar WHM, Glauner T, R W, Kilbey A, et al. Single-cell analysis uncovers 1 differential regulation of lung γδ T cell subsets by the co-inhibitory molecules, PD-1 and TIM-3. bioR- xiv. 2021;2021.07.04.451035;. 37. Guilliams M, De Kleer I, Henri S, Post S, Vanhoutte L, De Prijck S, et al. Alveolar macrophages develop from fetal monocytes that differentiate into long-lived cells in the first week of life via GM-CSF. J Exp Med. 2013; 210(10):1977–92. https://doi.org/10.1084/jem.20131199 PMID: 24043763 38. Olson GS, Murray TA, Jahn AN, Mai D, Diercks AH, Gold ES, et al. Type I interferon decreases macro- phage energy metabolism during mycobacterial infection. Cell Rep. 2021; 35(9):109195. https://doi.org/ 10.1016/j.celrep.2021.109195 PMID: 34077724 39. Huang L, Nazarova EV, Tan S, Liu Y, Russell DG. Growth of Mycobacterium tuberculosis in vivo segre- gates with host macrophage metabolism and ontogeny. J Exp Med. 2018; 215(4):1135–52. https://doi. org/10.1084/jem.20172020 PMID: 29500179 40. Griffiths KL, Ahmed M, Das S, Gopal R, Horne W, Connell TD, et al. Targeting dendritic cells to acceler- ate T-cell activation overcomes a bottleneck in tuberculosis vaccine efficacy. Nat Commun. 2016; 7:13894. https://doi.org/10.1038/ncomms13894 PMID: 28004802 41. Lim PN, Cervantes MM, Pham LK, Rothchild AC. Alveolar macrophages: novel therapeutic targets for respiratory diseases. Expert Rev Mol Med. 2021; 23:e18. https://doi.org/10.1017/erm.2021.21 PMID: 34823627 42. Correa-Macedo W, Fava VM, Orlova M, Cassart P, Olivenstein R, Sanz J, et al. Alveolar macrophages from persons living with HIV show impaired epigenetic response to Mycobacterium tuberculosis. J Clin Invest. 2021; 131(22). https://doi.org/10.1172/JCI148013 PMID: 34473646 43. Verma AK, Bansal S, Bauer C, Muralidharan A, Sun K. Influenza Infection Induces Alveolar Macro- phage Dysfunction and Thereby Enables Noninvasive Streptococcus pneumoniae to Cause Deadly Pneumonia. J Immunol. 2020; 205(6):1601–7. https://doi.org/10.4049/jimmunol.2000094 PMID: 32796026 44. D’Agostino MR, Lai R, Afkhami S, Khera A, Yao Y, Vaseghi-Shanjani M, et al. Airway Macrophages Mediate Mucosal Vaccine-Induced Trained Innate Immunity against Mycobacterium tuberculosis in Early Stages of Infection. J Immunol. 2020; 205(10):2750–62. https://doi.org/10.4049/jimmunol. 2000532 PMID: 32998983 PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 26 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium 45. Gu H, Zeng X, Peng L, Xiang C, Zhou Y, Zhang X, et al. Vaccination induces rapid protection against bacterial pneumonia via training alveolar macrophage in mice. Elife. 2021; 10. https://doi.org/10.7554/ eLife.69951 PMID: 34544549 46. Peters JM, Irvine EB, Rosenberg JM, Wadsworth MH, Hughes TK, Sutton M, et al. Protective intrave- nous BCG vaccination induces enhanced immune signaling in the airways. bioRxiv. 2023. 47. Khan N, Downey J, Sanz J, Kaufmann E, Blankenhaus B, Pacis A, et al. M. tuberculosis Reprograms Hematopoietic Stem Cells to Limit Myelopoiesis and Impair Trained Immunity. Cell. 2020; 183(3):752– 70 e22. 48. Grant RA, Morales-Nebreda L, Markov NS, Swaminathan S, Querrey M, Guzman ER, et al. Circuits between infected macrophages and T cells in SARS-CoV-2 pneumonia. Nature. 2021; 590(7847):635– 41. https://doi.org/10.1038/s41586-020-03148-w PMID: 33429418 49. Berry MP, Graham CM, McNab FW, Xu Z, Bloch SA, Oni T, et al. An interferon-inducible neutrophil- driven blood transcriptional signature in human tuberculosis. Nature. 2010; 466(7309):973–7. https:// doi.org/10.1038/nature09247 PMID: 20725040 50. Zak DE, Penn-Nicholson A, Scriba TJ, Thompson E, Suliman S, Amon LM, et al. A blood RNA signature for tuberculosis disease risk: a prospective cohort study. Lancet. 2016; 387(10035):2312–22. https:// doi.org/10.1016/S0140-6736(15)01316-1 PMID: 27017310 51. Esaulova E, Das S, Singh DK, Choreno-Parra JA, Swain A, Arthur L, et al. The immune landscape in tuberculosis reveals populations linked to disease and latency. Cell Host Microbe. 2021; 29(2):165–78 e8. https://doi.org/10.1016/j.chom.2020.11.013 PMID: 33340449 52. Antonelli LR, Gigliotti Rothfuchs A, Goncalves R, Roffe E, Cheever AW, Bafica A, et al. Intranasal Poly- IC treatment exacerbates tuberculosis in mice through the pulmonary recruitment of a pathogen-per- missive monocyte/macrophage population. J Clin Invest. 2010; 120(5):1674–82. https://doi.org/10. 1172/JCI40817 PMID: 20389020 53. Redford PS, Mayer-Barber KD, McNab FW, Stavropoulos E, Wack A, Sher A, et al. Influenza A virus impairs control of Mycobacterium tuberculosis coinfection through a type I interferon receptor-depen- dent pathway. J Infect Dis. 2014; 209(2):270–4. https://doi.org/10.1093/infdis/jit424 PMID: 23935205 54. Mayer-Barber KD, Andrade BB, Barber DL, Hieny S, Feng CG, Caspar P, et al. Innate and adaptive interferons suppress IL-1alpha and IL-1beta production by distinct pulmonary myeloid subsets during Mycobacterium tuberculosis infection. Immunity. 2011; 35(6):1023–34. 55. 56. 57. 58. Zhang B, Moorlag SJ, Dominguez-Andres J, Bulut O, Kilic G, Liu Z, et al. Single-cell RNA sequencing reveals induction of distinct trained-immunity programs in human monocytes. J Clin Invest. 2022; 132 (7). https://doi.org/10.1172/JCI147719 PMID: 35133977 Lipscomb MF, Lyons CR, Nunez G, Ball EJ, Stastny P, Vial W, et al. Human alveolar macrophages: HLA-DR-positive macrophages that are poor stimulators of a primary mixed leukocyte reaction. J Immu- nol. 1986; 136(2):497–504. PMID: 2934472 Lyons CR, Ball EJ, Toews GB, Weissler JC, Stastny P, Lipscomb MF. Inability of human alveolar mac- rophages to stimulate resting T cells correlates with decreased antigen-specific T cell-macrophage binding. J Immunol. 1986; 137(4):1173–80. PMID: 2426354 Toews GB, Vial WC, Dunn MM, Guzzetta P, Nunez G, Stastny P, et al. The accessory cell function of human alveolar macrophages in specific T cell proliferation. J Immunol. 1984; 132(1):181–6. PMID: 6228577 59. Andersen P, Scriba TJ. Moving tuberculosis vaccines from theory to practice. Nat Rev Immunol. 2019; 19(9):550–62. https://doi.org/10.1038/s41577-019-0174-z PMID: 31114037 60. Srivastava S, Ernst JD. Cutting edge: Direct recognition of infected cells by CD4 T cells is required for control of intracellular Mycobacterium tuberculosis in vivo. J Immunol. 2013; 191(3):1016–20. https:// doi.org/10.4049/jimmunol.1301236 PMID: 23817429 61. Hilligan KL, Namasivayam S, Clancy CS, O’Mard D, Oland SD, Robertson SJ, et al. Intravenous admin- istration of BCG protects mice against lethal SARS-CoV-2 challenge. J Exp Med. 2022; 219(2). https:// doi.org/10.1084/jem.20211862 PMID: 34889942 62. Rothchild AC, Mai D, Aderem A, Diercks AH. Flow Cytometry Analysis and Fluorescence-activated Cell Sorting of Myeloid Cells from Lung and Bronchoalveolar Lavage Samples from Mycobacterium tubercu- losis-infected Mice. Bio Protoc. 2020; 10(10). https://doi.org/10.21769/bioprotoc.3630 PMID: 32995363 63. Wu TD, Nacu S. Fast and SNP-tolerant detection of complex variants and splicing in short reads. Bioin- formatics. 2010; 26(7):873–81. https://doi.org/10.1093/bioinformatics/btq057 PMID: 20147302 64. Robinson MD, McCarthy DJ, Smyth GK. edgeR: a Bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics. 2010; 26(1):139–40. https://doi.org/10.1093/ bioinformatics/btp616 PMID: 19910308 PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 27 / 28 PLOS PATHOGENS Alveolar macrophage remodeling by Mycobacterium 65. Liberzon A, Birger C, Thorvaldsdottir H, Ghandi M, Mesirov JP, Tamayo P. The Molecular Signatures Database (MSigDB) hallmark gene set collection. Cell Syst. 2015; 1(6):417–25. https://doi.org/10.1016/ j.cels.2015.12.004 PMID: 26771021 66. Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A. 2005; 102(43):15545–50. https://doi.org/10.1073/pnas.0506580102 PMID: 16199517 67. Chen J, Cheung F, Shi R, Zhou H, Lu W, Consortium CHI. PBMC fixation and processing for Chromium single-cell RNA sequencing. J Transl Med. 2018; 16(1):198. https://doi.org/10.1186/s12967-018-1578- 4 PMID: 30016977 68. Stuart T, Butler A, Hoffman P, Hafemeister C, Papalexi E, Mauck WM 3rd, et al. Comprehensive Inte- gration of Single-Cell Data. Cell. 2019; 177(7):1888–902 e21. https://doi.org/10.1016/j.cell.2019.05.031 PMID: 31178118 69. Heng TS, Painter MW, Immunological Genome Project C. The Immunological Genome Project: net- works of gene expression in immune cells. Nat Immunol. 2008; 9(10):1091–4. https://doi.org/10.1038/ ni1008-1091 PMID: 18800157 70. Aran D, Looney AP, Liu L, Wu E, Fong V, Hsu A, et al. Reference-based analysis of lung single-cell sequencing reveals a transitional profibrotic macrophage. Nat Immunol. 2019; 20(2):163–72. https:// doi.org/10.1038/s41590-018-0276-y PMID: 30643263 71. Cao J, Spielmann M, Qiu X, Huang X, Ibrahim DM, Hill AJ, et al. The single-cell transcriptional land- scape of mammalian organogenesis. Nature. 2019; 566(7745):496–502. https://doi.org/10.1038/ s41586-019-0969-x PMID: 30787437 72. Khan A, Mathelier A. Intervene: a tool for intersection and visualization of multiple gene or genomic region sets. BMC Bioinformatics. 2017; 18(1):287. https://doi.org/10.1186/s12859-017-1708-7 PMID: 28569135 PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1011871 January 18, 2024 28 / 28 PLOS PATHOGENS
null
10.1371_journal.pcbi.1011196.pdf
Data Availability Statement: Data can be found in the following doi: https://doi.org/10.6084/m9. figshare.21938651.v1.
Data can be found in the following 10.6084/m9. figshare.21938651.v1 .
RESEARCH ARTICLE Dynamic recycling of extracellular ATP in human epithelial intestinal cells Nicolas Andres SaffiotiID Virginia Gentilini5,6, Gabriel Eduardo Gondolesi5,6, Pablo Julio Schwarzbaum1,2*, Julieta SchachterID 1,2,3, Cora Lilia Alvarez1,4, Zaher Bazzi1,2, Marı´a 1,2* a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Saffioti NA, Alvarez CL, Bazzi Z, Gentilini MV, Gondolesi GE, Schwarzbaum PJ, et al. (2023) Dynamic recycling of extracellular ATP in human epithelial intestinal cells. PLoS Comput Biol 19(6): e1011196. https://doi.org/10.1371/journal. pcbi.1011196 Editor: Melissa L. Kemp, Georgia Institute of Technology and Emory University, UNITED STATES Received: January 25, 2023 Accepted: May 17, 2023 Published: June 29, 2023 Copyright: © 2023 Saffioti et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: Data can be found in the following doi: https://doi.org/10.6084/m9. figshare.21938651.v1. Funding: Grants from Secretarı´a de Ciencia y Te´cnica, Universidad de Buenos Aires (PJS, UBACYT 20020170100152BA), Consejo Nacional de Investigaciones Cientı´ficas y Te´cnicas (PJS, CONICET PIP 1013) and Agencia Nacional de Promocio´n Cientı´fica y Tecnolo´gica (NAS, PICT- 2019 03218; JS, PICT 2019-0204; PJS, PICT 2021- 1 Instituto de Quı´mica y Fı´sico-Quı´mica Biolo´ gicas “Prof. Alejandro C. Paladini”, Universidad de Buenos Aires (UBA), Consejo Nacional de Investigaciones Cientı´ficas y Te´cnicas (CONICET), Facultad de Farmacia y Bioquı´mica, Buenos Aires, Argentina, 2 Universidad de Buenos Aires (UBA), Facultad de Farmacia y Bioquı´mica, Departamento de Quı´mica Biolo´gica, Ca´tedra de Quı´mica Biolo´ gica, Buenos Aires, Argentina, 3 Instituto de Nanosistemas, Universidad Nacional de General San Martin, Buenos Aires, Argentina, 4 Universidad de Buenos Aires (UBA), Facultad de Ciencias Exactas y Naturales, Departamento de Biodiversidad y Biologı´a Experimental, Buenos Aires, Argentina, 5 Fundacio´n Favaloro Hospital Universitario, Unidad de Insuficiencia, Rehabilitacio´n y Trasplante Intestinal, Buenos Aires, Argentina, 6 Instituto de Medicina Traslacional, Trasplante y Bioingenierı´a (IMETTyB, CONICET, Universidad Favaloro), Laboratorio de Inmunologı´a asociada al Trasplante, Buenos Aires, Argentina * pjs@qb.ffyb.uba.ar (PJS); jschachter@qb.ffyb.uba.ar (JS) Abstract Intestinal epithelial cells play important roles in the absorption of nutrients, secretion of elec- trolytes and food digestion. The function of these cells is strongly influenced by purinergic signalling activated by extracellular ATP (eATP) and other nucleotides. The activity of sev- eral ecto-enzymes determines the dynamic regulation of eATP. In pathological contexts, eATP may act as a danger signal controlling a variety of purinergic responses aimed at defending the organism from pathogens present in the intestinal lumen. In this study, we characterized the dynamics of eATP on polarized and non-polarized Caco-2 cells. eATP was quantified by luminometry using the luciferin-luciferase reaction. Results show that non-polarized Caco-2 cells triggered a strong but transient release of intracellular ATP after hypotonic stimuli, leading to low micromolar eATP accumulation. Subsequent eATP hydrolysis mainly determined eATP decay, though this effect could be counterbalanced by eATP synthesis by ecto-kinases kinetically characterized in this study. In polarized Caco-2 cells, eATP showed a faster turnover at the apical vs the basolateral side. To quantify the extent to which different processes contribute to eATP regulation, we cre- ated a data-driven mathematical model of the metabolism of extracellular nucleotides. Model simulations showed that eATP recycling by ecto-AK is more efficient a low micromo- lar eADP concentrations and is favored by the low eADPase activity of Caco-2 cells. Simula- tions also indicated that a transient eATP increase could be observed upon the addition of non-adenine nucleotides due the high ecto-NDPK activity in these cells. Model parameters showed that ecto-kinases are asymmetrically distributed upon polarization, with the apical side having activity levels generally greater in comparison with the basolateral side or the non-polarized cells. PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 1 / 26 PLOS COMPUTATIONAL BIOLOGY eATP recycling in human epithelial intestinal cells 00125). The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Finally, experiments using human intestinal epithelial cells confirmed the presence of functional ecto-kinases promoting eATP synthesis. The adaptive value of eATP regulation and purinergic signalling in the intestine is discussed. Author summary Intestinal epithelial cells play important roles in the absorption of nutrients, secretion of electrolytes and food digestion. When intracellular ATP is released into the intestinal milieu, either at the lumen or the internal side, the resulting extracellular ATP can act as an alert signal to engage cell surface purinergic receptors that activate the immune defence of the organism against pathogens. We worked with Caco-2 and primary human intestinal cell, and our results showed that extracellular ATP regulation is a complex network of reactions that simultaneously consume or generate ATP in whole viable intestinal epithelial cells. In particular, we cre- ated a mathematical model and fitted it to experimental data allowing to quantify the degree to which intracellular ATP release and the activity of a variety of ectoenzymes con- trol the concentration of extracellular ATP. 1. Introduction The surface of the intestine is covered by a layer of cells that form the intestinal epithelium. Intestinal epithelial cells play important roles in the absorption of nutrients, secretion of elec- trolytes, digestion of food and host defence mechanisms [1,2]. The function of intestinal epi- thelial cells is strongly influenced by extracellular nucleotides, supporting a complex signalling network that mediates short-term functions such as secretion and motility, and long-term functions like proliferation and apoptosis [3,4]. Among these nucleotides, extracellular ATP (eATP) was found to be an early danger signal response to infection with enteric pathogens that eventually promote inflammation of the gut [4,5]. An important source of eATP is the intracellular ATP (iATP) found in the cytosol and vesi- cles of many cell types [6]. Activation of iATP release was found in subepithelial intestinal fibroblasts, human epithelial cell lines and enteroendocrine cells in response to several stimuli, including agents that elevate cAMP, such as forskolin and cholera toxin [7], low medium phos- phate, hypoosmotic swelling and bacterial infection [7,8]. Currently, several mechanisms have been postulated to mediate regulated iATP release, and these mechanisms can vary according to the cell type and the stimuli [6]. For example, after a hypotonic shock, Schwann cells release ATP via the anionic channel pannexin-1 [9], while the treatment with lipopolysaccharide (LPS) induces ATP release via connexin-43 in macrophages [10]. Additionally, the vesicular release pathway for ATP was also described, as in the case of endothelial cells under hypoxia [11]. Extracellular ATP and other di- and tri-phosphonucleosides can activate purinergic recep- tors 2 (P2 receptors) unevenly distributed in the small and large intestine [12]. Purinergic sig- nalling is controlled by membrane bound ecto-nucleotidases and ecto-kinases capable of promoting the synthesis and/or hydrolysis of eATP, and/or its conversion into other extracel- lular nucleotides and nucleosides. For any cell type and metabolic context, a specific set of ecto-enzymes may control the rate, amount and timing of nucleotide turnover [13]. Ecto-nucleoside triphosphate diphosphohydrolases (Ecto-NTPDases) are a family of enzymes promoting the extracellular hydrolysis of eATP, eADP, eUTP and eUDP. One or PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 2 / 26 PLOS COMPUTATIONAL BIOLOGY eATP recycling in human epithelial intestinal cells more members of this family are present in almost every cell. Ecto-NTPDase-1, -2, and -3, which differ regarding the specific preferences for nucleotides, are responsible for the hydroly- sis of nucleoside diphosphates (NDPs) and nucleoside triphosphates (NTPs) in various tissues of the gastrointestinal tract [1]. Regarding eATP and eADP hydrolysis, ecto-NTPDase-1 hydrolyses both nucleotides at similar rates, while ecto-NTPDase-2 has a high preference for eATP over eADP and ecto-NTPDase3 is a functional intermediate which preferably hydrolyses eATP [14]. The intestinal cell line HT29 cells expressed functional ecto-NTPDase-2 displaying high ecto-ATPase activity [15], while Caco-2 cells and their exosomes were reported to exhibit ecto- NTPDases-1 and -2 at the cell membrane [16,17]. Extracellular ATP can be also metabolized by ecto-kinases, with ecto-adenylate kinase (Ecto-AK) facilitating the reversible conversion of eADP to eATP and eAMP, and ecto-nucleo- side diphosphate kinase (Ecto-NDPK) promoting the exchange of terminal phosphate between extracellular NDPs and NTPs [13]. All these ecto-enzymes, if present and active, should be able to control the concentration of eATP. Up to now, although some ecto-enzymes have been identified in intestinal cells, no attempts have been made to characterize the dynamic interaction of these membrane proteins on eATP regulation of intestinal cells. In this study, we aimed to characterize iATP release and eATP recycling by ecto-enzymes, contributing to the regulation of eATP concentration ([eATP]) in Caco-2 cell line. The Caco-2 cells derive from colorectal adenocarcinoma and easily differenti- ate into cells exhibiting the morphology and function of enterocytes, the absorptive cells of the small intestine [18]. The experimental studies on eATP dynamics in polarized and non-polar- ized Caco-2 were complemented with a mathematical model quantifying the complex relation- ship among the different processes contributing to [eATP] regulation. Our results provide a quantitative description of the eATP dynamics of human intestinal epithelial cells. 2. Results In this section we show experimental results on eATP kinetics of non-polarized and polarized Caco-2 cells. To understand the dynamics of the different processes contributing to [eATP] regulation, a mathematical model was fitted to experimental data, and predictions were made. Finally, for a comparative purpose, we show results of a few experiments made on epithelial cells obtained from intestinal surgical pieces. 2.1. Non-polarized Caco-2 cells 2.1.1. eATP kinetics after hypotonic shock. The kinetics of eATP accumulation, i.e., eATP kinetics, results from the dynamic balance between iATP release mechanisms and the activities of ecto-enzymes capable of degrading and/or synthetizing eATP. As a first step towards the characterization of eATP kinetics, iATP release was triggered by exposing Caco-2 cells to hypotonic media (Fig 1A–1D). Hypotonic swelling is a stimulus that influences the uptake of nutrients by epithelial cells [19] and represents an inducer of iATP release in most cell types and tissues [8]. Under unstimulated conditions, [eATP] remained stable. Whereas addition of isotonic medium triggered a slight increase of [eATP], hypotonic media (100–180 mOsm) activated a stronger iATP release with different kinetics according to the osmotic gradient imposed (Fig 1A–1C). As shown in Fig 1D, [eATP] increased non-linearly with the magnitude of the hypo- tonic stimulus. The experimental [iATP] amounted to 1.81 mM. By comparing [iATP] with [eATP] along eATP kinetics, it was possible to estimate the energy cost of iATP release. Calculations were PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 3 / 26 PLOS COMPUTATIONAL BIOLOGY eATP recycling in human epithelial intestinal cells Fig 1. eATP kinetics of hypotonically-stimulated Caco-2 cells. The time course of [eATP] from Caco-2 cells after a hypotonic shock was quantified by luminometry performed at room temperature. (A-C) Cells were maintained in isotonic medium and, at the times indicated by the arrow, were exposed to isotonic medium (grey) or to hypotonic media (blue) of 180 mOsm (A), 150 mOsm (B) and 100 mOsm (C). Results are expressed as means of [eATP] from 4, 3 and 2 independent experiments run in triplicate for the 180, 150 a 100 mOsm experiments, respectively. (D) Increases in [eATP] from data in A-C were evaluated as ΔATP, i.e., the difference between [eATP] at 30 minutes post-stimulus and basal [eATP]. Cells were exposed to 300 mOsm (light blue bars), 180 mOsm (blue bars), 150 mOsm (dark blue bars) and 100 mOsm (grey bars). Bars show mean values + standard error of the mean (s.e.m) from 2 to 5 independent experiments. Points represent the independents values for each condition. https://doi.org/10.1371/journal.pcbi.1011196.g001 made for cells exposed to isotonic or 180 mOsm media, two conditions where no lysis was detected [17]. During the isotonic shock, representing a mechanical stimulus in the absence of osmotic gradient, [eATP] amounted to 0.33% of [iATP], while under 180 mOsm this figure amounted to 3.6%. Thus, the energy cost of eATP production by iATP efflux was very small (see section 4.9 for further details). No iADP release was detected in the 180 mOsm stimulus (S1 Fig). In our previous work, we showed that ecto-nucleotidases present in Caco-2 cells catalyse significant rates of eATP hydrolysis, leading to eADP accumulation [17]. In principle, the resulting accumulated eADP could be used by the potential presence of ecto-kinases like ecto- AK and ecto-NDPK, present in several cell types, to synthetize eATP. Thus, in the following experiments the activities of ecto-AK and ecto-NDPK were assessed by quantifying eATP kinetics under different conditions. PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 4 / 26 PLOS COMPUTATIONAL BIOLOGY eATP recycling in human epithelial intestinal cells Fig 2. Synthesis of eATP from eADP in Caco-2 cells. (A) The time course of [eATP] synthetized from exogenous eADP (6–48 μM) in the extracellular medium of intact Caco-2 cells was quantified by luminometry. The cells were incubated with the luciferin-luciferase reaction mix and the [eADP] indicated in the figure were added at the time indicated by the arrow. Data are means of at least 3 independent experiments run in duplicate for each [eADP]. (B) Effect of treatment with Ap5A (adenylate kinase inhibitor) on eATP synthesis from eADP in Caco-2 cells. The cells were treated or not (w/o treatment) with 10 μM Ap5A and the [eATP] at 30 minutes was measured by luminometry under similar conditions as experiments in (A). The bars are means ± s.e.m. from at 3–5 independent experiments run in duplicate in the absence of Ap5A and 2 independent experiments in the presence of the inhibitor. ND means that there was non-detected [eATP]. https://doi.org/10.1371/journal.pcbi.1011196.g002 2.1.2. Ecto-AK activity in Caco-2 cells. AK catalyses the following reversible reaction: 2 eADP $ eATP + eAMP and is inhibited by Ap5A [20]. Ecto-AK activity was then assessed by following eATP synthesis when Caco-2 cells were incubated with exogenous eADP (6–48 μM). Non-linear [eATP] increases were proportional to [eADP] (Fig 2A). At 30 minutes post-stimu- lus, treatment with 10 μM Ap5A, which does not permeate cells, inhibited eATP synthesis by 100% (6–24 μM eADP) or by 90% (48 μM eADP) (Fig 2B), thus showing the presence of a functional ecto-AK in Caco-2 cells membrane. 2.1.3. Ecto-NDPK activity in Caco-2 cells. NDPK catalyses the transfer of a γ–phosphate from NTP to NDP. Thus, in the presence of eADP and a given eNTP, the following reaction: eADP + eNTP $ eATP + eNDP leads to eATP synthesis when eADP is phosphorylated by NDPK. Accordingly, incubation of cells with 100 μM eCTP at different [eADP] (3–12 μM) resulted in the rapid synthesis of eATP (Fig 3A). Maximal [eATP] values were obtained 30 minutes after the addition of substrates (Fig 3A). The experiments were conducted in the presence of 10 μM Ap5A to rule out any contribution of ecto-AK to the observed eATP kinetics. Addition of 5 mM eUDP, together with 100 μM eCTP and 12 μM eADP, decreased the eATP synthesis by 91% (Fig 3B), a result compatible with high [eUDP] favouring eUDP to eUTP conversion by ecto-NDPK, rather than eATP synthesis from eADP. In separate experiments, addition of increasing [eUTP] (1–100 μM) without the addition of exogenous eADP (only endogenous eADP present), resulted in a concentration-dependent increase of [eATP] (Fig 3C). Because this increase was abolished by 5 mM eUDP (S2 Fig), we hypothesized that eATP synthesis was due to ecto-NDPK activity using exogenous eUTP and endogenous eADP. This is because there is a basal eADP concentration in the extracellular media of 0.77 ± 0.47 μM eADP/mg protein (S3 Fig). A similar experiment using 100 μM eGTP, instead of eUTP, provided qualitatively similar results (Fig 3D). Overall results showed a functional ecto-NDPK activity capable of synthetizing eATP from different γ-phosphate donors (eCTP, eUTP and eGTP) in the presence of endogenous and exogenous eADP. PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 5 / 26 PLOS COMPUTATIONAL BIOLOGY eATP recycling in human epithelial intestinal cells Fig 3. Extracellular synthesis of eATP from eADP and eCTP, eUTP and eGTP in Caco-2 cells. (A) The time course of eATP synthesis from eCTP (100 μM) and eADP (light blue for 3 μM, blue for 6 μM and grey for 12 μM) in the extracellular medium of Caco-2 cells was quantified by luminometry. Cells were incubated with the reaction mix and eCTP and eADP were added at the time indicated by the arrow. Data are means of 3 to 5 independent experiments run in duplicate for each [ADP]. (B) Production of eATP after 30 minutes exposure of Caco-2 cells to 100 μM eCTP and 12 μM eADP. Experiments were run in the presence of 5 mM eUDP (grey bar and squares) or in its absence (blue bar and points). The [eATP] was measured under conditions similar to the experiments in (A). Results are expressed as [eATP] in μM/mg of protein, bars are means ± s.e.m from 4 to 7 independent experiments run in duplicate. * means p- value <0.01 in comparison with the condition without treatment. (C) and (D) The time course of eATP accumulation in the presence of eUTP (C; grey for 100 μM, dark blue for 10 μM, blue for 1 μM) or 100 μM eGTP (D). Data are the means from 4 independent experiments in the case of 100 μM eUTP, 3 in the case of 100 μM eGTP, and 2 independent experiments in the case of 10 μM or 1 μM eUTP. Nucleotides were added at the time indicated by the arrow. https://doi.org/10.1371/journal.pcbi.1011196.g003 2.1.4. Modelling eATP kinetics of non-polarized Caco-2 cells. Caco-2 cells regulate eATP kinetics by iATP release, eATP synthesis by the activities of ecto-AK and ecto-NDPK (as shown in this study), and hydrolysis by ecto-nucleotidases [17]. These processes are active simultaneously when measuring the eATP dynamics in Caco-2 cells, except when a specific inhibitor was added (like Ap5A in the experiments of Fig 3A and 3B). Thus, to quantify the individual contribution of these processes to eATP kinetics, we built a mathematical model that was then fitted to experimental data. Model parameters contain the kinetic information of each enzyme, allowing to assess the individual contribution of ecto-enzymes to eATP dynamics. A scheme of the model is depicted in Fig 4A. In the model, [eATP] can increase by iATP release, by lytic and by non-lytic mechanisms. In addition, [eATP] can be modulated by the activities of ecto-ATPases, ecto-AK and ecto-NDPK. PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 6 / 26 PLOS COMPUTATIONAL BIOLOGY eATP recycling in human epithelial intestinal cells Fig 4. A model of extracellular purinergic regulation in non-polarized Caco- 2 cells. (A) The scheme shows a representation of the model created to explain the experimental results in non-polarized cells. The yellow bolt indicates that the JATP depended on the application of a hypotonic shock. ADO means extracellular adenosine. The green star behind “eATP” indicates that this is the metabolite measured directly during experiments. (B) The plot shows, in red, the model fitting to eATP kinetics exposed to media of different osmolarities (experimental data correspond to those shown in Fig 1A and 1C). (C) iATP efflux (JATP) predicted by the model upon the hypotonic or isotonic shocks indicated in the figure. https://doi.org/10.1371/journal.pcbi.1011196.g004 The model provides functions describing each of the fluxes involved in transport and metabolism of extracellular nucleotides (see S1 Table and section 4.13.1). Fitting the model to the experimental eATP kinetics under the different conditions allowed to obtain the best-fit values for the parameters of these functions (S1 Table). In that way the contributions of each flux to eATP kinetics were quantified, and several predictions were made. 2.1.4.1. iATP release. For experiments under iso- and hypotonic media, the model found a good fit to experimental data (continuous lines in Fig 4B), thus allowing to predict the rate of iATP efflux (JATP) over time (Fig 4C). JATP was rapid and transient in nature, leading to a 12-fold increase of [eATP] to a maximum in less than 2 seconds under the 180 mOsm shock, followed by rapid inactivation. The magnitude of the JATP peak depended on the osmotic gra- dient imposed. Inactivation of JATP was observed under conditions where no lysis was detected (isotonic and 180 mOsm media). On the other hand, a lytic flux (JL) explains the continuous increase of [eATP] at 100 mOsm (Figs 1C and 4B). 2.1.4.2. Ecto-enzymes. Another factor shaping eATP kinetics is eATP hydrolysis by ecto- ATPase activity. We have previously observed that, in intact non-polarized Caco-2 cells, ecto- ATPase activity follows a linear function of micromolar [eATP] [17]. Thus, following a stimu- lus promoting iATP release, any increase of [eATP] should be at least partially counterbal- anced by an increase of ecto-ATPase activity. Model predictions made at 180 mOsm show that the initial peak of [eATP] increase due to JATP is about 8-fold higher than the rate of eATP hydrolysis, i.e., JATP was 1.2 μM iATP/min/ mg of protein (Fig 4C) and eATP hydrolysis was 0.15 μM eATP/min/mg of protein at 1.5 μM eATP (S1 Table). Thus, during the first seconds of [eATP] increase, eATP kinetics was mainly governed by iATP release. At later times, however, the JATP inactivated, and the ecto-ATPase activity progressively gained importance in controlling [eATP]. This is illustrated by modelling a change in the amount of ecto-ATPase over a wide range, showing that a 5-fold increase of ecto-ATPase activity could lead to rapid decay of [eATP], while a 5-fold decrease would pro- long high levels of [eATP] over the entire incubation period (Fig 5A). However, a similar pro- cedure, i.e, increasing or decreasing 5 times the activity of ecto-AK, had no influence on the PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 7 / 26 PLOS COMPUTATIONAL BIOLOGY eATP recycling in human epithelial intestinal cells Fig 5. Role of ecto-AK, ecto-ATPase and ecto-ADPase activity on eATP dynamics. (A) The simulation shows the [eATP] as a function of time upon a 180 mOsm hypotonic shock when the ecto-ATPase activity displayed its measured value (0.1, continuous line in blue), a 5-fold increase (0.5, dashed line in grey), and a 5-fold decrease (0.02, dashed line in light blue). The numbers in the plot indicate the kinetic constant of the activity in (μM eATP hydrolized)/(mg prot/ μM eATP/ min) units. (B) The simulation shows the [eATP] as a function of time upon a 180 mOsm shock at different initial [eADP] concentrations: calculated pre-stimulus eADP (0.22 μM continuous line in blue), a 3.5-fold increase (0.8 μM, dashed line in grey), and a 14-fold increase (3 μM, dashed line in light blue). (C) The simulation shows the [eATP] as a function of time upon addition of 6 μM eADP (the corresponding experimental results are shown in Fig 2A). The plot shows the eATP kinetics under various values of the kinetic constant for ecto-ADPase, i.e, the constant experimentally determined (0.008, continuous line in blue), a 5-fold increase (0.04, dashed line in grey), and a 12-fold increase (0.96, dashed line in light blue). The numbers in the plot indicate the kinetic constant of the activity in (μM eADP hydrolized)/(mg prot / μM eADP / min) units. (D) Ecto-ATPase, ecto-AK and ecto-ADPase activites as a function of their respective substrates, [eATP] for ecto-ATPase and [eADP] for ecto-AK and ecto-ADPase. The points show the initial velocities for eATP synthesis as a function of [eADP] by ecto-AK calculated from experimental data shown in Fig 2A. The points are means ± s.e.m. from 3 to 5 independent experiments run in duplicate. The continuous lines represent enzyme activities as a function of their respective substrates (see S1 Table for further details). Shadows behind the lines in panels A, B and C represent the uncertainty of the prediction calculated as indicated in section 4.13. Shadows behind the lines in panel D represent the interval ± the standard error of enzyme activity, calculated using the standard error of kinetic parameters of ecto-ATPase and ecto-ADPase activities. https://doi.org/10.1371/journal.pcbi.1011196.g005 [eATP] during the hypotonic shock (not shown). This can be attributed to the sigmoidal kinet- ics of ecto-AK, whose activity is very low below 3 μM eADP, but significantly higher above that concentration (Fig 5D). Thus, ecto-AK might influence eATP kinetics only when [eADP] is sufficiently high. Fig 5B shows a simulation where the initial [eADP] was raised up to 3 μM. At 3 μM eADP, eATP degradation was comparable to eATP synthesis by ecto-AK, indicating that ecto-AK can counterbalance ecto-ATPase activity. Note that a 180 mOsm hypoosmotic PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 8 / 26 PLOS COMPUTATIONAL BIOLOGY eATP recycling in human epithelial intestinal cells medium does not change the activity of ecto-ATPase, ecto-ADPase, ecto-AMPase [17] or ecto- AK [21] in comparison with isosmotic medium. Another factor to consider is ecto-ADPase activity. We have previously shown that Caco-2 cells displays high ecto-ATPase but very low ecto-ADPase activity [17]. Nevertheless, a hypo- thetical increase of ecto-ADPase activity could negatively modulate ecto-AK activity. For example, an increase of 5- and 12-fold of ecto-ADPase activity would result in a 17% and 33% decrease in the [eATP] production respectively, at 6 μM eADP (Fig 5C). Model predictions showed above implied that the expression of ecto-AK in Caco-2 cells may have an important role in eATP kinetics. To assess this hypothesis, we compared ecto- ATPase, ecto-ADPase and ecto-AK activities as a function of their respective substrate’s con- centrations, that is, [eATP] for ecto-ATPase and [eADP] for ecto-ADPase and ecto-AK (Fig 5D). In Fig 5D, symbols of ecto-AK activities represent the initial velocities for eATP synthesis as a function of [eADP] calculated from experimental data shown in Fig 2A, and the continu- ous line represents the fit to data of the ecto-AK function included in the model (details in S1 Table and in the work of Sheng and collaborators [22]). The ecto-ATPase and ecto-ADPase activities are predictions made from data of our previous work [17]. Ecto-ATPase displayed the highest rate of the three reactions. On the other hand, although at low [eADP], ecto-AK and ecto-ADPase activities are similar and have relatively low values, the sigmoidal kinetics of ecto-AK allows a strong activity increase as [eADP] is raised, thus reaching activity levels well above those of ecto-ADPase activity (Fig 5D). Finally, in the presence of non-adenosine nucleotides, the influence of ecto-NDPK on eATP dynamics was assessed and analysed. Caco-2 cells synthetized eATP by ecto-NDPK activity in the presence of eCTP, eUTP and eGTP as NTP donors, and exogenous and endoge- nous eADP (Fig 3A–3D). The model found a good fit to the experimental eATP kinetics in the presence of 100 μM eCTP and different [eADP] (Fig 6A). Model predictions of ecto-NDPK activity at different [eADP] agreed well with initial velocities of experimental ecto-NDPK activities shown in Fig 3A (Fig 6B). We also studied the effect of eUTP addition without the addition of exogenous eADP (a condition where only endogenous eADP was present, S3 Fig) on the transient rise of [eATP] (Fig 3C, replicated in Fig 6C). To understand the role of eNTPs on ecto-NDPK activity, it is important to recall that ecto-NTPDases of Caco-2 cells can hydrolyse non-adenine nucleotides (S4 Fig). Model predictions show changes in ecto-NDPK and ecto-ATPase activities (Fig 6D), and the corresponding dynamics of [eATP] and [eADP] (Fig 6E), and of [eUTP] and [eUDP] (Fig 6F). Kinetics of eATP (Fig 6C and 6E) could be analysed in 3 stages. First, [eATP] increases due to a high an ecto-NDPK/ecto-ATPase activities ratio in the presence of high [eUTP] and basal [eADP] (stage 1 in Fig 6D, 6E and 6F). The resulting elevated [eATP] activates ecto- ATPase activity, while ecto-NDPK decreases deeply because its substrates (eUTP and eADP) are consumed by ecto-NTPase activity and by ecto-NDPK activity itself. A balance is then estab- lished between ecto-NDPK and ecto-ATPase activities in stage 2, where [eATP] is transiently stable. Finally in stage 3, [eUTP] continues decreasing, leading to a high ecto-ATPase/ecto- NDPK activities ratio, causing [eATP] to decrease and [eADP] to rise again (Fig 6E). 2.2. eATP regulation in polarized Caco-2 cells Because several reports showed differential activities of enzymes and transporters at each side of polarized epithelia [23,24], we speculated that [eATP] regulation might be different at the apical and basolateral sides of polarized Caco-2 monolayers. We then used polarized Caco-2 cells to test the effect of hypotonic shock on iATP release and resulting eATP kinetics at the apical and basolateral sides. Similarly to the procedure PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 9 / 26 PLOS COMPUTATIONAL BIOLOGY eATP recycling in human epithelial intestinal cells Fig 6. Role of ecto-NDPK on eATP dynamics. (A) The plot shows the experimental results of eATP dynamics in the presence of 100 μM eCTP and various [eADP] (also shown in Fig 3A). Model fitting was applied to data and shown as continuous red lines. (B) The plot shows the ecto-NDPK activity expressed in μM of ATP synthetized per minute per mg of protein. The dots represent the initial velocity of ecto-NDPK (calculated from the experimental data shown in panel A) as a function of [eADP]. Points represent the means ± s.e.m. from 3 independent experiments run in duplicate. The continuous line represents the ecto-NDPK activity predicted by the model (details in S1 Table). (C) The plot shows the experimental eATP dynamics in the presence of various [eUTP] (also shown in Fig 3C) and the continuous red lines represent the model fitting. (D) The plot shows time changes of ecto-NDPK (blue line) and ecto- ATPase (grey line) activities predicted by the model. In the plot, the zones 1 (white background), 2 (pink background) and 3 (white background) represents the [eATP], increase, stabilization and decrease stages respectively. In (E) and (F) the plot shows the model predictions of [eATP] and [eADP], or [eUTP] and [eUDP] respectively as a function of time upon addition of 100 μM eUTP to non-polarized Caco-2 cells. Data is expressed in μM/mg protein, which was calculated by dividing the [eATP] at any time by the average protein mass in the experiments (0.25 mg in average). The shadows behind the lines in panels A, B and C represent the uncertainty of the prediction calculated as indicated in section 4.13. https://doi.org/10.1371/journal.pcbi.1011196.g006 employed for non-polarized cells, we fitted the model shown in Fig 4A to the experimental data to understand quantitatively the mechanisms involved in [eATP] regulation in differenti- ated monolayers of Caco-2 cells. Experimental results show that, following a 180 mOsm hypotonic shock, [eATP] increased at both sides of the monolayers, with qualitatively different kinetics. While at both sides the initial rate of [eATP] increase was fast, apical eATP kinetics achieved a maximum at 1.5 min- utes, followed by a rapid decay. This was not observed in the basolateral domain, where [eATP] continued increasing at a progressively lower rate, and a very slow [eATP] decay was observed only after 20 minutes (Fig 7A). The two different eATP kinetics suggested different activities of ecto-enzymes present at both sides of the monolayers. Therefore, we determined the activities of ecto-ATPase, ecto-AK and ecto-NDPK. For assessing ecto-ATPase activity, polarized Caco-2 cells were exposed to various [eATP] (0.2–7 μM) and eATP hydrolysis was estimated by quantifying [eATP] decay rates (S5 Fig). PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 10 / 26 PLOS COMPUTATIONAL BIOLOGY eATP recycling in human epithelial intestinal cells Fig 7. Apical and basolateral eATP regulation in Caco-2 monolayers. (A) Effect of hypotonic shock on eATP kinetics. At the times indicated by the arrow, cells were exposed to 180 mOsm medium on the basolateral (grey) or the apical (blue) compartments. Data are the means from 5 independent experiments. (B) Ecto-ATPase activity measured from the eATP kinetics at different [eATP]. Data was obtained from eATP hydrolysis kinetics like the ones shown in S5A and S5B Fig. The points in the plot represent the mean ± s.e.m. of 3 independent experiments. The dashed lines represent a linear regression to the data allowing to obtain the ecto-ATPase kinetic constant which was 1.70 ± 0.08 and 0.36 ± 0.22 mM ATP hydrolized mM ATP mg prot min basolateral compartments respectively. (C) Ecto-AK activity. eATP kinetics in the presence of 12 μM eADP added to the basolateral (grey) or apical (blue) compartments. Data are the means from 2 independent experiments. (D) Ecto-NDPK activity. eATP kinetics in the presence of 100 μM eCTP + 12 μM eADP added to the basolateral (grey) or the apical (blue) compartments. Experiments were run in the presence of 10 μM Ap5A (adenylate kinase blocker). Data are the means from 3 independent experiments. (E) Ecto-AK initial velocities in polarized Caco-2 cells. Data are means + s.e.m. of 4 independent experiments. * indicates a p-value < 0.05 in comparison with the apical condition. (F) Ecto-NDPK initial velocities in polarized Caco-2 cells. Data are means ± s.e.m. of 3 independent experiments. (G) Scheme of the results interpretation showing that the increased activity of Ecto-AK, Ecto-NDPK and Ecto-ATPase leads to a faster eATP turnover. for the apical and https://doi.org/10.1371/journal.pcbi.1011196.g007 PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 11 / 26 PLOS COMPUTATIONAL BIOLOGY eATP recycling in human epithelial intestinal cells The initial rate values of [eATP] decay were used to calculate ecto-ATPase activity at each [eATP], so as to build a substrate curve (Fig 7B). Linear fitting to experimental data showed that ecto-ATPase activity was 4-fold higher in the apical than in the basolateral domain. To assess ecto-AK activity, Caco-2 cells were exposed to 12 μM eADP at the basolateral or apical domains. In the apical domain, [eATP] increased rapidly to a maximum, followed by a rapid decay towards pre-stimulated levels, while basolateral [eATP] increased steadily at a lower rate (Fig 7C). Initial velocity estimations showed that ecto-AK activity was significantly higher in the apical than in the basolateral compartment (Fig 7E). Ecto-NDPK activity was quantified using polarized cells exposed to 100 μM eCTP plus 12 μM eADP in the basal and apical domains. Experiments were run in the presence of 10 μM Ap5A to block ecto-AK activity. Production of eATP by ecto-NDPK was much higher than that observed under conditions used to measure ecto-AK activity, though the domain specific pattern of eATP kinetics was similar when assessing the two ecto-kinases, i.e., a biphasic pat- tern in the apical domain, and a steady [eATP] increase, at a lower rate, in the basal domain (Fig 7D). The initial velocity of ecto-NDPK was higher in the apical than in the basolateral domain although differences were not significant (Fig 7F, p value = 0.1). A good fitting of the model to all experimental data was achieved (red lines in Fig 7A, 7C and 7D). The model fitting allowed to obtain the ecto-NDPK and ecto-AK maximal velocity (Vmax) and compared them with the ones obtained from non-polarized cells (S6A and S6B Fig and S2 Table). Results indicated that the ecto-NDPK maximal activity in the apical compart- ment is a slightly higher than that of the non-polarized cells and significantly higher than that of the basolateral compartment. On the other hand, the ecto-AK maximal activity is signifi- cantly higher compared with the basolateral compartment or the non-polarized cells. Thus, the differences between the apical and basolateral eATP dynamics can be explained by an increase or decrease in ecto-enzymes activities. Altogether experimental results showed significantly higher activities of the ecto-enzymes (ecto-ATPase, ecto-AK and ecto-NPDK) in the apical, as compared to the basolateral domain (Fig 7G). Fig 8. Ecto-AK and ecto-NDPK activities in IECs. Time course of eATP synthetized from exogenous eADP (A) or eCTP and eADP (B) in the extracellular medium of IECs. (A) The cells were incubated with the luciferase-luciferin reaction mix and 12 μM eADP was added at the time indicated by the arrow in presence (grey) or absence (blue) of 10 μM Ap5A (B) 100 μM eCTP plus 12 μM eADP, in the presence of 10 μM Ap5A were added at the time indicated by the arrow. [eATP] was quantified by luminometry. Values are the means of 3 independent experiments run in duplicate. https://doi.org/10.1371/journal.pcbi.1011196.g008 PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 12 / 26 PLOS COMPUTATIONAL BIOLOGY eATP recycling in human epithelial intestinal cells 2.3. Ecto-AK and ecto-NDPK are active in human primary small intestinal epithelial cells Having characterized ecto-AK and ecto-NDPK activities of Caco-2 cells, we wondered whether these ecto-enzymes would be functional in IECs extracted from human small intes- tine. Accordingly, we used samples obtained from small intestine biopsies from healthy donors (Fig 8A and 8B). Results show that exposure of IECs to both 12 μM eADP (to assess ecto-AK) (Fig 8A) or to 12 μM eADP + 100 μM eCTP (to assess ecto-NDPK, Fig 8B) led to significant eATP produc- tion in the micromolar range. Furthermore, as expected, the presence of ApA5 totally inhibited the ecto-AK activity (Fig 8A). 3. Discussion Intestinal epithelial cells can release iATP and express several ecto-enzymes capable of regu- lating the amount and metabolism of eATP at the cell surface. The main goal of this study was to characterize quantitatively the dynamic interplay of iATP release, eATP hydrolysis and eATP synthesis contributing to the dynamic regulation of [eATP] in Caco-2 cells. Spe- cial emphasis was given to the role of ecto-kinases promoting eATP production under differ- ent conditions. Since Caco-2 cells undergo spontaneous enterocytic differentiation in culture, we decided to first approach the complexity of eATP regulation using the relatively simpler non-polarized cell model, and later extend the study to fully differentiated cells. These form apical and baso- lateral poles where morphological and biochemical features are segregated [23]. When exposed to hypotonicity, non-polarized Caco-2 cells triggered a strong iATP efflux that rapidly inactivated, leading to low μM [eATP] accumulation. A number of studies have confirmed that such micromolar [eATP] are capable of activating P2 receptors with high affin- ity for that nucleotide, such as P2Y2, P2Y11 and almost all P2X receptors [25]. In Caco-2 cells, eATP dose dependently activates P2Y receptors involved in the activation of MAPK cascades and transcription factors that promote cell proliferation [26,27], while higher [eATP] can induce apoptosis via P2X7 receptor [3]. In principle, purinergic activation by eATP should be transient, due to the presence of ecto- nucleotidases, the activities of which promotes strong eATP hydrolysis in Caco-2 cells [17]. Accordingly, our results show that hypotonicity induced iATP release and concomitant eATP accumulation, where [eATP] decay was accelerated by constitutive ecto-ATPase activity. This decay was even higher for a model predicted upregulation of eATP hydrolysis by one or more ecto-nucleotidases, as occurs in various cells and tissues during pathogen infection [28], cell differentiation [29] or tumorigenesis [30]. The above results imply that iATP release and eATP hydrolysis constitute two opposing fluxes shaping eATP kinetics of Caco-2 cells. However, the presence of ecto-kinases found in this study suggest that the dynamic regulation of [eATP] should also take the activities of these enzymes into account. In this respect, addition of exogenous eADP to Caco-2 cells dose dependently increased [eATP]. The fact that eATP synthesis was almost fully blunted by Ap5A, an AK blocker that does not permeate intact cells, suggested the presence of a functional ecto-AK. Results of the mathematical model allowed to envisage the contribution of ecto-AK to eATP kinetics. In the absence of exogenous eADP, the contribution of ecto-AK to eATP kinetics was negligible, so that [eATP] depended mainly on the balance between the rates of iATP release and eATP hydrolysis. This is due to the low endogenous [eADP] present under the experimental condi- tions. However, due to the sigmoidal nature of the AK reaction, model predictions show that PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 13 / 26 PLOS COMPUTATIONAL BIOLOGY eATP recycling in human epithelial intestinal cells increasing [eADP] in the low micromolar range, suffices to promote significant eATP synthe- sis by ecto-AK, upregulating eATP kinetics. Thus, under certain conditions, e.g., when cell leak intracellular ADP (iADP) or eADP is supplied paracrinally by other cell types, eATP synthesis by ecto-AK of Caco-2 cells will transiently stabilize [eATP] levels, thereby favouring propaga- tion of eATP-dependent purinergic signalling. A similar stabilizing role of ecto-AK on [eATP] has been proposed for HT29 cells, lung epithelial cells and lymphocytes [14,15,31]. Modelling shows that ecto-ADPase activity, which facilitates eADP degradation, may com- pete with ecto-AK for the available eADP. However, Caco-2 cells -as HT29 cells [15]- displayed a relative low ecto-ADPase activity, in agreement with the presence of a functional ecto- NTPDase 2 in both cell types [15,17], and in addition the intrinsic sigmoidal nature of ecto- AK activity makes ecto-AK more sensitive to [eADP] than ecto-ADPase. Another consequence of ecto-AK activation relates to P1 signalling, since activity of this enzyme will provide eAMP from eADP for further hydrolysis to adenosine by ecto-5’NT present in Caco-2 cells [17], finally leading to extracellular adenosine accumulation. Our model predictions show how increasing [eADP] in the low μM range might lead to substantial adenosine accumulation, which may engage 4 different P1 receptors [32]. The con- sequences of P1 signalling on proliferation of Caco-2 cells and several other intestinal epithelial cell lines have been studied before [33]. In general, the balance between P1 and P2 receptors on epithelial cells regulate intestinal secretion [34–37] and absorption [38,39]; responses trig- gered by the P2 receptor stimulation by eATP and other nucleotides are sometimes counter- acted by P1 receptor stimulation by adenosine, though the potential role of ecto-AK was not considered in this context. Another factor affecting eATP kinetics is ecto-NDPK. Activity of this enzyme was detected in many cells and tissues such as astrocytoma cells [40], endothelial cells [41,42], lymphocytes [41], keratinocytes [43] and hepatocytes [44]. In general, ecto-NDPK will pri- marily serve to transfer phosphate groups between different extracellular nucleotides and thus potentially alter the pattern of P2 receptor activation. This is especially important since P2 receptor subtypes are differentially selective for adenine and uridine eNDPs and eNTPs [45,46]. Our results show that ecto-NDPK can use eCTP, eGTP and eUTP to phosphorylate eADP to eATP. As model predictions show, activities of ecto-NDPK (promoting eATP synthesis from eUTP and eADP) and ecto-nucleotidase (promoting eATP and eUTP hydrolysis) change in opposite directions to transiently stabilize [eATP]. Results analysed above show that, in non-polarized Caco-2 cells, [eATP] can increase by iATP release and ecto-kinase mediated eATP synthesis and decrease by ecto-nucleotidases mediated by eATP hydrolysis. Next, we studied eATP dynamics of polarized Caco-2 cells. These cells differentiate sponta- neously into polarized cells, with apical and basolateral domains exhibiting morphological and biochemical features of small intestine enterocytes [23,47]. In particular, the Caco-2 polarized phenotype is characterized by high levels of hydrolases typically associated with the brush bor- der membrane. The fact that in a variety of epithelia several ecto-nucleotidases and ecto-phos- phatases preferentially -but not exclusively- locate in the apical domain [48–50], anticipated a different eATP regulation at both poles of Caco-2 cells. Accordingly, hypotonically induced eATP kinetics had a faster resolution and was more effectively regulated at the apical, than at the basolateral side, a result in agreement with the observed higher apical (than basolateral) ecto-ATPase activity measured in this study. This is in agreement with several reports using intestinal epithelial cell from murine models and human intestinal cell lines, showing that various isoforms of ecto-NTPDases, ecto-phospha- tases and ecto-NPPases are preferentially located in the apical domain [50]. PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 14 / 26 PLOS COMPUTATIONAL BIOLOGY eATP recycling in human epithelial intestinal cells The mechanism mediating iATP release at both sides of the polarized Caco-2 cell mono- layer requires further investigation. We previously showed that iATP release in Caco-2 cells challenged by adrenergic stimulation or the presence of bacteria is reduced by treatment with carbenoxolone, a blocker of conductive iATP efflux [51]. Since the molecular mechanisms involved in iATP release may depend on the stimulus, further investigations will be conducted to clarify this topic. A qualitatively similar pattern was observed for ecto-AK and ecto-NDPK of Caco-2 cells, in that apical activities were much higher. Interestingly, the model describing eATP dynamics of non-polarized cells could be successfully fitted to eATP kinetics on each of the polarized domains, thus allowing to calculate the consequences of ectoenzymes sorting on eATP regulation. The results obtained with primary human IECs suggest that our model could be employed to explain the eATP kinetics in these cells, given the presence of both ecto-AK and ecto-NDPK. However, further studies would be required to adapt the Caco-2 model to IECs. These studies may need to account for bacterial periplasmic ATPases capable of hydrolyzing eATP, as we and others have shown that commensal bacteria can express periplasmic nucleotidases and release ATP into the intestinal lumen [51,52]. In this context, the model presented here may be taken as a starting point to progressively add other processes affecting [eATP] regulation in vivo. The fact that the apical domain exhibited a higher turnover of extracellular nucleotides, leading to higher eATP regulation may have adaptive value, considering that iATP release is a common response of epithelial intestinal cells to enteric pathogens [53]. Extracellular ATP may then act as a danger signal controlling a variety of purinergic responses aimed at defend- ing the organism from a variety of pathogens and their toxins present in the intestinal lumen. 4. Materials and methods 4.1. Ethics statement The protocol for handling human samples was approved by the Institutional Review Board of the Favaloro Foundation University Hospital (DDI [1587] 0621) and has been performed in accordance with the ethical standards laid down in the declarations of Helsinki and Istanbul. Informed written consent was obtained from donors. 4.2. Chemicals All reagents were of analytical grade. Bovine serum albumin (BSA), malachite green, adeno- sine 50 -triphosphate (ATP), adenosine 50 -diphosphate (ADP), cytidine 50-triphosphate diso- dium salt (CTP), adenosine 50 -monophosphate (AMP), uridine 5’-triphosphate (UTP), uridine 5’-diphosphate (UDP), guanosine-5’-triphosphate (GTP), phosphate-buffered saline (DPBS), 4-(2-hydroxyethyl)-1-piperazineetahnesulfonic acid (HEPES), ammonium molyb- date, Triton X-100, phenylmethylsulphonyl fluoride (PMSF), pyruvate kinase, phosphoenol- pyruvate (PEP), luciferase, coenzyme A and P1,P5-Di (adenosine-5´) pentaphosphate pentaso- dium salt (Ap5A) were purchased from Sigma-Aldrich (St Louis, MO, USA). D-luciferin was purchased from Molecular Probes Inc. (Eugene, OR, USA). 4.3. Solutions In the experiments to measure [eATP] by luminometry (section 4.5), cells were incubated with isotonic buffer called isosmotic DPBS (300 mOsm) containing: 137 mM NaCl, 2.7 mM KCl, 1 mM CaCl2, 2 mM MgCl2, 1.5 mM KH2PO4 and 8 mM Na2HPO4, pH 7.4 at 37˚C (assay medium). When applying a hypotonic shock to cells, the medium was changed for other con- taining the same components but with a lower NaCl concentration. Thus, DPBS with 100, 150 PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 15 / 26 PLOS COMPUTATIONAL BIOLOGY eATP recycling in human epithelial intestinal cells and 180 mOsm were prepared. The osmolarity of all media was measured with a vapor pres- sure osmometer (5100B, Lugan, USA). When measuring phosphate (section 4.8.4) the following medium without phosphate was employed instead of isotonic buffer: 145 mM NaCl, 5 mM KCl, 1 mM CaCl2, 10 mM HEPES, and 1 mM MgCl2, pH 7.4 at 37˚C. 4.4. Caco-2 cell culture Caco-2 cells (ATCC, Molsheim, France) were grown in Dulbecco’s modified Eagle’s medium (DMEM-F12, Gibco, Grad Island, NY, USA) containing 4.5 g/L glucose (Sigma-Aldrich, St Louis, MO, USA) supplemented with 10% v/v fetal bovine serum (Natocor, Co´rdoba, Argen- tina), 2 mM L-glutamine (Sigma-Aldrich, St Louis, MO, USA), 100 U/mL penicillin, 100 μg/ mL streptomycin and 0.25 μg/mL fungizone (Invitrogen, Carlsbad, CA, USA) in a humidified atmosphere of 5% CO2 at 37˚C. For eATP kinetics measurements cells were directly seeded on glass coverslips. For ecto-nucleotidase activity experiments using the malachite green method, cells were seeded in cell culture 24-well plates (Corning Costar, NY, USA). 4.4.1. Polarisation of Caco-2 cells. For preparation of polarized Caco-2 monolayers, cells were seeded in permeable supports (inserts) made of polyester (Transwell; 0.1 μm pore size, 1.12 cm2 cell growth area; Jet Biofil, China) in 12-well plates at a density of 3 × 104 cells/0.5 mL per insert. The medium was changed after 3 days, and then after every 3 or 4 days. The polar- ized Caco-2 monolayers were used for experiments after the transepithelial electrical resistance reached a plateau (approximately 21 days after seeding). In polarized and non-polarized cul- tures contamination (including Mycoplasma) was routinely tested. 4.5. Human Intestinal Epithelial Cells (IECs) isolation IECs were isolated from ileum biopsies collected from healthy volunteers who were endoscopi- cally evaluated for colon cancer (N = 3) at the Favaloro Foundation University Hospital. Sam- ples of non-tumoral, non-injured intestinal biopsies were collected and transported in ice-cold Hanks’s balanced salt solution (HBSS) for immediate processing. The biopsies were incubated in 5 mM ethylenediaminetetra-acetic acid (EDTA) and 1.5 mM dithiothreitol HBSS with agita- tion for 25–30 minutes at room temperature to obtain IECs. Cells were pelleted, re-suspended in DPBS and used immediately. 4.6. ATP measurements The [eATP] of non-polarized Caco-2, polarized Caco-2 monolayers or IECs was measured using the firefly luciferase reaction (EC 1.13.12.7, Sigma-Aldrich, St Louis, MO, USA), which catalyses the oxidation of D-luciferin in the presence of ATP to produce light [54]. As described below, using this method it was possible to determine eATP kinetics, the iATP con- tent and the activities of ecto-enzymes. Before the experiments, the cells were washed two times with the assay medium (isosmotic DPBS with or without Pi). In this work, the cells’ medium was substituted by the assay medium before any measure- ment, therefore exoenzymes (enzymes released to extracellular medium not bound to the membrane) were removed and only ecto-enzymes (membrane bound extracellular enzymes) were investigated. 4.7. eATP kinetics of non-polarized Caco-2 and IECs Non-polarized Caco-2 cells and IECs were seeded on glass coverslips. Under all conditions cells were mounted in the assay chamber of a custom-built luminometer, as previously PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 16 / 26 PLOS COMPUTATIONAL BIOLOGY eATP recycling in human epithelial intestinal cells described [55]. Because luciferase activity at 37˚C is only 10% of that observed at 20˚C [56], to maintain full luciferase activity, [eATP] measurements were performed at room temperature. The setup allowed continuous measurements of [eATP] by the luciferin-luciferase reaction. A calibration curve was used to transform the time course of light emission into [eATP] versus time. Increasing [eATP] from 13 to 1000 nM were sequentially added to the assay medium from a stock solution of pure ATP dissolved in isosmotic or hypotonic medium, according to the experiment. Calibration curves displayed a linear relationship within the range tested. After each experiment, cells were lysed with a solution containing 1 mM PMSF and 0.1% of Triton X-100 and the protein contents of each sample were quantified [57]. Results were expressed as [eATP] at every time point of a kinetics curve denoted as “eATP kinetics”, with [eATP] expressed as μM of eATP/mg protein in a final assay volume of 100 μL. 4.8. eATP kinetics of polarized monolayers Polarized Caco-2 cells monolayers were placed in the insert physically separating an apical and a basolateral compartment. Detection of eATP was performed separately on either side, by adding the luciferin-luciferase mixture in one compartment (apical or basolateral) and adding isosmotic DPBS to the other side. In preliminary experiments, we observed that the luciferin- luciferase mix added in one compartment did not cross the monolayer into the other compart- ment. Thus, luminescence registered when measuring the [eATP] in one compartment was not contaminated by light from the other compartment due to luciferin-luciferase leakage. When an hypoosmotic shock was applied, a luciferin-luciferase mix in DPBS with an osmo- larity of 180 mOsm was added to the compartment of interest while, isosmotic DPBS was added to the other side. 4.9. Activities of ecto-enzymes Ecto-ATPase, ecto-AK and ecto-NDPK activities of intact cells were measured by luminome- try (section 4.5). Ecto-nucleotidase activities were measured by measuring the inorganic phos- phate (Pi) release. 4.9.1. Ecto-ATPase activity Cells were exposed to different [eATP] (0.2, 1.2, 4.2 or 7 μM). Following an acute increase of [eATP], ecto-ATPase activity was estimated from the initial velocity of eATP decay at each [eATP]. 4.9.2. Ecto-AK activity. Cells were exposed to different [eADP] (6, 12, 24 or 48 μM) and the eATP kinetics was quantified in the absence and presence of 10 μM Ap5A (an AK blocker). Ecto-AK initial velocity was estimated as indicated in section 4.12. 4.9.3. Ecto-NDPK activity. Cells were exposed to different [eADP] (3, 6 or 12 μM) in the presence of eCTP (100 μM), eGTP (100 μM) or eUTP (1, 10 or 100 μM). Then, the eATP kinet- ics was quantified in the presence of Ap5A to block the eADP to eATP conversion by ecto-AK activity. In some experiments 5 mM eUDP was added to inhibit ecto-NDPK activity. Ecto- NDPK initial velocity was estimated as indicated in section 4.12. 4.9.4. Ecto-NTPDase activities. Cells were incubated with 500 μM of eCTP, eUTP or eGTP at 37˚C. Samples were taken at 30, 60, 90 and 120 minutes after nucleotides addition and, the inorganic phosphate concentration was measured by the malachite green method [17,58]. Activities measured in section 4.8.1 were expressed as μM of eATP hydrolysed per minute, normalized by the cell protein mass in the experimental sample (μM of eATP /mg protein/ min). Results from experiments explained in sections 4.8.2 and 4.8.3 were expressed as μM of PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 17 / 26 PLOS COMPUTATIONAL BIOLOGY eATP recycling in human epithelial intestinal cells eATP synthetized per minute, normalized by the cell protein mass in the experimental sample (μM of eATP /mg protein/min). Activities measured in section 4.8.4 were expressed as μM of inorganic phosphate released per minute, normalized by the cell protein mass in the experi- mental sample (μM of Pi /mg protein/min). 4.10. Intracellular ATP measurements Caco-2 (0–30,000 cells) were laid on coverslips, incubated with 45 μL of luciferin-luciferase reaction mix for 5 minutes and subsequently permeabilized with digitonin (1.6 mg/mL final concentration). Light emission was transformed into [eATP] as a function of time as indicated in section 4.6. After considering the total volume occupied by Caco-2 present in the chamber, and the relative solvent cell volume (3.66 μl per mg of protein) [59], [iATP] was calculated in mM. To calculate the % of iATP release, the following equation was employed: %iATP ¼ 100 x ATPcell ð1Þ where ATPcell represents the [ATP] obtained when iATP from all cells is released into the assay medium. The "x" denotes the [eATP] measured at any time. The value of ATPcell was 66 μM/mg protein and was calculated by multiplying the [iATP] (1.8 mM, section 2.1.1) by the Caco-2 cell volume (3.66 μl per mg of protein [59]) and diving by the assay volume (0.1 mL). 4.11. Extracellular ADP measurements For detection of eADP of intact Caco-2 cells, 3 U/100 μL of pyruvate kinase and 100 μM PEP were added to the luciferin-luciferase mix. Using PEP as a substrate, pyruvate kinase promotes the stoichiometric conversion of eADP into eATP [60]. The resulting eATP was then mea- sured by light emission using the luciferin-luciferase procedure described above. 4.12. Data analysis Statistical significance was determined using the non-parametric Mann-Whitney test. Data were analyzed and graphically represented using GraphPad Prism software v5.0 (Graph Pad Software, San Diego, CA, USA). Each independent experiment was carried out in an indepen- dent cell culture or tissue sample in a different day. 4.13. Initial velocity estimation To measure the initial velocity of Ecto-AK or Ecto-NDPK, the eATP dynamics were measured as indicated in section 4.8.2 and 4.8.3. Only the values of [eATP] obtained during the first 5 minutes after substrates addition were considered for further analysis. The following equation was fitted to experimental data: ½ eATP � ¼ Að1 (cid:0) e(cid:0) k timeÞ ð2Þ where A and k are parameters, whose value are optimized to achieve a good fitting of Eq 2 to experimental data. The initial velocity is the derivative of [eATP] as a function of time at time 0 (the time when substrates were added). Thus, the initial velocity was calculated by multiply- ing the value of A by the value of k. 4.14. Mathematical modelling Chemical models of extracellular nucleotides were built using COPASI (Complex Pathway Simulator) software in version 4.29 (source: https://copasi.org/) [61]. Parameter optimization PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 18 / 26 PLOS COMPUTATIONAL BIOLOGY eATP recycling in human epithelial intestinal cells was performed using COPASI “parameter estimation function” with Hooke & Jeeves (50 itera- tion steps, 10−5 tolerance and 0.2 rho factor), Levenberg-Marquardt (2000 iteration steps, 10−6 tolerance), or Evolutionary programming (200 generations with a 20 population size) as opti- mization methods. An initial guess of the parameter value was proposed based on literature data for each kinetic step. A detailed description of the models employed in this work can be found in S1 and S2 Tables. Parameters obtained from the model fitting are expressed as the best value ± standard deviation. The COPASI and SBML files of the models described in sec- tion 4.13.1 and 4.13.2 can be found in the data repository (see data availability statement). When performing time course simulations, we used the deterministic LSODA method with default parameters in COPASI version 4.29. Specifically, we set the relative tolerance to 10−6, the absolute tolerance to 10−12, and allowed for a maximum of 105 internal steps without a limit on maximal step size. Model prediction uncertainties were calculated using the parameter scan task in COPASI, which explores the parameter space within the interval given by the parameter best value ± the standard error (S1 Table). Four parameter values were considered within the range of the parameter best value ± the standard error, one at the lowest value of the interval, one at the highest and two in the middle. Simulations were performed by testing all possible combina- tions of the selected parameter values. The shaded regions depicted in Figs 5 and 6 correspond to the area that includes all simulations results obtained by varying the parameters values. In Fig 5, the prediction uncertainty was calculated by simulating the eATP kinetics while varying the parameters KATP, KADP, Vmax Ecto-AMPase and FtrAK. These are the kinetic parameters that control eATP kinetics under the hypotonic stimulus. In Fig 6D, 6E and 6F, the prediction uncertainty was calculated by simulating the nucleotides kinetics by varying the parameters KATP, KmAD, KmUTP Vmax NDPK and KNTPase. 4.1.4.1. A model of purinergic homeostasis in non-polarized Caco-2 cells. To explain the experimental observations, a data driven mathematical model was created (depicted in Fig 4A). The model has 7 reactions to explain the chemical fluxes of transformations or transport of extracellular nucleotides in Caco-2 cells: JATP, JEcto-ATPase, JEcto-ADPase, JEcto-AMPase, JEcto-AK, JEcto-NDPK and JEcto-NTPDase. A detailed description of each flux, its mathematical description and parameters can be found in S1 Table. In the model, the concentration of each species as a function of time was calculated from the following differential equations: � ½ @ eATP @t ¼ JATP (cid:0) ð JEcto(cid:0) ATPase þ JEcto(cid:0) AK þ JEcto(cid:0) NDPK Þ � ½ @ eADP @t ¼ JEcto(cid:0) ATPase (cid:0) JEcto(cid:0) ADPase þ 2∗JEcto(cid:0) AK þ JEcto(cid:0) NDPK � ½ @ eAMP @t ¼ JEcto(cid:0) ADPase (cid:0) ð JEcto(cid:0) AK þ JEcto(cid:0) AMPase Þ � ½ @ eADO @t ¼ JEcto(cid:0) AMPase � ½ @ eCTP @t ¼ JEcto(cid:0) NDPK � ½ @ eCDP @t ¼ (cid:0) JEcto(cid:0) NDPK ð3Þ ð4Þ ð5Þ ð6Þ ð7Þ ð8Þ PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 19 / 26 PLOS COMPUTATIONAL BIOLOGY eATP recycling in human epithelial intestinal cells � ½ @ eUTP @t ¼ JEcto(cid:0) NDPK (cid:0) JEcto(cid:0) NTPase � ½ @ eUDP @t ¼ (cid:0) JEcto(cid:0) NDPK þ JEcto(cid:0) NTPase ð9Þ ð10Þ Note that in the equations JEcto-AK and JEcto-NDPK were considered in the direction of eATP consumption, i.e., eAMP + eATP $ 2 eADP for JEcto-AK and eNDP + eATP $ eNTP + eADP for JEcto-NDPK. The model was written in COPASI 4.29 and was fitted simultaneously to all experimental data shown in Figs 1A, 1C, 2A, 3A and 3B. The fitting of the model to experi- mental data can be seen in Figs 4B, 5C, 6A and 6C as red lines. Some kinetic parameters of the enzymes catalyzing the reactions were obtained from the lit- erature. Parameters from the Jecto-ATPase and Jecto-ADPase were obtained from our previous work [17]. The Vmax of the Jecto-AMPase reaction was obtained from our previous work [17], while the Km was obtained from the work of Navarro et al. [62]. Kinetic parameters of the Jecto- AK activity were obtained from the work of Sheng et al [22]. The equilibrium constant (Keq) and the affinity for ATP (KmAT) of the Jecto-NDPK were obtained from the work of Garces and Cleland [63]. The affinity constants for product inhibition in Jecto-NDPK (KiNDP and KiADP) were estimated from the work from Lascu et Gonin [64]. The rest of the model parameters were obtained from model fitting to experimental data (see S1 and S2 Tables for more details). The shape of the JATP flux as a function of time was modeled based on findings of a previous work from our group [65]. 4.1.4.2. A model of purinergic homeostasis in polarized Caco-2 cells. The model fitted to experimental data from the apical and basolateral compartments data is the same model indicated in section 4.13.1, although the parameters of some reactions were fitted again (S2 Table). The JATP expression for the 180 mOsm hypotonic shock in the polarized cells was dif- ferent from the one employed on non-polarized cells (S2 Table). The mathematical expressions of the other 6 reactions were not modified. Four parameters were refitted to the data to account for differences in the ecto-ADPase, ecto-AK and ecto-NDPK activities after polariza- tion (values can be found in S2 Table). Moreover, in the case of ecto-NTPDase, the eCTP hydrolysis could not be neglected in the apical compartment and was necessary to achieve a good fit to experimental data. In contrast the eCTP hydrolysis could be avoided in the basolat- eral compartment without affecting model fitting. This suggest that the ecto-CTPase activity is greater in the apical than in the basolateral compartment, in agreement with the increased activity of other enzymes on the apical side. The differential equations for [eCTP] and [eCDP] are modified in the apical side model to account for the eCTP hydrolysis: � ½ @ eCTP @t ¼ JEcto(cid:0) NDPK (cid:0) JEcto(cid:0) NTPase � ½ @ eCDP @t ¼ (cid:0) JEcto(cid:0) NDPK þ JEcto(cid:0) NTPase ð11Þ ð12Þ The models for the apical and basolateral compartments were written in COPASI 4.29 and fit- ted to experimental data shown in Fig 7A, 7C and 7D. The COPASI files can be found at https://doi.org/10.6084/m9.figshare.21938651.v1. PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 20 / 26 PLOS COMPUTATIONAL BIOLOGY eATP recycling in human epithelial intestinal cells Supporting information S1 Fig. iADP release estimation. Increase in [eATP] after a 180 mOsm hypotonic shock in absence (blue) or presence (grey) of PK (3 U) and PEP (100 μM) were evaluated as ΔATP, i.e., the difference between [eATP] at 1 min post-stimulus and basal [eATP]. (TIF) S2 Fig. Inhibition by exogenous eUDP of ecto-NDPK activity in the presence of eCTP and eADP. Time course of eATP accumulation in the presence of 100 eUTP μM in the absence (blue) or in the presence of 5 mM eUDP (grey). The data showed are the means of 3 indepen- dent experiments. (TIF) S3 Fig. Measurement of eADP by the conversion to eATP. Caco-2 cells were incubated with luciferin-luciferase and, at the time indicated with the arrow, PK (3 U) and PEP (100 μM) were added. The value of the [eADP] in resting conditions was 0.77 ± 0.44 μM eADP/mg. Given a usual protein cell mass of 0.2 mg, the [eADP] in resting conditions is 0.15 ± 0.09 μM. The data showed are the means of 5 independent experiments. (TIF) S4 Fig. Ecto-nucleotidase activity of Caco-2 cells. Experiments were performed in assay medium without Pi at room temperature, and Pi production was measured by the malachite green method (section 4.9.4). The time course of Pi accumulation in the extracellular media of Caco-2 cells was measured and values of enzyme activity were derived from initial rates of nucleotides hydrolysis for 500 μM of eUTP (grey), eGTP (blue) and eCTP (light blue). The data are the means of ± s.e.m. from 3 to 5 independent experiments. (TIF) S5 Fig. Basolateral and apical ecto-ATPase activity of Caco-2 cells. eATP kinetics of cells exposed to [eATP] (0.2–7 μM). Levels of [eATP] were measured by luminometry at the baso- lateral (A) and apical (B) sides of the polarized Caco-2 monolayers. Data is the mean of 3 inde- pendent experiments run in duplicate. The initial velocity of the ecto-ATPase activity was calculated by linear regression to experimental data obtaining the slope and y-intercept of the line. The slope represented the eATP hydrolysis as a function of time, i.e. the ecto-ATPase activity at each [eATP] and in each compartment. (TIF) S6 Fig. Enzyme Vmax calculated from model fitting. The plot shows the enzymes’ Vmax in the apical and basolateral compartments, and in non-polarized cells. The ecto-NDPK Vmax (A) were obtained from model fitting to experimental data and are the same shown in S2 Table (for the apical and basolateral compartments) and in S1 Table (for the non-polarized cells). The ecto-AK Vmax (B) was calculated from the model parameters using the following formula: FtrAK k(cid:0) 2k1 , where the FtrAK was obtained from the model fitting (S2 Table for the apical and baso- k(cid:0) 2þk1 lateral compartments and S1 Table for the non-polarized cells). The k-2 and k1 parameters value can be found in S1 Table. (TIF) S1 Table. Mathematical model of eATP regulation in non-polarized Caco-2 cells. Numeri- cal values of constants were normalized by the protein cell mass in the experiments (Mcell), measured by the Bradford method (section 4.7 in the manuscript). Parameter fitting and simu- lations were performed by selecting the average cell mass in the experiments (Mcell = 0.2 mg). JL and JNL represent the lytic and non-lytic iATP release respectively upon an osmotic shock. PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 21 / 26 PLOS COMPUTATIONAL BIOLOGY eATP recycling in human epithelial intestinal cells The value of these terms was 0 before shock application. Jleakage represents a constant and small iATP release observed in the absence of any stimulus. The parameter values obtained from the model fitting are expressed as the best value ± standard deviation. (XLSX) S2 Table. A: JNL parameters obtained from fitting to experimental data at 180 mOsm shock in apical or basolateral compartments in polarized cells. The same value of kobs was considered for both compartments. Parameters obtained from the model fitting are expressed as the best value ± standard deviation. B: Parameters obtained from model fitting to experimental in apical or basolateral compartments in polarized cells. The model equations are the same shown in S1 Table, however, some parameters values were fitted again to experimental data from polarized cells. The parameters whose value has changed in comparison with the model of non-polarized cells are shown in this file. The rest of the parameters had the same value for non-polarized cells (shown in S1 Table). The KADPase and KNTPase (for eCTP) were considered 0 in the basolateral compartment. This does not mean that there is no ecto-ADPase or ecto- NTPase activity in the basolateral side but, they can be neglected in our experimental condi- tions. Parameters obtained from the model fitting are expressed as the best value ± standard deviation. (XLSX) Acknowledgments We are thankful to Dr. Cafferata for providing the Caco-2 cells. Author Contributions Conceptualization: Nicolas Andres Saffioti, Pablo Julio Schwarzbaum, Julieta Schachter. Data curation: Nicolas Andres Saffioti, Pablo Julio Schwarzbaum. Formal analysis: Nicolas Andres Saffioti, Julieta Schachter. Funding acquisition: Pablo Julio Schwarzbaum, Julieta Schachter. Investigation: Nicolas Andres Saffioti, Cora Lilia Alvarez, Zaher Bazzi, Marı´a Virginia Genti- lini, Julieta Schachter. Methodology: Nicolas Andres Saffioti, Cora Lilia Alvarez, Gabriel Eduardo Gondolesi, Pablo Julio Schwarzbaum, Julieta Schachter. Project administration: Pablo Julio Schwarzbaum, Julieta Schachter. Resources: Julieta Schachter. Supervision: Pablo Julio Schwarzbaum, Julieta Schachter. Writing – original draft: Nicolas Andres Saffioti, Pablo Julio Schwarzbaum, Julieta Schachter. Writing – review & editing: Nicolas Andres Saffioti, Cora Lilia Alvarez, Pablo Julio Schwarz- baum, Julieta Schachter. References 1. Lavoie EG, Gulbransen BD, Martı´n-Satue´ M, Aliagas E, Sharkey KA, Se´ vigny J. Ectonucleotidases in the digestive system: focus on NTPDase3 localization. Am J Physiol Liver Physiol. 2011; 300: G608– G620. https://doi.org/10.1152/ajpgi.00207.2010 PMID: 21233276 2. Stefan C, Jansen S, Bollen M. Modulation of purinergic signaling by NPP-type ectophosphodiesterases. Purinergic Signal. 2006; 2: 361–370. https://doi.org/10.1007/s11302-005-5303-4 PMID: 18404476 PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 22 / 26 PLOS COMPUTATIONAL BIOLOGY eATP recycling in human epithelial intestinal cells 3. Coutinho-Silva R, Stahl L, Cheung K-K, De Campos NE, de Oliveira Souza C, Ojcius DM, et al. P2X and P2Y purinergic receptors on human intestinal epithelial carcinoma cells: effects of extracellular nucleotides on apoptosis and cell proliferation. Am J Physiol Liver Physiol. 2005; 288: G1024–G1035. https://doi.org/10.1152/ajpgi.00211.2004 PMID: 15662049 4. Dal Ben D, Antonioli L, Lambertucci C, Spinaci A, Fornai M, D’Antongiovanni V, et al. Approaches for designing and discovering purinergic drugs for gastrointestinal diseases. Expert Opin Drug Discov. 2020; 15: 687–703. https://doi.org/10.1080/17460441.2020.1743673 PMID: 32228110 5. 6. Trautmann A. Extracellular ATP in the immune system: more than just a “danger signal.” Sci signal. 2009; 2: e6. Zimmermann H. Extracellular ATP and other nucleotides—ubiquitous triggers of intercellular messen- ger release. Purinergic Signal. 2016; 12: 25–57. https://doi.org/10.1007/s11302-015-9483-2 PMID: 26545760 7. Crane JK, Naeher TM, Choudhari SS, Giroux EM. Two pathways for ATP release from host cells in enteropathogenic Escherichia coli infection. Am J Physiol Liver Physiol. 2005; 289: G407–G417. https://doi.org/10.1152/ajpgi.00137.2005 PMID: 16093420 8. Ullrich N, Caplanusi A, Broˆ ne B, Hermans D, Larivière E, Nilius B, et al. Stimulation by caveolin-1 of the hypotonicity-induced release of taurine and ATP at basolateral, but not apical, membrane of Caco-2 cells. Am J Physiol Physiol. 2006; 290: C1287–C1296. 9. Wei Z-Y, Qu H-L, Dai Y-J, Wang Q, Ling Z-M, Su W-F, et al. Pannexin 1, a large-pore membrane chan- nel, contributes to hypotonicity-induced ATP release in Schwann cells. Neural Regen Res. 2021; 16: 899. https://doi.org/10.4103/1673-5374.290911 PMID: 33229726 10. Dosch M, Zindel J, Jebbawi F, Melin N, Sanchez-Taltavull D, Stroka D, et al. Connexin-43-dependent ATP release mediates macrophage activation during sepsis. Elife. 2019; 8: e42670. https://doi.org/10. 7554/eLife.42670 PMID: 30735126 11. To WKL, Kumar P, Marshall JM. Hypoxia is an effective stimulus for vesicular release of ATP from human umbilical vein endothelial cells. Placenta. 2015; 36: 759–766. 12. Burnstock G. Purinergic signalling in the gastrointestinal tract and related organs in health and disease. Purinergic Signal. 2014; 10: 3–50. https://doi.org/10.1007/s11302-013-9397-9 PMID: 24307520 13. Yegutkin GG. Enzymes involved in metabolism of extracellular nucleotides and nucleosides: functional implications and measurement of activities. Crit Rev Biochem Mol Biol. 2014; 49: 473–497. https://doi. org/10.3109/10409238.2014.953627 PMID: 25418535 Zimmermann H. Ectonucleoside triphosphate diphosphohydrolases and ecto-50-nucleotidase in puri- nergic signaling: how the field developed and where we are now. Purinergic Signal. 2021; 17: 117–125. https://doi.org/10.1007/s11302-020-09755-6 PMID: 33336318 14. 15. Bahrami F, Kukulski F, Lecka J, Tremblay A, Pelletier J, Rockenbach L, et al. Purine-metabolizing ectoenzymes control IL-8 production in human colon HT-29 cells. Mediators Inflamm. 2014; 2014. https://doi.org/10.1155/2014/879895 PMID: 25242873 16. Clayton A, Al-Taei S, Webber J, Mason MD, Tabi Z. Cancer exosomes express CD39 and CD73, which suppress T cells through adenosine production. J Immunol. 2011; 187: 676–683. https://doi.org/10. 4049/jimmunol.1003884 PMID: 21677139 17. Schachter J, Alvarez CL, Bazzi Z, Faillace MP, Corradi G, Hattab C, et al. Extracellular ATP hydrolysis in Caco-2 human intestinal cell line. Biochim Biophys Acta (BBA)-Biomembranes. 2021; 1863: 183679. https://doi.org/10.1016/j.bbamem.2021.183679 PMID: 34216588 18. Lu Y, Qi J, Wu W. Lipid nanoparticles: In vitro and in vivo approaches in drug delivery and targeting. Drug targeting and stimuli sensitive drug delivery systems. Elsevier; 2018. pp. 749–783. 19. Karlsson J, Ungell A-L, Gråsjo¨ J, Artursson P. Paracellular drug transport across intestinal epithelia: influence of charge and induced water flux. Eur J Pharm Sci. 1999; 9: 47–56. https://doi.org/10.1016/ s0928-0987(99)00041-x PMID: 10493996 20. Sinev MA, Sineva E V, Ittah V, Haas E. Domain closure in adenylate kinase. Biochemistry. 1996; 35: 6425–6437. https://doi.org/10.1021/bi952687j PMID: 8639589 21. Mayer RJ. Brain mitochondrial hexokinase and adenylate kinase: effect of osmotic conditions on enzyme activities. J Neurochem. 1972; 19: 2127–2138. https://doi.org/10.1111/j.1471-4159.1972. tb05122.x PMID: 5072389 22. Sheng XR, Li X, Pan XM. An iso-random Bi Bi mechanism for adenylate kinase. J Biol Chem. 1999; 274: 22238–22242. https://doi.org/10.1074/jbc.274.32.22238 PMID: 10428790 23. Harris DS, Slot JW, Geuze HJ, James DE. Polarized distribution of glucose transporter isoforms in Caco-2 cells. Proc Natl Acad Sci. 1992; 89: 7556–7560. https://doi.org/10.1073/pnas.89.16.7556 PMID: 1502167 PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 23 / 26 PLOS COMPUTATIONAL BIOLOGY eATP recycling in human epithelial intestinal cells 24. Maule´n NP, Henrı`quez EA, Kempe S, Ca´ rcamo JG, Schmid-Kotsas A, Bachem M, et al. Up-regulation and Polarized Expression of the Sodium-Ascorbic Acid Transporter SVCT1 in Post-confluent Differenti- ated CaCo-2 Cells *. J Biol Chem. 2003; 278: 9035–9041. https://doi.org/10.1074/jbc.M205119200 PMID: 12381735 25. Burnstock G, Kennedy C. P2X receptors in health and disease. Adv Pharmacol. 2011; 61: 333–372. https://doi.org/10.1016/B978-0-12-385526-8.00011-4 PMID: 21586364 26. Buzzi N, Boland R, de Boland AR. Signal transduction pathways associated with ATP-induced prolifera- tion of colon adenocarcinoma cells. Biochim Biophys Acta (BBA)-General Subj. 2010; 1800: 946–955. https://doi.org/10.1016/j.bbagen.2010.05.009 PMID: 20562007 27. Buzzi N, Bilbao PS, Boland R, de Boland AR. Extracellular ATP activates MAP kinase cascades through a P2Y purinergic receptor in the human intestinal Caco-2 cell line. Biochim Biophys Acta (BBA)-General Subj. 2009; 1790: 1651–1659. https://doi.org/10.1016/j.bbagen.2009.10.005 PMID: 19836435 28. Vuerich M, Robson SC, Longhi MS. Ectonucleotidases in intestinal and hepatic inflammation. Front Immunol. 2019; 10: 507. https://doi.org/10.3389/fimmu.2019.00507 PMID: 30941139 29. Matsumoto H, Erickson RH, Gum JR, Yoshioka M, Gum E, Kim YS. Biosynthesis of alkaline phospha- tase during differentiation of the human colon cancer cell line Caco-2. Gastroenterology. 1990; 98: 1199–1207. https://doi.org/10.1016/0016-5085(90)90334-w PMID: 2323513 30. Woods LT, Forti KM, Shanbhag VC, Camden JM, Weisman GA. P2Y receptors for extracellular nucleo- tides: Contributions to cancer progression and therapeutic implications. Biochem Pharmacol. 2021; 187: 114406. https://doi.org/10.1016/j.bcp.2021.114406 PMID: 33412103 31. Yegutkin GG. Nucleotide-and nucleoside-converting ectoenzymes: important modulators of purinergic signalling cascade. Biochim Biophys Acta (BBA)-Molecular Cell Res. 2008; 1783: 673–694. https://doi. org/10.1016/j.bbamcr.2008.01.024 PMID: 18302942 32. Kolachala VL, Bajaj R, Chalasani M, Sitaraman S V. Purinergic receptors in gastrointestinal inflamma- tion. Am J Physiol Liver Physiol. 2008; 294: G401–G410. https://doi.org/10.1152/ajpgi.00454.2007 PMID: 18063703 33. Lelièvre V, Muller J-M, Falcòn J. Adenosine modulates cell proliferation in human colonic carcinoma. II. Differential behavior of HT29, DLD-1, Caco-2 and SW403 cell lines. Eur J Pharmacol. 1998; 341: 299– 308. https://doi.org/10.1016/s0014-2999(97)01463-5 PMID: 9543252 34. Christofi FL. Purinergic receptors and gastrointestinal secretomotor function. Purinergic Signal. 2008; 4: 213–236. https://doi.org/10.1007/s11302-008-9104-4 PMID: 18604596 35. Grasl M, Turnheim K. Stimulation of electrolyte secretion in rabbit colon by adenosine. J Physiol. 1984; 346: 93–110. https://doi.org/10.1113/jphysiol.1984.sp015009 PMID: 6699790 36. Leipziger J, Kerstan D, Nitschke R, Greger R. ATP increases [Ca2+] i and ion secretion via a basolateral P2Y-receptor in rat distal colonic mucosa. Pflu¨gers Arch. 1997; 434: 77–83. https://doi.org/10.1007/ pl00008079 PMID: 9094258 37. Korman LY, Lemp GF, Jackson MJ, Gardner JD. Mechanism of action of ATP on intestinal epithelial cells: cyclic AMP-mediated stimulation of active ion transport. Biochim Biophys Acta (BBA)-Molecular Cell Res. 1982; 721: 47–54. 38. Yamamoto T, Suzuki Y. Role of luminal ATP in regulating electrogenic Na+ absorption in guinea pig dis- tal colon. Am J Physiol Liver Physiol. 2002; 283: G300–G308. https://doi.org/10.1152/ajpgi.00541.2001 PMID: 12121876 39. Kinoshita N, Takahashi T, Tada S, Shinozuka K, Mizuno N, Takahashi K. Activation of P2Y receptor enhances high-molecular compound absorption from rat ileum. J Pharm Pharmacol. 2006; 58: 195– 200. https://doi.org/10.1211/jpp.58.2.0006 PMID: 16451747 40. Lazarowski ER, Homolya L, Boucher RC, Harden TK. Identification of an ecto-nucleoside diphosphoki- nase and its contribution to interconversion of P2 receptor agonists. J Biol Chem. 1997; 272: 20402– 20407. https://doi.org/10.1074/jbc.272.33.20402 PMID: 9252347 41. Yegutkin GG, Henttinen T, Samburski SS, Spychala J, Jalkanen S. The evidence for two opposite, ATP-generating and ATP-consuming, extracellular pathways on endothelial and lymphoid cells. Bio- chem J. 2002; 367: 121–128. https://doi.org/10.1042/BJ20020439 PMID: 12099890 42. Yegutkin GG, Henttinen T, Jalkanen S. Extracellular ATP formation on vascular endothelial cells is mediated by ecto-nucleotide kinase activities via phosphotransfer reactions. FASEB J. 2001; 15: 251– 260. https://doi.org/10.1096/fj.00-0268com PMID: 11149913 43. Burrell HE, Wlodarski B, Foster BJ, Buckley KA, Sharpe GR, Quayle JM, et al. Human keratino- cytes release ATP and utilize three mechanisms for nucleotide interconversion at the cell surface. J Biol Chem. 2005; 280: 29667–29676. https://doi.org/10.1074/jbc.M505381200 PMID: 15958381 PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 24 / 26 PLOS COMPUTATIONAL BIOLOGY eATP recycling in human epithelial intestinal cells 44. Fabre ACS, Vantourout P, Champagne E, Terce´ F, Rolland C, Perret B, et al. Cell surface adenylate kinase activity regulates the F1-ATPase/P2Y13-mediated HDL endocytosis pathway on human hepato- cytes. Cell Mol Life Sci C. 2006; 63: 2829–2837. 45. Kennedy C. The P2Y/P2X divide: How it began. Biochem Pharmacol. 2021; 187: 114408. https://doi. org/10.1016/j.bcp.2021.114408 PMID: 33444568 46. Giuliani AL, Sarti AC, Di Virgilio F. Extracellular nucleotides and nucleosides as signalling molecules. Immunol Lett. 2019; 205: 16–24. https://doi.org/10.1016/j.imlet.2018.11.006 PMID: 30439478 47. Pieri M, Christian HC, Wilkins RJ, Boyd CAR, Meredith D. The apical (hPepT1) and basolateral peptide transport systems of Caco-2 cells are regulated by AMP-activated protein kinase. Am J Physiol Liver Physiol. 2010; 299: G136–G143. https://doi.org/10.1152/ajpgi.00014.2010 PMID: 20430871 48. Narumi K, Ohata T, Horiuchi Y, Satoh H, Furugen A, Kobayashi M, et al. Mutual role of ecto-5’-nucleo- tidase/CD73 and concentrative nucleoside transporter 3 in the intestinal uptake of dAMP. PLoS One. 2019; 14: e0223892. https://doi.org/10.1371/journal.pone.0223892 PMID: 31634358 49. Salem M, Lecka J, Pelletier J, Marconato DG, Dumas A, Vallières L, et al. NTPDase8 protects mice from intestinal inflammation by limiting P2Y6 receptor activation: identification of a new pathway of inflammation for the potential treatment of IBD. Gut. 2022; 71: 43–54. https://doi.org/10.1136/gutjnl- 2020-320937 PMID: 33452178 50. Zimmermann H, Zebisch M, Stra¨ter N. Cellular function and molecular structure of ecto-nucleotidases. Purinergic Signal. 2012; 8: 437–502. https://doi.org/10.1007/s11302-012-9309-4 PMID: 22555564 51. Alvarez CL, Corradi G, Lauri N, Marginedas-Freixa I, Leal Denis MF, Enrique N, et al. Dynamic regula- tion of extracellular ATP in Escherichia coli. Biochem J. 2017; 474: 1395–1416. https://doi.org/10.1042/ BCJ20160879 PMID: 28246335 52. Inami A, Kiyono H, Kurashima Y. ATP as a pathophysiologic mediator of bacteria-host crosstalk in the gastrointestinal tract. Int J Mol Sci. 2018; 19: 2371. https://doi.org/10.3390/ijms19082371 PMID: 30103545 53. Puhar A, Tronchère H, Payrastre B, Van Nhieu GT, Sansonetti PJ. A Shigella effector dampens inflam- mation by regulating epithelial release of danger signal ATP through production of the lipid mediator PtdIns5P. Immunity. 2013; 39: 1121–1131. https://doi.org/10.1016/j.immuni.2013.11.013 PMID: 24332032 54. Strehler BL. Bioluminescence assay: principles and practice. Methods Biochem Anal. 1968; 16: 99– 181. https://doi.org/10.1002/9780470110348.ch2 PMID: 4385967 55. Pafundo DE, Chara O, Faillace MP, Krumschnabel G, Schwarzbaum PJ. Kinetics of ATP release and cell volume regulation of hyposmotically challenged goldfish hepatocytes. Am J Physiol Integr Comp Physiol. 2008; 294: R220–R233. https://doi.org/10.1152/ajpregu.00522.2007 PMID: 17928510 56. Gorman MW, Marble DR, Ogimoto K, Feigl EO. Measurement of adenine nucleotides in plasma. Lumin J Biol Chem Lumin. 2003; 18: 173–181. https://doi.org/10.1002/bio.721 PMID: 12701093 57. Bradford MM. A rapid and sensitive method for the quantitation of microgram quantities of protein utiliz- ing the principle of protein-dye binding. Anal Biochem. 1976; 72: 248–254. https://doi.org/10.1006/abio. 1976.9999 PMID: 942051 58. Lanzetta PA, Alvarez LJ, Reinach PS, Candia OA. An improved assay for nanomole amounts of inor- ganic phosphate. Anal Biochem. 1979; 100: 95–97. https://doi.org/10.1016/0003-2697(79)90115-5 PMID: 161695 59. Blais A, Bissonnette P, Berteloot A. Common characteristics for Na+-dependent sugar transport in Caco-2 cells and human fetal colon. J Membr Biol. 1987; 99: 113–125. https://doi.org/10.1007/ BF01871231 PMID: 3123697 60. Mohammadi S, Nikkhah M, Nazari M, Hosseinkhani S. Design of a coupled bioluminescent assay for a recombinant pyruvate kinase from a thermophilic Geobacillus. Photochem Photobiol. 2011; 87: 1338– 1345. https://doi.org/10.1111/j.1751-1097.2011.00973.x PMID: 21790618 61. Hoops S, Sahle S, Gauges R, Lee C, Pahle J, Simus N, et al. COPASI—a COmplex PAthway SImula- tor. Bioinformatics. 2006; 22: 3067–3074. https://doi.org/10.1093/bioinformatics/btl485 PMID: 17032683 62. Navarro JM, Olmo N, Turnay J, Lo´ pez-Conejo MT, Lizarbe MA. Ecto-50-nucleotidase from a human colon adenocarcinoma cell line. Correlation between enzyme activity and levels in intact cells. Mol Cell Biochem. 1998; 187: 121–131. https://doi.org/10.1023/a:1006808232059 PMID: 9788749 63. Garces E, Cleland WW. Kinetic study of yeast nucleosidediphosphate kinase. Biochemistry. 1969; 8: 633–640. 64. Lascu I, Gonin P. The catalytic mechanism of nucleoside diphosphate kinases. J Bioenerg Biomembr. 2000; 32: 237–246. https://doi.org/10.1023/a:1005532912212 PMID: 11768307 PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 25 / 26 PLOS COMPUTATIONAL BIOLOGY eATP recycling in human epithelial intestinal cells 65. Espelt M V, de Tezanos Pinto F, Alvarez CL, Alberti GS, Incicco J, Denis MFL, et al. On the role of ATP release, ectoATPase activity, and extracellular ADP in the regulatory volume decrease of Huh-7 human hepatoma cells. Am J Physiol Physiol. 2013; 304: C1013–C1026. https://doi.org/10.1152/ajpcell.00254. 2012 PMID: 23485713 PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011196 June 29, 2023 26 / 26 PLOS COMPUTATIONAL BIOLOGY
null
10.1126_sciadv.adf9336.pdf
null
null
Supplementary Materials for Reconstitution of morphogen shuttling circuits Ronghui Zhu et al. Corresponding author: Michael B. Elowitz, melowitz@caltech.edu Sci. Adv. 9, eadf9336 (2023) DOI: 10.1126/sciadv.adf9336 The PDF file includes: Supplementary Text Figs. S1 to S11 Tables S1 to S4 Legends for movies S1 to S8 References Other Supplementary Material for this manuscript includes the following: Movies S1 to S8 Supplementary Text The mathematical model of BMP4-Chordin-Twsg1-BMP-1 circuit Here we introduce the mathematical model of BMP4-Chordin-Twsg1-BMP-1 circuit, which is based on previous models (10, 20, 30) but specifically incorporating the components analyzed here (Fig. 4A), including mobile extracellular components BMP4 ([𝐵𝑀𝑃4]), Chordin ([𝐶ℎ𝑜𝑟𝑑𝑖𝑛]), Twsg1 ([𝑇𝑤𝑠𝑔1]), BMP4-Chordin complex ([𝐵𝑀𝑃4-𝐶ℎ𝑜𝑟𝑑𝑖𝑛]) and BMP4-Chordin-Twsg1 ([𝐵𝑀𝑃4-𝐶ℎ𝑜𝑟𝑑𝑖𝑛-𝑇𝑤𝑠𝑔1]) complex, as well as immobile components receptors ([𝑅]), BMP4- receptor complex ([𝐵𝑅]), and fluorescent reporters ([𝐶𝑖𝑡𝑟𝑖𝑛𝑒]). We used a set of reaction-diffusion partial differential equations for mobile components. Thus, their equations have two parts. The first part is a diffusion term 𝐷∇!𝑐, where 𝑐 can be [𝐵𝑀𝑃4], [𝐶ℎ𝑜𝑟𝑑𝑖𝑛], [𝑇𝑤𝑠𝑔1], [𝐵𝑀𝑃4-𝐶ℎ𝑜𝑟𝑑𝑖𝑛], [𝐵𝑀𝑃4-𝐶ℎ𝑜𝑟𝑑𝑖𝑛-𝑇𝑤𝑠𝑔1], and 𝐷 can be 𝐷", 𝐷#, 𝐷$, 𝐷"#, 𝐷"#$, correspondingly. We used the effective diffusion coefficients for each mobile component, taking into account the effects of their interactions with extracellular matrix components such as heparan sulfate polyglycans, but not BMP4-receptor interactions. The second part contains reaction terms. First, BMP4 can bind to Chordin weakly, with estimated association rate 𝑘"# and dissociation rate 𝑟"#. Then BMP4-Chordin complex can further interact with Twsg1 to form BMP4-Chordin-Twsg1 complex, with association rate 𝑘"#$ and dissociation rate 𝑟"#$. Furthermore, BMP4 can bind to receptors with estimated association rate 𝑘% and dissociation rate 𝑟%. Finally, Chordin in its free form or in complex forms can be cleaved by BMP-1 secreted by Receiver-B1 cells with a rate 𝛽 dependent on BMP-1 expression level, and BMP4 and Twsg1 can be released from BMP4-Chordin and BMP4-Chordin-Twsg1 complex once Chordin is cleaved. Thus, we can write, 𝜕[𝐵𝑀𝑃4]/𝜕𝑡 = 𝐷"∇![𝐵𝑀𝑃4] − 𝑘"#[𝐵𝑀𝑃4][𝐶ℎ𝑜𝑟𝑑𝑖𝑛] + 𝑟"#[𝐵𝑀𝑃4-𝐶ℎ𝑜𝑟𝑑𝑖𝑛] − 𝑘%[𝐵𝑀𝑃4][𝑅] + 𝑟%[𝐵𝑅] + 𝛽[𝐵𝑀𝑃4-𝐶ℎ𝑜𝑟𝑑𝑖𝑛] + 𝛽[𝐵𝑀𝑃4-𝐶ℎ𝑜𝑟𝑑𝑖𝑛-𝑇𝑤𝑠𝑔1] 𝜕[𝐶ℎ𝑜𝑟𝑑𝑖𝑛]/𝜕𝑡 = 𝐷#∇![𝐶ℎ𝑜𝑟𝑑𝑖𝑛] − 𝑘"#[𝐵𝑀𝑃4][𝐶ℎ𝑜𝑟𝑑𝑖𝑛] + 𝑟"#[𝐵𝑀𝑃4-𝐶ℎ𝑜𝑟𝑑𝑖𝑛] − 𝛽[𝐶ℎ𝑜𝑟𝑑𝑖𝑛] 𝜕[𝑇𝑤𝑠𝑔1]/𝜕𝑡 = 𝐷$∇![𝑇𝑤𝑠𝑔1] − 𝑘"#$[𝑇𝑤𝑠𝑔1][𝐵𝑀𝑃4-𝐶ℎ𝑜𝑟𝑑𝑖𝑛] + 𝑟"#$[𝐵𝑀𝑃4-𝐶ℎ𝑜𝑟𝑑𝑖𝑛-𝑇𝑤𝑠𝑔1] + 𝛽[𝐵𝑀𝑃4-𝐶ℎ𝑜𝑟𝑑𝑖𝑛-𝑇𝑤𝑠𝑔1] 𝜕[𝐵𝑀𝑃4-𝐶ℎ𝑜𝑟𝑑𝑖𝑛]/𝜕𝑡 = 𝐷"#∇![𝐵𝑀𝑃4-𝐶ℎ𝑜𝑟𝑑𝑖𝑛] + 𝑘"#[𝐵𝑀𝑃4][𝐶ℎ𝑜𝑟𝑑𝑖𝑛] − 𝑟"#[𝐵𝑀𝑃4-𝐶ℎ𝑜𝑟𝑑𝑖𝑛] − 𝑘"#$[𝑇𝑤𝑠𝑔1][𝐵𝑀𝑃4-𝐶ℎ𝑜𝑟𝑑𝑖𝑛] + 𝑟"#$[𝐵𝑀𝑃4-𝐶ℎ𝑜𝑟𝑑𝑖𝑛-𝑇𝑤𝑠𝑔1] − 𝛽[𝐵𝑀𝑃4-𝐶ℎ𝑜𝑟𝑑𝑖𝑛] 𝜕[𝐵𝑀𝑃4-𝐶ℎ𝑜𝑟𝑑𝑖𝑛-𝑇𝑤𝑠𝑔1]/𝜕𝑡 = 𝐷"#$∇![𝐵𝑀𝑃4-𝐶ℎ𝑜𝑟𝑑𝑖𝑛-𝑇𝑤𝑠𝑔1] + 𝑘"#$[𝑇𝑤𝑠𝑔1][𝐵𝑀𝑃4-𝐶ℎ𝑜𝑟𝑑𝑖𝑛] − 𝑟"#$[𝐵𝑀𝑃4-𝐶ℎ𝑜𝑟𝑑𝑖𝑛-𝑇𝑤𝑠𝑔1] − 𝛽[𝐵𝑀𝑃4-𝐶ℎ𝑜𝑟𝑑𝑖𝑛-𝑇𝑤𝑠𝑔1] Note that Twsg1 can also interact with BMP4 and Chordin individually (71), so there exist multiple interaction routes of forming the final BMP4-Chordin-Twsg1 complex. However, since interactions between Twsg1 and BMP4 or Chordin are much weaker than interactions between BMP4 and Chordin, we only consider one interaction route (BMP4 first interacts with Chordin, then BMP4-Chordin interacts with Twsg1) in our model. For the immobile BMP4 receptors ([𝑅]), other than reversible interactions with BMP4 ligands, we also considered receptor-mediated internalization and degradation of BMP4 ligands (72). Once the BMP4-receptor complex ([𝐵𝑅]) is internalized, we assumed that the BMP4 ligand is degraded and the receptor is recycled with a rate 𝛾. Thus, we can write, 𝜕[𝑅]/𝜕𝑡 = −𝑘%[𝐵𝑀𝑃4][𝑅] + 𝑟%[𝐵𝑅] + 𝛾[𝐵𝑅] 𝜕[𝐵𝑅]/𝜕𝑡 = 𝑘%[𝐵𝑀𝑃4][𝑅] − 𝑟%[𝐵𝑅] − 𝛾[𝐵𝑅] From these equations we can see that the total receptor concentration 𝑅𝑇𝑜𝑡𝑎𝑙 = [𝑅] + [𝐵𝑅] is held constant. Finally, we assume fluorescent reporter production rate follows a Hill function with BMP4-receptor complex concentration as a variable. The Citrine degrades with a ~24hr turnover time (70). Thus, we can write, 𝜕[𝐶𝑖𝑡𝑟𝑖𝑛𝑒]/𝜕𝑡 = 𝑏[𝐵𝑅]&/(𝐾& + [𝐵𝑅]&) − 𝛿’()[𝐶𝑖𝑡𝑟𝑖𝑛𝑒] To obtain more precise estimates of parameters that have relatively large estimated ranges from previous studies, such as 𝑅𝑇𝑜𝑡𝑎𝑙, 𝑘%, 𝑟% and 𝛾, as we as parameters that cannot be estimated from previous studies, such as 𝑏, 𝑛 and 𝐾, we fitted a simplified model containing only [𝐵𝑀𝑃4], [𝑅], [𝐵𝑅], [𝐶𝑖𝑡𝑟𝑖𝑛𝑒], and without diffusion terms, to a time-lapse movie (Movie S2-4) of Receiver-B1 cells turning on Citrine fluorescence in response to 5, 10, 20 ng/ml recombinant BMP4. All estimated parameters used in the model are listed in Table S4. There is a time delay for Citrine fluorescence to become detectable after BMP signaling, due to transcription, translation and maturation of Citrine fluorescence proteins (Fig. S1). To incorporate this delay into the model, we added a [𝐶𝑖𝑡𝑟𝑖𝑛𝑒]∗ term to lump together species like Citrine mRNA and immature Citrine proteins that are produced by Citrine fluorescence reporter but have not been converted into detectable mature Citrine proteins. The [𝐶𝑖𝑡𝑟𝑖𝑛𝑒]∗ can be converted into [𝐶𝑖𝑡𝑟𝑖𝑛𝑒] with a conversion rate 𝑟’(), i.e., 𝜕[𝐶𝑖𝑡𝑟𝑖𝑛𝑒]∗/𝜕𝑡 = 𝑏[𝐵𝑅]&/(𝐾& + [𝐵𝑅]&) − 𝛿’()[𝐶𝑖𝑡𝑟𝑖𝑛𝑒]∗ − 𝑟’()[𝐶𝑖𝑡𝑟𝑖𝑛𝑒]∗ 𝜕[𝐶𝑖𝑡𝑟𝑖𝑛𝑒]/𝜕𝑡 = 𝑟’()[𝐶𝑖𝑡𝑟𝑖𝑛𝑒]∗ − 𝛿’()[𝐶𝑖𝑡𝑟𝑖𝑛𝑒] When parameter values were not available, we made arbitrary but physiologically reasonable assumptions, and later tested whether key conclusions were sensitive to these values. As shown in Fig. S6, key conclusions in this paper were relatively insensitive to the precise values of unknown parameters, or to the incorporation of time delay of Citrine fluorescence reporter. Fig. S1. Dynamics of citrine fluorescence reporter after recombinant BMP4 addition. (A) Images at selected timepoints of time-lapse imaging data (one of two replicates, corresponding to Movie S1-4) of Receiver-B1 cells treated with various concentrations of recombinant BMP4 (rBMP4). (B) Citrine fluorescence reporter shows significant elevation at 4 hours after the addition of 5 ng/ml rBMP4 (left), and at 3 hours after the addition of 10 ng/ml rBMP4 (middle) and 20 ng/ml rBMP4 (right). At each timepoint, we calculated the average citrine level per image column and acquired a distribution of average citrine per column. Then we used Welch’s t-test to test whether citrine distribution of samples with rBMP4 is significantly larger than citrine distribution of the sample without rBMP4 for each timepoint (ns: p≥0.05, not significant; *: p<0.05, significant). These results show that the time delay of Citrine fluorescence reporter is 3-4 hours, since BMP signaling can be detected by pSmad staining 20 minutes after rBMP4 addition (44). Fig. S2. BMP gradients can be reconstituted in vitro. (A) (Top) Images at selected timepoints of Movie S5 are from the same time-lapse imaging data as Fig. 2B (adding a 60hr image). (Bottom) mCherry expression at the sender region can be visualized more clearly by focusing on the RFP channel alone at the same selected timepoints. (B) Without 4-OHT induction (Movie S1), no gradients were formed (top) and no mCherry expression (bottom) was detected in the sender region. In both (A) and (B), the white line on each image labels the position of sender-receiver interface. The Citrine fluorescence is shown as yellow, and mCherry fluorescence is shown as red. Fig. S3. Raw flow cytometry traces and statistical tests of sender-receiver co-culture experiment in Fig. 3A. (A) These raw flow cytometry traces are from one of the three replicate experiments. To calculate the log2 fold change in Fig. 3A, we first obtained the mean Citrine from each trace, then calculated the log2 fold change by the equation log2[(Receiver Citrine - Cit0) / (Cit1 - Cit0)], where Cit0 is the receiver Citrine of the sample without 4-OHT (gray trace) and Cit1 is the receiver Citrine of the sample with only 4-OHT induction (red trace). The number in the parenthesis in the figure legend corresponds to the condition number in Fig. 3A. (B) For each matrix entry, the number is the p value of Welch’s t-test between the corresponding pair of conditions. The condition number corresponds to the condition number in Fig. 3A. Fig. S4. Noggin strongly inhibits BMP4 signaling. (A) We used the same doxycycline inducible system as Sender-C to construct an inducible Noggin sender cell line, Sender-N (Table S3). (B) Sender-receiver co-culture experiment verifies strong inhibition of BMP4 signaling by Noggin, which cannot be relieved by BMP-1 expression. This co-culture experiment was performed in a similar way to that in Fig. 3A, with Sender-N* substituting Sender-C*. Sender-N* cells were engineered from Sender-N cells by adding a constitutively expressed mTurquoise2 cassette (Table S3), so that they can be distinguished from Receiver-B1 cells in flow cytometry. (C) Noggin can form inhibitory gradients, which cannot be modulated by BMP-1 expression. In samples with Dox induction, Dox was added 8 hours before other components (rBMP4, rTwsg1 and ABA) to pre-induce Chordin expression. In all samples, 15 ng/ml recombinant BMP4 was added to the culture, and we took images 24 hours after rBMP4 was added. (D) Noggin expression completely abolishes BMP4 gradients. Cells were plated using the Fig. 2A protocol, with Sender-N cells substituting filler cells. 4-OHT, Dox and ABA were added together and images were taken 48 hours after induction. In both (B) and (D), 4-OHT = 4 µM. In (B), (C) and (D), Dox = 100 ng/ml, rTwsg1 = 10 nM, ABA = 1000 µM. In (C), N is the number of replicates. The white line on each image labels the position of sender-receiver interface. The Citrine fluorescence is shown as yellow, and mTurquoise2 fluorescence is shown as blue. White corresponds to yellow+blue on the computer screen. We removed the mCherry channel from images to avoid interfering with the Citrine visualization. Fig. S5. Full images and individual traces of Fig. 3B. The white line on each image labels the position of sender-receiver interface. The Citrine fluorescence is shown as yellow, and mTurquoise2 fluorescence is shown as blue. White corresponds to yellow+blue on the computer screen. We removed the mCherry channel from images to avoid interfering with the Citrine visualization. The replicate #3 of rBMP4+Dox+ rTwsg1+ABA group used a small field of view due to an acquisition error, making the individual trace and average trace for that condition noisier than other traces in Fig. 3B. Fig. S6. Qualitative shuttling behaviors in Fig. 4 were insensitive to the precise values of unknown parameters or to incorporation of a time delay of Citrine reporter. The qualitative shuttling behaviors include (1) Chordin lengthens the gradient, (2) Twsg1 with Chordin suppresses gradients, (3) BMP-1 with other components generates a displaced gradient. (A) In the original model (left), we arbitrarily chose a 𝑘"#$ to be the same as 𝑘"# (Table S4), both within the diffusion limited rate range (0.006 nM-1min-1 – 0.06 nM-1min-1) (73). Holding other parameters constant, changing 𝑘"#$ to the upper (middle) or lower (right) limit of the diffusion limited rate range does not affect the qualitative shuttling behaviors. (B) Previous studies showed that Twsg1 enhances BMP-1 cleavage of Chordin in vitro (18, 74). In the original model (left), we set the BMP-1 cleavage rate 𝛽 to be the same for Chordin in both BMP4-Chordin and BMP4- Chordin-Twsg1. Holding other parameters constant, setting a higher BMP-1 cleavage rate for Chordin in BMP4-Chordin-Twsg1 (right) does not affect the qualitative shuttling behaviors. (C) Twsg1 is not necessary for a displaced gradient, but the contrast between proximal and distal BMP signals (compared with the plots of the original model in (A) or (B)). (D) Incorporating the time delay of Citrine reporter (Fig. S1) does not affect the qualitative shuttling behaviors. We incorporate the time delay by introducing a [𝐶𝑖𝑡𝑟𝑖𝑛𝑒]∗ term to lump together species that are produced by Citrine fluorescence reporter but not detectable yet, such as Citrine mRNA and immature Citrine fluorescence proteins, and a conversion rate 𝑟’() between [𝐶𝑖𝑡𝑟𝑖𝑛𝑒]∗ and detectable mature Citrine fluorescence protein [𝐶𝑖𝑡𝑟𝑖𝑛𝑒] (Supplementary Text). The 𝑟’() is set to 0.00385 min-1, corresponding to a 3-hour conversion time (Table S4, Fig. S1). Fig. S7. Pairwise comparison of Citrine traces in Fig. 4. For each pair of traces, we performed pixel-wise Welch’s t-test along the distance to the sender- receiver interface. The red line on the distance versus log10 p value plot is log10(0.05). Fig. S8. Twsg1 addition or BMP-1 expression minimally affects BMP4 gradients without Chordin. In the sender region, Sender-B cells were mixed with filler cells (Sender-C, see Materials and Methods). 4-OHT and (A) rTwsg1 or (B) ABA were added together, and images were taken 48 hours after induction. In both (A) and (B), the white line on each image labels the position of sender-receiver interface. The Citrine fluorescence is shown as yellow. We removed the mCherry channel from images to avoid interfering with the Citrine visualization. 4 µM 4-OHT was added to all samples. N denotes the number of replicates. Fig. S9. The mathematical model recapitulates the dynamic properties of shuttling. We generated the simulated dynamic gradient formation data using the same mathematical model in Fig. 4 (also see Supplementary Text), with all the parameters being the same except for the scaling factor of Citrine fluorescence (b). (A) Simulated data recapitulates the dynamic features of gradient formation for 4-OHT only condition (Fig. 2B and Fig. S2A), including gradient shapes approach a steady state at around 30hr (red line of normalized citrine) and instantaneous signal first increases near the source, then spread to more distal regions, and begin to diminish near the source after 30hr, possibly due to receptor saturation. (B) Simulated data recapitulates the dynamic features of shuttling. The dotted lines and bidirectional arrows between dotted lines in the top two instantaneous signal plots show that, both in model and in experiment, Chordin addition by Dox causes a delay of the Citrine signal to reach a detectable level. The arrows in the bottom instantaneous signal plot show that, both in model and in experiment, the Citrine signal in 4-OHT+Dox+rTwsg1+ABA condition initiates near the final displaced peak position and spreads outwards. Fig. S10. Raw flow cytometry traces and statistical tests of sender-receiver co-culture experiment in Fig. 6A. (A) These raw flow cytometry traces are from one of the four replicate experiments. To calculate the log2 fold change in Fig. 6A, we first obtained the mean Citrine from each trace, then calculated the log2 fold change by the equation log2[(Receiver Citrine - Cit0) / (Cit1 - Cit0)], where Cit0 is the receiver Citrine of the sample without 4-OHT (gray trace) and Cit1 is the receiver Citrine of the sample with only 4-OHT induction (red trace). The number in the parenthesis in the figure legend corresponds to the condition number on the right. (B) For each matrix entry, the number is the p value of Welch’s t-test between the corresponding pair of conditions. The condition number corresponds to the condition number in (A). Fig. S11. Pairwise comparison of Citrine traces in Fig. 6C. For each pair of traces, we performed pixel-wise Welch’s t-test along the distance to the sender- receiver interface. The red line on the distance versus log10 p value plot is log10(0.05). Table S1. List of BMP ligands, receptors and modulators expressed in NMuMG cells (data from Antebi et al., 2017). Gene (Ligand) FPKM Gene (Receptor) FPKM Gene (Modulator) FPKM Bmp2 Bmp3 Bmp3b Bmp4 Bmp5 Bmp6 Bmp7 Bmp8a Bmp8b Bmp9 Bmp10 Bmp15 0.966488 0 0 6.17382 0 0.128872 1.06013 0.240426 0.0148391 0 0 0 Bmpr1a Bmpr1b Bmpr2 Acvr1 Acvr2a Acvr2b 18.6297 0 5.64062 65.3538 54.2251 105.359 Nog Chrd Twsg1 Bmp1 Bmper Bambi 0.0946048 0.132197 57.4595 24.9904 0.486271 0.956501 Dragon/Rgmb 6.41953 Chrdl1 Chrdl2 Fst Fstl1 Fstl5 Sost Sostdc-1 Nbl1 Cer1 Grem1 Grem2 Crim1 Kcp 0 0.0308965 3.12281 1.01323 0 0 0 37.5115 0 0 0 37.1051 0.433142 Gapdh 39.5971 Table S2. List of plasmids used in this study. Index Construct name Cell line pJM009 PB-PGK-ERT2-Gal4-T2A-H2B-Citrine-SV40-HygroR Sender-B pJM018 PB-UAS-BMP4-IRES-H2B-mCherry-BGHpA-SV40- BlastR Sender-B pRZ007 PB-EF1α-TET3G-IRES-H2B-citrine-BGHpA-SV40- NeoR Sender-C, Sender-N, Sender- S pRZ011 PB-TRE3G-Chordin-IRES-H2B-mTurquoise2-SV40- Sender-C ZeoR pRZ012 PB-TRE3G-Noggin-IRES-H2B-mTurquoise2-SV40- Sender-N ZeoR pRZ018 PB-TRE3G-Sog-IRES-H2B-mTurquoise2-SV40-ZeoR Sender-S pRZ044 PB-UAS-BMP-1-IRES-H2B-mCherry-SV40-BlastR Receiver-B1 pRZ056 PB-EF1α-NLS-VP16-PYL-IRES-NLS-Gal4DBD-ABI- Receiver-B1 SV40-NeoR pRZ032 PB-EF1α-IRES-H2B-mTurquoise2-BGHpA-SV40- Sender-B* NeoR ES006 PB-CAG-H2B-mTurqoise2-BGHpA-SV40-HygroR Sender-C*, Sender-N*, Sender-S* Note: PB = PiggyBac backbone; ERT2-Gal4 = Gal4-VP16 transcriptional activator fused with human estrogen receptor (variant ERT2) (46); Gal4DBD = DNA-binding domain of Gal4; Tet3G = Tet-On 3G transactivator protein from Takara Bio; PYL, ABI = domains from PYL1 and ABI1 genes that confer ABA-induced proximity (47); PGK = constitutive promoter from mouse phosphoglycerate kinase 1 gene; UAS = inducible promoter with ERT2-Gal4 binding site; EF1α = constitutive EF1α promoter; TRE3G = inducible promoter with Tet3G binding sites; CAG = constitutive CAG promoter (75); SV40 = constitutive promoter from the early promoter of the simian virus 40 NLS = nuclear localization sequence; IRES = internal ribosome entry site; BGHpA = bovine growth hormone polyadenylation signal; HygroR, BlastR, NeoR, ZeoR = antibiotics resistance genes for hygromycin, blasticidin, geneticin/neomycin and zeocin, respectively; Construct maps in GenBank format are available at data.caltech.edu/records/0sdrn-73r13. Table S3. List of stable cell lines constructed for this study and their use in the figures. Cell lines Parental cells Figures Polyclonal or Monoclonal Integrated constructs Sender-B NMuMG (ATCC) Monoclonal Sender-C NMuMG (ATCC) Monoclonal Sender-N NMuMG (ATCC) Monoclonal Sender-S NMuMG (ATCC) Monoclonal Receiver- B1 NMuMG Sensor Line from (45) Monoclonal Sender-B* Sender-B Monoclonal Sender-C* Sender-C Monoclonal Sender-N* Sender-N Sender-S* Sender-S Polyclonal Polyclonal pJM009, pJM018 pRZ007, pRZ011 pRZ007, pRZ012 pRZ007, pRZ018 pRZ044, pRZ056 pRZ032 ES006 ES006 ES006 2-5, 6C, S2, S4C, S4D, S5, S8, Movie S1, Movie S5-8 3B, 4, 5, S5, S8, Movie S1, Movie S5-8 S4C, S4D 6C 2- 6, S1-5, S8, S10, Movie S1-8 3A, 6B, S3, S4B, S10 3A, S3 S4B 6B, S10 Table S4. List of parameters used in the model. Estimated range in the literature Parameters Value used in the model References 𝐷", 𝐷#, 𝐷$, 𝐷"#, 𝐷"#$ 0.1 – 20 µm2/s 𝑘% 𝑟% 𝑅𝑇𝑜𝑡𝑎𝑙 𝛾 𝑘"# 𝑟"# 𝑘"#$ 𝑟"#$ 𝛽 𝑏 𝑛 𝐾 𝛿’() 𝑟’() 0.00168 – 0.0438 nM-1min-1 0.018 – 0.09 min-1 0.0375 – 2.7 nM (corresponding to 18 – 1300 molecules/µm2 and a 800 µm Matrigel layer on the cell surface) 0.00693 – 0.173 min-1 (corresponding to t1/2 of 4 – 100 min) 0.0168 – 0.0234x10-2 nM-1min-1 15 µm2/s 0.00326 nM-1min-1 (fitted within literature range) 0.09 min-1 (fitted within literature range) 0.57 nM (fitted within literature range) 0.00693 min-1 (fitted within literature range) 0.018 nM-1min-1 (33, 34, 76, 77) (78–81) (78–81) (82, 83) (84–92) (58, 93) (58, 93) 0.003 – 0.204 min-1 0.06 min-1 N/A 0.0554 – 0.9 min-1 (corresponding to Chordin-Twsg1 dissociation constant ranging 3.08 – 50 nM, calculated by chosen kBCT) N/A N/A N/A N/A 0.000481 min-1 (corresponding to t1/2 of 24 hr) N/A 0.018 nM-1min-1 * N/A 0.0554 min-1 (40, 71) 0.002 (basal) – 0.01 (fully induced) min-1 # N/A 15.8 a.u./min (fitted) N/A 1.4 (fitted) 0.0126 nM (fitted) 0.000481 min-1 N/A N/A (70) 0.00385 min-1 (corresponding to t1/2 of 3 hr) Fig. S1. * Lacking direct measurements of 𝑘"#$, we set it arbitrarily to the same value as 𝑘"#, which lies within the diffusion limited range. We also verified that major conclusions are not sensitive to its precise value (Fig. S6). # Twsg has been shown to enhance Chordin cleavage by BMP-1 (18, 74), but we did not find a good estimate of the magnitude of this effect. We set the cleavage rate to be the same for Chordin with or without Twsg, but also verified that major conclusions are not sensitive to whether cleavage rates are set to be the same or not (Fig. S6). Movie S1. One replicate of the time-lapse movie of BMP4 gradients in the no induction condition. Movie S2. One replicate of the time-lapse movie of Receiver-B1 cells turning on Citrine fluorescence after the addition of 5 ng/ml rBMP4. Movie S3. One replicate of the time-lapse movie of Receiver-B1 cells turning on Citrine fluorescence after the addition of 10 ng/ml rBMP4. Movie S4. One replicate of the time-lapse movie of Receiver-B1 cells turning on Citrine fluorescence after the addition of 20 ng/ml rBMP4. Movie S5. One replicate of the time-lapse movie of BMP4 gradients in the 4 µM 4-OHT only condition. Movie S6. Another replicate of the time-lapse movie of BMP4 gradients in the 4 µM 4-OHT only condition. Movie S7. One replicate of the time-lapse movie of BMP4 gradients in the 4 µM 4-OHT + 30 ng/ml Dox condition. Movie S8. One replicate of the time-lapse movie of BMP4 gradients in the 4 µM 4-OHT + 30 ng/ml Dox + 10 nM rTwsg1 + 500 µM ABA condition. REFERENCES AND NOTES 1. G. Struhl, K. Struhl, P. M. Macdonald, The gradient morphogen bicoid is a concentration-dependent transcriptional activator. Cell 57, 1259–1273 (1989). 2. W. Driever, C. Nüsslein-Volhard, A gradient of bicoid protein in Drosophila embryos. Cell 54, 83–93 (1988). 3. W. Driever, C. Nüsslein-Volhard, The bicoid protein determines position in the Drosophila embryo in a concentration-dependent manner. Cell 54, 95–104 (1988). 4. A. M. Turing, The chemical basis of morphogenesis. Philos. Trans. R. Soc. Lond., B, Biol. Sci. 237, 37– 72 (1952). 5. K. A. Wharton, R. P. Ray, W. M. Gelbart, An activity gradient of decapentaplegic is necessary for the specification of dorsal pattern elements in the Drosophila embryo. Development 117, 807–822 (1993). 6. L. Dale, F. C. Wardle, A gradient of BMP activity specifies dorsal–ventral fates in early Xenopus embryos. Semin. Cell Dev. Biol. 10, 319–326 (1999). 7. L. Marchant, C. Linker, P. Ruiz, N. Guerrero, R. Mayor, The inductive properties of mesoderm suggest that the neural crest cells are specified by a BMP gradient. Dev. Biol. 198, 319–329 (1998). 8. J. Raspopovic, L. Marcon, L. Russo, J. Sharpe, Digit patterning is controlled by a Bmp-Sox9-Wnt Turing network modulated by morphogen gradients. Science 345, 566–570 (2014). 9. H. Spemann, Embryonic Development and Induction (Taylor & Francis, 1988). 10. D. Ben-Zvi, B.-Z. Shilo, A. Fainsod, N. Barkai, Scaling of the BMP activation gradient in Xenopus embryos. Nature 453, 1205–1211 (2008). 11. H. Inomata, T. Shibata, T. Haraguchi, Y. Sasai, Scaling of dorsal-ventral patterning by embryo size- dependent degradation of spemann’s organizer signals. Cell 153, 1296–1311 (2013). 12. C. E. Peluso, D. Umulis, Y.-J. Kim, M. B. O’Connor, M. Serpe, Shaping BMP morphogen gradients through enzyme-substrate interactions. Dev. Cell 21, 375–383 (2011). 13. D. Umulis, M. B. O’Connor, S. S. Blair, The extracellular regulation of bone morphogenetic protein signaling. Development 136, 3715–3728 (2009). 14. E. Bier, E. M. De Robertis, EMBRYO DEVELOPMENT. BMP gradients: A paradigm for morphogen-mediated developmental patterning. Science 348, aaa5838 (2015). 15. S. Piccolo, Y. Sasai, B. Lu, E. M. De Robertis, Dorsoventral patterning in Xenopus: Inhibition of ventral signals by direct binding of chordin to BMP-4. Cell 86, 589–598 (1996). 16. J. J. Ross, O. Shimmi, P. Vilmos, A. Petryk, H. Kim, K. Gaudenz, S. Hermanson, S. C. Ekker, M. B. O’Connor, J. L. Marsh, Twisted gastrulation is a conserved extracellular BMP antagonist. Nature 410, 479–483 (2001). 17. C. Chang, D. A. Holtzman, S. Chau, T. Chickering, E. A. Woolf, L. M. Holmgren, J. Bodorova, D. P. Gearing, W. E. Holmes, A. H. Brivanlou, Twisted gastrulation can function as a BMP antagonist. Nature 410, 483–487 (2001). 18. I. C. Scott, I. L. Blitz, W. N. Pappano, S. A. Maas, K. W. Cho, D. S. Greenspan, Homologues of twisted gastrulation are extracellular cofactors in antagonism of BMP signalling. Nature 410, 475–478 (2001). 19. O. Shimmi, M. B. O’Connor, Physical properties of Tld, Sog, Tsg and Dpp protein interactions are predicted to help create a sharp boundary in Bmp signals during dorsoventral patterning of the Drosophila embryo. Development 130, 4673–4682 (2003). 20. A. Eldar, R. Dorfman, D. Weiss, H. Ashe, B.-Z. Shilo, N. Barkai, Robustness of the BMP morphogen gradient in Drosophila embryonic patterning. Nature 419, 304–308 (2002). 21. B.-Z. Shilo, M. Haskel-Ittah, D. Ben-Zvi, E. D. Schejter, N. Barkai, Creating gradients by morphogen shuttling. Trends Genet. 29, 339–347 (2013). 22. M. B. O’Connor, D. Umulis, H. G. Othmer, S. S. Blair, Shaping BMP morphogen gradients in the Drosophila embryo and pupal wing. Development 133, 183–193 (2006). 23. I. C. Scott, I. L. Blitz, W. N. Pappano, Y. Imamura, T. G. Clark, B. M. Steiglitz, C. L. Thomas, S. A. Maas, K. Takahara, K. W. Cho, D. S. Greenspan, Mammalian BMP-1/Tolloid-related metalloproteinases, including novel family member mammalian Tolloid-like 2, have differential enzymatic activities and distributions of expression relevant to patterning and skeletogenesis. Dev. Biol. 213, 283–300 (1999). 24. M. Serpe, D. Umulis, A. Ralston, J. Chen, D. J. Olson, A. Avanesov, H. Othmer, M. B. O’Connor, S. S. Blair, The BMP-binding protein Crossveinless 2 is a short-range, concentration-dependent, biphasic modulator of BMP signaling in Drosophila. Dev. Cell 14, 940–953 (2008). 25. A. Ralston, S. S. Blair, Long-range Dpp signaling is regulated to restrict BMP signaling to a crossvein competent zone. Dev. Biol. 280, 187–200 (2005). 26. L. Zakin, C. A. Metzinger, E. Y. Chang, C. Coffinier, E. M. De Robertis, Development of the vertebral morphogenetic field in the mouse: Interactions between Crossveinless-2 and Twisted Gastrulation. Dev. Biol. 323, 6–18 (2008). 27. L. Zakin, E. Y. Chang, J.-L. Plouhinec, E. M. De Robertis, Crossveinless-2 is required for the relocalization of Chordin protein within the vertebral field in mouse embryos. Dev. Biol. 347, 204–215 (2010). 28. B. Reversade, E. M. De Robertis, Regulation of ADMP and BMP2/4/7 at opposite embryonic poles generates a self-regulating morphogenetic field. Cell 123, 1147–1160 (2005). 29. S. Matsuda, O. Shimmi, Directional transport and active retention of Dpp/BMP create wing vein patterns in Drosophila. Dev. Biol. 366, 153–162 (2012). 30. C. M. Mizutani, Q. Nie, F. Y. M. Wan, Y.-T. Zhang, P. Vilmos, R. Sousa-Neves, E. Bier, J. L. Marsh, A. D. Lander, Formation of the BMP activity gradient in the Drosophila embryo. Dev. Cell 8, 915–924 (2005). 31. O. Shimmi, D. Umulis, H. Othmer, M. B. O’Connor, Facilitated transport of a Dpp/Scw heterodimer by Sog/Tsg leads to robust patterning of the Drosophila blastoderm embryo. Cell 121, 493 (2005). 32. Y.-C. Wang, E. L. Ferguson, Spatial bistability of Dpp–receptor interactions during Drosophila dorsal–ventral patterning. Nature 434, 229–234 (2005). 33. J. Zinski, Y. Bu, X. Wang, W. Dou, D. Umulis, M. C. Mullins, Systems biology derived source-sink mechanism of BMP gradient formation. eLife 6, e22199 (2017). 34. A. P. Pomreinke, G. H. Soh, K. W. Rogers, J. K. Bergmann, A. J. Bläßle, P. Müller, Dynamics of BMP signaling and distribution during zebrafish dorsal-ventral patterning. eLife 6, e25861 (2017). 35. M. Oelgeschläger, J. Larraín, D. Geissert, E. M. De Robertis, The evolutionarily conserved BMP- binding protein Twisted gastrulation promotes BMP signalling. Nature 405, 757–763 (2000). 36. S. Piccolo, E. Agius, B. Lu, S. Goodman, L. Dale, E. M. De Robertis, Cleavage of Chordin by Xolloid metalloprotease suggests a role for proteolytic processing in the regulation of Spemann organizer activity. Cell 91, 407–416 (1997). 37. J. Larrain, D. Bachiller, B. Lu, E. Agius, S. Piccolo, E. M. De Robertis, BMP-binding modules in chordin: A model for signalling regulation in the extracellular space. Development 127, 821–830 (2000). 38. H. Troilo, A. L. Barrett, A. V. Zuk, M. P. Lockhart-Cairns, A. P. Wohl, C. P. Bayley, R. Dajani, R. B. Tunnicliffe, L. Green, T. A. Jowitt, G. Sengle, C. Baldock, Structural characterization of twisted gastrulation provides insights into opposing functions on the BMP signalling pathway. Matrix Biol. 55, 49–62 (2016). 39. P. Li, J. S. Markson, S. Wang, S. Chen, V. Vachharajani, M. B. Elowitz, Morphogen gradient reconstitution reveals Hedgehog pathway design principles. Science 360, 543–548 (2018). 40. R. Sekine, T. Shibata, M. Ebisuya, Synthetic mammalian pattern formation driven by differential diffusivity of Nodal and Lefty. Nat. Commun. 9, 5456 (2018). 41. H. Yano, H. Uchida, T. Iwasaki, M. Mukai, H. Akedo, K. Nakamura, S. Hashimoto, H. Sabe, Paxillin α and Crk-associated substrate exert opposing effects on cell migration and contact inhibition of growth through tyrosine phosphorylation. Proc. Natl. Acad. Sci. 97, 9076–9081 (2000). 42. Y. E. Antebi, J. M. Linton, H. Klumpe, B. Bintu, M. Gong, C. Su, R. McCardell, M. B. Elowitz, Combinatorial signal perception in the BMP pathway. Cell 170, 1184–1196.e24 (2017). 43. H. E. Klumpe, M. A. Langley, J. M. Linton, C. J. Su, Y. E. Antebi, M. B. Elowitz, The context- dependent, combinatorial logic of BMP signaling. Cell Syst. 13, 388–407.e10 (2022). 44. S. S. Gerety, M. A. Breau, N. Sasai, Q. Xu, J. Briscoe, D. G. Wilkinson, An inducible transgene expression system for zebrafish and chick. Development 140, 2235–2243 (2013). 45. F.-S. Liang, W. Q. Ho, G. R. Crabtree, Engineering the ABA plant stress pathway for regulation of induced proximity. Sci. Signal. 4, rs2 (2011). 46. A. D. Lander, Q. Nie, F. Y. M. Wan, Do morphogen gradients arise by diffusion? Dev. Cell 2, 785– 796 (2002). 47. M. Paulsen, S. Legewie, R. Eils, E. Karaulanov, C. Niehrs, Negative feedback in the bone morphogenetic protein 4 (BMP4) synexpression group governs its dynamic signaling range and canalizes development. Proc. Natl. Acad. Sci. U.S.A. 108, 10202–10207 (2011). 48. L. Zakin, E. M. De Robertis, Extracellular regulation of BMP signaling. Curr. Biol. 20, R89–R92 (2010). 49. P. Tiwari, H. Rengarajan, T. E. Saunders, Scaling of internal organs during Drosophila embryonic development. Biophys. J. 120, 4264–4276 (2021). 50. R. Mateus, L. Holtzer, C. Seum, Z. Hadjivasiliou, M. Dubois, F. Jülicher, M. Gonzalez-Gaitan, BMP signaling gradient scaling in the zebrafish pectoral fin. Cell Rep. 30, 4292–4302.e7 (2020). 51. J.-L. Plouhinec, L. Zakin, Y. Moriyama, E. M. De Robertis, Chordin forms a self-organizing morphogen gradient in the extracellular space between ectoderm and mesoderm in the Xenopus embryo. Proc. Natl. Acad. Sci. 110, 20372–20379 (2013). 52. T. Kurata, J. Nakabayashi, T. S. Yamamoto, M. Mochii, N. Ueno, Visualization of endogenous BMP signaling during Xenopus development. Differentiation 67, 33–40 (2001). 53. J. Tucker, K. Mintzer, M. Mullins, The BMP signaling gradient patterns dorsoventral tissues in a temporally progressive manner along the anteroposterior axis. Dev. Cell 14, 108–119 (2008). 54. M.-C. Ramel, C. S. Hill, The ventral to dorsal BMP activity gradient in the early zebrafish embryo is determined by graded expression of BMP ligands. Dev. Biol. 378, 170–182 (2013). 55. H. Troilo, A. V. Zuk, R. B. Tunnicliffe, A. P. Wohl, R. Berry, R. F. Collins, T. A. Jowitt, G. Sengle, C. Baldock, Nanoscale structure of the BMP antagonist chordin supports cooperative BMP binding. Proc. Natl. Acad. Sci. U.S.A. 111, 13063–13068 (2014). 56. P. Blader, S. Rastegar, N. Fischer, U. Strähle, Cleavage of the BMP-4 antagonist chordin by zebrafish tolloid. Science 278, 1937–1940 (1997). 57. R. W. Padgett, J. M. Wozney, W. M. Gelbart, Human BMP sequences can confer normal dorsal- ventral patterning in the Drosophila embryo. Proc. Natl. Acad. Sci. U.S.A. 90, 2905–2909 (1993). 58. T. K. Sampath, K. E. Rashka, J. S. Doctor, R. F. Tucker, F. M. Hoffmann, Drosophila transforming growth factor beta superfamily proteins induce endochondral bone formation in mammals. Proc. Natl. Acad. Sci. U.S.A. 90, 6004–6008 (1993). 59. G. Marqués, M. Musacchio, M. J. Shimell, K. Wünnenberg-Stapleton, K. W. Cho, M. B. O’Connor, Production of a DPP activity gradient in the early Drosophila embryo through the opposing actions of the SOG and TLD proteins. Cell 91, 417–426 (1997). 60. A. Madamanchi, M. C. Mullins, D. M. Umulis, Diversity and robustness of bone morphogenetic protein pattern formation. Development 148, dev192344 (2021). 61. H. L. Ashe, M. Levine, Local inhibition and long-range enhancement of Dpp signal transduction by Sog. Nature 398, 427–431 (1999). 62. L. Wolpert, Positional information and the spatial pattern of cellular differentiation. J. Theor. Biol. 25, 1–47 (1969). 63. K. Miyazono, Y. Kamiya, M. Morikawa, Bone morphogenetic protein receptors and signal transduction. J. Biochem. 147, 35–51 (2009). 64. A. Bandyopadhyay, K. Tsuji, K. Cox, B. D. Harfe, V. Rosen, C. J. Tabin, Genetic analysis of the roles of BMP2, BMP4, and BMP7 in limb patterning and skeletogenesis. PLOS Genet. 2, e216 (2006). 65. K.-S. Choi, C. Lee, D. M. Maatouk, B. D. Harfe, Bmp2, Bmp4 and Bmp7 are co-required in the mouse AER for normal digit patterning but not limb outgrowth. PLOS ONE 7, e37826 (2012). 66. D. Ben-Zvi, A. Fainsod, B.-Z. Shilo, N. Barkai, Scaling of dorsal-ventral patterning in the Xenopus laevis embryo. Bioessays 36, 151–156 (2014). 67. N. Kitsera, A. Khobta, B. Epe, Destabilized green fluorescent protein detects rapid removal of transcription blocks after genotoxic exposure. Biotechniques 43, 222–227 (2007). 68. J.-L. Zhang, Y. Huang, L.-Y. Qiu, J. Nickel, W. Sebald, Von Willebrand factor type C domain- containing proteins regulate bone morphogenetic protein signaling through different recognition mechanisms. J. Biol. Chem. 282, 20002–20014 (2007). 69. A. Hartung, K. Bitton-Worms, M. M. Rechtman, V. Wenzel, J. H. Boergermann, S. Hassel, Y. I. Henis, P. Knaus, Different routes of bone morphogenic protein (BMP) receptor endocytosis influence BMP signaling. Mol. Cell. Biol. 26, 7791–7805 (2006). 70. M. Schlosshauer, D. Baker, Realistic protein-protein association rates from a simple diffusional model neglecting long-range interactions, free energy barriers, and landscape ruggedness. Protein Sci. 13, 1660–1669 (2004). 71. J. Larraín, M. Oelgeschläger, N. I. Ketpura, B. Reversade, L. Zakin, E. M. De Robertis, Proteolytic cleavage of Chordin as a switch for the dual activities of Twisted gastrulation in BMP signaling. Development 128, 4439–4447 (2001). 72. J. Miyazaki, S. Takaki, K. Araki, F. Tashiro, A. Tominaga, K. Takatsu, K. Yamamura, Expression vector system based on the chicken β-actin promoter directs efficient production of interleukin-5. Gene 79, 269–277 (1989). 73. S. Zhou, W.-C. Lo, J. L. Suhalim, M. A. Digman, E. Gratton, Q. Nie, A. D. Lander, Free extracellular diffusion creates the Dpp morphogen gradient of the Drosophila wing disc. Curr. Biol. 22, 668–675 (2012). 74. A. Kicheva, P. Pantazis, T. Bollenbach, Y. Kalaidzidis, T. Bittig, F. Jülicher, M. González-Gaitán, Kinetics of morphogen gradient formation. Science 315, 521–525 (2007). 75. T. Kirsch, J. Nickel, W. Sebald, BMP-2 antagonists emerge from alterations in the low-affinity binding epitope for receptor BMPR-II. EMBO J. 19, 3314–3324 (2000). 76. S. Saremba, J. Nickel, A. Seher, A. Kotzsch, W. Sebald, T. D. Mueller, Type I receptor binding of bone morphogenetic protein 6 is dependent on N-glycosylation of the ligand. FEBS J. 275, 172–183 (2008). 77. T. Hatta, H. Konishi, E. Katoh, T. Natsume, N. Ueno, Y. Kobayashi, T. Yamazaki, Identification of the ligand-binding site of the BMP type IA receptor for BMP-4. Biopolymers 55, 399–406 (2000). 78. V. Khodr, P. Machillot, E. Migliorini, J.-B. Reiser, C. Picart, High-throughput measurements of bone morphogenetic protein/bone morphogenetic protein receptor interactions using biolayer interferometry. Biointerphases 16, 031001 (2021). 79. Y. Chen, A. C. Munteanu, Y.-F. Huang, J. Phillips, Z. Zhu, M. Mavros, W. Tan, Mapping receptor density on live cells by using fluorescence correlation spectroscopy. Chemistry 15, 5327–5336 (2009). 80. A. F. Eckert, P. Gao, J. Wesslowski, X. Wang, J. Rath, K. Nienhaus, G. Davidson, G. U. Nienhaus, Measuring ligand-cell surface receptor affinities with axial line-scanning fluorescence correlation spectroscopy. eLife 9, e55286 (2020). 81. A. L. Schwartz, S. E. Fridovich, H. F. Lodish, Kinetics of internalization and recycling of the asialoglycoprotein receptor in a hepatoma cell line. J. Biol. Chem. 257, 4230–4237 (1982). 82. A. Ciechanover, A. L. Schwartz, A. Dautry-Varsat, H. F. Lodish, Kinetics of internalization and recycling of transferrin and the transferrin receptor in a human hepatoma cell line. Effect of lysosomotropic agents. J. Biol. Chem. 258, 9681–9689 (1983). 83. A. L. Schwartz, A. Bolognesi, S. E. Fridovich, Recycling of the asialoglycoprotein receptor and the effect of lysosomotropic amines in hepatoma cells. J. Cell Biol. 98, 732–738 (1984). 84. G. J. Strous, A. Du Maine, J. E. Zijderhand-Bleekemolen, J. W. Slot, A. L. Schwartz, Effect of lysosomotropic amines on the secretory pathway and on the recycling of the asialoglycoprotein receptor in human hepatoma cells. J. Cell Biol. 101, 531–539 (1985). 85. W. M. Pardridge, A. J. Van Herle, R. T. Naruse, G. Fierer, A. Costin, In vivo quantification of receptor-mediated uptake of asialoglycoproteins by rat liver. J. Biol. Chem. 258, 990–994 (1983). 86. P. R. Dragsten, D. B. Mitchell, G. Covert, T. Baker, Drug delivery using vesicles targeted to the hepatic asialoglycoprotein receptor. Biochim. Biophys. Acta. 926, 270–279 (1987). 87. C. T. H. Jonker, C. Deo, P. J. Zager, A. N. Tkachuk, A. M. Weinstein, E. Rodriguez-Boulan, L. D. Lavis, R. Schreiner, Accurate measurement of fast endocytic recycling kinetics in real time. J. Cell Sci. 133, jcs231225 (2019). 88. S. N. Roed, P. Wismann, C. R. Underwood, N. Kulahin, H. Iversen, K. A. Cappelen, L. Schäffer, J. Lehtonen, J. Hecksher-Soerensen, A. Secher, J. M. Mathiesen, H. Bräuner-Osborne, J. L. Whistler, S. M. Knudsen, M. Waldhoer, Real-time trafficking and signaling of the glucagon-like peptide-1 receptor. Mol. Cell. Endocrinol. 382, 938–949 (2014). 89. K. M. Mayle, A. M. Le, D. T. Kamei, The intracellular trafficking pathway of transferrin. Biochim. Biophys. Acta. 1820, 264–281 (2012). 90. F. Rentzsch, J. Zhang, C. Kramer, W. Sebald, M. Hammerschmidt, Crossveinless 2 is an essential positive feedback regulator of Bmp signaling during zebrafish gastrulation. Development 133, 801–811 (2006).
Could not heal snippet
10.1088_1361-6579_ad0f70.pdf
Data availability statement The data cannot be made publicly available upon publication because they contain commercially sensitive information. The data that support the findings of this study are available upon reasonable request from the authors.
Data availability statement The data cannot be made publicly available upon publication because they contain commercially sensitive information. The data that support the findings of this study are available upon reasonable request from the authors.
RECEIVED 10 October 2023 ACCEPTED FOR PUBLICATION 23 November 2023 PUBLISHED 11 December 2023 Physiol. Meas. 44 (2023) 125002 https://doi.org/10.1088/1361-6579/ad0f70 PAPER Magnetocardiography-based coronary artery disease severity assessment and localization using spatiotemporal features , Jiaojiao Pang3,4,5,∗, Dong Xu6, Ruizhe Wang1,2, Fei Xie3,4,5, Yanfei Yang1,2, Jiguang Sun7, Xiaole Han1,2 Yu Li3,4,5, Ruochuan Li3,4,5, Xiaofei Yin3,4,5, Yansong Xu3,4,5, Jiaxin Fan3,4,5, Yiming Dong3,4,5, Xiaohui Wu3,4,5, Xiaoyun Yang3,5,8, Dexin Yu3,5,9, Dawei Wang3,5,9, Yang Gao1,2,6,10 Jinji Sun1,2,6,11,10 1 Key Laboratory of Ultra-Weak Magnetic Field Measurement Technology, Ministry of Education, School of Instrumentation and , Yuguo Chen3,4,5,∗ and Xiaolin Ning1,2,6,11,10 , Min Xiang1,2,6,11,10,∗ , Feng Xu3,4,5, Optoelectronic Engineering, Beihang University, People’s Republic of China 2 Zhejiang Provincial Key Laboratory of Ultra-Weak Magnetic-Field Space and Applied Technology, Hangzhou Innovation Institute, Beihang University, Hangzhou 310051, People’s Republic of China 3 Shandong Key Laboratory for Magnetic Field-free Medicine & Functional Imaging, Institute of Magnetic Field-free Medicine & Functional Imaging, Shandong University, People’s Republic of China 4 Department of Emergency Medicine, Qilu Hospital of Shandong University, Shandong Provincial Clinical Research Center for Emergency and Critical Care Medicine, People’s Republic of China 5 National Innovation Platform for Industry-Education Intearation in Medicine-Engineering Interdisciplinary, Shandong University, People’s Republic of China 6 National Institute of Extremely-Weak Magnetic Field Infrastructure, Hangzhou, People’s Republic of China 7 Hangzhou Nuochi Life Science Co., Ltd, People’s Republic of China 8 Department of Gastroenterology, Qilu Hospital of Shandong University, Shandong Provincial Clinical Research Center for Digestive Disease, People’s Republic of China 9 Department of Radiology, Qilu Hospital of Shandong University, People’s Republic of China 10 Institute of Large-Scale Scientific Facility and Centre for Zero Magnetic Field Science, Beihang University, People’s Republic of China 11 Hefei National Laboratory, People’s Republic of China ∗ Authors to whom any correspondence should be addressed. E-mail: jiaojiaopang@email.sdu.edu.cn, xiang_min@buaa.edu.cn and chen919085@sdu.edu.cn Keywords: coronary artery disease, disease severity and location, feature extraction, magnetocardiography, machine learning Supplementary material for this article is available online Abstract Objective. This study aimed to develop an automatic and accurate method for severity assessment and localization of coronary artery disease (CAD) based on an optically pumped magnetometer magnetocardiography (MCG) system. Approach. We proposed spatiotemporal features based on the MCG one-dimensional signals, including amplitude, correlation, local binary pattern, and shape features. To estimate the severity of CAD, we classified the stenosis as absence or mild, moderate, or severe cases and extracted a subset of features suitable for assessment. To localize CAD, we classified CAD groups according to the location of the stenosis, including the left anterior descending artery (LAD), left circumflex artery (LCX), and right coronary artery (RCA), and separately extracted a subset of features suitable for determining the three CAD locations. Main results. For CAD severity assessment, a support vector machine (SVM) achieved the best result, with an accuracy of 75.1%, precision of 73.9%, sensitivity of 67.0%, specificity of 88.8%, F1-score of 69.8%, and area under the curve of 0.876. The highest accuracy and corresponding model for determining locations LAD, LCX, and RCA were 94.3% for the SVM, 84.4% for a discriminant analysis model, and 84.9% for the discriminant analysis model. Significance. The developed method enables the implementation of an automated system for severity assessment and localization of CAD. The amplitude and correlation features were key factors for severity assessment and localization. The proposed machine learning method can provide clinicians with an automatic and accurate diagnostic tool for interpreting MCG data related to CAD, possibly promoting clinical acceptance. © 2023 Institute of Physics and Engineering in Medicine Physiol. Meas. 44 (2023) 125002 1. Introduction X Han et al Coronary artery disease (CAD) is a leading cause of cardiovascular mortality and morbidity worldwide. Acute chest pain is the most common symptom of CAD (Xiao et al 2021). Approximately one-third of adults in the United States of America have some form of CAD, with more than 17 million suffering from CAD and nearly 10 million suffering from angina (Shin et al 2018, Tsao et al 2023). Some patients with CAD either experience atypical symptoms or present normal physical conditions, thus requiring further testing to detect CAD. The 12- lead electrocardiography (ECG) examination is the most common non-invasive technique for CAD detection. However, more than one-third of patients with acute coronary syndrome present normal results, which may represent an acute and severe manifestation of CAD (Roffi et al 2016). When troponin is used for diagnosis, many patients with acute coronary syndrome exhibit normal troponin levels early in the disease. The gold standard for diagnosing CAD is coronary angiography, which allows for determining the severity and location of CAD; however, its high cost and invasiveness make it difficult to use as a routine diagnostic tool. CAD is preventable, and early diagnosis and appropriate intervention are essential to reduce mortality. Magnetocardiography (MCG) is a non-invasive and passive technique that detects weak magnetic fields generated by the heart’s electrical activity (Stratbucker et al 1963). The main types of magnetometers are superconducting quantum interference devices (SQUIDs) and optically pumped magnetometers (OPMs). Compared with SQUIDs, OPM-based MCG features a small sensor array, light-weight, portable measurement ability, and room-temperature operation (Boto et al 2017, Hill et al 2019, Hill et al 2020, Yang et al 2021) as well as higher sensitivity and lower cost, rendering it more promising for clinical applications. Both ECG and MCG are functional examinations. Compared with the conventional 12-lead ECG, MCG enables non-contact measurements and has a higher sensitivity to tangential and eddy current signals caused by damaged myocardial tissue (Dutz et al 2006), and arrayed multichannel sensors achieve a higher spatial resolution (Tavarozzi et al 2002, Fenici et al 2003), thereby improving the acquisition of functional information from the heart (Smith et al 2007). Although MCG allows for the sensitive acquisition of functional signals from the heart, signal interpretation is time-consuming and requires experienced professionals, thus limiting its widespread clinical application. Therefore, automated MCG signal processing and disease detection may substantially improve clinical acceptance. With the maturing of machine learning algorithms, numerous researchers are introducing them into MCG and ECG research for tasks such as automated disease detection. The acute coronary syndrome was detected using 554 features extracted from 12-lead ECG examinations along with logistic regression, gradient boosting machine, and artificial neural network models, obtaining a sensitivity of 77%, specificity of 76%, positive predictive value of 43%, and negative predictive value of 94% (Al-Zaiti et al 2020). A logistic regression model was developed based on induction coil magnetometers using 10 features and diagnosed ischemic heart disease with a sensitivity of 95.4%, specificity of 35.0%, and negative predictive value of 97.8% (Mooney et al 2017, Ghasemi-Roudsari et al 2018). Six machine learning models based on SQUIDs were used to identify ischemic heart disease, and a back-propagation neural network achieved the highest performance with an accuracy of 78.43%, specificity of 68.18%, and sensitivity of 86.21% (Yosawin et al 2010). CAD detection in patients with chest pain was proposed using 10 two-dimensional (2D) MCG features and a multilayer perceptron based on SQUIDs, achieving an accuracy of 90.0%, sensitivity of 91.4%, and specificity of 87.7% (Xiao et al 2021). An automatic classification system for CAD based on SQUIDs MCG information entropy was proposed using a multilayer perceptron based on linear discriminant analysis to classify 10 patients with coronary stenosis, achieving a sensitivity of 99%, specificity of 97%, accuracy of 98%, positive predictive value of 96%, and negative predictive value of 99% (Steinisch et al 2013). Using 164 MCG features based on SQUIDs and a support vector machine (SVM)—an extreme gradient boosting model for ischemic heart disease detection was proposed with an accuracy of 94.03%. Furthermore, 18 time-domain 2D features were extracted to localize ischemic heart disease with accuracies of 74%, 68%, and 65% for detection at the left anterior descending artery (LAD), left circumflex artery (LCX), and right coronary artery (RCA), respectively (Tao et al 2019). Previous studies have shown promising results for the automated detection of cardiovascular disease. However, some problems remain to be addressed. First, MCG features are mainly extracted from 2D maps, with features extracted from one-dimensional (1D) signals being mostly neglected, particularly regarding the sensitivity assessment of features from different channels. In addition, previous studies have focused on T-wave MCG features, consequently discarding the effects of other waves. Moreover, in clinical situations, physicians must determine the treatment modality based on the severity of CAD. However, existing methods can only detect CAD but cannot estimate its severity. Finally, although models to localize CAD are available, their performance is low and far from meeting clinical applicability. In this study, we developed a system for severity assessment and localization of CAD using OPM-based MCG. The proposed system utilizes spatiotemporal features of 1D MCG signals, including amplitude, correlation, local binary pattern (LBP), and shape features. We modeled CAD severity assessment and 2 Physiol. Meas. 44 (2023) 125002 X Han et al Figure 1. Photographs of the OPM-based MCG system. (a) position of the OPMs array relative to the thorax of the subject, (b) arrangement of 36 OPMs. localization using both conventional and the proposed MCG features, respectively, and compared their effectiveness. In addition, we analyzed the proposed features used for modeling. 2. Materials Figure 1(a) shows the OPM-based MCG system used in this study. This consisted of a semi-open magnetic shielded enclosure, an OPMs array, a magnetic-free bed, and data acquisition device (Han et al 2023). The system noise floor was 7–10 fT/Hz1/2. The operational dynamic range was ±5 nT. The relative position between the sensor array and the subject remained fixed, including the sensor array lower edge center alignment with the xiphoid process of the subject and the vertical distance between the sensor array and the subject’s thorax (2 cm). Figure 1(b) depicts the arrangement of 36 OPMs, spanning a 275 × 275 mm area for acquisition at a 1000 Hz sampling frequency. The OPMs detected the magnetic field signal along the Z axis, which was perpendicular to the sensor array. For each subject, MCG data were collected for a period of 3 min. A dataset was constructed from a prospective study of cardiovascular disease at the Qilu Hospital of Shandong University. All the enrolled subjects underwent coronary angiography or computed tomography angiography (CTA), and each subject had detailed medical records. Data on cardiovascular disease were collected and recorded using an OPM-based MCG system from 402 patients with symptoms of chest pain. A total of 93 cases were excluded based on the inclusion criteria, and 309 cases were enrolled. Among the excluded cases, 67 cases showed other medical conditions, 14 cases had non-removable metal implants in the body, and 12 cases did not have the correct MCG data. The 309 enrolled patients included 70 with stable angina, 208 with unstable angina, 13 with ST-segment elevation myocardial infarction, and 18 with non-ST-segment elevation myocardial infarction. The data from the 309 patients included 257 cases of coronary stenosis and 52 cases without coronary stenosis. In addition, 117 patients underwent CTA, and 192 patients underwent coronary angiography. Based on a coronary stenosis threshold of 50%, LAD (48 cases), LCX (6 cases), RCA (9 cases), LAD and LCX (26 cases), LAD and RCA (21 cases), LCX and RCA (9 cases), and LAD, LCX, and RCA (93 cases) stenoses were detected. Notably, patients commonly exhibited multiple stenoses in their coronary arteries. 3. Methods 3.1. System overview Figure 2 shows the data processing flow of the CAD detection system. The raw MCG data were preprocessed to obtain a signal with a high signal-to-noise ratio. Preprocessing included median filtering to remove baseline noise, notch filtering to remove 50 Hz power frequency noise, lowpass filtering to remove high-frequency noise, and superposition of the averaged heartbeats to remove common-mode noise. In addition, different heartbeat waves (e.g. P, QRS, and T waves) were segmented for feature extraction. Subsequently, four types of MCG features were extracted, and important feature subsets were selected for the system. Finally, the selected features were fed into a machine learning model for severity assessment and localization of CAD. Hierarchical tenfold 3 Physiol. Meas. 44 (2023) 125002 X Han et al Figure 2. Overview of data processing for CAD detection system. cross-validation was applied to improve the robustness of the system, with each fold containing approximately the same proportion of grouped labels. 3.2. Feature extraction 3.2.1. Conventional features The conventional MCG features are listed in table 1. Time, singular value decomposition (SVD), and amplitude class features were extracted from the butterfly diagram. The most common features extracted from ECG are time features, which mainly include the time intervals of P, QRS, and T waves and ST, PR, and QT segments. These features mainly describe changes in different segments of a heartbeat over time (Seki et al 2008, Sutter et al 2020). SVD features depict the number of independent sources in a signal (Tao et al 2019). Amplitude features are typically extracted from ECG to calculate ST segment amplitudes, and common statistical parameters include the maximum value, range, mean, standard deviation, and amplitude area. In the magnetic field, the features typically extracted are the vector amplitude and direction from the negative magnetic pole pointing toward the positive magnetic pole (Beadle et al 2021) and the area ratio between positive and negative magnetic fields (Chen et al 2014). In the current field, the maximum current vector (MCV) and the total current vector (TCV) are extracted as features (Ogata et al 2009). 3.2.2. Proposed features Based on the 36 channels of 1D MCG signals, four classes of MCG features were proposed, as listed in table 2. The 36 OPM sensors in the MCG system and relative positions between the sensor array and the subject were fixed, resulting in a constant Pearson linear correlation coefficient among the sensor signals for normal individuals. As CAD damages the myocardium, the altered electrical conduction pathways in the damaged part were reflected in the sensor array as alterations in the signal of each channel, thus changing the Pearson linear correlation coefficient among the 36 channels. Figure 3(a) shows the MCG butterfly diagrams (Haberkorn et al 2006) of a healthy person and CAD patient, and figure 3(b) presents their 36-channel 1D MCG signals. Figure 3(c) depicts the Pearson linear correlation coefficient of the 36-channel MCG signals in the T wave for the two individuals. The Pearson linear correlation coefficient values and positions among the 36 channels drastically differed between the healthy person and the CAD patient. LBP is an operator used to describe the local texture of an image and is widely used in texture classification, texture segmentation, and face image analysis (Prakasa 2016). We use LBP to describe the variation in the 1D MCG signal by constructing histograms to determine the frequency values of binary patterns, where each 4 Table 1. Conventional MCG features for CAD detection system. Map Butterfly diagram Class Time 5 SVD Amplitude Magnetic field Vector Current field Area MCV TCV Feature P wave time interval QRS wave time interval T wave time interval ST segment time interval PR segment time interval QT segment time interval SVD value SVD entropy Feature identifier Time_P Time_QRS Time_T Time_ST Time_PR Time_QT SVD_Value# SVD_Entropy# Amp_Max Amp_Range Amp_Mean Amp_SD Amp_Area Mag_Dir Mag_Amp Feature description Number of features SVD_Value# contains the top 8 singular values SVD_Entropy# uses 8 singular values to calculate Shannon, Tsallis, and Renyi entropy values 6 11 × 8 5 × 36 × 8 2 1 2 2 1,541 Maximum value of wave Amplitude range of wave Mean value of wave Standard deviation of wave Area of wave Vector direction of negative magnetic pole pointing to the positive magnetic pole at T-wave peak Vector amplitude of negative magnetic pole pointing to the positive magnetic pole at T-wave peak Area ratio between positive and negative magnetic fields at T-wave peak Maximum current vector direction at T-wave peak Maximum current vector amplitude at T-wave peak Total current vector direction at T-wave peak Total current vector amplitude at T-wave peak Mag_Area MCV_Dir MCV_Amp TCV_Dir TCV_Amp Total number of features: The numbers 36 and 8 represent the spatial dimension with 36 channels and temporal dimension with 8 waves (P, QRS, and T waves, ST, ST-T, PR, and QT segments, and the whole heartbeat), respectively. P h y s i o l . M e a s . 4 4 ( 2 0 2 3 ) 1 2 5 0 0 2 X H a n e t a l Physiol. Meas. 44 (2023) 125002 X Han et al Figure 3. MCG signals and features of a healthy person (top) and CAD patient (bottom). (a) MCG butterfly diagram. (b) 36-channel 1D MCG signal. (c) Correlation features in T wave. Table 2. Proposed MCG features for CAD detection system. Map Class Feature Feature identifier Feature description Butterfly diagram Correlation LBP Amplitude Shape Pearson linear correlation coefficient of two-channel waveforms Bin values of LBP histogram Bin values of wave histogram Corr_channel# Corr_channel# contains 630 combinations LBP_Bin# Amp_Bin# LBP_Bin# contains 10 bins Amp_ Bin# contains 10 bins Kurtosis factor Skewness factor Waveform factor Peak factor Pulse factor Margin factor Volatility index Wave_Kurtosis Wave_Skewness Wave_Waveform Wave_Peak Wave_Pulse Wave_Margin Wave_Volatility Number of features 630 × 8 10 × 36 × 8 10 × 36 × 8 7 × 36 × 8 The numbers 36 and 8 represent the spatial dimension with 36 channels and temporal dimension with 8 waves (P, QRS, and T waves, ST, ST- T, PR, and QT segments, and the whole heartbeat), respectively. Total number of features: 12,816 pattern represents the probability of finding a binary pattern in the image. The number of histogram bins is set to 10. Figure 4(a) shows LBP features calculated from the T-wave MCG data of a healthy person and CAD patient at channel 1. The first and tenth bin values considerably differ, likely establishing the difference between the healthy person and CAD patient. We propose using the amplitude histogram as a statistical parameter to compute the amplitude features. Figure 4(b) presents the amplitude features calculated from the T-wave MCG data of a healthy person and CAD patient at channel 1. Ten bin values were distinguished between the individuals. The shape features were extracted from the amplitude and shape changes between individuals as determined from the acquired MCG data. Figure 4(c) depicts the shape features calculated from the T-wave MCG data of a healthy person and CAD patient at channel 1. Seven shape features were distinguished between the individuals. 3.2.3. Fine-grained spatiotemporal features We extracted fine-grained features in the spatiotemporal dimensions based on conventional and proposed MCG features (butterfly diagrams). Fine-grained spatial features are defined by calculating the values for each 6 Physiol. Meas. 44 (2023) 125002 X Han et al Figure 4. MCG features of a healthy person and CAD patient. (a) LBP, (b) amplitude, and (c) shape features in the T wave. channel instead of averaging across the 36 channels. The fine-grained spatial features are applied to amplitude, shape, correlation, and LBP features. The representative features and channels differ, possibly promoting CAD severity assessment and localization. Fine-grained temporal features are defined for different waves, including the P, QRS, and T waves, ST, ST-T, PR, and QT segments, and the whole heartbeat. The fine-grained temporal features are applied to the amplitude, SVD, shape, correlation, and LBP features. Furthermore, representative features selected by the models are distributed in different waves, and waves other than the T wave benefit the models. To consider the spatiotemporal dimensions, the total number of conventional and proposed MCG features used were 1541 and 12 816, respectively. Compared to the population size, the number of conventional and proposed features was enormous and completely imbalanced. Therefore, an effective feature reduction was required (Patel et al 2022). Feature selection reduces the dimensionality of the data by selecting only a subset of the measured features to build a model. The main benefits of feature selection are improved model prediction performance, faster and cost- effective feature subset, and improved data generation process understanding (Guyon and Elisseeff 2003, Richards 2022). The feature selection process was performed prior to inputting the features into the classification model. First, feature importance was evaluated using a feature selection algorithm, and the features were sorted according to the importance score. Second, features with a large Pearson linear correlation coefficient were removed (the threshold for the Pearson linear correlation coefficient was set at 0.55). Finally, the number of selected features was 1/10 of the number of observations in the model. A hybrid feature selection algorithm (Tasci et al 2023) was utilized this study because the obtained results were considered superior to those obtained by separately utilizing filter and embedded feature selection algorithms. Initially, filter-type feature selection, in particular, the Chi-square test (Thaseen et al 2019) was used, followed by embedded-type feature selection which employs the linear discriminant analysis classifier. 3.3. Classification models We evaluated six machine learning classification models: discriminant analysis, naïve Bayes, SVM, k-nearest neighbors (KNN), boosting tree, and bagging tree models. The model with the best classification performance was selected for severity assessment and localization of CAD. Five metrics were calculated to evaluate the models based on the confusion matrix: accuracy, precision, sensitivity, specificity, and F1-score. In addition, the receiver operating characteristic (ROC) curves were obtained, and the AUC was calculated to assess the classification performance. Because CAD severity assessment was a multi-label model, macro-average metrics were used for evaluation. 4. Results To validate the effectiveness of the MCG features proposed in this study, we developed models utilizing conventional MCG features (refer to table 1) and the proposed MCG features (refer to table 2) and then compared the modeling metrics obtained from the two models. 4.1. CAD severity assessment We used coronary angiography and CTA as references to determine the CAD severity. From CTA, we could only estimate mild, moderate, and severe stenoses. A machine learning model for estimating the CAD severity was developed to distinguish between individuals without stenosis and those with mild, moderate, or severe stenosis. Overall, data from 309 individuals were considered for modeling, with the cases distributed as listed in table 3. We selected 31 features based on the feature selection process, and the selected features were fed into the six classification models for training and validation, obtaining the macro-average results listed in supplementary table 1. According to the accuracy metrics, the results of modeling the optimum based on conventional and proposed features are selected, respectively, as shown in table 4. Compared to the conventional feature-based 7 Physiol. Meas. 44 (2023) 125002 X Han et al Figure 5. ROC curves of two models for estimating CAD severity. (a) Bagging tree model (Conventional features), and (b) SVM model (Proposed features). Table 3. Cases for estimating CAD severity. CAD severity No CAD Mild stenosis Moderate stenosis Severe stenosis Coronary stenosis range Class index Number of cases 0% (0, 50%) [50%, 70%) [70%,100%] 1 2 3 4 52 45 39 173 Table 4. Performance of the best classification models for CAD severity assessment. Best classifier Accuracy Precision Sensitivity Specificity F1-score Mean AUC Bagging tree (conventional features, N = 31) SVM (proposed features, N = 31) 58.3 75.1 42.7 73.9 33.7 67.0 79.3 88.8 33.7 69.8 0.656 0.876 N denotes the number of features selected, and the highest metrics are in bold. modeling metrics, the proposed feature-based modeling metrics show a noteworthy improvement, particularly in the increase in accuracy by 16.8%. Figure 5(a) shows the ROC curves of the Bagging tree model (conventional features), and figure 5(b) shows the ROC curves of the SVM model (proposed features). Compared to the AUC values of the four classes using conventional features, all four AUC values using the proposed features exhibit an increase. To further understand the model based on the proposed features, we analyzed the 31 features (refer to supplementary table 2) utilized for the modeling. The classes, waves, and channels of these features were statistically distributed separately, as shown in figure 6. Figure 6(a) indicates the class of features, with a high percentage of features in the amplitude and correlation classes. Figure 6(b) presents the wave of features. All waves are features except for the ST and PR segments, with the highest percentage of features calculated using the whole wave. Figure 6(c) depicts the MCG channels and the number of features extracted per channel. The features were predominantly located on the diagonal, and there are channels with multiple features. In addition, one feature in the correlation class involved two channels, whereas the amplitude, LBP, and shape features involved only one channel. Overall, the performance of the CAD severity model is improved by the combined use of the proposed feature classes, feature extraction waves (temporal dimension), and channels (spatial dimension). 8 Physiol. Meas. 44 (2023) 125002 X Han et al Figure 6. The selected proposed features for predicting CAD severity. Feature (a) classes, (b) waves, and (c) channels. Figure 7. Seven-label model converted into three binary classification models. Table 5. Data distribution for CAD localization formulated as binary classification. Binary classification Positive Negative LAD versus no LAD LCX versus no LCX RCA versus no RCA 188 134 132 24 78 80 4.2. CAD localization In ECG studies, changes in the ST segment of different leads are statistically counted to localize myocardial ischemia (Shu et al 2017). In MCG studies, features reflecting the 2D magnetic field pattern are used to localize coronary stenosis (Tao et al 2019). Accordingly, we investigated whether CAD could be localized using features extracted from 1D MCG signals. CAD localization was applied to patients with moderate and severe coronary stenosis data. A seven-label model was constructed to account for the possible coronary stenosis locations, as illustrated in figure 7. However, building an accurate seven-label model was difficult because of the high imbalance between labels in the collected data. To address this problem, we converted the seven-label model into three binary classification models to independently predict each stenosis location (Tao et al 2019). The data distribution for the three binary classification models is presented in table 5. Compared with the seven-label model, data were more balanced between the two classes of each binary classification model. The three CAD localization models were constructed following a similar process to that used for obtaining the CAD severity model. Note that feature selection was performed for each of the three location models, and three feature sets were selected, each with a number of 21 (the number of observations for the location detection model is 212 cases). We inputted the three selected feature sets into each of the six classification models for training and validation to obtain the metrics for the LAD, LCX, and RCA detection models, respectively (refer to 9 Physiol. Meas. 44 (2023) 125002 X Han et al Table 6. Performance of CAD localization using binary classification. Binary classification Best classifier Accuracy Precision Sensitivity Specificity F1-score AUC LAD versus no LAD LCX versus no LCX RCA versus no RCA Discriminant analysis (Conventional features, N = 21) SVM (Proposed features, N = 21) Discriminant analysis (Conventional features, N = 21) Discriminant analysis (Proposed features, N = 21) Naïve Bayes (Conventional features, N = 21) Discriminant analysis (Proposed features, N = 21) 69.8 94.3 70.8 84.4 62.7 84.9 92.5 98.4 75.0 86.3 69.6 87.3 71.8 95.2 80.6 89.6 71.2 88.6 54.2 87.5 53.8 75.6 48.8 78.8 80.8 0.652 96.8 0.949 77.7 0.711 87.9 0.921 70.4 0.662 88.0 0.900 N denotes the number of features selected, and the highest metrics are in bold. supplementary tables 3–5). Under the accuracy metric, the optimal modeling results based on conventional and proposed features were selected, respectively, as illustrated in table 6. In table 6, all three proposed feature-based location detection model metrics are significantly improved compared to the conventional feature-based modeling metrics. LAD detection has the highest accuracy, sensitivity, and specificity among the three location models. For further insight into the three localization models, we analyzed the three feature sets used for modeling. The three feature sets are specified in supplementary tables 6–8. The feature analysis was performed similarly to the feature analysis of the CAD severity model. The classes, waves, and channels of each of the three feature sets were statistically distributed, as shown in figure 8. Figure 8 shows three lines with the results of the feature sets analysis used by the LAD, LCX, and RCA models, respectively. Figure 8(a) illustrates the class distribution of the three feature sets; the amplitude class features are highest in the LCX and RCA models, and the correlation class features are highest in the LAD model. Figure 8(b) displays the wave distributions of the three feature sets; in the LAD model, the features are distributed in all waves, with the highest percentage in the P wave. In the LCX model, the features are selected to be computed in all waves except for the ST-T segment, with the highest percentage in the QT segment, and in the RCA model, the features are selected to be computed in all waves except the PR segment, with the highest percentage in the QT segment and the whole wave. Figure 8(c) presents the channel distributions of the three feature sets, which have different distributions of the channels where the features are located. In summary, the designed features contain three perspectives: feature class, feature wave (temporal dimension), and feature channel (spatial dimension). The features are extracted from these three perspectives separately for each CAD localization model, and the three CAD localization models obtained perform acceptably. 5. Discussion We investigated the severity assessment and localization of CAD. To determine the CAD severity, we selected a set of representative features evaluated in six classification models. The SVM achieved the best performance, with an accuracy of 75.1%. Overall, the accuracy of the CAD severity assessment model was lower than that of the three CAD localization models for the following reasons. CAD severity required the classification of four classes, whereas CAD localization was formulated as three binary classification models. In addition, there was a higher imbalance across the four classes of CAD severity. These aspects undermined the performance of CAD severity assessment. For CAD localization, we used three binary classification models to separately detect coronary stenosis at the LAD, LCX, and RCA. The average accuracy of CAD localization using proposed MCG features (average accuracy, 88%; LAD, 94%; LCX, 84%; RCA, 85%) improved by 19% compared with that obtained in a previous study (average accuracy, 69%; LAD, 74%; LCX, 68%; RCA, 65%) (Tao et al 2019). Three factors can explain the improved accuracy of CAD localization. (1) We used an MCG feature set combining spatial and temporal features, widely covering important information. (2) The important subsets of features for each of the three CAD localization models were selected separately. (3) We considered data from patients with moderate and severe stenoses and excluded those with mild stenosis. There are 94 features (31 + 21 × 3 = 94) used in the CAD severity and localization model, and we analyze these selected features as shown in figure 9. 10 Physiol. Meas. 44 (2023) 125002 X Han et al Figure 8. The selected proposed features for CAD localization. Feature (a) classes, (b) waves, and (c) channels. Figure 9(a) contains a higher percentage of amplitude and correlation class features, indicating that these two classes of features are more important in the CAD severity and localization model. In terms of physiology, various degrees of coronary artery stenosis or occlusion results in myocardial ischemia in the region supplied by a coronary artery (Nabel and Braunwald 2012). Local ischemia reduces the duration, resting potential, and propagation velocity of action potentials in the affected myocardium, resulting in wide variability in the conduction velocity between different myocardial regions. The variability in conduction velocity between the epicardial and endocardial walls of the affected region leads to signal variability in the specific MCG channels oriented to that region. To extract the signal variability information, we designed and selected channel-specific amplitude, LBP, and shape class features. The percentage of amplitude class features is much higher than that of LBP and shape class features, indicating that the amplitude class features better portray the variation of MCG signals affected by CAD. The variability in the ischemic region and normal myocardium leads to spatial variability between MCG channels. To extract the spatial variability information, we designed and selected the correlation features. The percentage of correlation class features is second only to the amplitude class features in CAD severity and location detection. Figure 9(b) presents the distribution of features in different waves, where the higher percentage is in the whole wave and QT segment rather than in the T wave. This reveals the potential benefits of selecting multiple waves for detecting CAD severity and location. Regarding physiology, the repolarization sequence propagates from the endocardium to the epicardium in the normal ventricle (Zhu et al 2009). When cardiac ischemia occurs, there is a corresponding change in conductivity within the ischemic region, resulting in slower repolarization and impaired ventricular diastole, which is reflected in the alteration of the T wave signal. Myocardial injury may also lead to cardiac electrical conduction system abnormalities, which may cause arrhythmia signals, which are reflected in the P and QRS wave signals. Upon evaluating the screened feature results using the data-driven feature selection method, a contribution from other waves in addition to the T wave could be observed, which is consistent with the myocardial injury physiological process. To summarize, the higher percentage of relevant waves containing T wave (T wave, ST-T segment, QT segment, and whole wave) indicates that ventricular repolarization signals have a major contribution to the detection of disease, but that the other waves (P wave, QRS wave, ST segment, and PR segment) also have a certain role. 11 Physiol. Meas. 44 (2023) 125002 X Han et al Figure 9. The selected proposed features for CAD severity and localization. Feature (a) classes, (b) waves, and (c) channels. Figure 9(c) shows that the features use 35 channels, in which the channels with a higher number of features are mainly distributed in the middle region of the sensing array. It illustrates that the number of sensors can be reduced to lower the cost of the OPM-based MCG system while ensuring the effectiveness of the CAD severity and localization model. A promising finding was that when the top 5 important features of each model (20 features in total) were selected for analysis, there were only 14 channels where the features were located (refer to supplementary figure 1), which gives us an insight that if, in the development of portable MCG devices, the requirements for the detection model metrics are relatively low, it is possible to drastically reduce the number of 36-channel MCG sensors by retaining the sensors only in specific locations. In the ECG, stenosis at different CAD locations resulted in ST segment changes in different leads (Yuan et al 1991, Shu et al 2017). Likewise, we observed that among the features selected for the three CAD location detection models, the distribution of channels where important features were located was different (figure 8(c)). Finally, the primary contribution of this research is the extraction of new MCG features and their validation using clinical data acquired by an OPM-based MCG system. Specifically, owing to the shortcomings of the aforementioned studies, this study proposes spatiotemporal features, including amplitude, correlation, LBP, and shape, obtained from 1D MCG signals. Based on the database and feature selection method used in this study, selected features are considered significant. However, because the database used in this study was relatively small, future research will focus on increasing the database to further validate and optimize the selected features and models. 6. Conclusion We propose a system to automatically estimate the severity and localize CAD. Our system can detect either the absence of stenosis or a mild, moderate, or severe case. In addition, it may promote the clinical diagnostic application of OPM-based MCG. Regarding the MCG features, we can draw three conclusions. (1) Amplitude and correlation features are essential for determining the severity and localizing CAD. (2) T wave signals are not the only waves useful for detecting CAD. Other waves also contribute to severity assessment and localization. (3) Not all channels in the 36-channel sensor array contribute to severity assessment and localization. There is a correlation between the channel distribution of important features and the location of coronary stenosis. Acknowledgments The authors would like to thank Dr Qinghua Sun for the scientific discussions pertaining to this study, the engineers from Hangzhou Nuochi Life Science Co., Ltd for their support during the experiments, and Huidong Wang, Sijia Zhang, Junliang Zhang, Chong Ma for their participation in the work data collection. The author’s have confirmed that any identifiable participants in this study have given their consent for publication. This work was supported in part by the Innovation Program for Quantum Science and Technology under Grant No. 2021ZD0300500, the Development and Application of Ultra-Weak Magnetic Measurement Technology based on Atomic Magnetometer under Grant No. 2022-189-181, the National Natural Science Foundation of China under Grant No. 62101017, and the Key R&D Program of Shandong Province under Grant No. 2022ZLGX03. 12 Physiol. Meas. 44 (2023) 125002 X Han et al Data availability statement The data cannot be made publicly available upon publication because they contain commercially sensitive information. The data that support the findings of this study are available upon reasonable request from the authors. Ethical statement The study was reviewed and approved by the Ethics Committee of Scientific Research of Shandong University Qilu Hospital; the ethics board protocol approval number was KYLL-202204-017, and the approval date was 20 April, 2022. All the subjects provided written informed consent for the experimental procedure conducted in accordance with the Declaration of Helsinki. The trial name was Magnetocardiography in the Accurate Identification of Severe Coronary Lesions and Myocardial Necrosis; the http://clinicalTrials.gov identifier was NCT05392712. ORCID iDs Xiaole Han Yang Gao Min Xiang Jinji Sun Xiaolin Ning https://orcid.org/0000-0001-9152-8182 https://orcid.org/0000-0002-6841-9276 https://orcid.org/0000-0002-0239-3392 https://orcid.org/0000-0002-4804-637X https://orcid.org/0000-0003-3563-3601 References Al-Zaiti S et al 2020 Machine learning-based prediction of acute coronary syndrome using only the pre-hospital 12-lead electrocardiogram Nat. Commun. 11 1–10 Beadle R, Mcdonnell D, Roudsari S G and Unitt L 2021 Assessing heart disease using a novel magnetocardiography device Biomed. Phys. Eng. Express 7 1–9 Boto E et al 2017 A new generation of magnetoencephalography: room temperature measurements using optically-pumped magnetometers Neuroimage 149 404–14 Chen T, Chen Z, Jiang S Q, Leeuwen P V and Grnemeyer D 2014 Noninvasively diagnosing coronary artery disease with 61-channel MCG data Sci. Bull. 59 1123–8 Dutz S, Bellemann M E, Leder U and Haueisen J 2006 Passive vortex currents in magneto- and electrocardiography: comparison of magnetic and electric signal strengths Phys. Med. Biol. 51 145–51 Fenici R, Brisinda D, Nenonen J and Fenici P 2003 Noninvasive study of ventricular preexcitation using multichannel magnetocardiography Pacing Clin. Electrophys. 26 431–5 Ghasemi-Roudsari S, Al-Shimary A, Varcoe B, Byrom R, Kearney L and Kearney M 2018 A portable prototype magnetometer to differentiate ischemic and non-ischemic heart disease in patients with chest pain PLoS One 13 10 Guyon I and Elisseeff A 2003 An introduction to variable and feature selection J. Mach. Learn. Res. 3 1157–82 Haberkorn W, Steinhoff U, Burghoff M, Kosch O, Morguet A and Koch H 2006 Pseudo current density maps of electrophysiological heart, nerve or brain function and their physical basis Biomagn. Res. Technol. 4 1–18 Han X, Xue X, Yang Y, Liang X, Gao Y, Xiang M, Sun J and Ning X 2023 Magnetocardiography using optically pumped magnetometers array to detect acute myocardial infarction and premature ventricular contractions in dogs Phys. Med. Biol. 68 1–14 Hill R M et al 2019 A tool for functional brain imaging with lifespan compliance Nat. Commun. 10 1–11 Hill R M et al 2020 Multi-channel whole-head OPM-MEG: helmet design and a comparison with a conventional system Neuroimage 219 1–20 Kangwanariyakul Y, Nantasenamat C, Tantimongcolwat T and Naenna T 2010 Data mining of magnetocardiograms for prediction of ischemic heart disease EXCLI J. 9 82–95 Mooney J W, Ghasemi-Roudsari S, Banham E R, Symonds C, Pawlowski N and Varcoe B H 2017 A portable diagnostic device for cardiac magnetic field mapping Biomed. Phys. Eng. Express 3 1–18 Nabel E G and Braunwald E 2012 A tale of coronary artery disease and myocardial infarction New Engl. J. Med. 366 54–63 Ogata K, Kandori A, Watanabe Y, Suzuki A and Tanaka K 2009 Repolarization spatial-time current abnormalities in patients with coronary heart disease Pacing Clin. Electrophysiol. 32 516–24 Patel R, Gireesan K, Baskaran R and Shekar N C 2022 Optimal classification of N-back task EEG data by performing effective feature reduction Sādhanā 47 281–92 Prakasa E 2015 Texture feature extraction by using local binary pattern INKOM J. 9 45–8 Richards J A 2022 Remote Sensing Digital Image Analysis (Switzerland: Springer) (https://doi.org/10.1007/978-3-030-82327-6) Roffi M, Patrono C, Collet J P, Mueller C and Valgimigli M 2016 2015 ESC Guidelines for the management of acute coronary syndromes in patients presenting without persistent ST-segment elevation Eur. Heart J. 37 267–315 Seki Y, Muneyuki K, Kandori A, Tsukada K, Terao K and Ageyama N 2008 Standardization of magnetocardiography in nonhuman primates Phys. Med. Biol. 53 1609–18 Shin E S, Park S G, Saleh A, Lam Y Y, Bhak J, Jung F, Morita S and Brachmann J 2018 Magnetocardiography scoring system to predict the presence of obstructive coronary artery disease Clin. Hemorheol. Microcirc. 70 1–9 Shu L, Adam M, Tan J et al 2017 Automated identification of coronary artery disease from short-term 12 lead electrocardiogram signals by using wavelet packet decomposition and common spatial pattern techniques J. Mech. Med. Biol. 17 1–19 13 Physiol. Meas. 44 (2023) 125002 X Han et al Smith F E, Langley P, Leeuwen P V, Hailer B, Trahms L, Steinhoff U, Bourke J P and Murray A 2006 Comparison of magnetocardiography and electrocardiography: a study of automatic measurement of dispersion of ventricular repolarization EP. Europace. 8 887–93 Steinisch M, Torke P R, Haueisen J, Hailer B, Gronemeyer D, Van Leeuwen P and Comani S 2013 Early detection of coronary artery disease in patients studied with magnetocardiography: an automatic classification system based on signal entropy Comput. Biol. Med. 43 144–53 Stratbucker R A, Hyde C M and Wixson S E 1963 The magnetocardiogram–a new approach to the fields surrounding the heart IEEE Trans. Biomed. Electron. 10 145–9 Sutter J U et al 2020 Recording the heart beat of cattle using a gradiometer system of optically pumped magnetometers Comput. Electron. Agric. 177 1–8 Tao R et al 2019 Magnetocardiography-based ischemic heart disease detection and localization using machine learning methods IEEE Trans. Biomed. Eng. 66 1658–67 Tasci E, Jagasia S, Zhuge Y, Camphausen K and Krauze A V 2023 GradWise: a novel application of a rank-based weighted hybrid filter and embedded feature selection method for glioma grading with clinical and molecular characteristics Cancers 15 1–19 Tavarozzi I et al 2002 Magnetocardiography: current status and perspectives: II. Clinical applications Italian Heart J. 3 151–65 Thaseen I S, Kumar C A and Ahmad A 2019 Integrated intrusion detection model using chi-square feature selection and ensemble of classifiers Arab. J. Sci. Eng. 44 3357–68 Tsao C W et al 2023 Heart disease and stroke statistics—2023 update: a report from the american heart association Circulation 147 93–621 Xiao H, Pengfei C, Fakuan T and Ning H 2021 Detection of coronary artery disease in patients with chest pain: an machine learning model based on magnetocardiography parameters Clin. Hemorheol. Microcirc. 78 227–36 Yang Y F, Xu M Z, Liang A M, Yin Y, Ma X, Gao Y and Ning X L 2021 A new wearable multichannel magnetocardiogram system with a SERF atomic magnetometer array Sci. Rep. 11 1–11 Yuan S, Blomström P, Pehrson S and Bertil Olsson S 1991 Localization of cardiac arrhythmias: conventional noninvasive methods Int. J. Cardiac Imaging 7 193–205 Zhu T G, Patel C, Martin S, Quan X, Wu Y, Burke J F, Chernick M, Kowey P R and Yan G-X 2009 Ventricular transmural repolarization sequence: its relationship with ventricular relaxation and role in ventricular diastolic function Eur. Heart J. 30 372–80 14
null
10.1093_molbev_msad069.pdf
Data availability No new data were generated in support of this research. TE models were deposited in the DFAM database.
Data availability No new data were generated in support of this research. TE models were deposited in the DFAM database.
Transposable Element Interactions Shape the Ecology of the Deer Mouse Genome Landen Gozashti ,1 Cedric Feschotte,2 and Hopi E. Hoekstra *,1 1Department of Organismic & Evolutionary Biology, Department of Molecular & Cellular Biology, Museum of Comparative Zoology and Howard Hughes Medical Institute, Harvard University, Cambridge, MA 2Department of Molecular Biology & Genetics, Cornell University, Ithaca, NY *Corresponding author: E-mail: hoekstra@oeb.harvard.edu. Associate editor: Dr. Irina Arkhipova Abstract The genomic landscape of transposable elements (TEs) varies dramatically across species, with some TEs demonstrat- ing greater success in colonizing particular lineages than others. In mammals, long interspersed nuclear element (LINE) retrotransposons are typically more common than any other TE. Here, we report an unusual genomic land- scape of TEs in the deer mouse, Peromyscus maniculatus. In contrast to other previously examined mammals, long terminal repeat elements occupy more of the deer mouse genome than LINEs (11% and 10%, respectively). This pat- tern reflects a combination of relatively low LINE activity and a massive invasion of lineage-specific endogenous ret- roviruses (ERVs). Deer mouse ERVs exhibit diverse origins spanning the retroviral phylogeny suggesting they have been host to a wide range of exogenous retroviruses. Notably, we trace the origin of one ERV lineage, which arose ∼5–18 million years ago, to a close relative of feline leukemia virus, revealing inter-ordinal horizontal transmission. Several lineage-specific ERV subfamilies have very high copy numbers, with the top five most abundant accounting for ∼2% of the genome. We also observe a massive amplification of Kruppel-associated box domain-containing zinc finger genes, which likely control ERV activity and whose expansion may have been facilitated by ectopic recombin- ation between ERVs. Finally, we find evidence that ERVs directly impacted the evolutionary trajectory of LINEs by outcompeting them for genomic sites and frequently disrupting autonomous LINE copies. Together, our results il- luminate the genomic ecology that shaped the unique deer mouse TE landscape, shedding light on the evolutionary processes that give rise to variation in mammalian genome structure. Key words: mobile genetic elements, genome evolution, endogenous retrovirus, genomic conflict, Peromyscus maniculatus. A r t i c l e Introduction Transposable elements (TEs) are parasitic genetic elements capable of mobilizing in genomes and function as import- ant drivers of genome evolution (Orgel and Crick 1980; Kazazian 2004; Bourque et al. 2018). In mammals, for ex- ample, TEs account for at least 20% of the genome and, in some cases, have been exapted for significant functional innovations (van de Lagemaat et al. 2003; Platt et al. 2018; Senft and Macfarlan 2021). When TEs insert into new posi- tions in the genome, they generate mutations and thus re- present a significant burden on host fitness. This cost is compounded by the fact that TEs can contain gene regu- latory sequences and cause structural rearrangements even after they have lost the ability to transpose (Bourque et al. 2018; Klein and O’Neill 2018). Thus, the evolutionary success of a given TE lineage is dictated by its ability to replicate faster than the host genome but lim- ited by its cost to host fitness (Ford Doolittle and Sapienza 1980; Orgel and Crick 1980). TE lineages are in a constant coevolutionary conflict with each other as well as their host (Brookfield 2005; Venner et al. 2009). As a conse- quence, hosts have evolved various ways to suppress TE ac- tivity (Cosby et al. 2019). These genetic conflicts embody the “ecology of the genome” and play an important role in shaping the genomic landscape of TEs in a given species as well as its genome structure more broadly (Brookfield 2005; Venner et al. 2009). TEs are remarkably diverse, and TE landscapes can vary dramatically across species (Wells and Feschotte 2020). TEs are classified into two broad categories based on their transposition mechanism: class I elements (retrotranspo- sons), which mobilize through an RNA intermediate, and class II elements (DNA transposons), which do not. Most eukaryotic lineages harbor a diversity of TEs from multiple taxonomic subgroups within each of these broad classes (Bourque et al. 2018; Wells and Feschotte 2020). By con- trast, some phylogenetic groups have TE landscapes that are relatively similar across species (Abrusán and Krambeck 2006; Sotero-Caio et al. 2017). One such clade is mammals (Platt et al. 2018). In most mammalian © The Author(s) 2023. Published by Oxford University Press on behalf of Society for Molecular Biology and Evolution. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/ licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. Open Access Mol. Biol. Evol. 40(4):msad069 https://doi.org/10.1093/molbev/msad069 Advance Access publication March 22, 2023 1 Gozashti et al. · https://doi.org/10.1093/molbev/msad069 genomes, DNA transposons cannot actively mobilize and only exist as relics of anciently active elements (Platt et al. 2018). Actively mobilizing retrotransposons include long terminal repeat (LTR) retrotransposons, which are mostly endogenous retroviruses (ERVs), as well as non-LTR retrotransposons represented by long inter- spersed nuclear elements (LINEs) and their nonautono- mous counterparts, short interspersed nuclear elements (SINEs) (Eickbush 1992; Platt et al. 2018). LINEs are nearly always the most abundant TEs, and most are represented by a single family, L1, which typically occupies hundreds of megabases of the mammalian genome (Platt et al. 2018). However, the dearth of examples of alternative TE land- scapes has limited our ability to investigate the evolution- ary processes driving mammalian genome structure evolution and specifically, the maintenance of LINE dom- inance (Furano et al. 2004; Platt et al. 2018). The North American deer mouse, Peromyscus manicu- latus, has become an important model for studying the genetic basis of adaptation (Bedford and Hoekstra 2015). Early studies of deer mice and closely related spe- cies used polymerase chain reaction methods to explore TE abundance and reported evidence for an unprece- dented expansion of ERVs (Wichman et al. 1985; Cantrell et al. 2005). However, the landscape of TEs in the deer mouse remains unexplored on a genomic scale. Here, we report a highly distinct genomic landscape of TEs in the deer mouse genome. We find that, in contrast to nearly all examined mammalian genomes, LTR retro- transposons are more abundant in the deer mouse gen- ome than LINEs. We investigate the evolutionary origins and implications of the deer mouse’s distinct genomic landscape, revealing ecological processes that shaped its evolution. Results and Discussion Deer Mice Exhibit a Unique Landscape of TEs To evaluate the genomic landscape of TEs in the deer mouse genome, we first generated a lineage-specific TE li- brary de novo from the deer mouse (P. maniculatus bair- dii) genome using a combination of systematic and manual methods (see Methods). We identified 48 LINE, 28 SINE, and 118 LTR deer mouse-specific subfamilies (fig. 1A and supplementary table S1, Supplementary Material online). We then merged this lineage-specific TE library with all curated mammalian TEs from the Dfam database (Hubley et al. 2016) and annotated the genome using the combined library. We define lineage-specific sub- families with respect to those observed in house mice, Mus musculus (strain C57BL6) (∼25 Myr diverged from the deer mouse; Kumar et al. 2017). Our annotation revealed a distinct genomic landscape of TEs in the deer mouse, relative to other mammals, in which LTR elements occupy more of the genome than LINEs (fig. 1A). Specifically, LTR elements occupy ∼11% of the genome, followed by LINEs (∼10%), SINEs (7%), and other TEs (<2%) (fig. 1A and 2 MBE supplementary table S2, Supplementary Material online). Notably, the 10% LINE occupancy observed for the deer mouse is much lower compared to house mouse, rat, and human. It is worth noting that the dearth of LINE con- tent observed in the deer mouse genome is unlikely an artifact of our inability to detect lineage-specific LINEs since vertical propagation of LINEs has been accompanied by relatively little sequence changes. In total, TEs occupy ∼30% of the deer mouse genome, reflecting an increase in TE content relative to other species in the rodent Family Cricetidae, such as the grasshopper mouse (Onychomys torridus, 24%) and prairie vole (Microtus ochrogaster, 17%) (fig. 1A), but a reduction relative to house mouse (M. musculus, >40%), although differences in genome assembly and TE annotation quality may con- tribute to these patterns (Platt et al. 2016; Peona et al. 2021). Nonetheless, most of the difference in TE content between the deer mouse and house mouse can be attrib- uted to decreased LINE content in the deer mouse, where- as most of the difference in TE content among cricetid species can be attributed to LTR elements. Based on these observations, we hypothesized that the distinct TE landscape of deer mice is the result of a com- bination of reduced accumulation of lineage-specific LINE1s (L1s) and a proliferation of lineage-specific LTR ele- ments. To investigate this possibility, we first compared genomic representation as a function of within-subfamily divergence (as a proxy for subfamily age) across LINEs, SINEs, and LTR elements (fig. 1B). Consistent with our hy- pothesis, we observe reduced representation of LINEs with lower divergence from the consensus, suggesting de- creased LINE accumulation in the deer mouse lineage on more recent timescales (fig. 1B). However, despite this de- cline in the accumulation of LINEs, we still find multiple candidate L1s with intact protein machinery, suggesting that LINEs are still active, consistent with previous reports of LINE activity in deer mice (supplementary table S3, Supplementary Material online; Casavant et al. 1996). We also observed evidence for lineage-specific SINE activity (fig. 1B). Since SINEs parasitize LINE machinery for mobil- ization, evidence of recently active SINEs suggests that re- cently active LINEs still exist in the genome. In addition, we find a lineage-specific proliferation of LTR elements (fig. 1B): LTR elements are significantly overrepresented among the youngest TEs in the genome (<1% divergence from the consensus; two-sided Fisher’s exact test, P < 0.00001). Furthermore, the observed decline of LINE gains in the genome coincides with the peak of LTR gains in the gen- ome (fig. 1B). Together, these results suggest that both re- duced LINE gain and lineage-specific LTR proliferation have contributed to the deer mouse’s unique TE land- scape, and that the two may be associated. DNA Loss Fails to Explain Reduced LINE Content In addition to gain, TE loss can be an important driver of genomic TE content (Kapusta et al. 2017). Although we find evidence for a decline of LINE gain, the low LINE Transposable Element Interactions Shape the Ecology of the Deer Mouse Genome · https://doi.org/10.1093/molbev/msad069 MBE FIG. 1. TE landscape. (A) Phylogeny highlighting the relationship of deer mice (P. maniculatus) to other mammalian species considered in this study. Branch lengths in millions of years were obtained from Timetree (Kumar et al. 2017). Pie charts show the relative percent of the genome occupied by TE subclasses for each species. Color corresponds to the percent of the genome attributed to each type of TE (see legend); gray represents the percent of the genome not occupied by TEs. Stacked bar plots show the proportion of TE content represented by each TE type. Note, in deer mouse, LTRs occupy more of the genome than LINEs. (B) Percent of the genome as a function of CpG corrected Kimura divergence from the consensus for each TE subfamily of LINEs, SINEs, and LTR elements. The vertical dotted line represents the start of observed LINE decline in all plots. content in the deer mouse genome, relative to house mouse, could also have resulted from higher rates of ances- tral DNA loss in the deer mouse (fig. 2A and B). To inves- tigate this possibility, we calculated the DNA loss coefficient k (following Lindblad-Toh et al. 2005), using the formula E = Ae − kt, where E is the amount of extant ancestral DNA in the species considered, A is the ancestral assembly size, and t is time. Larger values of k suggest high- er rates of lineage-specific DNA loss. We calculated a k co- efficient of ∼0.0047 for the deer mouse, a value similar to, and in fact slightly lower than the k value estimated for the house mouse (∼0.006; Kapusta et al. 2017). These data sug- gest that the reduced LINE content observed in the deer mouse genome cannot be explained by generally higher rates of DNA loss in deer mice (fig. 2C). While the results above indicate that the genome-wide rate of DNA loss cannot explain the low LINE content of the deer mouse genome, it is still possible that LINEs are lost at a higher rate than other types of TEs. To investigate this possibility, we compared the proportions of DNA at- tributed to ancient mammalian LINEs present in the com- mon ancestor of the deer mouse and house mouse as well as lineage-specific elements. If the relative absence of LINEs in the deer mouse is due to higher rates of loss of these types of elements, we expect to find a decreased amount of DNA attributed to ancient LINEs in the deer mouse rela- tive to house mouse. To the contrary, we find that although LINEs contribute twice as much content to the house mouse as to the deer mouse (∼575Mb vs. ∼250Mb), ancient LINEs are significantly underrepre- sented in the house mouse genome (∼16% of total LINEs in deer mouse vs. ∼9% of total LINEs in house mouse; two-sided Fisher’s exact test, P = 0.006; fig. 2D). These re- sults suggest that LINE DNA is lost at a slower rate in the deer mouse lineage than in the house mouse lineage, consistent with our k calculations above. Together these data suggest that the low LINE content observed in the deer mouse cannot be attributed to high rates of LINE DNA loss, and instead, is most likely the result of relatively low rates of LINE activity in the deer mouse lineage. ERVK Elements Amplified Predominantly as Nonautonomous Subfamilies Most mammalian LTR retrotransposons are represented by ERVs, which are derived from germline infiltrations of exogenous retroviruses (Mager and Stoye 2015). ERVs are divided into three broad classes depending on their retroviral origins: ERV1 (Gammaretroviridae), ERVK (Betaretroviridae), and ERVL (Spumaretroviridae), with a subgroup of nonautonomous ERVLs called MaLRs (Hubley et al. 2016; Gifford et al. 2018). In the deer mouse, we find a pattern in which ERVK and ERVL/MaLR elements together account for over 80% of genomic ERV content, 3 Gozashti et al. · https://doi.org/10.1093/molbev/msad069 MBE FIG. 2. Non-mutually exclusive evolutionary scenarios that may have shaped the TE landscape in the deer mouse. (A) Higher rates of lineage- specific loss (larger red arrows) could have resulted in the reduced LINE content observed in deer mouse (P. maniculatus) relative to house mouse (M. musculus) and/or (B) higher rates of lineage-specific gain (larger blue arrows) in the house mouse relative to the deer mouse. (C ) k coefficients of DNA loss suggest lower rates of loss in deer mouse relative to house mouse. (D) Ancient elements account for a greater pro- portion of total LINE content in deer mouse relative to house mouse. consistent with previous reports in other rodents (Hubley et al. 2016; Platt et al. 2018; fig. 3A). Lineage-specific ERVs as a whole represent over half of genomic ERV content (∼57%), suggesting that the deer mouse has experienced a substantial ERV expansion. When we compare the pro- portion of lineage-specific ERV content to the proportion of shared ERV content represented by ERVKs, we find that ERVKs account for a disproportionately large part of lineage-specific ERV content (two-sided Fisher’s exact test, P < 0.00001), representing over 75% of observed lineage-specific ERV sequence in the genome (fig. 3B and supplementary table S1, Supplementary Material online). Thus, ERVK activity has been particularly pronounced in the deer mouse lineage. In well-annotated mammalian genomes, among full- length proviral elements (with two LTRs), ERVKs are typic- ally represented by autonomous elements that encode their own machinery for mobilization (Mager and Stoye 2015). After manually inspecting deer mouse-specific ERV subfamilies, and annotating gag, pro, pol, and env genes as well as predicting protein domains required for autonomous transposition, we find that the most abun- dant ERVK families do not possess any internal open read- ing frames (ORFs) predicted to encode proteins with conserved domains, suggesting that they are largely com- posed of nonautonomous elements. Many ERVs contain assembly gaps that interrupt or truncate their internal se- quences (507 of 1,119 candidate full-length ERVs), making it challenging to reconstruct full-length elements and 4 assess the presence or absence of coding machinery. In light of this caveat, we required that a putatively nonauto- nomous ERV subfamily display at least five full-length cop- ies with no gaps for it to be classified as nonautonomous, regardless of its consensus sequence length or content. Even with this conservative filter, we find that the most abundant deer mouse-specific ERV subfamilies are nonautonomous ERVK-like elements lacking any obvious coding capacity (fig. 3C and supplementary table S4, Supplementary Material online). Pman_ERV2_4.24, for ex- ample, is the most abundant ERV in the genome, account- ing for ∼5% of total ERV content. Furthermore, for the subset of ERVKs in which we could confidently reconstruct full-length copies and assess their coding capacity, nonau- tonomous elements occupy more of the genome than au- tonomous ones (supplementary table S4, Supplementary Material online). Overall, our results suggest that autono- mous ERVKs and their nonautonomous counterparts have had a significant impact on the deer mouse’s genome structure. Nonautonomous TEs parasitize autonomous elements for mobilization. Studies on the nonautonomous ERVK subfamily, ETn, in house mouse showed that ETn exhibits regions of sequence similarity to fully coding MusD ele- ments, suggesting that ETn likely arose from the ancestors of Mus-D (Mager and Freeman 2000) and now hijacks MusD machinery trans- complementation (Ribet et al. 2004). Given the high copy numbers of nonautonomous ERVKs in the deer for mobilization via Transposable Element Interactions Shape the Ecology of the Deer Mouse Genome · https://doi.org/10.1093/molbev/msad069 MBE FIG. 3. Relative contribution of different broad ERV classes to ERV content in the deer mouse genome across (A) all ERVs and (B) lineage-specific ERVs. (C ) Respective genomic occupancy across the 18 most common lineage-specific ERV subfamilies. Red font denotes nonautonomous sub- families. (D) VISTA plot showing regions of homology between and relative position of autonomous mysTR subfamilies (gray) and nonautono- mous subfamilies (red). (E) Comparison of LTR percent identity aggregated across all lineage-specific autonomous and nonautonomous elements. (F) Comparison of LTR percent identity across all ERV subfamilies that display at least one candidate full-length copy showing that the youngest subfamilies are autonomous (gray). mouse genome, we sought to identify related autonomous elements that may have been their progenitors and/or In addition to several facilitated their mobilization. prolific nonautonomous subfamilies, we identified three autonomous ERVK subfamilies, Pman_ERVK_4.503, Pman_ERVK_6.7639, and Pman_ERVK_5.247, that to- gether occupy ∼1% of the genome (fig. 3C). These subfam- ilies show sequence similarity to mysTR (∼89%, ∼96%, and ∼90% identity to mysTR pro-pol sequence, respectively), an ERV in Peromyscus (Cantrell et al. 2005), suggesting a shared origin from mysTR. Together, these mysTR-related subfamilies re- present the most abundant autonomous ERVs in the genome. family previously identified Previous studies failed to identify full-length mysTR copies with intact gag and pro-pol genes required for mo- bilization, raising questions about its overall origin and ability to mobilize, although these studies lacked the gen- omic resources to analyze mysTR sequences comprehen- sively (Cantrell et al. 2005; Erickson et al. 2011). Our reveals multiple copies of genome-wide analysis mysTR-related ERVs, which display apparently intact ORFs with homology to gag and pro-pol genes and contain all the protein domains expected to be encoded by au- tonomous elements, suggesting that these ERVs are indeed autonomous and may still be capable of mobilizing (supplementary table S4, Supplementary Material online). We also find several nonautonomous subfamilies related to mysTR (fig. 3D and supplementary table S4, Supplementary Material online; Wichman et al. 1985; Lee et al. 1996). The most conserved region of nucleotide se- quence homology between mysTR-related subfamilies is just downstream of the pro-pol gene and upstream of the 3′ LTR (fig. 3D). Interestingly, most candidate nonauto- nomous and autonomous mysTR-related subfamilies do not display strong homology outside of this region, sug- gesting that nonautonomous subfamilies may have evolved through a recombination event in an autonomous element, which replaced the original internal sequence of the autonomous element with a nonhomologous sequence (Mager and Freeman 2000). Maintenance of sequence similarity in this region is also consistent with functional constraint due to a possible role in ERVK mobil- ization, although its function remains unknown. Since ERV LTRs are identical upon insertion, LTR se- quence identity can provide an estimate for how recently an ERV inserted. To investigate the evolutionary dynamics of deer mouse ERVs, we compared the distributions of LTR identity across lineage-specific ERV subfamilies. We find that nonautonomous ERVs overall exhibit similar ages to 5 Gozashti et al. · https://doi.org/10.1093/molbev/msad069 autonomous ERVs (fig. 3E). However, when we compare nonautonomous and autonomous subfamilies independ- ently, we observe evidence for waves of autonomous elem- ent activity followed by waves of nonautonomous element activity, with the most recently active subfamilies being autonomous (fig. 3F). These observations are consistent with the hypothesis that nonautonomous subfamilies evolved from autonomous subfamilies. Diverse Origins of ERVs ERVs arise in a species when an exogenous retrovirus in- fects the germline. Thus, new families of ERVs evolve de novo through horizontal introduction more frequently than other autonomous mammalian TEs such as LINEs, which primarily evolve through the diversification of verti- cally inherited elements (Mager and Stoye 2015). To inves- tigate the origins of ERVs in the deer mouse, we focused on full-length ERVs across all identified subfamilies with flank- ing LTRs, pol genes, and reverse transcriptase (RT) domains (>450 bp), which we used for classification and phylogen- etic analysis. We initially identified 148 candidate full- length ERVs with pol genes and evidence of an RT domain. However, many ERVs contained ambiguous sites or gaps that interrupted or truncated the RT domain, leaving only 52 ERVs that met our conservative requirements tables S5 and S6, Supplementary (supplementary Material online). Thus, we note that our estimate of ERV diversity is likely an underestimate. We initially used a hidden Markov model approach (Finn et al. 2011) to classify ERVs based on their RT do- mains (supplementary table S6, Supplementary Material online). Using this approach, we find that of the 52 deer mouse ERVs with full-length RT domains, 11 are derived from gammaretroviruses (ERV1), 39 from betare- troviruses (ERVK), and 2 from spumaretroviruses (ERVL) (supplementary table S6, Supplementary Material online). Phylogenetic analysis of RT domains from these ERVs and other known retroviruses supports these initial classifica- tions and shows that deer mouse ERVs form 14 distinct clusters representing at least 14 independent endogeniza- tion events spanning retroviral diversity (fig. 4A). Most of these are derived from diverse betaretroviruses (9 of the 14), consistent with observations in other rodents (Baillie et al. 2004; Cui et al. 2015). Additionally, four ERV clusters show evidence of gammaretroviral origin, and one ERV cluster shows evidence of spumaretroviral origin (fig. 4A). To determine the age of ERVs, we conducted searches for deer mouse ERVs in grasshopper mouse, prairie vole, and house mouse. We find that most (9 of the 14) deer mouse ERVs arose before the divergence of the deer mouse and its close relative, the grasshopper mouse (∼5–13 mil- lion years ago [MYA]) (León-Paniagua et al. 2007; Leite et al. 2014), but after the divergence of their ancestor lineage of the prairie vole (∼18 MYA) and the (Abramson et al. 2009; Kumar et al. 2017), and are thus lineage-specific relative to house mouse (fig. 4B). Additionally, one ERV (Beta_Pman-ERV_cluster-5) was 6 MBE introduced even more recently, after the divergence be- tween the deer mouse and grasshopper mouse (fig. 4B). Intra-element LTR identity for ERVs in each respective cluster generally concurs with the timing estimate of their successive endogenization (supplementary table S6, Supplementary Material online). Given the relatively re- cent origins of several deer mouse ERVs (since the diver- gence between Peromyscus and Microtus ∼18 MYA), we reasoned that it may be possible to trace more precisely their origins by searching the databases for their closest ex- ogenous retrovirus relatives. This search revealed one po- tential case of a recent endogenization of an exogenous Feline Leukemia Virus (FeLV), or a closely related virus, in the ancestor of the deer mouse and grasshopper mouse within the last ∼5–18 million years (fig. 4C; Abramson et al. 2009; Kumar et al. 2017). Some Deer Mouse ERVs may Still be Infectious Although ERVs only require gag and pol genes to mobilize in the germline via retrotransposition, ERVs with intact env genes can also replicate via reinfection (Belshaw et al. 2004). Given the relatively recent evolution of several ERVs in the deer mouse, we inspected all intact ERVs as well as ERV subfamily consensus sequences for intact env genes. We find no evidence for env genes in mysTR- related subfamilies, consistent with previous studies on mysTR (Cantrell et al. 2005; fig. 4D and supplementary table S4, Supplementary Material online). However, we find putatively full-length env genes in multiple other ERV clusters, suggesting that some deer mouse ERVs may still be capable of infection (supplementary table S4, Supplementary Material online). One of these is the previ- ously mentioned FeLV-related Gamma_Pman-ERV_ cluster-10. The observation of a putatively intact env in these ERVs is consistent with previous studies showing that leukemia viruses remain infectious in other species (Hoover and Mullins 1991; Polat et al. 2017; fig. 4D). We also observe evidence of an intact env gene for ERVs within the Beta_Pman-ERV_cluster-5. Beta_Pman-ERV_cluster-5 ERVs are absent in grasshopper mice and thus represent some of the most recent ERVs to infiltrate the deer mouse germline (fig. 4B and D). Interestingly, env genes from this family show ∼60% sequence similarity to the env encoded by intracisternal A-type particle (IAP) elements in house mice, which are also capable of intercellular transmission (Ribet et al. 2008), suggesting a possible origin from a similar exogenous retrovirus (supplementary tables S4 and S6, Supplementary Material online). Together these data sug- gest that several ERVs derived from exogenous retroviruses recently and some may still be infectious. Negative Selection Shapes TE Distributions in the Deer Mouse Genome Most TE insertions are deleterious or neutral, and the genomic distribution of TEs is shaped in large part by se- lection against deleterious insertions. In the deer mouse genome, TEs account for nearly 25% of nucleotides in Transposable Element Interactions Shape the Ecology of the Deer Mouse Genome · https://doi.org/10.1093/molbev/msad069 MBE FIG. 4. Origins of ERVs in the deer mouse genome. (A) Cladogram displaying a maximum likelihood tree constructed with RT domains of deer mouse ERVs and publicly available endogenous and exogenous retroviruses for context. Internal nodes across monophyletic retroviral clades are all strongly supported (>95% bootstrap support). Deer mouse ERVs form 14 distinct clusters spanning broad retroviral diversity (highlighted in red). (B) Cladogram showing the approximate time of origin of deer mouse (P. maniculatus) ERVs based on the presence or absence in three other species at increasing phylogenetic distances. ERV cluster numbers are shown on the branch corresponding to their approximate origin. For each species, green boxes represent presence and gray boxes represent absence for each respective ERV cluster found in deer mouse. (C ) Neighbor-joining tree showing the phylogenetic relationship between Pman-ERV_cluster-10 copies in deer mouse, related ERVs in grasshopper mouse (O. torridus), and FeLV. (D) Structure of ERVs found in deer mouse with ORFs (colored boxes); important protein domains are annotated. Overlapping boxes represent overlapping ORFs. protein coding genes and 30% in long noncoding RNAs (lncRNAs) nucleotides but are relatively absent from coding exons (permutation test, P < 0.001), suggesting strong purifying selection on new insertions in coding exons (fig. 5A). These patterns are consistent with ob- servations in other mammals (Nellåker et al. 2012; Kapusta et al. 2013; Platt et al. 2018). Comparison of TE occupancies across chromosomes reveals that ERVs and LINEs are prevalent on the X chromosome (occupy- ing ∼15% and ∼17 of the X chromosome, respectively, compared to an average of ∼12% and ∼10% for other chromosomes; fig. 5B). This pattern is not observed for SINEs and likely reflects the more frequent removal of longer TEs such as LINEs and ERVs on autosomes by re- combination or purifying selection against new inser- tions (Kent et al. 2017; Dechaud et al. 2019). ERV insertions around protein coding genes are also usually deleterious since ERVs contain complex internal regula- tory elements that can disrupt gene expression. Consistent with this, ERVs are generally distant from genes and significantly more distant from genes in the same orientation (Mann Whitney U, P < 0.0001; fig. 5C). It is worth noting that this bias is most pronounced for mysTR-related ERV subfamilies, suggesting that these ERVs are highly deleterious, perhaps containing regulatory sequences particularly prone to disrupt gene expression when inserted in the same orientation as surrounding genes. Some ERV Subfamilies Have Possible Regulatory Function While most ERV subfamilies show patterns suggesting deleterious effects affecting the expression of neighboring genes, others display patterns consistent with possible regulatory function. Indeed, ERV LTRs may be co-opted for important regulatory functions over evolutionary time (Chuong et al. 2017). We find a small subset of ERV subfamilies to be enriched in the 5-kb region upstream of gene transcription start sites, suggesting that these ERVs minimally affect neighboring gene expression or that they may contribute to host gene regulation as either promoters or enhancers (fig. 5D). These ERVs also display significantly higher within-subfamily divergence relative to other lineage-specific deer mouse ERVs (Mann Whitney U, P = 0.0048), suggesting that they primarily represent older, inactive subfamilies (fig. 5E). Additionally, some subfam- ilies, including MT2B1 and ORR1B1, represent lineages of ancestrally shared elements that have been co-opted for regulatory functions in other mammalians (Franke et al. 7 Gozashti et al. · https://doi.org/10.1093/molbev/msad069 MBE FIG. 5. Genomic distribution of TEs in the deer mouse genome. (A) Respective coverage, defined as the proportion of nucleotides attributed to TEs for a given feature, for different TE subclasses across genomic features. CDS = protein coding sequence; lncRNA = long noncoding RNA. (B) Respective coverage for TEs across chromosomes. (C ) Box plots showing the distribution of ERV distances from the closest gene on the same strand (red) versus when strand is ignored (gray). (D) Enrichment ratio (number observed/expected) and Bonferroni-corrected Fisher’s exact test P-values (Q-values) for ERV subfamilies enriched within the 5 kb region upstream of genes in the same orientation. (E) Within-subfamily CpG corrected Kimura divergence for ERV subfamilies enriched within the 5 kb region upstream of genes in the same orientation (red) compared to all other ERV subfamilies (gray). (F ) Genomic distribution of ERV hotspots (red) across chromosomes. Lineage-specific KZNF genes are indicated (purple triangles) and are enriched in ERV hotspots. (G) KZNFs in ERV hotspots (red) show lower Kimura divergence than other KZNFs (gray), suggesting that they are younger. (H ) Kernel density estimates for the distribution of Kimura divergences for KZNFs outside ERV hotspots (gray), KZNFs in ERV hotspots (red), and ERVKs (blue). Vertical dotted lines show the peak value for each distribution. (I ) Cartoon displaying an ERV insertion interrupting a formerly intact LINE. N represents the range of observed candidate instances of ERV-mediated LINE interruption. 2017). Given the frequent and recurrent lineage-specific ERV co-option events observed across mammals (Feschotte 2008; Sakashita et al. 2020; Fueyo et al. 2022), these subfamilies represent promising candidates for co- option events in the deer mouse lineage. ERVs Accumulate in “Hotspots” Enriched for Kruppel-Associated Box-Zinc Finger Genes The distribution of ERVs in the genome is largely biased to- wards specific regions, or “hotspots”, which are enriched in Kruppel-associated box (KRAB) domain-containing zinc finger genes (KZNFs). We define “hotspots” as regions of the genome in the top 95th percentile of ERV density, where ERV density is the proportion of nucleotides attrib- uted to ERVs in a given 100-kb genomic window (fig. 5F). Lineage-specific ERVKs constitute over 70% of ERVs in hot- spots, suggesting that these genomic associations are likely lineage-specific. Furthermore, neighboring ERVs are signifi- cantly more divergent in ERV hotspots than in other 8 regions of the genome (Mann Whitney U, P = 3.317e −06), suggesting that hotspots arose primarily through in- dependent insertions rather than segmental duplication (SD) of existing insertions. ERV hotspots are largely devoid of genes, and we observe a strong negative correlation be- tween gene density and ERV density overall (generalized linear model, P < 0.0001). However, we do observe some genes in ERV hotspots. We performed a gene ontology (GO) enrichment analysis for genes in ERV hotspots and found significant enrichment for one biological pro- cess term: “regulation of transcription, DNA-templated” (two-sided Fisher’s exact test, Q < 0.00001). Scrutiny of genes overlapping ERV hotspots that match this GO term reveals that ∼85% (100/118) are deer mouse- specific KZNFs (fig. 5C). We define deer mouse-specific KZNFs based on refseq’s annotation of genes that do not have orthologs in other species (O’Leary et al. 2016). We find that deer mouse-specific KZNFs specific- ally are enriched in ERV hotspots, with ∼32% (100/312) of KZNFs occurring in ERV hotspots, despite the fact Transposable Element Interactions Shape the Ecology of the Deer Mouse Genome · https://doi.org/10.1093/molbev/msad069 MBE that ERV hotspots only represent <5% of the genome (two-sided Fisher’s exact test, P < 0.00001). Coevolution of ERVs and KZNFs It has become increasingly clear that the primary function of KZNFs is to suppress retroelement activity (Thomas and Schneider 2011; Yang et al. 2017; Cosby et al. 2019). KZNF gene clusters evolve rapidly through a birth-death model under positive selection and often expand in response to the lineage-specific activity of retroelements, including ERVs (Emerson and Thomas 2009; Najafabadi et al. 2015; Wolf et al. 2020). The colocalization of KZNF genes and ERVs in genomic space is intriguing and has been observed previously in the house mouse (Kauzlaric et al. 2017). Although this observation could simply be explained by re- laxed selection on nonessential KZNF genes, two alterna- tive, non-mutually exclusive hypotheses could explain the observed colocalization between KZNFs and ERVs: (1) KZNFs use neighboring ERVs as regulatory sequences to respond to the global derepression of ERVs (Pontis et al. 2019; Ito et al. 2020) or (2) ERVs contribute to KZNF gene family evolution by facilitating rapid gene du- plication and deletion (i.e., turnover) in these regions. Indeed, ERVs are known to facilitate structural rearrange- ments via ectopic recombination, and ERV-rich regions of the genome can be highly plastic (Hughes and Coffin 2001; Doxiadis et al. 2008; Jern and Coffin 2008; Hermetz et al. 2012). Interestingly, lineage-specific KZNF duplicates in ERV hotspots exhibit significantly lower divergence compared to other KZNFs, suggesting that genes in ERV hotspots duplicated relatively recently (Mann Whitney U, P = 0.0043; fig. 5G). This observation supports the idea that KZNFs overlapping with ERV hotspots duplicate more often, although the evolutionary processes driving this pattern remain unclear. Next, we examine whether lineage-specific KZNF gene family expansion located in ERV hotspots coincides with lineage-specific ERV activity. To do so, we compared the age (sequence divergence) distribution of KZNF gene du- plicates located within ERV hotspots to that of KZNFs res- iding outside ERV hotspots and that of ERVK subfamilies, as measured by intra-subfamily copy divergence (fig. 5H). The distribution of duplicate divergence for KZNFs in ERV hotspots suggests that the largest KZNF expansion oc- curred just before or around the same time as the peak of ERVK amplification (fig. 5H). Indeed, the median percent divergence for lineage-specific KZNF gene duplicates in ERV hotspots is ∼17.2%, while the within-subfamily diver- gence for the top three most abundant ERVs in the deer mouse genome is ∼17.4%. This pattern is consistent with a KZNF expansion driven by the amplification of highly ac- tive lineage-specific ERVKs. In contrast, the distribution of duplicate divergence for KZNFs not overlapping ERV hot- spots shows no obvious relationship to lineage-specific ERVK activity (fig. 5H), further suggesting that the ob- served colocalization between ERVKs and KZNFs may be causally associated. Furthermore, some KZNF gene clusters display much larger expansions than others: for example, a cluster on chromosome 1 contains >90 genes, represent- ing about one-third of lineage-specific KZNFs in the deer mouse genome (fig. 5F). This observation suggests that KZNFs in this chromosome 1 cluster may play an import- ant role in suppressing ERVKs. We observe another ex- ample on chromosome 22, which displays a cluster of 48 lineage-specific KZNF genes. Since members of the same KZNF clusters often bind to related ERV families, the mas- sive invasion of closely related ERVs predicts expansions of closely related KZNFs (Wolf et al. 2020). Together, these re- sults suggest that KZNFs in the deer mouse underwent a large expansion in response to lineage-specific ERV activity. Lineage-Specific ERV Insertions Interrupt LINE Sequences In addition to evaluating ERV distributions with respect to genes, we also assessed ERV distributions with respect to other TEs. We were specifically interested in how the ob- served ERV invasion in the deer mouse might directly im- pact pre-existing LINEs. Specifically, we sought to examine whether ERVs could have directly impacted L1 activity by inserting into and interrupting transposition-competent L1. L1 families typically only have from a hundred to a few thousand “master genes” that are transposition- competent in mammalian genomes (Deininger et al. 1992; Brouha et al. 2003; Zemojtel et al. 2007; Platt et al. 2018). Furthermore, the L1 retrotransposition mechanism is fairly inefficient, and the vast majority of new L1 inser- tions are defective and incapable of mobilizing thereafter (Fanning 1983; Grimaldi et al. 1984). Thus, disruption of many master genes could have a considerable impact on the evolutionary trajectory of L1s in a species. To explore the direct impacts of ERV insertions on L1s, we searched for ERV insertions directly flanked by L1 se- quences from the same L1 subfamily. We then filtered for cases in which flanking L1 sequences conjoined at the correct coordinates with respect to the subfamily con- sensus, forming a full-length L1. We also initially filtered for L1s that did not contain any additional TE insertions. These results revealed 322 prospective lineage-specific ERV inser- tions that interrupt full-length L1 (supplementary table S7, Supplementary Material online). However, this number is likely a large underestimate, since it does not include frag- mented L1s that, for example, have accumulated multiple indels. If we include fragmented L1s as well, we find 2,664 prospective ERV insertions interrupting LINEs, 900 of which are attributed to the two most abundant ERVK related subfamilies (Pman_ERV2_4.503 and Pman_ERV2_4.24; supplementary table S8, Supplementary Material online). Interestingly, within-subfamily percent divergences for these subfamilies (16.05% and 15.86%) suggest that they invaded just before the decline of L1 gain (see fig. 1B, ∼15%). We speculate that this association is no coincidence, and that ERV insertions within potentially active L1s were a signifi- cant driver in reducing L1 activity in the deer mouse lineage. Punctuated L1 interruptions on these scales (322 - > 2,500 9 Gozashti et al. · https://doi.org/10.1093/molbev/msad069 L1 interruptions) would eliminate most functional L1s in many mammalian species, and even on much smaller scales, could have a catastrophic effect on the evolutionary trajec- tory of L1s in a genome, especially given the poor success rates of the L1 retrotransposition mechanism in producing new fully functional L1 copies (Kazazian and Moran 1998; Szak et al. 2002). We also explored quantitative patterns of ERV content in LINEs more broadly. To do so, we tested for an enrichment of ERVs inserted within L1s across all ERV subfamilies. Interestingly, we find that several lineage-specific ERVK sub- families show significant enrichment within L1s relative to ran- dom expectation (Bonferroni-corrected permutation test, α = 0.01, N = 1000; supplementary table S9, Supplementary Material online). For example, Pman_ERVK_4.63 elements interrupt L1s more than 46 times more than expected by chance (Bonferroni-corrected permutation test, Q < 0.001). The observed enrichment for lineage-specific ERVK subfam- ilies, representing 29 of the 36 enriched subfamilies, and absence of enrichment for other (older) ERV subfamilies, is consistent with our hypothesis that lineage-specific ERVK expansion directly impacted L1 viability (supplementary table S9, Supplementary Material online). Together, these results suggest an intriguing model whereby ERVs and LINEs compete for genomic sites and that ERVKs may have directly impacted the evolutionary trajectory of LINEs in the deer mouse lineage. An “Ecology of the Genome” Model for the Evolution of the Deer Mouse Genome In the same way that species compete for space and re- sources, TEs compete with each other for sites in the gen- ome as well as metabolic resources (Brookfield 2005). TEs can occupy specific niches, which can allow them to coex- ist with limited competition, but TEs that occupy similar niches are more likely to compete and thereby drive one or another to extinction (Brookfield 2005; Venner et al. 2009). Furthermore, the relative success of a given TE also depends on host suppression mechanisms and their targets. For example, differential host targeting between two TE families in direct competition could limit the suc- cess of one family that would, in the absence of host de- fense mechanisms, be more fit than the other (Venner et al. 2009). Also, because TEs could threaten to kill their host in the absence of host-mediated suppression, it can be advantageous (for both the host and TEs) that host de- fenses evolve to suppress TE activity (Venner et al. 2009). Our model for the evolution of the distinctive TE land- scape of the deer mouse supports the notion of “genomic ecology” (Brookfield 2005). We postulate that the intro- duction of mysTR-related ERVs caused a shift in the deer mouse TE landscape through the following processes (fig. 6): first, mysTR ERVs evaded host defenses upon germ- line infiltration, which allowed them to expand to large numbers. This hypothesis is supported by the observation that mysTR ERVs are highly divergent from other known retroviruses as well as the remarkable expansion of deer 10 MBE mouse-specific KZNFs following peak ERV activity (Cantrell et al. 2005). In mammals, ERVs are the primary targets of KZNF suppression, whereas LINEs and SINEs are less frequently targeted, probably because ERV inser- tions are more regulatorily potent and therefore more deleterious (Wolf et al. 2015; Zhou et al. 2020). These host defenses keep ERVs in check, despite evidence that LINEs and ERVs compete for similar sites in the genome. First, many ERVs and LINEs both preferentially integrate into AT rich regions (Medstrand et al. 2002; Babushok et al. 2006; Nellåker et al. 2012; Campos-Sánchez et al. 2016). Thus, ERVs and LINEs often inhabit similar regions of the genome and frequently insert within each other (Campos-Sánchez et al. 2016). Second, ERV insertions in LINEs (or vice versa) are likely invisible to selection and exhibit a higher rate of fixation relative to deleterious insertions (Campos-Sánchez et al. 2016). Under these circumstances, in the absence of host defense mechanisms, we expect the pri- mary driver of ERV or LINE success in the genome to be rela- tive rates of gain of transposition-competent copies. Thus, we postulate that the massive expansion of mysTR ERVs nearly drove LINEs to extinction in the deer mouse genome. Since this initial ERV invasion, we postulate that expansions of host KZNF repertoires helped stabilize ERV activity in the gen- ome and were likely aided by the proliferation of nonautono- mous ERV derivatives. These suppression mechanisms likely enabled more sustainable ERV activity by limiting the rate of ERV expansion and reducing fitness cost. More generally, we propose that this model may explain the loss of LINE activity in other mammals. A subclade of sigmodontine rodents for example (∼13–18 MYA di- verged from the deer mouse; Abramson et al. 2009; Leite et al. 2014; Kumar et al. 2017; Gonçalves et al. 2020) repre- sents one of the few mammalian lineages to have experi- enced LINE extinction (Yang et al. 2019). Consistent with our model, previous studies suggest that LINE extinction in this group followed an invasion of mysTR-related ERVs on a similar or possibly larger scale to that observed for the deer mouse (Cantrell et al. 2005; Erickson et al. 2011). At present, the lack of genome assemblies for sigmodontine rodents makes it challenging to study TEs in these species. However, a recent study examining TE content in mammals notes that cricetid rodents exhibit the highest rates of recent LTR retrotransposon accumulation among mammals (Osmanski et al. 2022). Overall, we hypothesize that for an ERV invasion to have a similar effect on LINE ac- tivity in another mammal, the causal ERV must arise from a divergent retrovirus (unfamiliar to host suppression machin- ery), show similar integration preference to that of LINEs, and rapidly evolve nonautonomous derivatives. Future studies in other species, which show unique patterns of mammalian genome composition, will shed further light on evolutionary conflicts that drive mammalian genome evolution. Concluding Remarks Although TE landscapes differ drastically across species, most mammalian genomes are similarly dominated by Transposable Element Interactions Shape the Ecology of the Deer Mouse Genome · https://doi.org/10.1093/molbev/msad069 MBE FIG. 6. Model for deer mouse genome evolution. LINE (and SINE) non-LTR retrotransposons. The deer mouse, P. maniculatus, represents one of the few excep- tions to this pattern: LTR elements occupy more of the genome than LINEs. We find that the distinct genomic landscape of TEs in the deer mouse reflects a massive ex- pansion of ERVs as well as a dearth of LINE activity, and that the two phenomena are likely associated. Our results show that a broad diversity of ERVs invaded the deer mouse genome and that the infiltration of one ERV family in particular, mysTR, played a prominent role in establish- ing its unique TE landscape. Furthermore, we note that re- ported ERV copy numbers and diversity are likely a vast underestimate since the current deer mouse genome as- sembly was generated with short reads (Peona et al. 2021). Based on these findings, we postulate that the propen- sity for a mammalian genome to undergo a shift in TE con- tent and/or experience LINE extinction is directly related to its susceptibility to invasion by divergent TEs—in this case, ERVs. Furthermore, the accumulation of ERVs in specific genomic hotspots raises additional questions about how TE-dense regions can affect mammalian gen- ome evolution. Indeed, we would expect such regions to experience structural rearrangements more often than other regions of the genome. Previous studies in the deer mouse have identified many large inversions (>1 Mb in length), which are polymorphic, even within popu- lations (Hager et al. 2022; Harringmeyer and Hoekstra 2022). Could these ERV hotspots have played a role in fa- cilitating deer mouse inversions? We also observe enrich- ment of KZNF gene families which evolve rapidly via duplication within ERV hotspots. Is the colocalization of KZNFs and ERVs advantageous for the host due to the in- creased propensity for KZNF gene family expansion? We show that KZNFs in ERV hotspots are indeed younger than other KZNFs, providing some support for the coevolution of these genomic features. However, within- population studies are critical to further elucidate this coevolutionary relationship. Together, our results have 11 Gozashti et al. · https://doi.org/10.1093/molbev/msad069 MBE broad implications and open up a range of opportunities to investigate the evolutionary processes that give rise to the evolution of mammalian genome structure. deer mouse proteins with parameters (-max_target_seqs 25 -culling_limit 2 -evalue 10e-10) and filtered all TEs with unknown classifications that shared homology with proteins. Methods Obtaining Relevant Genomic Data We downloaded publicly available TE annotations for human, Homo sapiens (GCF_000001405.40; genome contig N50 = 57,879,411; contig L50 = 18); house mouse, M. musculus (GCF_000001635.27; genome contig N50 = 59,462,871; contig L50 = 15); Norway rat, Rattus norvegicus (GCF_015227675.2; genome contig N50 = 29,198,295; contig L50 = 27); and prairie vole, M. ochrogaster (GCF_000317375.1; genome contig N50 = 21,250; contig L50 = 29,205) from RepeatMasker (http://www.repeat masker.org/genomicDatasets/RMGenomicDatasets.html) and for grasshopper mouse, O. torridus from NCBI (GCF_903995425.1; genome contig N50 = 2,276,141; con- tig L50 = 308). We used the deer mouse, P. maniculatus, genome assembly available (refseq GCF_003704035.1; contig N50 = 30,111; contig L50 = 23,323) for all genomic analyses. Retroviral sequences for ERV phylogenetic analysis were downloaded from NCBI. Genbank accession numbers and for these sequences are shown in fig. 4A. through NCBI TE Discovery and Annotation We used a combination of systematic and manual techni- ques to identify and annotate TEs in the deer mouse gen- ome. We started by using an approach similar to the EarlGrey pipeline (github.com/TobyBaril/EarlGrey/; Baril and Hayward 2022). We first identified known rodent TEs in the deer mouse genome using RepeatMasker (ver- sion 4.1.2, https://www.repeatmasker.org/) with a curated set of rodent TEs from the DFAM database (Hubley et al. 2016) and the flags -nolow, -norna, and -s. Next, we constructed a de novo repeat library using RepeatModeler2 (version 2.0.1), with RECON (version 1.08) and RepeatScout (version 1.0.5) (Bao and Eddy 2002; Price et al. 2005; Flynn et al. 2020). Maximum-length consensus sequences were generated for putative de novo TEs identified by RepeatModeler using an automated version of the “Basic Local Alignment Search Tool (BLAST), Extract, Extend” process through EarlGray (Platt et al. 2016). Briefly, EarlGray first performs a BLASTn search to obtain the top hits for each TE subfamily (Camacho et al. 2009). Then, it aligns the 1,000 base pairs of flanking retrieved se- quences using multiple alignment using fast fourier trans- form (MAFFT; version 7.453; Katoh and Standley 2013). Following this, alignments are trimmed using trimAl (ver- sion 1.4) with -cons 60; Capella-Gutiérrez et al. 2009). Finally, consensus sequences are updated using European molecular biology open soft- ware suite cons (-plurality 3; Rice et al. 2000). This process is then repeated five times. Following this, we performed blastx (Camacho et al. 2009) searches against all known the options (-gt 0.6 12 families were Following the automated processes described above, alignments for TE families were individually inspected using AliView (Larsson 2014) and poorly represented po- sitions were manually trimmed as recommended by Storer et al. (2021). Families were also manually realigned using extract_align.py (Platt et al. 2016) and MAFFT (ver- sion 7.453; Katoh and Standley 2013) and then reexa- mined. Manually curated TE then re-clustered using cd-hit-est (Fu et al. 2012) and families were merged based on the 80-80-80 rule criterion (Wicker et al. 2007). We also used TE-Aid (https:// github.com/clemgoub/TE-Aid) to identify TE-associated ORFs and sequence features such as LTRs when classify- ing TEs. We combined our final de novo TE library with the Rodent DFAM TE library (Hubley et al. 2016) and an- notated TEs the deer mouse genome using RepeatMasker. To identify full-length LTR elements, we used LTR_FINDER (Xu and Wang 2007) and LTRharvest (Ellinghaus et al. 2008) through EDTA_raw with the flag -type ltr (version 2.0.0; Ou et al. 2019), which also re- port LTR divergence for each element. TE annotations were defragmented and refined using RepeatCraft with the flag -loose (Wong and Simakov 2019), and overlap- ping annotations were resolved in favor of the longer element using MGKit (version 0.4.1) filter-gff (Rubino and Creevey 2014). in Identifying Functional Machinery for Putatively Autonomous TEs To identify the protein machinery of potentially autono- mous LINE and LTR elements, we extracted all LINE ele- ments longer than 2700 bp and LTR elements longer than 5000 bp from the deer mouse genome. Then, we also used TE-Aid (https://github.com/clemgoub/TE-Aid) to identify ORFs in each retrieved LTR and LINE element with homology to known TE genes. We used hmmer (Finn et al. 2011) and relevant hmms available from GyDB (Llorens et al. 2011) and PFAM (Mistry et al. 2021) to identify retroviral protein domains as well as NCBI’s conserved domain search tool (Marchler-Bauer et al. 2015; Marchler-Bauer et al. 2017). Calculating k Coefficients We calculated the DNA loss coefficient k (Lindblad-Toh et al. 2005), using the formula E = Ae − kt, where E is the amount of extant ancestral DNA in the species considered, A is the ancestral assembly size, and t is the time. We cal- culated E for each species by subtracting the amount of genomic DNA attributed from the amount of DNA attributed to ancient shared mamma- lian TEs (retrieved from Kapusta et al. 2017). We used 2.8 Gb for A and 100 million years for t as in Kapusta et al. (2017). lineage-specific TEs Transposable Element Interactions Shape the Ecology of the Deer Mouse Genome · https://doi.org/10.1093/molbev/msad069 MBE Identifying Nonautonomous ERV-Like Elements Since the deer mouse genome was produced primarily with short reads, most ERVs have internal gaps or strings of low quality or ambiguous nucleotides. Thus, to decipher nonautonomous ERV-like elements from autonomous ERVs, we used a strict criterion. For a given ERV subfamily to be considered nonautonomous, we required at least five full-length copies which lack identifiable ORFs as well as ambiguous nucleotides. We performed global pairwise alignments between nonautonomous and autonomous ERVK consensus sequences using the global alignment software AVID with default parameters (Bray et al. 2003). We visualized alignments using VISTA (Frazer et al. 2004). ERV Classification and Phylogenetic Analysis We used two complementary approaches to classify deer mouse ERVs. First, we examined e-value statistics in the output from our GyDB hmm scans to discern which viral RT domain hmm best fit each ERV. In addition, we also used a phylogenetic approach. We annotated ERVs with their viral origin as predicted by our hmm scans. Next, we downloaded several endogenous and exogenous retro- viruses from NCBI (accessions shown in fig. 4A), extracted their RT domains, and annotated them with their respect- ive viral clade. Then, we filtered sequences with large strings of ambiguous characters, performed a multiple se- quence alignment of RT genes using MAFFT (Katoh and Standley 2013), and generated a maximum likelihood- based phylogeny using IQ-TREE (Minh et al. 2020) with a GTR + G model (general time reversible model with un- equal rates and unequal base frequencies and discrete gamma rate heterogeneity; Yang 1994). We constructed a consensus tree across 1,000 replicates using IQ-TREE’s -bb flag (Yang 1994). All internal nodes separating mono- phyletic ERV clades were strongly supported (>95% boot- strap support). We analyzed and edited the resulting phylogeny using ete3 (version 3.1.2) (Huerta-Cepas et al. 2016), collapsing clusters of deer mouse ERVs into repre- sentative nodes. We visualized the phylogenetic tree using the interactive tree of life (Letunic and Bork 2021) and FigTree (version 1.4.4; https://github.com/rambaut/ figtree). To search for homology between deer mouse ERVs and MysTR, we used blastn (Camacho et al. 2009) to align deer mouse ERV consensus sequences to a previ- ously (Genbank DQ139737.1; Cantrell et al. 2005). isolated MysTR pol-pro sequence Searching for Deer Mouse ERVs in Other Species To search for deer mouse ERVs in the house mouse, prairie vole, and grasshopper mouse genomes, we performed local BLASTn (Camacho et al. 2009) queries for each full-length deer mouse ERV to each respective genome. We ran BLASTn (Camacho et al. 2009) with the flag -outfmt 6 and required a minimum alignment length of 400 bp and minimum percent identity of 75 to limit possible erro- neous hits. As a proof of concept, we also made sure our results were consistent with expectations based on LTR divergences. For example, we would not expect an ERV with highly divergent LTRs (a signature of a more ancient insertion) to be specific to the deer mouse. We also per- formed broader BLASTn queries against NCBI’s nucleotide database. Queries for Gamma_Pman-ERV_cluster-10 se- quences yielded high-confidence hits in deer mouse spe- cies, the grasshopper mouse, and a FeLV reference genome (Genbank AB060732.3). BLAST queries of FeLV (AB060732.3) back to the non-redundant nucleotide data- base also showed best hits to the deer mouse and grass- hopper mouse genomes when other FeLV genomes were excluded. A neighbor-joining phylogeny constructed from deer mouse Gamma_Pman-ERV_cluster-10 se- quences, homologous ERVs in the grasshopper mouse gen- ome, and FeLV (AB060732.3) suggest a scenario in which Gamma_Pman-ERV_cluster-10 originated in the common ancestor the deer mouse and grasshopper mouse from FeLV or another closely related exogenous virus between 5 and 18 MYA. TE Distribution Analysis We used bedtools intersect (Quinlan and Hall 2010) to find overlaps between TE annotations and gene feature an- notations. We used bedtools closest (Quinlan and Hall 2010) with the parameter -s to identify TE distances from the nearest gene on the same strand and again with default parameters to ignore strand. All functional enrichment tests were performed using goatools (Klopfenstein et al. 2018). We also tested for enrichment or depletion of TEs 5 kb upstream of genes in the same orientation. Specifically, for each TE subfamily, we rando- mized all TE locations on each chromosome and com- pared the number of TEs within 5 kb of genes upstream in the same orientation with the observed value. We per- formed two-sided Fisher’s exact tests comparing the num- ber of observed and expected elements within these regions to obtain P-values. Fisher’s exact P-values and per- mutation test P-values were adjusted using the Bonferroni method to obtain Q-values. This revealed eight subfamilies that displayed enrichment for the 5 kb regions upstream of genes in the same orientation (supplementary table S10, Supplementary Material online). However, enrich- ment of SDs in these regions could also cause a similar pat- tern. To test this alternative, we used SEDEF (version 1.1) (Numanagic et al. 2018) with default parameters to iden- tify SDs in the deer mouse genome. Then, for each en- riched ERV subfamily, we intersected SD coordinates with the 5 kb regions upstream of genes harboring at least one element in the same orientation. We then compared SD coverage in these regions for each ERV subfamily with expectations from randomization for 1,000 permuta- tions. This revealed that genes containing RMER15 copies within 5 kb upstream in the same orientation also dis- played enrichment for SDs, suggesting that SDs could al- ternatively explain RMER15 (supplementary table S10, Supplementary Material online). Thus, we did not include RMER15 in figure 5D. We used bedtools coverage (Quinlan 13 Gozashti et al. · https://doi.org/10.1093/molbev/msad069 and Hall 2010) to calculate ERV and gene density along 100 kb windows in the genome. ERV hotspots were de- fined as windows which exhibit ERV densities within the top 95th percentile. ERV hotspots could arise through two possible non-mutually exclusive mechanisms: inde- pendent insertion of ERVs in specific genomic regions and SD of pre-existing ERV insertions. One expectation of the latter is that neighboring ERVs of the same subfamily would exhibit more similar divergences from the consen- sus in ERV hotspots (if they arose from a duplication of the one original insertion) than in other regions of the genome. To assess this possibility, we compared delta divergence from the consensus between neighboring ERVs (|neighbor_1_div—neighbor_2_div|) in ERV hot- spots to other regions of the genome and report a signifi- cant trend in the opposite direction (Mann Whitney U, P = 3.317e−06), suggesting that SD is not the primary con- tributing mechanism to ERV hotspot formation. Figure 5F was produced using RIdeogram (Hao et al. 2020). KZNF Gene Family Analysis We defined deer mouse-specific zinc finger (ZF) genes as genes which do not have recognizable orthologs as anno- tated by NCBI. We employed hmmscan (Finn et al. 2011) using KRAB hmms downloaded from PFAM (Mistry et al. 2021) to identify KRAB domain-containing ZFs (KZNFs). Then, we performed a multiple sequence align- ment of all KZNFs using Clustal omega (Sievers et al. 2011) with the parameters -use-kimura and -full in order to simultaneously produce a pairwise Kimura divergence matrix across all genes. We constructed a subsequent phyl- ogeny using IQ-TREE (Minh et al. 2020) with a general time reversible model. To test for phylogenetic clustering of KZNF that overlapped ERV hotspots, we used phyloclust through RRphylo R package (Castiglione et al. 2018) with 100 simulations. Since KZNF genes evolve via a birth-death process, we define duplicate genes as genes that exhibit the lowest divergence among all pairwise comparisons. ERV-Mediated LINE Interruption To identify candidate LINEs interrupted by ERVs, we searched for LINE fragments which would be full length (>5000 bp) if connected but exhibit an ERV sequence which splits them with respect to their subfamily consen- sus (supplementary table S7, Supplementary Material on- line). This yielded 322 candidate ERV-mediated LINE interruptions, 121 of which represented lineage-specific LINEs. In this first analysis, we excluded LINEs which showed more than two fragments. If we include those as well, we find 2,664 candidate ERV-mediated LINE interrup- tions. We employed a permutation test to quantitatively assess biased representation of ERVs in LINEs. We did this separately for each ERV subfamily. To do this, we com- pared the observed number of ERV insertions inside LINEs (ERV sequences flanked on both sides by LINE sequences from the same subfamily) to expectations by randomiza- tion 1,000 times. We calculated the proportion of 14 MBE iterations that ERVs interrupted LINEs more than expected to obtain a P-value for each ERV subfamily. Then we per- formed a Bonferroni correction to obtain Q-values (supplementary table S8, Supplementary Material online). Acknowledgments The authors thank Daniel Hartl, Russell Corbett-Detig, Andreas Kautt, and Scott W. Roy for feedback on this Imbeault and Timothy manuscript, and Michael B. Sackton is an Investigator of the Howard Hughes Medical Institute. for helpful discussions. H.E.H. Supplementary material Supplementary data are available at Molecular Biology and Evolution online. Data availability No new data were generated in support of this research. TE models were deposited in the DFAM database. References Abramson NI, Lebedev VS, Bannikova AA, Tesakov AS. 2009. Radiation events in the subfamily Arvicolinae (Rodentia): evi- dence from nuclear genes. Dokl Biol Sci. 428:458–461. Abrusán G, Krambeck H-J. 2006. Competition may determine the di- versity of transposable elements. Theor Popul Biol. 70:364–375. Babushok DV, Ostertag EM, Courtney CE, Choi JM, Kazazian HH Jr. 2006. L1 integration in a transgenic mouse model. Genome Res. 16:240–250. Baillie GJ, van de Lagemaat LN, Baust C, Mager DL. 2004. Multiple groups of endogenous betaretroviruses in mice, rats, and other mammals. J Virol. 78:5784–5798. Bao Z, Eddy SR. 2002. Automated de novo identification of repeat sequence families in sequenced genomes. Genome Res. 12: 1269–1276. Baril T, Hayward A. 2022. Migrators within migrators: exploring in the monarch butterfly, transposable element dynamics Danaus plexippus. Mob DNA. 13:5. Bedford NL, Hoekstra HE. 2015. The natural history of model organ- isms: peromyscus mice as a model for studying natural variation. Elife. 4:e06813. Belshaw R, Pereira V, Katzourakis A, Talbot G, Paces J, Burt A, Tristem M. 2004. Long-term reinfection of the human genome by en- dogenous retroviruses. Proc Natl Acad Sci. 101:4894–4899. Bourque G, Burns KH, Gehring M, Gorbunova V, Seluanov A, Hammell M, Imbeault M, Izsvák Z, Levin HL, Macfarlan TS, et al. 2018. Ten things you should know about transposable ele- ments. Genome Biol. 19:199. Bray N, Dubchak I, Pachter L. 2003. AVID: a global alignment pro- gram. Genome Res. 13:97–102. Brookfield JFY. 2005. The ecology of the genome—mobile DNA ele- ments and their hosts. Nat Rev Genet. 6:128–136. Brouha B, Schustak J, Badge RM, Lutz-Prigge S, Farley AH, Moran JV, Kazazian HH Jr. 2003. Hot L1s account for the bulk of retrotran- sposition in the human population. Proc Natl Acad Sci. 100: 5280–5285. Camacho C, Coulouris G, Avagyan V, Ma N, Papadopoulos J, Bealer K, Madden TL. 2009. BLAST+: architecture and applications. BMC Bioinformatics. 10:421. Transposable Element Interactions Shape the Ecology of the Deer Mouse Genome · https://doi.org/10.1093/molbev/msad069 MBE Campos-Sánchez R, Cremona MA, Pini A, Chiaromonte F, Makova KD. 2016. Integration and fixation preferences of human and mouse endogenous retroviruses uncovered with functional data analysis. PLoS Comput Biol. 12:e1004956. Cantrell MA, Ederer MM, Erickson IK, Swier VJ, Baker RJ, Wichman HA. 2005. MysTR: an endogenous retrovirus family in mammals that is undergoing recent amplifications to unprecedented copy numbers. J Virol. 79:14698–14707. Capella-Gutiérrez S, Silla-Martínez JM, Gabaldón T. 2009. Trimal: a tool for automated alignment trimming in large-scale phylogen- etic analyses. Bioinformatics. 25:1972–1973. Casavant NC, Sherman AN, Wichman HA. 1996. Two persistent LINE-1 lineages in Peromyscus have unequal rates of evolution. Genetics. 142:1289–1298. Castiglione S, Tesone G, Piccolo M, Melchionna M, Mondanaro A, Serio C, Di Febbraro M, Raia P. 2018. A new method for testing evolutionary rate variation and shifts in phenotypic evolution. Methods Ecol Evol. 9:974–983. Chuong EB, Elde NC, Feschotte C. 2017. Regulatory activities of trans- posable elements: from conflicts to benefits. Nat Rev Genet. 18: 71–86. Cosby RL, Chang N-C, Feschotte C. 2019. Host-transposon interac- tions: conflict, cooperation, and cooption. Genes Dev. 33: 1098–1116. Cui J, Tachedjian G, Wang L-F. 2015. Bats and rodents shape mam- malian retroviral phylogeny. Sci Rep. 5:16561. Dechaud C, Volff J-N, Schartl M, Naville M. 2019. Sex and the TEs: transposable elements in sexual development and function in animals. Mob DNA. 10:42. Deininger PL, Batzer MA, Hutchison CA 3rd, Edgell MH. 1992. Master genes in mammalian repetitive DNA amplification. Trends Genet. 8:307–311. Doxiadis GGM, de Groot N, Bontrop RE. 2008. Impact of endogenous intronic retroviruses on major histocompatibility complex class II diversity and stability. J Virol. 82:6667–6677. Eickbush TH. 1992. Transposing without ends: the non-LTR retro- transposable elements. New Biol. 4:430–440. Ellinghaus D, Kurtz S, Willhoeft U. 2008. LTR harvest, an efficient and flexible software for de novo detection of LTR retrotransposons. BMC Bioinformatics. 9:18. Emerson RO, Thomas JH. 2009. Adaptive evolution in zinc finger transcription factors. PLoS Genet. 5:e1000325. Erickson IK, Cantrell MA, Scott L, Wichman HA. 2011. Retrofitting the genome: l1 extinction follows endogenous retroviral expan- sion in a group of muroid rodents. J Virol. 85:12315–12323. Fanning TG. 1983. Size and structure of the highly repetitive BAM HI element in mice. Nucleic Acids Res. 11:5073–5091. Feschotte C. 2008. Transposable elements and the evolution of regu- latory networks. Nat Rev Genet. 9:397–405. Finn RD, Clements J, Eddy SR. 2011. HMMER Web server: interactive sequence similarity searching. Nucleic Acids Res. 39:W29–W37. Flynn JM, Hubley R, Goubert C, Rosen J, Clark AG, Feschotte C, Smit AF. 2020. Repeatmodeler2 for automated genomic discovery of transposable element families. Proc Natl Acad Sci. 117:9451–9457. Ford Doolittle W, Sapienza C. 1980. Selfish genes, the phenotype paradigm and genome evolution. Nature. 284:601–603. Franke V, Ganesh S, Karlic R, Malik R, Pasulka J, Horvat F, Kuzman M, Fulka H, Cernohorska M, Urbanova J, et al. 2017. Long terminal repeats power evolution of genes and gene expression programs in mammalian oocytes and zygotes. Genome Res. 27:1384–1394. Frazer KA, Pachter L, Poliakov A, Rubin EM, Dubchak I. 2004. VISTA: computational tools for comparative genomics. Nucleic Acids Res. 32:W273–W279. Fu L, Niu B, Zhu Z, Wu S, Li W. 2012. CD-HIT: accelerated for cluster- ing the next-generation sequencing data. Bioinformatics. 28: 3150–3152. Fueyo R, Judd J, Feschotte C, Wysocka J. 2022. Roles of transposable elements in the regulation of mammalian transcription. Nat Rev Mol Cell Biol. 23:1–17. Furano AV, Duvernell DD, Boissinot S. 2004. L1 (LINE-1) retrotrans- poson diversity differs dramatically between mammals and fish. Trends Genet. 20:9–14. Gifford RJ, Blomberg J, Coffin JM, Fan H, Heidmann T, Mayer J, Stoye J, Tristem M, Johnson WE. 2018. Nomenclature for endogenous retrovirus (ERV) loci. Retrovirology. 15:59. Gonçalves PR, Christoff AU, Machado LF, Bonvicino CR, Peters FB, Percequillo AR. 2020. Unraveling deep branches of the Sigmodontinae tree (Rodentia: Cricetidae) in Eastern South America. J Mamm Evol. 27:139–160. Grimaldi G, Skowronski J, Singer MF. 1984. Defining the beginning and end of KpnI family segments. EMBO J. 3:1753–1759. Hager ER, Harringmeyer OS, Wooldridge TB, Theingi S, Gable JT, McFadden S, Neugeboren B, Turner KM, Jensen JD, Hoekstra HE. 2022. A chromosomal inversion contributes to divergence in mul- tiple traits between deer mouse ecotypes. Science. 377:399–405. Hao Z, Lv D, Ge Y, Shi J, Weijers D, Yu G, Chen J. 2020. RIdeogram: drawing SVG graphics to visualize and map genome-wide data on the idiograms. PeerJ Comput Sci. 6:e251. Harringmeyer OS, Hoekstra HE. 2022. Chromosomal inversion poly- morphisms shape the genomic landscape of deer mice. Nat Ecol Evol. 6:1965–1975. Hermetz KE, Surti U, Cody JD, Rudd MK. 2012. A recurrent transloca- is mediated by homologous recombination between tion HERV-H elements. Mol Cytogenet. 5:6. Hoover EA, Mullins JI. 1991. Feline leukemia virus infection and dis- eases. J Am Vet Med Assoc. 199:1287–1297. Hubley R, Finn RD, Clements J, Eddy SR, Jones TA, Bao W, Smit AFA, Wheeler TJ. 2016. The dfam database of repetitive DNA families. Nucleic Acids Res. 44:D81–D89. Huerta-Cepas J, Serra F, Bork P. 2016. ETE 3: reconstruction, analysis, and visualization of phylogenomic data. Mol Biol Evol. 33:1635–1638. Hughes JF, Coffin JM. 2001. Evidence for genomic rearrangements mediated by human endogenous retroviruses during primate evolution. Nat Genet. 29:487–489. Ito J, Kimura I, Soper A, Coudray A, Koyanagi Y, Nakaoka H, Inoue I, Turelli P, Trono D, Sato K. 2020. Endogenous retroviruses drive KRAB zinc-finger protein family expression for tumor suppres- sion. Sci Adv. 6:eabc3020. Jern P, Coffin JM. 2008. Effects of retroviruses on host genome func- tion. Annu Rev Genet. 42:709–732. Kapusta A, Kronenberg Z, Lynch VJ, Zhuo X, Ramsay L, Bourque G, Yandell M, Feschotte C. 2013. Transposable elements are major contributors to the origin, diversification, and regulation of ver- tebrate long noncoding RNAs. PLoS Genet. 9:e1003470. Kapusta A, Suh A, Feschotte C. 2017. Dynamics of genome size evolution in birds and mammals. Proc Natl Acad Sci. 114: E1460–E1469. Katoh K, Standley DM. 2013. MAFFT multiple sequence alignment software version 7: improvements in performance and usability. Mol Biol Evol. 30:772–780. Kauzlaric A, Ecco G, Cassano M, Duc J, Imbeault M, Trono D. 2017. The mouse genome displays highly dynamic populations of KRAB-zinc finger protein genes and related genetic units. PLoS One. 12:e0173746. Kazazian HH Jr. 2004. Mobile elements: drivers of genome evolution. Science. 303:1626–1632. Kazazian HH Jr, Moran JV. 1998. The impact of L1 retrotransposons on the human genome. Nat Genet. 19:19–24. Kent TV, Uzunović J, Wright SI. 2017. Coevolution between transpos- able elements and recombination. Philos Trans R Soc Lond B Biol Sci. 372:1736. Klein SJ, O’Neill RJ. 2018. Transposable elements: genome innovation, chromosome diversity, and centromere conflict. Chromosome Res. 26:5–23. Klopfenstein DV, Zhang L, Pedersen BS, Ramírez F, Warwick Vesztrocy A, Naldi A, Mungall CJ, Yunes JM, Botvinnik O, Weigel M, et al. 2018. GOATOOLS: a python library for gene ontology analyses. Sci Rep. 8:10872. 15 Gozashti et al. · https://doi.org/10.1093/molbev/msad069 Kumar S, Stecher G, Suleski M, Hedges SB. 2017. Timetree: a resource for timelines, timetrees, and divergence times. Mol Biol Evol. 34: 1812–1819. Larsson A. 2014. Aliview: a fast and lightweight alignment viewer and editor for large datasets. Bioinformatics. 30:3276–3278. Lee RN, Jaskula JC, van den Bussche RA, Baker RJ, Wichman HA. 1996. Retrotransposon mys was active during evolution of the Peromyscus leucopus-maniculatus complex. J Mol Evol. 42:44–51. Leite RN, Kolokotronis S-O, Almeida FC, Werneck FP, Rogers DS, Weksler M. 2014. In the wake of invasion: tracing the historical biogeography of the South American cricetid radiation (Rodentia, Sigmodontinae). PLoS One. 9:e100687. León-Paniagua L, Navarro-Sigüenza AG, Hernández-Baños BE, Morales JC. 2007. Diversification of the arboreal mice of the genus Habromys (Rodentia: Cricetidae: Neotominae) in the mesoamerican highlands. Mol Phylogenet Evol. 42:653–664. Letunic I, Bork P. 2021. Interactive tree of life (iTOL) v5: an online tool for phylogenetic tree display and annotation. Nucleic Acids Res. 49:W293–W296. Lindblad-Toh K, Wade CM, Mikkelsen TS, Karlsson EK, Jaffe DB, Kamal M, Clamp M, Chang JL, Kulbokas EJ 3rd, Zody MC, et al. 2005. Genome sequence, comparative analysis and haplotype structure of the domestic dog. Nature. 438:803–819. Llorens C, Futami R, Covelli L, Domínguez-Escribá L, Viu JM, Tamarit D, Aguilar-Rodríguez J, Vicente-Ripolles M, Fuster G, Bernet GP, et al. 2011. The gypsy database (GyDB) of mobile genetic ele- ments: release 2.0. Nucleic Acids Res. 39:D70–D74. Mager DL, Freeman JD. 2000. Novel mouse type D endogenous pro- viruses and ETn elements share long terminal repeat and internal sequences. J Virol. 74:7221–7229. Mager DL, Stoye JP. 2015. Mammalian endogenous retroviruses. Microbiol Spectr. 3:1. Marchler-Bauer A, Bo Y, Han L, He J, Lanczycki CJ, Lu S, Chitsaz F, Derbyshire MK, Geer RC, Gonzales NR, et al. 2017. CDD/SPARCLE: functional classification of proteins via subfamily domain architectures. Nucleic Acids Res. 45:D200–D203. Marchler-Bauer A, Derbyshire MK, Gonzales NR, Lu S, Chitsaz F, Geer LY, Geer RC, He J, Gwadz M, Hurwitz DI, et al. 2015. CDD: NCBI’s conserved domain database. Nucleic Acids Res. 43:D222–D226. Medstrand P, van de Lagemaat LN, Mager DL. 2002. Retroelement distributions in the human genome: variations associated with age and proximity to genes. Genome Res. 12:1483–1495. Minh BQ, Schmidt HA, Chernomor O, Schrempf D, Woodhams MD, von Haeseler A, Lanfear R. 2020. IQ-TREE 2: new models and ef- ficient methods for phylogenetic inference in the genomic era. Mol Biol Evol. 37:1530–1534. Mistry J, Chuguransky S, Williams L, Qureshi M, Salazar GA, Sonnhammer ELL, Tosatto SCE, Paladin L, Raj S, Richardson LJ, et al. 2021. Pfam: the protein families database in 2021. Nucleic Acids Res. 49:D412–D419. Najafabadi HS, Mnaimneh S, Schmitges FW, Garton M, Lam KN, Yang A, Albu M, Weirauch MT, Radovani E, Kim PM, et al. 2015. C2h2 zinc finger proteins greatly expand the human regu- latory lexicon. Nat Biotechnol. 33:555–562. Nellåker C, Keane TM, Yalcin B, Wong K, Agam A, Belgard TG, Flint J, Adams DJ, Frankel WN, Ponting CP. 2012. The genomic land- scape shaped by selection on transposable elements across 18 mouse strains. Genome Biol. 13:R45. Numanagic I, Gökkaya AS, Zhang L, Berger B, Alkan C, Hach F. 2018. Fast characterization of segmental duplications in genome as- semblies. Bioinformatics. 34:i706–i714. O’Leary NA, Wright MW, Brister JR, Ciufo S, Haddad D, McVeigh R, Rajput B, Robbertse B, Smith-White B, Ako-Adjei D, et al. 2016. Reference se- quence (RefSeq) database at NCBI: current status, taxonomic expan- sion, and functional annotation. Nucleic Acids Res. 44:D733–D745. Orgel LE, Crick FH. 1980. Selfish DNA: the ultimate parasite. Nature. 284:604–607. Osmanski AB, Paulat NS, Korstian J, Grimshaw JR, Halsey M, Sullivan KAM, Moreno-Santillán DD, Crookshanks C, Roberts J, Garcia C, 16 MBE et al. 2022. Insights into mammalian TE diversity via the curation of 248 mammalian genome assemblies. bioRxiv. [cited 2023 Jan 20]. Available from: https://www.biorxiv.org/content/10.1101/ 2022.12.28.522108v1.full Ou S, Su W, Liao Y, Chougule K, Agda JRA, Hellinga AJ, Lugo CSB, Elliott TA, Ware D, Peterson T, et al. 2019. Benchmarking trans- posable element annotation methods for creation of a stream- lined, comprehensive pipeline. Genome Biol. 20:275. Peona V, Blom MPK, Xu L, Burri R, Sullivan S, Bunikis I, Liachko I, Haryoko T, Jønsson KA, Zhou Q, et al. 2021. Identifying the causes and consequences of assembly gaps using a multiplatform genome assembly of a bird-of-paradise. Mol Ecol Resour. 21: 263–286. Platt RN 2nd, Blanco-Berdugo L, Ray DA. 2016. Accurate transpos- able element annotation is vital when analyzing new genome as- semblies. Genome Biol Evol. 8:403–410. Platt RN 2nd, Vandewege MW, Ray DA. 2018. Mammalian transpos- impacts on genome evolution. able elements and their Chromosome Res. 26:25–43. Polat M, Takeshima S-N, Aida Y. 2017. Epidemiology and genetic di- versity of bovine leukemia virus. Virol J. 14:209. Pontis J, Planet E, Offner S, Turelli P, Duc J, Coudray A, Theunissen TW, Jaenisch R, Trono D. 2019. Hominoid-specific transposable elements and KZFPs facilitate human embryonic genome activa- tion and control transcription in naive human ESCs. Cell Stem Cell. 24:724–735.e5. Price AL, Jones NC, Pevzner PA. 2005. De novo identification of repeat families in large genomes. Bioinformatics. 21:i351–i358. Quinlan AR, Hall IM. 2010. BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics. 26:841–842. Ribet D, Dewannieux M, Heidmann T. 2004. An active murine trans- poson family pair: retrotransposition of “master” MusD copies and ETn trans-mobilization. Genome Res. 14:2261–2267. Ribet D, Harper F, Dupressoir A, Dewannieux M, Pierron G, Heidmann T. 2008. An infectious progenitor for the murine IAP retrotransposon: emergence of an intracellular genetic para- site from an ancient retrovirus. Genome Res. 18:597–609. Rice P, Longden I, Bleasby A. 2000. EMBOSS: the European molecular biology open software suite. Trends Genet. 16:276–277. Rubino F, Creevey CJ. 2014. MGkit: metagenomic framework for the study of microbial communities. [Internet]. Available from: https://figshare.com/articles/poster/MGkit_Metagenomic_ Framework_For_The_Study_Of_Microbial_Communities/1269288/1 Sakashita A, Maezawa S, Takahashi K, Alavattam KG, Yukawa M, Hu Y-C, Kojima S, Parrish NF, Barski A, Pavlicev M, et al. 2020. Endogenous retroviruses drive species-specific germline tran- scriptomes in mammals. Nat Struct Mol Biol. 27:967–977. Senft AD, Macfarlan TS. 2021. Transposable elements shape the evolution of mammalian development. Nat Rev Genet. 22: 691–711. Sievers F, Wilm A, Dineen D, Gibson TJ, Karplus K, Li W, Lopez R, McWilliam H, Remmert M, Söding J, et al. 2011. Fast, scalable generation of high-quality protein multiple sequence alignments using clustal Omega. Mol Syst Biol. 7:539. Sotero-Caio CG, Platt RN 2nd, Suh A, Ray DA. 2017. Evolution and in vertebrate genomes. diversity of transposable elements Genome Biol Evol. 9:161–177. Storer JM, Hubley R, Rosen J, Smit AFA. 2021. Curation guidelines for de novo generated transposable element families. Curr Protoc. 1: e154. Szak ST, Pickeral OK, Makalowski W, Boguski MS, Landsman D, Boeke JD. 2002. Molecular archeology of L1 insertions in the hu- man genome. Genome Biol. 3:research0052. Thomas JH, Schneider S. 2011. Coevolution of retroelements and tandem zinc finger genes. Genome Res. 21:1800–1812. van de Lagemaat LN, Landry J-R, Mager DL, Medstrand P. 2003. Transposable elements in mammals promote regulatory vari- ation and diversification of genes with specialized functions. Trends Genet. 19:530–536. Transposable Element Interactions Shape the Ecology of the Deer Mouse Genome · https://doi.org/10.1093/molbev/msad069 MBE Venner S, Feschotte C, Biémont C. 2009. Dynamics of transposable elements: towards a community ecology of the genome. Trends Genet. 25:317–323. Wells JN, Feschotte C. 2020. A field guide to eukaryotic transposable elements. Annu Rev Genet. 54:539–561. Wichman HA, Potter SS, Pine DS. 1985. Mys, a family of mammalian isolated by phylogenetic screening. transposable elements Nature. 317:77–81. Wicker T, Sabot F, Hua-Van A, Bennetzen JL, Capy P, Chalhoub B, Flavell A, Leroy P, Morgante M, Panaud O, et al. 2007. A unified classification system for eukaryotic transposable elements. Nat Rev Genet. 8:973–982. Wolf G, de Iaco A, Sun M-A, Bruno M, Tinkham M, Hoang D, Mitra A, Ralls S, Trono D, Macfarlan TS. 2020. KRAB-zinc finger protein gene expansion in response to active retrotransposons in the murine lineage. Elife. 9:e56337. Wolf G, Greenberg D, Macfarlan TS. 2015. Spotting the enemy with- in: targeted silencing of foreign DNA in mammalian genomes by the Krüppel-associated box zinc finger protein family. Mob DNA. 6:17. Wong WY, Simakov O. 2019. Repeatcraft: a meta-pipeline for repeti- tive element de-fragmentation and annotation. Bioinformatics. 35:1051–1052. Xu Z, Wang H. 2007. LTR_FINDER: an efficient tool for the prediction of full-length LTR retrotransposons. Nucleic Acids Res. 35: W265–W268. Yang Z. 1994. Maximum likelihood phylogenetic estimation from DNA sequences with variable rates over sites: approximate methods. J Mol Evol. 39:306–314. Yang L, Scott L, Wichman HA. 2019. Tracing the history of LINE and SINE extinction in sigmodontine rodents. Mob DNA. 10:22. Yang P, Wang Y, Macfarlan TS. 2017. The role of KRAB-ZFPs in trans- posable element repression and mammalian evolution. Trends Genet. 33:871–881. Zemojtel T, Penzkofer T, Schultz J, Dandekar T, Badge R, Vingron M. 2007. Exonization of active mouse L1s: a driver of transcriptome evolution? BMC Genomics. 8:392. Zhou W, Liang G, Molloy PL, Jones PA. 2020. DNA methylation en- ables transposable element-driven genome expansion. Proc Natl Acad Sci. 117:1935. 17
null
10.1088_2752-5295_acf4b5.pdf
Data availability statement The data that support the findings of this study are openly available in the Harvard Dataverse at https://doi. org/10.7910/DVN/2UT4GM (Hauer 2023).
Data availability statement The data that support the findings of this study are openly available in the Harvard Dataverse at https://doi. org/10.7910/DVN/2UT4GM (Hauer 2023) .
OPEN ACCESS RECEIVED 14 December 2022 REVISED 14 August 2023 ACCEPTED FOR PUBLICATION 29 August 2023 PUBLISHED 18 September 2023 Original Content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. Environ. Res.: Climate 2 (2023) 045004 https://doi.org/10.1088/2752-5295/acf4b5 PAPER Sea level rise already delays coastal commuters Mathew E Hauer1,4, Valerie Mueller2,3,∗ and Glenn Sheriff2 1 Department of Sociology, Florida State University, Tallahassee, FL, United States of America 2 School of Politics and Global Studies, Arizona State University, Tempe, AZ, United States of America 3 International Food Policy Research Institute, Washington, DC United States of America 4 Mathew Hauer is the primary author of the paper. Map data copyrighted OpenStreetMap contributors and available from www. openstreetmap.org. This material is based upon work supported by the National Science Foundation under Grant Number 1939841. ∗ Author to whom any correspondence should be addressed. E-mail: vmuelle1@asu.edu Keywords: sea level rise, adaptation, transportation, climate change impacts, king tides Supplementary material for this article is available online Abstract Although the most dire societal impacts of sea-level rise (SLR) typically manifest toward the end of the 21st century, many coastal communities face challenges in the present due to recurrent tidal flooding. Few studies have documented transportation disruptions due to tidal flooding in the recent past. Here, we address this issue by combining home and work locations for approximately 500 million commuters in coastal US counties from 2002 to 2017. We find tidal flooding delays coastal commuters by approximately 22 min per year in 2015–2017, increasing to between 200 and 650 min by 2060 under various SLR scenarios. Adjustments in residential and work locations reduce the growth in commuting delays for approximately 40% of US counties. For residents in coastal counties, SLR is not a distant threat—it is already lapping at their toes. 1. Main text Sea-level rise (SLR) is one of the most visible and costly impacts of climate change (Church et al 2013, Oppenheimer et al 2019). With sea levels expected to rise up to 2.5 m by 2100 (Sweet et al 2017, Oppenheimer et al 2019), scientists routinely quantify the potential multitude of impacts of SLR over the next eighty to 2000+ years (Strauss et al 2015, Clark et al 2016, Oppenheimer et al 2019), focusing on impacts deep into the future when they will be presumed greatest. Although the most worrying societal impacts of SLR (displacement and permanent submergence) typically manifest toward the end of the 21st century (Kulp and Strauss 2019, Oppenheimer et al 2019, Hauer et al 2020), many coastal communities face daunting impacts in the present due to recurrent tidal or high tide flooding (Dahl et al 2017, Moftakhari et al 2017, Sweet et al 2018). We use ‘tidal flooding’ to refer to only tide-based inundation, i.e. excluding precipitation (Hague and Taylor 2021). Our definition differs from that used by other studies focusing on tidal flooding above minimum thresholds (e.g. Sweet et al 2018)4. Studies focusing exclusively on directly affected areas find tidal flooding causes coastal erosion (Hinkel et al 2013), saltwater intrusion (Chang et al 2011), reduced property values (McAlpine and Porter 2018), damaged property (Bukvic and Harrald 2019), submerged transportation routes (Jacobs et al 2018), and exposure of thousands of people to general flood risks (Kulp and Strauss 2019). Given the interconnectedness of coastal communities, e.g. via commuting (Kasmalkar et al 2021), SLR impacts will likely indirectly affect inland, higher elevation coastal areas. The extent to which coastal commutes might be delayed due to inundation on roadways is presently uncertain for the coastal United States. Some research has linked tidal flooding to transportation networks. These studies use prescribed flood levels rather than historical tide data (Jacobs et al 2018, Kasmalkar et al 2020), limited transportation routes 4 Many factors can influence sea levels including storm surge, precipitation, currents, trapped waves, etc (Ezer and Atkinson 2014, Barnard et al 2017). Our definition of ‘tidal flooding’ includes all factors as measured at NOAA stations. © 2023 The Author(s). Published by IOP Publishing Ltd Environ. Res.: Climate 2 (2023) 045004 M E Hauer et al to major roads (Jacobs et al 2018, Kasmalkar et al 2020), annual daily traffic data rather than actual commuter data (Jacobs et al 2018), or limit their analysis to specific areas (Jacobs et al 2018, Kasmalkar et al 2020, Shen and Kim 2020, Hauer et al 2021, Praharaj et al 2021). Most notably, Jacobs et al (2018) and Fant et al (2021) link average annual traffic on major roadways to road inundation and calculate road-segment-specific travel delays in the absence of alternative routing to ‘dry’ routes. No such national-level estimate of commuting delays presently exist that account for alternative routing to avoid or minimize delays along flooded roadways. These shortcomings combine to create incomplete estimates of SLR and tidal flooding impacts on commuting in coastal communities. Using national-level routing, commuting, and elevation data, we overcome these issues to estimate the burden SLR driven tidal flooding imposes on commuting times with a dynamic routing algorithm to allow for commuters to route around flooding-related delays. By combining commuter data with detailed road networks in a flood hazard model we assess the commuting delays associated with present and future SLR driven tidal flooding. Our analytical framework addresses three primary questions concerning SLR impacts in coastal communities in the United States: (1) What is the current commuting time burden imposed by SLR driven tidal flooding and what areas does SLR driven tidal flooding burden the most? (2) How have changes in commuting behavior and residential location affected this burden? And (3) What is the future commuting time burden, absent further behavioral changes? We estimate commuting delays attributable to recurrent tidal flooding by combining multiple large-scale georeferenced data sources within a travel optimization routine. Namely, we combine (1) the home and work (HW) locations for 74 million census block group (CBG) located in 222 coastal counties and 158 non-coastal counties for 500 million commuter-years in the period 2002–2017 (U.S. Census Bureau 2020), (2) a complete road network from Open Street Map (OpenStreetMap contributors 2017), (3) data from eighty-four tide stations across the US from National Oceanographic and Atmospheric Administration’s (NOAA) Center for Operational Oceanographic Products and Services (CO-OPs) tide gauge database (Chamberlain 2020, NOAA CO-OPs 2020), (4) the National Levee Database (US Army Corps of Engineers 2020), and (5) digital elevation models from both NOAA (NOAA 2020) and the US Geological Survey (Gesch et al 2002). We converted OSM’s street network into dual-weighted directed graphs and calculated the minimum travel times between HW CBGs, conditional on flood depth on roadways, between 2002 and 2017 and projected in 2060 using NOAA’s global mean SLR scenarios (Sweet et al 2017). As roadways become inundated over time and travel velocity slows, we dynamically adjust travel routes to ensure commuters select the path with the least travel time for a given amount of roadway inundation. Both the change in the commuters along a HW pair and the change in tide heights due to SLR contribute to changes in commuting delays over time. Changes in commuters along a HW pair can theoretically reflect adaptive residential or work location changes to reduce flood-related commuting delays—all else equal—and we isolate this behavior using a two-factor decomposition (Gupta 1991). We refer to this adaptive behavior as ‘accommodation’, as opposed to ‘retreat’ or ‘adaptation’ broadly, since it is an adaptation response that reduces vulnerability to commuting delays without protection or necessarily retreat (Oppenheimer et al 2019). 2. Methods and materials First, we describe the data sets used in our analysis. Second, we describe the methods to estimate commuting delays. Data. We use five primary data sources to estimate commuting delays: commuting information, road network data, tide gauge data, levee data, and digital elevation models. Commuting information comes from the U.S. Census Bureau’s Longitudinal Employer-Household Dynamics (LEHD) Origin-Destination Employment Statistics (LODES) for the period 2002–2017 (U.S. Census Bureau 2020). LEHD-LODES is a partially synthetic dataset for paired CBG residential (home) and employment (work) locations in the United States. We downloaded these data for all coastal counties (n = 222), limiting our analysis to respondents with differing home and work CBGs. For computational tractability, we subset the complete ‘commuting area’ of any given coastal county to those counties that are (i) within 100 miles of the coastal county centroid, and that (ii) contain at least 1% of the commuters into the coastal county. As illustrated in figure 1, on average this procedure captures nearly 90% of a county’s commuters (median 89%, 80% of counties capture at least 81% of commuters). Overall, it yields 74 million CBG origin-destination pairs. Due to unavailability of historical data, data sharing limitations, or quality assurance problems, three states in our analysis have incomplete commuting data: New Hampshire in 2002, Mississippi in 2002 and 2003, and Massachusetts from 2002–2010. Massachusetts lack of data for 2002–2010 also impacts New 2 Environ. Res.: Climate 2 (2023) 045004 M E Hauer et al Figure 1. Distribution of county commuters within 100 miles of destination county and containing at least 1% of commuters. The median county captures approximately 90% of commuters. Hampshire, as nearly 30% of New Hampshire commuters are missing between 2003 and 2010. All other states contain complete commuting data for the entire period. We therefore exclude Mississippi commuters prior to 2004 and Massachusetts and New Hampshire commuters prior to 2011. We obtain road network data from Open Street Map (OSM). OSM is an open-source, free, editable map of world street networks and features built from volunteers. OSM data include speed limits, network connectivity, and road segment characteristics useful for our analysis such as ‘tunnel’ and ‘bridge.’ Previous analyses of the completeness and accuracy of OSM data suggest sufficient georeferenced accuracy (El-Ashmawy 2016, Brovelli et al 2017). We obtain all road network types (primary, motorway, residential, etc) except for service road types for any destination coastal county, any origin commuter county within 100 miles and containing at least 1% of commuters, and any county in between for our analysis. We obtain tide gauge data for 84 tide stations across the US from NOAA’s Center for Operational Oceanographic Products and Services (CO-OPs) tide gauge database. Figure 2 shows the geographic distribution of tide gauages. We download hourly verified water heights for the period 2002–2019 using the ‘rnoaa’ package (Chamberlain 2020), in the R programming language and subset the tide gauges for the max water height during readings between the hours of 05:00 and 20:00 for Monday through Friday—the primary commuting days and hours in most areas. We normalize the annual number of commuting days to a standardized 250-day year. To correct for extreme water levels which might correspond with extreme weather events such as tropical cyclones, we use an outlier detection algorithm (Chen and Liu 1993, de Lacalle 2019) to search and correct for these extreme water levels by building counterfactual tide gauge series. We use the counterfactual value in time periods where the observed gauge reading exceeds τ ⩾ 10 (t-score), correcting only the most extreme water levels. We assign a tide gauge to each county based on shortest distance as the crow flies. For computational tractability, we group tide gauge readings into five discrete bins corresponding to the <90th, <98th, <99.5th, <99.9th, and >99.9th percentiles for each gauge. Notably, the two periods in our historical analysis (2002–2004 and 2015–2017) take place at different points in both the perigean and El Ni˜no cycles (Goodman et al 2018). Consequently, this analysis is best interpreted as measuring the effects of changes in water levels between these periods resulting from the combined impact of all influences of water levels at tide gauges, rather than those attributable to global warming-induced sea level rise alone. Levee data come from the National Levee Database (NLD) maintained by the US Army Corps of Engineers. The NLD is a nationally comprehensive database on levee systems in the United States. We obtained georeferenced levee information for the 22 states in our analysis. Digital elevation models (DEMs) come from two sources. For areas threatened by SLR we use NOAA’s Digital Coast, which is based on 1/9 arcsecond (10 m) or less horizontal resolution. For counties outside of 3 Environ. Res.: Climate 2 (2023) 045004 M E Hauer et al Figure 2. Geographic distribution of the 84 tide gauges used. NOAA’s coverage, we download DEMs from the National Elevation Dataset (NED) at 1/3 arcsecond resolution (Gesch et al 2002). Methods. We calculate commuting delay in five steps (Hauer et al 2021). First, we model road inundation depth as a function of a given tide gauge reading. Second, we model travel velocity for each road segment as a function of inundation. Third, we select the fastest route between any two points, conditional on this velocity. Fourth, we use the observed distribution of annual tide gauge readings to calculate the average annual flooding delay for a given pair of points. Fifth, we obtain aggregate commuting delays for any given area as the weighted average annual delay over all home-work location pairs, with weights corresponding to the number of commuters traveling from a given home location to a work location in the period. Step 1: Road inundation as a function of tide guage. We combine road network, DEM, levee, and tide gauge data to calculate flood inundation for each road segment. We apply a ‘bathtub’ hydrological model, similar to NOAA’s tidal flooding mapping procedure 5, calculating road segment inundation levels as the difference between the water levels reported at the tide gauge and the road surface elevation. Any road segment in the OSM wholly located inside of the area of a levee is given an elevation of 100 m, setting its elevation far above any flood height, and any road segment partially located inside of the area of a levee (e.g. when a road segment begins outside of a levee and terminates inside of a levee) we assume the road segment elevation is the mean of the two points. Any road segment located outside of any DEM for any reason (due to being either very far inland via a circuitous route) is also assigned an elevation of 100 m. To aid in computational tractability, we divide tide gauge data into the five discrete bins outlined above. With each road segment assigned an elevation value, z, we subtract tide gauge bin gb from road segment z to produce the flood depth i for any given segment. Step 2: Travel velocity as a function of local inundation. We model travel velocity v in mph along any given road segment as a function of the speed limit L and the mm of road inundation i: 8 < v(i) = : (cid:8) L min 1 L, 0.6[0.0009i2 − 0.5529i + 86.9448] (cid:9) i <= 0 0 < i < 300 300 < i. (1) This is the depth-disruption function fitted by Pregnolato et al (2017) but modified in four key ways. First, if a road segment elevation z is above the tide gauge level gb and thus road inundation i is less than 0, we assume vehicle speed is the speed limit, i.e. the road segment is unaffected by tidal flooding. Second, if 5 https://coast.noaa.gov/data/digitalcoast/pdf/slr-inundation-methods.pdf. 4 Environ. Res.: Climate 2 (2023) 045004 M E Hauer et al flooding is less than 300 mm, we assume vehicle speed to be the minimum of the speed limit or the maximum safe speed estimated by Pregnolato et al (2017). This ensures when the maximum safe speed is above the speed limit, we assign travel velocity as the speed limit. Third, we assume a maximum safe speed of one mph, rather than zero, after inundation reaches 300 mm, which is the average depth at which passenger vehicles start to float. Fourth, any road segment in the OSM listed as either a ‘tunnel’ or a ‘bridge’ is presumed to be unaffected by road inundation and thus is assigned the speed limit. Step 3: Calculating optimal travel times. To find the optimal travel route between two points we first identify each individual origin (home) and destination (work) location conditional on tide gauge bin gb. We assume that the precise location of each origin and destination is the population weighted mean center for each CBG. We then locate the nearest road segment to each population weighted mean centroid as the origin and destination. Road networks can be conceptualized as dual-weighted directed graphs where the weight is given as the travel velocity. We use the dodgr (Padgham 2019) package in R to convert each county’s road network into a dual-weighted directed graph. The travel time along each road segment is calculated based on the segment’s length and its speed limit, producing a ‘weighted’ travel time along a road segment. The travel time between any two points is then calculated as the sum of the weight values. The ‘fastest’ route is simply calculated as the minimal total sum of weight values for all routes between two points. We calculate the fastest route between each origin and destination CBG under each tide gauge bin b and one ‘dry’ route, assuming the absence of any flooding. Step 4: Average annual flooding delays for each home-work pair. Let µh,w y denote daily mean annual flooding delay in min per commuter for any HW pair (h, w) in year y. It is the weighted average difference between the commute time conditional on the inundation level implied by tide bin b, mh,w , with weights pby, the number of days in year y with tide levels in bin b: , and the ‘dry’ commute time mh,w b b h P 5 b=1 µh,w y = i − mh,w by mh,w b P 5 b=1 pby · pby . Since the number of workdays varies year to year, we calculate the total annual round trip tide-induced delay per commuter for a normalized 250 workday year as Given a number of commuters, c h,w y , the total min of delay for (h, w) is then y = 2 · 250 · µh,w t h,w y . y = t h,w dh,w y c h,w y . Step 5: Projected SLR commuting delays. To compute anticipated water depths in 2060, we add the change in NOAA global mean low, intermediate, and extreme SLR scenarios in Sweet et al (2017) to the underlying tide distributions for each tide gauge. These scenarios correspond with 2100 SLR values of 0.3 m, 0.9 m, and 2.5 m, respectively, which translate to 2060 values of 0.19 m, 0.45 m, and 0.9 m. Our projections assume no changes in the shape of the distribution of annual tide values, just a shift in the distribution (Kirezci et al 2020, Taherkhani et al 2020). Accommodation effects from tide effects. For any given (h, w) pair between 2002 and 2017, changes in total commuting delays can come from two main changes: a change in the number of commuters (which we refer to as the ‘accommodation effect’) and a change in the tide values (which we refer to as the ‘tide effect’). We employ a Das Gupta decomposition (Gupta 1991) to isolate the change in commuting delays attributable to these two factors. For a given (h, w) pair and two periods (1,2) the change in the total delay can be expressed: dh,w 2 − dh,w 1 = 3. Results h 2 1 + th,w th,w 2 | c h,w 2 {z Accomodation effect − c h,w 1 i } + h 1 + c h,w c h,w 2 | 2 th,w 2 {z Tide effect − th,w 1 i } . (2) Our models show that tidal flooding delayed the average commuter by a total of 23.3 min in 2017 (figure 3). The total delay increased from 11.9 in 2002—more than doubling in fifteen years. Even with some of the lowest SLR scenarios (0.3 m by 2100), this commuting delay is projected to increase by a factor of seven by 5 Environ. Res.: Climate 2 (2023) 045004 M E Hauer et al Figure 3. Commuting delays in the United States, 2002–2017 and in 2060 under varying SLR scenarios. Estimates reflect total annual commuting delay for workers employed in coastal communities. Colored dots in 2060 correspond to predicted tide levels based on NOAA SLR curves (Sweet et al 2017) for low (0.3 m), intermediate (1.0 m), and extreme (2.5 m) by 2100. Uncertainty reflects the 90th percentile confidence or prediction interval. Sea-level rise is already delaying US coastal commuters. Figure 4. Commuting delays in US States, 2002–2017. Estimates reflect total annual commuting delay for workers employed in coastal communities. Results for Mississippi in 2002/2003, and New Hampshire and Massachusetts in 2002–2010 excluded due to data quality issues. Every coastal state experiences some amount of SLR commuting delay and all states experience increased delays since 2002. 2060 (183 min) and could increase by a factor of twenty-four under high end SLR scenarios by 2060 (643 min with 2.5 m of SLR by 2100). All coastal US states experience tidal-flood induced commuting delays and an increase in these delays since 2002 (figure 4). In 2015–2017, Georgia (251 min), North Carolina (92.4 min), and Massachusetts (64.5 min) were the only states with more than one hour of annual commuting delays (table 1). By 2060, 6 Environ. Res.: Climate 2 (2023) 045004 M E Hauer et al Table 1. Average Annual Commuting Delays in min for US States, 2002–2060. 2060 Intermediate (low-extreme) values refer to SLR under three 2100 scenarios — 1.0 m [0.3 m - 2.5 m]. These correspond to SLR in 2060 of 0.45 m [0.19 m - 0.9 m]. State 2002–2004 2015–2017 % Increase 2060 Intermediate [low-extreme] state 2002–2004 2015–2017 % Increase TOT 9.94 7.59 AL CA 19.6 CT 0.469 2.59 DE FL 1.64 GA 86.1 LA 6.11 ME 4.05 MD 2.3 MA — 22.4 16.5 32.2 0.797 3.94 7.54 251 7.98 7.54 5.46 64.5 126% 117% 66% 69% 51% 360% 192% 30% 87% 138% — MS — 359 [183 - 643] NH — 326 [136 - 495] NJ 9.45 457 [246 - 725] NY 11.4 27.6 [9.58 - 62.9] NC 55.8 223 [58.2 - 545] OR 13.1 109 [53.4 - 159] 0.108 2794 [1451 - 5296] RI SC 10.6 687 [155 - 1189] TX 0.0753 91.3 [51.1 - 187] 186 [68.8 - 284] VA 5.97 1051 [560 - 2148] WA 4.73 0.0711 13.4 20.9 18.4 92.4 19 0.178 34.2 0.148 14.9 8.3 — — 123% 60% 66% 45% 63% 225% 96% 150% 75% 2060 Intermediate [low-extreme] 16.5 [2.42 - 31.9] 227 [118 - 513] 411 [193 - 782] 325 [174 - 652] 2052 [864 - 3669] 209 [119 - 385] 5.3 [2.14 - 9.55] 362 [194 - 586] 3.7 [1.19 - 7.38] 269 [130 - 476] 92.6 [53.4 - 168] Figure 5. Annual Commuting delays in 2015–2017 and in 2060 with intermediate SLR (0.9 m by 2100) for US counties. Counties not included in the study area are colored in gray. even low-end SLR (0.3 m by 2100) will increase commuting delays in coastal areas by an order of magnitude within the next forty years without some form of adaptation. Commuting delays are unevenly distributed across space (figure 5). For many coastal counties, SLR commuting delays are presently minimal with the median annual commuting delay of just over 3 min in 2017 but by 2060 with intermediate SLR, we estimate the median annual commuting delay will increase to 99.5 min. Three counties—Washington NC (3057 min), McIntosh GA (1275 min), and Tyrrell NC (695 min)—experience over 500 min of annual delay. Twenty-three counties experienced more than 100 min of commuter delay in 2017—approximately the median delay in 2060 under intermediate SLR. For the residents in these counties, SLR is not a distant threat—it is already lapping at their toes. Unlike population exposure to inundation (Hauer et al 2016), which heavily threatens the US South, SLR impacts are not isolated to specific regions of the US. Places like San Francisco CA, Boston MA, and Savannah GA already see significant commuting delays due to recurrent tidal flooding. These are areas with high king tides, generally longer distance commutes, roadways along low-elevation coastlines, and generally, few, if any, alternative routes. Other major population centers, like Houston TX, Los Angeles CA, and New York NY, presently experience minimal, if any, commuting delays attributable to more alternative routes along higher-elevation inland areas. Areas with the greatest delays tend to contain more long-distance commutes, along roadways with higher maximum speed, containing fewer alternative, ‘drier’ routes. Additionally, we investigate how adjustments to commuter home or work locations (or both) have affected commuting delays. Some HW pairs might experience increasing delays due to tidal flooding but decreasing delays due to changes in the number of commuters along that HW pair. These changes are unlikely to be due to technological shifts to increased remote work as the number of non-commuters 7 Environ. Res.: Climate 2 (2023) 045004 M E Hauer et al Figure 6. Accommodation to commuting delays between 2002–2017 for Home-Work pairs with non-zero impacts in 2015-2017 for US counties . (a) shows the top and bottom eight counties with the greatest and least Accommodation effects. ‘Commuters’ refers to the number of commuters with non-zero commuting delays in 2015–2017. ‘Tot Increase’ refers to the total increase in commuting delays between 2002 and 2017. ‘Tide Effect’ is the raw increase in commuting delays if the home-work commuters remained unchanged in 2002–2004 but exposed to tides in 2015–2017. ‘Accommodation’ refers to the decomposed effect due to changes in either home or work locations between 2002–2004 and 2015–2017 with tides kept constant in 2002–2004. (b) maps these accommodation effects. Counties not included in the study area or excluded due to data limitations are colored in gray. (i.e. those commuting to the CBG in which they reside) accounted for 1.33% of commuters in 2002 and 1.39% in 2017, suggesting little increase in ‘home commuting.’ We decompose the change between 2002–2004 and 2015–2017 into the contributions from rising water levels and to changes in the allocation of commuters along HW pairs (what we call the ‘accommodation effect’(Hauer et al 2021)) using a two-factor Das Gupta style decomposition (Gupta 1991) for HW pairs with non-zero commuting delays in 2015–2017 (see Methods). Changes in commuter behavior regarding choices of residential and workplace location reduced the impact of flooding delays in 95 (43%) counties (figure 6). However, these accommodation effects are not uniform across the US coast. In parts of California, Georgia, and Florida, for example, the number of commuters increased in highly affected routes, further exacerbating flood-induced delays. 4. Discussion Understanding of SLR thresholds at which people adopt adaptive behavior to reduce their exposure and vulnerability to SLR impacts remains severely limited (Hauer et al 2020), though there is a general understanding that people theoretically must adapt to or accommodate higher water levels. Most examples of contemporary SLR adaptation tend to focus on large governmental infrastructure projects such as deployment of pumps or protective infrastructure (Fu et al 2017), whereas examples of individual adaptation strategies are mostly limited to either theoretical options for coastal residents (Kwadijk et al 2010) or localized case studies with minimal identification of thresholds or tipping points (Jamero et al 2017). Here, we describe how changes in commuter behavior across the coastal US have amplified or reduced the impact of rising tides on commute times. It remains to be seen whether some of the anticipated commuting delays (figure 5 and supplementary material) over the coming decades leave enough space for further accommodation to occur without large scale intervention. Much of the literature surrounding SLR impacts assesses flood impacts on housing (Hallegatte et al 2013, Hinkel et al 2014), critical infrastructure (Heberger et al 2011, Storlazzi et al 2018), or populations (Strauss et al 2015, Kulp and Strauss 2019). Temporal horizons associated with these impact assessments can be as short as 100 years (Hinkel et al 2014) and as long as 2000 years (Strauss et al 2015), pushing much of these impacts into the deep future. Additionally in many of these impact assessments, what is ‘impacted’ is defined by residence within a flooded area. Our results demonstrate two important considerations for SLR impact assessments. First, SLR impacts in the form of commuting delays are presently occurring in US coastal 8 Environ. Res.: Climate 2 (2023) 045004 M E Hauer et al communities. In principle, changes in commuter behavior could ameliorate these delays if people were to choose HW locations less vulnerable to tidal flooding. Although we are not able to determine a causal relationship between flood exposure and commuting choices, it does not appear that such accommodating behavior adaptations have been sufficiently widespread as to effectively counter the effect of rising tides. Changes in commuter behavior between 2002 and 2017 reduce flooding delays in less than half of the counties compared to a counterfactual in which residential and workplace locations did not change. Assuming a value of time equal to the median hourly wages of workers in 2019 in each county6, the cost of commuting delays increased from $154.7 M in 2002 to more than $439.7 M in 2017. Without significant additional accommodation, adaptation, or mitigation of carbon emissions, these costs could top $3.4 B in 2060 with low SLR or $11.8 B with extreme SLR. These estimates of lost wages are considerably lower than Fant et al (2021)’s estimates of economic damages of $1.3–$1.5 B in 2020 and $28–$37 B in 2050 likely due to our inclusion of alternative routing along ‘drier’ routes. Second, impacts are not limited to residents immediately adjacent to a coastline but extend to commuters elsewhere (see supplementary figure 1 for an interactive map). Repeated obstruction of transportation routes can impede the flow of goods, affecting business operating costs, employment opportunities, and the cost of living in the long term. Limiting research to flooding impacts of storm surges or inundation on property values underestimates potential SLR impacts in flooding-adjacent areas. Calls to uncover more specific SLR impacts beyond just flood risk (Hauer et al 2020) make clear the importance analyzing SLR impacts as integrated natural and social systems. For example, recent findings concerning reduced property values in repetitive flood loss areas (McAlpine and Porter 2018), soil salinization on agricultural livelihoods (Chen and Mueller 2018), and climate gentrification responses to SLR flooding (Keenan et al 2018) are some such mechanisms integrating natural and social systems together. In this article, we investigate another specific mechanism in the form of recurrent tidal flooding on commuting delays. In the future, scientists should consider more integrated natural-social systems when investigating SLR impacts, moving beyond just flood hazard and flood risk into the actual social systems in coastal communities. We make several simplifying assumptions for computational tractability. We assess road water depth with a bathtub model that assumes perfect hydrologic connectivity, potentially overestimating the inundation on roadways. Tidal flooding adversely impacts local drainage systems and since we do not model precipitation, we do not include the possibility of roadway inundation due to rainfall and tidal flooding, potentially underestimating the inundation on roadways (Gold et al 2023). Our model does not account for traffic congestion or public transportation, nor does it include the less than 1.5% of people who live and work within the same CBG, potentially underestimating the commuting delays. Finally, most tidal flooding occurs for several hours, possibly coinciding with morning, evening, or both commutes, depending on when and for how long the high tide lasts. By simplifying to round trip, it is possible we overstate our results. Data availability statement The data that support the findings of this study are openly available in the Harvard Dataverse at https://doi. org/10.7910/DVN/2UT4GM (Hauer 2023). ORCID iDs Mathew E Hauer  https://orcid.org/0000-0001-9390-5308 Valerie Mueller  https://orcid.org/0000-0003-1246-2141 Glenn Sheriff  https://orcid.org/0000-0001-9642-5529 References Barnard P L et al 2017 Extreme oceanographic forcing and coastal response due to the 2015–2016 El Ni˜no Nat. Commun. 8 14365 Brovelli M A, Minghini M, Molinari M and Mooney P 2017 Towards an automated comparison of openstreetmap with authoritative road datasets Trans. GIS 21 191–206 Bukvic A and Harrald J 2019 Rural versus urban perspective on coastal flooding: the insights from the U.S. mid-atlantic communities Clim. Risk Manag. 23 7–18 Chamberlain S 2020 rnoaa: NOAA weather data from R. R package version 1.0.0. (available at: https://CRAN.R-project.org/ package=rnoaa) Chang S W, Clement T P, Simpson M J and Lee K-K 2011 Does sea-level rise have an impact on saltwater intrusion? Adv. Water Resour. 34 1283–91 6 US Census Bureau’s American Community Survey tableID: S2001 median hourly wages of workers over age 16 with earnings in 2019. 9 Environ. Res.: Climate 2 (2023) 045004 M E Hauer et al Chen C and Liu L-M 1993 Forecasting time series with outliers J. Forecast. 12 13–35 Chen J and Mueller V 2018 Coastal climate change, soil salinity and human migration in Bangladesh Nat. Clim. Change 8 981–5 Church J et al 2013 Sea level change Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change (Cambridge University Press) pp 1137–216 Clark P U et al 2016 Consequences of twenty-first-century policy for multi-millennial climate and sea-level change Nat. Clim. Change 6 360–9 Dahl K, Fitzpatrick M and Spanger-Siegfried E 2017 Sea level rise drives increased tidal flooding frequency at tide gauges along the US east and gulf coasts: projections for 2030 and 2045 PLoS One 12 e0170949 de Lacalle J L 2019 Tsoutliers: detection of outliers in time series R package version 0.6-8 (available at: https://CRAN.R-project.org/ package=tsoutliers) El-Ashmawy K L 2016 Testing the positional accuracy of openstreetmap data for mapping applications Geodesy Cartogr. 42 25–30 Ezer T and Atkinson L P 2014 Accelerated flooding along the U.S. East Coast: on the impact of sea-level rise, tides, storms, the Gulf Stream, and the North Atlantic Oscillations Earth’s Future 2 362–82 Fant C, Jacobs J M, Chinowsky P, Sweet W, Weiss N, Sias J E, Martinich J and Neumann J E 2021 Mere nuisance or growing threat? The physical and economic impact of high tide flooding on us road networks J. Infrastruct. Syst. 27 04021044 Fu X, Gomaa M, Deng Y and Peng Z-R 2017 Adaptation planning for sea level rise: a study of US coastal cities J. Environ. Plan. Manag. 60 249–65 Gesch D, Oimoen M, Greenlee S, Nelson C, Steuck M and Tyler D 2002 The national elevation dataset Photogramm. Eng. Remote Sens. 68 5–32 Gold A, Anarde K, Grimley L, Neve R, Srebnik E R, Thelen T, Whipple A and Hino M 2023 Data from the drain: a sensor framework that captures multiple drivers of chronic coastal floods Water Resour. Res. 59 e2022WR032392 Goodman A C, Thorne K M, Buffington K J, Freeman C M and Janousek C N 2018 El Ni˜no increases high-tide flooding in tidal wetlands along the US Pacific Coast J. Geophys. Res. Biogeosci. 123 3162–77 Gupta P D 1991 Decomposition of the difference between two rates and its consistency when more than two populations are involved Math. Popul. Stud. 3 105–25 Hague B S and Taylor A J 2021 Tide-only inundation: a metric to quantify the contribution of tides to coastal inundation under sea-level rise Nat. Hazards 107 675–95 Hallegatte S, Green C, Nicholls R J and Corfee-Morlot J 2013 Future flood losses in major coastal cities Nat. Clim. Change 3 802–6 Hauer M E, Evans J M and Mishra D R 2016 Millions projected to be at risk from sea-level rise in the continental United States Nat. Clim. Change 6 691–5 Hauer M E, Fussell E, Mueller V, Burkett M, Call M, Abel K, McLeman R and Wrathall D 2020 Sea-level rise and human migration Nat. Rev. Earth Environ. 1 28–39 Hauer M 2023 Sea level rise already delays coastal commuters Harvard Dataverse V1 (https://doi.org/10.7910/DVN/2UT4GM) Hauer M, Mueller V, Sheriff G and Zhong Q 2021 More than a nuisance: measuring how sea level rise delays commuters in Miami, FL Environ. Res. Lett. 16 064041 Heberger M, Cooley H, Herrera P, Gleick P H and Moore E 2011 Potential impacts of increased coastal flooding in California due to sea-level rise Clim. Change 109 229–49 Hinkel J, Lincke D, Vafeidis A T, Perrette M, Nicholls R J, Tol R S, Marzeion B, Fettweis X, Ionescu C and Levermann A 2014 Coastal flood damage and adaptation costs under 21st century sea-level rise Proc. Natl Acad. Sci. 111 3292–7 Hinkel J, Nicholls R J, Tol R S, Wang Z B, Hamilton J M, Boot G, Vafeidis A T, McFadden L, Ganopolski A and Klein R J 2013 A global analysis of erosion of sandy beaches and sea-level rise: an application of diva Glob. Planet. Change 111 150–8 Jacobs J M, Cattaneo L R, Sweet W and Mansfield T 2018 Recent and future outlooks for nuisance flooding impacts on roadways on the U.S. East Coast Transp. Res. Rec. 2672 1–10 Jamero M L, Onuki M, Esteban M, Billones-Sensano X K, Tan N, Nellas A, Takagi H, Thao N D and Valenzuela V P 2017 Small-island communities in the Philippines prefer local measures to relocation in response to sea-level rise Nat. Clim. Change 7 581–6 Kasmalkar I G, Serafin K A, Miao Y, Bick I A, Ortolano L, Ouyang D and Suckale J 2020 When floods hit the road: resilience to flood-related traffic disruption in the San Francisco bay area and beyond Sci. Adv. 6 eaba2423 Kasmalkar I G, Serafin K A and Suckale J 2021 Integrating urban traffic models with coastal flood maps to quantify the resilience of traffic systems to episodic coastal flooding MethodsX 8 101483 Keenan J M, Hill T and Gumber A 2018 Climate gentrification: from theory to empiricism in Miami-Dade County, Florida Environ. Res. Lett. 13 054001 Kirezci E, Young I R, Ranasinghe R, Muis S, Nicholls R J, Lincke D and Hinkel J 2020 Projections of global-scale extreme sea levels and resulting episodic coastal flooding over the 21st century Sci. Rep. 10 11629 Kulp S A and Strauss B H 2019 New elevation data triple estimates of global vulnerability to sea-level rise and coastal flooding Nat. Commun. 10 4844 Kwadijk J C et al 2010 Using adaptation tipping points to prepare for climate change and sea level rise: a case study in the Netherlands WIREs Clim. Change 1 729–40 McAlpine S and Porter J 2018 Estimating recent local impacts of sea-level rise on current real-estate losses: a housing market case study in Miami-Dade, Florida Popul. Res. Policy Rev. 37 871–95 Moftakhari H, AghaKouchak A, Sanders B and Matthew R 2017 Cumulative hazard: the case of nuisance flooding Earth’s Future 5 214–23 NOAA CO-OPs 2020 Tides and currents (available at: tidesandcurrents.noaa.gov) (Accessed 9 November 2020) NOAA 2020 Sea level rise viewer tools high tide flooding (available at: https://coast.noaa.gov/digitalcoast/tools/slr.html) (Accessed 13 November 2020) OpenStreetMap Contributors 2017 Planet dump (available at: www.openstreetmap.org) (Accessed 17 November 2020) Oppenheimer M et al 2019 Sea level rise and implications for low lying islands, coasts and communities ed H.-O. Pörtner et al Padgham M 2019 dodgr: an r package for network flow aggregation Transp. Findings (https://doi.org/10.32866/6945) Praharaj S, Chen T, Zahura F, Behl M and Goodall J 2021 Estimating impacts of recurring flooding on roadway networks: a Norfolk, Virginia case study Nat. Hazards 107 2363–87 Pregnolato M, Ford A, Glenis V, Wilkinson S and Dawson R 2017 Impact of climate change on disruption to urban transport networks from pluvial flooding J. Infrastruct. Syst. 23 04017015 Shen S and Kim K 2020 Assessment of transportation system vulnerabilities to tidal flooding in Honolulu, Hawaii Transp. Res. Rec. 2674 207–19 10 Environ. Res.: Climate 2 (2023) 045004 M E Hauer et al Storlazzi C D et al 2018 Most atolls will be uninhabitable by the mid-21st century because of sea-level rise exacerbating wave-driven flooding Sci. Adv. 4 eaa9741 Strauss B H, Kulp S and Levermann A 2015 Carbon choices determine us cities committed to futures below sea level Proc. Natl Acad. Sci. 112 13508–13 Sweet W V, Dusek G, Obeysekera J and Marra J J 2018 Patterns and projections of high tide flooding along the U.S. coastline using a common impact threshold Technical Report (NOAA) Sweet W V, Kopp R E, Weaver C P, Obeysekera J, Horton R M, Thieler E R and Zervas C 2017 Global and regional sea level rise scenarios for the United States Technical Report (NOAA) Taherkhani M, Vitousek S, Barnard P L, Frazer N, Anderson T R and Fletcher C H 2020 Sea-level rise exponentially increases coastal flood frequency Sci. Rep. 10 1–17 U.S. Census Bureau 2020 LEHD origin-destination employment statistics (2002–2017), LODES 7.4 (Longitudinal-Employer Household Dynamics Program) (available at: onthemap.ces.census.gov) (Accessed 7 October 2020) US Army Corps of Engineers 2020 National levee database (available at: levees.sec.usace.army.mil) (Accessed 20 October 2020) 11
null
10.1371_journal.pbio.3000080.pdf
null
The data for individual figures are available as Excel files (labeled, e.g., S1 Data) with links in the relevant figure captions. The full listing of these data files can be found following the captions for supplementary figures, as well as in the Excel file DataFileListings.xlsx. In addition, the full data are available. The corresponding author (Aniruddha Das) will maintain the data at Columbia University until all datasets will be shared openly with qualified scientists. Access will be granted by request to the corresponding
RESEARCH ARTICLE Task-related hemodynamic responses are modulated by reward and task engagement Mariana M. B. Cardoso1,2, Bruss LimaID 1,3, Yevgeniy B. Sirotin1,4, Aniruddha DasID 1,5* a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Cardoso MMB, Lima B, Sirotin YB, Das A (2019) Task-related hemodynamic responses are modulated by reward and task engagement. PLoS Biol 17(4): e3000080. https://doi.org/10.1371/ journal.pbio.3000080 Academic Editor: Frank Tong, Vanderbilt University, UNITED STATES Received: October 23, 2018 Accepted: March 29, 2019 Published: April 19, 2019 Copyright: © 2019 Cardoso et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The data for individual figures are available as Excel files (labeled, e.g., S1 Data) with links in the relevant figure captions. The full listing of these data files can be found following the captions for supplementary figures, as well as in the Excel file DataFileListings.xlsx. In addition, the full data are available. The corresponding author (Aniruddha Das) will maintain the data at Columbia University until publication. Once published, all datasets will be shared openly with qualified scientists. Access will be granted by request to the corresponding author. It will be our intent to collaborate with 1 Department of Neuroscience, Columbia University, New York, New York, United States of America, 2 Center for Neural Science, New York University, New York, New York, United States of America, 3 Institute of Biophysics Carlos Chagas Filho, Federal University of Rio de Janeiro, Rio de Janeiro, Brazil, 4 Identity and Data Science Laboratory of Science Applications International Corporation, Annapolis Junction, Maryland, United States of America, 5 Zuckerman Mind Brain and Behavior Institute, Columbia University, New York, New York, United States of America * Aniruddha.Das@columbia.edu Abstract Hemodynamic recordings from visual cortex contain powerful endogenous task-related responses that may reflect task-related arousal, or “task engagement” distinct from atten- tion. We tested this hypothesis with hemodynamic measurements (intrinsic-signal optical imaging) from monkey primary visual cortex (V1) while the animals’ engagement in a peri- odic fixation task over several hours was varied through reward size and as animals took breaks. With higher rewards, animals appeared more task-engaged; task-related responses were more temporally precise at the task period (approximately 10–20 seconds) and mod- estly stronger. The 2–5 minute blocks of high-reward trials led to ramp-like decreases in mean local blood volume; these reversed with ramp-like increases during low reward. The blood volume increased even more sharply when the animal shut his eyes and disengaged completely from the task (5–10 minutes). We propose a mechanism that controls vascular tone, likely along with local neural responses in a manner that reflects task engagement over the full range of timescales tested. Introduction The use of functional magnetic resonance imaging (fMRI) in humans, complemented with electrode measurements from animal studies, has considerably advanced our understanding of cortical visual processing. This combination of tools has been particularly useful in under- standing exogenous, stimulus-evoked responses. Models of neural responses in humans based on electrophysiological recordings in animals, combined with linear models linking neural to hemodynamic responses, have been effective in accounting for stimulus-evoked fMRI mea- surements in human subjects and in quantitatively predicting the corresponding sensory per- cepts [1–9]. However, fMRI measurements from subjects performing visual tasks also contain large endogenous hemodynamic responses in the absence of or independent of visual stimuli, even at the earliest stages of visual processing [10–15]. There are at least two types of endogenous PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 1 / 34 individuals who contact us about sharing the data. But if that is not possible or does not make sense, then we will simply provide the data to them. Because of the large size of the optical imaging and electrode recording data, it might not be possible to keep it online. Hence, it will also be distributed (upon request) as a DVD box set for a nominal fee (sufficient to cover the costs of the DVDs, the time for a lab technician to burn the discs, and shipping expenses). Funding: National Institutes of Health (NIH) Grants R01EY025330, R01EY025673, R01 EY019500, and R01 NS063226 were given to AD, and a National Research Service Award was given to YBS, as well as grants from the Columbia Research Initiatives in Science and Engineering, the Gatsby Initiative in Brain Circuitry, and The Dana Foundation Program in Brain and Immuno Imaging and the Kavli Institute for Brain Science (to AD). BL received a fellowship from the The Italian Academy for Advanced Studies in America, Columbia University. MMBC was supported by Fundac¸ão para a Ciência e a Tecnologia (FCT), scholarship SFRH/BD/33276/2007. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Abbreviations: BF-ACh, basal forebrain- cholinergic; BOLD, blood oxygen level–dependent; CBV, cerebral blood volume; EEG, electroencephalographic; fMRI, functional magnetic resonance imaging; HR, heart rate; HRF, hemodynamic response function; IACUC, Institutional Animal Care and Use Committees; ISOI, intrinsic-signal optical imaging; LC-NA, locus coeruleus-adrenergic; LFP, local field potential; NIH, National Institutes of Health; NS, not significant. Modulating task-related hemodynamics response, “attention-like” and “task related” [16]. Unlike the case with exogenous responses, there has been mixed success in interpreting these endogenous hemodynamic responses. Selective visual attention has been characterized extensively through studies in human fMRI [10–15] with close parallels seen in animal electrophysiology [17–25]. Although likely driven by a unified mechanism [26], attention can take different forms. It could be selective for stimu- lus location [10,11,27–29], features (e.g., color versus motion [30]), or timing [28]. The related hemodynamic responses reflect corresponding attributes of the expected stimuli. Attentional responses also increase in strength along the visual cortical hierarchy [11,20]. Much less is known about the task-related endogenous hemodynamic response, including whether it comprises one or multiple types. It appears to be distinct from selective attention. It entrains to task structure and extends over large sections of cortical areas (e.g., primary visual cortex—i.e., V1) independent of the stimulus [16,31–33], where it can even be substantially stronger than stimulus-selective responses [34]. It is also strongest in V1 and progressively weaker in higher visual areas [16]. These differences may reflect distinct brain processes underlying these two endogenous responses. There is growing evidence of the importance of the task-related endogenous response. It may play a role in sensory processing, in temporally grouping otherwise unrelated sensory stimuli [33] or in switching between stimulus modalities [35]. As yet, relatively little is understood about the mechanism of the task-related response even though its presence has been known for over a decade [16,33,35–41]. This is largely due to the paucity of studies comparing hemodynamics with electrophysiology in behaving subjects. The current work derives from a task-related hemodynamic response measured using intrinsic-signal optical imaging (ISOI) [42,43] in V1 of behaving macaques performing cued visual tasks [31]. The observed task-related response entrained to task timing independent of visual stimulation, with amplitudes that could compare with or even exceed vigorous visually evoked responses [44]. It appeared to be spatially nonselective, being homogeneous over the optical imaging window and presumably extending beyond [32]. It is thus likely a good model for investigating the mechanism underlying the task-related response seen in humans. Con- current electrode recordings showed it to be poorly predicted by changes in local firing rates or local field potential (LFP) power at any frequency band [31], unlike stimulus-evoked hemo- dynamic responses that were well predicted by local electrophysiology [44,45]. Additionally, at a vascular level, this response corresponded to a coordinated contraction–dilation cycle engag- ing the arterial blood supply into the imaged cortical region [31]. These observations suggested an underlying mechanism distinct from exogenous, stimulus-evoked responses. Here, we explore the link between this task-related hemodynamic response and the level of engagement in a task. The link was suggested by earlier measurements showing correlations between the measured task-related response and task performance [32], as well as with sympa- thetic-like markers of mental effort in a task [46] such as phasic pupil dilation [31] and heart rate (HR) fluctuations [31]. To modulate the level of engagement, we changed reward size sys- tematically [47] while the monkeys performed a periodic visual fixation task over several hours. Using ISOI and electrophysiology, we looked for effects on the measured task-related hemodynamic response at multiple timescales: of individual trials (approximately 10–20 sec- onds), of blocks of trials (150–300 seconds), and finally, of extended segments of task engage- ment versus disengagement as the animal switched between working and resting with eyes closed (many minutes). Based on our results, we propose that the task-related hemodynamic response reflects mechanisms that entrain brain processing more sharply to a task during peri- ods of higher task engagement, possibly as a means of temporally filtering or binding compo- nents of a task. Although we use the term “task engagement” as a shorthand for the set of behavioral and hemodynamic responses described here, in the Discussion, we consider PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 2 / 34 Modulating task-related hemodynamics possible links with states of task-specific arousal that have variously been labeled “sustained attention,” “vigilance,” or “alertness” [48–50]. Additionally, we propose an overarching mech- anism that controls vascular tone over multiple timescales in coordination with ongoing changes in the level of engagement during a task. Understanding these links would be an important step forward in understanding the dynamic allocation of brain resources in the con- text of a task. Results Overview Two male rhesus macaques performed a cued, periodic visual fixation task, receiving a juice reward following every correct fixation with no time-out or other punishment for errors (see Methods). The task is known to evoke a robust task-related hemodynamic response in the monkeys’ V1 independently of visual stimulation [31,44]. Here, we systematically manipulated the size of the reward per correct trial as a means of modulating the animals’ level of engage- ment in the task. This was done either in alternating blocks of high and low reward or in sequences of progressively changing reward (see Methods). We recorded V1 hemodynamics using ISOI [42], a high-resolution optical analog of fMRI [51–53]. This technique deduces brain hemodynamic responses at the exposed cortical surface by measuring changes in reflected light intensity at wavelengths absorbed by hemoglobin. Here, we used a wavelength specific to total hemoglobin, which provides a measurement analogous to cerebral blood vol- ume (CBV) [54] (see Methods). Imaging was combined with concurrent extracellular elec- trode recording of multiunit spiking and LFP. All experimental procedures were performed in accordance with the National Institutes of Health (NIH) Guide for the Care and Use of Labo- ratory Animals and were approved by the Institutional Animal Care and Use Committees (IACUC) of Columbia University and the New York State Psychiatric Institute. We observed distinct effects on the task-related V1 hemodynamic responses at the three different timescales tested. At the shortest timescale (individual trials—a few seconds), higher reward led to crisper temporal alignment of the task-related response to each trial, accompa- nied by a significant, if modest, improvement in response amplitude. At a slower timescale of blocks of alternating high versus low reward (10 to 20 trials—i.e., 150 to 300 seconds per block), we observed consistent alternating ramp-like changes in the mean local cortical blood volume. The sign of the ramps was such as to decrease blood volume for blocks of high reward while increasing it for low. Finally, periods of disengagement from the task during which the animal shut his eyes and rested over many minutes led to further large, sustained increases in the mean local blood volume. None of these effects at any timescale could be accounted for by changes in local spike rate. The majority of the reported results came from tasks performed in essentially complete darkness (“dark-room fixation” N = 30 sites, 3 hemispheres, 2 animals). This allowed measure- ment of the effects on the endogenous task-related hemodynamic response while minimizing exogenous visual confounds [31]. A complementary section (N = 33 sites, 2 hemispheres, 2 animals) confirmed that the observed results generalized to the presence of visual stimuli. Timescale of single trials: Higher reward leads to greater temporal precision A section of recording made while the animal fixated periodically in the dark illustrates the pattern of task-entrained responses, as well as changes to these responses with reward size (Fig 1A). Despite the near-total absence of visual stimulation, the V1 hemodynamic recording PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 3 / 34 Modulating task-related hemodynamics Fig 1. High reward leads to greater engagement in the task. (A-C) Example data set, periodic fixation task in the dark. (A) Continuous records of hemodynamic response (“Hemo”), radial eye position (“Eye pos”), heart rate (“HR”), and pupil size (“Pupil”) while reward level alternated between high (“Hi”; 0.375 ml per correct trial) and low (“Lo”; 0.11 ml) in blocks of 10 correct trials (showing roughly 3 of 32 blocks total, with time indexed relative to the start of the experiment; red = high reward, cyan = low. Same color code is used all through the paper.). Trials with no color indicate incorrect fixation (compare “Eye pos”). Each continuous sequence of incorrect trials counts as one error (gray arrows). Monkeys made more frequent errors in low-reward blocks (0.29 for “Lo” versus 0.19 for “Hi” as fraction of correct trials [N = 330]). Hemodynamic response (dR R fractional change in light reflected off cortical surface; down indicates increasing light absorption (i.e., increasing local blood volume). (B) Comparing pupil dilation during the fix period, high- versus low-reward trials. “All Hi, Lo” compares all correct high-reward trials (N = 160) with low (N = 170). “Lo to Hi” and “Hi to Lo” compare the first trial after a change in reward size to the immediately preceding trial (N = 15 “Lo to Hi” transitions, 16 “Hi to Lo” [data in S3 Data]). Gray shaded rectangle indicates a period of steady fixation starting 1 second after fix onset, which is used for quantifying pupil dilation. Inset histograms show dilation difference (high minus low reward) for all experiments with reliable pupil recording (N = 9; x-axis labeling, shown only for the third histogram [“Pupil Hi to Lo”] to avoid clutter, is common to all [data in S2 Data]). Rewards were given at the end of each correct fixation (gray arrowheads below time line; the same reward timing was used in all experiments reported here [data in S1 Data]). (C) Comparing amplitude (defined as standard deviation) of mean trial-linked heart rate fluctuations, high versus low reward (0.038 s−1: high, 0.025 s−1: low [data in S4 Data]). Traces in (B, C) are shown as mean +/− SEM (lighter ribbon). (D) Scatterplot comparing errors as fraction of correct trials, high versus low reward, all experiments (N = 30 [data in S5 Data]). (E) Comparing amplitudes of mean HR fluctuation (standard deviation as in panel C), high versus low reward, all experiments (“expts”). Each data point in (D, E) corresponds to one recording site (data in S6 Data). In (D), error trials were counted as high or low reward based on the block in which they occurred (see Methods). In (E), values were averaged separately across all correct high-reward versus correct low-reward trials for the given recording site. p-Values in (D and E): Wilcoxon signed rank test. ) plots https://doi.org/10.1371/journal.pbio.3000080.g001 showed robust task-related fluctuations in local tissue blood volume [31]. These were accom- panied, as noted earlier [31], by sympathetic-like responses [46]—i.e., phasic pupil dilation and HR fluctuations—also entrained to the task period. These sympathetic-like responses increased with higher reward. The pupils dilated more per trial, switching dilation size across single trials at block transitions (Fig 1B). The mean HR fluctuations were stronger (Fig 1C and 1E). Furthermore, animals made fewer errors (fixations broken or never acquired) in high- PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 4 / 34 Modulating task-related hemodynamics reward blocks (Fig 1A and 1D). The mean hemodynamic response also appeared to ramp slowly upwards during the high-reward block—i.e., reducing mean local blood volume—as indicated by the slope of a linear regression line (red, Fig 1A); this observation is addressed in a later section on slow changes. We did note a weak fluctuation in recorded spiking that was periodic in the mean and appeared to relate to hemodynamics in some data sets. However, the correlation was unreliable and did not generalize (see S1 Fig), consistent with our earlier find- ings that the task-related response is not predicted by local spiking or LFP [31]. At the timescale of individual trials, the primary correlate of high reward on the task-related response appeared to be greater temporal precision—i.e., tighter alignment to trial timing. This was evident qualitatively in lower trial-to-trial temporal jitter for high reward (Fig 2A, left panel). The mean of these trial-by-trial responses, averaged across all correct trials, was also higher for high-reward trials. But it was unclear how much of that was due to a true difference in amplitude, as opposed to better temporal alignment of individual responses. To resolve this issue, it was necessary to separately estimate the timing and amplitude of the task-related response for each trial. We used a template-matching approach based on the observation that, other than temporal jitter, individual responses appeared similar to each other in shape independent of reward size (Fig 2A; also see [32]). The full hemodynamic recording was thus modeled implicitly as a sequence of task-related responses of stereotyped shape, one per trial, varying only in ampli- tude and timing from trial to trial. The template was defined to be the trial-triggered average response over all correct trials. This template was slid in a one-trial-long moving window over the recorded response, calculating the normalized local dot product at each time point (“Tem- plate Match” in Fig 2B, Methods, Eqs 1–3). The dot product is closely analogous to Pearson’s correlation (see Methods). We thus surmised that it would have maxima (peaks) at points of high correlation where the recorded hemodynamics locally matched the template in shape—in effect, defining locations of putative task-related responses. But in addition, unlike Pearson’s r, which is scale-invariant, dot products scale linearly with the amplitude of their arguments and thus provide a measure of response strength (Fig 2B and Methods). We therefore defined our estimates of task-related response time and amplitude per trial to be the location and height of the corresponding template match peak. After estimating response times and amplitudes as described above, we wanted to check our starting assumption that the measured hemodynamics are well modeled as a sequence of jittered but stereotyped shapes. If the assumption is valid, the segments of recorded hemody- namics centered on each peak of the template match should match each other closely in shape. To test, we centered each putative task-related response, as picked out through template matching, by its response time as estimated from the same template match (Fig 2C). Indeed, the realigned responses were strikingly well correlated with each other. This can be appreciated visually by normalizing realigned responses by their amplitudes to help compare shapes (Fig 2D) and quantitatively by correlating realigned responses to the template used for matching (Fig 2E and 2F). The strength of this correlation supports our approach. With the task-related response times and amplitudes thus quantified, we confirmed that the primary effect of higher reward was greater temporal precision. Response times were better aligned to the task period, with consistently tighter distributions (quantified by the 2 standard deviation width of the distribution). This was evident for the example data set (Fig 3A) as well as in essentially every other data set (Fig 3C). High reward also led to significantly higher response amplitude for the example data set (Fig 3B). However, that pattern was less consistent over the full set of experiments, with only a relatively modest improvement in median response amplitudes overall (Fig 3D). PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 5 / 34 Modulating task-related hemodynamics Fig 2. Estimating trial-by-trial timing and amplitude of task-related responses by template matching. (A) All correct trials in one data set, separated by reward size: high (N = 140 trials; left, red bar on top as in Fig 1A) and low (N = 148, right, cyan bar). Gray indicates individual trials, and black indicates the mean. The time axis is shared with (C), (D) (0 = trial onset; yellow indicates a fixation period [data in S7 Data]). (B) Elements of the template match. Black (“Hemo”) indicates a section of recorded hemodynamic response (z-scored, shifted down for visibility; time indexed from an arbitrary t = 0); vertical dashed lines indicate trial onsets. Green (“Template Match”) marks the sliding- window dot product of “Hemo” with “Tmplt” (inset: defined as mean hemodynamic response, all correct trials). The locations and heights of template match peaks (red dots) define estimated timing and amplitude of task-related response per trial. “Match Peak #1, #2” are examples illustrating the information carried by peaks. Both #1 and #2 mark locations where “Hemo” matches “Tmplt” in shape (see “Hemo” segments on green shading. Compare with “Match Trough,” gray shading, phase-reversed “Hemo”). Greater height of peak #1 versus #2 quantifies higher amplitude of “Hemo” fluctuation at #1. However, location of peak #2 is better centered in its trial. (C) Same traces as in (A), aligned by response times estimated from template match (data in S8 Data). (D) Same data as (C), normalized by amplitude (standard deviation; “SD-norm”). Orange indicates responses with standard deviation in the lowest 10th percentile over the full set. Gray marks the upper 90th percentile. Black indicates the mean of gray traces. The red dotted line marks the template. Gray traces match each other and the template well, particularly near the midpoint of the trial (data in S9 Data). (E) Histogram of correlations (“corr”) of aligned responses with the template (Pearson’s r; all correct trials, high and low reward, including responses in the lowest 10th percentile of standard deviation). (F) Histogram of correlation medians as in (E), all experiments (“expts”; data in S10 Data). https://doi.org/10.1371/journal.pbio.3000080.g002 We wondered if these results were due to our particular choice of template. We tested by repeating the analysis shown here across all data sets using a range of alternate templates. The alternate templates were also each one trial long and constructed from measured responses but using different criteria: for example, being phase shifted in time or using only “high signal-to- noise” responses with amplitudes exceeding a threshold. The task-related response times and amplitudes estimated by matching to these alternate templates showed a strikingly similar overall relationship to reward size as in Fig 3. This is illustrated in S2 Fig for a particular alter- nate template with timing and shape distinct from the one used in the main text. This result highlights the overall robustness of our findings. It also suggests that high reward leads to a state of greater temporal regularity and periodicity overall for the duration of the block, accounting for the higher temporal precision in estimated response times independent of the details of the template used. PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 6 / 34 Modulating task-related hemodynamics Fig 3. Higher reward leads to more temporally precise task-related responses. (A) Distributions of response (“Resp.”) times per trial, estimated as positions of the corresponding template match peaks (“Match Peak pos.”), same data set as in Fig 2A–2E. Distributions are separated by reward size, with color coding as indicated in the key (common to (A, B) and to all later figures). Data are shown as a vertical “violin plot” histogram with numbers of trials increasing from 0 (middle) upwards for high reward (“Hi”) and downwards for low (“Lo”). Similar displays are used for all such comparisons of distributions to avoid clutter (e.g., with interleaved histograms). Clustering of response times per reward size was quantified as the 2 standard deviation (“2x StdDev”) width of timing distributions. (B). Distributions of response amplitudes per trial, estimated from template matching (“Tmplt Match”), shown separated by reward size following the same conventions as in (A). Response amplitude per reward size was quantified as the median (indicated by arrowheads; medians are indicated similarly in all later amplitude distributions). Significance (p- values) in (A,B) were obtained from bootstrap with 10,000 resamples. (The 2x StdDev values per reward size [“hi”,”lo” in panel A] and median amplitude per reward size in panel B shown in these and other panels are not the sample medians from the measured distributions but rather are medians obtained from the same bootstrap procedure used to get p-values.) (C) Comparing the 2 standard deviation width of the response time distribution for all correct high- reward trials versus that for all correct low-reward trials, per experiment (“expt”; N = 30). (D) Comparing median response amplitudes for all correct high- versus all correct low-reward trials, per experiment. p-Values in (C), (D): Wilcoxon signed rank tests for pairwise comparisons (data in S11 Data). https://doi.org/10.1371/journal.pbio.3000080.g003 Timing precision is robust to noise in template match We were concerned that the apparently lower temporal precision with low reward could be an artifact of a noisier template match. When task-related responses had lower amplitudes, the template match could be poorer simply because of lower signal to noise. This could lead to noisier estimates of response time with wider distribution and thus apparently poorer preci- sion but, because of the poorer signal to noise alone, independent of reward size (Fig 4A; also consider, e.g., the responses with poor shape match in Fig 2D). Since lower rewards were asso- ciated with somewhat lower response amplitudes, this increased noise could make the low- reward responses appear artifactually less precise. PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 7 / 34 Modulating task-related hemodynamics Fig 4. Temporal precision is not an artifact of higher signal for high-reward responses. (A) An outline of the null hypothesis using a data set in which low-reward responses had substantially lower amplitudes. Left panel: Scatterplot of response amplitude versus time per trial, colored by reward size. Gray and white shading indicates quintiles along the amplitude axis, combining high (“hi”) and low (“lo”) rewards. Right panel: 2 standard deviation width of response time distribution in each quintile. The time axis is scaled to match that for the left panel. These 2 standard deviation widths (“2x StdDev”) increased progressively for lower response amplitudes, which were also more dominated by low-reward trials. The null hypothesis is that this covariance alone gives low-reward trials larger timing scatter. Arrows mark the y-axis locations indicating median amplitudes of quintiles (data in S12 Data). (B) Plot of response amplitude versus timing for a large data set (1,285 correct trials; 629 low reward; 626 high reward) separated into quartiles by response amplitude (gray/white shading). Each quartile also roughly matched for numbers of low- versus high-reward trials (data in S13 Data). (C1, C2, D1, D2, E1, E2, F1, F2) Pairs of distributions of response amplitude and timing per quartile separated by reward size. The numbers “N” in parentheses in (C1–F1) indicate numbers of high- and low-reward trials. The high-reward responses are significantly more precise than the corresponding low-reward ones in each quartile despite the similarity in response amplitudes (C2–F2). All distributions are shown as “violin plots” using the same conventions as Fig 3A and 3B. Panels (C1–F1) share a common abscissa scale, as do panels (C2–F2). The “NTrials” label for the ordinate is shown only for (F1) to avoid clutter. p-Values, bootstrap, 10,000 resamples (data in S14 Data). https://doi.org/10.1371/journal.pbio.3000080.g004 To test, we selected subsets from each experiment in which the high- and low-reward responses were matched in amplitude and in numbers of data points (Fig 4B, 4C1, 4D1, 4E1 and 4F1). If our concern about signal to noise in the template match were valid, high- and low-reward responses in each amplitude-matched subset should exhibit similar distributions of response times independent of reward size. Instead, even after matching for amplitudes, the high-reward responses remained consistently and significantly more temporally precise (see, particularly, Fig 4D1, 4D2, 4E1 and 4E2). Timing precision is independent of eye fixation timing and eye movements We wondered if there were some simple oculomotor explanation or correlate of our observa- tion. We considered two possible scenarios under which this could happen. PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 8 / 34 Modulating task-related hemodynamics We considered the null hypothesis that the timing of the task-related response per trial is determined by fix onset, with a stereotyped response time course and hence constant delay fol- lowing fixation (see S3 Fig). If that were the case, response times should be correlated to fix onset times, with unity slope and constant delay. The higher precision with high reward could reflect a behavioral pattern in which the animal is more precise in its fix onsets prior to those trials (S3 Fig, panel a). This null hypothesis turned out not to be the case, and task-related response times were uncorrelated with fixation onset. Parenthetically, both animals’ fixation behavior changed over the many months that we tested them intermittently on this task. Ini- tially, both animals tended to maintain fixation for long periods with very few breaks even dur- ing intertrial intervals. This led to extended periods of fixation prior to the start of each trial, or even across multiple trials, without any breaks but unrelated to the timing of the task-related response or reward size (S3 Fig, panel b). Later, both animals showed a different behavioral pattern, moving their eyes around during intertrial intervals and reacquiring fixation shortly before trial onset. This led to a pattern of brief fixation periods prior to each trial (S3 Fig, panel c). Task-related hemodynamic response times remained more precise with high reward, inde- pendent of the changing pattern of fixation. We next considered the possibility that animals may have steadier fixation or smaller eye movements during high-reward blocks, due to generally higher engagement in the task (S4 Fig). We failed to see any consistent patterns. There were no consistent differences in fixational jitter between high- and low-reward trials at the resolution of our measurements (60 Hz, 0.33 deg). There were also no consistent differences in eye movements during the intertrial periods during which the animals were free to look around. The animals also changed their patterns of intertrial eye movements over the many months of recording. In earlier sessions, they did move their eyes less during high-reward blocks (S4 Fig, panels a1-a3). Later, however, the ani- mals adopted a behavioral pattern of greater intertrial eye excursions for high-reward trials (S4 Fig, panels b1-b3). However, the task-related responses remained more precise for higher- reward trials (smaller 2 standard deviation width for task-related response time distributions), independent of this changing pattern of eye movements. Timing precision generalizes to the presence of visual stimulation The question that remained was whether reward size affected task-related responses only in the unnatural circumstance of visual tasks in the near absence of all visual stimulation or whether such effects generalized to the presence of visual stimuli. To test, we analyzed data from a separate set of experiments in which the animals were passively shown visual stimuli— gratings of different contrasts—while performing the same cued, periodic fixation task (Fig 5). Rewards were comparable to the dark room, if slightly higher (see Methods), ranging typically from 0.2 ml/trial (low) to 0.6 ml/trial (high). For this, we first needed to estimate the task- related response from recorded hemodynamics by estimating and removing stimulus-evoked responses. We did so by modeling the overall measured hemodynamics as a linear sum of the stimulus-evoked and task-related components, which we fitted to get the optimal kernels for the two components[55] (Methods, Eqs 4–6; also, S5 Fig). The optimal hemodynamic response function (HRF) kernel thus obtained was then convolved with the recorded spiking to estimate the stimulus-evoked component of hemodynamics and regress it away from the full hemody- namics. The residual—that was, by construction, the component of hemodynamics not pre- dicted by local spiking—was then defined to be the task-related component of the hemodynamic response, equivalent to the full hemodynamic response in the dark room. The task-related response thus estimated in the presence of visual stimuli was again tempo- rally more precise with high reward, just as with the task undertaken in the dark room. This PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 9 / 34 Modulating task-related hemodynamics Fig 5. Task-related responses in the presence of visual stimuli are temporally more precise and modestly higher in amplitude for high reward as in the dark room. (A-D) One example data set: (A) Residual task-related responses (“Task Rel. Resp.”) separated by trial and by reward size were obtained by regressing away stimulus-evoked responses (see S5 Fig) (data in S15 Data). (B) Comparing pupil dilation for high- (“Hi Reward”) versus low-reward (“Lo Reward”) trials. Gray shading indicates the period over which pupil dilations are compared, starting 1 second after fixation. (These pupil measurements were made in the presence of visual stimulation, unlike dark-room results [Fig 1B], likely accounting for different shape of trace including initial constriction on fixation) (data in S16 Data). (C) Distribution of response times from template match in this data set. (D) Distribution of response amplitudes in this data set. Conventions for “violin plot” histograms are used as in Fig 3A and 3B (data in S19 Data). (E) Comparing the 2 standard deviation widths of response times (“Resp time 2x StdDev”) for high versus low reward, per experiment (“expt”; N = 33) (data in S17 Data). (F) Comparing median response amplitudes (“Resp Amp Median”) for high versus low rewards, per experiment (N = 33). p-Values in (E), (F): Wilcoxon signed rank tests for the pairwise comparisons. The inset in panel F indicates overall behavioral performance as the total numbers of error trials as a fraction of the correct trials, per experiment (data in S18 Data). Eye pos, eye position; Fix on, fixation on; norm., normalized; NS, not significant; Stim on, stimulus on. https://doi.org/10.1371/journal.pbio.3000080.g005 can be seen qualitatively after separating the estimated task-related response into individual trials and segregating trials by reward size. These trial-wise responses were visibly less tempo- rally jittered for high reward (Fig 5A). The timing and amplitude of these task-related PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 10 / 34 Modulating task-related hemodynamics responses were quantified by matching to a template just as for recordings in the dark room; the template was taken to be the optimal mean kernel for the task-related component as esti- mated from the fit (Methods, Eqs 7 and 8. S5 Fig). The results of this template match closely paralleled those obtained in the dark room fixation task. The estimated response times were again more tightly clustered for high-reward trials, both for this specific data set (Fig 5A and 5C) and over the set of visually stimulated experiments (Fig 5E). Response amplitudes showed only a modest improvement (Fig 5D and 5F). High-reward trials were also associated with greater pupil dilation (Fig 5B). Timescale of blocks of trials: Mean blood volume decreases for high reward and increases for low Analyses up to this point were restricted to the scale of single trials—i.e., about 10 to 20 sec- onds. However, we also noted slow ramp-like drifts in the mean local blood volume over blocks of 10 to 20 trials of a given reward size—i.e., about 150 to 300 seconds (Fig 1A; Fig 6). The ramps decreased blood volume for high-reward blocks while increasing it for low. Regres- sion lines fitted through sequences of correct trials per block clustered into distinct sets of neg- ative slopes (increasing absorption of light during imaging—i.e., increasing blood volume) for low-reward blocks and positive slopes (decreasing blood volume) for high (Fig 6B and 6C). These slow hemodynamic drifts were not driven by slow changes in spiking (see S7 Fig). Blocks of trials with alternating ramps of mean blood volume failed to show similar alternating ramps of mean spiking (S7 Fig, Panels a-d). To test more quantitatively, we first simulated the spiking patterns required to generate the measured hemodynamic slopes on convolving with the corresponding optimal fitted HRF per experiment (S7 Fig, Panels e, f). The slopes of the simulated spiking ramps alternated in sign with reward size, as expected. Each measured spik- ing slope was then divided by the slope of its corresponding simulation to compare. If the mea- sured slopes had the same sign as their simulations, these ratios would consist of positive numbers, with some magnitude reflecting a scale factor. This was not the case; the ratios were equally likely to be positive or negative for both high and low reward. The measured spiking slopes were thus uncorrelated with those required to generate the measured hemodynamic slopes. Switching from task engagement to rest with eyes closed: Further profound increases in blood volume We wondered if the slow increase in mean local blood volume accompanying reduced reward could be part of a broader pattern of shifts in mean local blood volume accompanying shifts in the level of engagement. A potential clue was seen in the continuous measurements during long dark-room recording sessions lasting up to 3 hours (Fig 7). In these sessions, in between extended stretches of working well, the animals would take occasional breaks of many minutes during which they stopped working and rested with their eyes shut. The mean local blood vol- ume in V1 increased strikingly during these breaks, returning to baseline when the animal resumed working (Fig 7A). This pattern appeared to be an extreme manifestation of the ramp- like changes in blood volume with reward size in which lower reward, with its lower level of engagement (Fig 1), led to increasing mean blood volume (Fig 6). Before ascribing an association with reduced engagement in the task, we needed to test whether the increased blood volume could be accounted for simply by concurrent changes in neural or physiological drivers. As possible drivers, we considered the mean HR and the mean local multiunit spike rate (recorded separately at two electrodes spaced 4 mm apart in the recording chamber). We also measured the pairwise noise correlation of spike rates between PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 11 / 34 Modulating task-related hemodynamics Fig 6. Mean local cortical blood volume increases for low-reward blocks and decreases for high, in alternating ramp-like drifts. (A) Recordings from a sequence of correct trials, alternating between high (“Hi Reward”) and low reward (“Lo Reward”) in blocks of 10. Lines indicate regression fits to each block separately. Increasing slope implies decreasing local tissue blood volume. Correct trials were concatenated after excising incorrect ones while maintaining vertical position (see S6 Fig). This panel shows 20 blocks of 102 total: 1,160 trials total, 1,019 correct. (B) Histogram of regression slopes in high- versus low-reward blocks. Same data set as (A) (p-values, bootstrap, 10,000 resamples) (data in S21 Data). (C) Comparing median slopes of high-reward versus low-reward blocks over the set of all experiments (“expt.”). All the statistically significant data points lie in the upper left quadrant, “Lo(-)/Hi(+)”—i.e., with negative slopes for low- and positive slope for high-reward blocks (N = 19 experiments: using only those with at least 10 pairs of alternating blocks of 10 trials each). The results shown here were based on correct trials alone. Analyses that utilized all trials including incorrect ones gave results that were broadly similar but were sometimes harder to interpret because of the arbitrary duration of sequences of incorrect trials (S6 Fig) (data in S20 Data). Hemo, hemodynamic response; NS, not significant. https://doi.org/10.1371/journal.pbio.3000080.g006 the two electrodes over a 1-second sliding window, since that was expected to increase at rest [56]. To focus on slow changes, all recordings were downsampled to get, in effect, the smoothed average in a 60-second window (see Methods). We then assessed changes in these physiological and neural measurements as the animal switched state, marking the state based on the fraction of time within the 60-second window that the eyes were closed (Fig 7A, top row, “Eyes closed”). Spontaneous eye closures in the dark have been shown to provide a useful measure of drops in “vigilance” [49], correlating well with electroencephalographic (EEG) and fMRI indicators [57]. For this study, epochs during which the eyes were shut more than 60% of the time were defined as “rest,” whereas those with less than 5% of time with eye closure were considered “engaged.” To check, we compared with an LFP measure of arousal based on the ratio of power in the beta- and theta-range frequency bands (15–25 Hz and 3–7 Hz, respectively) as suggested by earlier studies ([57]; also reviewed in [48]). To get a measure that was low when the animal was engaged in his task and high when at rest [48], as with eye closure, we placed the theta power in the numerator. The square root of this ratio further compressed the dynamic range to roughly 0–1, as with eye closure. These two measures based on eye closure and on the LFP were closely comparable (Fig 7A, upper two rows and inset), supporting the use of eye closure to segregate physiological mea- surements by the state of task engagement. PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 12 / 34 Modulating task-related hemodynamics Fig 7. Mean local blood volume, spiking, and heart rate trace switches between states of task engagement and rest. (A) Traces show continuous 2.5-hour records of measured variables as indicated by adjoining labels, smoothed and downsampled to a 60-second sampling rate to track slow changes. (See text and Methods for more details.) Red and green bars on top mark epochs of rest (defined as “Eyes closed” > 0.6—i.e., >60% of the 60-second sample time) and task engagement (sections with eyes open, defined as “Eyes closed” < 0.05—i.e., <5% of the time). “Eyes closed” is highly correlated with the LFP measure (see text for definition. Pearson’s r = 0.94 for the example data set. Inset shows histogram of Pearson’s r for similar pairwise correlations over all data sets used for this analysis, N = 11). Performance in the task is quantified as the (smoothed) fraction of trials initiated in the 60-second (i.e., approximately 4-trial) window. “Hemo” marks the mean hemodynamic response (dR/R); “Spike1” and “Spike2” mark multiunit responses recorded from two electrodes spaced 4 mm apart in the imaged region. “Spk1-Spk2 Corr” marks the pairwise correlation between these two recordings over a 1-second moving window. “HR” marks mean heart rate. The red box indicates the section of hemodynamic and corresponding Spike1 spiking measurements (red asterisk) that are analyzed at a higher temporal resolution in Fig 8A. (B1-B6) Scatterplots of the measured values for the given experiment, as indicated, versus “Eyes closed.” Each data point represents a single smoothed, nonoverlapping 60-second sample. Data points are segregated into “task-engaged” (black) and “rest” (gray) using the value of “Eyes closed” as described. Red lines connect medians (data in S22 Data). (C1-C6) Lines connecting medians as in (B1-B6) for all experiments (“expts”) used (N = 11) (data in S23 Data). LFP, local field potential. https://doi.org/10.1371/journal.pbio.3000080.g007 The mean neural and HR measurements thus segregated showed systematic changes as the animal switched between states of rest and task engagement but in a direction opposite to that expected to increase blood volume at rest (Fig 7A). Thus, the mean HR, averaged over the PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 13 / 34 Modulating task-related hemodynamics moving 60-second window, reduced systematically relative to its local baseline value each time the animal disengaged from work and rested with eyes shut (Fig 7A: bottom trace [“HR”]: see shaded areas indicating rest). This is consistent with the abrupt falls in mean HR and blood pressure seen at sleep onset in human subjects [58]. But it suggests that the concurrent increase observed in V1 local blood volume is not a passive consequence of cardiovascular changes, as that would require an increase rather than a decrease in HR [59]. Similarly, the mean spike rate recorded at individual electrodes typically decreased as the animal rested. If the blood vol- ume at rest were driven linearly by local spiking, then the mean spike rate should have increased [1]. Although the mean spike rates at individual electrodes largely decreased, the pairwise correlation of spike rates over the pair of electrodes showed the expected [56] striking increases for the epochs of rest versus engagement. Comparing hemodynamics and spiking for the same data set at the higher imaging tempo- ral resolution (15 frames/second; Fig 8) supported our contention that the large blood volume increases at rest are not predicted from spiking. This conclusion was not immediately apparent on qualitative inspection (Fig 8A; same data segment as enclosed in the red box in Fig 7A). At the higher temporal resolution, the spiking response showed expected [60] bursts of high instantaneous spike rate (red arrow, Fig 8A) that stood out despite the overall reduction in mean spike rate as the animal rested with eyes shut (red asterisk in “Spike1,” Fig 7A). The cor- responding blood volume measurements showed large swings in amplitude that appeared, qualitatively, to follow the bursts of spiking. Our earlier work showed that the recorded hemo- dynamics is poorly predicted by spiking when the animal is engaged in his task, because of the presence of the task-related response (see S1 Fig; [31,44,55]). But there should be no task- related response, by definition, when the animal is disengaged from the task with his eyes shut and the hemodynamics could in principle be predictable from spiking. It was thus important to test the relationship between the two at this higher temporal resolution. We tested using deconvolution—i.e., multilinear regression (see Methods, Eqs 9–12), which has the advantage that it makes no assumptions about HRF shape [61]. The deconvolu- tion was done over partially overlapping 150-second windows (75-second steps; 150 seconds typically covered 10 trials) to get adequate temporal resolution for tracking rest states (e.g., the rest epoch in Fig 8A, indicated by the red bar, lasts about 400 seconds. Shorter deconvolution windows led to excessive noise). Each design matrix contained not only the spiking regressor for the given deconvolution window but also additional intercept and slope terms. The inter- cept is analogous to the “y intercept” in 1D linear regression, quantifying an inhomogeneous addition to a homogeneous linear equation. Here, we defined it as an estimated inhomoge- neous “mean shift” in the hemodynamic response, in addition to hemodynamic components that are linearly predictable from spiking. The full prediction using the deconvolved HRF ker- nel plus additional “mean shift” matched measured hemodynamics very well overall (Fig 8A, compare “Hemo, full pred,” green with “Hemo, meas,” black. Goodness of fit, R2 = 0.94, over this rest epoch). The HRFs from deconvolution windows falling within the rest epoch also matched each other well and resembled canonical HRFs (inset “HRFs,” Fig 8A; also see S5 Fig for an example canonical HRF). They predicted the high-frequency fluctuations in the mea- sured hemodynamics well, indicating that these high-frequency terms are accounted for by spiking (Fig 8A, “Hemo, spiking pred,” red). However, they failed to account for the increase in the mean blood volume, predicting a decrease instead (prediction rising above baseline), which is consistent with the decrease in the local mean spiking. The measured increase in the blood volume was well accounted for, on the other hand, by the fitted “mean shift” (Fig 8A, “Hemo, mean shift”; the slope terms made only small contributions). The same pattern was seen over the full 2.5-hour recording (Fig 8B). Linear predictions from spiking matched the high-frequency fluctuations of hemodynamics during rest epochs whereas the additional PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 14 / 34 Modulating task-related hemodynamics Fig 8. Hemodynamic response during rest is the sum of a high-frequency component predicted linearly by spiking plus an additional mean shift not predicted by spiking. The same data set as in Fig 7 is shown at a temporal resolution of 66.7 ms (15 Hz camera frame rate). (A) Expanded view of the sections of “Eyes closed,” “Hemo,” and “Spike1” data enclosed by a red box in Fig 7A, along with three alternate predictions of the measured “Hemo” response. “Eyes closed” is shown as in Fig 7A, top trace, with green and red bars indicating periods of task engagement and rest. Red arrow over the “Spike1” points to peaks of high instantaneous spike rate, indicating burst of spiking despite lower mean spike rate over this epoch (24.7 spikes/second average under “Eyes closed” red bar, Panel A, versus 29.4 spikes/second average in the two flanking green sections where eyes were open; same data as in Fig 7A, red asterisk). “Hemo” refers to 4 different hemodynamic traces color-coded as in the key. Black (“meas.”) = the measured response, same data as in Fig 7A. Green (“full pred.”) = full prediction following deconvolution. Red (“spiking pred.”) = linear prediction from spiking using deconvolved HRFs. The inset (“HRFs”) shows optimal HRF kernels from deconvolution windows in the “Eyes-closed” segment; colors are arbitrary. Magenta (“mean shift”) = fit to the intercept term in the design matrix, estimating components not predicted by spiking (see accompanying text). Black arrowheads pertain to an additional analysis in supplementary data; they point to two segments marked for comparison with an alternate deconvolution and prediction made without intercept terms in the design matrix (see S8 Fig). (B) Results of the deconvolution and prediction as in (A), shown over the full experiment. (The location of the expanded section in panel A is also indicated.) Only measured spiking, measured hemodynamic trace, and PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 15 / 34 Modulating task-related hemodynamics deconvolved “mean shift” are shown; full and “spiking” predictions are not shown to avoid clutter. Red arrow marks a burst of high instantaneous spiking despite lower overall mean; this burst is expanded in (A). HRF, hemodynamic response function. https://doi.org/10.1371/journal.pbio.3000080.g008 mean shift tracked the mean measured hemodynamic response (compare Fig 8B, “mean shift” with “Hemo” trace in Fig 7A). This result supports the suggestion that the large changes in mean blood volume during rest were likely driven by a mechanism acting in addition to spik- ing. Similar results were obtained for all extended recording sessions, including ones in which the mean spike rate increased during rest. It could be argued that the deconvolved “mean shift” is just the fit to the intercept term that we chose to include in the design matrix. It would thus necessarily fit the mean of the mea- sured response, by design, with no additional physiological significance. We tested by fitting the same data, using an identical deconvolution approach but without intercept terms in the design matrix (see S8 Fig). The resulting prediction was much worse at matching the measured hemodynamics. In addition, the deconvolved HRFs obtained from this new fit were markedly different from canonical spiking HRFs in two ways. First, the HRFs now incorporated the mean CBV for their respective deconvolution windows. They thus acquired large mean shifts that made them apparently acausal with large nonzero values prior to time zero. In addition, HRFs from successive deconvolution windows were noisy and matched each other poorly. This distinctly poorer fit without the intercept term suggests that the “mean shift” in the full fit (Fig 8) represents a physiological component of the hemodynamic response during the eyes- closed, disengaged behavioral state. Discussion Our goal is to understand the task-related endogenous component of hemodynamic responses recorded from visual cortex of subjects engaged in cued, predictable tasks. The existence of such responses has been known for more than a decade [16,33,35–41]. Their substantial strength relative to other brain hemodynamic components is well recognized [34]. Recent studies suggest their relevance to sensory processing [33,35]. Yet they have not been adequately studied, and little is known about their underlying mechanism or behavioral significance. Here, we consider the behavioral correlates of one particularly prominent task-related response recorded by us in V1 of behaving macaques [31], which is likely analogous to responses seen in human visual cortex [16,37]. Our work here suggests that this task-related response reflects brain mechanisms associated with the degree of task engagement. On increasing reward size to get the animal more engaged, the most notable effect, trial-by-trial, was improved temporal precision: the response became consistently more crisply aligned to task timing. It also became modestly stronger. At a slower timescale, different levels of task engagement led to consistent shifts in the mean local blood volume. High-reward blocks led to consistent decreases in the mean blood volume, whereas low-reward blocks led to corresponding increases. This effect was even more pronounced when the animal disengaged completely from his task and rested with eyes closed. The mean blood volume increased strikingly during these breaks, returning to baseline when the animal resumed working (also see [62]). Other than a high-frequency component while the animal slept, none of the hemodynamic measurements at any time scale—whether trial-by-trial or averaged across blocks while the animal worked, or the mean while the animal slept—could be accounted for by concurrent local spiking. On a methodological note, we recorded the hemo- dynamic response using ISOI at a wavelength tuned for measuring cortical blood volume. Such recordings have a steadier and more reliable baseline than blood oxygen level–dependent PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 16 / 34 Modulating task-related hemodynamics (BOLD) fMRI, suffering much less from instrumental noise and drift. This allowed us to moni- tor the response continuously over many hours, thus obtaining the reported results, including the prominent mean shifts. We propose that the task-related hemodynamic response and the effects reported in this paper comprise a marker of task-specific arousal tied to the level of engagement during an extended and possibly repetitive task. The term “arousal” is used in different senses in different areas of research, from arousal during tasks such as here to arousal in the face of fear or danger to nonspecific arousal along the sleep–wake axis [48]. The literature on task-specific arousal thus suggests avoiding the term “arousal” in favor of “vigilance,” “alertness,” or “sustained attention” [48–50]. This condition of sustained engagement in a task is known to fluctuate between states of higher stability, which are less prone to error (“in the zone”), and states that are more unstable and error-prone (out of “the zone”; see, e.g., [50]). The state of being “in the zone” is marked by higher regularity and temporal precision; responses “in the zone” show less variability in reaction time, trial-by-trial, even when the mean reaction time remains unchanged overall [50]. The state is further enhanced by reward, which leads to even less vari- ability in reaction times, in a manner that appears distinct from increased arousal [63]. This behavioral result may have a physiological analog in our finding of improved temporal preci- sion or regularity of the task-related hemodynamic response during high reward (Fig 3 and accompanying text). An attractive possibility is that the task-related response reflects the behavioral state variously labeled “vigilance,” “alertness,” or “sustained attention” in the cited literature. Our term “task engagement” is a shorthand reflecting this possibility as well as a nod to the initial report of this hemodynamic response component, which described it as “task structure” related [16]. Much remains to be done to flesh out these connections. An important question remains: How closely does the task-related hemodynamic response we record in macaque V1 correspond to the response identified in human visual cortex? And how distinct is it from selective visual attention [16]? Currently, the strongest evidence is that although varying substantially in time, the task-related response in the macaque is spatially homogeneous over the imaging window (a circular region 15 mm in diameter, typically extending approximately 1–6˚ eccentricity [32]). This makes it unlikely to be selective atten- tion at the fovea (e.g., for the task of discriminating the fixation cue color) that should lead to spatially graded activation over this cortical extent [16,64,65]. Additional evidence comes from the response timing. If it were selective attention cued to the fixation point, its time course at fix onset should be stereotyped independent of trial length. That was not the case in an earlier test; the starting time course even switched sign when switching between blocks of short versus long trials (e.g., 8 versus 20 seconds; see Fig 3 and Supplementary Fig S9 in [31]). This result also speaks to a corollary question arising from our describing the response as being entrained to task timing. It could be argued that the hemodynamics and the sympathetic-like changes in HR and pupil are, instead, responses to the reward acting as a stimulus. However, our earlier results noted above showed the hemodynamics to be entrained to the expected timing of upcoming trials rather than a stereotyped response to the reward. It should thus be interpreted as task-entrained albeit modulated by the current reward. However, although the evidence is strongly suggestive, the questions are interesting and open and remain topics of ongoing research in the lab. What could be the underlying mechanism or function? Although the poor prediction by recorded spiking does not rule out control by a small, hard-to-measure set of specialized neu- rons, it also suggests a different underlying mechanism such as neuromodulatory input (e.g., see [66]). The strong sympathetic-like responses (Fig 1) suggest the basal forebrain-cholinergic (BF-ACh) or the locus coeruleus-adrenergic (LC-NA) systems, both of which are linked to wakeful states, arousal, and attention. They powerfully facilitate hemodynamic responses via PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 17 / 34 Modulating task-related hemodynamics modulation of stimulus-evoked neural responses (reviewed in [67]) with additional control of cortical blood flow [68] through direct modulation of microvessels [69,70] or via astrocytes [71] or pericytes [72]. Other neuromodulators such as dopamine could also be involved [73]. All of this can lead to robust neuromodulator-mediated increases [68,74,75] or decreases [70– 72,76] in cortical blood flow. Our finding of a stereotyped response shape independent of tem- poral precision could be accounted for by a mechanism in which the response timing is deter- mined in a distal nucleus through temporal dynamics local to that nucleus. The result could then be transmitted in an all-or-nothing manner, like an action potential, to the target (here, V1), where it could release a fixed quantum of neurotransmitter. In addition, there could be a contribution from myogenic mechanisms independent of neural input, as suggested in a recent study of ongoing vascular fluctuations [77]. A combination of such mechanisms could modulate neural responsivity and vascular tone in a manner that reflects the level of task engagement. In single trials, this could help refresh the local blood supply ahead of task onsets [78]. Over more extended periods, it could also shift the mean vascular tone—e.g., by slow accumulation of the active substance—to higher values for high task engagement and the con- verse for low. Such a mean shift could account for the surprising finding of progressively lower mean blood volume for higher task engagement, since higher vascular tone does imply nar- rower blood vessels and thus lower tissue fraction occupied by blood. The increased vascular tone could have additional functional benefits of higher precision in stimulus-evoked hemody- namic responses. For example, adrenergic increase in vascular tone has been shown to lead to spatially and temporally sharper vascular responses to neural activity [71]. Exploring these issues through targeted experiments in behaving animals would be crucial to understanding brain mechanisms of task engagement. Methods Experimental model and subject details Animal use procedures were in accordance with the United States NIH Guide for the Care and Use of Laboratory Animals and were approved by the IACUC of Columbia University and the New York State Psychiatric Institute (Animal Care Protocol AC-AAAU1456). Two male rhe- sus macaques (Macaca mulatta) were used in the study. Access to water was scheduled to training or recording sessions that lasted 3–5 hours per day. Eye fixation and pupil diameter were recorded using an infrared eye tracker (ISCAN [79]). Before training, each animal was implanted with a stainless steel or titanium head post. After training, craniotomies were per- formed over the animals’ V1, and glass-windowed stainless steel or titanium recording cham- bers were implanted for subsequent ISOI in the behaving animal (see section “ISOI” below). The craniotomy exposed a 20-mm diameter area of V1 covering visual eccentricities from about 1 to 10˚. The exposed dura was resected and replaced with a soft, clear silicone artificial dura (GE Silicone RTV615 001). Recording chambers and artificial dura were fabricated in our laboratory following published designs [80,81]. Chambers were opened regularly for clean- ing, testing for infection, and treating if necessary, following published protocols [43]. Method details Summary. Extracellular electrode recording was carried out simultaneously with ISOI from V1 of behaving monkeys performing a periodic visual fixation task. Task and recording methods are essentially identical to those in earlier papers from our lab [31,44,45,55]. Task and reward schedules. All experiments were based on a simple fixation task carried out either under essentially complete darkness or in the presence of visual stimuli. In both con- ditions, animals held fixation periodically, cued by the color of a fixation spot (fixation PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 18 / 34 Modulating task-related hemodynamics window: 1.0–3.5 deg. diameter; monitor distance: 133 cm; fixation duration: 3–5 seconds; trial duration: 9–22 seconds; all parameters fixed for a given experiment but variable between experiments). A juice reward followed every correct (unbroken) fixation, with no time-out or other punishment for errors. The primary behavioral manipulation consisted of systematically changing reward size. For fixation trials in the dark room, the monitor was covered and the fixation point was presented behind a pinhole [31]. Reward sizes were alternated between high (typically 0.45 ml per correct trial, ranging from 0.35 to 0.6 ml) and low (typically 0.15 ml per correct trial, rang- ing from 0.1 to 0.2 ml. High- and low-reward sizes were fixed for an experiment; they were selected per day based on the animal’s willingness to work for the low reward). Rewards were alternated in blocks (typically 10 correct trials each; some experiments had longer blocks; some experiments had blocks of variable size). The animal had to correctly complete the full set of trials per block—i.e., not counting error trials—before the reward switched. Trials were grouped into “high-reward” and “low-reward” blocks for analysis. Errors were identified by the block in which they occurred—i.e., as “high” or “low” reward based on the preceding cor- rect trial. Continuous sequences of error trials were counted as a single error to avoid arbitrary overcounting during epochs in which the animal disengaged from work and took a nap. Thus, each counted error corresponded to a break following one or more correct trials. These experi- ments accounted for the majority of the reported results. For visually stimulated trials (Fig 5 and S5 Fig), the animals were passively shown gratings of different contrasts while holding fix- ation (sine-wave gratings; contrasts doubled in steps ranging typically from 6.25% to 100%; mean luminance = background luminance = 46 cd/m2; spatial frequency: 2 cycles/deg; drift speed 4 deg/second; diameter 2–4 deg; orientation optimized for the electrode recording site. These data are reanalyses of earlier experiments designed to relate hemodynamics to electro- physiology over a wide dynamic range of stimulated responses [44,45,55]). Reward sizes for these experiments increased progressively from a baseline (typically 0.2 ml per correct trial) to a maximum value (typically 0.6 ml per correct trial) for each successive correct fixation to keep the animals motivated. Again, the lowest reward size per day was chosen based on the animals’ willingness to work. For analysis, trials were grouped into “high-” and “low-reward” sets rela- tive to the median reward. Sequences of errors were also counted as single errors for these experiments, as with the dark room. However, errors were not identified as “high” or “low” reward. Since rewards were increased progressively for correct trials, errors (which often occurred at the end of a sequence of trials) typically followed a high reward; but that associa- tion was not informative. ISOI. ISOI is based on the finding that in vivo and in the visible spectrum, changes in light absorption in cortical tissue primarily measure changes in oxy- and deoxyhemoglobin in the blood flowing through cortical blood vessels [51,82,83]. ISOI deduces hemodynamic responses by imaging changes in light reflection at relevant wavelengths off the exposed corti- cal surface. CBV and oxygenation changes measured using ISOI can be used to predict concur- rently measured fMRI responses [53], making ISOI in effect an optical analog of fMRI albeit restricted to upper layers of exposed cortex. We imaged at 530 nm (green), an isosbestic wave- length that is equally absorbed by oxy- and deoxyhemoglobin. Increased absorption of light at this wavelength thus measures increased cortical tissue fraction of hemoglobin—in effect local cortical blood volume, independent of oxygenation state [54]. After the animals had recovered from surgery, we used this technique to image their V1 through the glass window of the recording chamber routinely while they engaged in the fixation task. Imaging hardware con- sisted of the following: camera (Dalsa 1M30P; binned to 256 × 256 pixels, 7.5 or 15 frames per second) and frame grabber (Optical PCI Bus Digital; Coreco Imaging). Imaging software was developed in our laboratory in C++ based on a previously described system [84]. Illumination PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 19 / 34 Modulating task-related hemodynamics was provided by high-intensity LEDs (Agilent Technologies, Purdy Technologies). The lens was a macroscope [85] of back-to-back camera lenses focused on the cortical surface. Imaging, trial data (trial onset, stimulus onset, identity and duration, etc.), and behavioral data (eye position, pupil size, timing of fixation breaks, fixation acquisitions, trial outcome) were acquired continuously. All data analyses were performed offline using custom software in MATLAB (MathWorks; RRID:nlx_153890). Electrophysiology. Electrode recordings were made simultaneously with optical imaging. Recording electrodes (FHC, AlphaOmega; typical impedances approximately 600–1,000 kO) were advanced into the recording chamber through a silicone-covered hole in the external glass window, using a custom-made low-profile microdrive. Recording sites were mostly but not exclusively confined to upper layers. Signals were recorded and amplified using a Plexon recording system (RRID:nif-0000-10382). The electrode signal was split into spiking (100 Hz to 8 kHz bandpass) and LFP (0.7–170 Hz). Subsequently, an additional analog 2-pole 250-Hz high-pass filter was applied to spiking, effectively eliminating any spectral power overlap between LFP and spiking. No attempt was made at isolating single units, and all measured spiking was multiunit activity (MUA) defined as each negative-going crossing of a threshold = roughly 4× the r.m.s. of the baseline obtained while the animal looked at a gray screen. The LFP recording was analyzed to obtain two bandpass-limited measurements in the beta- and theta-range frequency bands (15–25 Hz and 3–7 Hz, respectively; multitaper spectral analysis using the Chronux MATLAB toolbox). This gave an LFP measure of (low) vigilance defined by the square root of the ratio of power in theta versus beta. Analysis: Preprocessing. The imaging measurement was averaged over the imaged area, frame by frame (frame rate: 7.5 or 15 frames/second), and then divided by the mean value of this quantity for the given experiment (over all trials). This converted the measurement per image frame into dR R particular imaging wavelength of 530 nm, the negative of this quantity ((cid:0) the fractional increase in local tissue hemoglobin—i.e., the fractional increase in local cortical blood volume [54]. The dR was then detrended, and a prominent pulse artifact was filtered out R from the measured hemodynamics using Runline (Chronux) with a window of 2 seconds. This filtered dR R (i.e., the fractional change in light reflected off the cortical surface). At the ) is proportional to defined the measured hemodynamic response for all calculations. dR R The pulse artifact was used to estimate the instantaneous HR after upsampling 8× and iden- tifying peak times and thus the local pulse rate. Both the estimated HR and the spiking mea- surements were then resampled and aligned to the imaging frames. Neither the imaging nor spiking nor estimated HR were further temporally filtered. Unlike in our earlier papers, we did not high-pass filter to remove slow fluctuations [31,44,45,55], specifically so as to be able to estimate fluctuations over slow timescales of many minutes. Template matching (dark-room experiments: Fig 2). The amplitude and timing of the task-related response, per trial, were estimated as the height and location of the corresponding peak of a Template Match. This Template Match consisted of the continuous, normalized dot product of a template with the measured hemodynamic response. The calculation involved the following steps: 1. The default template “Tmplt” was defined to be the one-trial-long mean hemodynamic recording (z-scored to give “H(t)”) aligned to trial onsets, averaged across all correct trials, and mean-subtracted. 2. This Tmplt was then slid over H(t) in unit time steps (at the resolution of the imaging frame rate; e.g., 66.7 ms for 15 frames/s). At every time point t, the Template Match was defined to be the local dot product over the one-trial-long section of H centered on t, normalized by PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 20 / 34 Modulating task-related hemodynamics the (fixed) sum of squares of the Tmplt (Fig 2B): Template Match ¼ HðtÞ:Tmplt P jTmpltj2 ð1Þ This expression is identical in form to the Pearson’s correlation between the Tmplt and the same one-trial-long segment of H(t), other than the normalization. Thus, like Pearson’s r, this expression would have local maxima where the local one-trial-long segment of H(t) matched the Tmplt in shape (Fig 2B). The Template Match is also invariant to shifts in the mean of H(t), since the Tmplt is mean-subtracted and thus integrates to zero over any addi- tional constant. However, unlike Pearson’s r, this expression carries scale information. Pearson’s r has the standard deviation of both arguments in the denominator, making it scale-invariant. The Template Match on the other hand, with its fixed normalization inde- pendent of H(t), scales linearly with the amplitude of fluctuation in H(t). Thus, peaks of the Template Match carry information about both timing and amplitude of the task-related response per trial. 3. For computational efficiency in MATLAB, the above expression was rewritten as the nor- malized convolution of H(t) with the time-reversed version of the template Tmplt: Template Match ¼ HðtÞ � TmpltTR jTmpltj2 P ð2Þ where TmpltTR is the time-reversed Tmplt; i.e., TmpltTR(t) = Tmplt(−t), and the symbol � denotes convolution. The denominator for normalization remains unchanged. This expres- sion translated to the following script using MATLAB functions conv and sum: Template Match ¼ convðH; TmpltTR;0same0Þ sumðjTmpltj2Þ ð3Þ Peaks of Template Match were then identified as zero crossings of the first derivative at points where the second derivative was negative (marked by red dots in Fig 2B). 4. Alternate template matches used the same formalism but with different definitions of Tmplt. Thus, the Tmplt in S2 Fig was defined as the one-trial-long mean H(t) aligned to a point that was one-quarter trial ahead of trial onsets, averaged across all correct trials, and mean-subtracted. All other steps were the same. Template matching (with visual stimuli present: Fig 5, S5 Fig). This involved two sepa- rate sets of steps. 1. Estimating the task-related response from the net recorded hemodynamic response: i. We modeled the net recorded response as a linear sum of stimulus-evoked and task- related components. The stimulus-evoked response was modeled as the convolution of concurrent spiking with an “HRF” kernel. The task-related component was estimated iteratively. Our earlier approach [55] had modeled it as a stereotyped task-related func- tion (“TRF”) that was identical in timing and amplitude for each correct trial. Here, how- ever, we specifically need to estimate trial-by-trial variations in response timing and amplitude. As a first step, we assumed that the TRF had a fixed shape that could be esti- mated from the mean across trials. Optimal mean HRF and TRF kernels were obtained by PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 21 / 34 Modulating task-related hemodynamics fitting the mean recorded responses, separated by contrast, with the following equation, identical to Eq 1 in [55]: HðtÞ ¼ HRF � SðtÞ þ TRF � TrlðtÞ ð4Þ H(t) is the recorded hemodynamics and S(t) the concurrently measured spiking, and HRF�S(t) models the stimulus-evoked response. The second term on the RHS models the task-related response as a TRF kernel convolved with the set of delta functions at trial onsets, "Trl(t)”. The symbol � denotes convolution over time. The HRF kernel was parametrized, as before [31,44,45], as a gamma-variate function of time t: HRF t; t; W; A ð Þ ¼ A � �a t t � exp (cid:0) a � t (cid:0) t t ð5Þ The HRF parameters fitted during optimization are the amplitude A, time to peak τ, and full width at half maximum W [31,86,87]. The factor a ¼ 8:0 � log 2:0ð The TRF kernel was parametrized as the finite sum of a Fourier time series: � � � (cid:0) �2. Þ � t W � XN n¼1 � an cos n 2p PT t � þ bn sin n 2p PT t ð6Þ TRF t; a; b; P; N ð Þ ¼ Although the Fourier series was based on the trial period T, the fundamental Fourier period was allowed to vary as a fraction P of the trial period and optimized in the fit. The parameters an and bn, (with n ranging from 1 to the total number of terms in the Fourier series, N) are the pairs of cosine and sine coefficients, respectively, for the nth Fourier term. We showed earlier that only the fundamental and first harmonic—i.e., N = 2, carry significant information [55]. Thus, there are eight parameters in the model: three for the HRF, the two pairs of an and bn, and P. ii. All parameters were optimized simultaneously by matching the predicted to the mea- sured hemodynamics using a downhill simplex algorithm (fminsearch, MATLAB meth- ods as in [45]). To keep contrast information, we made concatenated sequences of the mean response per distinct contrast, randomized per contrast (same random sequence for hemodynamics, and spiking), and over multiple blocks (an arbitrarily large number 52, about 100× larger than a single HRF kernel convolution length, to minimize edge effects; we only matched traces two convolution lengths in from the edge). The error to � � be minimized was defined as the normalized sum squared error SSerror SStotal calculated sepa- rately per contrast and then averaged over all contrasts including the blank. This was intended to give equal weight to the fractional error at each stimulus contrast. The good- ness of fit R2 for the optimal prediction was defined as the coefficient of determination � 1 (cid:0) calculated separately per contrast and averaged across contrasts. This, again, � SSerror SStotal was intended to give equal weight to errors at each contrast. In order to reduce the chances of getting caught in a local minimum, we started with large sets of initial param- eter values, independently covering an order of magnitude for each fitted parameter. The fits were robust and converged to the same optimal parameters from multiple starting values, giving us confidence that we had reached global and not local minima. PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 22 / 34 Modulating task-related hemodynamics iii. Next, we used the optimal fitted HRF thus obtained from the mean hemodynamics and spiking, averaged per contrast, to get a continuous estimate of the exogenous, visually evoked component of measured hemodynamics. This was done by convolving the opti- mal HRF with the measured spiking, including both the spiking from the controlled visual stimuli and from uncontrolled visual stimulation as the animal looked around in between fixations: HSTIMULATEDðtÞ ¼ HRF � SðtÞ ð7Þ iv. This estimate of HSTIMULATED was subtracted from the full measured hemodynamics to get an estimate of the endogenous, task-related component of hemodynamics as the residual not accounted for by spiking: HTASK(cid:0) RELATEDðtÞ ¼ HðtÞ (cid:0) HSTIMULATEDðtÞ ð8Þ 2. Estimating task-related response peaks and amplitudes: i. The estimate of the task-related response HTASK−RELATED(t) as defined above was then used, exactly like the full hemodynamic response in the dark room, to estimate response times and amplitudes trial-by-trial. As a template, we used the optimal TRF obtained above by fitting to mean responses. The steps for template matching and identifying and analyzing peaks were identical to those outlined in Eqs 1–3, with the H(t) being replaced with HTASK−RELATED(t) and the Tmplt(t) being replaced with the optimal TRF. Peak times and amplitudes, per trial, were obtained exactly as in the dark-room template match. Tracking measured variables as animal switches from task-engaged to disengaged with eyes closed (Fig 7). To track slow changes in all measured variables, we downsampled the data. Data were averaged using a 15-second box car that corresponded roughly to a single trial and then decimated 4×, giving in effect a smoothed 60-second sample rate. Along with MUA spike rates at individual electrodes and the measured hemodynamics, the following measure- ments were thus tracked: 1. “Eyes closed”: Fraction of time over the 60-second averaging window that eyes are closed. Eye closure was monitored using the output from the IR eye tracker. All blinks or eye clo- sures appeared as sequences of missing points or “rails” (saturated output). Spontaneous eye blinks in macaques last roughly 200 ms (see [88,89]). Our own data showed a bimodal distribution with blink durations peaking either at 200 ms or multiple seconds to minutes. Thus, sequences of missing points lasting <500 ms were considered regular spontaneous blinks while awake and were marked as having duration = 0. Sequences lasting >500 ms were categorized as eye closures, and their durations were included in the moving average. 2. “LFP measure”: square root of the ratio of spectral band–limited power in the theta (3–7 Hz) and beta (15–25 Hz) frequency bands, each normalized by its standard deviation over the entire experiment. We chose this particular ratio to get a measure that was high during epochs of low engagement in the task to match “Eyes closed,” since theta power increases sharply on transitions from high to low engagement or to sleep [48]. We took the square root of the ratio to compress the measure to approximately 0–1 to make it comparable to “Eyes closed.” There was no attempt to separate the resting state more finely into sleep PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 23 / 34 Modulating task-related hemodynamics stages, since the goal was a broad separation into states of “task-engaged” versus “resting” with a time resolution of 60 seconds. 3. “Spike1,” “Spike2”: MUA responses recorded from two electrodes spaced 1 mm apart, in imaged V1. 4. “Spk1-Spk2 Corr”: Pairwise correlation of MUA spike rate from the two electrodes, calcu- lated over a 1-second moving window. 5. “HR”: Obtained from the pulse artifact in the measured hemodynamics, after upsampling 8× and identifying peak times and thus the local pulse rate. This instantaneous pulse rate, estimated at pulse time points, was then interpolated with spline smoothing to the imaging time base (7.5-Hz or 15-Hz sample rate depending on the experiment). Deconvolution, i.e., multilinear regression (Fig 8). We started with the assumption that the measured hemodynamics H(t) can be predicted from local spiking S(t) using a homoge- neous linear equation, along with two inhomogeneous terms: a scalar Intercept and a linear Slope in each 150-second window, at the resolution of the camera frame rate: HðtÞ ¼ HRF � SðtÞ þ Intercept þ Slope ð9Þ There are no assumptions about the shape of the HRF other than that it does not extend more than 10 seconds prior to time 0 and is back to baseline about 25 seconds after time 0. Using the formalism of deconvolution, this expression can be rewritten as a matrix equation H ¼ S � HRF þ Intercept þ Slope ð10Þ where H is a column vector of recorded hemodynamic responses (at the temporal resolution of the camera frame rate, 15 Hz); S is the spiking regressor expanded into a stimulus convolu- tion matrix (SCM) [37,61]; the symbol × indicates matrix multiplication; and HRF, Intercept, and Slope here refer to the same terms as in Eq 10 but expressed as column vectors. The SCM was constructed as a Toeplitz matrix comprising a horizontal concatenation of spiking column vectors, with circular time shifts ranging from −10 seconds to +25 seconds relative to t = 0. Formally, the SCM S can be extended (“Se”) to incorporate the Intercept and Slope by horizon- tally concatenating the two additional column vectors: a column of ones for the Intercept and a linear ramp from −1 to 1 for the Slope. The corresponding HRF can be formally extended (“HRFe”) by two coefficients, one for the Intercept and another for the Slope. H ¼ Se � HRFe ð11Þ Assuming that any noise is Gaussian and has zero mean, the optimal deconvolved HRFe can then be estimated using a least-squares solution to the linear regression [37,61]: HRFe ¼ ðSe T � SeÞ(cid:0) 1 � Se T � H ð12Þ where the superscript T indicates the matrix transpose, -1 indicates the matrix inverse, and × indicates matrix multiplication. The full prediction using the optimal deconvolved HRFe kernel is then computed using Eq 11. Similarly, just the (linear, homogeneous) prediction from spiking is obtained by taking the matrix multiplication over all column vectors of Se, save the last two, with all coefficients of the deconvolved HRFe, save the last two. Conversely, just the Intercept term or just the Slope term are obtained by appropriately multiplying the last two column vectors of Se with the corresponding last two coefficients of the deconvolved HRFe. The optimal deconvolved Intercept is defined as the additional “mean shift” not predicted by PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 24 / 34 Modulating task-related hemodynamics spiking. As with the model-based approach to fitting used earlier (see Eqs 4–6), the goodness � of fit R2 for the optimal prediction was defined as the coefficient of determination 1 (cid:0) SSerror SStotal This was used to compare fits made with versus without including an intercept term in the design matrix (Fig 8A versus S8 Fig). � . Getting bootstrap estimates for significance (p-values). All comparisons between distri- butions of amplitudes or peak times were tested for significance by bootstrapping, typically using 10,000 resamples with replacement. In cases with different numbers of trials for high and low reward, the smaller number of trials was chosen to make the bootstrap comparison. For comparing response amplitudes (e.g., Fig 3B), we tested for the median of high-reward amplitudes being less than that for low-reward amplitudes over the set of all resamples, against the null hypothesis that this difference has zero mean. We also tested for the complement— i.e., that median of low-reward amplitudes is less than that of high-reward amplitudes. For comparing widths of peak time distributions (e.g., Fig 3A), we first calculated the 2 standard deviation width (specifically, the +/− 34th percentile around the median, given the non-nor- mal distribution) of each bootstrapped set of peak times separately for high and low reward. We then tested for 2 standard deviation for high reward being less than that for low reward over the set of all resamples, against the null hypothesis that the difference has zero mean. We also tested for the complement—i.e., that 2 standard deviations for low reward was less than that for high reward. Fitting spiking to dark-room hemodynamic response using gamma-variate HRF (S1 Fig). To link to spiking, the dark-room response was modeled as a homogeneous prediction from spiking, fitted by optimizing a gamma-variate HRF kernel using fminsearch as described above: HðtÞ ¼ HRF � SðtÞ ð13Þ The fitting was done separately for the high-reward and low-reward trials at each recording site. Stimulus-evoked responses were fitted using a model with a task-related component: Eqs 4–6. In each case, the optimal fitted HRF kernel was then convolved with the continuous recorded spiking response to give a continuous prediction. Since the spiking response included both high-reward and low-reward segments, the prediction included sections of “same” pre- diction (e.g., low-reward spiking convolved with the low-reward kernel) and sections of cross predictions (e.g., high-reward spiking convolved with the low-reward kernel). Supporting information S1 Fig. Local spiking, although appearing to predict mean hemodynamic responses in indi- vidual recording conditions, is a poor and unreliable predictor of task-related responses overall. (a,b) Mean measured responses and optimal predictions for low-reward and high- reward trials, respectively, of a data set recorded in the dark-room task. In each case, the lower panel shows the mean measured spiking; the upper panel shows the mean measured hemody- namics as well as the prediction from spiking using the corresponding optimal fitted gamma- variate HRF kernel (see color code in each column). Low reward (N = 148 trials), R2 = 0.73 for the optimal prediction. High reward (N = 140 trials), R2 = 0.42 for the optimal prediction. (c) A separate set of visually stimulated trials at the same recording site, using visual stimuli con- sisting of optimally oriented drifting gratings at different contrasts, as indicated by the gray- scale coding (orange bar below depicts the visual stimulation period). Again, the top panel shows mean measured hemodynamics and optimal predictions grouped by stimulus contrast; predictions are shifted to the right for visibility (N = 141 trials total, i.e., 47 trials / contrast. R2 PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 25 / 34 Modulating task-related hemodynamics = 0.95). (d) The optimal fitted gamma-variate HRF kernels for the three recording conditions, color coded as shown. Note how poorly they match each other. (e) Comparing the measured low-reward hemodynamics to predictions using the low-reward dark-room set of spiking responses (as in panel a)—but with different optimal HRF kernels—from low-reward, high- reward, and stimulus-evoked sets. The cross predictions are poor (R2 of prediction using high- reward HRF = −0.014; stimulated HRF = −0.011). (f) Optimal HRFs from the full set of dark- room experiments, normalized in each case to the amplitude of the corresponding visually stimulated HRF (N = 56: pairs of high- and low-reward HRFs for each of 28 sets with electrode recordings). Scale truncates some HRFs of high absolute amplitude to help visualize those of smaller amplitude. Colors are arbitrary (MATLAB default). The different optimal HRFs match each other poorly, with some even reversed in sign. This makes cross prediction meaningless and suggests that apparently good predictions of mean responses in individual experiments are fortuitous. HRF, hemodynamic response function. (TIF) S2 Fig. Increase in temporal precision with reward size is not sensitive to the choice of tem- plate used to estimate response time and amplitude. (a-d) Same example data set as in Figs 2 and 3. (a) Orange indicates the alternate template defined as the mean hemodynamic response across correct trials, aligned to a time point one-quarter cycle ahead of trial onset (i.e., starting at the dashed vertical line 4.1 seconds ahead of time 0. Single trials are shown in gray). Green background (time points 0–16.4 second) marks the timing of the earlier template for compari- son (see Fig 2B, “Tmplt”). (b) New template match (orange, “Tmplt Match,” upper row) illus- trated using the same segment of recorded hemodynamics (“Hemo”) as in Fig 2B. The earlier template match from Fig 2B is shown alongside for comparison (green, dashed line). Black dots identify the peaks of the new Template Match, marking locations where the “Hemo” is locally best phase-matched to the new template (see “Match Peak,” compared to “Match Trough”). (c) Distributions of response times, defined as the positions of the new template match peaks. Compare with Fig 3A (same conventions). (d) Distributions of response ampli- tudes using the new template match. Compare with Fig 3B (same conventions). (e, f) New response timing distribution 2 standard deviation widths and amplitude medians for high- versus low-reward trials across all experiments, including p-values from Wilcoxon signed rank test for the pairwise comparisons. Compare with Fig 3C and 3D (data in S24 Data). (TIF) S3 Fig. Response timing does not correlate with fixation onset. (a) Simulation of the null hypothesis. The task-related response has a stereotyped time course following the onset of fixa- tion. Response times would then have a constant delay following fix onset, leading to a linear relation between the two with unity slope (the delay was taken to be 10 seconds for this simula- tion). The observed tighter clustering of response times for high reward could result from a corresponding clustering of fixation onsets (consider projection of red dots versus blue dots on the Response Time axis). (b) Relationship between measured response time (estimated as usual with a template match) and fixation onset in an early recording session. Animals tended to hold fixation for extended periods prior to trial onset, even across multiple trials. (c) Rela- tionship between response time and fix onset in a late recording session. Animals tended to move their eyes a lot during intertrial intervals, fixating shortly before trial onset. For both cases (b) and (c), response times were independent of fixation onset and very different from the pattern expected for the null hypothesis. In both data sets, response times for high-reward trials showed visibly lower scatter independent of fix onset (data in S25 Data). (TIF) PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 26 / 34 Modulating task-related hemodynamics S4 Fig. High reward does not correlate with tighter eye movements. (a1) Mean radial eye movement per intertrial interval in an early recording session. Each dot represents a single trial (mean eye movement during 7-second intertrial intervals per 9-second trial). Horizontal lines indicate median eye movement per block of high or low reward (blocks with varying numbers [13–31] of correct trials each). Intertrial eye movements were higher in low-reward blocks. (a2) Histogram of mean eye movement per trial. (a3) Relationship between response time and eye movement per trial, colored by reward size. (b1) Mean radial eye movement in a later recording session (12-second intertrial intervals in 16-second trials; alternating blocks of 10 correct trials each; all other conventions as in panel A1). Eye movements were higher in high-reward blocks. (b2) Corresponding histogram of mean eye movements per trial. (b3) Response time versus eye movement per trial colored by reward size. Low reward leads to wider scatter of response times in both panels (a3) and (b3) despite opposite effects on inter- trial eye movement (data in S26 Data). (TIF) S5 Fig. Estimating task-related response and its template match in the presence of visual stimulation (one example data set). (a-c) Estimating optimal fitted parameters (see Methods, Eqs 4–6). (a) The mean hemodynamic response per stimulus contrast (see key), averaged across trials. The response is modeled as the sum of the stimulus-evoked component (b) and the task-related component (c). The stimulus-evoked component is modeled as the convolu- tion (�) of the measured spiking with a gamma-variate HRF kernel (inset). The mean task- related component is modeled as the convolution of delta functions at trial onset with a “Mean TRF” kernel comprising a partial Fourier sum with its fundamental at the trial period (inset). Earlier work showed that the fundamental and the first harmonic terms of the Fourier series are adequate. Insets show the optimal fitted gamma-variate HRF (in b) and optimal mean TRF (in c), respectively. (d) Set of traces illustrating the process of estimating the residual task- related response and then estimating its timing and amplitude per trial by matching to a tem- plate (see Methods, Eqs 7 and 8). “Spiking,” “Hemo”: full measured responses, individually z- scored. “Hemo (predicted from spiking)” is the convolution of the spiking response with the optimal fitted HRF (b, inset). Subtracting this from the measured hemodynamic response gives the residual “Hemo (Unpredicted by spiking),” which we defined to be the task-related response. The moving-window dot product of this residual with the template (the optimal fit- ted mean TRF [c), inset]) gives the “Template Match” (shifted up for visibility). Timing and amplitude of task-related responses, per trial, are defined to be the location and height of each Template Match peak, as for the dark-room task. Showing a section of the full experiment of 483 trials (122 correct). (e) Set of all residual task-related responses, converted back from z- scored values, separated into trials grouped by reward size. The same data are shown in Fig 5A. HRF, hemodynamic response function; TRF, task-related function. (TIF) S6 Fig. Comparing regression lines through alternating blocks of high and low reward, before (top panel) and after (bottom panel) removing error trials. Color coding for high (red) and low reward (cyan) is the same as in the main text. Error trials are indicated in lighter colors and are grouped with the reward block corresponding to the immediately preceding correct trial. Straight lines show regression fits. Letters (“A,” “B,” “C”) and arrows identify correspond- ing blocks. Blocks A and B contain individual or short stretches of error trials. C includes a roughly 400-second stretch during which the animal napped. The time axis has the same scale for both top and bottom panels, with time 0 indicating the start of the experiment; the bottom concatenates time points for correct trials. Six consecutive blocks are shown from an PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 27 / 34 Modulating task-related hemodynamics experiment comprising 47 blocks (482 correct trials of 684 total). (TIF) S7 Fig. Ramp-like drifts in local blood volume are not accounted for by slow changes in local spiking. (a, c) Hemodynamics and spiking, respectively, showing correct trials from alternating blocks of high and low reward. Lines show regression fits per block (same data set as Figs 2 and 3). (b, d) Histograms with slopes of regression fits from (a), (c). (e) Simplified simulation of slow mean hemodynamic responses: triangle wave of matching period, with slopes equal to the median (absolute) slopes of the regression lines in (a) (= 4.1 × 10−5/second). (f) Simulated spiking response that generates the model hemodynamic response in (e) on convolving with the visually stimulated HRF for this recording site (see “HRF kernels,” S1 Fig; also, Methods). Measured spiking regression slopes (d) are only about 4× weaker than those in the simulation; but they do not alternate in sign with reward size. (g) Distributions of the ratios of measured spiking regression slope per block to the slope of the corresponding simulation, as in (e), (f), across all experiments (N = 752 blocks of 10 trials each, 376 blocks/reward size; from N = 11 experiments with electrode recordings and at least 10 blocks per reward size). p- Values test for the probability of the distributions being centered on zero (bootstrap, 10,000 resamples) (data in S27 Data). HRF, hemodynamic response function. (TIF) S8 Fig. Deconvolution fit of the same data segment as in Fig 8A but with no intercept term in the design matrix. The full prediction here matches the measured response reasonably well except for a few locations with large mismatches (black arrowheads; compare with the same locations in Fig 8A). The overall goodness of fit R2 = 0.76, averaged over this rest epoch, is worse than for the fit with an intercept (R2 = 0.94; see Fig 8A, text). The inset shows HRFs from the deconvolution windows covering this rest epoch, as in Fig 8A; colors identify corre- sponding HRFs for the two fits. HRF, hemodynamic response function. (TIF) S1 Data. Data for “Eye Pos” traces in Fig 1B. (XLSX) S2 Data. Data for “Hi–Lo” reward pupil dilation histograms in Fig 1B. (XLSX) S3 Data. Data for pupil traces in Fig 1B. (XLSX) S4 Data. Data for Fig 1C. (XLSX) S5 Data. Data for Fig 1D. (XLSX) S6 Data. Data for Fig 1E. (XLSX) S7 Data. Data for Fig 2A. (XLSX) S8 Data. Data for Fig 2C. (XLSX) PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 28 / 34 Modulating task-related hemodynamics S9 Data. Data for Fig 2D. (XLSX) S10 Data. Data for Fig 2E and 2F. (XLSX) S11 Data. Data for Fig 3. (XLSX) S12 Data. Data for Fig 4A. (XLSX) S13 Data. Data for Fig 4B. (XLSX) S14 Data. Data for Fig 4C–4F. (XLSX) S15 Data. Data for Fig 5A. (XLSX) S16 Data. Data for Fig 5B. (XLSX) S17 Data. Data for Fig 5E. (XLSX) S18 Data. Data for Fig 5F. (XLSX) S19 Data. Data for Fig 5C and 5D. (XLSX) S20 Data. Data for Fig 6C. (XLSX) S21 Data. Data for Fig 6A and 6B. (XLSX) S22 Data. Data for Fig 7B. (XLSX) S23 Data. Data for Fig 7C. (XLSX) S24 Data. Data for S2 Fig. (XLSX) S25 Data. Data for S3 Fig. (XLSX) S26 Data. Data for S4 Fig. (XLSX) S27 Data. Data for S7 Fig. (XLSX) PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 29 / 34 Modulating task-related hemodynamics Acknowledgments Thanks to Maria Bezlepkina and Elena Glushenkova for technical support and lab manage- ment, to Liam Paninski for suggestions about analyzing slow changes in physiological responses, and to David Heeger, Mike Shadlen, Elisha Merriam, Charles Burlingham, and Saghar Mirbagherion for comments. Author Contributions Conceptualization: Yevgeniy B. Sirotin, Aniruddha Das. Data curation: Mariana M. B. Cardoso, Bruss Lima. Formal analysis: Mariana M. B. Cardoso, Bruss Lima, Aniruddha Das. Funding acquisition: Aniruddha Das. Investigation: Mariana M. B. Cardoso, Bruss Lima, Yevgeniy B. Sirotin. Methodology: Mariana M. B. Cardoso, Aniruddha Das. Project administration: Aniruddha Das. Resources: Aniruddha Das. Software: Mariana M. B. Cardoso, Aniruddha Das. Supervision: Aniruddha Das. Validation: Mariana M. B. Cardoso. Writing – original draft: Aniruddha Das. Writing – review & editing: Mariana M. B. Cardoso, Bruss Lima, Yevgeniy B. Sirotin, Anirud- dha Das. References 1. Boynton GM, Engel SA, Glover GH, Heeger DJ. Linear Systems Analysis of Functional Magnetic Reso- nance Imaging in Human V1. J Neurosci 1996; 16(13):4207–21. PMID: 8753882 2. Rees G, Friston KJ, Koch C. A direct quantitative relationship between the functional properties of human and macaque V5. Nat Neurosci 2000; 3(7):716–23. https://doi.org/10.1038/76673 PMID: 10862705 3. Engel SA, Zhang X, Wandell BA. Colour tuning in human visual cortex measured with functional mag- netic resonance imaging. Nature 1997; 388:68–71. https://doi.org/10.1038/40398 PMID: 9214503 4. Buckner RL, Goodman J, Burrock M, Rotte M, Koutstaal W, Schachter DL, et al. Functional-Anatomic Correlates of Object Priming in Humans Revealed by Rapid Presentation Event-Related fMRI. Neuron 1998; 20(2):285–96. PMID: 9491989 5. Boynton GM, Demb JB, Glover GH, Heeger DJ. Neuronal basis of contrast discrimination. Vision Res 1999; 39:257–69. PMID: 10326134 6. Grill-Spector K, Malach R. fMR-adaptation: a tool for studying the functional properties of human cortical neurons. Acta Psychologica 2001; 107(1–3):293–321. PMID: 11388140 7. Logothetis NK, Pauls J, Augath M, Trinath T, Oeltermann A. Neurophysiological investigation of the basis of the fMRI signal. Nature 2001; 412:150–7. https://doi.org/10.1038/35084005 PMID: 11449264 8. Mukamel R, Gelbard H, Arieli A, Hasson U, Fried I, Malach R. Coupling between neuronal firing, field potentials and fMRI in human auditory cortex. Science 2005; 309:951–4. https://doi.org/10.1126/ science.1110913 PMID: 16081741 9. Gardner JL, Sun P, Waggoner RA, Ueno K, Tanaka K, Cheng K. Contrast adaptation and representa- tion in human early visual cortex. Neuron 2005; 47(4):607–20. https://doi.org/10.1016/j.neuron.2005. 07.016 PMID: 16102542 10. Ress D, Backus BT, Heeger DJ. Activity in primary visual cortex predicts performance in a visual detec- tion task. Nat Neurosci 2000; 3(9):940–5. https://doi.org/10.1038/78856 PMID: 10966626 PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 30 / 34 Modulating task-related hemodynamics 11. Kastner S, Pinsk MA, DeWeerd P, Desimone R, Ungerleider LG. Increased activity in human visual cor- tex during directed attention in the absence of visual stimulation. Neuron 1999; 22:751–61. PMID: 10230795 12. Somers DC, Dale AM, Seiffert AE, Tootell RBH. Functional MRI reveals spatially specific attentional modulation in human primary visual cortex. Proc Natl Acad Sci USA 1999; 96:1663–8. PMID: 9990081 13. Gandhi SP, Heeger DJ, Boynton GM. Spatial attention affects brain activity in human primary visual cor- tex. Proc Natl Acad Sci USA 1999; 96:3314–9. PMID: 10077681 14. Shulman GL, McAvoy MP, Cowan MC, Astafiev SV, Tansy AP, d’Avossa G, et al. Quantitative analysis of attention and detection signals during visual search. J Neurophysiol 2003; 90(5):3384–97. https://doi. org/10.1152/jn.00343.2003 PMID: 12917383 15. Serences JT, Yantis S, Culberson A, Awh E. Preparatory activity in visual cortex indexes distractor sup- pression during covert spatial orienting. J Neurophysiol 2004; 92:3538–45. https://doi.org/10.1152/jn. 00435.2004 PMID: 15254075 16. 17. Jack AI, Shulman GL, Snyder AZ, McAvoy MP, Corbetta M. Separate modulations of human V1 associ- ated with spatial attention and task structure. Neuron 2006; 51:135–47. https://doi.org/10.1016/j. neuron.2006.06.003 PMID: 16815338 Luck SJ, Chelazzi L, Hillyard SA, Desimone R. Neural mechanisms of spatial selective attention in areas V1, V2 and V4 of Macaque visual cortex. J Neurophysiol 1997; 77:24–42. https://doi.org/10.1152/ jn.1997.77.1.24 PMID: 9120566 18. McAdams CJ, Maunsell JHR. Effects of attention on orientation-tuning functions of single neurons in macaque cortical area V4. J Neurosci 1999; 19(1):431–41. PMID: 9870971 19. McAdams CJ, Reid RC. Attention modulates the responses of simple cells in monkey primary visual cortex. J Neurosci 2005; 25(47):11023–33. https://doi.org/10.1523/JNEUROSCI.2904-05.2005 PMID: 16306415 20. Motter BC. Focal Attention Produces Spatially Selective Processing in Visual Cortical Areas V1, V2, and V4 in the Presence of Competing Stimuli. J Neurophysiol 1993; 70(3):909–19. https://doi.org/10. 1152/jn.1993.70.3.909 PMID: 8229178 21. Roelfsema PR, Lamme VAF, Spekreijse H. Object-based attention in the primary visual cortex of the macaque monkey. Nature 1998; 395:376–81. https://doi.org/10.1038/26475 PMID: 9759726 22. Vidyasagar TR. Gating of neuronal responses in macaque primary visual cortex by an attentional spot- light. Neuroreport 1998; 9(9):1947–52. PMID: 9674572 23. Ito M, Gilbert CD. Attention modulates contextual influences in the primary visual cortex of alert mon- keys. Neuron 1999; 22:593–604. PMID: 10197538 24. Grunewald A, Bradley DC, Andersen RA. Neural correlates of structure-from-motion perception in macaque V1 and MT. J Neurosci 2002; 22(14):6195–207. PMID: 12122078 25. Marcus DS, Van Essen DC. Scene Segmentation and Attention in Primate Cortical Areas V1 and V2. J Neurophysiol 2002; 88:2648–58. https://doi.org/10.1152/jn.00916.2001 PMID: 12424300 26. Cohen MR, Maunsell JHR. Using Neuronal Populations to Study the Mechanisms Underlying Spatial and Feature Attention. Neuron 2011; 70(6):1192–204. https://doi.org/10.1016/j.neuron.2011.04.029 PMID: 21689604 27. Sylvester CM, Shulman GL, Jack AI, Corbetta M. Asymmetry of Anticipatory Activity in Visual Cortex Predicts the Locus of Attention and Perception. J Neurosci 2007; 27(52):14424–33. https://doi.org/10. 1523/JNEUROSCI.3759-07.2007 PMID: 18160650 28. Ghose GM, Maunsell JHR. Attentional modulation in visual cortex depends on task timing. Nature 2002; 419:616–20. https://doi.org/10.1038/nature01057 PMID: 12374979 29. Sylvester CM, Jack AI, Corbetta M, Shulman GL. Anticipatory Suppression of Nonattended Locations in Visual Cortex Marks Target Location and Predicts Perception. J Neurosci 2008; 28(26):6549–56. https://doi.org/10.1523/JNEUROSCI.0275-08.2008 PMID: 18579728 30. Kastner S, DeWeerd P, Desimone R, Ungerleider LG. Mechanisms of directed attention in the human extrastriate cortex as revealed by functional MRI. Science 1998; 282:108–11. PMID: 9756472 31. Sirotin YB, Das A. Anticipatory haemodynamic signals in sensory cortex not predicted by local neuronal activity. Nature 2009; 457:475–9. https://doi.org/10.1038/nature07664 PMID: 19158795 32. Sirotin YB, Cardoso MMB, Lima BR, Das A. Spatial homogeneity and task-synchrony of the trial-related hemodynamic signal. NeuroImage 2012; 59:2783–97. https://doi.org/10.1016/j.neuroimage.2011.10. 019 PMID: 22036678 33. Swallow KM, Makovski T, Jiang YV. Selection of events in time enhances activity throughout early visual cortex. J Neurophysiol 2012; 108(12):3239–52. https://doi.org/10.1152/jn.00472.2012 PMID: 22993261 PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 31 / 34 Modulating task-related hemodynamics 34. Albers AM, Meindertsma T, Toni I, de Lange FP. Decoupling of BOLD amplitude and pattern classifica- tion of orientation-selective activity in human visual cortex. NeuroImage 2017;pii: S1053-8119(17) 30794-2. https://doi.org/10.1016/j.neuroimage.2017.09.046 PMID: 28951159 35. Griffis JC, Elkhetali AS, Vaden RJ, Visscher KM. Distinct effects of trial-driven and task Set-related con- trol in primary visual cortex. NeuroImage 2015; 120:285–97. https://doi.org/10.1016/j.neuroimage. 2015.07.005 PMID: 26163806 36. Sylvester CM, Shulman GL, Jack AI, Corbetta M. Anticipatory and Stimulus-Evoked Blood Oxygenation Level-Dependent Modulations Related to Spatial Attention Reflect a Common Additive Signal. J Neu- rosci 2009; 29(34):10671–82. https://doi.org/10.1523/JNEUROSCI.1141-09.2009 PMID: 19710319 37. Donner TH, Sagi D, Bonneh YS, Heeger DJ. Opposite Neural Signatures of Motion-Induced Blindness in Human Dorsal and Ventral Visual Cortex. J Neurosci 2008; 28(41):10298–310. https://doi.org/10. 1523/JNEUROSCI.2371-08.2008 PMID: 18842889 38. Silver MA, Ress D, Heeger DJ. Neural correlates of sustained spatial attention in human early visual cortex. J Neurophysiol 2007; 97:229–37. https://doi.org/10.1152/jn.00677.2006 PMID: 16971677 39. Offen S, Gardner JL, Schluppeck D, Heeger DJ. Differential roles for frontal eye fields (FEFs) and intra- parietal sulcus (IPS) in visual working memory and visual attention. J Vis 2010; 10(11):28. https://doi. org/10.1167/10.11.28 PMID: 20884523 40. Liu T, Pestilli F, Carrasco M. Transient Attention Enhances Perceptual Performance and fMRI Response in Human Visual Cortex. Neuron 2005; 45(3):469–77. https://doi.org/10.1016/j.neuron.2004. 12.039 PMID: 15694332 41. Pestilli F, Carrasco M, Heeger DJ, Gardner JL. Attentional enhancement via selection and pooling of early sensory responses in human visual cortex. Neuron 2011; 72(5):832–46. https://doi.org/10.1016/j. neuron.2011.09.025 PMID: 22153378 42. Bonhoeffer T, Grinvald A. Optical imaging based on intrinsic signals: The methodology. In: Toga AW, Mazziotta JC, editors. Brain Mapping: The Methods. San Diego: Academic Press; 1996. p. 55–97. 43. Shtoyerman E, Arieli A, Slovin H, Vanzetta I, Grinvald A. Long-term optical imaging and spectroscopy reveal mechanisms underlying the intrinsic signal and stability of cortical maps in V1 of behaving mon- keys. J Neurosci 2000; 20(21):8111–21. PMID: 11050133 44. Cardoso MMB, Sirotin YB, Lima BR, Glushenkova E, Das A. The Neuroimaging Signal is a Linear Sum of Neurally Distinct Stimulus- and Task-Related Components. Nat Neurosci 2012; 15(9):1298–306. https://doi.org/10.1038/nn.3170 PMID: 22842146 45. 46. Lima BR, Cardoso MMB, Sirotin YB, Das A. Stimulus-Related Neuroimaging in Task-Engaged Subjects Is Best Predicted by Concurrent Spiking. J Neurosci 2014; 34(42):13878–91. https://doi.org/10.1523/ JNEUROSCI.1595-14.2014 PMID: 25319685 Tursky B, Shapiro D, Crider A, Kahneman D. Pupillary, heart rate, and skin resistance changes during a mental task. J Exp Psych 1969; 79(1):164–7. 47. Pleger B, Blankenburg F, Ruff CC, Driver J, Dolan RJ. Reward Facilitates Tactile Judgments and Modu- lates Hemodynamic Responses in Human Primary Somatosensory Cortex. J Neurosci 2008; 28 (33):8161–8. https://doi.org/10.1523/JNEUROSCI.1093-08.2008 PMID: 18701678 48. Oken BS, Salinsky MC, Elsas SM. Vigilance, alertness, or sustained attention: physiological basis and measurement. Clin Neurophysiol 2006; 117(9):1885–901. https://doi.org/10.1016/j.clinph.2006.01.017 PMID: 16581292 49. Abe T, Nonomura T, Komada Y, Asaoka S, Sasai T, Ueno A, et al. Detecting deteriorated vigilance using percentage of eyelid closure time during behavioral maintenance of wakefulness tests. Int J Psy- chophysiol 2011; 82(3):269–74. https://doi.org/10.1016/j.ijpsycho.2011.09.012 PMID: 21978525 50. Esterman M, Noonan SK, Rosenberg MD, DeGutis J. In the zone or zoning out? Tracking behavioral and neural fluctuations during sustained attention. Cereb Cortex 2013; 23(11):2712–23. https://doi.org/ 10.1093/cercor/bhs261 PMID: 22941724 51. Malonek D, Grinvald A. Interactions Between Electrical Activity and Cortical Microcirculation Revealed by Imaging Spectroscopy: Implications for Functional Brain Mapping. Science 1996; 272:551–4. PMID: 8614805 52. Fukuda M, Moon C-H, Wang P, Kim S-G. Mapping iso-orientation columns by contrast agent-enhanced functional magnetic resonance imaging: reproducibility, specificity, and evaluation by optical imaging of intrinsic signal. J Neurosci 2006; 26(46):11821–32. https://doi.org/10.1523/JNEUROSCI.3098-06.2006 PMID: 17108155 53. Kennerley AJ, Berwick J, Martindale J, Johnston D, Zheng Y, Mayhew JEW. Refinement of optical imaging spectroscopy algorithms using concurrent BOLD and CBV fMRI. NeuroImage 2009; 47 (4):1608–19. https://doi.org/10.1016/j.neuroimage.2009.05.092 PMID: 19505581 PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 32 / 34 Modulating task-related hemodynamics 54. Sirotin YB, Hillman EMC, Bordier C, Das A. Spatiotemporal precision and hemodynamic mechanism of optical point-spreads in alert primates. Proc Natl Acad Sci USA 2009; 106(43):18390–5. https://doi.org/ 10.1073/pnas.0905509106 PMID: 19828443 55. Herman MC, Cardoso MMB, Lima BR, Sirotin YB, Das A. Simultaneously estimating the task-related and stimulus-evoked components of hemodynamic imaging measurements. Neurophotonics 2017; 4 (3):031223. https://doi.org/10.1117/1.NPh.4.3.031223 PMID: 28721355 56. Destexhe A, Contreras D, Steriade M. Spatiotemporal Analysis of Local Field Potentials and Unit Dis- charges in Cat Cerebral Cortex during Natural Wake and Sleep States. J Neurosci 1999; 19(11):4595– 608. PMID: 10341257 57. Chang C, Leopold DA, Scho¨lvinck ML, Mandelkow H, Picchioni D, Liu X, et al. Tracking brain arousal fluctuations with fMRI. Proc Natl Acad Sci U S A 2016; 113(16):4518–23. https://doi.org/10.1073/pnas. 1520613113 PMID: 27051064 58. Carrington MJ, Barbieri R, Colrain IM, Crowley KE, Kim Y, Trinder J. Changes in cardiovascular function during the sleep onset period in young adults. J Appl Physiol 2005; 98(2):468–76. https://doi.org/10. 1152/japplphysiol.00702.2004 PMID: 15448124 59. Huo BX, Greene SE, Drew PJ. Venous cerebral blood volume increase during voluntary locomotion reflects cardiovascular changes. NeuroImage 2015; 118:301–12. https://doi.org/10.1016/j.neuroimage. 2015.06.011 PMID: 26057593 60. Hubel DH. Single unit activity in striate cortex of unrestrained cats. J Physiol 1959; 147:226–38. PMID: 14403678 61. Dale AM. Optimal experimental design for event-related fMRI. Hum Brain Mapp 1999; 8:109–14. PMID: 10524601 62. Bergel A, Deffieux T, Demene´ C, Tanter M, Cohen I. Local hippocampal fast gamma rhythms precede brain-wide hyperemic patterns during spontaneous rodent REM sleep. Nat Commun 2018; 9(1):5364. https://doi.org/10.1038/s41467-018-07752-3 PMID: 30560939 63. Esterman M, Poole V, Liu G, DeGutis J. Modulating Reward Induces Differential Neurocognitive Approaches to Sustained Attention. Cereb Cortex 2017; 27(8):4022–32. https://doi.org/10.1093/cercor/ bhw214 PMID: 27473320 64. Kleinschmidt A, Muller NG. The blind, the lame, and the poor signals of brain function—a comment on Sirotin and Das (2009). NeuroImage 2010; 50(2):622–5. https://doi.org/10.1016/j.neuroimage.2009.12. 075 PMID: 20044008 65. Das A, Sirotin YB. What could underlie the trial-related signal? A response to the commentaries by Drs. Kleinschmidt and Muller, and Drs. Handwerker and Bandettini. NeuroImage 2011; 55:1413–8. https:// doi.org/10.1016/j.neuroimage.2010.07.005 PMID: 20637876 66. Pisauro MA, Benucci A, Carandini M. Local and global contributions to hemodynamic activity in mouse cortex. J Neurophysiol 2016, 115:2931–2936. https://doi.org/10.1152/jn.00125.2016 PMID: 26984421 67. Lecrux C, Hamel E. Neuronal networks and mediators of cortical neurovascular coupling responses in normal and altered brain states. Phil Trans R Soc Lond B 2016; 371:20150350. 68. Sato A, Sato Y, Uchida S. Activation of the intracerebral cholinergic nerve fibers originating in the basal forebrain increases regional cerebral blood flow in the rat’s cortex and hippocampus. Neurosci Lett 2004; 361(1–3):90–3. https://doi.org/10.1016/j.neulet.2004.01.004 PMID: 15135901 69. Vaucher E, Hamel E. Cholinergic basal forebrain neurons project to cortical microvessels in the rat: electron microscopic study with anterogradely transported Phaseolus vulgaris leucoagglutinin and cho- line acetyltransferase immunocytochemistry. J Neurosci 1995; 15(11):7427–41. PMID: 7472495 70. Raichle ME, Hartman BK, Eichling JO, Sharpe LG. Central noradrenergic regulation of cerebral blood flow and vascular permeability. Proc Natl Acad Sci USA 1975; 72(9):3726–30. PMID: 810805 71. Bekar LK, Wei SS, Nedergaard M. The Locus Coeruleus-Norepinephrine Network Optimizes Coupling of Cerebral Blood Volume with Oxygen Demand. J Cereb Blood Flow Metab 2012; 32(12):2135–45. https://doi.org/10.1038/jcbfm.2012.115 PMID: 22872230 72. Peppiatt CM, Howarth C, Mobbs P, Attwell D. Bidirectional control of CNS capillary diameter by peri- cytes. Nature 2006; 443:700–9. https://doi.org/10.1038/nature05193 PMID: 17036005 73. 74. Tan CO. Anticipatory Changes in Regional Cerebral Hemodynamics: A New Role for Dopamine? J Neurophysiol 2009; 101:2738–40. https://doi.org/10.1152/jn.00141.2009 PMID: 19321643 Zhang F, Xu S, Iadecola C. Role of nitric oxide and acetylcholine in neocortical hyperemia elicited by basal forebrain stimulation: evidence for an involvement of endothelial nitric oxide. Neuroscience 1995; 69(4):1195–204. PMID: 8848107 75. Choi J-K, Chen YI, Hamel E, Jenkins BG. Brain hemodynamic changes mediated by dopamine recep- tors: Role of the cerebral microvasculature in dopamine-mediated neurovascular coupling. NeuroImage 2006; 30:700–12. https://doi.org/10.1016/j.neuroimage.2005.10.029 PMID: 16459104 PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 33 / 34 Modulating task-related hemodynamics 76. Krimer LS, Muly EC, Williams GV, Goldman-Rakic PS. Dopaminergic regulation of cerebral cortical microcirculation. Nat Neurosci 1998; 1(4):286–9. https://doi.org/10.1038/1099 PMID: 10195161 77. Winder AT, Echagarruga C, Zhag Q, Drew PJ. Weak correlations between hemodynamic signals and ongoing neural activity during the resting state. Nat Neurosci 2017; 20(12):1761–9. https://doi.org/10. 1038/s41593-017-0007-y PMID: 29184204 78. Drew PJ, Shih AY, Kleinfeld D. Fluctuating and sensory-induced vasodynamics in rodent cortex extend arteriole capacity. Proc Natl Acad Sci USA 2011; 108(20):8473–8. https://doi.org/10.1073/pnas. 1100428108 PMID: 21536897 79. Matsuda K, Nagami T, Kawano K, Yamane S. A new system for measuring eye position on a personal computer [abstract]. Abstracts, Society for Neuroscience 26, 744.2. 2000. 80. Arieli A, Grinvald A, Slovin H. Dural substitute for long-term imaging of cortical activity in behaving mon- keys and its clinical implications. J Neurosci Methods 2002; 114:119–33. PMID: 11856563 81. Arieli A, Grinvald A. Optical imaging combined with targeted electrical recordings, microstimulation or tracer injections. J Neurosci Methods 2002; 116:15–28. PMID: 12007980 82. Kohl M, Lindauer U, Royl G, Kuhl M, Gold L, Villringer A, et al. Physical model for the spectroscopic analysis of cortical intrinsic optical signals. Phys Med Biol 2000; 45:3749–64. PMID: 11131197 83. Hillman EMC, Devor A, Bouchard MB, Dunn AK, Krauss GW, Skoch J, et al. Depth-resolved optical imaging and microscopy of vascular compartment dynamics during somatosensory stimulation. Neuro- Image 2007, 35:89–104. https://doi.org/10.1016/j.neuroimage.2006.11.032 PMID: 17222567 84. Kalatsky VA, Stryker MP. New paradigm for optical imaging: temporally encoded maps of intrinsic sig- nals. Neuron 2003; 38:529–45. PMID: 12765606 85. Ratzlaff EH, Grinvald A. A tandem-lens epifluorescence macroscope: hundred-fold brightness advan- tage for wide-field imaging. J Neurosci Methods 1991; 36:127–37. PMID: 1905769 86. Madsen MT. A simplified formulation of the gamma variate function. Phys Med Biol 1992; 37(7):1597. 87. Cohen MS. Parametric Analysis of fMRI Data Using Linear Systems Methods. NeuroImage 1997; 6:93– 103. https://doi.org/10.1006/nimg.1997.0278 PMID: 9299383 88. Tada H, Omori Y, Hirokawa Y, Hirokawa K, Ohira H, Tomonaga M. Eye-Blink Behaviors in 71 Species of Primates. PLoS ONE 2013; 8(5):e66018. https://doi.org/10.1371/journal.pone.0066018 PMID: 23741522 89. Guipponi O, Odouard S, Pinède S, Wardak C, Ben Hamed S. fMRI Cortical Correlates of Spontaneous Eye Blinks in the Nonhuman Primate. Cereb Cortex 2015; 25(9):2333–45. https://doi.org/10.1093/ cercor/bhu038 PMID: 24654257 PLOS Biology | https://doi.org/10.1371/journal.pbio.3000080 April 19, 2019 34 / 34
Could not heal snippet
10.1021_acssynbio.1c00142.pdf
null
null
pubs.acs.org/synthbio Research Article Sequence Preference and Initiator Promiscuity for De Novo DNA Synthesis by Terminal Deoxynucleotidyl Transferase Erika Schaudy, Jory Lietard, and Mark M. Somoza* Cite This: ACS Synth. Biol. 2021, 10, 1750−1760 Read Online ACCESS Metrics & More Article Recommendations *sı Supporting Information ABSTRACT: The untemplated activity of terminal deoxynucleotidyl transferase (TdT) represents its most appealing feature. Its use is well established in applications aiming for extension of a DNA initiator strand, but a more recent focus points to its potential in enzymatic de novo synthesis of DNA. Whereas its low substrate specificity for nucleoside triphosphates has been studied extensively, here we interrogate how the activity of TdT is modulated by the nature of length, chemistry, and the initiating strands, in particular nucleotide composition. Investigation of full permutational libraries of mono- to pentamers of D-DNA, L-DNA, and 2′O-methyl-RNA of differing directionality immobilized to glass surfaces, and generated via the efficiency of photolithographic in situ synthesis, shows that extension strongly depends on the nucleobase sequence. We also show TdT being catalytically active on a non-nucleosidic substrate, hexaethylene glycol. These results offer new perspectives on constraints and strategies for de novo synthesis of DNA using TdT regarding the requirements for initiation of enzymatic generation of DNA. KEYWORDS: TdT polymerase, microarray, synthetic biology, L-DNA, enzymatic DNA synthesis, photolithographic synthesis their T erminal deoxynucleotidyl transferase (TdT) is a member of the polX family of DNA polymerases first purified from calf thymus glands.1,2 In contrast to template-dependent DNA polymerases, TdT extends DNA strands at their 3′ hydroxy terminus in the presence of divalent cation cofactors3 and deoxynucleoside triphosphates (dNTPs), but in the absence of a template strand. This activity is of major importance in the diversification of immunoglobulins and T cell receptors in the process of V(D)J recombination of the adaptive immune system via random addition of nucleotides to nicked DNA strands.4,5 TdT’s unique ability to mediate template- independent polymerization has made it a valuable tool in a variety of molecular biology applications including finding strand breaks,6 modifying DNA oligomers with various NTPs,7 and identifying DNA damage and epigenetic modifications.8 Furthermore, the enzyme has proven useful for the generation of polynucleotides of high molecular weight9 and amphiphilic structures upon extension with BODIPY-dUTP,10 for detection of DNA and RNA on surfaces,11,12 and immobiliza- tion of DNA on solid supports.13 In the context of synthetic biology, template-independent DNA polymerization by TdT is, along with enzyme-based approaches,14 a promising alternative to chemical synthesis as many of the shortcomings of the phosphoramidite approach can be potentially avoided. In particular, coupling failures and depurination during the deblocking step limit chemical to about 200 nucleotides. The atom economy of phosphoramidite synthesis synthesis of DNA is also very poor, producing an approximately 1000- fold excess of chemical waste. Since polymerases work in aqueous solutions and are capable of fast and high-fidelity synthesis of almost arbitrary length, they promise a greener and far more efficient approach to DNA synthesis. Beyond genomics and biotechnological applications, DNA is an attractive medium for archiving digital information since it can achieve a storage density of hundreds of petabytes per gram,15 and data can be reliably recovered after being stored for thousands of years.16 Useful DNA data storage may depend on successful implementation of enzymatic synthesis since even high throughput chemical approaches are economically uncompetitive with, storage technologies.17 e.g., magnetic or optical Several recent publications have addressed sequence control in TdT-based enzymatic synthesis. In the context of digital information storage, a looser definition of sequence control can be tolerated, allowing dNTP degradation with apyrase to limit TdT-catalyzed extension to a controlled series of short Received: April 6, 2021 Published: June 22, 2021 © 2021 The Authors. Published by American Chemical Society 1750 https://doi.org/10.1021/acssynbio.1c00142 ACS Synth. Biol. 2021, 10, 1750−1760 ACS Synthetic Biology pubs.acs.org/synthbio Research Article homopolymers.18 Precise sequence control has been achieved using photocleavable TdT-dNTP conjugates,19,20 3′ photoc- aged dNTPs,21 and through the controlled release of divalent ion cofactors from photosensitive chelators.22,23 While it is too soon to tell which approach to sequence-controlled de novo DNA synthesis will be optimal, here we explore another factor critical to practical and efficient enzymatic synthesis with TdT, its initiator preferences. Experiments demonstrating TdT- based synthesis have used relatively long primer DNA oligonucleotides20 to 60mersas starting substrates, an impractically large number since these are made chemically and remain attached.18,20,21,23 In the phosphoramidite chemistry approach it is standard practice to start with one of four solid-phase columns preloaded with the first DNA nucleoside of the desired sequence. Such an approach might also be feasible in enzymatic synthesis if the initiator sequence length can be limited to one or two nucleotides, resulting respectively in 4 or 16 starting sequences or columns. This seems possible since very early research on TdT suggests a lower limit in length of the initiating DNA strand of at least 3 nt24 or as low as 2 nt.25 At the same time, we should ask whether some sequences are extended more efficiently than others, as this affects not just the initiation, but potentially each subsequent cycle of the synthesis. for A related question is whether TdT is able to extend initiator molecules other than the 3′ terminus of DNA, enabling enzymatic synthesis of chimeric nucleic acid sequences, DNA/ RNA hybrids, or even conjugates where an unnatural initiator is extended with dNTPs or rNTPs. Regarding the differences in efficiency in the use of dNTPs and rNTPs, there appears to be limited ability for extension of DNA initiator strands with ribonucleotides.26,27 Furthermore, TdT was found to catalyze the extension of oligonucleotide strands with a variety of modified nucleoside triphosphates, instance biotiny- lated,11,28,29 fluorescence-tagged,30 photo-cross-linkable31 or light-cleavable21 dNTPs and non-nucleosidic substrates,32 as well as fluorescent nucleobase analogues33 and metal base- pairs,34 showing rather low substrate specificity in contrast to other DNA polymerases, which could be further loosened by protein engineering efforts.35 An investigation of nucleoside triphosphate analogues, including arabinonucleosides and acyclic triphosphates of acyclovir and penciclovir, and their L- and D-stereoisomers showed that the stereochemistry of the triphosphates had a profound effect on substrate recognition by TdT.36 Whereas nucleoside triphosphate substrate specific- ity is rather flexible, DNA analogues in the initiating strand seem to hamper extension, for instance upon replacement of natural DNA nucleotides at the 3′ terminus with L-DNA,37 or when using RNA initiator strands.38,39 Herein, we report on the ability of TdT to extend ssDNA initiators between 1 and 5 nt in length and immobilized on a glass surface, as well as other nucleosidic and non-nucleosidic primers. Our the enzymatic results, which encompass extension of all 1364 possible sequence permutations of mono- up to pentamers for each of several nucleic acid chemistries, are based on the use of nucleic acid photo- lithography for the massively parallel synthesis of initiator strands on a common surface.40 We have recently expanded the toolbox of light-sensitive DNA phosphoramidites used in photolithographic synthesis beyond the standard 3′ → 5′ (“forward”) direction,41 and we are using this chemical diversity to investigate the activity of the TdT polymerase on from DNA oligonucleotides with a variety of initiators, accessible 3′ or 5′-OH groups (from “reverse” or “forward” DNA synthesis, respectively), to RNA-like nucleic acids with 2′O-methyl RNA (2′OMe-RNA), to mirror-image (L-)DNA primer strands with a terminal 5′-OH. We also examined the potential of non-nucleosidic molecules to act as initiators for TdT-mediated enzymatic synthesis by preparing polymers of hexaethylene glycol (HEG) linkers. Surprisingly, with the exception of 5′-OH D-DNA, all tested substrates were able to support some level of enzymatic extension, but with 3′ hydroxy terminated DNA clearly the optimal initiator. The extension efficiency of 3′ hydroxy terminated ssDNA by TdT is also strongly sequence dependent, with a factor of 3 efficiency difference between the best and worse pentamer initiator sequences. ■ MATERIALS AND METHODS Approach. In order to investigate the ability of TdT to extend terminal hydroxy groups of different nucleic acid chemistries, multiple replicates of each oligonucleotide strand were synthesized on the same array, each present in two versions: one where the final light-sensitive protecting group was removed at the end of the synthesis, exposing an accessible hydroxy group, whereas in the other version, the terminal hydroxy group was capped with a DMTr-dT phosphorami- dite.42 We have previously measured the coupling efficiency of most non-RNA phosphoramidites for light-directed synthesis to be ∼99.9%, including DMTr-dT in its role as capping agent; G being the exception at 97−98%.41,43−46 After synthesis and deprotection, the surface-bound oligonucleotides serve as initiator sequences for dT homopolymer extension with TdT polymerase. The efficiency of polymerization was evaluated by hybridization to the extension product. Absolute fluorescent signal intensities of the capped and uncapped versions present on a single surface were compared in order to evaluate the ability of TdT to extend short oligonucleotide strands of differing chemistry and nucleotide composition. In order to allow investigation of all different monomers as initiators, and to distance the terminal hydroxy group from the glass surface, the synthesis was started with coupling of a hexaethylene glycol phosphoramidite as linker in an initial synthesis cycle. to the glass Photolithographic in Situ Synthesis. The detailed procedure for photolithographic in situ synthesis has already been described elsewhere.47,48 Briefly, microscopy glass slides (Schott NEXTERION glass D) were functionalized with a 2% N-(3-triethoxysilylpropyl)-4-hydroxybutyramide (95%; abcr) solution in ethanol/water/acetic acid (95:5:0.1), washed, and cured at 120 °C under a vacuum for 2 h. An Expedite 8909 nucleic acid synthesizer was used to deliver reagents for substrate. Anhydrous acetonitrile synthesis (Biosolve) and DCI activator (Sigma-Aldrich, L032000) were maintained dry under molecular sieves (Sigma-Aldrich, Z509027). The exposure solvent consisted of 1% imidazole (Sigma-Aldrich, 56750) in anhydrous DMSO (Biosolve). The oxidizer was 20 mM I2 in H2O/pyridine/THF (Sigma-Aldrich L060060). Cyanoethyl phosphoramidites were used as 0.03 M solutions in dry acetonitrile and obtained from Orgentis (5′- BzNPPOC D-DNA, 3′-BzNPPOC D-DNA), ChemGenes (5′- NPPOC L-DNA; 3′-NPPOC 2′OMe-RNA; NPPOC-hexa- ethylene glycol), and LINK (DMTr-dT). Phosphoramidite purity and 3′ phosphitylation selectivity was verified by 31P and 2D 1H−31P NMR. Coupling times varied depending on the type of phosphoramidite, between 15 s (D-DNA), 60 s (L-DNA and 2′OMe-RNA), 120 s (DMTr-dT), and 300 s (hexa- 1751 https://doi.org/10.1021/acssynbio.1c00142 ACS Synth. Biol. 2021, 10, 1750−1760 ACS Synthetic Biology pubs.acs.org/synthbio Research Article ethylene glycol). After synthesis, cyanoethyl and base protecting groups were removed by treating the array with ethylenediamine/ethanol (1:1) for either 2 or 15 h (3′- BzNPPOC D-DNA, NPPOC-hexaethylene glycol). An optical system, focusing UV light from a 365 nm high- power UV-LED source (Nichia NVSU333A)49 onto a digital micromirror device (Texas Instruments 0.7 XGA DMD) with 1024 × 768 individually addressable micromirrors, and via an Offner optical relay, further onto a functionalized glass slide, allows the photosensitive protecting groups according to a set of digital masks generated by a MATLAB program. spatially resolved removal of Synthesis Design. Oligonucleotide microarrays used in this study are based on the same layout and design. Using the full 1024 × 768 synthesis space, a 9:25 layout (blocks of 3 × 3 synthesis pixels surrounded by 2 pixel-wide unused margins) allowed for photolithographic synthesis of 31 008 individual sequences in parallel. The full permutation library of 1 to 5 nt length was synthesized with both free and capped terminal hydroxy groups. A 25mer (“QC25”: 5′-GTCATCATCATG- AACCACCCTGGTC-3′) was synthesized in parallel in order the synthesis quality via a to allow for evaluation of standardized hybridization. Furthermore, synthesis of T or U 18mers enabled the hybridization efficiency to be assessed during the detection of enzymatically generated dT homopol- ymers. All strands were grown on a single hexaethylene glycol (HEG) moiety as a non-nucleotide linker. Distribution of the sequencesand all replicates of individual sequenceson the array surface was randomized in order to compensate for any spatial effects possibly occurring upon reaction and/or hybridization. The microarrays for the investigation of non- nucleotide initiator strands were synthesized using only HEG phosphoramidites in order to obtain strands of up to nine HEG units in length, both with accessible and blocked termini. Extension and Detection. After removal of cyanoethyl and nucleobase protecting groups, extension reactions were performed with a mix of 0.2 u/μL calf thymus TdT (NEB M0315; 20 u/μL stock) and 100 μM dTTP (Carl Roth; 100 mM stock) in 1× TdT buffer (NEB; 50 mM potassium acetate, 20 mM tris-acetate, 10 mM magnesium acetate, pH 7.9 at 25 °C) supplemented with 0.25 mM CoCl2 (NEB; 2.5 mM stock) at 37 °C in a hybridization oven with rotation for 120 min in an adhesive chamber (Grace Biolabs). After incubation, the reaction mix was removed from the hybridization chamber and the array rinsed briefly by pipetting in and out nonstringent washing buffer (NSWB) (6× SSPE, 0.01% Tween-20), followed by a short wash (ca. 10 s) of the entire slide in final washing buffer (FWB) (0.1× SSC) and drying in a microarray centrifuge. A hybridization solution containing probe rA18-Cy3 (IDT; 5′-Cy3-GDDDD(rA)18-3′; with D being either A,G,T; 90 nM) and acetylated BSA (Promega; 0.44 mg/mL) in 1× MES buffer (100 mM MES, 1 M Na+, 20 mM EDTA, 0.01% Tween-20) was applied to the array surface for incubation at 4 °C without rotation for 120 min. Stringency washes were performed by washing the slide for 2 min in NSWB, 1 min in stringent washing buffer (SWB) (100 mM MES, 0.1 M Na+, 0.01% Tween-20) and 10 s in FWB at 4 °C. After drying, the slides were scanned at 532 nm at a resolution of 5 μm using a GenePix Personal 4100A scanner. Data Analysis. The Cy3 fluorescent signal intensities observed upon hybridization to the enzymatically generated homopolymer served as a measure of successful extension of initiator strands. Alignment of the scans with the underlying design using NimbleScan 2.1.68 (NimbleGen) allowed for data extraction for each individual feature. The data were analyzed using Microsoft Excel. Fluorescent signal intensities observed on features with blocked termini were treated as background noise and subtracted from the signal measured for the version with an accessible terminal hydroxy group. Sequence logos were created using WebLogo (weblogo.berkeley.edu).50 ■ RESULTS AND DISCUSSION The ability of TdT to extend all possible mono- to pentamers of nucleotide chains with differing sugar chemistries was investigated via hybridization to the product of extension. This setup allowed not only for a comparison of the extension efficiency of different chemistries, but also for the identification of preferences in nucleotide composition as well as the minimal length still allowing for enzymatic polymerization. Besides nucleic acid pentamers, we also prepared polymers of hexaethylene glycol (HEG) containing up to nine units. Due to the uncontrolled mode of action of TdT randomly adding nucleoside triphosphates to the growing chain, we restricted our study to only dTTP as a substrate in order to generate poly dT strands detectable in a hybridization-based assay with a fluorescently labeled complementary rA18 probe, as shown in Figure 1. Synthesis only using a single type of dNTP allows us to isolate the impact of the initiating sequence on extension efficiency from biases in the incorporation of dNTPs that have been observed in vivo51 and in vitro.19,24 The analysis of fluorescent signal intensities measured upon hybridization to the enzymatic reaction products allowed for extension efficiencies to be compared. Figure 2 shows the intensities observed for the extension of range of signal initiators of differing nucleic acid chemistries, with lowest and highest fluorescent intensities detected and after background subtraction (sequences with blocked terminal the threshold to evaluate TdT’s hydroxy group). We set general ability to extend an initiator as the average of signal intensities for blocked sequences plus three times their standard deviation. D-DNA 3′-OH Extension. With the cognate substrate of TdT being single-stranded DNA with a 3′-OH terminus, we expected the highest extension efficiency for this substrate. Indeed, signal intensities plotted in Figure 2 clearly show 3′- terminated D-DNA as the favored substrate of all five different chemistries. Focusing on the left panel of Figure 2, the extension reaction efficiency increases with the length of the initiating strands. However, there is also a clear dependence on the nucleotide composition, to the extent that some sequences of longer oligonucleotides can be less efficient initiators than the shortest. signal Investigation of the full permutation library allowed us to identify which nucleotide sequences are preferentially ex- tended. Figure 3a provides an overview of the trends of initiating sequences yielding the 10% highest (left, framed in framed in red) extension green) and lowest (right panel, efficiency. Consensus sequence logos illustrate these trends. While the nucleobase sequence is less relevant for mono- and dimers, for tetra- and pentamers extension is least efficient in the case of a deoxycytidine in the 3′ terminal position, with the lowest signal detected for the sequences TAGAC and GATC (all sequences 5′ → 3′). In the case of trimers, the two isolated data points at the top end of the range correspond to the sequences GGG and CGG, emphasizing the preference for G in the terminal positions of efficient initiators of this length. 1752 https://doi.org/10.1021/acssynbio.1c00142 ACS Synth. Biol. 2021, 10, 1750−1760 ACS Synthetic Biology pubs.acs.org/synthbio Research Article Figure 2. Fluorescent signal intensities after background subtraction for five different types of initiating strands. The structure of the corresponding dimer (monomer for HEG), immobilized to the surface is illustrated. For each type and length of oligonucleotide strands, the 0th, 25th, 75th, and 100th intensity percentiles are shown, based on all possible 4n data points for each initiator strand of length n. Hexaethylene glycol n-mers are plotted with dots. The greyed-out insert for 5′-OH D-DNA, 5′-OH L-DNA, and HEG shows this lower range of signal intensity in more detail. The two data points at the low end of the sigmoidal curve represent results for ACC and TGC, with the corresponding consensus sequence again clearly showing that a terminal C is not favored for extension. Investigation of the effect of strand length is shown in Figure 3b, where the average signal intensity of all sequences of a specific length are normalized to the average signal Independent of intensity of monomers. nucleotide composition, the results show that the efficiency initiation increases with strand length, with pentamers of facilitatingon average2.2× higher initiation efficiency compared to monomers. Applying a second order polynomial fit as guide suggests elongation efficiency asymptotically approaches a maximum, hinting that increasing initiating strand length further may not significantly improve average efficiency. Still, the wide range of signal intensities detected for each length emphasizes even more the impact of nucleotide composition. 2′OMe-RNA 3′-OH Extension. Investigation of enzymatic extension of short strands of 2′OMe-RNA with a terminal 3′ hydroxy group synthesized on the surface clearly shows that TdT is able to use it as a substrate, albeit at lower efficiency than its D-DNA counterpart. In comparison to 3′-OH D-DNA, TdT exhibits distinct preferences for sequence composition in 2′OMe-RNA initiator strands, as shown in Figure 4a. Indeed, the nucleobase in the terminal position of the strand and the impact on the efficiency of adjacent one have a major extension, with adenine and cytidine nucleotides being favored in the terminal position when next to adenine or guanosine nucleotides. In contrast, both guanosine and uracil nucleotides at the 3′ terminus have a negative impact on the efficiency of strand extension. Of note, we found that extension of a 2′OMe uracil nucleotide is disfavored in almost all cases, including for mono- and dinucleotides. Comparison of the efficiency of initiation based on strand length shows a significant leap from monomers to pentamers (Figure 4b). The second order 1753 https://doi.org/10.1021/acssynbio.1c00142 ACS Synth. Biol. 2021, 10, 1750−1760 Figure 1. Schematic representation of the experimental design and assays. (1) Two variants of all possible permutations of mono- to pentamers, either with accessible terminal hydroxy group (OH) or with DMTr-blocked terminus (×), were synthesized on a glass slide via photolithography. (2) The immobilized initiator strands were then extended enzymatically by TdT using dTTP as substrate, generating dT homopolymers. (3) Poly dT strands were detected via hybridization with a Cy3 labeled complementary probe. (4) Scanning of the microarray allows for fluorescent signal intensities at different positions to be assigned to specific sequences. The scan to the left corresponds to 2.4% of the total synthesis area (scale bar 300 μm). In more detail, the close-up of 16 features (scale bar 100 μm) and the corresponding layout beneath are shown with a grid next to it, indicating the sequences synthesized at specific positions. Features with blocked termini (×) exhibit much lower fluorescence signal intensity than those with strands accessible for extension. TdT model adapted from PDB: 1JMS. ACS Synthetic Biology pubs.acs.org/synthbio Research Article Figure 3. Analysis of extension of 3′-OH D-DNA initiating strands. (a) Fluorescent signal intensities were normalized to the maximum and clustered according to length, with representative SEM error bars. Panels to the left illustrate sequence patterns (5′ → 3′ direction) from the data for the 10% highest signal intensities framed in green, whereas panels to the right show the data for the 10% lowest signal intensities for penta-, tetra-, and trimers (top to bottom) framed in red. Data for pentamers are repeated in gray in the subsequent plots for comparison. For monomers and dimers, data are plotted from highest to lowest signal intensity with the corresponding sequence specified by the labeling of the top and bottom x-axis for dimers and monomers, respectively. Next to this plot, the chemical structure of a dimer immobilized to the glass surface serves as a guide for straightforward identification of differences between the chemical variants tested for initiation in this and subsequent figures. (b) Fluorescent signal intensities normalized to the average of all monomers and clustered according to strand length. Averages for each strand length are indicated by an “×”. The dotted second order polynomial fit through the averages serves as a visual guide. polynomial fit to the average signal of all sequences of a specific length levels off for tetramers and pentamers, suggesting a close-to-maximum efficiency already for initiating strands with five nucleotides in length. In comparison to the D-DNA (3′- OH) substrate, the position of the average signal intensity relative to the range of signal observed for longer initiators is striking. The distribution of data points clearly indicates that most of the sequences are being extended with low efficiency, keeping the average efficiency of initiation of pentamers at a level of approximately 4× compared to monomers, whereas some outstanding variants even show initiating efficiencies of more than 10× that of monomers. D-DNA 5′-OH Extension. In order to investigate if 5′-OH DNA extension is possible, the TdT reaction mix was applied to a microarray populated only with D-DNA tethered to the surface at the 3′ end and with a terminal 5′-OH. In this case, only very low fluorescent signal intensities were detected (see Figure 4. Extension analysis of 2′OMe-RNA initiating strands with terminal 3′-OH. (a) Fluorescent signal intensities were normalized to the maximum signal detected and clustered according to their length, with representative error bars corresponding to 2× SEM for better visibility. Panels to the left illustrate sequence patterns (5′ → 3′) emerging from the data for the 10% highest signal intensities (framed in green), whereas panels to the right show the data for the 10% lowest signal intensities framed in red. Data for pentamers are also shown in the following plots for comparison, pointing to the similarity in shape between graphs for differing strand lengths. For monomers and dimers, data are plotted from highest to lowest signal intensity with the corresponding sequence specified on the labeling of the top and bottom x-axis for dimers and monomers, respectively. Next to this plot, the chemical structure of a dimer immobilized to the glass surface serves as a guide for identification of differences between the chemical variants initiation. (b) Fluorescent signal intensities normalized to the average of all monomers and clustered according to strand length. Averages for each strand length are indicated by “×”. A polynomial fit through the averages serves as a visual guide. tested for Figure 2). The average values for all sequence permutations and for lengths between monomers and pentamers were below the limits of detection (determined using the data for DMTr- capped strands as unextendable controls), indicating that strands of D-DNA with terminal 5′ hydroxy group are not suitable substrates for extension with TdT. L-DNA 5′-OH Extension. Our recent report establishing photolithographic in situ synthesis for mirror-image DNA (L- DNA)46 motivated us to investigate the activity of TdT on this non-natural substrate. Surprisingly, we indeed were able to lower than for 3′-OH detect significant extension, albeit terminated D-DNA (Figure 2). The sigmoidal curves generated to show the distribution of fluorescent signal in order intensities among all sequence permutations of equal length in Figure 5a cover a considerable range, indicating that the sequence of the initiating strand has a critical impact on the 1754 https://doi.org/10.1021/acssynbio.1c00142 ACS Synth. Biol. 2021, 10, 1750−1760 ACS Synthetic Biology pubs.acs.org/synthbio Research Article intensity. Comparing the average signal intensities for each initiator length with one another in Figure 5b once again emphasizes the significant increase in the efficiency of initiation with strand length. A second order polynomial fit to the average serves as a visual guide and suggests a maximum efficiency of initiation for strands approximately five nucleo- tides in length. On average, signal intensities for pentamers are 4.3× higher than for monomers. However, a few isolated data points at the top end of the range show that the efficiency of initiation is strongly influenced by the L-DNA sequence, as initiating efficiencies for individual pentamers can be up to 20× higher than for the monomer average. Hexaethylene Glycol Extension. In order to assess the ability of TdT to act on primary hydroxy groups of non- nucleosidic substrates, microarrays with strands of hexa- ethylene glycol (molecular structure shown in Figure 6a), Figure 5. Analysis of 5′-OH L-DNA strand extension data. (a) Fluorescent signal intensities were normalized to the maximum and grouped by length, with representative error bars corresponding to 2× SEM for better visibility. Panels framed in green illustrate sequence patterns for the 10% highest signal intensities, whereas panels framed in red show the data for the 10% lowest. Pentamer data are repeated in subsequent plots for comparison, pointing to the similarity in shape between graphs for differing strand lengths. For monomers and dimers, data are plotted from highest to lowest signal intensity with the corresponding sequence specified on the labeling of the top and bottom x-axis for dimers and monomers, respectively. Next to this plot, the chemical structure of a dimer on the glass surface serves as a guide for identification of differences between the chemical variants tested for initiation. (b) Fluorescent signal intensities normalized to the average of all monomers and clustered according to strand length. Averages for each strand length are indicated by “×”. The dotted line is a second order polynomial fit through the averages. efficiency of extension. Analysis of nucleotide composition of the L-DNA initiator strands unambiguously shows a strong preference for L-dT at both the 5′-OH terminus and at the adjacent position for TdT extension for all initiator lengths investigated. In contrast, the identities of nucleotides more from the site of extension are mostly irrelevant. distant Interestingly, poorly extended substrates fall into the same low fluorescence regardless of primer length, as range of indicated by overlapping the greyed-out curve for pentamers with data points of tetramers and trimers. Whereas short strands do not allow for considerable extension, with signal intensities for monomers on average being hardly above the limit of detection, thymine is the favored nucleobase even in this context. Exceptionally high signal intensities compared to other sequences of the same length were measured for the pentamers TTAAA, TTAAG, and TTAAT, the tetramers TTAA and TTAT, and the trimers TTT, TTC, and TTA (all 5′ → 3′), as illustrated by their prominent positions as individually discernible data points at maximum signal Figure 6. Extension of hexaethylene glycol strands. (a) Molecular structure of a single HEG unit. (b) Fluorescent signal intensities, normalized to maximum signal detected for dimers, show a decreasing trend with increasing number of HEG units in the initiating strand (error bar representative for SEM). ranging from one to nine units in length, were synthesized. these initiator strands were extended by the Surprisingly, enzyme, with fluorescence signals clearly above the LOD and in the same range as for 2′OMe-RNA and L-DNA (Figure 2). Investigating the dependence of fluorescent signal intensities as a function of initiating strand length hints at shorter strands being extended more efficiently than longer ones (Figure 6b). DNA extension with TdT has been studied for over 60 years, but surprisingly few specific details have been established regarding the initiator preferences of this unique polymerase. Particularly in the context of de novo DNA synthesis, these preferences are crucial to developing a practical and efficient approach competitive with phosphoramidite chemistry. Recent efforts in this field have used initiator strands 20 to 60 nt in length and of heterogeneous nucleobase composition. Although earlier research has shown that short TdT initiators are also functional, a lack of information on the initiator length dependence for TdT polymerization efficiency may have contributed to the choice of very long initiators. The crystal structure of murine TdT indicates that only three nucleotides at the 3′-hydroxy end are ordered within the polymerase, whereas additional ones are outside the polymerase and disordered.52 This along with the 3 nt minimum initiator length indicated by Kato et al.24 suggests that any benefit to longer initiators would be due to more indirect mechanisms 1755 https://doi.org/10.1021/acssynbio.1c00142 ACS Synth. Biol. 2021, 10, 1750−1760 ACS Synthetic Biology pubs.acs.org/synthbio Research Article such as 1D diffusion along the strand facilitating the localization of TdT to the 3′-hydroxy end. While 1D diffusion along DNA has been identified for the T7 RNA polymerase,53 there is no evidence of a similar process for DNA polymerases in the absence of accessory sliding clamp factors.54 Although our data only extends to pentamers, it clearly shows that initiators longer than about 5 nt are unlikely to significantly enhance TdT polymerization efficiency. This is true for TdT’s natural substrate, 3′-hydroxy terminated DNA, for which we observe polymerization efficiency flattening beyond an initiator length of 4 nt (Figure 3b), as well as for the non-natural substrates 2′OMe-RNA and 5′-hydroxy terminated L-DNA (Figures 4b and 5b). On the short end of initiator length, we were able to observe significant polymerization for both monomers and dimers. This observation stands in contrast to the 3 nt lower limit of Kato et al. However, in our experiments these short DNA strands are linked to relatively long hexaethylene glycol strands, which themselves can function as initiator strands. For 2′OMe-RNA, lower efficiencies of TdT extension were expected considering earlier reports of RNA primers not being extended, neither with dNTPs nor with rNTPs.38 Extension of a DNA primer with rNTPs showed an upper limit of 3−4 added nucleotides,26 leading to the hypothesis that the enzyme stops extension as soon as the initiator strand transitions from DNA to RNA. Comparing these reports with our own results, we observe that methylated RNA analogues can indeed be extended. Since a minimum extension length of seven dT nucleotides are necessary to provide a detectable hybridization signal with the rA18-Cy3 probe, our data show that the initiating strand must have been extended by at least seven dT nucleotides. Considering the additional steric hindrance from the 2′-methyl compared to unmodified RNA, the extension of 2′OMe-RNA with bulkier methyl groups suggests a more complex gating mechanism. Since 2′OMe-RNA, HEG, and L- initiating strands, DNA are functional, albeit whereas 5′-OH extension of D-DNA does not occur, the results suggest that TdT has evolved to exclude this last substrate rather than to be highly specific for 3′-OH DNA extension. inefficient Regarding the activity of TdT on mirror-image DNA substrates, only a few reports exist. Already in 1995, Focher et al.55 demonstrated the ability of calf thymus terminal transferase to extend a dT20 primer of D-DNA (with a blocked 5′ terminus) upon addition of L-dTTP. However, extension stopped after 1−2 nt, indicating that this short stretch of L- DNA with a terminal 3′-OH is not a functional initiator. Another study on the extension of a D-DNA primer with a single L-dT incorporation at the 3′ end showed the extension using D-dNTPs is aborted after 1−2 nt. The authors speculated that a distortion of orientation initiated by presence of the L- nucleotide could result in termination of extension.37 However, all these investigations focus on extension of oligonucleotides with a terminal 3′ hydroxy group in solution. In contrast, the present study used L-DNA phosphoramidites in 3′ → 5′ synthesis direction using pure 5′-NPPOC 3′-L phosphorami- dites, resulting in strands immobilized to the surface and with an accessible terminal 5′ hydroxy group. In this context, comparing the results with those for the corresponding 5′-OH D-DNA initiator strands is especially surprising. As shown in the inset in Figure 2, signal intensities for hybridization after applying TdT to L-DNA initiator strands were significantly higher relative to the corresponding 5′-OH D-DNA initiators, which were simply not extended at all, also indicating the absence of D-DNA contamination in the L-DNA building blocks. We surmise that the structural differences between D- and L-DNA play a role in the mirror-image form acting as a potential substrate. The left-handed conformation of L-DNA prevents not only hybridization to D-DNA, but also interaction with L-enzymes in the active center.56,57 Since the structure of D-DNA oligonucleotides with a 5′ terminal hydroxy group did not prove suitable as a substrate for extension, the conforma- tional change to its mirror-image pendant seems to represent the variation required to fit the active center of the polymerase and allow for strand extension, albeit with much lower efficiency than at the 3′-OH of D-DNA substrates. Interaction of mirror-image DNA oligonucleotides with a natural DNA polymerase has been reported recently, however, with a substantial difference in location of the binding site compared to D-DNA.58 To the best of our knowledge, this is the first report of a native DNA polymerase in L-conformation showing cross-chiral activity via catalysis of a reaction on a mirror-image DNA substrate, thereby generating chimeric L-/D-DNA strands. TdT was found to preferentially extend L-DNA strands featuring a thymidine residue at the 5′ terminus, and efficiency of initiation was enhanced considerably compared to extension of monomers by increasing the length of strands with one or more terminal T nucleotides. Given the enhanced intracellular stability of mirror-image oligonucleotides,59 their potential as drug delivery vehicles in the form of micelles generated via TdT-mediated extension of L-DNA aptamers is an alluring prospect.10 That TdT can elongate even short initiator sequences is of major importance for enzymatic de novo synthesis of DNA since any initiator must be either removed after synthesis or chosen to match the 5′ end of the desired sequence. Presumably, any initiator must be synthesized chemically, negating many benefits of enzymatic synthesis, at least for longer initiators. Fortunately, short initiators work reasonably well, such that in the manner of current solid phase synthesis of DNA, synthesis columns preloaded with the first 5′ nucleoside on a long and cleavable linker could be used. The elongation efficiency of monomers is about a third of that of pentamers, thus requiring longer initial cycles until a more optimal length is reached. The extension of non-natural initiators such as HEG and L-DNA by TdT could also be used as a workaround; even retained as a 5′ extension to the desired DNA sequence, these initiators are largely bio-orthogonal and would not interfere in many downstream applications, or could potentially be selectively removed chemically or enzymatically after synthesis. Nevertheless, the demonstrated success of nucleotide monomer initiators for de novo TdT synthesis seems more useful in most contexts. The use of alternative initiator chemistries still supports the possibility to use TdT to create mixed nucleic acid chimeras, particularly since several non-DNA nucleoside triphosphates have been found to be accepted by TdT.11,27,28,30−32 The strong sequence dependence of TdT initiator extension is a potential complication in TdT-based de novo synthesis. Crystallographic studies of murine TdT indicate that three consecutive nucleotides are at well-defined positions within the polymerase,52 suggesting that TdT processivity is potentially sensitive to the identity of the last three bases, but unlikely to be significantly affected by further upstream bases. This hypothesis is largely confirmed by our data. Consensus logos for the 3′-hydroxy DNA initiators extended most efficiently by 1756 https://doi.org/10.1021/acssynbio.1c00142 ACS Synth. Biol. 2021, 10, 1750−1760 ACS Synthetic Biology pubs.acs.org/synthbio Research Article Table 1. Results Summary Regarding Sequence and Length Dependence of Initiation Efficiency on Oligonucleotide Extension a with TdT Polymerase and dTTP for Various Types of Initiator Chemistries sequence motifs (5′→3′) for 10% highest/lowest signal most/least efficiently extended substrate dimers trimers tetramers pentamers highest lowest highest G _ T _ _ _ _ G G _ _ T/C _ G/A C/A _ _ _ _ _ _ _ C _ _ G/A C/A _ _ _ _ _c _ _ _ _ C _ _ _ G/A C/Ac sequence TTCAT TAGAC GGUGC corresponding normalized signald 1.000 0.315 0.264 normalized average signalb D-DNA 3′-OH 2′OMe-RNA 3′- OH L-DNA 5′-OH 0.652 0.086 0.021 lowest highest lowest n.s. n.e. U _ T _ _ _ n.s. n.e. U U U/G T T _ _ G _ n.s. n.e. _ _ U G/U T T _ _ A G _ _ n.s. n.e. _ _ U U G/U T T _ _ _c A G G _ _ n.s. n.e. UGUUG TTAAA AGT (HEG)2 n.e. 0.018 0.102 0.004 0.125 n.e. 0.083 0.002 HEG D-DNA 5′-OH a“_”, no distinct nucleotide occurring at higher frequency at this position; n.s., no sequence dependence; n.e., no extension. bFluorescent signal intensities averaged over all lengths and sequences, then normalized to highest signal intensity (1 = D-DNA 3′-OH “TTCAT”). cInitiator length showing highest fluorescent signal for extension. dFluorescent signal intensity of best or worst sequence for initiation, respectively, normalized to highest signal intensity. TdT include only the last three 3′ bases for both the pentamers and the tetramers, and the last two 3′ bases in the case of the trimers (Figure 3a). In the case of the most poorly extended initiators, a consensus only appears for the terminal 3′ base for the pentamers and tetramers, whereas there is a small contribution from the second nucleobase in the case of the the 2′O-methyl-RNA and L-DNA trimers. In the case of initiating strands, again only two or three bases adjacent to the 3′-hydroxy end contribute significantly, either positively or to the polymerase extension efficiency. For negatively, extension of TdT’s natural substrate, we found a 3-fold range in efficiency between the best and worse initiator sequences for pentamers, with the worst sequences resulting in polymer- ization yields similar to the average values obtained for monomer extension, about 2.5-fold lower than the average for pentamers. Poorly extended pentamers are characterized by a 3′ cytosine, whereas the more optimal initiators are less well- defined but are generally missing cytosines in the two terminal positions. Very similar trends are apparent for tetramers, and for trimers the pattern is less well-defined, but guanines in the first two 3′ positions and cytosine or thymine at the 3′ are correlated with best and worse extension, respectively. For monomers and dimers, the reduced number of possible initiators and the smaller range between the best and worse initiators prevents a similar sequence assessment. In the case of 2′OMe-RNA initiating strands, we measured a ∼12-fold range in initiator extension efficiency between the best and worst pentamer sequences. For tetramers, trimers and dimers, the range decreases with length but is far larger than for 3′-hydroxy D-DNA initiators of the same length (Figure 4). Only in the case of the monomers is the efficiency largely independent of nucleobase identity. This strong sequence dependence results in well-defined consensus sequence logos. The nucleobases immediately adjacent to the 3′ terminus are consistently cytosines and adenosines for the best initiators and guanines and uracils for the worst initiators. That the sequence dependence for 2′OMe-RNA is completely different from that of D-DNA is not surprising given that the methoxy group must substantially alter the conformation of the initiator within TdT, such that, apparently, only sequences with the rather specific pattern revealed by the consensus logos are able to function as a substrate for polymerization. As for 2′OMe-RNA, TdT is also able to extend the 5′ hydroxy of L-DNA with low but clearly measurable yield. Similarly, the sequence-dependent range of extension efficiency is very large, about 20-fold, and associated with specific sequence patterns. Better initiators share a pair of terminal thymines, whereas the worse initiators omit this base in these positions and instead favor adenine and guanine. Since this substrate is the wrong end of the enantiomorph of the natural substrate of TdT, it appears that the polymerase is rather unspecific and will add nucleotides to many hydroxy-bearing molecules that fit within its binding site. This hypothesis is supported by the extension of hexaethylene glycol, which has little resemblance to single-stranded DNA other than flexibility and a terminal hydroxy group. Figure 6 clearly indicates fluorescent intensity, corresponding to extension efficiency, reaching a maximum for two linked HEG molecules. For the cases of both one and two HEG units, the extension efficiency is greater than for any of the substrates except the natural DNA substrate and the ∼10% best 2′OMe-RNA initiators. We attribute the loss of efficiency with further extension to the primary alcohol becoming less accessible within a polyethylene glycol tangle. signal By comparing absolute signal intensities (Figure 2) for the different chemistries and averaged values across all sequence variations (summarized in Table 1), D-DNA with available 3′- OH represents the most efficient polymerization initiator. The data shown in Table 1 indicate that the extension of even the poorest initiator sequence made of 3′-OH D-DNA remains a better primer substrate. These differences should be taken into account when considering nonstandard initiators. In such cases, the reaction conditions should be adapted, with for instance longer reaction times or an increase of TdT concentration. than any other type of ■ CONCLUSIONS Our study brings important new information to the activity spectrum of TdT polymerase. In addition to its already well- described broad range of acceptance for different types of modified (d)NTPs and their analogues, we show here its ability to extend other types of initiators as well. Although the natural substrate of TdT, the 3′ terminus of DNA, clearly outperforms 2′OMe-RNA (3′-OH), L-DNA (5′-OH), and in enzymatic extension efficiency, that hexaethylene glycol 1757 https://doi.org/10.1021/acssynbio.1c00142 ACS Synth. Biol. 2021, 10, 1750−1760 ACS Synthetic Biology these initiators are extended at all is remarkable. With the investigation of sequence dependence on the efficiency of extension, and the detection of initiation of extension even for single nucleotides, our results open up new opportunities for decoupling approaches for enzymatic de novo synthesis from chemical synthesis of DNA and illustrate substrate diversity coexisting with sequence specificity for the template- independent TdT polymerase. ■ ASSOCIATED CONTENT *sı Supporting Information The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acssynbio.1c00142. Description of SI contents (PDF) Data S1: Fluorescent intensity data of all experimental data in spreadsheet format (XLSX) Data S2: Layout design file with the location and identity of all probes for the HEG arrays (TXT) Data S3: Layout design file with the location and identity of all probes for the oligonucleotide arrays (TXT) Data S4: High resolution fluorescent scan data for the 2′OMe-RNA data (TIF) Data S5: High resolution fluorescent scan data for the D- DNA 3′-OH extension data (TIF) Data S6: High resolution fluorescent scan data for the D- DNA 5′-OH extension data (TIF) Data S7: High resolution fluorescent scan data for the HEG extension data (TIF) Data S8: High resolution fluorescent scan data for the L- DNA 5′-OH data (TIF) ■ AUTHOR INFORMATION Corresponding Author Mark M. Somoza − Institute of Inorganic Chemistry, Faculty of Chemistry, University of Vienna, 1090 Vienna, Austria; Chair of Food Chemistry and Molecular Sensory Science, Technical University of Munich, 85354 Freising, Germany; Leibniz-Institute for Food Systems Biology at the Technical University of Munich, 85354 Freising, Germany; orcid.org/0000-0002-8039-1341; Email: mark.somoza@ univie.ac.at Authors Erika Schaudy − Institute of Inorganic Chemistry, Faculty of Chemistry, University of Vienna, 1090 Vienna, Austria; orcid.org/0000-0002-2803-6684 Jory Lietard − Institute of Inorganic Chemistry, Faculty of Chemistry, University of Vienna, 1090 Vienna, Austria; orcid.org/0000-0003-4523-6001 Complete contact information is available at: https://pubs.acs.org/10.1021/acssynbio.1c00142 Author Contributions M.M.S. and E.S. conceived the study. E.S. performed the experiments and analyzed the data. E.S. and M.M.S. wrote the manuscript. E.S., M.M.S., and J.L. discussed the results and carefully revised the manuscript. Notes The authors declare no competing financial interest. pubs.acs.org/synthbio ■ ACKNOWLEDGMENTS Research Article The authors thank Orgentis GmbH for the synthesis of D-DNA phosphoramidites and ChemGenes for L-DNA, 2′OMe-RNA, and HEG monomers. This work was supported by the Austrian Science Fund (FWF P30596) and the Faculty of Chemistry of the University of Vienna. ■ REFERENCES (1) Bollum, F. J. (1960) Calf Thymus Polymerase. J. Biol. Chem. 235, 2399−2403. (2) Fowler, J. D., and Suo, Z. (2006) Biochemical, Structural, and Physiological Characterization of Terminal Deoxynucleotidyl Trans- ferase. Chem. Rev. 106, 2092−2110. (3) Gouge, J., Rosario, S., Romain, F., Beguin, P., and Delarue, M. (2013) Structures of Intermediates along the Catalytic Cycle of Terminal Deoxynucleotidyltransferase: Dynamical Aspects of the Two-Metal Ion Mechanism. J. Mol. Biol. 425, 4334−4352. (4) Desiderio, S. V., Yancopoulos, G. D., Paskind, M., Thomas, E., Boss, M. A., Landau, N., Alt, F. W., and Baltimore, D. (1984) Insertion of N regions into heavy-chain genes is correlated with expression of terminal deoxytransferase in B cells. Nature 311, 752− 755. (5) Schatz, D. G., Oettinger, M. A., and Schlissel, M. S. (1992) V(D)J Recombination: Molecular Biology and Regulation. Annu. Rev. Immunol. 10, 359−383. (6) Kordon, M. M., Zarębski, M., Solarczyk, K., Ma, H., Pederson, T., and Dobrucki, J. W. (2020) STRIDEa fluorescence method for direct, specific in situ detection of individual single- or double-strand DNA breaks in fixed cells. Nucleic Acids Res. 48, e14−e14. (7) Jang, E. K., Son, R. G., and Pack, S. P. (2019) Novel enzymatic single-nucleotide modification of DNA oligomer: prevention of incessant incorporation of nucleotidyl transferase by ribonucleotide- borate complex. Nucleic Acids Res. 47, e102−e102. (8) Cao, B., Wu, X., Zhou, J., Wu, H., Liu, L., Zhang, Q., DeMott, M. S., Gu, C., Wang, L., You, D., et al. (2020) Nick-seq for single- nucleotide resolution genomic maps of DNA modifications and damage. Nucleic Acids Res. 48, 6715−6725. (9) Tang, L., Navarro, L. A., Jr., Chilkoti, A., and Zauscher, S. (2017) High-Molecular-Weight Polynucleotides by Transferase-Catalyzed Living Chain-Growth Polycondensation. Angew. Chem., Int. Ed. 56, 6778−6782. (10) Tang, L., Tjong, V., Li, N., Yingling, Y. G., Chilkoti, A., and Zauscher, S. (2014) Enzymatic polymerization of high molecular weight DNA amphiphiles that self-assemble into star-like micelles. Adv. Mater. 26, 3050−3054. (11) Wolff, N., Hendling, M., Schönthaler, S., Geiss, A. F., and Barišić, I. (2019) Low-cost microarray platform to detect antibiotic resistance genes. Sens. Biosensing Res. 23, 100266. (12) Tjong, V., Yu, H., Hucknall, A., and Chilkoti, A. (2013) Direct Fluorescence Detection of RNA on Microarrays by Surface-Initiated Enzymatic Polymerization. Anal. Chem. 85, 426−433. (13) Sarac, I., and Hollenstein, M. (2019) Terminal Deoxynucleo- tidyl Transferase in the Synthesis and Modification of Nucleic Acids. ChemBioChem 20, 860−871. (14) Hoff, K., Halpain, M., Garbagnati, G., Edwards, J. S., and Zhou, W. (2020) Enzymatic Synthesis of Designer DNA Using Cyclic Reversible Termination and a Universal Template. ACS Synth. Biol. 9, 283−293. (15) Erlich, Y., and Zielinski, D. (2017) DNA Fountain enables a robust and efficient storage architecture. Science 355, 950−954. (16) Grass, R. N., Heckel, R., Puddu, M., Paunescu, D., and Stark, W. J. (2015) Robust Chemical Preservation of Digital Information on DNA in Silica with Error-Correcting Codes. Angew. Chem., Int. Ed. 54, 2552−2555. (17) Antkowiak, P. L., Lietard, J., Darestani, M. Z., Somoza, M. M., Stark, W. J., Heckel, R., and Grass, R. N. (2020) Low cost DNA data storage using photolithographic synthesis and advanced information reconstruction and error correction. Nat. Commun. 11, 5345. 1758 https://doi.org/10.1021/acssynbio.1c00142 ACS Synth. Biol. 2021, 10, 1750−1760 ACS Synthetic Biology pubs.acs.org/synthbio Research Article (18) Lee, H. H., Kalhor, R., Goela, N., Bolot, J., and Church, G. M. (2019) Terminator-free template-independent enzymatic DNA syn- thesis for digital information storage. Nat. Commun. 10, 2383. (19) Barthel, S., Palluk, S., Hillson, N. J., Keasling, J. D., and Arlow, D. H. (2020) Enhancing Terminal Deoxynucleotidyl Transferase Activity on Substrates with 3′ Terminal Structures for Enzymatic De Novo DNA Synthesis. Genes 11, 102. (20) Palluk, S., Arlow, D. H., de Rond, T., Barthel, S., Kang, J. S., Bector, R., Baghdassarian, H. M., Truong, A. N., Kim, P. W., Singh, A. K., et al. (2018) De novo DNA synthesis using polymerase-nucleotide conjugates. Nat. Biotechnol. 36, 645. (21) Mathews, A. S., Yang, H., and Montemagno, C. (2016) Photo- cleavable nucleotides for primer free enzyme mediated DNA synthesis. Org. Biomol. Chem. 14, 8278−8288. (22) Gurney, A. M., and Lester, H. A. (1987) Light-flash physiology with synthetic photosensitive compounds. Physiol. Rev. 67, 583−617. (23) Lee, H., Wiegand, D. J., Griswold, K., Punthambaker, S., Chun, H., Kohman, R. E., and Church, G. M. (2020) Photon-directed multiplexed enzymatic DNA synthesis for molecular digital data storage. Nat. Commun. 11, 5246. (24) Kato, K.-i., Gonalves, J. M., Houts, G.E., and Bollum, F.J. (1967) Deoxynucleotide-polymerizing enzymes of calf thymus gland. II. Properties of the terminal deoxynucleotidyltransferase. J. Biol. Chem. 242, 2780−2789. (25) Hayes, F. N., Mitchell, V. E., Ratliff, R. L., Schwartz, A. W., and Williams, D. L. (1966) Incorporation Efficiency of Small Oligo-5′- nucleotide Initiators in the Terminal Deoxyribonucleotide Trans- ferase Reaction. Biochemistry 5, 3625−3629. J.-B., Rougeon, F., and Papanicolaou, C. (2001) (26) Boulé, Terminal Deoxynucleotidyl Transferase Indiscriminately Incorporates Ribonucleotides and Deoxyribonucleotides. J. Biol. Chem. 276, 31388−31393. (27) Roychoudhury, R., and Kössel, H. (1971) Synthetic Polynucleotides. Enzymic Synthesis of Ribonucleotide Terminated Oligodeoxynucleotides and Their Use as Primers for the Enzymic Synthesis of Polydeoxynucleotides. Eur. J. Biochem. 22, 310−320. (28) Rosemeyer, V., Laubrock, A., and Seibl, R. (1995) Non- radioactive 3′-End-Labeling of RNA Molecules of Different Lengths by Terminal Deoxynucleotidyltransferase. Anal. Biochem. 224, 446− 449. (29) Yang, Y.-J., Song, L., Zhao, X.-C., Zhang, C., Wu, W.-Q., You, H.-J., Fu, H., Zhou, E.-C., and Zhang, X.-H. (2019) A Universal Assay for Making DNA, RNA, and RNA−DNA Hybrid Configurations for Single-Molecule Manipulation in Two or Three Steps without Ligation. ACS Synth. Biol. 8, 1663−1672. (30) Guerra, C. E. (2006) Analysis of oligonucleotide microarrays by 3′ end labeling using fluorescent nucleotides and terminal transferase. BioTechniques 41, 53−56. (31) Tauraitė, D., Jakubovska, J., Dabužinskaitė, J., Bratchikov, M., and Meškys, R. (2017) Modified Nucleotides as Substrates of Terminal Deoxynucleotidyl Transferase. Molecules 22, 672. (32) Matyugina, E. S., Alexandrova, L. A., Jasco, M. V., Ivanov, A. V., Vasiliev, I. A., Lapteva, V. L., Khandazhinskaya, A. L., and Kukhanova, M. K. (2009) Structure-functional analysis of interactions of terminal deoxynucleotidyl transferase with new non-nucleoside substrates. Russ. J. Bioorg. Chem. 35, 342−348. (33) Jarchow-Choy, S. K., Krueger, A. T., Liu, H., Gao, J., and Kool, E. T. (2011) Fluorescent xDNA nucleotides as efficient substrates for a template-independent polymerase. Nucleic Acids Res. 39, 1586− 1594. (34) Röthlisberger, P., Levi-Acobas, F., Sarac, I., Baron, B., England, P., Marlière, P., Herdewijn, P., and Hollenstein, M. (2017) Facile immobilization of DNA using an enzymatic his-tag mimic. Chem. Commun. 53, 13031−13034. (35) Chua, J. P. S., Go, M. K., Osothprarop, T., McDonald, S., Karabadzhak, A. G., Yew, W. S., Peisajovich, S., and Nirantar, S. (2020) Evolving a Thermostable Terminal Deoxynucleotidyl Trans- ferase. ACS Synth. Biol. 9, 1725−1735. (36) Kukhanova, M. K., Ivanov, A. V., and Jasko, M. V. (2005) StructuralFunctional Relationships between Terminal Deoxynu- cleotidyltransferase and 5′-Triphosphates of Nucleoside Analogs. Biochemistry (Moscow) 70, 890−896. (37) Sosunov, V. V., Santamaria, F., Victorova, L. S., Gosselin, G., Rayner, B., and Krayevsky, A. A. (2000) Stereochemical control of DNA biosynthesis. Nucleic Acids Res. 28, 1170−1175. (38) Thomas, C., Rusanov, T., Hoang, T., Augustin, T., Kent, T., Gaspar, I., and Pomerantz, R. T. (2019) One-step enzymatic modification of RNA 3′ termini using polymerase θ. Nucleic Acids Res. 47, 3272−3283. (39) Winz, M.-L., Samanta, A., Benzinger, D., and Jäschke, A. (2012) Site-specific terminal and internal labeling of RNA by poly(A) polymerase tailing and copper-catalyzed or copper-free strain- promoted click chemistry. Nucleic Acids Res. 40, e78. (40) Hölz, K., Schaudy, E., Lietard, J., and Somoza, M. M. (2019) Multi-level patterning nucleic acid photolithography. Nat. Commun. 10, 3805. (41) Hölz, K., Hoi, J. K., Schaudy, E., Somoza, V., Lietard, J., and Somoza, M. M. (2018) High-Efficiency Reverse (5′→3′) Synthesis of Complex DNA Microarrays. Sci. Rep. 8, 15099. (42) Chen, S., Phillips, M. F., Cerrina, F., and Smith, L. M. (2009) Controlling Oligonucleotide Surface Density in Light-Directed DNA Array Fabrication. Langmuir 25, 6570−6575. (43) Agbavwe, C., Kim, C., Hong, D., Heinrich, K., Wang, T., and Somoza, M. M. (2011) Efficiency, error and yield in light-directed maskless synthesis of DNA microarrays. J. Nanobiotechnol. 9, 57. (44) Lietard, J., Abou Assi, H., Gomez-Pinto, I., Gonzalez, C., Somoza, M. M., and Damha, M. J. (2017) Mapping the affinity landscape of Thrombin-binding aptamers on 2′F-ANA/DNA chimeric G-Quadruplex microarrays. Nucleic Acids Res. 45, 1619− 1632. (45) Lietard, J., Ameur, D., Damha, M. J., and Somoza, M. M. (2018) High-Density RNA Microarrays Synthesized In Situ by Photolithography. Angew. Chem., Int. Ed. 57, 15257−15261. (46) Schaudy, E., Somoza, M. M., and Lietard, J. (2020) l-DNA Duplex Formation as a Bioorthogonal in Nucleic Acid-Based Surface Patterning. Chem. - Eur. J. 26, 14310− 14314. (47) Sack, M., Kretschy, N., Rohm, B., Somoza, V., and Somoza, M. M. (2013) Simultaneous Light-Directed Synthesis of Mirror-Image Microarrays in a Photochemical Reaction Cell with Flare Suppression. Anal. Chem. 85, 8513−8517. (48) Kretschy, N., Holik, A.-K., Somoza, V., Stengele, K.-P., and Somoza, M. M. (2015) Next-Generation o-Nitrobenzyl Photolabile Groups for Light-Directed Chemistry and Microarray Synthesis. Angew. Chem., Int. Ed. 54, 8555−8559. (49) Hölz, K., Lietard, J., and Somoza, M. M. (2017) High-Power for Photo- 365 nm UV LED Mercury Arc Lamp Replacement chemistry and Chemical Photolithography. ACS Sustain. ACS Sustainable Chem. Eng. 5, 828−834. (50) Schneider, T. D., and Stephens, R. M. (1990) Sequence logos: a new way to display consensus sequences. Nucleic Acids Res. 18, 6097− 6100. J. (2010) Terminal (51) Motea, E. A., and Berdis, A. deoxynucleotidyl transferase: The story of a misguided DNA polymerase. Biochim. Biophys. Acta, Proteins Proteomics 1804, 1151− 1166. (52) Delarue, M., Boulé, J. B., Lescar, J., Expert-Bezançon, N., Jourdan, N., Sukumar, N., Rougeon, F., and Papanicolaou, C. (2002) Crystal structures of a template-independent DNA polymerase: murine terminal deoxynucleotidyltransferase. EMBO J. 21, 427−439. (53) Kim, J. H., and Larson, R. G. (2007) Single-molecule analysis of 1D diffusion and transcription elongation of T7 RNA polymerase along individual stretched DNA molecules. Nucleic Acids Res. 35, 3848−3858. (54) Daitchman, D., Greenblatt, H. M., and Levy, Y. (2018) Diffusion of ring-shaped proteins along DNA: case study of sliding clamps. Nucleic Acids Res. 46, 5935−5949. Information Channel 1759 https://doi.org/10.1021/acssynbio.1c00142 ACS Synth. Biol. 2021, 10, 1750−1760 ACS Synthetic Biology pubs.acs.org/synthbio Research Article (55) Focher, F., Maga, G., Bendiscioli, A., Capobianco, M., Colonna, F., Garbesi, A., and Spadari, S. (1995) Stereospecificity of human DNA polymerases α,β,γ,§ and ε, HIV-reverse transcriptase, HSV- DNA polymerase, calf thymus terminal transferase and Escherichia coli DNA polymerase I in recognizing D- and L-thymidine 5′- triphosphate as substrate. Nucleic Acids Res. 23, 2840−2847. (56) Hauser, N. C., Martinez, R., Jacob, A., Rupp, S., Hoheisel, J. D., and Matysiak, S. (2006) Utilising the left-helical conformation of L- DNA for analysing different marker types on a single universal microarray platform. Nucleic Acids Res. 34, 5101−5111. (57) Urata, H., Shinohara, K., Ogura, E., Ueda, Y., and Akagi, M. (1991) Mirror-image DNA. J. Am. Chem. Soc. 113, 8174−8175. (58) An, J., Choi, J., Hwang, D., Park, J., Pemble, C. W., Duong, T. H. M., Kim, K.-R., Ahn, H., Chung, H. S., and Ahn, D.-R. (2020) The crystal structure of a natural DNA polymerase complexed with mirror DNA. Chem. Commun. 56, 2186−2189. (59) Zhong, W., and Sczepanski, J. T. (2021) Direct Comparison of d-DNA and l-DNA Strand-Displacement Reactions in Living Mammalian Cells. ACS Synth. Biol. 10, 209−212. 1760 https://doi.org/10.1021/acssynbio.1c00142 ACS Synth. Biol. 2021, 10, 1750−1760
null
10.1088_2632-2153_ad0e17.pdf
Data availability statement All data that support the findings of this study are included within the article (and any supplementary files).
Data availability statement All data that support the findings of this study are included within the article (and any supplementary files).
OPEN ACCESS RECEIVED 16 May 2023 REVISED 6 November 2023 ACCEPTED FOR PUBLICATION 20 November 2023 PUBLISHED 28 November 2023 Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence. Any further distribution of this work must maintain attribution to the author (s) and the title of the work, journal citation and DOI. Mach. Learn.: Sci. Technol. 4 (2023) 045039 https://doi.org/10.1088/2632-2153/ad0e17 PAPER Artificial intelligent identification of apatite fission tracks based on machine learning Zuoting Ren1, Shichao Li1,2,∗, Perry Xiao3, Xiaopeng Yang1 and Hongtao Wang1 1 College of Earth Sciences, Jilin University, Changchun 130061, People’s Republic of China 2 Key Laboratory of Mineral Resources Evaluation in Northeast Asia, Ministry of Natural Resources, Changchun 130061, People’s Republic of China 3 School of Engineering, London South Bank University, London SE1 0AA, United Kingdom ∗ Author to whom any correspondence should be addressed. E-mail: lsc@jlu.edu.cn Keywords: apatite fission track, OpenCV cascade classifier, TensorFlow object detection API, precision, recall, F1-Score Abstract Over the past half century, apatite fission track (AFT) thermochronometry has been widely used in the studies of thermal histories of Earth’s uppermost crust. The acquired thermal histories in turn can be used to quantify many geologic processes such as erosion, sedimentary burial, and tectonic deformation. However, the current practice of acquiring AFT data has major limitations due to the use of traditional microscopes by human operators, which is slow and error-prone. This study uses the local binary pattern feature based on the OpenCV cascade classifier and the faster region-based convolutional neural network model based on the TensorFlow Object Detection API, these two methods offer a means for the rapid identification and measurement of apatite fission tracks, leading to significant improvements in the efficiency and accuracy of track counting. We employed a training dataset consisting of 50 spontaneous fission track images and 65 Durango standard samples as training data for both techniques. Subsequently, the performance of these methods was evaluated using additional 10 spontaneous fission track images and 15 Durango standard samples, which resulted in higher Precision, Recall, and F1-Score values. Through these illustrative examples, we have effectively demonstrated the higher accuracy of these newly developed methods in identifying apatite fission tracks. This suggests their potential for widespread applications in future apatite fission track research. 1. Introduction The fission track method (Silk and Barnes 1959, Price and Walker 1962, Fleischer et al 1965) has been applied to resolve many geological problems such as determining the thermal history of sedimentary basins (Gleadow et al 1986, Emmel et al 2014), the timing of fault activities (Roden-Tice and Wintsch 2002, Abbey and Niemi 2018), source-sink coupling during mountain building processes (Ruiz et al 2004, Chen et al 2020), uplift histories of orogenic belts (He et al 2018, Bonilla et al 2020, Wang et al 2023), regional tectonic evolution (Grist and Zentilli 2003), and mineralization (Chakurian et al 2003). In the 1980s and 1990s, the apatite fission track dating method has been greatly improved after the introduction of ζ age parameter reference, the performance of annealing experiments, the development of annealing models, and the quantification of apatite multivariate kinetic annealing processes (Hurford and Green 1983, Laslett et al 1987, Green et al 1989, Vrolijk et al 1992). The fission track dating method is a valuable tool in geology as it provides information not only on the age of a geologic event but also on the thermal history of a sample. For instance, fission track lengths can be used to determine the age and cooling rate of an orogenic uplift event. The thermal history of orogenic development can also be modeled using software like HeFty and QtQt (Vermeesch and Tian 2014). There are several methods for fission track identification and counting including the external-detector method, subtraction method, re-etching method, and re-polishing method. Currently, the external detector method is © 2023 The Author(s). Published by IOP Publishing Ltd Mach. Learn.: Sci. Technol. 4 (2023) 045039 Z Ren et al the most widely used method for fission track thermochronology. This method determines the fission track age by counting the fossil fission tracks in minerals and induced-fission tracks in external detectors. In the present study, fission tracks were obtained using the external detector method. The procedure involved collecting rock samples, isolating apatite crystals, creating thin sections, etching the sections, irradiating the samples with thermal neutrons, processing post-irradiation, and counting the number and distribution of fission tracks. Determining the number and length of fission tracks may be the final step in the dating procedure, but it is also the most important and challenging step. This work forms the foundation for subsequent studies such as dating and thermal history simulation. Traditionally, fission track statistics were obtained through manual observations using a microscope to count the track lengths. This process can be time-consuming for the operator and prone to counting errors, leading to a low efficiency of fission track identification. To improve efficiency, technology with automatic recognition capabilities is necessary. The initial investigations into the automatic recognition of fission tracks utilized image morphology analysis (Petford et al 1993, Gleadow et al 2009). This approach entailed the conversion of the transmitted light-reflected light image of fission tracks into a binary image followed by using a threshold segmentation procedure to separate the two images. The overlapping features in the two binary images were then extracted and used for recognition. In recent years, machine learning has developed rapidly and has applications in many fields that work with large data sets. Many algorithms have been developed for data mining with particular applications to Earth science research (Fleming et al 2021, Recanati et al 2021, Zhang et al 2021), such as rock classification, stratigraphic analysis, earthquake prediction, landslide statistics, and geochemistry (Li et al 2018, Baraboshkin et al 2020, de Lima et al 2020, Xu et al 2021). Object detection technology is a widely researched area in machine learning. It can be divided into traditional algorithms and deep learning algorithms, with the latter being the current mainstream. Traditional object detection algorithms have been applied in image processing and face recognition, but they suffer from low efficiency and low recognition rates. The advent of regional convolution neural networks has propelled object detection technology into the deep learning era, resulting in a significant improvement in detection accuracy (Girshick 2015). In recent years, significant advancements have been made in the field of intelligent identification of apatite fission tracks, thanks to the progress in target detection technology. Researchers have achieved notable progress by employing cutting-edge techniques. For instance, Nachtergaele et al successfully developed a deep neural network that demonstrates exceptional capabilities in intelligently identifying apatite fission tracks, yielding highly accurate results (Nachtergaele and Grave 2021). Similarly, Li et al utilized a convolutional neural network (CNN) to extract semi-tracks through image semantic segmentation, thereby contributing to the study of intelligent identification methods for apatite fission tracks (Li et al 2022). These advancements highlight the promising trajectory of research in this area. This paper employs two novel object detection algorithms. The first is the TensorFlow Object Detection API based on faster region-based convolutional neural network (R-CNN) (Ren et al 2017), which integrates feature extraction, candidate area extraction, object location, and object classification into a single network, thereby enhancing recognition capabilities. We also employed the local binary pattern (LBP) feature-based OpenCV cascade classifier object detection method for comparison (Ojala et al 2000). After selecting the two approaches, we gathered and filtered the respective experimental data (photographs of apatite fission track samples). We carefully selected clear and representative photos of apatite fission track samples. These sample photos were preprocessed to establish the necessary experimental conditions for running the two methods. Subsequently, each method was individually trained while continuously adjusting the training duration, the amount of training samples, and the model parameters. The experimental results of both methods were calculated, and a comparison and discussion were conducted on their Precision, Recall, and F1-Score. 2. Data and methods Tracks that have not undergone chemical etching are referred to as latent tracks, while fission tracks that have been etched become linear grooves with defined lengths. Usually, these tracks lack a preferred orientation. However, in some cases, ‘noise’ on the etched surface of apatite, such as defects and scratches, can interfere with track identification. These challenges make the identification of fission tracks difficult. To overcome these issues, we use two newly developed methods for removing these effects. 2 Mach. Learn.: Sci. Technol. 4 (2023) 045039 Z Ren et al Figure 1. LBP schematic. 2.1. Experimental method 2.1.1. OpenCV processing image recognition OpenCV stands for open source computer vision library, which primarily uses image processing and machine learning algorithms to address related problems. Compared to other computer vision recognition libraries, OpenCV boasts several advantages, such as multiple language interfaces, cross-platform compatibility, robust development, and a plentiful API. In this paper, we use the OpenCV cascade classifier based on LBP features for our method. LBP is a tool for capturing the local texture features of an image, primarily utilized for texture feature extraction (Ojala et al 1996). It operates within a 3 × 3 window where the center pixel is used as a threshold and the grayscale values of the eight surrounding pixels are compared to it. If the peripheral pixel value is greater than the center pixel value, the pixel’s position is marked as 1; otherwise, it is marked as 0. By comparing the 8 points in the 3 × 3 neighborhood, 8-bit binary numbers are generated, which represent the LBP value of the center pixel and depict the texture information of the area (figure 1, Ojala et al 2000). The process is described by the equation: P−1∑ ( LBP(P,R) = s gp − ga ) 2p p=0 { 1x ⩾ 0 0x < 0 . s (x) = (1) (2) In the equation, gp is the gray value of the pixel on the circle with the center pixel and R as the radius, ga is the gray value of the center pixel in the corresponding local neighborhood, and P is the number of surrounding pixels points. The experiment employs the use of the cascade classifier, which is based on the LBP algorithm and provided by OpenCV. This classifier comprises many weak classifiers that are designed to classify different features of detection targets. The classifiers at each level are more complex than those at the previous level. Multiple weak classifiers work together, and different features are extracted from each window, which are then fed into different weak classifiers for judgment. If all the labels judged by the weak classifiers are positive samples, the target is detected in the smoothing window. The classifier at each layer of the cascade is trained and optimized based on the results of the previous layer. This enables negative samples to be quickly eliminated and reduces the number of misclassified samples, improving the overall classification performance without increasing the computational complexity. 2.1.2. TensorFlow object detection API processing image recognition The TensorFlow Object Detection API is a programming interface that utilizes TensorFlow to tackle object recognition issues such as real-time object identification. It boasts a reliable API, a streamlined workflow, excellent system compatibility, and highly efficient recognition capabilities. The API comprises object detection frameworks, including single shot multibox detector (SSD), region with CNN feature (R-CNN), and region-based fully convolutional network (R-FCN). For these models, integrated experiments can also be conducted by combining different feature extraction networks. Typical feature extraction networks include VGG, Inception v3, ResNet-101, Inception ResNet, etc (Al-Azzo et al 2018). The Faster R-CNN algorithm is the method employed in this study. As a typical two-stage object detection model, Faster R-CNN builds on the R-CNN and Fast R-CNN algorithms. The key difference is the use of a region proposal network (RPN) to generate candidate proposal windows. The Faster R-CNN detection process is composed of four major components: (1) a feature extraction network, which extracts features from the input image and outputs a feature map for use by the RPN and fully connected layer; (2) a 3 Mach. Learn.: Sci. Technol. 4 (2023) 045039 Z Ren et al Figure 2. Flow chart of Faster R-CNN detection. region candidate network (RPN), which generates anchor boxes and determines whether they contain an object, and also performs a bounding box regression to form a region proposal; (3) ROI Pooling Layer. Combine the feature map obtained from the first two modules with the region proposals to obtain a fixed-size proposal feature map which is then passed into the fully connected network for classification; (4) classification and regression. The proposal and feature map are used to calculate the specific class of the object, and a bounding box regression is done to obtain the exact position of the detection box (figure 2) (Ren et al 2017). When training RPN, the Anchor is divided into two categories. An anchor with a target in the box is labeled as a positive sample. An anchor without a target in the box is labeled as a negative sample. The loss function of the RPN network consists of two components, which are classification loss (Lcls) and boundary regression loss (Lreg). The equations are as follows (Ren et al 2017): L ({pi} , {ti}) = 1 Ncls ∑ i Lcls (pi, p ∗ i ) + λ 1 Nreg ∑ i ∗ i Lreg (ti, t ∗ i ) p (3) where pi represents the probability that the ith Anchor predicted by the network is the target, p∗ i represents the true value corresponding to pi. If the Anchor is positive, p∗ is 1, and if the Anchor is negative, it is 0, ti is i a vector representing the 4 parameterized coordinates of the prediction bounding box, indicating the offset between the prediction box and the Anchor box, t∗ i represents the true value corresponding to ti, indicating the offset between the true value and the Anchor box. Ncls is set to the size of the batch, and Nreg is set to the total number of Anchors, λ is the balance parameter used for the two loss functions. The faster_rcnn_inception_v2 model uses the Inception V2 network as its base network. This network is a pretrained CNN that has been trained on large-scale image data and has good feature extraction capabilities. It consists of multiple convolutional layers, pooling layers, and fully connected layers. In the Inception V2 network, advanced features of the image are gradually extracted through multiple convolution operations. Each convolutional layer slides a convolutional kernel over the input image to extract features and obtain an output feature map. As the number of layers increases, the receptive field gradually increases, allowing the model to model a larger range of image information (Ioffe and Szegedy 2015). 2.2. Data construction In this study, we utilized apatite fission tracks obtained through the external detector method. We employed two Durango samples, which are recognized as the standard in apatite fission track dating. The ages of the two samples are 31.4 ± 0.5 Ma (Wang et al 2018) and 31.02 ± 1.01 Ma (Mcdowell et al 2005), respectively. Our sample collection consisted of 20 images of Dur1 (Durango sample 1) and 60 images of Dur2 (Durango sample 2). A Zeiss microscope equipped with the Autoscan TrackWorks software, with a magnification setting of 1000, was used to generate sample images. The images of spontaneous fission tracks, with dimensions of 2048 px × 1536 px, were taken and numbered Qi (60 in total). These images were sourced from granite samples of the Qimen Tagh Range located on the northeastern margins of the Tibetan Plateau. 4 Mach. Learn.: Sci. Technol. 4 (2023) 045039 Z Ren et al Dataset Sample name Number of images Image format Image resolution Table 1. Fission track data table. Training samples Testing samples Qi Dur1 Dur2 Qi Dur1 Dur2 50 50 15 10 10 5 .jpg .jpg .jpg .jpg .jpg .jpg 2048 × 1536 2048 × 1536 2048 × 1536 2048 × 1536 2048 × 1536 2048 × 1536 The apatite fission tracks from these images dated from 58.7 ± 3.6 Ma to 239.5 ± 29.8 Ma. A total number of 50 images of Qi, 15 images of Dur1, and 50 images of Dur2 were selected as the training data images. Meanwhile, ten images of Qi, five images of Dur1, and ten images of Dur2 were used as test data images. These test samples were manually counted and compared against machine learning methods (table 1). 3. Experimental design Object detection is a subcategory of image recognition. In image recognition, the goal is to identify different objects present in an image, while in object detection, not only do the objects need to be recognized but also their specific location must be identified. Conducting experiments on object detection requires image labeling of the experimental data, where the objects to be recognized are assigned specific labels to distinguish them from other objects in the image. In this experiment, two methods were used, and the same training sample was used for both methods. However, the data processing methods were different. In the first method, which used OpenCV’s cascade classifier, the entire training image was cut into positive and negative samples, and this was used for data processing. In contrast, for the second method, which used TensorFlow Object Detection API, Labelimg software was used to label the entire image, and the fission track was divided into two labels: opaque and transparent. 3.1. OpenCV cascade classifier for image processing The experiment employed OpenCV’s cascade classifier to preprocess the images. The primary step in data preprocessing involved uniformly cutting the parts of the large image containing tracks into 40 × 40 pixel images and then converting them to grayscale to be used as positive samples (figure 3). The images without fission tracks were also cropped and utilized as negative samples. The sample size is specified below (table 2). The objective of creating positive and negative sample description files with corresponding samples is to produce vectorized data. The opencv_createsamples.exe program, which is part of OpenCV, can be used to create a vectorized positive sample set (.vec file) from the positive samples and their description files. The subsequent step involves utilizing OpenCV’s training classification tool (opencv_traincascade.exe) for the classification training. Upon completion of the training, a final model (cascade.xml) will be generated. Finally, the trained cascade classifier is employed for detecting the target of apatite fission track images. 3.2. TensorFlow object detection API for image processing The TensorFlow Object Detection API simplifies image preprocessing compared to the OpenCV cascade classifier, as it does not require the preparation of positive and negative samples. Instead, it is necessary to annotate the apatite fission track images. During experiments, the tracks exhibit diverse shapes, so to ease training, we categorize the training samples into two groups based on the transparency of the tracks: transparent and opaque (figure 4). We calculate the results of the two labels in a unified way in the final result testing stage. The training sample used is based on the data shown in the training set in table 1. The marking data quantity of opaque and opaque labels is shown in the following table (table 3). Labelimg is a graphical tool for image annotation, which converts label information into XML files for storage and exchange. The datasets were annotated manually using Labelimg by drawing bounding boxes around the targets and labeling them accordingly. Upon completion of the annotation, an XML file is generated automatically. However, to be compatible with TensorFlow’s operating environment, it is necessary to convert the XML file to a CSV file and then to a TensorFlow-compatible TFRECORD file. After the conversion is completed, the parameter setting and training of the model can be performed. 5 Mach. Learn.: Sci. Technol. 4 (2023) 045039 Z Ren et al Figure 3. Positive sample (A) negative sample (B). Table 2. OpenCV positive and negative sample data table. Sample name Positive samples Negative samples Number 1000 2000 Figure 4. Examples of labels. Table 3. TensorFlow object detection API label sample data table. Sample name Opaque samples Transparent samples Number 5200 800 Table 4. The important hyperparameters of TensorFlow object detection API. Name Num_classes Batch_size Initial_learning_rate Num_steps Eval_config:{num_examples} Eval_config:{max_evals} Number 2 1 0.0002 100 000 57 10 The training model used in this experiment is faster_rcnn_inception_v2. Using the TensorFlow-GPU version, compared to the TensorFlow-CPU version, the GPU operation will process graphics and images faster, shortening the training time. The important parameter settings in this training, are the batch size (the number of training samples at a time) is set to 10, and the training steps are set to 100 000. The important hyperparameters settings in this training are as follows (table 4): 6 Mach. Learn.: Sci. Technol. 4 (2023) 045039 Z Ren et al Table 5. OpenCV test result table. OpenCV Xa Xm TP FP FN Precision Recall F1-Score 23 29 16 23 29 35 27 39 7 28 27 24 28 26 23 30 44 34 37 41 31 39 29 34 41 17 25 16 18 19 24 19 29 6 22 23 22 22 22 18 20 30 21 27 32 20 27 20 24 31 6 4 0 5 10 11 8 10 1 6 4 2 6 4 5 10 14 13 10 9 11 12 9 10 10 5 9 5 5 6 8 9 10 3 5 5 7 1 9 5 5 15 2 10 11 8 11 6 7 8 Image1 Image2 Image3 Image4 Image5 Image6 Image7 Image8 Image9 Image10 Image11 Image12 Image13 Image14 Image15 Image16 Image17 Image18 Image19 Image20 Image21 Image22 Image23 Image24 Image25 Average 22 34 21 23 25 32 28 39 9 27 28 29 23 31 23 25 45 23 37 43 28 38 26 31 39 Qi Dur1 Dur2 All 73.9% 86.2% 100.0% 78.3% 65.5% 68.6% 70.4% 74.4% 85.7% 78.6% 85.2% 91.7% 78.6% 84.6% 78.3% 66.7% 68.2% 61.8% 73.0% 78.0% 64.5% 69.2% 69.0% 70.6% 75.6% 78.1% 83.7% 69.7% 75.9% 77.3% 73.5% 76.2% 78.3% 76.0% 75.0% 67.9% 74.4% 66.7% 81.5% 82.1% 75.9% 95.7% 71.0% 78.3% 80.0% 66.7% 91.3% 73.0% 74.4% 71.4% 71.1% 76.9% 77.4% 79.5% 74.7% 80.6% 76.2% 76.4% 75.6% 79.4% 86.5% 78.3% 70.4% 71.6% 69.1% 74.4% 75.0% 80.0% 83.6% 83.0% 86.3% 77.2% 78.3% 72.7% 67.4% 73.7% 73.0% 76.2% 67.8% 70.1% 72.7% 73.8% 77.5% 76.0% 81.7% 72.5% 75.7% 4. Experiment results The experimental results of different methods are evaluated by the same evaluation method. Evaluation indicators are an important basis for evaluating the goodness of target detection algorithms. There are many kinds of evaluation indicators, among which the more typical ones are Precision (P) and Recall (R), with the following formulas: P = R = TP TP + FP TP TP + FN . (4) (5) The true positive (TP) represents correctly identified fission tracks, the false positive (FP) represents the incorrectly identified fission tracks, and the false negative (FN) represents the fission tracks that were not identified. We used two different target detection models, thus it was necessary to choose between the two sets of Precision and Recall. Hence, we used F1-Score (F1) as a measure of the classification problem. It is the average of Precision and Recall, which integrates the results and ranges from 0 to 1, with 1 being the best and 0 the worst output of the model. The formula for the F1-Score is as follows: F1 = 2 × P × R P + R . (6) Tables 5 and 6 show the results of the OpenCV cascade classifier and TensorFlow Object Detection API, in which Image1–Image10 is sample Qi, Image11–Image15 is sample Dur1, and Image16–Image25 is sample Dur2. The number of fission tracks in artificially identified test pictures is Xa, and the number of fission tracks identified by the machine learning method is Xm. The experimental results using the OpenCV cascade classifier method are as follows (table 5). 7 Mach. Learn.: Sci. Technol. 4 (2023) 045039 Z Ren et al Table 6. TensorFlow object detection API test result table. TensorFlow object detection API Xa Xm TP FP FN Precision Recall F1-Score 17 26 10 17 18 28 22 32 9 22 28 27 23 26 20 28 43 25 37 41 30 34 24 32 40 16 24 10 15 17 26 21 32 9 21 28 27 22 26 20 24 39 23 33 38 25 32 24 28 38 1 2 0 2 1 2 1 0 0 1 0 0 1 0 0 4 4 2 4 3 5 2 0 4 2 5 8 11 8 8 6 7 7 0 6 0 2 1 5 3 1 6 0 4 5 3 6 2 3 1 Image1 Image2 Image3 Image4 Image5 Image6 Image7 Image8 Image9 Image10 Image11 Image12 Image13 Image14 Image15 Image16 Image17 Image18 Image19 Image20 Image21 Image22 Image23 Image24 Image25 Average 22 34 21 23 25 32 28 39 9 27 28 29 23 31 23 25 45 23 37 43 28 38 26 31 39 Qi Dur1 Dur2 All 94.1% 92.3% 100.0% 88.2% 94.4% 92.9% 95.5% 100.0% 100.0% 95.5% 100.0% 100.0% 95.7% 100.0% 100.0% 85.7% 90.7% 92.0% 89.2% 92.7% 83.3% 94.1% 100.0% 87.5% 95.0% 95.3% 99.1% 91.0% 94.4% 76.2% 75.0% 47.6% 65.2% 68.0% 81.3% 75.0% 82.1% 100.0% 77.8% 100.0% 93.1% 95.7% 83.9% 87.0% 96.0% 86.7% 100.0% 89.2% 88.4% 89.3% 84.2% 92.3% 90.3% 97.4% 74.8% 91.9% 91.4% 84.9% 84.2% 82.8% 64.5% 75.0% 79.1% 86.7% 84.0% 90.1% 100.0% 85.7% 100.0% 96.4% 95.7% 91.2% 93.0% 90.6% 88.6% 95.8% 89.2% 90.5% 86.2% 88.9% 96.0% 88.9% 96.2% 83.2% 95.3% 91.1% 88.8% The experimental results using the TensorFlow Object Detection API method are as follows (table 6). For the validation of the Qi sample (Wang et al 2018), a total of ten images (Image1–Image10) were used, with the manual identifications ranging from 9 to 39. The OpenCV cascade classifier had an average Precision of 78.1%, an average Recall of 74.7%, and an average F1-Score of 76.0%. The TensorFlow Object Detection API based on the Faster R-CNN algorithm had an average Precision of 95.3%, an average Recall of 74.8%, and an average F1-Score of 83.2%. For the verification of the Dur1 sample (Wang et al 2018), a total of five images (Image11–Image15) were used, with manual identifications ranging from 23 to 31. The OpenCV cascade classifier had an average Precision of 83.7%, an average Recall of 80.6%, and an average F1-Score of 81.7%. The TensorFlow Object Detection API based on the Faster R-CNN algorithm had an average Precision of 99.1%, an average Recall of 91.9%, and an average F1-Score of 95.3%. For the verification of the Dur2 sample (Mcdowell et al 2005), a total of ten images (Image16–Image25) were used, with manual identifications ranging from 23 to 45. The OpenCV cascade classifier had an average Precision of 69.7%, an average Recall of 76.2%, and an average F1-Score of 72.5%. The TensorFlow Object Detection API based on the Faster R-CNN algorithm had an average Precision of 91.0%, an average Recall of 91.4%, and an average F1-Score of 91.1%. For the apatite fission tracks tested using the OpenCV cascade classifier, the average Precision was 75.9%, the average Recall was 76.4%, and the average F1-Score was 75.7%. On the other hand, the TensorFlow Object Detection API, based on the Faster R-CNN algorithm, recorded an average Precision of 94.4%, Recall of 84.9%, and F1-Score of 88.8%. Figures 5 and 6 show the results of the sample OpenCV cascade classifier based on LBP and the TensorFlow Object Detection API based on the Faster R-CNN algorithm, respectively. 8 Mach. Learn.: Sci. Technol. 4 (2023) 045039 Z Ren et al Figure 5. Image 9, image 13, and image 23 are identified by the OpenCV cascade classifier. 5. Discussion 5.1. Experimental discussion From the experimental results table (table 5), it can be seen that the average accuracy of Precision, Recall, and F1-Score using the OpenCV cascade classifier based on LBP is above 75%. And it can be found that almost all of the test images show fission track recognition errors, and the number of recognition errors is high compared to that of using the TensorFlow Object Detection API. On the whole, the overall average accuracy rate of the Recall of the verified samples is higher than the overall average accuracy rate. This shows that there are more identified fission tracks than unidentified ones. In the experimental results (table 6), the TensorFlow Object Detection API based on the Faster R-CNN algorithm achieves an average accuracy of over 84% for Precision, Recall, and F1-Score. It outperforms the OpenCV cascade classifier approach in terms of average Precision, with a value close to 95%. However, the overall average Precision is higher than the overall average Recall, indicating that more fission tracks are not identified compared to the misidentified ones. From the experimental results table, it can be observed that the average Precision, Recall, and F1-Score accuracy of both methods is higher for the Dur1 sample than for the Qi and Dur2 samples. This is due to the lower background interference in the training images and fewer overlapping fission tracks in the Dur1 sample. The average Precision accuracy of both methods for the Dur1 and Qi samples is higher than the average Recall accuracy, indicating that there are more unidentified fission tracks than misidentified ones. In contrast, for the Dur2 sample, the average Precision accuracy is lower than the average Recall accuracy, meaning that there are more misidentified fission tracks than unidentified ones. 9 Mach. Learn.: Sci. Technol. 4 (2023) 045039 Z Ren et al Figure 6. Image 10, image 12, and image 18 are identified by tensorflow object detection API. The results of Precision, Recall, and F1-Score, depicted in the graphs (figure 7), demonstrate that the TensorFlow Object Detection API, based on the Faster R-CNN algorithm, outperforms the OpenCV cascade classifier based on LBP. The average accuracy of Precision, Recall, and F1-Score for the TensorFlow Object Detection API is higher, proving its superior performance in recognizing apatite fission tracks compared to the OpenCV cascade classifier. The results of both methods are based on trained data. By drawing Precision–Recall Curve (figure 8), the TensorFlow Object Detection API method has a more comprehensive algorithm and feature analysis, resulting in a higher accuracy rate in recognizing apatite fission tracks with many overlapping tracks, inconspicuous features, and short tracks. The OpenCV cascade classifier, on the other hand, has a certain accuracy in dealing with scattered, single, and well-defined fission tracks. Both methods can automatically identify apatite fission tracks. The OpenCV cascade classifier has the advantage of being easy to set up and faster in training speed, but its accuracy may be low in complex situations due to its algorithmic limitations. 5.2. Advantages and disadvantages From the data result table and experimental result chart of the two aforementioned experimental methods, it is evident that the average accuracy of Precision, Recall, and F1-Score achieved through the utilization of TensorFlow Object Detection API exceeds 84%. On the other hand, the average accuracy of Precision, Recall, and F1-Score obtained by employing the OpenCV cascade classifier surpasses 75%. While most fission tracks can be successfully identified, the identification efficiency requires enhancement when compared to other studies on intelligent identification of apatite fission tracks. There are still numerous instances of both non-identification and misidentification, indicating a lack of effective handling of overlapping fission tracks. 10 Mach. Learn.: Sci. Technol. 4 (2023) 045039 Z Ren et al Figure 7. TensorFlow object detection API and OpenCV cascade classifier Precision (A), Recall (B), F1-Score (C) result. P1–P25 are the test images 1–25. Figure 8. Precision–recall curve of tensorflow object detection API. In addition to the aforementioned issues, the two methods employed in this experiment offer certain advantages when compared to other existing studies on apatite fission track. Notably, both methods are characterized by their convenience of use. The TensorFlow Object Detection API boasts a comprehensive framework that allows for the utilization of various target detection models without necessitating extensive parameter adjustments. On the other hand, the OpenCV cascade classifier stands out due to its simple model configuration, short training duration, and minimal environmental requirements, setting it apart from alternative approaches. 11 Mach. Learn.: Sci. Technol. 4 (2023) 045039 Z Ren et al Figure 9. Fission track and scratch (1 is scratch and 2 is fission track). 5.3. Problems encountered in the experiment Compared to other intelligent research on apatite fission track, the two methods employed in this study exhibit lower recognition efficiency. The specific reasons for this are summarized as follows: (1). the number of training data samples is small compared to the amount of training data in most experiments. (2). Some of the training data samples have poor picture quality and many overlapping fission tracks. (3). The data samples have many background interference, which affects the recognition process. Firstly, the limited number of training data samples used in this experiment contributes to the insufficient diversity of the objectives. Additionally, the quality of the training data samples, which were captured through microscopy, is partially dependent on the accuracy of the experimental equipment. This can lead to inaccuracies in the pre-processing marking. To address these issues, we recommend increasing the sample size of the training data, improving the quality of the training data, and adjusting the model parameters. Secondly, during the verification process with fission track images, it is common to come across overlapping tracks, which greatly affects the object identification statistics. Although simple overlapping tracks can still be identified, a large number of overlapping tracks in some samples make some tracks unidentifiable. This presents the biggest challenge so far. Even manual recognition of overlapping track statistics is difficult. However, our study reveals that the overlapping track portion is not predominant. Hence, we suggest combining machine recognition with human recognition, enabling human recognition to correct machine recognition results to improve experiment accuracy. Finally, background interference in the test sample can also negatively impact the accuracy of fission track recognition. In some cases, ‘noise’ on the etched surface of apatite, such as defects and scratches, can interfere with track identification (figure 9). This makes it challenging to differentiate and identify fission tracks. Our solution to this issue is to minimize the interference factors in the sample through manual elimination during the pre-processing stage of the training data. 6. Conclusions The study presented in this paper employed two machine learning methods for fission track identification, with the following main conclusions: (1) Results from the experiments revealed that the TensorFlow Object Detection API outperforms the OpenCV cascade classifier in terms of fission track recognition. The API had an average Precision of 94.4%, Recall of 84.9%, and F1-Score of 88.8%, while the cascade classifier had an average Precision of 75.9%, Recall of 76.4%, and F1-Score of 75.7%. (2) The experiments encountered three challenges: a small number of training data samples with low quality, high levels of overlapping fission tracks, and high background interference in the samples. The solutions proposed were: increasing the number and quality of training data, combining machine 12 Mach. Learn.: Sci. Technol. 4 (2023) 045039 Z Ren et al recognition with manual recognition, and manually eliminating background interference factors during sample pre-processing. (3) In terms of ease of operation, the OpenCV cascade classifier environment is simpler and more convenient to build and operate compared to the TensorFlow Object Detection API. However, the API offers better integrity, allowing for monitoring of the entire training session. (4) The two target detection methods explored in this study offer the potential for wider use in the geology field. For instance, in rock research, target detection can aid in identifying rock types. Additionally, in the study of geological structures, utilizing target detection can assist in locating significant geological features such as fracture zones or volcanic craters through remote sensing images. To fully utilize the capabilities of deep learning in traditional geology, it is crucial to adopt a technically savvy approach and effectively integrate multiple methods to maximize their potential in furthering the development of geology. Data availability statement All data that support the findings of this study are included within the article (and any supplementary files). Acknowledgments This study was financially supported by the National Natural Science Foundation of China (Grant No. 41872234) and Science and Technology Research Project of Jilin Provincial Education Department (Grant No. JJKH20241255KJ). We are grateful to Professor An Yin for help, comments, and discussions on an earlier version of the manuscript. Thank you also to the all anonymous reviewers for their key comments on the revision of the manuscript. Conflict of interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. ORCID iD Shichao Li  https://orcid.org/0000-0003-3928-2132 References Abbey A L and Niemi N A 2018 Low-temperature thermochronometric constraints on fault initiation and growth in the northern Rio Grande rift, upper Arkansas River valley, Colorado, USA Geology 46 627–30 Al-Azzo F, Taqi A M and Milanova M 2018 Human related-health actions detection using android camera based on tensorflow object detection API Int. J. Adv. Comput. Sci. Appl. 9 9–23 Baraboshkin E E, Ismailova L S, Orlov D M, Zhukovskaya E A, Kalmykov G A, Khotylev O V, Baraboshkin E Y and Koroteev D A 2020 Deep convolutions for in-depth automated rock typing Comput. Geosci. 135 104330 Bonilla A, Franco J A, Cramer T, Poujol M, Cogné N, Nachtergaele S and de Grave J 2020 Apatite LA-ICP-MS U–Pb and fission-track geochronology of the Ca˜no Viejita gabbro in E-Colombia: evidence for Grenvillian intraplate rifting and Jurassic exhumation in the NW Amazonian Craton J. South Am. Earth Sci. 98 102438 Chakurian A, Arehart G, Donelick R, Zhang X and Reiners P 2003 Timing constraints of gold mineralization along the Carlin trend utilizing apatite fission-track, 40Ar/39Ar, and apatite (U-Th)/He methods Econ. Geol. 98 1159–71 Chen J, Wang Q, Qiao L, Liu X and Zhang Q 2020 Cretaceous exhumation history of the southwestern South China Block: constraints from fission-track thermochronology Geol. J. 55 6718–31 de Lima R P, Duarte D, Nicholson C, Slatt R and Marfurt K J 2020 Petrographic microfacies classification with deep convolutional neural networks Comput. Geosci. 142 104481 Emmel B, Kumar R, Jacobs J, Ueda K, Van Zuilen M and Matola R 2014 The low-temperature thermochronological record of sedimentary rocks from the central Rovuma Basin (N Mozambique)—constraints on provenance and thermal history Gondwana Res. 25 1216–29 Fleischer R L, Price P, Walker R and Leakey L 1965 Fission-track dating of Bed I, Olduvai Gorge Science 148 72–74 Fleming S W, Watson J R, Ellenson A, Cannon A J and Vesselinov V C 2021 Machine learning in Earth and environmental science requires education and research policy reforms Nat. Geosci. 14 878–80 Girshick R 2015 Fast r-CNN Proc. IEEE Int. Conf. on Computer Vision pp 1440–48 Gleadow A J, Duddy I, Green P F and Lovering J 1986 Confined fission track lengths in apatite: a diagnostic tool for thermal history analysis Contrib. Mineral. Petrol. 94 405–15 Gleadow A J, Gleadow S J, Belton D X, Kohn B P, Krochmal M S and Brown R W 2009 Coincidence mapping-a key strategy for the automatic counting of fission tracks in natural minerals Geol. Soc. 324 25–36 Green P, Duddy I, Laslett G, Hegarty K, Gleadow A W and Lovering J 1989 Thermal annealing of fission tracks in apatite 4. Quantitative modelling techniques and extension to geological timescales Chem. Geol. (Isot. Geosci. Sect.) 79 155–82 13 Mach. Learn.: Sci. Technol. 4 (2023) 045039 Z Ren et al Grist A M and Zentilli M 2003 Post-Paleocene cooling in the southern Canadian Atlantic region: evidence from apatite fission track models Can. J. Earth Sci. 40 1279–97 He P, Song C, Wang Y, Meng Q, Chen L, Yao L, Huang R, Feng W and Chen S 2018 Cenozoic deformation history of the Qilian Shan (Northeastern Tibetan Plateau) constrained by detrital apatite fission-track thermochronology in the northeastern Qaidam Basin Tectonophysics 749 1–11 Hurford A J and Green P F 1983 The zeta age calibration of fission-track dating Chem. Geol. 41 285–317 Ioffe S and Szegedy C 2015 Batch normalization: accelerating deep network training by reducing internal covariate shift (arXiv:1502.03167) Laslett G, Green P F, Duddy I and Gleadow A 1987 Thermal annealing of fission tracks in apatite 2. A quantitative analysis Chem. Geol. (Isot. Geosci. Sect.) 65 1–13 Li R, Xu Z, Su C and Yang R 2022 Automatic identification of semi-tracks on apatite and mica using a deep learning method Comput. Geosci. 162 105081 Li Z, Meier M A, Hauksson E, Zhan Z and Andrews J 2018 Machine learning seismic wave discrimination: application to earthquake early warning Geophys. Res. Lett. 45 4773–9 Mcdowell F W, Mcintosh W C and Farley K A 2005 A precise 40Ar–39Ar reference age for the Durango apatite (U–Th)/He and fission-track dating standard Chem. Geol. 214 249–63 Nachtergaele S and Grave J D 2021 AI-Track-tive: open-source software for automated recognition and counting of surface semi-tracks using computer vision (artificial intelligence) Geochronology 3 383–94 Ojala T, Pietikäinen M and Harwood D 1996 A comparative study of texture measures with classification based on featured distributions Pattern Recognit. 29 51–59 Ojala T, Pietikäinen M and Mäenpää T 2000 Gray scale and rotation invariant texture classification with local binary patterns Computer Vision—ECCV 2000 (Springer) pp 404–20 Petford N, Miller J A and Briggs J 1993 The automated counting of fission tracks in an external detector by image analysis Comput. Geosci. 19 585–91 Price P B and Walker R M 1962 Chemical etching of charged-particle tracks in solids J. Appl. Phys. 33 3407–12 Recanati A, Grozavu N, Bennani Y, Gautheron C and Missenard Y 2021 Apatite (U-Th-Sm)/He date dispersion: first insights from machine learning algorithms Earth Planet. Sci. Lett. 554 116655 Ren S, He K, Ross G and Sun J 2017 Faster R-CNN: towards real-time object detection with region proposal networks IEEE Trans. Pattern Anal. Mach. Intell. 39 1137–49 Roden-Tice M K and Wintsch R P 2002 Early Cretaceous normal faulting in southern New England: evidence from apatite and zircon fission-track ages J. Geol. 110 159–78 Ruiz G, Seward D and Winkler W 2004 Detrital thermochronology–a new perspective on hinterland tectonics, an example from the Andean Amazon Basin, Ecuador Basin Res. 16 413–30 Silk E and Barnes R 1959 Examination of fission fragment tracks with an electron microscope Phil. Mag. 4 970–2 Vermeesch P and Tian Y 2014 Thermal history modelling: HeFTy vs QTQt. Earth-Sci. Rev. 139 279–90 Vrolijk P, Donelick R A, Queng J, Cloos M, Larson R and Lancelot Y 1992 Testing models of fission track annealing in apatite in a simple thermal setting: site 800, leg 129 Proc. Ocean Drilling Program, Scientific Results (Citeseer) pp 169–176 Wang H, Li S, Zhang L, Sheldrick T C, Liu F, Zhao Z, Yang X and Wang Y 2023 Exhumation history of the greater Khingan mountains (NE China) since the late Mesozoic: implications for the tectonic regime change of Northeast Asia Lithosphere 2023 Wang Y, Zheng J and Zheng Y 2018 Mesozoic-Cenozoic exhumation history of the Qimen Tagh Range, northeastern margins of the Tibetan Plateau: evidence from apatite fission track analysis Gondwana Res. 58 16–26 Xu Z, Ma W, Lin P, Shi H, Pan D and Liu T 2021 Deep learning of rock images for intelligent lithology identification Comput. Geosci. 154 104799 Zhang W, Ching J, Goh A T C and Leung A Y F 2021 Big data and machine learning in geoscience and geoengineering: introduction Geosci. Front. 12 327–9 14
null
10.1186_s12875-019-0972-1.pdf
Availability of data and materials Table 1 provides a list of the 26 included papers and Additional file 1 shows the database search strategy.
Availability of data and materials Table 1 provides a list of the 26 included papers and Additional file 1 shows the database search strategy.
Wadsworth et al. BMC Family Practice (2019) 20:97 https://doi.org/10.1186/s12875-019-0972-1 R E S E A R C H A R T I C L E Open Access Shared medical appointments and patient- centered experience: a mixed-methods systematic review Kim H. Wadsworth1* Adam S. Hoverman3 , Trevor G. Archibald1, Allison E. Payne1, Anita K. Cleary1, Byron L. Haney1,2 and Abstract Background: Shared medical appointments (SMAs), or group visits, are a healthcare delivery method with the potential to improve chronic disease management and preventive care. In this review, we sought to better understand opportunities, barriers, and limitations to SMAs based on patient experience in the primary care context. Methods: An experienced biomedical librarian conducted literature searches of PubMed, Cochrane Library, PsycINFO, CINAHL, Web of Science, ClinicalTrials.gov, and SSRN for peer-reviewed publications published 1997 or after. We searched grey literature, nonempirical reports, social science publications, and citations from published systematic reviews. The search yielded 1359 papers, including qualitative, quantitative, and mixed method studies. Categorization of the extracted data informed a thematic synthesis. We did not perform a formal meta-analysis. Results: Screening and quality assessment yielded 13 quantitative controlled trials, 11 qualitative papers, and two mixed methods studies that met inclusion criteria. We identified three consistent models of care: cooperative health care clinic (five articles), shared medical appointment / group visit (10 articles) and group prenatal care / CenteringPregnancy® (11 articles). Conclusions: SMAs in a variety of formats are increasingly employed in primary care settings, with no singular gold standard. Accepting and implementing this nontraditional approach by both patients and clinicians can yield measurable improvements in patient trust, patient perception of quality of care and quality of life, and relevant biophysical measurements of clinical parameters. Further refinement of this healthcare delivery model will be best driven by standardizing measures of patient satisfaction and clinical outcomes. Keywords: Shared medical appointment, Group visit, Cooperative health care clinic, Group prenatal care, Patient satisfaction, Patient experience, Health services, Primary care, Primary health care, Coproduction Background Shared medical appointments (SMAs), or group visits, are a healthcare delivery innovation arising from the changing demands of patient-centered medical home (PCMH) set- tings and the primary care context. The model emphasizes prompt access and improved service, increased doctor- patient contact time, greater patient education, enhanced prevention and disease self-management, closer attention to routine health maintenance and performance measures, * Correspondence: kim_ha@stanfordalumni.org 1Pacific Northwest University of Health Sciences, College of Osteopathic Medicine, Yakima, WA, USA Full list of author information is available at the end of the article and the central role of patient and clinician experience within the Triple Aim: enhancing patient experience, im- proving population health, and reducing costs [1–3]. More recently, Bodenheimer and Sinsky recommended that “the Triple Aim be expanded to a Quadruple Aim, adding the goal of improving the work life of health care providers, including clinicians and staff [4].” We chose SMA as the overarching term to encompass shared visit, group appointment, group medical appoint- ment, group visit (GV), group medical clinic, shared in- group medical appointment, group prenatal care (GPNC) and group-based antenatal care. SMAs prioritize the deliv- ery of care within interprofessional environments utilizing © The Author(s). 2019 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated. Wadsworth et al. BMC Family Practice (2019) 20:97 Page 2 of 13 peer-to-peer interactions [5]. Multiple standardized SMA delivery models have been established, from the drop-in group medical appointment, cooperative health care clinic (CHCC) and physicals shared medical appointment, to CenteringPregnancy® (CP) and parenting visits [3, 6]. These visits frequently emphasize the “coproduction” roles of patients as experts in their own circumstances and health professionals as facilitators rather than fixers, thus fostering a shared experience of illness and health to bet- ter inform, empower, and support [2]. SMAs have garnered a body of evidence in chronic disease management and preventive care. The various interpretations of the group clinical model have been ap- plied to a wide array of settings and a myriad of health promotion and disease-focused visits, including patients with diabetes, hypertension, congestive heart failure, chronic lung disease, asthma, arthritis, stroke, kidney disease, cancer, hearing impairment, and prenatal care, among other conditions [7–15]. Several systematic reviews summarize the effects of SMAs on healthcare delivery, economic factors, and bio- physical outcomes. Health systems have begun to em- brace the need for this transformative approach in achieving patient goals [2, 16–18]. In an era recognizing the role of patient-centeredness in improving healthcare quality, numerous authors have highlighted the need for a review that addresses the impacts of SMAs on patient experience of care [3, 7, 16, 17, 19]. This review aims to meet this need by examining the patient experience from the published literature alongside an assessment of SMAs to improve biophysical outcomes in the adult pri- mary care setting. Analyzing the existing body of evidence for shared medical appointments, we sought to understand the op- portunities, barriers, and limitations to SMAs based on self-reported patient experience, a notable component of the Triple Aim [2]. Specifically, our goal was to highlight effective approaches for patients participating in SMAs and determinants of effectiveness. Methods librarian conducted pre- An experienced biomedical planned literature searches of PubMed, Cochrane Li- brary, PsycINFO, Cumulative Index of Nursing and Allied Health Literature (CINAHL), Web of Science, ClinicalTrials.gov, and Social Science Research Network (SSRN) for peer-reviewed publications, using controlled vocabulary, keywords, and text words (see Additional file 1 for search strategy details). The search was limited to publications from 1997 or after. We also searched grey literature, non-empirical reports, social science publica- tions, and citations from published systematic reviews. The search yielded 1359 papers, including qualitative, quantitative, and mixed-methods studies. Case studies, pilot/feasibility studies, protocols, opinions, or advocacy articles were excluded. Eligibility criteria and methods of analysis were specified a priori. Two researchers independently reviewed citation titles, abstracts, and full-text articles to determine eligibility as well as extracted the data and performed quality and risk of bias assessment on included articles, as detailed below. Before general use, we pilot-tested the abstraction form templates on a sample of included articles and then re- vised accordingly to ensure that all relevant data elements were captured. Disagreements were resolved by consensus of the two reviewers or by obtaining a third investigator’s opinion when consensus could not be reached. Studies were required to meet five process (p) and out- come (o) criteria: clinical intervention (o), clinician-led visit (p), patient experience of care (o), primary care (p), and availability of individual clinical consultation (p), as detailed below. Studies were excluded if any participants were < 18 years of age. To limit potential bias, we ex- cluded studies involving addiction medicine, substance dependence / rehabilitation treatment, inpatient settings (both short and long term) or chronic care clinics that implemented multiple interventions, and SMAs requir- ing management by a specialist. We deemed SMAs to be clinician led if led by an inde- pendent licensed prescriber or clinician. This included medical doctors (MDs), doctors of osteopathy (DOs), ad- vanced registered nurse practitioners (ARNPs), certified nurse midwives (CNMs), and in some regions, nurse practitioners (NPs). We verified prescriptive authority and care responsibility by consulting organizational web- sites from the countries in which our identified studies were conducted [20–22]. Our review emphasized biophysical metrics of adult pa- tients in primary care environments. The study team in- cluded articles focused on SMAs that implemented a clinical intervention, such as vital sign measurements, lab checks (e.g., hemoglobin A1c, lipid panels), medication ad- justments, or physical exams. We excluded studies if the intervention was limited to patient education, facilitation, peer-facilitated support groups, or group talk therapy. We tracked confounders within targeted studies, such as participant inclusion/exclusion criteria, local barriers to implementation, reimbursement framework, types of SMA interventions, and patient characteristics including language, culture, and socioeconomic status. In our consideration of quantitative research, we in- cluded only those studies with a comparative control group. Studies with quantitative primary outcomes were evaluated using the modified Jadad score, which assesses the overall quality of the individual studies, including risk of bias, and has shown high inter-rater reliability [23–26]. To evaluate qualitative studies, our team used the “Trustworthiness of Qualitative Inquiry” framework to Wadsworth et al. BMC Family Practice (2019) 20:97 Page 3 of 13 assess credibility, transferability, dependability, and ob- jectivity [27]. Inter-rater reliability was assessed during the data ex- traction phase via two-way mixed measures intraclass cor- relation (ICC) value for average agreement presented [28]. In consideration of ENTREQ and PRISMA frameworks for this mixed-methods systematic review, categorization of the extracted data informed a thematic synthesis [29– 32]. We did not perform a formal meta-analysis. Results Thirteen quantitative controlled trials, 11 qualitative pa- pers, and two mixed methods studies met inclusion cri- teria. Three models were identified: CHCC (five articles), SMA / GV (10 articles) and GPNC / CP (11 articles). Figure 1 shows the Preferred Reporting Items for Sys- tematic Reviews and Meta-Analyses (PRISMA) flowchart for all included studies [32]. Summary of included studies SMA / GV is the most frequently mentioned model in quantitative studies whereas the GPNC / CP model is the most common in qualitative studies in this re- is the least represented in view. The CHCC model this review (Table 1). Table 2 breaks down the included articles into locale, healthcare system, reimbursement model, study design, single site or multiple sites, and study duration. Table 3 provides details of the typical configuration of the three models included in this review: CHCC, SMA / GV, and GPNC / CP. Generally, CHCC has a larger group size compared to SMA / GV and GPNC / CP. Physician- led intervention teams were cited in most SMA / GV studies, whereas certified nurse midwives were most often cited as leaders of the GPNC / CP visits. Per inclusion criteria, all 26 articles reported patient satisfaction and experience (Table 4). Only one article reported outcomes for all four aims [8]. Patient experience and satisfaction Methodologies for tracking patient experience and satis- faction were grouped by data collection method into the following five categories: One-on-One Interviews (via tele- phone or in person), Focus Group Style Interviews, Self- Efficacy / Participation / Satisfaction Questionnaires, Diabetes-Related Quality of Life (DQoL) Related Scales; and Primary Care Assessment Tool / Trust in Provider Outcomes (Table 5). When comparing the results of the patient experience / satisfaction data in these 26 articles, the following six Citations identified through database searches (n = 1537) Citations identified through grey literature searches (n = 22) Additional citations identified through bibliography sources (n = 73) Citations available for initial screening (n = 1632) Titles and abstracts screened, after duplicates removed (n = 1359) Full-text articles assessed for eligibility (n = 299) Citations included in mixed- methods systematic review (n = 26) Citations excluded at title / abstract level (n = 1060) Full-text articles excluded, with reasons (n = 273) Quantitative articles (n = 13) Qualitative articles (n = 11) Mixed methods articles (n = 2) Fig. 1 The PRISMA flowchart for all included studies Wadsworth et al. BMC Family Practice (2019) 20:97 Page 4 of 13 Table 1 List of 26 included articles in the primary care setting, categorized by model of group clinic and study type Model: CHCC SMA / GV GPNC / CP Quantitative (13 articles) X X X X X Beck, 1997 Clancy, 2007 Jafari F, 2010 Junling, 2015 Kennedy, 2011 Naik, 2011 Scott, 2004 Tandon, 2013 Trento, 2001 Trento, 2002 Trento, 2004 Trento, 2005 Trento, 2010 Qualitative (11 articles) Andersson, 2012 Andersson, 2013 Capello, 2008 Clancy, 2003 Herrman, 2012 Kennedy, 2009 McDonald, 2014 McNeil, 2012 Novick, 2011 Raballo, 2012 Wong, 2015 Mixed-methods (2 articles) Heberlein, 2016 Krzywkowski-Mohn,2008 Total no. of articles (26) 5 X X X X X X X X X X 10 X X X X X X X X X X X 11 Abbreviations: CHCC Cooperative health care clinic, CP CenteringPregnancy®, GPNC Group prenatal care, GV Group visit, SMA Shared medical appointment major themes emerged (also see Additional file 2 for more details). Patient-clinician dynamic Overall, data on the patient-clinician dynamic that emerged during SMAs were positive. SMAs saw quantita- tive advantages over individual visits in domains ranging from improved communication to overall satisfaction with the visit [7, 15, 33]. In SMA environments, more time was allotted to discuss healthcare issues with the clinician compared to traditional individual visits, and physicians were perceived as less hurried [7, 14]. One study indicated that SMA experiences resulted in markedly enhanced trust in one’s primary care physician [33]. Qualitative feedback similarly supported the patient- clinician dynamic as a notable aspect of SMAs. Inter- views with CP patients indicated that extra time with cli- nicians helped them to develop strong, supportive, and positive relationships with their healthcare clinicians, and reduced anxiety about potentially not being familiar with the practitioner who would oversee their obstetric deliveries [9–11, 34]. Feedback from patients indicated that room for further improvement of the patient-clinician dynamic in SMAs lies in the avoidance of a paternalistic, didactic style of communication from the clinician leader [12]. Patients appreciated being empowered by their clinicians and preferred a more encouraging and empowering commu- nication style within their groups. Overall quality of care Multiple studies demonstrated that patients participating in SMAs were significantly more satisfied with their care than those in individual models of care [7, 13–15]. When compared to patients receiving traditional individ- ual care, those participating in SMAs were more likely to describe their overall quality of care as excellent, to feel that their care was meeting all their needs, and to feel that their care was well coordinated [8, 35]. No studies showed significant decreases in patient percep- tions of quality of care in SMAs. Overall quality of care was not a direct theme ex- tracted from qualitative investigations of SMAs. How- interviews from Herrman’s research on the CP ever, program revealed that “multiparous women frequently commented that [SMAs were] far superior to their pre- vious experiences” [11]. Quality of life Trento’s research thoroughly addressed the theme of quality of life, using a modified version of the Diabetes Quality of Life Measure (DQoL) questionnaire consisting of 39 questions ranked along a 5-point Likert scale. This assessment scale was used across all five of Trento’s arti- cles, and demonstrated consistent results over 10 years of varied research on SMAs for patients with Diabetes Mellitus, Type 2 (T2DM). In all five of Trento’s studies discussed in this paper, DQoL scores significantly im- proved among group participants while worsening or remaining the same in control subjects [36–40]. Sense of community Patients in multiple studies reported that the feeling that they were not alone in their experience was central to the positive impact of SMAs and persisted whether the subject of the SMA was pregnancy, navigation of the VA Wadsworth et al. BMC Family Practice (2019) 20:97 Page 5 of 13 Table 2 Characteristics of included studies in the primary care setting Study characteristics No. of studies, by medical condition N studies (participants) Diabetes 10 (1881) Country United States Canada Europe (Italy, Sweden) Middle East (Iran) Asia (China) Healthcare system Govt (VA, FQHC, NHS, PHD) Private (HMO, MCO) University-affiliated clinic Healthcare payment model Public (Medicaid, Medicare, govt funded) Private (fee-for-service, managed care) Uninsured /underinsured Study design Randomized controlled trial Non-randomized controlled trial Observational / interviews / focus groups Mixed methods Sites Single Multisite Study duration < 6 months 6 months 7 to 11 months 12 to 18 months 24 months > 2 years 4 (426) 0 6 (1455) 0 0 3 (362) 1 (120) 6 (1399) 8 (1575) 0 2 (306) 9 (1848) 0 1 (33) 0 9 (1066) 1 (815) 1 (87) 1 (120) 0 2(219) 3 (1169) 3 (286) HTN 2 (1262) 1 (58) 0 0 0 1 (1204) 2 (1262) 0 0 2 (1262) 0 0 1 (1204) 0 1 (58) 0 1 (58) 1 (1204) 1 (1204) 1 (58) 0 0 0 0 MCC 3 (645) 2 (616) 1 (29) 0 0 0 1 (29) 2 (616) 0 3 (645) 0 0 2 (616) 0 1 (29) 0 1 (321) 2 (324) 0 0 0 2 (350) 1 (295) 0 Pregnancy 11 (2010) 6 (926) 2 (21) 2 (435) 1 (628) 0 8 (1908) 3 (102) 0 8 (1908) 3 (102) 0 4 (1591) 1 (268) 5 (122) 1 (29) 4 (84) 7 (1926) 0 0 11 (2010) 0 0 0 Abbreviations: FQHC Federally qualified health center, HMO Health maintenance organization, HTN Hypertension, MCC Multiple chronic conditions, MCO Managed care organization, NHS National health service, PHD Public health district, VA Veterans Administration system, or hypertension [6, 10, 12, 33, 41–44]. Creation of community via SMAs supported patients’ emotional health by providing validation and stemming the isola- tion often experienced when managing chronic condi- tions. This sense of community was viewed as a benefit, though one study referenced a member who reported that at times she avoided discussion of “disturbing topics for fear that it would negatively impact her cohort” [34]. Patient empowerment / role in healthcare This body of research suggests that a strength of SMAs over usual care is the ability to engage and empower pa- tients as active participants in their own healthcare. This empowerment bore out in both qualitative and quantita- tive research participants. Quantitatively, patients reported that they were more able to participate in their care and had significant improvements on scales of Coping Skills and Health Distress as compared to their counterparts [13, 14, 43]. In the realm of qualitative analyses, it was de- scribed that patients felt they were better able to interpret their medical data, thus making them more likely to dis- cuss their issues with their clinicians [42]. Within the CP model, patients reported feeling “reassured, prepared, less anxious, and confident,” and they felt that the group ses- sions made them more proactive with respect to their own health [9]. Raballo’s research also indicated that after Wadsworth et al. BMC Family Practice (2019) 20:97 Page 6 of 13 Table 3 Typical configuration of group models, as represented by included studies in the primary care setting Model (no. of articles) CHCC (5) Duration of each group session 90– 120 min Duration of individual consultation 5–10 min each at end of group session Group size 6–20 SMA / GV(10) 60– 90 min Optional 10 mins each or 24 mins total allotted at end of group session 5–15 GPNC / CPa(11) 90– 120 min 10 mins each at beginning of group session 8–12 Clinical intervention Nonclinical components Intervention team Disciplines (no. of articles) Size Vital signs Lab results review and medical records update Medication management Preventive measures Scheduling Medical-related paperwork requested by pts Brief 1:1 visits with physician, as necessary Vital signs Lab results review and medical records update Routine lab test orders 1:1 indiv consultation with physician, as necessary Health risk assessment Medication management Referrals, coordination of public health services Vital signs Physical exam Routine prenatal screening and labs Routine ultrasound Flu vaccine (seasonal) Postpartum visit Individual assessments prior to prenatal care within group setting Socialization Health education Group cohesion Orientation and socialization Interactive health education Group cohesion Self-monitoring Group discussion Medication compliance Group discussion, self-care, skills- building Active tracking of pregnancy changes (done by pts) Tour of birth unit, labor and delivery nurse Pediatric care resources Postpartum reunion 2–5 2–7 PCP (5) Nurse, RN or diabetes nurse educator (5) Clinical pharmacist (2) PT, OT (2) Dietitian (2) Community health worker (1) 1–2 physicians (9) Nurse, NP, RN (2) Diabetes educator/ RD (4) Clin psychologist, psychopedagogist (3) 1–2 postgraduate med students (1) Others (2) 2 + others invited 1–2 CNMs (8) NP (3) Medical asst (3) Physician (2) Health / perinatal educator (1) Others (1) Abbreviations: CHCC Cooperative health care clinic, CNM Certified nurse midwife, CP CenteringPregnancy®, GPNC Group prenatal care, GV Group visit, NP Nurse practitioner, OT Occupational therapist, PCP Primary care physician, PT Physical therapist, RD Registered dietitian, RN Registered nurse, SMA Shared medical appointment aWk 5–10: First visit w/ nurse. Wk 10–12: First visit with clinician. Wk 12–16: Start CP program experiencing SMAs, patients were significantly more likely to describe an internal locus of control for their health than those followed by usual care [45]. communication between clinicians, decreased waiting times, increased opportunities for learning throughout their visits, and improved administrative support [41, 42, 46]. Access / efficiency Several articles also establish benefits of SMAs with respect to access and efficiency. Quantitatively, participants re- ported that appointments were easily scheduled “as soon as [they liked]” and were more likely to report that visit waiting time was acceptable [8, 14]. Qualitatively, patients described experiencing “more comprehensive services,” smoother Biophysical outcomes Less than half of the included articles reported biophys- ical outcomes by health condition—either diabetes mellitus (DM) or hypertension (HTN)—as summarized in Table 6 [36–40, 42, 43, 45, 47, 48]. These studies claimed significant and non-significant improvements in biophysical metrics; however, heterogeneity of study Table 4 Quadruple aim reported in included studies Model (no. of articles) CHCC (5) SMA / GV (10) GPNC / CP (11) No. of articles Patient experience Population health 5 10 11 2 1 3 Cost 2 1 0 Clinician experience 3 3 1 Abbreviations: CHCC Cooperative health care clinic, CP CenteringPregnancy®, GPNC Group prenatal care, GV Group visit, SMA Shared medical appointment Wadsworth et al. BMC Family Practice (2019) 20:97 Page 7 of 13 Table 5 Methods used to collect patient experience data Method 1:1 phone or in-person interviewsa Focus group style interviewsa Self-efficacy / participation / satisfaction questionnaires Diabetes-related quality of life scales (DQoL) Primary care assessment tool & trust in clinician outcomes Total: aAndersson 2012 is double coded as it included both 1:1 and group interviews No. of articles 10 3 6 6 2 27 populations, methods and outcomes did not allow data across studies to be combined and analyzed. (one article) This data subset was categorized into quantitative (seven articles), qualitative (two articles), and mixed to include additional studies methods details (Table 7). Eight articles had a control comparator of usual care while two articles (one qualitative study and one mixed methods study) only compared pre- and post-group intervention. Only one article utilized the CHCC model while the remaining nine articles were SMAs / GVs. From the ten studies included in this sub- set, the reported biophysical profile data varied, keeping with previous systematic reviews on SMAs by Booth et al. and Edelman et al. [17, 18]. Barriers to implementation Few studies addressed barriers, as shown in Additional file 3. Prior reviews by Edelman et al., Booth et al., and Jones et al. cite several barriers to implementation of SMAs overall, including patient participation and at- tendance, group dynamic incompatibilities, cost-benefit concerns, and staff/facilities inadequacies [16, 17, 49]. Prior studies cited poor attendance at SMAs [7, 13, 33]. In tracking attendance and patient-centered out- comes through different group visit formats, durations and patient populations, a great variation of attendance rates was found, as shown in Additional file 4. Inter-rater reliability As shown in Additional file 5, the ICC(2,k) inter-rater reliability values are 0.956 for Jadad-modified score of quantitative studies, 0.923 for trustworthiness score of qualitative studies, and indeterminable for mixed method studies due to sample size of n = 2 studies. Values greater than 0.90 indicate excellent reliability [28]. Table 6 Overview of biophysical data from available studies, categorized by health condition (no. of articles = 10) HbA1c FBG Lipids BP BMI Body First author, year Diabetes X HDL, TG X HDL X HDL, TG X TC, HDL, TG X TC, LDL, HDL, TG X TC, HDL, TG X LDL X X X Trento, 2001 Trento, 2002 Trento, 2004 Trento, 2005 Trento, 2010 Naik, 2011 X X X X X X Raballo, 2012 X Krzywkowski- Mohn, 2008 X Hypertension Junling, 2015 Capello, 2008 wt X X X X X X X X X X X X X SBP X X X CV risk DM Rx dosage Kidney Eye Foot Physical activity X X X retinopathy X insulin X Cr X ACR X Cr X foot ulcers X retinal exam X foot exam X Abbreviations: ACR Albumin/Creatinine ratio, BMI Body mass index, BP Blood pressure, Cr Creatinine, CV Cardiovascular, DM Diabetes mellitus, FBG Fasting blood glucose, HbA1c Glycated hemoglobin, HDL High-density lipoprotein, LDL Low-density lipoprotein, Rx Prescription, SBP Systolic blood pressure, TC Total cholesterol, TG Triglycerides Wadsworth et al. BMC Family Practice (2019) 20:97 Page 8 of 13 Table 7 Biophysical data from available studies, categorized by research type (no. of articles = 10) First author, year Quantitative Model Health cond(s) Sample size (n) Biophysical measures Reported findings (with p-values) Junling, 2015 CHCC HTN 600 group, 604 control ● BP ● BMI SBP decreased significantly in both group (p < 0.001) and control (p = 0.001) from baseline to follow-up, although decreases in group > control. ● Physical activity DBP decreased significantly in group (p = 0.001) but did not decrease significantly in control. Trento, 2001 SMA / GV T2DM 56 group, 56 control ● HbA1c ● BMI ● HDL ● Fasting TG Trento, 2002 SMA / GV T2DM 56 group, 56 control ● Dosage of anti- hyperglycemic agents ● Body wt, BP and CV risk ● Metabolic control: - HbA1c - BMI - HDL - Retinopathy BMI did not change in both. Increases in physical activity in group (p < 0.001) more remarkable than in control. HbA1c stable in group, worsened in control (p < 0.002). Tendency toward lower BMI in group (p = 0.06). HDL cholesterol initially similar in both but later lower in group only (p < 0.05). Trend toward lower TG in group (p = 0.053). Dosage of hypoglycemic agents decreased (p < 0.001) among group compared to control. Body wt (p < 0.001) and BMI (p < 0.001) decreased in group but not in control. Similar reductions in BP and CV risk in group vs control, but diff significant only for DBP (p < 0.001). Significant decrease in HbA1c (p < 0.001) in group. HDL increased (p < 0.001) in group but not in control. Retinopathy progressed less in group (p = 0.009). Trento, 2004 SMA / GV T2DM (NIDDM) 56 group, 56 control ● HbA1c ● BMI ● HDL, TG ● Cr HbA1c remained stable in group but progressively increased among control (p < 0.001). BMI, HDL, TG and Cr improved over 5 yrs. in group, but not significantly different from control. Trento, 2005 SMA / GV T2DM 31 group, 31 control ● HbA1c HbA1c decreased in both, though not significantly. Trento, 2010 SMA / GV T2DM (NIDDM) 421 group, 394 control TC decreased in controls (p < 0.05), while HDL increased in group (p = 0.027). No significant modifications in other clinical variables monitored (body wt, BMI, FBG, insulin dosage, TG, ACR, foot ulcers). FBG, HbA1c, TC, TG, LDL cholesterol, SBP, DBP, and BMI decreased in group from baseline to year 4 compared to control (p < 0.001, for all measures). HDL increased in group (p < 0.001). Cr did not change significantly in group. BMI, HbA1c, TG, and Cr increased in control, whereas total, HDL, and LDL cholesterol and SBP did not change and DBP decreased. ● Lipids (TC, HDL, TG) ● Body wt, BMI ● FBG ● Insulin dosage ● ACR ● Foot ulcers ● FBG ● HbA1c ● TC, LDL, HDL, TG ● BP ● BMI ● Cr Naik, 2011 SMA / GV T2DM 45 group, 42 control ● HbA1c ● SBP ● BMI Significantly greater improvements in HbA1c immediately following active Intervention and persisted at 1-year follow-up (p = 0.05). SBP and BMI were only reported at baseline, but not significantly different between both. Wadsworth et al. BMC Family Practice (2019) 20:97 Page 9 of 13 Table 7 Biophysical data from available studies, categorized by research type (no. of articles = 10) (Continued) First author, year Qualitative Reported findings (with p-values) Biophysical measures Sample size (n) Health cond(s) Model Capello, 2008 SMA / GV HTN 58 group (no control) Raballo, 2012 SMA / GV T1DM, T2DM 121 group, 121 control Mixed Methods Krzywkowski- Mohn, 2008 SMA / GV T2DM 33 group (no control) ● BP Significant effects on SBP and DBP (p < 0.01). ● HbA1c ● Lipids (TC, HDL, TG) ● FBG ● BMI HbA1c lower in T1DM group than in control (p = 0.001) and not significantly so in T2DM (NS). Lower HDL in T1DM control (p = 0.002), but no other significant differences among both. Lower HbA1c after group intervention (p < 0.05). Diabetic clinical indicators: ● HbA1c ● LDL ● BP ● Retinal exam Increase in diabetic eye exams. ● Foot exam Lower LDL after 18 mos (p < 0.05). No significant diff. in SBP or DBP after 18 mos. No diff in diabetic foot exams (96.9% pre + post). Abbreviations: ACR Albumin/Creatinine ratio, BMI Body mass index, BP Blood pressure, Cr Creatinine, CV Cardiovascular, DBP Diastolic blood pressure, FBG Fasting blood glucose, HbA1c Glycated hemoglobin, HDL High density lipoprotein, HTN Hypertension, LDL Low density lipoprotein, NIDDM Non-insulin dependent diabetes mellitus, SBP Systolic blood pressure, T1DM Diabetes mellitus, type 1, T2DM Diabetes mellitus, type 2, TC Total cholesterol, TG Triglycerides Discussion This review limited SMA models to three general cat- egories: cooperative health care clinic, shared medical appointment / group visit, and group prenatal care / the focus on group CenteringPregnancy®. To meet in- intervention, we considered visits clinical cluded the following clinical components: review of labs, medication management, physical examination, or other medical interventions. From a strength of evidence perspective, 16 of the studies reflected a ran- domized controlled design and one non-randomized controlled design. The remaining nine studies were cohort and case study designs, with a median study duration of 12 months. that As SMAs are generalizable to primary care environ- ments, we prioritized reviews that included Internal Medicine, Obstetrics/Gynecology, Family Medicine, and Psychiatry. Though non-clinician-led SMAs have been applied in myriad ways in primary care settings, such as group-based acupuncture clinics, group psy- chotherapy for post-traumatic stress disorder and group interventions for disabled adults, we excluded them to evaluate SMAs as a variation of clinician-led primary care. To the best of our knowledge, our current review up- dates the evidence base to date and provides a necessary segue to patient-oriented outcomes. In the spirit of the Triple Aim, SMAs uniquely enhance patient-centered experience, thus we limited our review to settings that provide individual primary care consultation alongside the group visit. Individual consultation provides a re- served space for private concerns. This is an important distinction as privacy concerns have been a prominent drawback of the model identified by prior research [13, 15, 34]. We prioritized this element, recognizing the trust it fosters in the patient-clinician relationship. Summary of findings In sum, designing, promoting, and running SMAs from tested and proven formats proves to be vital for implemen- tation. Model and content fidelity demonstrate significant outcome improvement, most notably in the prenatal care and birth outcomes through the CenteringPregnancy® group process. Standardized training also improves facilita- tion of group care. Therefore, clinicians learning to facilitate group care are encouraged to receive training in facilitative leadership with emphasis on the role that a participatory at- mosphere has in improving outcomes [50]. Several models describe a physical design component to enhance the effect on patient experience or group process [3, 42, 51]. Some studies use displayed patient biophysical data for comparison and a visual aid for decision-making. Patient seating design has also been identified as a driver, both circular and U-shaped formats. Krzywokwski-Mohn stipulates that SMAs occur with participants seated around a circular conference table, with no one at the “head of the table,” balancing power and significantly influencing SMA participant outcomes [42]. Additionally, the emergence of the patient-centered medical home motivates improvement in patient educa- tion, experience of care, and measurable outcomes Wadsworth et al. BMC Family Practice (2019) 20:97 Page 10 of 13 without increasing clinical workload [3]. The interprofes- sional team plays a prominent role in SMAs across the lit- erature, including nurses, nutritionists, NPs, pharmacists, physical therapists, PAs, primary healthcare coordinators and nurse midwives [7, 8, 14, 34, 52]. Despite these reallo- cation of tasks, roles, and resources, SMAs demonstrate efficacy and feasibility across a wide range of healthcare systems [39, 53]. Despite SMAs objectively providing patients more time with their clinicians, the degree to which this affects satis- faction is unknown and patient characteristics and outside influences can affect satisfaction outcomes [7, 13, 49, 54]. Furthermore, evaluating and effectively responding to the social determinants of health requires improved identifica- tion of patient needs and outcomes assessment [55]. Nonetheless, our evaluation includes consideration of pa- tient experience fundamental for evaluating health-related quality of life, including disease-related health locus of control, health behaviors, self-efficacy, and other measures of patient perspective of care and quality of life. Lastly, studies emphasizing biophysical outcomes re- port statistically significant improvement in at least one biophysical metric, yet are too heterogeneous to com- pare across studies. Nonetheless, results are consistent with other systematic reviews by Booth et al., Edelman et al., and Jones et al. [17, 18, 49]. Limitations of review Our inclusion criteria and focus on the primary care context limited the number of articles that we evaluated in this review, which may impact the generalizability of our conclusions. Previous systematic reviews looked at a broader number of articles, though their approach also introduced more heterogeneity [17, 18, 49]. Single center studies, representing the majority for our included arti- cles on diabetes patients, may also limit generalizability. We also note that half of our included articles for the SMA / GV format were authored by the same researcher [36–40]. Other previous reviews have mentioned the im- possibility of blinding the participant and clinician / care team. Given that trials of SMA interventions cannot be designed in a traditional double-blinded manner, our quality assessment scores for quantitative studies could only receive a maximum of seven out of a total of eight points on the modified Jadad score. However, a few studies described minimizing performance bias by hav- ing the same clinician and care team manage the same intervention and control subjects and by measuring outcomes blindly for the treatment group. Furthermore, there may be sampling bias in nonrandomized con- trolled trials as well as focus groups and interviews due to the possibility that patients who are high frequency attenders may self-select to be included in the interven- subjects who have negative tion group; likewise, experiences with SMAs may decline to be interviewed or refuse to be randomized into the intervention group. Moreover, information bias may have appeared due to variation in attendance and/or completion of visits within our sample. Critiques exist concerning the evaluation of patient ex- perience through patient satisfaction measures. Aside from a lack of agreement on a converging definition of “satisfaction,” there are methodological challenges in re- liably and precisely measuring and interpreting percep- tions of the healthcare environment (survey content, mode and timing of survey administration, bias, con- founding, need for post-hoc adjustment, and subjective nature of interpersonal experiences, including patient- clinician communication as a unique dimension of qual- ity). Despite these challenges, patient experience has a meaningful role in quality improvement discussions and determination of perceived quality and sense of commu- nity [56]. Implications for practice, policy, and future research Improved resilience and coping skills, in concert with pa- tient agency and activation, are valuable outcomes of the spectrum of SMAs [34]. The primary care environment is an optimum setting to build the necessary trust, health lit- eracy, and awareness of health beliefs required for suc- cessful intersection with the broader healthcare system [35, 38]. Honoring adult learning strategies often requires nonclinical skill sets for interdisciplinary care clinicians [38]; yet, few studies focused on interprofessional practice despite widespread presence across differing SMA models. SMAs emphasize patient empowerment through peer ac- countability, socialization, and appreciation of local cul- tural context as well as patients’ familiarity and comfort with the setting [40, 43, 53]. Engaging group members in the design of these SMAs can maximize responsiveness to [43]. cultural context and acceptability of GPNC / CP have demonstrated efficacy in increasing health-related knowledge, social support, personal locus of control, emotional care, and self-care [52, 57]. the model In general, to improve quality and validity of report- ing patient experience as well as improved reporting of population health outcomes, we recommend longer duration of follow up in each study setting. We also recommend specific evaluation of team-based care, in- cluding perspectives of administrators and supporting clinical staff. As provision of healthcare is a service, measures of quality should include assessment of the extent to which patients and care teams reach a com- mon understanding of treatment course and health out- comes [2]. This intersection of shared well-being with health improvement warrants further evaluation to optimize healthcare delivery models, such as SMAs, to achieve the quadruple aim. Wadsworth et al. BMC Family Practice (2019) 20:97 Page 11 of 13 Conclusions Shared medical appointments are increasingly employed in primary care settings. This mixed-methods systematic review concludes that accepting and implementing this nontraditional approach by both patients and clinicians can yield measurable improvements in patient trust, pa- tient perception of quality of care and quality of life, and relevant biophysical measurements of clinical parameters. Compared to usual care, SMAs have a greater ability to engage and empower patients as active participants in their own healthcare while improving patient access and healthcare efficiency. The cumulative benefits of SMAs are most notable when implemented within a conducive environment such as a PCMH. No singular model of SMA best serves all settings. Similarly, there does not appear to be a priority set of their outcome measures nor consistent means for evaluation from our review. Our analysis indicates that both quantitative and qualitative methods are equally valid for evaluating patient experience. Further refine- ment of this healthcare delivery model will benefit from standardizing measures of patient satisfaction and clin- ical outcomes. Not surprisingly, critiques and cost-benefit concerns remain. Demonstration of global payment models result- ing in improved population health outcomes alongside economies of scale may be essential for wider acceptance of SMAs. We recommend further evaluation of the en- ablers and barriers to advance SMA integration in pri- mary care practice settings. We also recommend more thorough and longitudinal evaluations to better describe the consumer-minded approach for care delivery design and responsiveness to the voice of the customer to achieve the most efficient models possible. Additional files Additional file 1: Database search strategies (DOCX 27 kb) Additional file 2: Description of data: Reported significant findings related to patient experience and satisfaction, as reported in included articles (DOCX 31 kb) Additional file 3: Description of data: Barriers to implementation from available studies (no. of articles = 8) (DOCX 28 kb) Additional file 4: Description of data: SMA patient-centered variables vs. attendance and outcomes (no. of articles = 26) (DOCX 40 kb) Additional file 5: Inter-rater reliability of included articles using two-way mixed measures intraclass correlation (ICC) value for average agreement presented. (DOCX 29 kb) Abbreviations CHCC: Cooperative health care clinic; CP: CenteringPregnancy®; DQoL: Diabetes-Related Quality of Life; GPNC: Group prenatal care; GV: Group visit; PCMH: Patient centered medical home; SMA: Shared medical appointment Acknowledgments The authors thank Mary Giovanini for her help with full-text citations; Drs. Sean Cleary, MD, PhD, and Jennifer Best, MD, for their thorough edits of our manuscript; Dr. William Elliott, MD, PhD, for his critical suggestions on quality assessment of included articles; Dr. Bernadette Howlett, PhD, for her early in- put on our research methodology; Dr. Michele McCarroll, PhD, Carla S. Case and Anita Quintana, MA, for their kind assistance and support; and Tracy Dana, MLS, and Sarah Safranek, MLIS, for reviewing our literature search strategies. Authors’ contributions All listed authors significantly contributed to this project. KHW, AKC and ASH developed the study protocol. KHW, TGA and ASH conducted the title and abstract screening. KHW, TGA, AEP, and ASH conducted the full-text screen- ing, data extraction, quality assessments, and data synthesis. BLH provided the content expertise. AKC is the biomedical librarian who conducted the lit- erature search and managed the citations. All authors had access to the data, played a role in writing the manuscript, and read and approved the final manuscript. Funding This project was made possible with a Mapping the Landscape, Journeying Together grant from the Arnold P. Gold Foundation (APGF). The APGF did not have any role in the design of the study, collection, analysis and interpretation of data, nor writing the manuscript. Availability of data and materials Table 1 provides a list of the 26 included papers and Additional file 1 shows the database search strategy. Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. Author details 1Pacific Northwest University of Health Sciences, College of Osteopathic Medicine, Yakima, WA, USA. 2Family Health Care of Ellensburg, Ellensburg, WA, USA. 3Multnomah County Health Department, Oregon Health and Science University–Portland State University School of Public Health, Portland, OR, USA. Received: 23 February 2018 Accepted: 31 May 2019 References 1. Berwick DM, Nolan TW, Whittington J. The triple aim: care, health, and cost. Health Aff Proj Hope. 2008;27:759–69. Batalden M, Batalden P, Margolis P, Seid M, Armstrong G, Opipari-Arrigan L, et al. Coproduction of healthcare service. BMJ Qual Saf. 2016;25:509–17. Noffsinger EB. Group visits -- the “secret sauce” of the medical home. Med Home News. 2013;5:1,6-8. Bodenheimer T, Sinsky C. From triple to quadruple aim: care of the patient requires care of the provider. Ann Fam Med. 2014;12:573–6. Noffsinger EB. Running group visits in your practice. New York: Springer; 2009. Novick G. CenteringPregnancy and the current state of prenatal care. J Midwifery Womens Health. 2004;49:405–11. Scott JC, Conner DA, Venohr I, Gade G, McKenzie M, Kramer AM, et al. Effectiveness of a group outpatient visit model for chronically ill older health maintenance organization members: a 2-year randomized trial of the cooperative health care clinic (structured abstract). J Am Geriatr Soc. 2004;52: 1463–70. Beck A, Scott J, Williams P, Robertson B, Jackson D, Gade G, et al. A randomized trial of group outpatient visits for chronically ill older HMO 2. 3. 4. 5. 6. 7. 8. Wadsworth et al. BMC Family Practice (2019) 20:97 Page 12 of 13 9. members: the cooperative Health care clinic. J Am Geriatr Soc. 1997;45: 543–9. Heberlein EC, Picklesimer AH, Billings DL, Covington-Kolb S, Farber N, Frongillo EA. Qualitative comparison of women’s perspectives on the functions and benefits of group and individual prenatal care. J Midwifery Womens Health. 2016;61:224–34. 10. McDonald SD, Sword W, Eryuzlu LE, Biringer AB. A qualitative descriptive study of the group prenatal care experience: perceptions of women with low-risk pregnancies and their midwives. BMC Pregnancy Childbirth. 2014;14:334. 11. Herrman JW, Rogers S, Ehrenthal DB. Women’s perceptions of centering pregnancy: a focus group study. MCN Am J Matern Nurs. 2012;37:19–28. 14. 13. 12. Andersson E, Christensson K, Hildingsson I. Parents’ experiences and perceptions of group-based antenatal care in four clinics in Sweden. Midwifery. 2012;28:442–8. Kennedy HP, Farrell T, Paden R, Hill S, Jolivet RR, Cooper BA, et al. A randomized clinical trial of group prenatal care in two military settings. Mil Med. 2011;176:1169–77. Tandon SD, Cluxton-Keller F, Colon L, Vega P, Alonso A. Improved adequacy of prenatal care and healthcare utilization among low-income Latinas receiving group prenatal care. J Women's Health. 2013;22:1056–61. Jafari F, Eftekhar H, Mohammad K, Fotouhi A. Does group prenatal care affect satisfaction and prenatal care utilization in Iranian pregnant women? Iran J Public Health. 2010;39:52–62. Edelman D, McDuffie JR, Oddone E, Gierisch JM, Nagi A, Williams JWJ. Shared medical appointments for chronic medical conditions: a systematic review. Washington (DC): Department of Veterans Affairs (US); 2012. https://www.ncbi. nlm.nih.gov/books/NBK99785/ 15. 16. 17. Booth A, Cantrell A, Preston L, Chambers D, Goyder E. What is the evidence for the effectiveness, appropriateness and feasibility of group clinics for patients with chronic conditions? A systematic review. Southampton: NIHR Journals Library; 2015. http://www.ncbi.nlm.nih.gov/books/NBK333454/. Accessed 9 Feb 2016 Edelman D, Gierisch JM, McDuffie JR, Oddone E, Williams JW. Shared medical appointments for patients with diabetes mellitus: a systematic review. J Gen Intern Med. 2015;30:99–106. 18. 19. Berwick DM. What “patient-centered” should mean: confessions of an extremist. Health Aff Proj Hope. 2009;28:w555–65. 20. Where Can Nurse Practitioners Work Without Physician Supervision? - Blog. 2016. https://onlinenursing.simmons.edu/nursing-blog/nurse-practitioners- scope-of-practice-map/. Accessed 5 Feb 2017. Survey of European Midwifery Regulators: Second Issue. 2010. http://www. ordre-sages-femmes.fr/wp-content/uploads/2015/11/Etude-NEMIR-2010-EN. pdf. Accessed 7 Feb 2017. 21. 22. CNM independent prescribing map | NCSBN. https://www.ncsbn.org/index. 23. htm. Accessed 5 Feb 2017. Jadad AR, Moore RA, Carroll D, Jenkinson C, Reynolds DJ, Gavaghan DJ, et al. Assessing the quality of reports of randomized clinical trials: is blinding necessary? Control Clin Trials. 1996;17:1–12. 24. Oremus M, Wolfson C, Perrault A, Demers L, Momoli F, Moride Y. Interrater reliability of the modified Jadad quality scale for systematic reviews of Alzheimer’s disease drug trials. Dement Geriatr Cogn Disord. 2001;12:232–6. Johannsen M, Farver I, Beck N, Zachariae R. The efficacy of psychosocial intervention for pain in breast cancer patients and survivors: a systematic review and meta-analysis. Breast Cancer Res Treat. 2013;138:675–90. 25. 26. Greenland S. On the bias produced by quality scores in meta-analysis, and a hierarchical view of proposed solutions. Biostatistics. 2001;2:463–71. 27. Howlett B, Rogo E, Shelton TG. Implementation and evaluation in evidence based practice. In: Evidence based practice for Health professionals. Burlington: Jones & Bartlett Publishers; 2013. p. 309–58. Koo TK, Li MY. A guideline of selecting and reporting intraclass correlation coefficients for reliability research. J Chiropr Med. 2016;15:155–63. 28. 29. Voils CI, Sandelowski M, Barroso J, Hasselblad V. Making sense of qualitative and quantitative findings in mixed research synthesis studies. Field Methods. 2008;20:3–25. 30. Creswell JW, Klassen AC, Plano Clark VL, Smith KC, For The Office of Behavioral and Social Sciences Research. Best practices for mixed methods research in the health sciences: National Institutes of Health; 2011. https:// obssr.od.nih.gov/training/online-training-resources/mixed-methods- research/. Accessed 19 Oct 2015 Tong A, Flemming K, McInnes E, Oliver S, Craig J. Enhancing transparency in reporting the synthesis of qualitative research: ENTREQ. BMC Med Res Methodol. 2012;12:181. 31. 32. Moher D, Liberati A, Tetzlaff J, Altman DG. The PRISMA group. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. PLoS Med. 2009;6:e1000097. 33. Clancy DE, Cope DW, Magruder KM, Huang P, Salter KH, Fields AW. Evaluating group visits in an uninsured or inadequately insured patient population with uncontrolled type 2 diabetes. Diabetes Educ. 2003;29:292–302. 34. Novick G, Sadler LS, Kennedy HP, Cohen SS, Groce NE, Knafl KA. Women’s experience of group prenatal care. Qual Health Res. 2011;21:97–116. 35. Clancy D, Yeager D, Huang P, Magruder K. Further evaluating the 36. 37. 38. 39. 40. acceptability of group visits in an uninsured or inadequately insured patient population with uncontrolled type 2 diabetes. Diabetes Educ. 2007;33:309–14. Trento M, Passera P, Bajardi M, Tomalino M, Grassi G, Borgo E, et al. Lifestyle intervention by group care prevents deterioration of type II diabetes: a 4- year randomized controlled clinical trial. Diabetologia. 2002;45:1231–9. Trento M, Passera P, Borgo E, Tomalino M, Bajardi M, Brescianini A, et al. A 3-year prospective randomized controlled clinical trial of group care in type 1 diabetes. Nutr Metab Cardiovasc Dis NMCD. 2005;15:293–301. Trento M, Passera P, Tomalino M, Bajardi M, Pomero F, Allione A, et al. Group visits improve metabolic control in type 2 diabetes - a 2-year follow- up. Diabetes Care. 2001;24:995–1000. Trento M, Gamba S, Gentile L, Grassi G, Miselli V, Morone G, et al. Rethink organization to iMprove education and outcomes (ROMEO): a multicenter randomized trial of lifestyle intervention by group care to manage type 2 diabetes. Diabetes Care. 2010;33:745–7. Trento M, Passera P, Borgo E, Tomalino M, Bajardi M, Cavallo F, et al. A 5- year randomized controlled study of learning, problem solving ability, and quality of life modifications in people with type 2 diabetes managed by group care. Diabetes Care. 2004;27:670–5. 42. 41. McNeil DA, Vekved M, Dolan SM, Siever J, Horn S, Tough SC. Getting more than they realized they needed: a qualitative study of women’s experience of group prenatal care. BMC Pregnancy Childbirth. 2012;12:17. Krzywkowski-Mohn SM. Diabetic control and patient perception of the scheduled in group medical appointment at the Cincinnati veterans administration medical center: University of Cincinnati; 2008. https://etd. ohiolink.edu/pg_10?0::NO:10:P10_ACCESSION_NUM:ucin1210103113. Accessed 9 Aug 2016 Junling G, Yang L, Junming D, Pinpin Z, Hua F. Evaluation of group visits for Chinese hypertensives based on primary health care center. Asia-Pac J Public Health Asia-Pac Acad Consort Public Health. 2015;27: NP350–60. Kennedy HP, Farrell T, Paden R, Hill S, Jolivet R, Willetts J, et al. ‘I wasn’t alone’--a study of group prenatal care in the military. J Midwifery Womens Health. 2009;54:176–83. 43. 44. 45. Raballo M, Trevisan M, Trinetta AF, Charrier L, Cavallo F, Porta M, et al. A study of patients’ perceptions of diabetes care delivery and diabetes: propositional analysis in people with type 1 and 2 diabetes managed by group or usual care. Diabetes Care. 2012;35:242–7. 46. Wong ST, Browne A, Lavoie J, Macleod MLP, Chongo M, Ulrich C. Incorporating group medical visits into primary healthcare: are there benefits? Healthc Policy Polit Santé. 2015;11:27–42. 47. Naik AD, Palmer N, Petersen NJ, Street RL, Rao R, Suarez-Almazor M, et al. Comparative effectiveness of goal setting in diabetes mellitus group clinics: randomized clinical trial. Arch Intern Med. 2011;171:453–9. 48. Capello J. An evaluation of the doctor interactive group medical Appointment : assessing changes in health behaviors attributed to an integrated healthcare model. 2008. https://repositories.lib.utexas.edu/ handle/2152/17781. Accessed 21 Jan 2016. Jones KR, Kaewluang N, Lekhak N. Group visits for chronic illness management: implementation challenges and recommendations. Nurs Econ. 2014;32:118–34 147. 49. 50. Novick G, Reid AE, Lewis J, Kershaw TS, Rising SS, Ickovics JR. Group prenatal care: model fidelity and outcomes. Am J Obstet Gynecol. 2013; 209:112.e1–6. 51. Northern Health. The group medical appointment manual first edition. 2007. http://www.hqontario.ca/Portals/0/documents/qi/learningcommunity/ roadmap%20resources/advanced%20access%20and%20efficiency/step%205/ pc-nha-group-medical-appointments-manual-en.pdf. Accessed 21 Jan 2016. 52. Andersson E, Christensson K, Hildingsson I. Mothers’ satisfaction with group antenatal care versus individual antenatal care – a clinical trial. Sex Reprod Healthc. 2013;4:113–20. Wadsworth et al. BMC Family Practice (2019) 20:97 Page 13 of 13 53. Campbell BB, Shah S, Gosselin D. Success with Men’s educational group appointments (MEGA): subjective improvements in patient education. Am J Mens Health. 2009;3:173–8. 54. Aharony L, Strasser S. Patient satisfaction: what we know about and what we still need to explore. Med Care Rev. 1993;50:49–79. 55. Asadi-Lari M, Tamburini M, Gray D. Patients’ needs, satisfaction, and health related quality of life: towards a comprehensive model. Health Qual Life Outcomes. 2004;2:32. 56. Manary MP, Boulding W, Staelin R, Glickman SW. The patient experience and health outcomes. N Engl J Med. 2013;368:201–3. 57. Baldwin K, Phillips G. Voices along the journey: midwives’ perceptions of implementing the CenteringPregnancy model of prenatal care. J Perinat Educ. 2011;20:210–7. Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
null
10.1186_s13059-023-02963-4.pdf
Availability of data and materials The results published here are in part based upon data generated by the TCGA Research Network (https:// www. cancer. gov/ tcga), METABRIC (https:// ega‑ archi ve. org/ studi es/ EGAS0 00000 00083), MSK‑IMPACT (https:// www. mskcc. org/ msk‑ impact) or deposited at cBioPortal (https:// www. cbiop ortal. org/). The following expression datasets from the Gene Expression Omnibus (GEO, https:// www. ncbi. nlm. nih. gov/ geo/) have also been employed: GSE114012 [48], GSE131594 [45], GSE137912 [12], GSE152699 [49], GSE75367 [47], GSE83142 [46], GSE93991 [44], GSE134836 [13], GSE134838 [13], GSE134839 [13], GSE124854 [93], GSE135215 [94], GSE99116 [93], GSE178839 [149], GSE149224 [100], GSE139944 [102], GSE191127 [150], GSE109211 [151], GSE50509 [152], GSE65185 [153], GSE66399 [154], GSE68871 [155] and GSE99898 [156]. The GEO datasets employed in the analyses are summarised in Additional file 1: Tables S2 and S3. All codes developed for the purpose of this study can be found at the following repository, released under a GNU Gen‑ eral Public License v3.0 at github: https:// github. com/ secri erlab/ Cance rG0Ar rest [157] and Zenodo (doi: 1
Availability of data and materials The results published here are in part based upon data generated by the TCGA Research Network ( https:// www. cancer. gov/ tcga ), METABRIC ( https:// ega-archi ve. org/ studi es/ EGAS0 00000 00083 ), MSK-IMPACT ( https:// www. mskcc. org/ msk- impact ) or deposited at cBioPortal ( https:// www. cbiop ortal. org/ ). The following expression datasets from the Gene Expression Omnibus (GEO, https:// www. ncbi. nlm. nih. gov/ geo/ ) have also been employed: GSE114012 [48] , GSE131594 [45] , GSE137912 [12] , GSE152699 [49] , GSE75367 [47] , GSE83142 [46] , GSE93991 [44] , GSE134836 [13] , GSE134838 [13] , GSE134839 [13] , GSE124854 [93] , GSE135215 [94] , GSE99116 [93] , GSE178839 [149], GSE149224 [100], GSE139944 [102], GSE191127 [150], GSE109211 [151], GSE50509 [152], GSE65185 [153], GSE66399 [154], GSE68871 [155] and GSE99898 [156]. The GEO datasets employed in the analyses are summarised in Additional file 1: Tables S2 and S3 . All codes developed for the purpose of this study can be found at the following repository, released under a GNU General Public License v3.0 at github: https:// github. com/ secri erlab/ Cance rG0Ar rest [157] and Zenodo (doi: 10. 5281/ zenodo. 78406 72 ) [158].
Wiecek et al. Genome Biology (2023) 24:128 https://doi.org/10.1186/s13059-023-02963-4 RESEARCH Genome Biology Open Access Genomic hallmarks and therapeutic implications of G0 cell cycle arrest in cancer Anna J. Wiecek1, Stephen J. Cutty2, Daniel Kornai1, Mario Parreno‑Centeno1, Lucie E. Gourmet1, Guidantonio Malagoli Tagliazucchi1, Daniel H. Jacobson1,3, Ping Zhang4, Lingyun Xiong4, Gareth L. Bond5, Alexis R. Barr2,6 and Maria Secrier1* *Correspondence: m.secrier@ucl.ac.uk 1 UCL Genetics Institute, Department of Genetics, Evolution and Environment, University College London, London, UK 2 Institute of Clinical Sciences, Faculty of Medicine, Imperial College London, London, UK 3 UCL Cancer Institute, Paul O’Gorman Building, University College London, London, UK 4 Wellcome Centre for Human Genetics, University of Oxford, Oxford, UK 5 Institute of Cancer and Genomic Sciences, University of Birmingham, Edgbaston, Birmingham, UK 6 Cell Cycle Control Team, MRC London Institute of Medical Sciences (LMS), London, UK Abstract Background: Therapy resistance in cancer is often driven by a subpopulation of cells that are temporarily arrested in a non‑proliferative G0 state, which is difficult to capture and whose mutational drivers remain largely unknown. Results: We develop methodology to robustly identify this state from transcriptomic signals and characterise its prevalence and genomic constraints in solid primary tumours. We show that G0 arrest preferentially emerges in the context of more stable, less mutated genomes which maintain TP53 integrity and lack the hallmarks of DNA damage repair deficiency, while presenting increased APOBEC mutagenesis. We employ machine learning to uncover novel genomic dependencies of this process and validate the role of the centrosomal gene CEP89 as a modulator of proliferation and G0 arrest capacity. Lastly, we demonstrate that G0 arrest underlies unfavourable responses to various therapies exploiting cell cycle, kinase signalling and epigenetic mechanisms in single‑cell data. Conclusions: We propose a G0 arrest transcriptional signature that is linked with therapeutic resistance and can be used to further study and clinically track this state. Keywords: Cell cycle arrest, G0, Cancer, Persister cells, Genomic dependencies, Machine learning, Data integration, Bulk/single‑cell sequencing Background Tumour proliferation is one of the main hallmarks of cancer development [1] and has been extensively studied. While most of the cells within the tumour have a high prolif- erative capacity, occasionally under stress conditions, some cells will become arrested temporarily in the G0 phase of the cell cycle, in a reversible state often referred to as ‘quiescence’, ‘dormancy’, ‘diapause-like’ or a potentially irreversible state called ‘senes- cence’, where they maintain minimal basal activity [2–5]. It has been proposed that G0 arrest enables cells to become resistant to anti-cancer compounds that target actively dividing cells, such as chemotherapy [5–7]. Moreover, a drug-tolerant ‘persister’ cell © The Author(s) 2023. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the mate‑ rial. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/. The Creative Commons Public Domain Dedication waiver (http:// creat iveco mmons. org/ publi cdoma in/ zero/1. 0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data. Wiecek et al. Genome Biology (2023) 24:128 Page 2 of 35 state represented by slow cycling, entirely quiescent or even senescent cells [4, 8–11] has been observed in a variety of pre-existing or acquired resistance scenarios, also in the context of targeted therapies [12, 13]. As neoplastic cells evolve, G0 arrest can also be employed as a mechanism to facilitate immune evasion [14, 15] or adaptation to new environmental niches during metastatic seeding [16, 17]. In the context of disseminated tumour cells, these G0 cycle arrest states can facilitate minimal residual disease, a major cause of relapse in the clinic [18]. Although G0 arrest is a widely conserved cellular state, essential for the normal devel- opment and homeostasis of eukaryotes [2, 19], and has been extensively studied in a variety of organisms including bacteria and yeast [20, 21], its role and different facets in cancer are still poorly defined. Hampering our understanding is the fact that it rep- resents a number of heterogeneous states [19, 22]. Canonically, cells can be forced into G0 arrest through serum starvation, mitogen withdrawal or contact inhibition [19]. Cells can also undergo G0 arrest spontaneously in response to cell-intrinsic factors like rep- lication stress [23–25]. This process is controlled by p53 [26], which triggers the inhibi- tion of cyclin-CDK complexes by activating p21 [24]. This in turn allows the assembly of the DREAM complex—a key effector responsible for repression of cell-cycle-dependent gene expression [27]. Min and Spencer [28] recently demonstrated a much broader sys- temic coordination of 198 genes underlying distinct types of G0 arrest by profiling the transcriptomes of cells that entered this state either spontaneously or upon different stimuli. Additionally, proliferation-G0 decisions can be impacted by oncogenic changes such as MYC amplification [29] or altered p38/ERK signalling [30]. Despite these advances, the identification of G0-arrested cells within tumours pre- sents an ongoing challenge due to their scarcity and lack of universal, easily measura- ble markers for the activation and maintenance of this state. As they are often defined by a lack of proliferative markers [31, 32], different forms of G0 arrest such as quies- cence, senescence, dormancy and (to a lesser extent) stemness might sometimes be used interchangeably [4, 33]. Quiescent and dormant cells can readily resume their prolif- erative state, senescent cells are irreversibly arrested [28] while cancer stem cells have a high capacity for self-renewal and sit at the top of the differentiation hierarchy [34]. Even though the same cell cycle arrest programme underlies all of these states, they are linked with distinct environmental stimuli and drive cancer progression and therapeu- tic resistance in different ways [11, 12, 35, 36]. Increasing evidence from the literature points towards the rapid adaptation of tumour cells to drug treatments being enabled by a slow dividing or a quiescent state that persists for a short period of time before the cells start reproliferating [37]. Thus, quiescence could more frequently be encountered at the early stages of therapeutic resistance compared to other cell cycle arrest phenotypes, although senescence and stemness are also often discussed in this context. Biomarkers of cell cycle arrest and persistence that are sufficiently specific and robust to be clinically useful are clearly needed. Furthermore, our understanding of how cancer evolution is shaped by proliferation and G0 arrest decisions is limited. The proliferative heterogeneity of cancer cell popula- tions has been previously described and linked with FAK/AKT1 signalling [38], but the constraints and consequences of these cell state switches have not been systematically profiled across cancer tissues. The extent to which G0 arrest in cancer is enacted through Wiecek et al. Genome Biology (2023) 24:128 Page 3 of 35 transcriptional or genetic control is unknown [5, 39], and neither are the mutational processes and genomic events modulating this state. Understanding the evolutionary triggers and molecular mechanisms that enable cancer cells to enter and maintain G0 arrest would enable us to develop pharmacological strategies to selectively eradicate these arrested cancer cells or prevent them from re-entering proliferative cycles. To address these challenges, we have developed a new method to reliably quantify G0 arrest in cancer using transcriptomic data, and employed it to characterise this phenom- enon in bulk and single-cell datasets from a variety of solid tumours. We describe the spectrum of proliferation and G0 arrest decisions in primary tumours, which reflects a range of stress adaptation mechanisms during the course of cancer development from early to advanced disease. We identify and validate mutational constraints for the emer- gence of G0 arrest, hinting at potential new therapeutic targets that could exploit this mechanism. We also demonstrate the relevance of G0 arrest to responses to a range of compounds targeting cell cycle, kinase signalling and epigenetic mechanisms in single- cell datasets and propose an expression signature that could be employed to detect treat- ment resistance induced by G0 arrested tumour cells. Results Evaluating G0 arrest in cancer from transcriptomic data We hypothesised that primary tumours contain varying numbers of cells temporarily or permanently arrested in the cell cycle, which reflect evolutionary adaptations to cellular stress and may determine their ability to overcome antiproliferative therapies. To capture this elusive phenotype, we developed a computational framework that would allow us to quantify G0 arrest signals in bulk and single-cell sequenced cancer samples (Fig. 1a). To define a signature of G0 arrest, we focused on genes that have been shown by Min and Spencer [28] to be specifically activated or inactivated during quiescence that arises spontaneously or as a response to serum starvation, contact inhibition, MEK inhibition or CDK4/6 inhibition. The activity of 139 of these genes changed in a coordinated man- ner across all these five distinct forms of quiescence, likely representing generic tran- scriptional consequences of G0 arrest. The expression levels of these markers were used to derive a score reflecting the relative abundance of G0 arrested cells within individual tumours (see ‘Methods’, Additional file 1: Table S1). (See figure on next page.) Fig. 1 Methodology for quantifying G0 arrest in cancer. a Workflow for evaluating G0 arrest from RNA‑seq data; 139 genes differentially expressed in multiple forms of quiescence were employed to score G0 arrest across cancer tissues. b Receiver operating characteristic (ROC) curves illustrating the performance of the Z‑score methodology on separating actively proliferating and G0 arrested cells in seven single‑cell (continuous curves) and bulk RNA‑seq (dotted curves) datasets. AUC area under the curve. c Compared classification accuracies of the G0 arrest Z‑score approach and classic cell proliferation markers across the seven single‑cell/bulk RNA‑seq validation datasets. d G0 arrest levels of embryonic fibroblast cells under serum starvation for various amounts of time. Replicates are depicted in the same colour. e Representative images of lung cancer cell lines immunostained and analysed to detect the G0 arrest fraction. Hoechst (labels all nuclei) is in blue, phospho‑Rb in green and EdU in red in the merged image. White dashed circles highlight G0 arrested cells that are negative for both phospho‑Rb and EdU signals. Scale bar: 100 µm. f Graphs show single‑cell quantification of phospho‑Rb and EdU intensities taken from images and used to define the cut‑off to calculate the G0 arrest fraction (green boxes). Images in e and graphs in f are taken from the A549 cell line. g–h Correlation between theoretical estimates of a G0 or G1 state and the fraction of cells entering G0 arrest in nine lung adenocarcinoma cell lines, as assessed through g phospho‑Rb assays and h 3 is shown for the average percentage of G0 arrested cells EdU assays. Mean of n = Wiecek et al. Genome Biology (2023) 24:128 Page 4 of 35 Fig. 1 (See legend on previous page.) To validate this signature and select the optimal method to score G0 arrest in indi- vidual samples amongst different enrichment/rank-based scoring methodologies [40–43], we used seven single-cell and bulk datasets [12, 44–49] where actively prolif- erating and quiescent/dormant cells had been independently isolated and sequenced (Additional file 1: Table S2, Methods). We tested the performance of our signature and scoring methodology, as well as that of other commonly used gene signatures, in dis- tinguishing between the truly quiescent/dormant and truly proliferating cells in these seven datasets while varying the expression cut-offs for labelling cells as G0 arrested or proliferating based on the respective signature. A combined Z-score approach had Wiecek et al. Genome Biology (2023) 24:128 Page 5 of 35 the highest accuracy in detecting signals of G0 arrest, with a 91% mean performance in classifying cells as G0 arrested or cycling (Fig.  1b, Additional file  2: Fig. S1a-b). Indeed, the individual cells that had been identified as arrested in G0 in the experi- ments showed a significantly higher Z-score than the dividing cells across all data- sets (Additional file 2: Fig. S1c). Our signature reflected an expected increase in p27 protein levels, which are elevated during G0 arrest [50] (Additional file  2: Fig. S1d). It also outperformed classical cell cycle and arrest markers, such as the expression of targets of the DREAM complex, CDK2, Ki67 and of mini-chromosome replication maintenance (MCM) protein complex genes—which are involved in the initiation of eukaryotic genome replication, as well as recently defined G1/S and G2/M signatures [51] (Fig. 1c). Importantly, our approach provided a good separation between G0 and proliferating samples across a variety of cancer types and models including cancer cell lines, 3D organoid cultures, circulating tumour cells and patient-derived xeno- grafts (Additional file 1: Table S2), thereby demonstrating its broad applicability. Fur- thermore, the strength of the score appeared to reflect the duration of G0 arrest [52] (Fig. 1d). We further experimentally validated our methodology in nine lung adenocarci- noma cell lines. We estimated the fraction of G0 cells in each of these cell lines using quantitative, single-cell imaging of phospho-Ser807/811-Rb (phospho-Rb, which labels proliferative cells [53]) and 24-h EdU proliferation assays (Fig.  1e–h). In these assays, cells were pulse-labelled with the nucleotide analogue, EdU, for 24  h before fixation and immunostaining. Only cells which have proliferated in the last 24 h will be labelled by EdU. EdU-negative cells are classed as G0. Cells that were negative for phospho-Rb were also defined as G0, and not G1, since they have not yet passed the restriction point (phospho-Rb negative; see ‘Methods’, Fig.  1e–f ). This G0 frac- tion was further validated in A549 and NCI-H1944 cells where endogenous PCNA has been labelled with an mRuby fluorophore to enable tracking of cell cycle phase lengths by live-cell imaging (Zerjatke et  al. [54], ‘Methods’). By quantifying the G0/ G1 length in individual cells (i.e. time taken to enter S-phase after mitotic exit) over a 48-h period, we could see that these cells were quiescent and not senescent (or in deep quiescence), as all G0/G1 cells did eventually enter S-phase, albeit with variable timing (Additional file 2: Fig. S1e). There was a remarkably good correlation between our predicted G0 arrest levels based on the expression of these cell lines from the Cancer Cell Line Encyclopedia (CCLE) and the fraction of G0 cells in the experiment as assessed by lack of EdU incorporation over a 24-h period (EdU incorporation only occurs during S phase) but particularly by lack of Rb phosphorylation. Phosphorylation and inactivation of the retinoblastoma protein is often used to define the boundary between G0 and G1 and was specifically shown to distinguish the G0 state recently by Stallaert et al. [53]. Furthermore, a G1 signature (see ‘Methods’) was not associated with these experimental measurements, suggesting our method recovers a state more similar to G0 arrest rather than a prolonged G1 state (Fig. 1g–h). The G0 arrest correlations appeared robust to random removal of individual genes from the signature, with no single gene having an inordinate impact on the score (Supplementary Fig.  1f-h). This provided further reassurance that our Z-score-based methodology is successful in capturing G0 arrest signals from bulk tumour data. Wiecek et al. Genome Biology (2023) 24:128 Page 6 of 35 The spectrum of G0 arrest capacity in solid primary tumours Having established a robust framework for quantifying G0 cell cycle arrest in cancer, we next profiled 8005 primary tumour samples across 31 solid cancer tissues from The Can- cer Genome Atlas (TCGA). After accounting for potential confounding signals of non- cycling non-tumour cells from the microenvironment by correcting for tumour purity (see ‘Methods’, Supplementary Fig. 1i-j), we observed an entire spectrum of fast prolif- erating to slowly cycling tumours, with the latter presenting stronger G0-linked signals (Fig. 2a). While we acknowledge that no tumour would be entirely quiescent/senescent and we cannot identify individual G0 arrested cells within the tumour, this analysis does capture a broad range of phenotypes reflecting varying proliferation and cell cycle arrest rates, which suggests that G0 arrest is employed to different extents by tumours as an adaptive mechanism to various extrinsic and intrinsic stress factors. Cancers known to be frequently dormant, such as glioblastoma [6, 44], were amongst the highest ranked in terms of G0 arrest levels, along with kidney and adrenocortical carcinomas (Fig. 2b). This is likely explained by the innate proliferative capacity of the respective tissues. Indeed, tissues with lower stem cell division rates presented a greater propensity for G0 arrest (Fig. 2c) [55]. Our score showed strong negative correlations with the expression of proliferation markers (Fig.  2d), suggesting that it captures a cellular state that could potentially act as a baseline for all major forms of cell cycle arrest, including quiescence, senescence, stemness and clinical dormancy. Indeed, we found that our signature could to a certain Fig. 2 Pan‑cancer evaluation of proliferative heterogeneity and linked tumour hallmarks. a PHATE plot illustrating the wide spectrum of proliferative to slow cycling/arrested states across 8005 primary tumour samples from TCGA. Each sample is coloured according to the relative G0 arrest level. b Variation in tumour G0 arrest levels across different cancer tissues. c Correlation between mean G0 arrest capacity and stem cell division estimates for various tissue types. d Correlating tumour G0 arrest scores with cancer cell stemness (Stemness Index), telomerase activity (EXTEND score), p21 activity (CDKN1A) and the expression of several commonly used proliferation markers. The Pearson correlation coefficient is displayed. RC replication complex. e Consistently higher levels of G0 arrest are detected in samples with functional p53. f Lower G0 arrest scores are observed in tumours with one or two whole‑genome duplication events. Wilcoxon rank‑sum test p‑values are displayed in boxplots, *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001 Wiecek et al. Genome Biology (2023) 24:128 Page 7 of 35 extent also separate senescent cells from proliferating ones in single-cell data from Her- nandez-Segura et al. [56], which is unsurprising since these cells are also in the G0 phase (Additional file 2: Fig. S2a). Indeed, the authors of this resource highlight that some of the pathways uncovered in these senescent cells may be shared with quiescence, which is also backed up by a study from Fujimaki and Yao [57] suggesting similarities between deep quiescence and senescence. Our score did not show strong correlations with other markers of senescence such as the senescence-associated secretory phenotype (SASP) and β-galactosidase activity [58–60] in single cells or TCGA samples (Fig.  2d, Supple- mentary Fig. 2b-d), although we cannot exclude the possibility of a senescent state being captured occasionally given that neither β-galactosidase nor the SASP are obligatory for maintaining senescence [61]. However, the underpinning programme appears to be dis- tinct from that of cancer stem cells, marked by signatures associated with high telomer- ase activity and an undifferentiated state [62, 63] (Additional file 2: Fig. S2e-f ). Lastly, we confirmed expected dependencies on the p53/p21/DREAM activation axis: tumours that were proficient in TP53 or the components of the DREAM complex, as well as those with higher p21 expression, had elevated G0 arrest levels across numerous tissues (Fig. 2e, Supplementary Fig. 2g-h), although only 8 out of 139 genes in our signa- ture are directly transcriptionally regulated by p53 [64]. Nevertheless, p53 proficiency appears to be a non-obligatory dependency of G0 arrest, which is also observed to arise in p53 mutant scenarios in 21% of cases. p53 has also been shown to play a role in pre- venting the occurrence of larger structural events and polyploidy [65–67], potentially explaining the lower G0 arrest levels we observed in tumours that had undergone whole- genome duplication (Fig. 2f ). The genomic background of G0 arrest in cancer Cancer evolution is often driven by a variety of genomic events, ranging from single base substitutions to larger scale copy number variation and rearrangements of genomic seg- ments. It is reasonable to expect that certain mutations accumulated by the cancer cells might enable a more proliferative phenotype, impairing the ability of cells to enter G0 arrest, or - on the contrary - might favour cell cycle exit as a temporary adaptive mecha- nism to extreme levels of stress. Having obtained G0 arrest estimates for primary tumour samples, we set out to identify potential genomic triggers or constraints that may shape proliferation versus G0 arrest decisions in cancer. We identified 285 cancer driver genes that were preferentially altered (via mutations or copy number alterations) either in slow cycling or fast proliferating tumours (Fig. 3a). Reassuringly, this list included genes pre- viously implicated in driving cell cycle exit decisions such as TP53 and MYC [26, 29]. We also investigated associations with mutagenic footprints of carcinogens (termed ‘muta- tional signatures’), which can be identified as trinucleotide substitution patterns in the genome [68, 69]. Fifteen mutational signatures were linked with G0 arrest levels either within individual cancer studies or pan-cancer (Additional file 2: Fig. S2i). Following the initial prioritisation of putative genomic constraints of G0 arrest, we employed machine learning to identify those events that could best distinguish slow cycling tumours with higher abundance of G0 arrested cells from fast proliferating ones, while accounting for tissue effects. An ensemble elastic net selection approach similar to the one described by Pich et al. [70] was applied for this purpose (Fig. 3b, see ‘Methods’). Wiecek et al. Genome Biology (2023) 24:128 Page 8 of 35 Our pan-cancer model identified tissue type to be a major determinant of G0 arrest lev- els (Additional file  2: Fig. S3a). It also uncovered a reduced set of 57 genomic events linked with proliferation/G0 arrest switches, including SNVs and copy number losses in 17 cancer genes, as well as amplifications of 10 cancer genes (Fig. 3c). These events could then be successfully employed to predict G0 arrest in a separate test dataset, thus inter- nally validating our model (Additional file 2: Fig. S3b). Thus, while these events are not necessarily causative, the link is strong enough to be identifying G0 arrest states from genomic data alone. Such events may also pinpoint cellular vulnerabilities that could be exploited therapeutically. Overall, the genomic dependencies of G0 arrest mainly comprised genes involved in cell cycle pathways, p53 regulation and ubiquitination (most likely of cell cycle targets), and RUNX3 regulation, which have previously been shown to play a role in controlling proliferation and cell cycle entry [71] (Additional file 2: Fig. S3c). Invariably, this analysis has captured several events that are well known to promote cellular proliferation in can- cer: this is expected and confirms the validity of our model. It was reassuring that a func- tional TP53, lack of MYC amplification and lower mutation rates (Fig. 3c) were amongst the top ranked characteristics of tumours with high levels of G0 arrest, which also dis- played less aneuploidy. However, our analysis has also uncovered novel dependencies of G0 arrest-proliferation decisions that have not been reported previously, such as CEP89 and LMNA amplifications observed in fast cycling tumours, or ZMYM2 deletions preva- lent in samples with high levels of G0 arrest. ZMYM2 has recently been described as a novel binding partner of B-MYB and has been shown to be important in facilitating the G1/S cell cycle transition [72]. p16 (CDKN2A) deletions, one of the frequent early events during cancer evolution [73, 74], were enriched in tumours with high proportions of cells in G0. RB1 deletions and amplifications were both associated with a reduction in G0 arrest, which might reflect the dual role of RB1 in regulating proliferation and apop- tosis [75]. Our model also calls to attention to the broader mutational processes associated with this cellular state. Such processes showed fairly weak and heterogeneous correlations with G0 arrest within individual cancer tissues (Additional file  2: Fig. S2g), but their contribution becomes substantially clearer pan-cancer once other genomic sources (See figure on next page.) Fig. 3 Genomic landscape of G0 arrest decisions in cancer. a Cancer drivers with mutations or copy number alterations depleted pan‑cancer in a G0 arrest context. Features further selected by the pan‑cancer model are highlighted. b Schematic of the ensemble elastic net modelling employed to prioritise genomic changes associated with G0 arrest. c Genomic events significantly associated with G0 arrest, ranked according to their importance in the model (highest to lowest). Each point depicts an individual tumour sample, coloured by the value of the respective feature. For discrete variables, purple indicates the presence of the feature and green its absence. The Shapley values indicate the impact of individual feature values on the G0 arrest score prediction. d G0 arrest levels are significantly reduced in microsatellite unstable (MSI) samples in stomach adenocarcinoma (STAD) and uterine corpus endometrial carcinoma (UCEC), with the same trend (albeit not significant) shown in colon adenocarcinoma (COAD). Wilcoxon rank‑sum test *p < 0.05; **p < 0.01. e Genomic alterations are depleted across DNA repair pathways during G0 arrest. Odds ratios of mutational load on pathway in G0 arrest are depicted, along with confidence intervals. CS, chromosome segregation; p53, p53 pathway; UR, ubiquitylation response; CPF, checkpoint factors; TM, telomere maintenance; CR, chromatin remodelling; TLS, translesion synthesis; NHEJ, non‑homologous end joining; NER, nucleotide excision repair; MMR, mismatch repair; FA, Fanconi Anaemia; BER, base excision repair. f G0 arrest scores are increased in cell lines with slow doubling time across MCF7 strains, which also show lower prevalence of PTEN mutations. g Tissue‑specific changes in G0 arrest between samples with/without quiescence‑associated deletions (blue), amplifications (red) and SNVs (brown) within the TCGA cohort (top) and external validation datasets (bottom) Wiecek et al. Genome Biology (2023) 24:128 Page 9 of 35 Fig. 3 (See legend on previous page.) are accounted for. In particular, we identified an association between G0 arrest and mutagenesis induced by the AID/APOBEC family of cytosine deaminases as denoted by signature SBS2 [68] (Fig.  3c). As highlighted by Mas-Ponte and Supek [76], APOBEC/ AID driven mutations tend to be directed towards early-replicating, gene-rich regions of the genome, inducing deleterious events on several genes including ZMYM2, which our pan-cancer model has linked with G0 arrest. In turn, defective DNA mismatch repair, as evidenced by signatures SBS44, SBS20, SBS15, SBS14 and SBS6 [68], was prevalent in fast cycling tumours (Fig. 3c). Mismatch Wiecek et al. Genome Biology (2023) 24:128 Page 10 of 35 repair deficiencies lead to hypermutation in a phenomenon termed ‘microsatellite insta- bility’ (MSI), which has been linked with increased immune evasion [77]. Cancers par- ticularly prone to MSI include colon, stomach and endometrial carcinomas [78], where this state was indeed linked with reduced G0 arrest (Fig.  3d). Furthermore, tumours with high proportions of cells in G0 were depleted of alterations across all DNA damage repair pathways (Fig. 3e). Our measurements of G0 arrest also reflected expected cycling patterns across 27 MCF7 strains [79]: cell lines with longer doubling times exhibited increased G0 arrest (Fig. 3f ). This coincided with a depletion of PTEN mutations, a dependency highlighted by the pan-cancer model. When checking for dependencies in individual cancer tissues, 24 out of the 25 genes identified by the model were significantly associated with G0 arrest or proliferation deci- sions in at least one tissue, most prominently in breast, lung and liver cancers which also represent the largest studies within TCGA (Fig. 3g, top panel). Most of these genomic insults were linked with a decrease in G0 arrest. In external validation datasets, these associations, including deletions in PTEN and LRP1B or amplifications of MYC, CEP89 and ETV6, featured most prominently in the largest cohort of breast cancer samples (Fig.  3g, bottom panel). These results highlight the fact that although a pan-cancer approach is suited to capture genomic events that are universally associated with cell cycle exit, certain genetic alterations may facilitate a higher or lower propensity of G0 arrest in a single tissue only. Indeed, when building a tissue-specific breast cancer model of G0 arrest using a com- bined ANOVA and random forest classification approach (Additional file  2: Fig. S4a), we not only recovered the associations with the TP53, MYC, LMNA and ETV6 events already seen in the pan-cancer model (Additional file  2: Fig. S4b) but also identified additional events which validated in the METABRIC cohort and were also seen in sev- eral other cancers, e.g. bladder, lung and lower grade glioma (Additional file 2: Fig. S4c). Notably, the APOBEC mutational signature SBS2 was the strongest genomic signal linked with G0 arrest in breast cancer (Supplementary Fig. 4b,d) and was most prevalent in Her2+ tumours, although the Luminal A subtype showed the highest levels of G0 arrest overall, as expected given its well-known lower proliferative capacity [80] (Sup- plementary Fig. 4e-f ). Validation of CEP89 as a modulator of G0 arrest capacity To gain more insight into the underlying biology of G0 arrest in cancer, we sought to experimentally validate associations highlighted by the pan-cancer model. We focused on the impact of CEP89 activity on proliferation/arrest decisions due to the high ranking of this putative oncogene in the model, the relatively unexplored links between CEP89 and cell cycle control, as well as its negative association with G0 arrest across a variety of cancer cell lines (Supplementary Fig. 5a-c). The function of CEP89 is not well char- acterised; however, the encoded protein has been proposed to function as a centroso- mal-associated protein [81, 82]. Centrosomes function as major microtubule-organising centres in cells, playing a key role in mitotic spindle assembly [83] and the mitotic entry checkpoint [84]. Moreover, centrosomes act as sites of ubiquitin-mediated proteolysis of cell cycle targets [85], and members of several growth signalling pathways, such as Wiecek et al. Genome Biology (2023) 24:128 Page 11 of 35 Wnt and NF-kB, localise at these structures [86, 87]. Several genetic interactions have also been reported between CEP89 and key cell cycle proteins, including cyclin D2 [88] (Fig. 4a). Our model linked CEP89 amplification with fast cycling tumours (Fig.  3c). Centro- some amplification is a common feature of tumours with high proliferation rates and high genomic instability [89], and overexpression of centrosomal proteins can alter cen- triole structure [90, 91]. Indeed, CEP89 amplified tumours presented elevated expres- sion of a previously reported centrosome amplification signature (CA20) [89] (Fig. 4b), which was strongly anticorrelated with G0 arrest levels (Fig.  4c). Furthermore, CEP89 expression was prognostic across multiple cancer tissues (Fig. 4d) and linked with toxic- ity of several cancer compounds in cell line models (Additional file 2: Fig. S5d). Fig. 4 CEP89 amplification is associated with lower G0 arrest capacity. a Network illustrating CEP89 interactions with cell cycle genes (from GeneMania). The edge colour indicates the interaction type, with green representing genetic interactions, orange representing predicted interactions and purple indicating pathway interactions. The edge width illustrates the interaction weight. b CA20 scores are significantly increased in TCGA primary tumours containing a CEP89 amplification. c Pan‑cancer relationship between CA20 and G0 arrest scores across the TCGA cohort. d Cox proportional hazards analysis estimates of the log hazards ratio for the impact of CEP89 expression on patient prognosis within individual cancer studies, after adjusting for tumour stage. Patients with high expression of CEP89 show significantly worse prognosis within ACC, LUSC, LIHC, KIRC and STAD, but significantly better prognosis within HNSC, PAAD and KIRP studies. e Western blot showing depletion of Cep89 protein 48 h after siRNA transfection of NCI‑H1299 cells. Mock is lipofectamine only; NTC is non‑targeting control siRNA. B‑actin is used as a loading control. f Graphs show that Cep89 depletion in NCI‑H1299 cells leads to a reduction in nuclear number and an increase in the fraction of G0 arrested cells, measured by an increase in the percentage of EdU negative (24 h EdU pulse) and Phospho‑Ser 807/811 Rb negative cells. One‑way ANOVA, *p < 0.05, **p < 0.01. Mean of n 3 = Wiecek et al. Genome Biology (2023) 24:128 Page 12 of 35 We validated this target in the lung adenocarcinoma cell line NCI-H1299 showing high levels of CEP89 amplification. Cep89 depletion via siRNA knockdown caused a consistent decrease in cell number, in the absence of any detectable cell death, and an increase in the fraction of G0 cells as measured by phospho-Rb and EdU assays (Fig. 4e– f ). Thus, we propose CEP89 as a novel cell proliferation regulator that may be exploited in certain scenarios to control tumour growth. Characterisation of individual stress response programmes of G0 arrest While we had previously examined a generic programme of G0 arrest, cancer cells can enter this state due to different stimuli [19] and this may inform its aetiology and mani- festation. To explore this, we re-scored tumours based on gene expression programmes specific to serum starvation, contact inhibition, MEK inhibition, CDK4/6 inhibition or spontaneously occurring quiescence as defined by Min and Spencer [28] (see ‘Methods’). We observed a good correlation between our estimates representing individual stress response programmes and the expression of genes associated with the corresponding form of G0 arrest in the literature (Fig. 5a–e, Additional file 2: Fig. S6, see ‘Methods’). Specifically, strong inverse correlations were seen between our CDK4/6 inhibition scores and the mean expression of CDK4 and CDK6, or between our MEK inhibition scores and the expression of genes involved in the MAPK pathway [92]. Spontaneous quies- cence and serum starvation scores were most correlated with the activity of p21, or of genes involved in the cellular response to starvation, respectively. The contact inhibition programme was also captured, but with lesser specificity. CDK4/6 inhibition-induced G0 arrest levels were further validated using exter- nal RNA-seq datasets from cancer cell lines and xenograft mice sequenced before and after treatment with the CDK4/6 inhibitor palbociclib [93, 94] (Fig. 5f, Additional file 1: Table S3). The fact that the estimates for a generic G0 arrest phenotype were equally or, in some cases, more discriminative of cells treated with palbociclib confirms the gener- alisability of this score, which may be more broadly applicable to different tissues and/ or model systems, as shown previously in the single-cell validation data. The CDK4/6 inhibition scores outperformed all the other stress response subtype scores, suggesting that a combination of individual programmes and the generic score might best identify a specific stimulus driving G0 arrest. Interestingly, we also observed significant differ- ences in spontaneous quiescence scores before and after treatment. Indeed, p21 activ- ity has been linked with the palbociclib mechanism of action [95, 96], and this analysis suggests potential similarities between CDK4/6 inhibition and p21-dependent G0 arrest phenotypes. Having validated our framework for quantifying stimulus-specific G0 arrest pro- grammes, we proceeded to estimate the dominant form of stress that may induce cell cycle arrest in different cancer types (Fig. 5g). We found a range of G0 arrest aetiolo- gies across most tissues, while a minority of cancers were dominated by a single form of stress response, e.g. serum starvation in all G0 arrested pheochromocytomas and paragangliomas, contact inhibition in 88% of head and neck carcinomas and CDK4/6 inhibition in 80% of adrenocortical carcinomas. While we do not wish to claim that the state of cell cycle arrest will have necessarily been induced by the actual pre- dicted stimulus (impossible in the case of CDK4/6 or MEK inhibition, as the analysed Wiecek et al. Genome Biology (2023) 24:128 Page 13 of 35 Fig. 5 Pan‑cancer characterisation of individual G0 stress response programmes. a‑e Comparison of correlation coefficients between stress response programme scores and a mean expression of CDK4 and CDK6, b mean expression of curated contact inhibition genes, c a transcriptional MAPK Pathway Activity Score (MPAS), d mean expression of curated serum starvation genes and e CDKN1A expression (encoding for p21), across TCGA cancers. The correlations expected to be strongest (either negative or positive) are denoted by an asterisk. The generic G0 arrest score refers to scores calculated using the original list of 139 genes differentially expressed across all 5 forms of G0 arrest. f Comparison of stress response programme scores measured in cancer cell lines before (grey) and after (red) palbociclib treatment across three validation studies. Datasets used for validation are denoted by their corresponding GEO series accession number. g Predicted stress response diversity in samples with high levels of G0 arrest across individual cancer types. The same colour legend as in a is applied. Grey bars represent the proportion of samples for which the G0 arrest inducer could not be estimated samples are all treatment-naïve), we suggest that the downstream signalling cascade may resemble that triggered by such stimuli, e.g. via CDK4/6 or MEK loss of function mutations. Some of the differences observed might be explained by the dependency between p53 activity and the form of stress response that is enacted. Amongst the five differ- ent forms, spontaneous G0 arrest appeared most strongly dependent on p53 func- tionality, with a nearly two-fold enrichment of p53 proficient tumours in this group (Additional file  2: Fig. S6f ). Indeed, significantly higher levels of spontaneous G0 arrest were observed in the majority of cancers (56%) when p53 was functional rather than mutated. The second most dependent state was that of CDK4/6 inhibition, with increased levels in 36% of cancer types displaying p53 proficiency (Additional file 2: Fig. S6g). Overall, these analyses of stress response states point to common transcriptional features of drug-tolerant G0 arrested cells in different cancer settings that could be employed in designing ways to eradicate these cells in the future. Wiecek et al. Genome Biology (2023) 24:128 Page 14 of 35 Role of G0 arrest in driving therapeutic resistance in cancer uncovered from single‑cell data Overall, G0 arrest appears to be beneficial for the long-term outcome of cancer patients, even when accounting for potential confounders such as stage, sex and tissue (Fig. 6a, Additional file  2: Fig. S7a). No clear relation was observed between G0 arrest levels within the primary tumour and risk of relapse, although higher G0 arrest was occasion- ally deemed favourable to avoiding disease recurrence or progression (Additional file 2: Fig. S7b-e). Indeed, such slow cycling, indolent tumours would have higher chances of being eradicated earlier in the disease, which is consistent with reported worse prog- nosis of patients with higher tumour cell proliferation rates [97]. As expected, G0 arrest levels were increased in stage 1 tumours, although later stages also exhibited this phe- notype occasionally (Additional file  2: Fig. S7f ). However, outcomes do vary depend- ing on the stress source, with worse survival observed upon contact inhibition (Fig. 6b). The outcomes also vary by tissue: when the cut-offs between increased G0 arrest and high proliferation were defined on an individual cancer basis rather than pan-cancer, we found that lung, colon or oesophageal carcinoma patients displayed significantly worse prognosis in the context of high proportions of G0 arrested cells in the tumour (Fig. 6c, see ‘Methods’). Indeed, p53 wild-type colorectal cancers expressing a quiescence-linked fetal phenotype have been recently associated with metastasis and poor prognosis [98]. In contrast, adrenocortical and kidney papillary cell carcinoma ranked in the top of can- cers with improved survival. It is noticeable that the cancers in the former, worse prog- nosis group are also amongst the ones displaying lower than average G0 arrest (Fig. 2b), so the observed inferior outcomes could in part be linked to these cancers being intrinsi- cally faster progressing. It is possible there is a lower limit below which G0 arrest stops being useful for delaying growth and becomes detrimental instead, perhaps in conjunc- tion with treatment. Indeed, other factors such as the type of therapy received could play a role too. While we are limited in the investigation of such factors in TCGA due to the incomplete records available, these discrepancies should be subject to future research. While G0 arrest may confer an overall survival advantage in most cancers, it can also provide a pool of cells that are capable of developing resistance to therapy [12, 99]. Using our methodology, we indeed observed an increase in G0 arrest levels in cell lines (See figure on next page.) Fig. 6 Impact of G0 arrest on patient prognosis and treatment response. a Disease‑specific survival based on proliferation/G0 arrest levels for patients from TCGA within 15 years of follow‑up. Patients with increased levels of G0 arrest in primary tumours showed significantly better prognosis than patients with fast proliferating tumours. b–c Hazard ratio ranges illustrating the impact of different forms of G0 induction (b) and different tissues (c) on patient prognosis, after taking into account potential confounding factors. Values above 0 indicate significantly better prognosis when tumours contain high proportions of cells arrested in G0. d Change in G0 arrest scores inferred from bulk RNA‑seq across breast, pancreatic, colorectal and skin cancer cells in response to treatment with the CDK4/6 inhibitor palbociclib, 5‑FU or the BRAF inhibitor vemurafenib. e–f UMAP plot illustrating the response of the TP53‑proficient RKO colorectal cancer cell line to various 5‑FU doses and the corresponding proportions of cells predicted to be arrested/proliferating. Each dot is an individual cell, coloured according to its G0 arrest level. g–h The same as previous, but for the TP53‑deficient SW480 cell line. i–j UMAP plot illustrating the response of individual PC9 NSCLC cells to the EGFR inhibitor erlotinib across several time points and the corresponding proportion of cells predicted to be arrested/proliferating. k Principal component analysis illustrating the superimposition of single‑cell RNA‑seq profiles (circles) of G0 arrested NSCLC cells before/after EGFR inhibition onto the bulk RNA‑seq reference data (triangles) for MCF10A cells occupying various stress response states. l The proportion of NSCLC cells in k predicted to occupy different stress response states across several time points Wiecek et al. Genome Biology (2023) 24:128 Page 15 of 35 Fig. 6 (See legend on previous page.) following treatment with EGFR, BRAF and CDK4/6 inhibitors, as well as conventionally used chemotherapies such as 5-fluorouracil (5-FU) in multiple bulk RNA-seq datasets (Fig. 6d). Furthermore, the recent widespread availability of single-cell transcriptomics offers the opportunity to investigate the impact of G0 arrest on such therapies with much greater granularity than is allowed by bulk data. Using our G0 arrest signature and single-cell data from RKO and SW480 colon cancer cell lines treated with 5-FU [100], we could observe G0 arrest and proliferation decisions following conventional chemotherapy Wiecek et al. Genome Biology (2023) 24:128 Page 16 of 35 treatment. Within the p53 proficient cell line RKO, the fraction of G0 arrested cells increased from 41 to 93% after treatment with a low dose (10  μM) of 5-FU and per- sisted at higher doses (Fig.  6e–f ). In contrast, a comparable increase in G0 arrest was not observed in TP53 mutant SW480 cells, further emphasizing the key role of p53 as a regulator of cell cycle exit (Fig. 6g–h). This implies that although TP53 mutations confer a more aggressive tumour phenotype and may drive resistance via other mechanisms, TP53 wild-type tumour cells are more likely to be capable of entering a G0 ‘persistent’ state associated with drug resistance. SW480 cells showed higher apoptotic activity following treatment compared to RKO cells, particularly within actively cycling cells, further corroborating that cells capable of entering G0 arrest may be intrinsically less vulnerable to this therapy (Supplementary Fig. 8a-b). Similarly, using single-cell data from an EGFR mutant non-small cell lung cancer (NSCLC) cell line treated with the EGFR inhibitor erlotinib [13], we predicted that 40% of cells were likely to exist in a G0 arrest state prior to treatment. EGFR inhibition led to a massive decrease in cell numbers immediately after treatment, mostly due to pro- liferating cells dying off (Supplementary Fig.  8c-d), while the proportion of arrested cells increased to 96% at day 1, indicating an immediate selective advantage for such cells (Fig. 6i–j). These cells appear to gradually start proliferating again in the following days during continuous treatment, with the percentage of proliferating cells approach- ing pre-treatment levels by day 11 (Fig. 6j). The same trend captured by our signature could be observed upon KRAS and BRAF inhibition in different cell line models (Addi- tional file 2: Fig. S8e-h, Additional file 1: Table S3) [12, 13]. Furthermore, during the first days of treatment, the NSCLC cells that survived EGFR inhibition appeared to reside in a state most resembling that induced by serum starvation (Fig. 6k–l). Both EGFR kinase inhibitors and serum starvation have been shown to trigger autophagy [101], which may explain the convergence between this inhibitory trigger and the type of stress response. At day 11, most of the remaining arrested cells appeared in a state similar to that preced- ing the treatment (Fig. 6l). Thus, G0 arrest appears to explain resistance to broad acting chemotherapy agents as well as targeted molecular inhibitors of the Ras/MAPK signalling pathway, being either selected for, or induced immediately upon treatment, and gradually waning over time as cells start re-entering the cell cycle. Using massively multiplexed chemical transcrip- tomic data, we also analysed responses to 188 small molecule inhibitors in cell lines at single-cell resolution [102] (Additional file 2: Fig. S9). We observed a large increase in G0 arrest following treatment with not only compounds targeting cell cycle regulation and tyrosine kinase signalling, consistent with our previous results, but also for com- pounds modulating epigenetic regulation, e.g. histone deacetylase inhibitors—thus high- lighting the broad relevance of G0 arrest. While links between G0 arrest and therapeutic resistance are prevalently observed in cell lines, one would question whether this translates to similar pathology in cancer patients. While we observed significantly higher G0 arrest levels in pre-treated tumours of non-responders to neoadjuvant chemotherapy in a breast cancer study by Hatzis et al. [103] (Additional file  2: Fig. S10a), surveying various targeted therapy datasets from the SELECT study [104] and the TCGA data for links with response to various thera- pies (single agent and combinations) showed little to no evidence for a bulk signature Wiecek et al. Genome Biology (2023) 24:128 Page 17 of 35 of G0 being useful for predicting resistance in these studies (Supplementary Fig.  10b- c). Although the studies available for inspection are rather sparse, evidence from all the analyses presented here suggests there is no universal a priori role that G0 arrest has within the pre-treated primary tumour in determining response to treatment: favour- able overall outcomes are observed occasionally due to slower progressing malignancy, but resistance is also observed in the case of chemotherapy in breast cancer. Instead, the role of G0 arrest in enabling therapeutic resistance as a short-lived acquired phenotype as demonstrated in single-cell datasets appears more consistent. Tumour cell G0 arrest signature for use in single‑cell transcriptomics data Our ability to probe the nature of G0 arrest phenotypes in single-cell RNA-seq data using a defined G0 signature could aid the development of methods to selectively tar- get G0 arrested drug-resistant persister cells. However, a major challenge of single-cell RNA-seq data analysis is the high percentage of gene dropout, which could impact our ability to evaluate G0 arrest using the full 139 gene signature. The single-cell RNA-seq datasets we analysed exhibited an average drop-out of 8.5 genes out of the full gene sig- nature. While our scoring method remains robust to such levels of dropout (Supple- mentary Fig. 1d-f ), we also employed machine learning to reduce our initial list of 139 markers of quiescence to a robust 35-gene signature, comprised mainly of RNA metabo- lism and splicing-regulating factors, and also of genes involved in cell cycle progression, ageing and senescence, which could be applied to sparser datasets with larger levels of gene dropout (see ‘Methods’, Fig.  7a-b, Additional file  1: Table  S4). The optimised sig- nature of G0 arrest performed similarly to the initial broadly defined programme in distinguishing fast cycling tumours from those containing high proportions of G0 cells (Fig. 7c). It also showed an average dropout of only 0.5 genes across the single-cell RNA- seq datasets used in this study (Fig. 7d), was similarly prognostic (p = 0.004) and showed comparable profiles of resistance to treatment (Fig. 7e, Additional file 2: Fig. S11). This minimal expression signature could be employed to track and further study emerging G0 arrest-enabled resistance in a variety of therapeutic scenarios. Discussion Despite its crucial role in cancer progression and resistance to therapies, G0 arrest in all its forms remains poorly characterised due to the scarcity of suitable models and bio- markers for large-scale tracking in the tissue or blood. The lack of proliferative mark- ers such as Ki67 or CDK2 [31, 105] does not uniquely distinguish G0 arrest from other cell cycle phases, e.g. G1. Miller et al. [32] have shown that the Ki67 is expressed at the mRNA level but the protein is degraded continuously both in G0 and G1, and it rather acts as a graded marker of S/G2/M. Similarly, CDK2 activity is low in G0 and G1, builds up at the restriction point, is high in the S phase and is then replaced by CDK1 in mito- sis. Reduced CDK2 expression can manifest due to not only quiescence and mitosis but also to DNA damage [106], and thus cut-offs to uniquely distinguish its activity in G0 would be difficult to define. Furthermore, these and other reliable markers of G0 arrest such as p27 or p130 [50] are best captured at protein level, which is much more sparsely measured, and expression does not accurately reflect their activity. This study overcame this limitation by employing genes active in different forms of quiescence whose patterns Wiecek et al. Genome Biology (2023) 24:128 Page 18 of 35 Fig. 7 Optimisation of the G0 arrest signature for use in single‑cell RNA‑seq data. a Methodology for refining the gene signature of G0 arrest: random forest classifiers are trained to distinguish arrested from cycling tumours on three high confidence datasets; Gini index thresholding is optimised to prioritise a final list of 35 genes. b Gini index variation, correlation with experimentally measured quiescence via EdU and phospho‑Rb staining assays, and corresponding p‑values are plotted as the number of genes considered in the model is increased. The vertical black dashed line indicates the threshold chosen for the final solution of 35 genes. The horizontal grey dotted line indicates the threshold for p‑value significance. c Additional external validation of the 35 gene signature acting as a classifier of G0 arrested and proliferating cells in single‑cell and bulk datasets. d Dropout in single‑cell data by gene signature. The percentage of genes out of the 35 (red) and 139 (grey) gene lists with reported expression across the single‑cell RNA‑seq datasets analysed in this study. e Proportion of cycling and G0 arrested cells estimated in single‑cell datasets of p53 wild‑type and mutant lines treated with 5FU, as well as cells treated with EGFR inhibitors. Data as in Fig. 6 of expression are distinct from markers of a longer G1 phase and capture cell cycle arrest as also observed in senescence, stemness or dormancy. We have extensively validated our method and signature in single-cell datasets and cancer cell lines and have demon- strated that it can reliably and robustly capture signals of G0 arrest both in bulk tissue as well as in single cells. Within bulk tissue, we are limited in our capacity to distinguish between large fractions of cells residing in short-lived G0 arrest and a smaller fraction of cells that are in deeper Wiecek et al. Genome Biology (2023) 24:128 Page 19 of 35 G0 arrest, as our score seems to reflect both parameters to a certain extent. Bearing in mind this limitation, our score could potentially also be used in a single-cell setting to capture longer-lived cell cycle arrest states such as ones demonstrated in senescence or dormancy and could assist in identifying such states, but only with the help of additional cell-state specific, immune or secretory biomarkers. Indeed, gene activity linked to cell cycle arrest is not exclusive to quiescence, but can be shared with senescence or dor- mancy in certain scenarios, as also demonstrated in some of our analyses of senescent cells. This makes it difficult to clearly distinguish states like dormancy, senescence and quiescence (particularly deep quiescence), as even their definitions can be contentious at times both in the context of human cancers [4, 57, 107–109] as well as in physiological conditions in other organisms [110, 111]. Since our signature was derived and validated in experiments that were tailored specifically to induce and/or measure quiescence, we believe the signature proposed in this study best reflects a quiescent-like, reversible G0 cell cycle arrest state. While senescent and dormant cells could be distinguished from their quiescent counterparts simply based on additional senescence and dormancy markers, further research is nevertheless required in the future to delineate signatures that are both necessary and sufficient to unambiguously discriminate all three states. In the meantime, future studies utilising our G0 signature should also test for such addi- tional markers like β-galactosidase activity, the SASP and other senescence markers, or NRF2, NR2F1, SOX9, RARβ [112, 113] and other dormancy markers to ascertain the type of G0 arrest that is being captured. The versatility of our signature is evidenced by high classification accuracies across a variety of solid cancer datasets. More variable performance was observed when applied to haematopoietic stem cells as it was not designed to capture signals in this context (Additional file  2: Fig. S1b). While we cannot exclude that the patterns captured may also occasionally reflect cell cycle arrest in G1 or G2, this broad signature would still capture phenotypes resulting from intrinsic or extrinsic cellular stress that reflect tem- porary tumour adaptation during the course of cancer evolution or upon treatment with drugs. Thus, studying such states is relevant for identifying vulnerabilities that could be exploited at different time points during the course of cancer treatment. We show that G0 arrest is pervasive across different solid cancers and generally associ- ated with more stable, less mutated genomes with intact DNA damage repair pathways. We also find a link between APOBEC mutagenesis and higher levels of G0 arrest. Some neoplastic events enriched in tumours with increased G0 arrest, such as p16 or ZMYM2 deletions, could mark elevated genomic stress that renders cells more prone to cell cycle exit. We also identified mutational events affecting a variety of genes such as PTEN, CEP89, CYLD and LMNA that appear unfavourable to cell cycle arrest, thus potentially implicating them in influencing G0 arrest-proliferation decisions. Amongst these, we propose and validate CEP89 as a novel modulator of G0 arrest capacity in non-small cell lung cancer. A recent paper describes how increased CEP89 copy number and expres- sion correlates with a worse prognosis in ovarian cancer [114], which we hypothesise could be linked to Cep89’s role in modulating G0 arrest. Although we do not yet know how Cep89 regulates G0 arrest, two roles have been ascribed to Cep89 which could be significant. First, Cep89 is required for primary cilium assembly [115, 116]. The primary cilium acts as a signalling hub, transducing extracellular signals to intracellular signalling Wiecek et al. Genome Biology (2023) 24:128 Page 20 of 35 networks, many of which regulate growth and proliferation [117]. Cep89 deficiency also leads to defects in Complex IV assembly in the electron transport chain in mitochon- dria, leading to decreased mitochondrial function and ATP production [118]. Decreased ATP would impair the ability of cells to proliferate. Since Cep89 is a coiled-coil protein with no obviously targetable regulatory domains, it will be important to ascertain which Cep89 function is key to regulating the balance between proliferation and arrest in can- cer cells to be able to potentially target that process, rather than Cep89 itself, to induce or maintain G0 arrest. These large-scale genomic associations with G0 arrest phenotypes are only currently feasible in bulk datasets. However, bulk sequenced data has a major limitation in cap- turing an average signal across all cells within the tumour, which prevents individual cell state identification and counting. Our subsequent exploration of single-cell datasets across 193 therapeutic scenarios complements this analysis and illustrates the power of applying our signature in single cells. Our signature of G0 arrest is prognostic and marks primary tumours with a lower proliferative capacity before treatment, but we also clearly demonstrate that it can be employed to track resistance to multiple cell cycle, kinase signalling and epigenetic tar- geting regimens, where it often appears as a short-lived phenotype. While this discrep- ancy may appear incompatible at first glance, it is not unlike other cellular processes that have been shown to present dual roles in a cancer setting, such as reactive oxygen spe- cies [119], but also p38α [120] or NRF2 [121], both of which have been implicated in qui- escence or dormancy [113, 122]. It is possible that there is a tipping point between G0 arrest acting beneficially or detrimentally during tumour development and treatment. Furthermore, this is likely influenced by a myriad of other complex factors that we have not had the chance to analyse in depth here, and in some cases, it may just be the base- line for acquiring cancer cell stemness or senescent properties. While we acknowledge this conundrum requires further study, we believe this phenotype also offers a unique opportunity to further understand mechanisms of tumour resistance. A key open ques- tion remains: if G0 arrest drives resistance, does it do so in a Darwinian fashion, as a pre- existing population that is selected for upon drug treatment, or is it instead an acquired phenotype? Our single-cell analyses cannot exclude either scenario. Given the variable links to treatment response and lack of clear evidence for relapse when surveying G0 arrest in primary tumours before treatment, it is likely our G0 arrest signature in its cur- rent form cannot be used to predict resistance to chemotherapy or targeted therapy, and we would not recommend it for this purpose unless further validated in a specific cancer setting. We have also not inspected the role of G0 arrest in the context of immunother- apy, which remains an area of future study. However, we believe our signature has high value for the study of emerging resistance in an in vitro/in vivo setting, as a short-lived enabler of drug tolerance. The optimised signature we propose for single-cell data makes it tractable to a variety of future studies in this area. In a treatment setting, vulnerabilities of G0 arrested cells could be exploited for com- bination therapies. Cells which have exited the cell cycle utilise several mechanisms to achieve drug resistance, including upregulation of stress-induced pathways such as anti-apoptotic BCL-2 signalling [123], anti-ROS programmes [28] or immune evasion Wiecek et al. Genome Biology (2023) 24:128 Page 21 of 35 [15]. Further studies are needed to elucidate which of these mechanisms are specifically employed on a case-by-case basis. Our findings contribute to the understanding of the aetiology and genetic context of G0 arrest in cancer. This is particularly relevant to not only identifying new anti-prolif- erative targets but also for the detection and eradication of drug-tolerant persister cells, which have been frequently, although not always, observed to be slow cycling or entirely quiescent/senescent [8, 9]. Importantly, the state of G0 arrest that we have studied here is distinct from that of disseminated tumour cells causing clinical dormancy and can- cer relapse, often after many years from the treatment of the primary tumour [4, 124]. Here, we have focused on understanding how tumours make proliferation and G0 arrest decisions during the earlier stages of cancer development, within the treatment-naïve primary tumour and as an immediate response to anti-cancer therapies. However, since the dormancy of disseminated tumour cells is fundamentally enabled through a long but temporary cell cycle arrest, we believe our findings of the fundamental processes linked with G0 arrest could in the future help inform a better characterisation of dor- mant tumour cells when combined with specific microenvironmental signatures that are critical for enabling that process. Conclusions Overall, our study provides, for the first time, a pan-cancer view of G0 arrest and its evolutionary constraints, underlying novel mutational dependencies which could be exploited in the clinic. We propose a G0 arrest signature which can be robustly meas- ured in bulk tissue or single cells and could potentially inform therapeutic strategies in the longer term. This signature could be assessed in the clinic to track rapidly emerg- ing resistance, e.g. through liquid biopsies or targeted gene panels. We hope these insights can be used as building blocks for future studies into the different regulators of G0 arrest, including epigenetics and microenvironmental interactions, as well as the mechanisms by which it enables therapeutic resistance both in solid and haematological malignancies. Methods Selection of G0 arrest marker genes Generic G0 arrest markers Differential expression analysis results comparing cycling immortalised, non-trans- formed human epithelial cells and cells in five different forms of quiescence (sponta- neous quiescence, contact inhibition, serum starvation, CDK4/6 inhibition and MEK inhibition) were obtained from Min and Spencer [28]. A total of 195 genes were differen- tially expressed in all five forms of quiescence under an adjusted p-value cut-off of 0.05. This gene list, reflective of a generic G0 arrest phenotype, was subjected to the follow- ing refinement and filtering steps: (1) selection of genes with a unidirectional change of expression across all five forms of quiescence; (2) removal of genes involved in other cell cycle stages included in the ‘KEGG_CELL_CYCLE’ gene list deposited at MSigDB; (3) removal of genes showing low standard deviation and low levels of expression within the TCGA dataset, or which showed low correlation with the pan-cancer expression of Wiecek et al. Genome Biology (2023) 24:128 Page 22 of 35 the transcriptional targets of the DREAM complex, the main effector of quiescence, in TCGA. The resulting 139-gene signature is presented in Additional file 1: Table S1. G0 arrest stress response‑specific markers Gene lists representing spontaneous quiescence, contact inhibition, serum starvation, CDK4/6 inhibition and MEK inhibition programmes were obtained using genes differ- entially expressed in each individual quiescence form using an adjusted p-value cut-off of 0.05. The gene lists were subjected to filtering steps 2 and 3 described above. Follow- ing the refinement steps, 10 upregulated and 10 downregulated genes with highest log2 fold changes were selected for each stress response type. Quantification of G0 arrest in tumours The GSVA R package was used to implement the combined Z-score [40], ssGSEA [41] and GSVA [42] gene set enrichment methods. For the above three methods, a sepa- rate score was obtained for genes upregulated in quiescence and genes downregulated in quiescence, following which a final G0 arrest score was obtained by subtracting the two scores. The singscore single-sample gene signature scoring method [43] was imple- mented using the singscore R package. In addition to these, we also calculated a mean scaled G0 arrest score based on the refined list of genes upregulated and downregulated in quiescence, as well as a curated housekeeping genes from the ‘HSIAO_HOUSEKEEP- ING_GENES’ list deposited at MSigDB, as follows: 1 n G0m = GD GU − 1 n 1 GH n G0m = mean scale G0 arrest score GU = expression of genes upregulated in quiescence GD = expression of genes downregulated in quiescence GH = expression of housekeeping genes n = number of genes in each gene set G0 arrest scores for the TCGA cohort were derived from expression data scaled by tumour purity estimates. The pan-cancer TCGA samples were also classified into groups with ‘high’ or ‘low’ levels of G0 arrest based on k-means clustering (k = 2) on the expres- sion data of 139 G0 biomarker genes, following the removal of tissue-specific expression differences using the ComBat function from the sva R package [125]. Measuring the duration of G0 arrest We employed the GSE124109 dataset from Fujimaki et  al. [52] where rat embryonic fibroblasts were transcriptomically profiled as they moved from short- to long-term qui- escence in the absence of growth signals. The derived G0 arrest scores using our com- bined Z-score methodology increased from short- to longer-term quiescence. Wiecek et al. Genome Biology (2023) 24:128 Page 23 of 35 Validation of G0 arrest scoring methodologies Single‑cell RNA‑sequencing validation datasets Datasets were obtained from the ArrayExpress and Gene Expression Omnibus (GEO) databases though the following GEO Series accession numbers: GSE83142, GSE75367, GSE137912, GSE139013, GSE90742 and E-MTAB-4547. Quality control analysis was standardised using the SingleCellExperiment [126] and scater [127] R packages. Normal- isation was performed using the scran [128] R package. Bulk RNA‑sequencing validation datasets Datasets were obtained from the GEO database through the following GEO Series accession numbers: GSE93391, GSE114012, GSE131594, GSE152699, GSE124854, GSE135215, GSE99116, GSE124109, GSE61130, GSE64553 and GSE63577. GSE114012 count data were normalised to TPM values using the GeoTcgaData R package. All nor- malised datasets were log-transformed before further analysis. The accuracy with which the G0 arrest scoring methods could separate proliferating and quiescent samples within the validation datasets was determined by calculating the area under the curve of the receiver operating characteristic (ROC) curves, using the plotROC R package. Experimental validation in lung adenocarcinoma cell lines The average fraction of cancer cells spontaneously entering quiescence was estimated for nine lung adenocarcinoma cell lines (NCIH460, A549, NCIH1666, NCIH1944, NCIH1563, NCIH1299, NCIH1650, H358, L23) using EdU and phospho-Rb staining proliferation assays. Cell lines were obtained from ATCC or Sigma and regularly checked for mycoplasma. A549 and NCIH460 were cultured in DMEM (Gibco). NCIH358, NCIH1299 and NCIH1563 were maintained in RPMI-1640 (Gibco) supplemented with 5  mM sodium pyruvate and 0.5% glucose. NCIH1944, NCIH1666, NCIH1650 and L23 were grown in RPMI-1640 ATCC formulation (Gibco). A427 were cultured in EMEM (ATCC). A549, NCIH460, H358, NCIH1299, NCIH1563, and A427 were supplemented with 10% heat inactivated FBS. NCIH1666 with 5% heat-inactivated FBS and all other cell lines with 10% non-heat inactivated FBS. All cell lines had penicillin–streptomycin (Gibco) added to 1%. Cells were maintained at 37 °C and 5% CO2. To calculate the quiescent fraction, A549 and NCIH460 cells were plated at a density of 500 cells/well, and all other cell lines at a density of 1000/well, in 384-well CellCarrier Ultra plates (PerkinElmer) in the rel- evant media. Twenty-four hours later, 5  μM EdU was added and cells were incubated for a further 24 h before fixing in a final concentration of 4% formaldehyde (15 min, RT), permeabilization with PBS/0.5% Triton X-100 (15 min, RT) and blocking with 2% BSA in PBS (60 min, RT). The EdU signal was detected using Click-iT chemistry, according to the manufacturer’s protocol (ThermoFisher). Cells were also labelled for phospho- Ser807/811 Rb (phospho-Rb) using Rabbit mAb 8516 (CST) at 1:2000 in blocking solu- tion, overnight at 4 °C. Unbound primary antibody was washed three times in PBS and secondary Alexa-conjugated antibodies were used to detect the signal (ThermoFisher, 1:1000, 1 h at RT). Finally, nuclei were labelled with Hoechst 33258 (1 μg/ml, 15 min RT) Wiecek et al. Genome Biology (2023) 24:128 Page 24 of 35 before imaging on a high-content widefield Operetta microscope, 20 × N.A. 0.8. Auto- mated image analysis (Harmony, PerkinElmer) was used to segment and quantify nuclear signals in imaged cells. Quiescent cells were defined by the absence of EdU or phospho- Rb staining, determined by quantification of their nuclear expression (Fig. 1e–f ). Endogenous PCNA was labelled at the N-terminus with a cDNA encoding mRuby in both A549 and NCI-H1944 cells, using AAV-mediated gene-targeting, according to methods described in Zerjatke et al. [54]. mRuby-expressing cells were sorted into 50:50 conditioned:fresh media at single-cell density into 96-well plates by FACS and single-cell clones expanded. For live-cell imaging, 500 cells in phenol-red free media were plated per well of a 384 CellCarrierUltra plate (PerkinElmer) the day before imaging. Prior to imaging, a breathable membrane was applied to the plate and cells were imaged on the Operetta HCS microscope (PerkinElmer) at 37 °C, 5% CO2 using the 20 × N.A. 0.8 objective and at 10–12  min intervals for 48  h. Images were then exported and G0/G1 length (time from mitotic exit to S-phase entry) was analysed manually in FIJI. The G0 arrest scores for cancer cell lines were calculated using corresponding log- transformed RPKM normalised bulk RNA-seq data from the Cancer Cell Line Encyclo- pedia (CCLE) database [129]. CEP89 was depleted by ON-Target siRNA Pool from Horizon. NCI-H1299 cells were reverse transfected in 384-well plates with 20  nM of non-targeting control (NTC) or CEP89-targeting siRNA using Lipofectamine RNAiMax (ThermoFisher), according to the manufacturer’s instructions. Cells were left for 24  h, before 5  μM EdU was added for the final 24  h and then cells were processed as above to determine the quiescent fraction. To determine the level of Cep89 depletion by western blot, cells were reverse transfected with siRNA in 24-well plates. Forty-eight hours after transfection, cells were lysed directly in 1 × SDS sample buffer with 1 mM DTT (ThermoFisher). Samples were separated on pre-cast 4–20% Tris-Glycine gels, transferred to PVDF using the iBlot2 system and membranes blocked in blocking buffer (5% milk in TBS) for 1 h at RT. The membrane was then cut and the upper half was incubated in 1:1000 Cep89 antibody (Sigma, HPA040056), the bottom half in B-actin antibody 1:2000 (CST; 3700S) diluted in blocking buffer overnight at 4 °C. Membranes were washed three times in TBS-0.05% TritonX-100 before being incubated in secondary anti-rabbit (Cep89) or anti-mouse (B-actin) HRP conjugated antibodies (CST 7074P2 and CST 7076P2, respectively) diluted 1:2000 in blocking buffer for 1  h at RT. Membranes were washed three times again and signal detected using Clarity ECL solution (BioRad) and scanned on an Amer- sham ImageQuant 800 analyser. Multi‑omics discovery cohort FPKM normalised RNA-sequencing expression data, copy number variation gene- level data, RPPA levels for p27 as well as mutation annotation files aligned against the GRCh38 human reference genome from the Mutect2 pipeline were downloaded using the TCGABiolinks R package [130] for 9712 TCGA primary tumour samples across 31 solid cancer types. Haematological malignancies were excluded as the G0 markers were derived in epithelial cells and might not be equally suited to capture this pheno- type in blood. For patients with multiple samples available, one RNA-seq barcode entry was selected for each individual patient resulting in 9631 total entries. All expression Wiecek et al. Genome Biology (2023) 24:128 Page 25 of 35 data were log-transformed for downstream analysis. During G0 arrest score calculation, expression data for the primary tumour samples was scaled according to tumour purity estimates reported by Hoadley et al. [131] to account for potential confounding cell cycle arrest signals coming from non-tumour cells in the microenvironment. Samples with purity estimates lower than 30% were removed, leaving 8005 samples for downstream analysis. The mutation rates of all TCGA primary tumour samples were determined by log- transforming the total number of mutations in each sample divided by the length of the exome capture (38 Mb). TP53 functional status was assessed based on somatic mutation and copy num- ber alterations as described in Zhang et al. [132]. TP53 mutation and copy number for the TCGA tumours were downloaded from cBioPortal (http:// www. cbiop ortal. org). Tumours with TP53 oncogenic mutations (annotated by OncoKB) and copy-number alterations (GISTIC score ≤ − 1) were assigned as TP53 mutant and CNV loss. Tumours without these TP53 alterations were assigned as TP53 wild type. The effects of the TP53 mutation status on G0 arrest were then determined with a linear model approach with the G0 arrest score as a dependent variable and mutational status as an independent variable. The P values were FDR-adjusted. APOBEC mutagenesis enriched samples were determined through pan-cancer clus- tering of mutational signature contributions as described in Wiecek et  al. [133]. The APOBEC mutagenesis cluster was defined as the cluster with highest mean SBS2 and SBS13 contribution. This was repeated 100 times and only samples which appeared in the APOBEC cluster at least 50 times were counted as being APOBEC enriched. Aneuploidy scores and whole-genome duplication events across TCGA samples were obtained from Taylor et  al. [134]. Microsatellite instability status for uterine corpus endometrial carcinoma, as well as stomach and colon adenocarcinoma samples were obtained from Cortes-Ciriano et  al. [78]. Telomerase enzymatic activity ‘EXTEND’ scores were obtained from Noureen et  al. [62]. Expression-based cancer cell stemness indices were obtained from Malta et al. [63]. Centrosome amplification transcriptomic signature (CA20) scores were obtained from Almeida et al. [89]. PHATE dimensionality reduction The phateR R package [135] was used to perform the dimensionality reduction with a constant seed for reproducibility. The ComBat function from the sva R package [136] was used to remove tissue-specific expression patterns from the TCGA RNA-seq data. Cancer stem cell division estimates The mean stem cell division estimates for different cancer types used in this study were obtained from Tomasetti and Vogelstein [55]. Mutational signature estimation Mutational signature contributions were inferred as described in Wiecek et al. [133]. Wiecek et al. Genome Biology (2023) 24:128 Page 26 of 35 Machine learning of G0 arrest‑linked features via ensemble elastic net regression models The COSMIC database was used to source a list of 723 known drivers of tumorigenesis (tiers 1 + 2); 285 oncogenes and tumour suppressors from a curated list showed a sig- nificant enrichment or depletion of mutations or copy number variants in samples with high levels of G0 arrest either pan-cancer or within individual TCGA studies. To classify G0 arrest-prone from fast proliferating tumours, the 285 genes were used as input features for an ensemble elastic net regression model along the tumour muta- tional rate, whole-genome doubling estimates, ploidy, aneuploidy scores and 15 muta- tional signatures, which showed a significant correlation with G0 arrest scores either pan-cancer or within individual TCGA studies. The caret R package was used to build an elastic net regression model 1000 times on the training dataset of 3753 TCGA pri- mary tumour samples (80% of the total dataset). Only samples with at least 50 mutations were used in the model, for which mutational signatures could be reliably estimated. For each of the 1000 iterations, we randomly selected 90% of the samples from the training dataset to build the model. Only features which were included in all 1000 model itera- tions were selected for further analysis. To test the performance of our approach, a linear regression model was built using the reduced list of genomic features and their corre- sponding coefficients averaged across the 1000 elastic net regression model iterations. When applying the resulting linear regression model on the internal validation dataset of 936 samples, we found a strong correlation between the observed and predicted G0 arrest scores (R = 0.73, p < 2.2e − 16). SHAP values for the linear regression model used to predict G0 arrest scores were obtained using the fastshap R package. Gene enrichment and network analysis Gene set enrichment analysis was carried out using the ReactomePA R package, as well as GeneMania [137] and ConsensusPathDB [138]. Interactions between CEP89 and other cell cycle components were inferred using the list of cell cycle genes provided by cBioPortal and GeneMania to reconstruct the expanded network with direct interactors (STAG1, CCND2, STAT3). Networks were visualised using Cytoscape [139]. Gene lists Genes associated with the G1 phase of the cell cycle were obtained from the curated ‘REACTOME_G1_PHASE’ list deposited at MSigDB. Genes associated with the G1/S and G2/M phases of the cell cycle were obtained from Tirosh et al. [51]. Genes associated with apoptosis were obtained from the curated ‘HALLMARK_ APOPTOSIS’ list deposited at MSigDB. Genes associated with the senescence-associated secretory phenotype were obtained from Basisty et al. [60]. Lists of genes making up the various DNA damage repair path- ways were derived from Pearl et al. [140]. Genes associated with contact inhibition were obtained from the curated ‘contact inhi- bition’ gene ontology term. Genes associated with serum starvation were obtained from the curated ‘REACTOME_CELLULAR_RESPONSE_TO_STARVATION’ list deposited at MSigDB. MEK inhibition was assessed based on the activity of the MAPK pathway as Wiecek et al. Genome Biology (2023) 24:128 Page 27 of 35 determined using an expression signature (MPAS) consisting of 10 downstream MAPK transcripts [92]. Validation of the genomic constraints of G0 arrest For elastic net model feature validation, RNA-seq data was downloaded for six cancer studies from cBioPortal [141], along with patient-matched whole-genome, whole-exome and targeted sequencing data. The 6 datasets used comprise breast cancer (SMC [142] and METABRIC [143]), paediatric Wilms’ tumour (TARGET [144]), bladder cancer, prostate adenocarcinoma and sarcoma (MSKCC [145–147]) studies. The data were pro- cessed and analysed in the same manner as the TCGA data. RNA-seq data for 27 MCF7 cell line strains, alongside cell line growth rates and targeted mutational sequencing data were obtained from Ben-David et al. [79]. Genomic dependency modelling in breast cancer An ANOVA-based feature importance classification was used and identified 30 genomic features most discriminative of samples with lower and higher than average G0 arrest scores. A random forest model was then built using the identified features and correctly classified samples according to their G0 arrest state with a mean accuracy of 74% across five randomly sampled test datasets from the cohort. Survival analysis Multivariate Cox proportional hazards analysis was carried out using the coxph func- tion from the survival R package. The optimal quiescence score cut-off value of 2.95 was determined pan-cancer using the surv_cutpoint function. We also used this function to determine optimal cut-offs for individual cancer types, as presented in Fig. 6c. Treatment response single‑cell and bulk RNA‑seq datasets Datasets have been obtained from the GEO database through the following GEO Series accession numbers for the cell line experiments: GSE134836, GSE134838, GSE134839, GSE137912, GSE149224, GSE124854, GSE135215, GSE99116, GSE152699, GSE178839, GSE139944, and the following accession numbers for the patient sample datasets: GSE191127, GSE109211, GSE50509, GSE65185, GSE66399, GSE68871, GSE99898 (Additional file  1: Table  S3). Unified treatment response data for TCGA was obtained from Moiso, medRxiv 2021 [148]. The umap R package was used for dimensionality reduction with constant seed for reproducibility. Stress response subtype determination TCGA cohort studies Samples with evidence of cell cycle arrest characterised by a generic G0 score > 0 were further subclassified based on the most likely form of stress response, among CDK4/6 inhibition, contact inhibition, MEK inhibition, spontaneous quiescence or serum Wiecek et al. Genome Biology (2023) 24:128 Page 28 of 35 starvation, using stress-specific expression signatures. We opted for a conservative approach and classed each sample with a high level of G0 arrest into a specific stress response subtype if the arrest score for the corresponding programme was higher than one standard deviation of the distribution across the TCGA cohort and if the score was significantly higher than for the remaining programmes when assessed using a Student’s t test. Samples which could not be classified into any of the five stress response states characterised in this study were classified as ‘uncertain’. Single‑cell RNA seq treatment response datasets The stress response subtype of individual single cells was inferred by mapping such individual cells onto the reference dataset of MCF10A cells reflecting different forms of G0 arrest obtained from Min and Spencer [28]. The ComBat R package was used to remove the study batch effect between the expression data to be classified and the reference bulk RNA-seq data. PCA dimensionality reduction analysis was then used on the combined datasets using the prcomp R function. For each patient sample or single-cell expression data entry, a k-nearest neighbour algorithm classification was performed using the knn function from the class R package. During the classifica- tion, the three nearest reference bulk RNA-seq data points were considered, with two nearest neighbours with identical class needed for classification. Optimisation of the G0 arrest signature We investigated if a subset of the 139 G0 phase-related genes could act as a more reli- able marker of cell cycle arrest that would bypass dropout issues in single-cell data. This was performed in three steps: (1) Assessment of individual importance as G0 arrest marker for a given gene We collected three high confidence single-cell expression datasets separating arrested from proliferating cells. A random forest model was trained on each dataset sepa- rately to predict the state (G0 arrest/cycling) of a given cell based on the expression levels of the 139 genes in the signature. The Gini indices corresponding to each gene in the model were normalised to a range of values between 0 and 1, which would reflect how important an individual gene was for determining G0 arrest state relative to the other 138 genes. The procedure was repeated 1000 times for each of the three datasets, and the average Gini coefficients across iterations were stored. (2) Prioritisation of gene subsets based on cumulative importance in the model Genes were placed in the candidate subset if their importance metric was above a given threshold in at least one of the datasets. By gradually increasing the threshold from 0 to 1, different gene combinations were produced. (3) External validation of candidate subsets The gene combinations in (2) were tested for their ability to predict G0 arrest. For this, a separate validation dataset was utilized, which contained gene expression levels for the 139 genes in the 10 lung cancer cell lines previously employed for experimental Wiecek et al. Genome Biology (2023) 24:128 Page 29 of 35 validation, along with the quiescence state of the lines as inferred by phospho-Rb and EdU staining. For each gene subset, a combined Z-score of G0 arrest was cal- culated from the expression levels as described previously. The correlations between this Z-score and the two experimental measurements of quiescence were used to establish the ability of a gene combination to predict quiescence. Among the top performing subsets, a 35 gene signature with a mean correlation of 78% between predicted and measured G0 arrest levels in the test data (p = 0.016) showed the highest correlation with phospho-Rb measurements capturing short-lived G0 arrest, the more common state observed in single-cell treatment datasets. Therefore, this signature was deemed to achieve the best trade-off between gene numbers and sig- nal capture. The optimised gene signature is provided in Additional file 1: Table S4. Statistical analysis Groups were compared using a two-sided Student’s t test, Wilcoxon rank-sum test or ANOVA, as appropriate. p-values were adjusted for multiple testing where appropriate using the Benjamini–Hochberg method. Graphs were generated using the ggplot2 and ggpubr R packages. Supplementary Information The online version contains supplementary material available at https:// doi. org/ 10. 1186/ s13059‑ 023‑ 02963‑4. Additional file 1: Table S1. The 139‑gene signature of G0 arrest. The up‑ or downregulation status in G0 arrest is indicated for each gene. Table S2. Summary of external G0 arrest score validation datasets. Table S3. Summary of external cancer cell line treatment response data. Table S4. Results of the signature optimisation analysis. Additional file 2: Fig. S1. G0 arrest score evaluation and validation. Fig. S2. Expression and genomic patterns of G0 arrest in tumours. Fig. S3. Modelling and validating a pan‑cancer classifier of G0 arrest. Fig. S4. Genomic landscape of G0 arrest in breast cancer. Fig. S5. CEP89 expression is associated with G0 arrest and has prognostic value. Fig. S6. Stress response programme validation and links with p53 status. Fig. S7. Relevance of G0 arrest to clinical outcome in cancer. Fig. S8. G0 arrest signatures of drug tolerance in single cell data. Fig. S9. G0 arrest dynamics upon various treatment modalities. Fig. S10. G0 arrest levels in patient samples compared between responders and non‑respond‑ ers to various cancer treatments. Fig. S11. Application of the reduced G0 arrest signature to scRNA‑seq data. Additional file 3. Review history. Acknowledgements We would like to thank Prof Chris Barnes for the very helpful discussions and input on the findings of the study. Review history The review history is available as Additional file 3. Peer review information Wenjing She was the primary editor of this article and managed its editorial process and peer review in collaboration with the rest of the editorial team. Authors’ contributions MS designed the study and supervised the computational analyses. ARB designed and supervised the experimental validation in cell lines. GB supervised the analysis of p53 functional association. AJW developed the quiescence scoring methodology and performed all computational analyses to validate and apply it in bulk and single‑cell datasets, as well as link it to genomic features. SC performed the experimental validation of G0 arrest prevalence and CEP89 association with G0 arrest in cell lines. DK performed the inference of the minimal signature of G0 arrest applicable in single‑cell data. MPC performed the random forest modelling and feature selection in breast cancer. LG helped with the gene prior‑ itisation. GMT wrote the code for batch effect correction and for PCA mapping of single‑cell data on a reference dataset. DHJ performed the APOBEC enrichment classification. PZ and LX performed the G0 arrest comparison of p53 wild‑type and mutated cancers. MS, AJW, ARB and SC wrote the manuscript, with contributions from all other authors. All authors read and approved the manuscript. Authors’ Twitter handles Twitter handles: @Alexis_Barr (Alexis R. Barr), @mariasecrier (Maria Secrier). Wiecek et al. Genome Biology (2023) 24:128 Page 30 of 35 Funding AJW and DHJ were supported by MRC DTP grants (MR/N013867/1). MPC was supported by an Academy of Medical Science Springboard award (SBF004\1042). GMT was supported by a Wellcome Seed Award in Science (215296/Z/19/Z). MS was supported by a UKRI Future Leaders Fellowship (MR/T042184/1). Work in MS’s lab was supported by a BBSRC equipment grant (BB/R01356X/1) and a Wellcome Institutional Strategic Support Fund (204841/Z/16/Z). ARB and SC are supported by a CRUK CDF (C63833/A25729) and work in ARB’s lab is supported by MRC core‑funding to the London Institute of Medical Sciences (MC‑A658‑5TY60). Availability of data and materials The results published here are in part based upon data generated by the TCGA Research Network (https:// www. cancer. gov/ tcga), METABRIC (https:// ega‑ archi ve. org/ studi es/ EGAS0 00000 00083), MSK‑IMPACT (https:// www. mskcc. org/ msk‑ impact) or deposited at cBioPortal (https:// www. cbiop ortal. org/). The following expression datasets from the Gene Expression Omnibus (GEO, https:// www. ncbi. nlm. nih. gov/ geo/) have also been employed: GSE114012 [48], GSE131594 [45], GSE137912 [12], GSE152699 [49], GSE75367 [47], GSE83142 [46], GSE93991 [44], GSE134836 [13], GSE134838 [13], GSE134839 [13], GSE124854 [93], GSE135215 [94], GSE99116 [93], GSE178839 [149], GSE149224 [100], GSE139944 [102], GSE191127 [150], GSE109211 [151], GSE50509 [152], GSE65185 [153], GSE66399 [154], GSE68871 [155] and GSE99898 [156]. The GEO datasets employed in the analyses are summarised in Additional file 1: Tables S2 and S3. All codes developed for the purpose of this study can be found at the following repository, released under a GNU Gen‑ eral Public License v3.0 at github: https:// github. com/ secri erlab/ Cance rG0Ar rest [157] and Zenodo (doi: 10. 5281/ zenodo. 78406 72) [158]. Declarations Ethics approval and consent to participate All data employed in this study comply with ethical regulations, with approval and informed consent for collection and sharing already obtained by the relevant consortia where the data were obtained from (TCGA, METABRI, MSK‑IMPACT). Consent for publication Not applicable. Competing interests The authors declare that they have no competing interests. Received: 1 June 2022 Accepted: 7 May 2023 References 1. Hanahan D, Weinberg RA. The hallmarks of cancer. Cell. 2000;100(1):57–70. 2. van Dijk D, Dhar R, Missarova AM, Espinar L, Blevins WR, Lehner B, et al. Slow‑growing cells within isogenic popula‑ tions have increased RNA polymerase error rates and DNA damage. Nat Commun. 2015;6:7972. 3. Davis JE, Kirk J, Ji Y, Tang DG. Tumor dormancy and slow‑cycling cancer cells. Adv Exp Med Biol. 2019;1164:199–206. 4. Phan TG, Croucher PI. The dormant cancer cell life cycle. Nat Rev Cancer. 2020;20(7):398–411. 5. Rehman SK, Haynes J, Collignon E, Brown KR, Wang Y, Nixon AML, et al. Colorectal cancer cells enter a diapause‑ like DTP state to survive chemotherapy. Cell. 2021;184(1):226‑42.e21. 6. Chen J, Li Y, Yu TS, McKay RM, Burns DK, Kernie SG, et al. A restricted cell population propagates glioblastoma growth after chemotherapy. Nature. 2012;488(7412):522–6. 7. Puig I, Tenbaum SP, Chicote I, Arqués O, Martínez‑Quintanilla J, Cuesta‑Borrás E, et al. TET2 controls chemoresistant slow‑cycling cancer cell survival and tumor recurrence. J Clin Invest. 2018;128(9):3887–905. 8. Cabanos HF, Hata AN. Emerging insights into targeted therapy‑tolerant persister cells in cancer. Cancers (Basel). 2021;13(11):2666. 9. Sharma SV, Lee DY, Li B, Quinlan MP, Takahashi F, Maheswaran S, et al. A chromatin‑mediated reversible drug‑ tolerant state in cancer cell subpopulations. Cell. 2010;141(1):69–80. 10. Swayden M, Chhouri H, Anouar Y, Grumolato L. Tolerant/persister cancer cells and the path to resistance to tar‑ geted therapy. Cells. 2020;9(12):2601. 11. Hsu CH, Altschuler SJ, Wu LF. Patterns of early p21 dynamics determine proliferation‑senescence cell fate after chemotherapy. Cell. 2019;178(2):361‑73.e12. 12. Xue JY, Zhao Y, Aronowitz J, Mai TT, Vides A, Qeriqi B, et al. Rapid non‑uniform adaptation to conformation‑specific KRAS(G12C) inhibition. Nature. 2020;577(7790):421–5. 13. Aissa AF, Islam ABMM, Ariss MM, Go CC, Rader AE, Conrardy RD, et al. Single‑cell transcriptional changes associated with drug tolerance and response to combination therapies in cancer. Nat Commun. 2021;12(1):1628. 14. Malladi S, Macalinao DG, Jin X, He L, Basnet H, Zou Y, et al. Metastatic latency and immune evasion through auto‑ crine inhibition of WNT. Cell. 2016;165(1):45–60. 15. Ribas A. Adaptive immune resistance: how cancer protects from immune attack. Cancer Discov. 2015;5(9):915–9. 16. Sosa MS, Bragado P, Aguirre‑Ghiso JA. Mechanisms of disseminated cancer cell dormancy: an awakening field. Nat Rev Cancer. 2014;14(9):611–22. 17. Barkan D, El Touny LH, Michalowski AM, Smith JA, Chu I, Davis AS, et al. Metastatic growth from dormant cells induced by a col‑I‑enriched fibrotic environment. Cancer Res. 2010;70(14):5706–16. Wiecek et al. Genome Biology (2023) 24:128 Page 31 of 35 18. Masago K, Fujita S, Yatabe Y. Targeting minimal residual disease after surgery with molecular targeted therapy: the real path to a cure? J Thorac Dis. 2018;10(Suppl 17):S1982–5. 19. Coller HA, Sang L, Roberts JM. A new description of cellular quiescence. PLoS Biol. 2006;4(3):e83. 20. Rittershaus ES, Baek SH, Sassetti CM. The normalcy of dormancy: common themes in microbial quiescence. Cell Host Microbe. 2013;13(6):643–51. 21. Miles S, Bradley GT, Breeden LL. The budding yeast transition to quiescence. Yeast. 2021;38(1):30–8. 22. Marescal O, Cheeseman IM. Cellular mechanisms and regulation of quiescence. Dev Cell. 2020;55(3):259–71. 23. Arora M, Moser J, Phadke H, Basha AA, Spencer SL. Endogenous replication stress in mother cells leads to quies‑ cence of daughter cells. Cell Rep. 2017;19(7):1351–64. 24. Barr AR, Cooper S, Heldt FS, Butera F, Stoy H, Mansfeld J, et al. DNA damage during S‑phase mediates the proliferation‑quiescence decision in the subsequent G1 via p21 expression. Nat Commun. 2017;8:14728. 25. Heldt FS, Barr AR, Cooper S, Bakal C, Novák B. A comprehensive model for the proliferation‑quiescence decision in 26. response to endogenous DNA damage in human cells. Proc Natl Acad Sci U S A. 2018;115(10):2532–7. Itahana K, Dimri GP, Hara E, Itahana Y, Zou Y, Desprez PY, et al. A role for p53 in maintaining and establishing the quiescence growth arrest in human cells. J Biol Chem. 2002;277(20):18206–14. 27. Sadasivam S, DeCaprio JA. The DREAM complex: master coordinator of cell cycle‑dependent gene expression. Nat Rev Cancer. 2013;13(8):585–95. 28. Min M, Spencer SL. Spontaneously slow‑cycling subpopulations of human cells originate from activation of stress‑ response pathways. PLoS Biol. 2019;17(3):e3000178. 29. García‑Gutiérrez L, Delgado MD, León J. MYC oncogene contributions to release of cell cycle brakes. Genes (Basel). 2019;10(3):244. 30. Aguirre‑Ghiso JA, Estrada Y, Liu D, Ossowski L. ERK(MAPK) activity as a determinant of tumor growth and dor‑ mancy; regulation by p38(SAPK). Cancer Res. 2003;63(7):1684–95. 31. Spencer SL, Cappell SD, Tsai FC, Overton KW, Wang CL, Meyer T. The proliferation‑quiescence decision is controlled by a bifurcation in CDK2 activity at mitotic exit. Cell. 2013;155(2):369–83. 32. Miller I, Min M, Yang C, Tian C, Gookin S, Carter D, et al. Ki67 is a graded rather than a binary marker of proliferation versus quiescence. Cell Rep. 2018;24(5):1105‑12.e5. 33. Reya T, Morrison SJ, Clarke MF, Weissman IL. Stem cells, cancer, and cancer stem cells. Nature. 2001;414(6859):105–11. 34. Kleinsmith LJ, Pierce GB Jr. Multipotentiality of single embryonal carcinoma cells. Cancer Res. 1964;24:1544–51. 35. Ashraf HM, Moser J, Spencer SL. Senescence evasion in chemotherapy: a sweet spot for p21. Cell. 2019;178(2):267–9. 36. Basile KJ, Aplin AE. Resistance to chemotherapy: short‑term drug tolerance and stem cell‑like subpopulations. Adv Pharmacol. 2012;65:315–34. 37. Basu S, Dong Y, Kumar R, Jeter C, Tang DG. Slow‑cycling (dormant) cancer cells in therapy resistance, cancer relapse and metastasis. Semin Cancer Biol. 2022;78:90–103. 38. Dey‑Guha I, Alves CP, Yeh AC, Salony, Sole X, Darp R, et al. A mechanism for asymmetric cell division resulting in proliferative asynchronicity. Mol Cancer Res. 2015;13(2):223–30. 39. Turati VA, Guerra‑Assunção JA, Potter NE, Gupta R, Ecker S, Daneviciute A, et al. Chemotherapy induces canaliza‑ tion of cell state in childhood B‑cell precursor acute lymphoblastic leukemia. Nat Cancer. 2021;2(8):835–52. 40. Lee E, Chuang HY, Kim JW, Ideker T, Lee D. Inferring pathway activity toward precise disease classification. PLoS Comput Biol. 2008;4(11):e1000217. 41. Barbie DA, Tamayo P, Boehm JS, Kim SY, Moody SE, Dunn IF, et al. Systematic RNA interference reveals that onco‑ genic KRAS‑driven cancers require TBK1. Nature. 2009;462(7269):108–12. 42. Hänzelmann S, Castelo R, Guinney J. GSVA: gene set variation analysis for microarray and RNA‑seq data. BMC Bioinformatics. 2013;14:7. 43. Foroutan M, Bhuva DD, Lyu R, Horan K, Cursons J, Davis MJ. Single sample scoring of molecular phenotypes. BMC Bioinformatics. 2018;19(1):404. 44. Aulestia FJ, Néant I, Dong J, Haiech J, Kilhoffer MC, Moreau M, et al. Quiescence status of glioblastoma stem‑like + cells involves remodelling of Ca(2 ) signalling and mitochondrial shape. Sci Rep. 2018;8(1):9731. 45. Kurppa KJ, Liu Y, To C, Zhang T, Fan M, Vajdi A, et al. Treatment‑induced tumor dormancy through YAP‑mediated transcriptional reprogramming of the apoptotic pathway. Cancer Cell. 2020;37(1):104‑22.e12. 46. Ebinger S, Özdemir EZ, Ziegenhain C, Tiedt S, Castro Alves C, Grunert M, et al. Characterization of rare, dormant, and therapy‑resistant cells in acute lymphoblastic leukemia. Cancer Cell. 2016;30(6):849–62. 47. Jordan NV, Bardia A, Wittner BS, Benes C, Ligorio M, Zheng Y, et al. HER2 expression identifies dynamic functional states within circulating breast cancer cells. Nature. 2016;537(7618):102–6. 48. Buczacki SJA, Popova S, Biggs E, Koukorava C, Buzzelli J, Vermeulen L, et al. Itraconazole targets cell cycle hetero‑ geneity in colorectal cancer. J Exp Med. 2018;215(7):1891–912. 49. Grigore F, Yang H, Hanson ND, VanBrocklin MW, Sarver AL, Robinson JP. BRAF inhibition in melanoma is associated with the dysregulation of histone methylation and histone methyltransferases. Neoplasia. 2020;22(9):376–89. 50. Vairo G, Soos TJ, Upton TM, Zalvide J, DeCaprio JA, Ewen ME, et al. Bcl‑2 retards cell cycle entry through p27(Kip1), pRB relative p130, and altered E2F regulation. Mol Cell Biol. 2000;20(13):4745–53. 51. Tirosh I, Izar B, Prakadan SM, Wadsworth MH, Treacy D, Trombetta JJ, et al. Dissecting the multicellular ecosystem of metastatic melanoma by single‑cell RNA‑seq. Science. 2016;352(6282):189–96. 52. Fujimaki K, Li R, Chen H, Della Croce K, Zhang HH, Xing J, et al. Graded regulation of cellular quiescence depth between proliferation and senescence by a lysosomal dimmer switch. Proc Natl Acad Sci U S A. 2019;116(45):22624–34. 53. Stallaert W, Kedziora KM, Taylor CD, Zikry TM, Ranek JS, Sobon HK, et al. The structure of the human cell cycle. Cell Syst. 2022;13(1):103. 54. Zerjatke T, Gak IA, Kirova D, Fuhrmann M, Daniel K, Gonciarz M, et al. Quantitative cell cycle analysis based on an endogenous all‑in‑one reporter for cell tracking and classification. Cell Rep. 2017;19(9):1953–66. Wiecek et al. Genome Biology (2023) 24:128 Page 32 of 35 55. Tomasetti C, Vogelstein B. Cancer etiology. Variation in cancer risk among tissues can be explained by the number of stem cell divisions. Science. 2015;347(6217):78–81. 56. Hernandez‑Segura A, de Jong TV, Melov S, Guryev V, Campisi J, Demaria M. Unmasking transcriptional heteroge‑ neity in senescent cells. Curr Biol. 2017;27(17):2652‑60.e4. 57. Fujimaki K, Yao G. Cell dormancy plasticity: quiescence deepens into senescence through a dimmer switch. Physiol Genomics. 2020;52(11):558–62. 58. Dimri GP, Lee X, Basile G, Acosta M, Scott G, Roskelley C, et al. A biomarker that identifies senescent human cells in culture and in aging skin in vivo. Proc Natl Acad Sci. 1995;92(20):9363. 59. Debacq‑Chainiaux F, Erusalimsky JD, Campisi J, Toussaint O. Protocols to detect senescence‑associated beta‑galactosidase (SA‑betagal) activity, a biomarker of senescent cells in culture and in vivo. Nat Protoc. 2009;4(12):1798–806. 60. Basisty N, Kale A, Jeon OH, Kuehnemann C, Payne T, Rao C, et al. A proteomic atlas of senescence‑associated secretomes for aging biomarker development. PLoS Biol. 2020;18(1):e3000599. 61. Lee BY, Han JA, Im JS, Morrone A, Johung K, Goodwin EC, et al. Senescence‑associated beta‑galactosidase is lysosomal beta‑galactosidase. Aging Cell. 2006;5(2):187–95. 62. Noureen N, Wu S, Lv Y, Yang J, Alfred Yung WK, Gelfond J, et al. Integrated analysis of telomerase enzymatic activity unravels an association with cancer stemness and proliferation. Nat Commun. 2021;12(1):139. 63. Malta TM, Sokolov A, Gentles AJ, Burzykowski T, Poisson L, Weinstein JN, et al. Machine learning identifies stemness features associated with oncogenic dedifferentiation. Cell. 2018;173(2):338‑54.e15. 64. Fischer M. Census and evaluation of p53 target genes. Oncogene. 2017;36(28):3943–56. 65. Andreassen PR, Lohez OD, Lacroix FB, Margolis RL. Tetraploid state induces p53‑dependent arrest of nontrans‑ formed mammalian cells in G1. Mol Biol Cell. 2001;12(5):1315–28. 66. Di Leonardo A, Khan SH, Linke SP, Greco V, Seidita G, Wahl GM. DNA rereplication in the presence of mitotic spin‑ dle inhibitors in human and mouse fibroblasts lacking either p53 or pRb function. Cancer Res. 1997;57(6):1013–9. 67. Ganem NJ, Cornils H, Chiu SY, O’Rourke KP, Arnaud J, Yimlamai D, et al. Cytokinesis failure triggers hippo tumor suppressor pathway activation. Cell. 2014;158(4):833–48. 68. Alexandrov LB, Kim J, Haradhvala NJ, Huang MN, Tian Ng AW, Wu Y, et al. The repertoire of mutational signatures in human cancer. Nature. 2020;578(7793):94–101. 69. Alexandrov LB, Nik‑Zainal S, Wedge DC, Aparicio SAJR, Behjati S, Biankin AV, et al. Signatures of mutational pro‑ cesses in human cancer. Nature. 2013;500(7463):415–21. 70. Pich O, Muiños F, Lolkema MP, Steeghs N, Gonzalez‑Perez A, Lopez‑Bigas N. The mutational footprints of cancer therapies. Nat Genet. 2019;51(12):1732–40. 71. Miyazono K, Maeda S, Imamura T. Coordinate regulation of cell growth and differentiation by TGF‑beta superfam‑ ily and Runx proteins. Oncogene. 2004;23(24):4232–7. 72. Cibis H, Biyanee A, Dörner W, Mootz HD, Klempnauer KH. Characterization of the zinc finger proteins ZMYM2 and ZMYM4 as novel B‑MYB binding proteins. Sci Rep. 2020;10(1):8390. 73. Maley CC, Reid BJ. Natural selection in neoplastic progression of Barrett’s esophagus. Semin Cancer Biol. 2005;15(6):474–83. 74. Bardeesy N, DePinho RA. Pancreatic cancer biology and genetics. Nat Rev Cancer. 2002;2(12):897–909. 75. Indovina P, Pentimalli F, Casini N, Vocca I, Giordano A. RB1 dual role in proliferation and apoptosis: cell fate control and implications for cancer therapy. Oncotarget. 2015;6(20):17873–90. 76. Mas‑Ponte D, Supek F. DNA mismatch repair promotes APOBEC3‑mediated diffuse hypermutation in human cancers. Nat Genet. 2020;52(9):958–68. 77. Kloor M, von Knebel DM. The immune biology of microsatellite‑unstable cancer. Trends Cancer. 2016;2(3):121–33. 78. Cortes‑Ciriano I, Lee S, Park W‑Y, Kim T‑M, Park PJ. A molecular portrait of microsatellite instability across multiple cancers. Nat Commun. 2017;8(1):15180. 79. Ben‑David U, Siranosian B, Ha G, Tang H, Oren Y, Hinohara K, et al. Genetic and transcriptional evolution alters cancer cell line drug response. Nature. 2018;560(7718):325–30. 80. Yersal O, Barutca S. Biological subtypes of breast cancer: prognostic and therapeutic implications. World J Clin Oncol. 2014;5(3):412–24. 81. Jakobsen L, Vanselow K, Skogs M, Toyoda Y, Lundberg E, Poser I, et al. Novel asymmetrically localizing components of human centrosomes identified by complementary proteomics methods. EMBO J. 2011;30(8):1520–35. 82. Sillibourne JE, Specht CG, Izeddin I, Hurbain I, Tran P, Triller A, et al. Assessing the localization of centrosomal pro‑ teins by PALM/STORM nanoscopy. Cytoskeleton (Hoboken). 2011;68(11):619–27. 83. Bettencourt‑Dias M, Glover DM. Centrosome biogenesis and function: centrosomics brings new understanding. Nat Rev Mol Cell Biol. 2007;8(6):451–63. 84. Grallert A, Chan KY, Alonso‑Nuñez ML, Madrid M, Biswas A, Alvarez‑Tabarés I, et al. Removal of centrosomal PP1 by NIMA kinase unlocks the MPF feedback loop to promote mitotic commitment in S. pombe. Curr Biol. 2013;23(3):213–22. 85. Wigley WC, Fabunmi RP, Lee MG, Marino CR, Muallem S, DeMartino GN, et al. Dynamic association of proteasomal 86. machinery with the centrosome. J Cell Biol. 1999;145(3):481–90. Itoh K, Jenny A, Mlodzik M, Sokol SY. Centrosomal localization of Diversin and its relevance to Wnt signaling. J Cell Sci. 2009;122(Pt 20):3791–8. 87. Kfoury Y, Nasr R, Favre‑Bonvin A, El‑Sabban M, Renault N, Giron ML, et al. Ubiquitylated Tax targets and binds the IKK signalosome at the centrosome. Oncogene. 2008;27(12):1665–76. 88. Lin A, Wang RT, Ahn S, Park CC, Smith DJ. A genome‑wide map of human genetic interactions inferred from radia‑ tion hybrid genotypes. Genome Res. 2010;20(8):1122–32. 89. de Almeida BP, Vieira AF, Paredes J, Bettencourt‑Dias M, Barbosa‑Morais NL. Pan‑cancer association of a centro‑ some amplification gene expression signature with genomic alterations and clinical outcome. PLoS Comput Biol. 2019;15(3):e1006832. Wiecek et al. Genome Biology (2023) 24:128 Page 33 of 35 90. Kohlmaier G, Loncarek J, Meng X, McEwen BF, Mogensen MM, Spektor A, et al. Overly long centrioles and defec‑ tive cell division upon excess of the SAS‑4‑related protein CPAP. Curr Biol. 2009;19(12):1012–8. 91. Tang C‑JC, Fu R‑H, Wu K‑S, Hsu W‑B, Tang TK. CPAP is a cell‑cycle regulated protein that controls centriole length. Nat Cell Biol. 2009;11(7):825–31. 92. Wagle MC, Kirouac D, Klijn C, Liu B, Mahajan S, Junttila M, et al. A transcriptional MAPK Pathway Activity Score (MPAS) is a clinically relevant biomarker in multiple cancer types. NPJ Precis Oncol. 2018;2(1):7. 93. Hafner M, Mills CE, Subramanian K, Chen C, Chung M, Boswell SA, et al. Multiomics profiling establishes the polyp‑ harmacology of FDA‑approved CDK4/6 inhibitors and the potential for differential clinical activity. Cell Chem Biol. 2019;26(8):1067‑80.e8. 94. Salvador‑Barbero B, Álvarez‑Fernández M, Zapatero‑Solana E, El Bakkali A, Menéndez MDC, López‑Casas PP, et al. CDK4/6 inhibitors impair recovery from cytotoxic chemotherapy in pancreatic adenocarcinoma. Cancer Cell. 2020;37(3):340‑53.e6. 95. Guiley KZ, Stevenson JW, Lou K, Barkovich KJ, Kumarasamy V, Wijeratne TU, et al. p27 allosterically activates cyclin‑ dependent kinase 4 and antagonizes palbociclib inhibition. Science. 2019;366(6471):eaaw2106. 96. Pack LR, Daigh LH, Chung M, Meyer T. Clinical CDK4/6 inhibitors induce selective and immediate dissociation of p21 from cyclin D‑CDK4 to inhibit CDK2. Nat Commun. 2021;12(1):3356. 97. Zhu X, Chen L, Huang B, Wang Y, Ji L, Wu J, et al. The prognostic and predictive potential of Ki‑67 in triple‑negative breast cancer. Sci Rep. 2020;10(1):225. 98. Solé L, Lobo‑Jarne T, Álvarez‑Villanueva D, Alonso‑Marañón J, Guillén Y, Guix M, et al. p53 wild‑type colorectal cancer cells that express a fetal gene signature are associated with metastasis and poor prognosis. Nat Commun. 2022;13(1):2866. 99. Brown JA, Yonekubo Y, Hanson N, Sastre‑Perona A, Basin A, Rytlewski JA, et al. TGF‑β‑induced quiescence mediates chemoresistance of tumor‑propagating cells in squamous cell carcinoma. Cell Stem Cell. 2017;21(5):650‑64.e8. 100. Park SR, Namkoong S, Friesen L, Cho CS, Zhang ZZ, Chen YC, et al. Single‑cell transcriptome analysis of colon cancer cell response to 5‑fluorouracil‑induced DNA damage. Cell Rep. 2020;32(8):108077. 101. Tan X, Thapa N, Sun Y, Anderson RA. A kinase‑independent role for EGF receptor in autophagy initiation. Cell. 2015;160(1–2):145–60. 102. Srivatsan SR, McFaline‑Figueroa JL, Ramani V, Saunders L, Cao J, Packer J, et al. Massively multiplex chemical tran‑ scriptomics at single‑cell resolution. Science. 2020;367(6473):45–51. 103. Hatzis C, Pusztai L, Valero V, Booser DJ, Esserman L, Lluch A, et al. A genomic predictor of response and survival following taxane‑anthracycline chemotherapy for invasive breast cancer. JAMA. 2011;305(18):1873–81. 104. Lee JS, Nair NU, Dinstag G, Chapman L, Chung Y, Wang K, et al. Synthetic lethality‑mediated precision oncology via the tumor transcriptome. Cell. 2021;184(9):2487‑502.e13. 105. Moser J, Miller I, Carter D, Spencer SL. Control of the restriction point by Rb and p21. Proc Natl Acad Sci U S A. 2018;115(35):E8219–27. 106. Bačević K, Lossaint G, Achour TN, Georget V, Fisher D, Dulić V. Cdk2 strengthens the intra‑S checkpoint and coun‑ teracts cell cycle exit induced by DNA damage. Sci Rep. 2017;7(1):13429. 107. Terzi MY, Izmirli M, Gogebakan B. The cell fate: senescence or quiescence. Mol Biol Rep. 2016;43(11):1213–20. 108. Kleffel S, Schatton T. Tumor dormancy and cancer stem cells: two sides of the same coin? Adv Exp Med Biol. 2013;734:145–79. 109. Miller AK, Brown JS, Enderling H, Basanta D, Whelan CJ. The evolutionary ecology of dormancy in nature and in cancer. Front Ecol Evol. 2021;9:676802. 110. Considine MJ, Considine JA. On the language and physiology of dormancy and quiescence in plants. J Exp Bot. 2016;67(11):3189–203. 111. Velappan Y, Signorelli S, Considine MJ. Cell cycle arrest in plants: what distinguishes quiescence, dormancy and differentiated G1? Ann Bot. 2017;120(4):495–509. 112. Sosa MS, Parikh F, Maia AG, Estrada Y, Bosch A, Bragado P, et al. NR2F1 controls tumour cell dormancy via SOX9‑ and RARβ‑driven quiescence programmes. Nat Commun. 2015;6:6170. 113. Fox DB, Garcia NMG, McKinney BJ, Lupo R, Noteware LC, Newcomb R, et al. NRF2 activation promotes the recurrence of dormant tumour cells through regulation of redox and nucleotide metabolism. Nat Metab. 2020;2(4):318–34. 114. Berkel C, Cacan E. Copy number and expression of CEP89, a protein required for ciliogenesis, are increased and predict poor prognosis in patients with ovarian cancer. Cell Biochem Funct. 2022;40(3):298–309. 115. Tanos BE, Yang HJ, Soni R, Wang WJ, Macaluso FP, Asara JM, et al. Centriole distal appendages promote membrane docking, leading to cilia initiation. Genes Dev. 2013;27(2):163–8. 116. Sillibourne JE, Hurbain I, Grand‑Perret T, Goud B, Tran P, Bornens M. Primary ciliogenesis requires the distal append‑ age component Cep123. Biol Open. 2013;2(6):535–45. 117. Wheway G, Nazlamova L, Hancock JT. Signaling through the primary cilium. Front Cell Dev Biol. 2018;6:8. 118. van Bon BW, Oortveld MA, Nijtmans LG, Fenckova M, Nijhof B, Besseling J, et al. CEP89 is required for mitochondrial metabolism and neuronal function in man and fly. Hum Mol Genet. 2013;22(15):3138–51. 119. Huang R, Chen H, Liang J, Li Y, Yang J, Luo C, et al. Dual role of reactive oxygen species and their application in cancer therapy. J Cancer. 2021;12(18):5543–61. 120. Gupta J, del Barco BI, Igea A, Sakellariou S, Pateras IS, Gorgoulis VG, et al. Dual function of p38α MAPK in colon cancer: suppression of colitis‑associated tumor initiation but requirement for cancer cell survival. Cancer Cell. 2014;25(4):484–500. 121. Moon EJ, Giaccia A. Dual roles of NRF2 in tumor prevention and progression: possible implications in cancer treat‑ ment. Free Radic Biol Med. 2015;79:292–9. 122. Sosa MS, Avivar‑Valderas A, Bragado P, Wen HC, Aguirre‑Ghiso JA. ERK1/2 and p38α/β signaling in tumor cell quiescence: opportunities to control dormant residual disease. Clin Cancer Res. 2011;17(18):5850–7. 123. Minassian LM, Cotechini T, Huitema E, Graham CH. Hypoxia‑induced resistance to chemotherapy in cancer. Adv Exp Med Biol. 2019;1136:123–39. Wiecek et al. Genome Biology (2023) 24:128 Page 34 of 35 124. Aguirre‑Ghiso JA, Sosa MS. Emerging topics on disseminated cancer cell dormancy and the paradigm of metasta‑ sis. Annu Rev Cancer Biol. 2018;2(1):377–93. 125. Johnson WE, Li C, Rabinovic A. Adjusting batch effects in microarray expression data using empirical Bayes meth‑ ods. Biostatistics. 2007;8(1):118–27. 126. Amezquita RA, Lun ATL, Becht E, Carey VJ, Carpp LN, Geistlinger L, et al. Orchestrating single‑cell analysis with bioconductor. Nat Methods. 2020;17(2):137–45. 127. McCarthy DJ, Campbell KR, Lun ATL, Wills QF. Scater: pre‑processing, quality control, normalization and visualiza‑ tion of single‑cell RNA‑seq data in R. Bioinformatics. 2017;33(8):1179–86. 128. Lun AT, McCarthy DJ, Marioni JC. A step‑by‑step workflow for low‑level analysis of single‑cell RNA‑seq data with Bioconductor. F1000Res. 2016;5:2122. 129. Barretina J, Caponigro G, Stransky N, Venkatesan K, Margolin AA, Kim S, et al. The Cancer Cell Line Encyclopedia enables predictive modelling of anticancer drug sensitivity. Nature. 2012;483(7391):603–7. 130. Colaprico A, Silva TC, Olsen C, Garofano L, Cava C, Garolini D, et al. TCGAbiolinks: an R/Bioconductor package for integrative analysis of TCGA data. Nucleic Acids Res. 2016;44(8):e71. 131. Hoadley KA, Yau C, Hinoue T, Wolf DM, Lazar AJ, Drill E, et al. Cell‑of‑origin patterns dominate the molecular clas‑ sification of 10,000 tumors from 33 types of cancer. Cell. 2018;173(2):291‑304.e6. 132. Zhang P, Kitchen‑Smith I, Xiong L, Stracquadanio G, Brown K, Richter PH, et al. Germline and somatic genetic variants in the p53 pathway interact to affect cancer risk, progression, and drug response. Cancer Res. 2021;81(7):1667–80. 133. Wiecek AJ, Jacobson DH, Lason W, Secrier M. Pan‑cancer survey of tumor mass dormancy and underlying muta‑ tional processes. Front Cell Dev Biol. 2021;9:1820. 134. Taylor AM, Shih J, Ha G, Gao GF, Zhang X, Berger AC, et al. Genomic and functional approaches to understanding cancer aneuploidy. Cancer Cell. 2018;33(4):676‑89.e3. 135. Moon KR, van Dijk D, Wang Z, Gigante S, Burkhardt DB, Chen WS, et al. Visualizing structure and transitions in high‑ dimensional biological data. Nat Biotechnol. 2019;37(12):1482–92. 136. Leek JT, Johnson WE, Parker HS, Jaffe AE, Storey JD. The sva package for removing batch effects and other unwanted variation in high‑throughput experiments. Bioinformatics. 2012;28(6):882–3. 137. Warde‑Farley D, Donaldson SL, Comes O, Zuberi K, Badrawi R, Chao P, et al. The GeneMANIA prediction server: bio‑ logical network integration for gene prioritization and predicting gene function. Nucleic Acids Res. 2010;38(Web Server issue):W214‑20. 138. Kamburov A, Pentchev K, Galicka H, Wierling C, Lehrach H, Herwig R. ConsensusPathDB: toward a more complete picture of cell biology. Nucleic Acids Res. 2011;39(Database issue):D712‑7. 139. Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT, Ramage D, et al. Cytoscape: a software environment for inte‑ grated models of biomolecular interaction networks. Genome Res. 2003;13(11):2498–504. 140. Pearl LH, Schierz AC, Ward SE, Al‑Lazikani B, Pearl FMG. Therapeutic opportunities within the DNA damage response. Nat Rev Cancer. 2015;15(3):166–80. 141. Cerami E, Gao J, Dogrusoz U, Gross BE, Sumer SO, Aksoy BA, et al. The cBio cancer genomics portal: an open plat‑ form for exploring multidimensional cancer genomics data. Cancer Discov. 2012;2(5):401–4. 142. Kan Z, Ding Y, Kim J, Jung HH, Chung W, Lal S, et al. Multi‑omics profiling of younger Asian breast cancers reveals distinctive molecular signatures. Nat Commun. 2018;9(1):1725. 143. Curtis C, Shah SP, Chin SF, Turashvili G, Rueda OM, Dunning MJ, et al. The genomic and transcriptomic architecture of 2,000 breast tumours reveals novel subgroups. Nature. 2012;486(7403):346–52. 144. Gadd S, Huff V, Walz AL, Ooms AHAG, Armstrong AE, Gerhard DS, et al. A Children’s Oncology Group and TARGET 145. initiative exploring the genetic landscape of Wilms tumor. Nat Genet. 2017;49(10):1487–94. Iyer G, Al‑Ahmadie H, Schultz N, Hanrahan AJ, Ostrovnaya I, Balar AV, et al. Prevalence and co‑occurrence of action‑ able genomic alterations in high‑grade bladder cancer. J Clin Oncol. 2013;31(25):3133–40. 146. Taylor BS, Schultz N, Hieronymus H, Gopalan A, Xiao Y, Carver BS, et al. Integrative genomic profiling of human prostate cancer. Cancer Cell. 2010;18(1):11–22. 147. Barretina J, Taylor BS, Banerji S, Ramos AH, Lagos‑Quintana M, Decarolis PL, et al. Subtype‑specific genomic altera‑ tions define new targets for soft‑tissue sarcoma therapy. Nat Genet. 2010;42(8):715–21. 148. Moiso E. Manual curation of TCGA treatment data and identification of potential markers of therapy response. medRxiv. 2021:2021.04.30.21251941. 149. Therizols G, Bash‑Imam Z, Panthu B, Machon C, Vincent A, Ripoll J, et al. Alteration of ribosome function upon 5‑fluorouracil treatment favors cancer cell drug‑tolerance. Nat Commun. 2022;13(1):173. 150. Hoogstraat M, Lips EH, Mayayo‑Peralta I, Mulder L, Kristel P, van der Heijden I, et al. Comprehensive characteriza‑ tion of pre‑ and post‑treatment samples of breast cancer reveal potential mechanisms of chemotherapy resist‑ ance. NPJ Breast Cancer. 2022;8(1):60. 151. Pinyol R, Montal R, Bassaganyas L, Sia D, Takayama T, Chau GY, et al. Molecular predictors of prevention of recurrence in HCC with sorafenib as adjuvant treatment and prognostic factors in the phase 3 STORM trial. Gut. 2019;68(6):1065–75. 152. Rizos H, Menzies AM, Pupo GM, Carlino MS, Fung C, Hyman J, et al. BRAF inhibitor resistance mechanisms in meta‑ static melanoma: spectrum and clinical impact. Clin Cancer Res. 2014;20(7):1965–77. 153. Hugo W, Shi H, Sun L, Piva M, Song C, Kong X, et al. Non‑genomic and immune evolution of melanoma acquiring MAPKi resistance. Cell. 2015;162(6):1271–85. 154. Dieci MV, Prat A, Tagliafico E, Paré L, Ficarra G, Bisagni G, et al. Integrated evaluation of PAM50 subtypes and immune modulation of pCR in HER2‑positive breast cancer patients treated with chemotherapy and HER2‑ targeted agents in the CherLOB trial. Ann Oncol. 2016;27(10):1867–73. 155. Terragna C, Remondini D, Martello M, Zamagni E, Pantani L, Patriarca F, et al. The genetic and genomic background of multiple myeloma patients achieving complete response after induction therapy with bortezomib, thalidomide and dexamethasone (VTD). Oncotarget. 2016;7(9):9666–79. Wiecek et al. Genome Biology (2023) 24:128 Page 35 of 35 156. Kakavand H, Rawson RV, Pupo GM, Yang JYH, Menzies AM, Carlino MS, et al. PD‑L1 Expression and immune escape in melanoma resistance to MAPK inhibitors. Clin Cancer Res. 2017;23(20):6054–61. 157. Wiecek A, Secrier M. Genomic hallmarks and therapeutic implications of G0 cell cycle arrest in cancer. Github; 2023. https:// github. com/ secri erlab/ Cance rG0Ar rest. 158. Wiecek A, Secrier M. Genomic hallmarks and therapeutic implications of G0 cell cycle arrest in cancer. Zenodo; 2023. https:// zenodo. org/ record/ 78406 73. Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. • fast, convenient online submission • thorough peer review by experienced researchers in your field• rapid publication on acceptance• support for research data, including large and complex data types• gold Open Access which fosters wider collaboration and increased citations maximum visibility for your research: over 100M website views per year • At BMC, research is always in progress.Learn more biomedcentral.com/submissionsReady to submit your researchReady to submit your research ? Choose BMC and benefit from: ? Choose BMC and benefit from:
null
10.1371_journal.pone.0261071.pdf
Data Availability Statement: Data are available from the Zenodo database (DOI: 10.5281/zenodo. 5774499).
Data are available from the Zenodo database (DOI: 10.5281/zenodo. 5774499 ).
RESEARCH ARTICLE Effects of weather and moon phases on emergency medical use after fall injury: A population-based nationwide study Min Ah Yuh1, Kisung KimID Jinwoo Kim5, Sungyoup HongID 1* 2, Seon Hee Woo3, Sikyoung Jeong1, Juseok OhID 4, a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Yuh MA, Kim K, Woo SH, Jeong S, Oh J, Kim J, et al. (2021) Effects of weather and moon phases on emergency medical use after fall injury: A population-based nationwide study. PLoS ONE 16(12): e0261071. https://doi.org/10.1371/journal. pone.0261071 Editor: Quan Yuan, Tsinghua University, CHINA Received: May 8, 2021 Accepted: November 23, 2021 Published: December 31, 2021 Copyright: © 2021 Yuh et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 1 Department of Emergency Medicine, Daejeon St Mary’s Hospital, The Catholic University of Korea College of Medicine, Seoul, Republic of Korea, 2 BioBrain Inc, Daejeon, Republic of Korea, 3 Department of Emergency Medicine, Incheon St Mary’s Hospital, The Catholic University of Korea College of Medicine, Seoul, Republic of Korea, 4 Department of Emergency Medicine, Uijeongbu St Mary’s Hospital, The Catholic University of Korea College of Medicine, Seoul, Republic of Korea, 5 Department of Emergency Medical Service, Daejeon Health Institute of Technology, Daejeon, Republic of Korea * emhong@catholic.ac.kr Abstract Background Previous studies reported that changes in weather and phases of moon are associated with medical emergencies and injuries. However, such studies were limited to hospital or com- munity level without explaining the combined effects of weather and moon phases. We investigated whether changes in weather and moon phases affected emergency depart- ment (ED) visits due to fall injuries (FIs) based on nationwide emergency patient registry data. Methods Nationwide daily data of ED visits after FI were collected from 11 provinces (7 metropolitan cities and 4 rural provinces) in Korea between January 2014 and December 2018. The daily number of FIs was standardized into FI per million population (FPP) in each province. A mul- tivariate regression analysis was conducted to elucidate the relationship between weather factors and moon phases with respect to daily FPP in each province. The correlation between weather factors and FI severity was also analyzed. Data Availability Statement: Data are available from the Zenodo database (DOI: 10.5281/zenodo. 5774499). Results Funding: We declare that this study was supported by Daejeon St. Mary’s Hospital, Clinical Research Institute Grant No. CMCDJ-P-2021013. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. The study analyzed 666,912 patients (418,135 in metropolitan and 248,777 in rural areas) who visited EDs on weekdays. No regional difference was found in age or gender distribu- tion between the two areas. Precipitation, minimum temperature and wind speed showed a significant association with FI in metropolitan areas. In addition, sunshine duration was also substantial risk factors for FI in rural areas. The incidence of FIs was increased on full moon days than on other days in rural areas. Injury severity was associated with weather factors such as minimum temperature, wind speed, and cloud cover. PLOS ONE | https://doi.org/10.1371/journal.pone.0261071 December 31, 2021 1 / 14 PLOS ONE Effects of weather and moon phases on EM use Conclusion Weather changes such as precipitation, minimum temperature, and wind speed are associ- ated with FI in metropolitan and rural areas. In addition, sunshine duration and full moon are significantly associated with FI incidence only in rural areas. Weather factors are associated with FI severity. Introduction Accurate prediction of the need for emergency medical care is critical to provide appropriate services for patients with injuries. Therefore, many countries have an emergency medical sys- tem data registry collected during pre-hospital and in-hospital phases to design emergency medical service (EMS) and implement public health monitoring and planning. Fall injury (FI) is the second major cause of accidental or unintended injury-related deaths worldwide [1]. The mechanism of injury for falls is vertical deceleration due to the force of gravity high place or loss of balance on a slippery surface. FIs in older or disabled individuals increase in winter due to low temperatures and long nights [2]. The slippery ground caused by melting ice, snow-covered ice, and ice is a typical cause of FI in winter [3]. The incidence of FIs is known to have a seasonal variation depending on geographical location, such as coun- tries with a cold climate (Russia, Canada, Sweden, Finland, and Norway) [4, 5] and countries with a warm tropical climate, such as Hong Kong [6]. Weather conditions have been reported to influence the occurrence of trauma and disease. Poor weather conditions may lead to traumatic events [7, 8]. However, other studies reported that outdoor activities even in good weather are related to increased incidence of all kinds of injuries [9]. If we target FI only, snowfall and icy surfaces were associated with FIs in late autumn and winter [6, 7, 10]. But another study reported the increased frequency of FIs was found in better weather with medium mean air temperature and atmospheric pressure during warm season [4]. A full moon has been reportedly associated with potential emergency department (ED) visits after traffic accidents [11] and mortality after motorcycle crashes and accidents [12]. However, Stomp et al. reported that phases other than full moon increased ED visits after all kinds of trauma [9]. Such difference might be attributed to the use of nationwide statistics of road car accidents in two of the three studies, whereas Stomp et al. [9] used all types of trauma data from one ED located in a small suburban area of Netherlands. Therefore, findings from these studies were limited by small sample size in specific regions, restricted data sources or target injury. In summary, previous studies evaluated the role of weather and lunar phases; however, the findings were limited to specific region or involved a small sample size. To prevent unwanted medical errors due to ED crowding and provide prompt and appropriate EMS, it is vital to foresee the demand for emergency medical use due to FIs. The primary purpose of this study was to assess the FI prevalence according to regional characteristics, weather changes and moon phases using the nationwide longitude data. The secondary goal was to determine the effects of weather factors and moon phases on FI severity. Materials and methods Study design and data collection To analyze the association of FI incidence with weather factors and lunar phases, a nationwide epidemiological analysis was conducted based on emergency department usage data obtained PLOS ONE | https://doi.org/10.1371/journal.pone.0261071 December 31, 2021 2 / 14 PLOS ONE Effects of weather and moon phases on EM use from all emergency centers in Korea. The population of mainland Korea and its affiliated islands of 99,000 km2 is approximately 50 million. Korea has a total of 420 registered ERs that are open to all beneficiaries without restriction. The National Emergency Department Infor- mation System (NEDIS) operated by the National Emergency Medical Center (NEMC) pro- spectively collected data of patients who visited all Korean EDs since 2005. This study used the NEDIS data of patients who visited emergency centers after FI, including age, gender, region of occurrence, onset time, injury mechanism, injury severity with Korean Triage and Acuity Scale (KTAS), and outcome of emergency care from January 2014 to December 2018. No per- sonal identifier was included in these data. Data were stored in a secured personal computer. KTAS score consists of five levels of acuity: level 1 (resuscitation), level 2 (urgent), level 3 (emergent), level 4 (non-urgent), and level 5 (delayed). The KTAS was developed as a single triage tool for emergency patients in Korea and has since become nationalized [13]. Daily weather data including precipitation, minimum temperature, mean wind speed (wind speed), sunshine and fog duration, and cloud cover were obtained from the Korea Mete- orological Agency (KMA). Daily precipitation was calculated as the sum of hourly measure- ments for 24 hours. Cloud cover was calculated in integers ranging from 0 to 10 tenths based on visual cloud cover observations from each observation site. A weather station located in the capital city of each province in Korea was selected to represent the weather data collection point. Moon phase data were obtained from a website (https://www.timeanddate.com/moon/ phases/south-korea/). First, pediatric patients under the age of 15 years were excluded from the collected data. FI patients on weekdays excluding Saturday, Sunday, and public holidays (New Year’s Day, Lunar New Year’s Day, Children’s Day, Korean Independence Day, Lunar Thanksgiving Day and Christmas) were also excluded from this study. We analyzed data from a total of 743 nights (182 new moon days, 186 first quarter and full moon days, and 189 3rd quarter nights). The full moon period was determined for three days starting from one night before to one night after the peak full moon night. The same rule was applied to other moon phases. Annual mid-population data of each province were obtained from the central organization for Statistics of Korea (http://kostat.go.kr/portal/eng/index.action). Outcome measurement We counted the daily number of patients who visited EDs after a FI for each province. The number of daily FIs was standardized into FIs per million population (FPP) by dividing with an annual mid-population of the province. This study compared FPP and severity of FI between two regions: 1) metropolitan areas consisting of provinces with a population of more than one million; and 2) rural areas without metropolitan cities within the perimeter of the provincial limit (S1 Fig). The primary outcome was the daily number of ED visits due to FI. It was calculated as the number of patients per million people. The secondary outcome was injury severity of patients and was determined by the mean KTAS score of all daily FI patients in the designated area. Statistical analysis The chi-square test is extremely sensitive to sample size. If the sample size is too large (> 500), any small differences appear statistically significant [14]. Hence, we used Cramer’s V statistics instead of Chi-square test to estimate the association of ordinary factors between two regions. The means of the continuous variable were compared using Student’s t-test between two regions. The number of FI events in a fixed time interval was modeled using Poisson distribu- tion, and thus a generalized linear model (GLM) with a Poisson distribution and log-linear PLOS ONE | https://doi.org/10.1371/journal.pone.0261071 December 31, 2021 3 / 14 PLOS ONE Effects of weather and moon phases on EM use function was used to assess the significance of association between dependent variable (FPP for each day) and independent variables including weather factors and lunar phases. All vari- ables with a p value < 0.2 in the univariate analysis were entered into multivariate analysis. The incidence risk ratios (IRRs) and their 95% confidence intervals (CIs) were calculated for each independent variable. Theoretical FI incidence-factor curve was approximated via non- linear curve fitting with Boltzmann sigmoidal function and illustrated with scatter plots. The correlation between weather factors and daily mean KTAS was analyzed using Pearson corre- lation coefficients. All statistical analyses were analyzed using Origin Pro (OriginLab, North- ampton, MA) and Rstudio 1.4.1717 (RStudio Inc, Boston, MA). Statistically significant difference was indicated by a p value 0.050 or less. Ethical approval The study protocol was reviewed and approved by the Institutional Review Board of Daejeon St Mary’s Hospital, The Catholic University of Korea (DC21ZIS10034). Results Demographic characteristics of the study subjects Of 1,476,652 people included in the registry with FI, 1,065,637 were older than 15 years in Korea between Jan 2014 and Dec 2018 (Fig 1). FPPs were significant higher on weekends than on weekdays (P < 0.010, S2 Fig). Hence, we excluded FI cases on weekend and holidays to pre- vent bias. Finally, 666,912 patients (418,135 in metropolitan and 248,777 in rural areas) on weekdays were analyzed in this study. The mean age of patients finally enrolled was 54.4 ± 20.4 years. There was no significant association with age distribution between the two regions, but the proportion of male patients was significantly higher in the rural areas (Table 1, P < 0.01). The distribution of FI between the two regions was balanced with no monthly difference. Of the total patients, 69, 563, 66, 755, 69, 573, and 69,783 patients suffered FIs on the new moon, 1st quarter, full moon, and 3rd quarter days, respectively. A notably higher number of FIs occurred on full moon days (Table 1, P < 0.010) and a significantly higher proportion of patients were brought to ED in an ambulance in the rural areas (Table 1, p = 0.017). In the rural areas of this study, the proportion of severe patients with KTAS scores of 1 to 2 was significantly lower, and the proportion of patients with KTAS scores from 3 to 5 was higher (Table 1, Cramer’s V = 0.076). The proportion of mentally alert patients was higher (V = 0.022) than in the metropolitan areas. The systolic and diastolic blood pressure and pulse rate per minute of FI patients in the rural areas were significantly higher in rural areas than in metropolitan areas. Relationship of FI incidence with weather and moon phase Pooled associations of the daily FPP with weather and moon phase are presented in Table 2, Fig 2 (metropolitan area), and Table 3, Fig 3 (rural area). Among weather factors, precipita- tion, minimum temperature, and wind speed showed a significant association with FI in met- ropolitan areas (Table 2). FIs occurred frequently on days with lower precipitation, lower minimum temperature, and low-wind days in metropolitan areas (Fig 2). In rural areas, FIs have been shown to increase significantly on days with lower precipitation levels, higher mini- mum temperatures, higher wind speed and longer sunshine duration (Table 3, Fig 3). The distribution of FI patients was compared according to moon phase. The frequency of FIs was higher on full moon days than on new moon days in rural areas (Table 1, p < 0.010). Full moon was a significant predictor of FIs in univariate analysis in rural area (Table 3, PLOS ONE | https://doi.org/10.1371/journal.pone.0261071 December 31, 2021 4 / 14 PLOS ONE Effects of weather and moon phases on EM use Fig 1. Schematic diagram showing the selection of study population for this study. https://doi.org/10.1371/journal.pone.0261071.g001 p = 0.048) but not significant in multivariate analysis. Based on the interaction analysis, the new moon phase showed a significant interaction with precipitation and wind speed in rural areas. However, there was no significant difference in the incidence of FI as similar FIs occurred on all days in metropolitan areas (Tables 1 and 2). Correlation of weather factors with FI severity The severity of FI was measured using KTAS assessed upon ED arrival. KTAS 1 refers to a state warranting emergency resuscitation, and KTAs 5 indicates absence of emergency. Injury severity (daily mean KTAS for each province) was significantly correlated with minimum tem- perature and wind speed and thus the injury severity was increased on cold windless days in both areas (Table 4). Additionally, in rural areas, the daily mean KTAS was significantly corre- lated with cloud cover. Precipitation, sunshine, and fog duration were not associated with the PLOS ONE | https://doi.org/10.1371/journal.pone.0261071 December 31, 2021 5 / 14 PLOS ONE Table 1. Demographic features of subjects who visited ED after a fall injury. Effects of weather and moon phases on EM use Variable Age Sex Month Moon phase Route 15–19 20–24 25–29 30–34 35–39 40–44 45–49 50–54 55–59 60–64 65–69 70–74 75–79 80–84 85–89 90–94 95–99 100–104 105–109 110–120 Male Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec New moon 1st quarter Full moon Last quarter other Ambulance Private car Ambulation Metropolitan (n = 418,135) N (%) Rural Cramer’s V or p value (n = 248,777) N (%) 25,423 (5.0) 28,389 (5.6) 29,192 (5.6) 29,566 (5.8) 27,869 (5.5) 30,840 (6.1) 36,172 (7.1) 43,583 (8.6) 48,404 (9.5) 38,986 (7.7) 35,154 (6.9) 37,268 (7.3) 39,169 (7.7) 31,347 (6.2) 17,844 (3.5) 6,660 (1.3) 1,393 (0.3) 183 (0.0) 24 (0.0) 8 (0.0) 7,299 (5.7) 7,864 (5.5) 7,795 (4.7) 7,988 (5.0) 7,780 (5.6) 8,706 (6.5) 10,245 (7.7) 12,272 (9.1) 13,521 (9.1) 10,742 (7.3) 9,593 (6.3) 10,214 (6.8) 10,960 (7.9) 8,844 (6.6) 5,034 (3.7) 1,893 (1.4) 409 (0.3) 59 (0.1) 13 (0.0) 3 (0.0) 258,579 (51.0) 132,101(53.1) 44,515(8.8) 38,357 (7.7) 39,581 (7.8) 40,136 (7.9) 44,224 (8.7) 38,029 (7.5) 39,968 (7.9) 41,564 (8.2) 43,425 (8.6) 46,051 (9.1) 42,183 (8.3) 49,410 (9.7) 44,461 (10.6) 41,531 (9.9) 43,107 (10.3) 44,059 (10.5) 244,713 (58.6) 439,439 (86.6) 65,574 (12.9) 2,227 (0.4) 12,288 (8.7) 10,381 (7.4) 10,448 (7.4) 10,939 (7.8) 12,420 (8.8) 10,928 (7.7) 11,315 (8.0) 12,131(8.6) 12,371 (8.8) 13,404 (9.5) 11,675 (8.3) 12,931 (9.2) 25,102 (10.1) 25,224 (10.1) 26,466 (10.6) 25,679 (10.3) 146,306 (58.8) 124,180 (87.9) 16,504 (11.7) 520 (0.3) V = 0.001 P< 0.010 V = 0.020 p <0.010 p = 0.017 (Continued ) PLOS ONE | https://doi.org/10.1371/journal.pone.0261071 December 31, 2021 6 / 14 PLOS ONE Table 1. (Continued) Variable KTAS Mental SBP DBP PR 1 2 3 4 5 Alert Verbal response Pain response Unresponsive Effects of weather and moon phases on EM use Metropolitan (n = 418,135) N (%) Rural Cramer’s V or p value (n = 248,777) N (%) 920 (0.80) 5334 (4.61) 34519 (29.83) 63769 (55.11) 11155 (9.64) 487775 (96.1) 11928 (2.4) 5038 (1.0) 2544 (0.5) 134.0 ± 26.6 79.8 ± 19.3 88.8 ± 40.4 223 (0.67) 1195 (3.59) 12620 (37.95) 15897(47.81) 3317(9.97) 137035 (97.0) 2358 (1.7) 1089 (0.8) 738 (0.5) 135.7 ± 27.1 80.7 ± 19.5 89.5 ± 19.5 V = 0.076 V = 0.022 p = 0.001 p <0.010 p <0.010 KTAS, Korean triage and acuity scale; SBP, systolic blood pressure; DBP, diastolic blood pressure; PR, pulse rate. https://doi.org/10.1371/journal.pone.0261071.t001 severity of FI in the rural or metropolitan areas. There was no significant difference in mean KTAS depending on the lunar phase in the metropolitan or rural areas (p = 0.394, p = 0.457, respectively). Discussion During the study period of five years, we found that the prevalence and severity of FI were associated with multiple weather factors such as daily precipitation, minimum temperature, Table 2. Multivariate regression analysis of relationships between weather factors and moon phase with fall injuries in metropolitan areas. Univariate analysis Multivariate analysis IRR 0.99 0.98 0.78 1.07 1.03 1.00 1.10 1.12 1.25 p value 0.043 <0.001 0.003 0.269 0.020 0.677 0.567 0.532 0.724 Precipitation Minimum temperature Wind speed Cloud cover Sunshine duration Fog duration Moon phase (versus full moon) New moon 1st quarter 3rd quarter Interaction effects precipitation:minimum precipitation:wind precipitation:sunshine minimum:wind minimum:sunshine wind:sunshine IRR, incidence risk ratio; SE, standard error; CI, confidence interval. https://doi.org/10.1371/journal.pone.0261071.t002 IRR 0.93 1.27 0.80 1.01 1.00 1.01 1.01 0.99 1.02 0.99 p value 0.045 0.005 <0.001 0.516 0.001 0.755 0.414 0.001 0.023 0.131 CI 0.93–0.97 0.97–0.98 0.95–1.01 1.01–1.02 1.00–1.01 1.01–1.01 1.01–1.01 0.99–1.00 1.01–1.03 0.99–0.99 PLOS ONE | https://doi.org/10.1371/journal.pone.0261071 December 31, 2021 7 / 14 PLOS ONE Effects of weather and moon phases on EM use PLOS ONE | https://doi.org/10.1371/journal.pone.0261071 December 31, 2021 8 / 14 PLOS ONE Fig 2. Scatter plots of the number of ED visits per million people after fall injuries versus weather components in metropolitan areas of Korea. A theoretical Gaussian regression line estimated by nonlinear curve fitting with the Boltzmann sigmoidal function is shown in red. https://doi.org/10.1371/journal.pone.0261071.g002 Effects of weather and moon phases on EM use and wind speed in both metropolitan and rural areas. However, we found that the longer the sunshine duration was linked with the higher FI in rural area. Moon phases were weakly asso- ciated with FI, especially in rural areas. The FI severity was closely related to weather factors. This study enrolled the largest dataset ever collected to determine the association between weather factors and ED visits due to FI in all 11 provinces of Korea over a 5-year period. A pre- vious study conducted in a small city of 23,000 people in northern Netherlands reported that better weather conditions were associated with the incidence of all types of trauma [9]. The study location was similar to rural South Korea, where weather components including maxi- mum temperature, sunshine duration, humidity, and precipitation were associated with all kinds of injury. The present study also found that precipitation, minimum temperature, and wind speed were typically related to FI. Additionally, sunshine duration was a significant pre- dictor of FI in rural areas with high agricultural activities. These findings suggest that it is essential to consider a variety of factors such as geographic location, main industry in the region, and weather changes during the investigation of injury prevalence. Ramgopal et al. [8] investigated the association of weather factors with all EMS dispatches using longitudinal data of ambulance transport in western Pennsylvania and reported increased EMS responses with rising temperature, snowfall, and rain based on a stratified anal- ysis of seasonal variables and a day-of-the-week effect week. We found that additional factors such as wind speed, cloud cover, and sunshine duration were associated with emergency Table 3. Multivariate regression analysis of relationships between weather factors and moon phase with fall injuries in rural areas. Univariate analysis Multivariate analysis IRR p value IRR p value 0.96 1.25 1.24 2.01 1.19 0.03 0.75 0.61 0.68 <0.001 <0.001 <0.001 0.383 <0.001 0.817 0.002 0.948 0.984 Precipitation Minimum temperature Wind speed Cloud cover Sunshine duration Fog duration Moon phase (versus full moon) new moon 1st quarter 3rd quarter Interaction effects precipitation:wind precipitation:minimum precipitation:sunshine minimum:wind minimum:sunshine wind:sunshine new moon:precipitation new moon:wind new moon:sunshine IRR, incidence risk ratio; SE, standard error; CI, confidence interval. https://doi.org/10.1371/journal.pone.0261071.t003 0.98 1.20 1.18 1.20 0.78 0.60 0.68 1.01 1.00 0.98 1.01 1.00 1.08 1.01 1.02 0.99 <0.001 <0.001 <0.001 0.040 0.043 0.556 0.984 <0.001 <0.001 <0.001 <0.001 0.590 <0.001 0.027 0.028 0.055 CI 0.90–0.97 0.99–1.51 1.23–1.25 0.67–2.20 0.61–0.85 0.95–1.02 0.99–1.03 1.01–1.01 1.00–1.00 0.96–0.99 1.01–1.02 1.00–1.00 1.06–1.11 0.00–14.6 0.79–1.16 0.02–3.13 PLOS ONE | https://doi.org/10.1371/journal.pone.0261071 December 31, 2021 9 / 14 PLOS ONE Effects of weather and moon phases on EM use PLOS ONE | https://doi.org/10.1371/journal.pone.0261071 December 31, 2021 10 / 14 PLOS ONE Fig 3. Scatter plots of the number of ED visits per million people after fall injuries versus weather components in rural areas of Korea. A theoretical Gaussian regression line estimated by nonlinear curve fitting with the Boltzmann sigmoidal function is shown in red. https://doi.org/10.1371/journal.pone.0261071.g003 Effects of weather and moon phases on EM use resource use after FI. However, seasonal changes were not included as independent variables in this study because changes in minimum temperature and precipitation implicated seasonal variations in weather. We also excluded FI on weekends and holidays because of increased trauma due to enhanced outdoor leisure activity and distant travel on days that might act as a confounding variable. We expected no major challenges in the analysis of FIs on weekdays because of a sufficient number of cases using 5-year large-scale longitudinal data for at the nationwide level. Stomp et al. [9] reported that better weather conditions in rural areas were associated with the incidence of all traumas. Our analysis also found that the frequency of FIs in rural areas was increased under less precipitation, higher minimum temperature, and longer sunshine duration such as busy farming seasons. This is the first study to compare FI-related factors between developed metropolitan cities and rural areas. During the course of our study, another research paper reported the correla- tion between weather changes and FI in a small Russian city [4]. It was the only longitudinal study for FIs like this study but was limited by geographic location of the study area or by small number of subjects. The daily average of outdoor falls in the cold season was 20.2 per 100,000 people and the slippery surfaces covered with wet snow or ice and temperatures between -7.0˚C and -0.7˚C were risk factors. As mentioned above, our study results showed a distinct increase in FIs according to regional characteristics, with a lower temperature trigger- ing falls on slippery surface in metropolitan areas, and a higher temperature during increased agricultural activity in rural areas associated with increased FIs. They also reported that the FIs were increased when the 12-hour precipitation was greater than 0.4 mm; however, the present study showed that the FIs were increased under low precipitation. This difference is probably explained by the falling of snow leading to slippery surfaces in Russia with a high altitude, whereas in Korea located in mid-latitude weather, rain accompanied by summer storms with strong winds reduced the frequency of outdoor activities. Northern Russia is located at the highest latitude among countries in the world. As the highest temperature in summer was near zero, the study failed to reflect changing weather patterns in mid-latitude areas with four clear seasons. Good weather conditions accompanied by active agricultural activities and increased night visibility under moonlight on a full moon might be associated with FIs in rural areas. A moon phase occurs every 29.53 days and 12.37 times in a year. The four principal moon phases include: new moon, the 1st quarter, full moon, and the last quarter. Moon phases are known to drive periodic changes in nighttime illumination, geomagnetic fields, gravitational pull, and other factors associated with major meteorological and biological changes [15]. We found that Table 4. Results of Pearson’s correlation analysis between injury severity (mean KTAS) with weather factors. Precipitation Minimum temperature Mean wind speed Cloud cover Sunshine duration Fog duration Metro CC Rural p value CC p value 0.006 0.701 -0.008 0.718 CC, correlation coefficient. 0.049 0.003 0.065 0.004 https://doi.org/10.1371/journal.pone.0261071.t004 0.157 <0.001 0.108 <0.001 -0.045 0.700 0.298 0.007 -0.012 0.957 0.005 0.809 <0.001 0.995 0.030 0.175 PLOS ONE | https://doi.org/10.1371/journal.pone.0261071 December 31, 2021 11 / 14 PLOS ONE Effects of weather and moon phases on EM use FI-related ED visits on full moon days were significantly increased than on new moon days only in rural areas. Two of the four rural provinces in this study border the sea and another province is an island with active fishing activities (S1 Fig). The full moon is a time of full tide and active fishing activities due to vertical migration of fishes [16]. Therefore, active nighttime activities and fishing activities might increase FIs in rural areas. However, FIs in metropolitan areas were less affected by lunar phase due to good visibility under night light that offset the effects of lunar phases. A previous study from Japan revealed a significant increase in the risk of emergency trans- port after traffic accidents on full moon days among those aged �40 years [11]. This finding is consistent with the results of our study showing a significant increase in FIs during full moon days especially in rural areas where the elderly individuals reside under weak artificial lighting at night. A population-based double control study conducted in the United States reported that deaths from motor traffic accidents are more frequent on full moon nights [12]. The authors postulated that a full moon might be associated with speeding, long distances, and unknown routes, resulting in more frequent deaths. In our study, FIs were increased on full moon days only in rural areas with weak artificial lighting, suggesting that increased visibility and outdoor activity under moonlight on full moon days are associated with increased FIs. The analysis of interaction between lunar phases and weather factors showed that the new moon phase interacted with precipitation, wind speed, and cloud cover, which is consistent with the findings of a previous study showing an increased number of storms during new moon phases [17]. The finding suggests that the decrease in FIs in rural areas during new phases may be a result of weather changes. Thus, the effect of the lunar phase is complex with increased near-field vision due to moonlight mainly in rural areas and secondary weather changes associated with lunar phases. We found that weather factors were correlated with FI severity measured by KTAS, a uni- fied triage scoring system. KTAS is a five-level triage scale developed in Korea based on Cana- dian Triage and Acuity Scale (CTAS) and the score is a strong predictor of severity of patients with higher 30-day mortality [18]. The strength of the study is that it is a population-based analysis of longitudinal data involv- ing FIs in a mid-latitude country with four clearly distinguished seasons. A few unknown envi- ronmental factors may confound the study results. Future studies should use more complex modeling methods and evaluate the effects of moon phases and weather changes. Patients sus- taining FIs may visit the ED the next day or later instead of on the day of injuries. Morency et al. [19] reported a significant increase in outdoor falls on days 1–3 after falling temperatures or snowfall. Therefore, it might be a challenge to compare changes in weather phenomena and patients visiting the hospital on the same day. We believed that the interval between the weather change and FIs is not a hindrance because of the gradual changes in weather and FI incidence over a period of several days. We enrolled subjects regardless of indoor or outdoor injuries because exposure to slippery terrain under snow or rain can still trigger injuries indoors. Additionally, snowy and rainy days lead to behavioral changes due to thick clothes and protective gears. Our study was conducted using large-scale nationwide databases without analyzing clinical data of patients with emotional stress, alcohol use, and violence. Further, the effects of other natural events such as earthquakes leading to mass casualties were not considered. In summary, we found that the incidence of FI is related to weather factors. Emergency medical personnel should understand that FIs occur frequently during days of low precipita- tion, high temperature and low winds linked with active outdoor activity in metropolitan areas. Additional weather factors have been shown to affect FI incidence in rural areas so that increased FI rates were noticed on days of low precipitation, high temperature, low winds and PLOS ONE | https://doi.org/10.1371/journal.pone.0261071 December 31, 2021 12 / 14 PLOS ONE Effects of weather and moon phases on EM use longer sunshine duration in rural areas. Moon phases are weakly linked to FI incidence rates. FIs increased only in rural areas during the full moon days compared with new moon days. FI severity is also affected by weather factors. In both urban and rural areas, the severity of FI sig- nificantly increased on cold and windy days. Supporting information S1 Fig. Areas to be studied were selected by dividing them into A) metropolitan areas (red color) including seven metropolitan cities with a population exceeding one million and B) rural areas (blue color) consisting of four provinces without containing metropolitan cities within its perimeter. (PDF) S2 Fig. Distribution of daily fall injuries by weekday for metropolitan and rural areas. (PDF) Author Contributions Conceptualization: Kisung Kim, Sungyoup Hong. Data curation: Min Ah Yuh, Kisung Kim, Juseok Oh, Jinwoo Kim, Sungyoup Hong. Formal analysis: Kisung Kim, Seon Hee Woo, Jinwoo Kim, Sungyoup Hong. Funding acquisition: Juseok Oh, Sungyoup Hong. Methodology: Min Ah Yuh. Project administration: Seon Hee Woo. Software: Kisung Kim. Supervision: Sungyoup Hong. Validation: Kisung Kim, Juseok Oh. Visualization: Min Ah Yuh, Sungyoup Hong. Writing – original draft: Sikyoung Jeong, Sungyoup Hong. Writing – review & editing: Seon Hee Woo, Sikyoung Jeong. References 1. Organization WH, Ageing WHO, Unit LC. WHO global report on falls prevention in older age: World Health Organization; 2008. 2. Smulders E, Enkelaar L, Weerdesteyn V, Geurts A, van Schrojenstein Lantman-de Valk H. Falls in older persons with intellectual disabilities: fall rate, circumstances and consequences. J Intellect Disabil Res. 2013; 57(12):1173–82. https://doi.org/10.1111/j.1365-2788.2012.01643.x PMID: 23106830 3. Lepy E, Rantala S, Huusko A, Nieminen P, Hippi M, Rautio A. Role of Winter Weather Conditions and Slipperiness on Tourists’ Accidents in Finland. Int J Environ Res Public Health. 2016; 13(8):822. https:// doi.org/10.3390/ijerph13080822 PMID: 27537899 4. Unguryanu TN, Grjibovski AM, Trovik TA, Ytterstad B, Kudryavtsev AV. Weather conditions and out- door fall injuries in Northwestern Russia. Int J Environ Res Public Health. 2020; 17(17):6096. https://doi. org/10.3390/ijerph17176096 PMID: 32825697 5. Sundfør HB, Sagberg F, Høye AJAA, Prevention. Inattention and distraction in fatal road crashes– Results from in-depth crash investigations in Norway. Accid Anal Prev. 2019; 125:152–7. https://doi. org/10.1016/j.aap.2019.02.004 PMID: 30763812 6. Yeung P-Y, Chau P-H, Woo J, Yim VW-T, Rainer TH. Higher incidence of falls in winter among older people in Hong Kong. J of Clin Gerontol Geriatr. 2011; 2(1):13–6. PLOS ONE | https://doi.org/10.1371/journal.pone.0261071 December 31, 2021 13 / 14 PLOS ONE Effects of weather and moon phases on EM use 7. Lin L-W, Lin H-Y, Hsu C-Y, Rau H-H, Chen P-L. Effect of weather and time on trauma events deter- mined using emergency medical service registry data. Injury. 2015; 46(9):1814–20. https://doi.org/10. 1016/j.injury.2015.02.026 PMID: 25818056 8. Ramgopal S, Dunnick J, Owusu-Ansah S, Siripong N, Salcido DD, Martin-Gill C. Weather and temporal factors associated with use of emergency medical services. Prehosp Emerg Care. 2019; 23(6):802–10. https://doi.org/10.1080/10903127.2019.1593563 PMID: 30874455 9. Stomp W, Fidler V, ten Duis HJ, Nijsten MW. Relation of the weather and the lunar cycle with the inci- dence of trauma in the Groningen region over a 36-year period. J Trauma. 2009; 67(5):1103–8. https:// doi.org/10.1097/TA.0b013e3181986941 PMID: 19901675 10. Gevitz K, Madera R, Newbern C, Lojo J, Johnson CC. Risk of fall-related injury due to adverse weather events, Philadelphia, Pennsylvania, 2006–2011. Public Health Rep. 2017; 132(1_suppl):53S–8S. https://doi.org/10.1177/0033354917706968 PMID: 28692393 11. Onozuka D, Nishimura K, Hagihara AJSotte. Full moon and traffic accident-related emergency ambu- lance transport: A nationwide case-crossover study. Sci Total Environ. 2018; 644:801–5. https://doi. org/10.1016/j.scitotenv.2018.07.053 PMID: 29990928 12. Redelmeier DA, Shafir E. The full moon and motorcycle related mortality: population based double con- trol study. Br Med J. 2017; 359:j5367. https://doi.org/10.1136/bmj.j5367 PMID: 29229755 13. Park J, Lim T. Korean triage and acuity scale (KTAS). J Korean Soc Emerg Med. 2017; 28(6):547–51. 14. McHugh ML. The chi-square test of independence. Biochem Med. 2013; 23(2):143–9. https://doi.org/ 10.11613/bm.2013.018 PMID: 23894860 15. Chakraborty U. Effects of different phases of the lunar month on living organisms. Biol Rhythm Res. 2020; 51(2):254–82. 16. Das D, Pal S, Bhaumik U, Paria T, Mazumdar D, Pal SJIJoF, et al. The optimum fishing day is based on moon. Int J Fish Aquat Stud. 2015; 2(4):304–9. 17. Pickering WHJPA. Relation of the Moon to the Weather. Pop Astron. 1903; 11:327–8. 18. Lim YD, Lee DH, Lee BK, Cho YS, Choi G. Validity of the Korean Triage and Acuity Scale for predicting 30-day mortality due to severe trauma: a retrospective single-center study. Eur J Trauma Emerg Surg. 2018; 46(4):1–7. https://doi.org/10.1007/s00068-018-1048-y PMID: 30456416 19. Morency P, Voyer C, Burrows S, Goudreau S. Outdoor falls in an urban context: winter weather impacts and geographical variations. Can J Public Health. 2012; 103(3):218–22. https://doi.org/10.1007/ BF03403816 PMID: 22905642 PLOS ONE | https://doi.org/10.1371/journal.pone.0261071 December 31, 2021 14 / 14 PLOS ONE
null
10.1186_s12866-023-02901-1.pdf
Data Availability Raw sequences from this study are available and were deposited in the European Nucleotide Archive (ENA) with bio project accession PRJEB56537 in the ENA bio project database: https://www.ebi.ac.uk/ena/browser/view/ PRJEB56537.
Data Availability Raw sequences from this study are available and were deposited in the European Nucleotide Archive (ENA) with bio project accession PRJEB56537 in the ENA bio project database: https://www.ebi.ac.uk/ena/browser/view/ PRJEB56537 .
Akinyemi et al. BMC Microbiology (2023) 23:164 https://doi.org/10.1186/s12866-023-02901-1 BMC Microbiology Whole genome sequencing of Salmonella enterica serovars isolated from humans, animals, and the environment in Lagos, Nigeria Kabiru Olusegun Akinyemi1*, Christopher Oladimeji Fakorede1, Jörg Linde2, Ulrich Methner2, Gamal Wareth2,3,4, Herbert Tomaso2 and Heinrich Neubauer2 Abstract Background Salmonella infections remain an important public health issue worldwide. Some serovars of non- typhoidal Salmonella (NTS) have been associated with bloodstream infections and gastroenteritis, especially in children in Sub-Saharan Africa with circulating S. enterica serovars with drug resistance and virulence genes. This study identified and verified the clonal relationship of Nigerian NTS strains isolated from humans, animals, and the environment. Methods In total, 2,522 samples were collected from patients, animals (cattle and poultry), and environmental sources between December 2017 and May 2019. The samples were subjected to a standard microbiological investigation. All the isolates were identified using Microbact 24E, and MALDI-TOF MS. The isolates were serotyped using the Kauffmann-White scheme. Antibiotic susceptibility testing was conducted using the disc diffusion method and the Vitek 2 compact system. Virulence and antimicrobial resistance genes, sequence type, and cluster analysis were investigated using WGS data. Results Forty-eight (48) NTS isolates (1.9%) were obtained. The prevalence of NTS from clinical sources was 0.9%, while 4% was recorded for animal sources. The serovars identified were S. Cotham (n = 17), S. Give (n = 16), S. Mokola (n = 6), S. Abony (n = 4), S. Typhimurium (n = 4), and S. Senftenberg (n = 1). All 48 Salmonella isolates carried intrinsic and acquired resistant genes such as aac.6…Iaa, mdf(A), qnrB, qnrB19 genes and golT, golS, pcoA, and silP, mediated by plasmid Col440I_1, incFIB.B and incFII. Between 100 and 118 virulence gene markers distributed across several Salmonella pathogenicity islands (SPIs), clusters, prophages, and plasmid operons were found in each isolate. WGS revealed that strains of each Salmonella serovar could be assigned to a single 7-gene MLST cluster, and strains within the clusters were identical strains and closely related as defined by the 0 and 10 cgSNPs and likely shared a common ancestor. The dominant sequence types were S. Give ST516 and S. Cotham ST617. *Correspondence: Kabiru Olusegun Akinyemi kabiru.akinyemi@lasu.edu.ng Full list of author information is available at the end of the article © The Author(s) 2023. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data. RESEARCHOpen Access Page 2 of 17 Conclusion We found identical Salmonella sequence types in human, animal, and environmental samples in the same locality, which demonstrates the great potential of the applied tools to trace back outbreak strains. Strategies to control and prevent the spread of NTS in the context of one’s health are essential to prevent possible outbreaks. Keywords Salmonella, WGS, MLST, Virulence gene, Resistant gene, Serotyping, Nigeria Background Salmonellosis in humans is caused by Gram-negative zoonotic bacteria of the species Salmonella enterica and Salmonella bongori and remains an important public health problem worldwide. Salmonellosis accounts for 93.8  million cases of gastroenteritis, with an estimated 155,000 deaths recorded globally every single year [1]. Most cases of salmonellosis are acquired from contami- nated food such as dairy and poultry products [2]. Of the more than 2,600 different serovars of S. enterica, only a few non-typhoidal serovars (NTS) are respon- sible for most human infections [2]. Salmonella Typhi and S. Paratyphi, which are human-restricted, cause sys- temic illness, i.e., typhoid or paratyphoid fever [3]. NTS serovars are diverse in their host range and vary in their pathogenicity [4]. For severe cases of salmonellosis like sepsis or SIRS (Systemic Inflammatory Response Syn- drome), antibiotic treatment is indicated. Unfortunately, multidrug resistance (MDR) with resistance to ampicil- lin, chloramphenicol, and trimethoprim-sulfamethox- azole has become a growing concern in NTS and some invasive non-typhoidal Salmonella (iNTS) infections [5]. Fluoroquinolone resistance in isolates from African countries has increased in recent decades, and resistance to β-lactam antibiotics, particularly third- and fourth generation cephalosporins, was found in Nigeria [6]. The development of NTS isolates resistant to extended spectrum cephalosporins such as ceftriaxone represents another substantial public health issue [7]. In Africa, NTS strains appear to be different from those that cause diarrheal disease in industrialized countries and cause invasive disease with bacteraemia more often in children, with 4100 deaths per year [8, 9]. The role of animal reser- voirs and human-to-human transmission of iNTS strains is unclear [10, 11]. The control of zoonotic diseases in Nigeria is difficult due to the diversity of reservoirs and lack of surveillance. Zoonotic pathogens can contami- nate the environment, spill over into the food chain, and appear at any time during processing and post-process- ing procedures. 29% of the national burden of human dis- ease has been linked to the environment in Nigeria, while the remaining 71% has been traced to diarrheal diseases, malaria, respiratory infections, etc. [11]. A high preva- lence of Salmonella in commercial poultry farms (43.6%) was reported [12]. Products from cattle and poultry have been identified as major sources of human Salmonella infections in Nigeria [13]. Furthermore, poultry and poul- try products are the major sources of protein for humans. The domestic industry (Poultry Industry) rose from an estimated 350,000 metric tons (MT) of eggs and 200,000 MT of poultry meat produced in 2003 to an estimated production of 650,000 MT of eggs and 290,000 MT of poultry meat in 2013 [14, 15]. The industry is estimated to be worth $600 million, i.e., approximately 165 million birds in 2013. By 2015, these numbers had risen to about 180 million birds, with the bulk of the poultry production coming from backyard poultry farming. This increase is attributed to the ban on the import of chicken (excluding day-old chicks) in 2003 [14, 15]. A recent survey (2008– 2015) on Salmonella bacteraemia in children in central and northwest Nigeria revealed that 20.7–23.6% of the Salmonella bacteraemia cases were due to non-typhoidal Salmonella, with up to 45% and 39% of the isolates being Salmonella Typhimurium and S. Enteritidis, respectively [16]. Bacterial resistance to antibiotics can be attributed to irrational use due to advertising strategies, application without previous susceptibility testing, or consumption of food produced from previously treated animals [17]. Antimicrobial resistance due to the acquisition of resis- tance gene clusters either horizontally or vertically is on the rise [18]. These gene clusters have been reported to have an epidemiological link with community-associated infections in many countries, including Nigeria [18–20]. The benefit of the recent explosion of genomic sequence information for Salmonella to study drug resistance in Salmonella serotypes cannot be overemphasized. In outbreak investigations and epidemiological surveil- lance, many laboratories, including reference labora- tories, have employed serotyping and phage typing for decades [21]. But due to the polyphyletic nature of Sal- monella serovars, evolutionary groupings may fail if phe- notypic methods such as serotyping are used alone [21]. Thus, in the last two decades, pathogen characteriza- tion has already shifted to genomic analysis techniques, e.g., multi-locus sequence typing (MLST) and the vari- able number of tandem repeats (VNTR) [22, 23]. Since whole-genome sequencing (WGS) has become widely available and affordable as a tool for genotyping bacteria, it is replacing older genomic techniques [24]. The Salmo- nella Typhi genome was first reported in 2001, and sev- eral thousand Salmonella strains of various serovars have been sequenced [7, 24]. The challenge of discriminating highly related lineages of bacteria was resolved by analy- sis of the genome sequence using WGS and bioinformat- ics pipelines [24, 25]. In Nigeria, there are currently few Akinyemi et al. BMC Microbiology (2023) 23:164 Page 3 of 17 studies on the molecular detection of virulence and anti- microbial resistance genes [26, 27]. The aim of this study is to investigate virulence and antimicrobial resistance genes among the Salmonella enterica serovars and to identify potential clonal relation- ships between strains from different sources using whole genome sequencing. Materials and methods Ethical approvals Ethical approval from the ethics committee of the follow- ing institutions was obtained before patients’ enrolment: The Human Research and Ethics Committee of the Lagos State University Teaching Hospital with reference num- ber LREC/06/10/1012 and the Lagos State Health Service Commission with reference number LSHSC/2222/VOL. VC/352. Preliminary investigation A total of 2,522 samples were collected between Decem- ber 2017 and January 2019, comprising 2,002 human samples (blood 1,042 and stool 960) from hospitalized and outpatients with clinical diagnoses of febrile illness and diarrheal disease, 150 samples of cattle dung (animal from the point of slaughter) and 270 samples of poul- try feces (birds ready for sale), and 100 hospital effluent (wastewater) samples. Samples were screened for Sal- monella growth using standard protocols [28, 29] at the Department of Microbiology, Lagos State University, Nigeria. Bacteria identification A total of 51 presumptive Salmonella isolates, compris- ing 13 strains from in-patients of the general hospitals (Alimosho General Hospital and Massy Street Children Hospital), 9 strains from out-patients of the Lagos State University Teaching Hospital, 11 and 6 strains from dung and faecal samples of cattle and poultry respectively, and 12 strains from wastewater of Gbagada General Hospital, were initially identified using a MICROBACT 24E iden- tification system (Oxoid Ltd, Basingstoke, UK). The iso- lates were stored in cryotubes with Nutrient-Agar (Oxoid Ltd., Basingstoke, UK) at room temperature and sent to the Institute of Bacterial Infections and Zoonoses (IBIZ) of the Friedrich-Loeffler-Institut (FLI) in Jena, Germany. These isolates were further analyzed following DIN EN ISO 6579-1:2017-07. Briefly, they were re-suspended in 3 mL of buffered peptone water for 6–18 ± 2  h (Oxoid Ltd., Basingstoke, UK) at 37 °C. Enrichment in a selective liquid medium was done in Rappaport-Vassiliadis (RVS) Soya Peptone Broth (RVS Broth) (OMNILAB-Laborzen- trum GmbH, Bremen, Germany) for 24 ± 3  h at 41.5  °C. For selective media cultivation, two methods according to DIN EN ISO 6579-1:2017-07 were applied. Animal isolates were cultured on Xylose-Lysin-Deoxycholate Agar (XLD) (Oxoid Ltd., Basingstoke, UK) and Rambach Agar (Merck KGaA, Darmstadt, Germany) at 37  °C for 24 ± 3  h. Human and sewage isolates were cultivated on Rambach Agar and Bismuth Sulfite Agar (Becton Dick- inson, Franklin Lakes, USA) at 37 °C for 24 ± 3 h for the early screening and detection of typhoidal and paraty- phoid Salmonella. For species confirmation of the animal isolates, an Ultraflex II MALDI-TOF MS instrument (Bruker Dal- tonik GmbH, Bremen, Germany) was used as described by [30]. After positive Salmonella confirmation, serologi- cal testing was done according to ISO/TR 6579-3:2014 for serovar identification. The early identification of typhoidal and paratyphoi- dal Salmonella serovars for the human and sewage iso- lates was done with serological tests according to ISO/ TR 6579-3:2014. The species confirmation for Salmonella isolates was done with anti-Salmonella A-67 + Vi omni- valent (Sifin Diagnostics GmbH, Berlin, Germany). Then, the O groups for S. Typhi (O: 9), S. Paratyphi A (O: 2), S. Paratyphi B (O: 4), and S. Paratyphi C (O: 7) were tested. A BSL-3 laboratory was available to further characterize isolates that would be found positive for anti-Salmonella O: 9 (Sifin Diagnostics GmbH, Berlin, Germany) and therefore suspicious to be S. Typhi. The non-suspicious isolates were also identified using an Ultraflex II MALDI- TOF MS instrument as Salmonella at the genus level (Bruker Daltonik GmbH, Bremen, Germany). Analysis was carried out with the Biotyper 3.1 software (Bruker Daltonik GmbH). MALDI-TOF MS Isolates were identified using MALDI-TOF MS [30]. Briefly, bacteria from overnight cultures were suspended in 300  µl of bi-distilled water and mixed with 900  µl of 96% ethanol (Carl Roth GmbH, Karlsruhe, Germany) for precipitation. After centrifugation for 5  min at 10,000 x g, the supernatant was removed, and the pellet was re- suspended in 50 µl of 70% (vol/vol) formic acid (Sigma- Aldrich Chemie GmbH, Steinheim, Germany). Fifty microliters of acetonitrile (Carl Roth GmbH) were added, mixed, and centrifuged for 5 min at 10,000 x g. One and a half microliters of the supernatant were transferred onto an MTP 384 Target Plate Polished Steel TF (Bruker Daltonik GmbH, Bremen, Germany). After air-drying, the material was overlaid with 2  µl of a saturated solu- tion of -cyano-4-hydroxycinnamic acid (Sigma-Aldrich Chemie GmbH) in a mix of 50% acetonitrile and 2.5% trifluoroacetic acid (Sigma-Aldrich Chemie GmbH). After air-drying, spectra were acquired with an Ultraflex II instrument (Bruker Daltonik GmbH). The instrument was calibrated using the IVD Bacterial Test Standard (Bruker Daltonik GmbH). Analysis was carried out with Akinyemi et al. BMC Microbiology (2023) 23:164 Page 4 of 17 the Biotyper 3.1 software (Bruker Daltonik GmbH). The following interpretation of results was performed accord- ing to the manufacturer’s recommendation: A score of ≥ 2.3 represented reliable species-level identification; a score of 2.0–2.29 represented probable species level iden- tification; a score of 1.7–1.9 represented probable genus- level identification; and a score ≤ 1.7 was considered an unreliable identification [31]. Pure cultures were stored appropriately in cryo-tubes at -80  °C (Mast Diagnostica GmbH, Reinfeld, Germany). Serotyping using the traditional White-Kaufman Le-Minor serotyping of 48Salmonellaisolates All Salmonella strains were serotyped using poly- and monovalent anti-O as well as anti-H sera (SIFIN, Ger- many) according to the Kauffmann-White scheme [32]. Antibiotic resistance testing The isolates were re-cultivated on Columbia agar plates with 5% sheep blood for 24 h at 37 °C for antibiotic sus- ceptibility testing with the Vitek 2 Compact system (Bio- Mérieux, Marcy-etoile, France) according to EUCAST guidelines [33] for 24 different antibiotics on two cards (AST-N195 and AST-N248): ampicillin, amoxicillin + cla- acid, piperacillin, piperacillin-tazobactam, vulanic cefalexin, cefuroxime, cefuroxime-acetyl, cefotaxime, ceftazidime, cefepime, azithronam, ertapenem, imipe- nem, meropenem, amikacin, gentamicin, Tobramycin, ciprofloxacin, tigecycline, fosfomycin, nitrofurantoin, colistin, trimethoprim, and trimethoprim-sulfamethox- azole. Appropriate dilutions of the colonies were made according to the manufacturer’s instructions for MIC evaluation. The strains with Vitek Extended Spectrum ß-Lactamase (ESBL) were further characterized phe- notypically: ESBL (CTX-M like), AmpC (High-Level Case (AmpC)), or/and carbapenemase (Carbapenemase (+ Oder - ESBL) phenotype were confirmed using a combination disk test (CDT) according to the manufac- turer’s instructions and EUCAST guidelines. For ESBL resistance testing, MASTDISCS® Combi Extended Spec- trum ß lactamase (ESBL) Set (CPD10) (Mast Diagnostica GmbH, Reinfeld, Germany) with Ceftazidime 30 µg and Ceftazidime 30  µg + Clavulanic Acid 10  µg, Cefotaxime 30  µg, and Cefotaxime 10  µg + Clavulanic Acid 10 and cefpodoxime 30 µg and Cefpodoxime 10 µg + Clavulanic Acid 1  µg were used. For the detection of carbapen- emases, KPC, MBL, and OXA-48 MASTDISCS® were used: carbapenemases (Rosco, Taastrup, Denmark) with Meropenem 10  µg, Meropenem 10  µg + Phenylboronic acid, Meropenem 10  µg + Dipicolinic acid, Merope- nem 10  µg + Cloxacillin, and Temocillin. For the AmpC detection, the 69  C AmpC Detection Disc Set (Mast Diagnostica GmbH, Reinfeld, Germany) was used with cefpodooxime 10  µg, AmpC stimulator, an ESBL inhibi- tor, and an AmpC inhibitor. Whole genomic sequence and bioinformatics analysis Next-generation-sequencing (NGS) The QIAGEN® Genomic-tip 20/G kit (QIAGEN, Ger- many) was used to prepare genomic DNA. NGS librar- ies were prepared using the NextEra XT DNA Library Preparation Kit (Illumina Inc., USA). An Illumina MiSeq instrument (Illumina Inc., USA) was used for paired-end sequencing. Raw sequences from this study are available and were deposited in the European Nucleotide Archive (ENA) with bio project accession PRJEB56537 in the ENA bio project database: https://www.ebi.ac.uk/ena/ browser/view/PRJEB56537. Bioinformatics analysis: data analysis 2.2.0 The Linux-based bioinformatics pipeline WGSBAC v. (https://gitlab.com/FLI_Bioinfo/WGSBAC) was used to analyze raw sequencing data as previously described [34, 35]. For quality control, WGSBAC used FastQC v. 0.11.7 [36] and calculated sequencing cover- age. As a next step, the pipeline assembled sequencing reads using Shovill v. 1.0.4 [37] and accessed assembly quality using QUAST v. 5.0 [38]. To check for poten- tial contamination, Kraken v2.1.1 [39] was used to clas- sify the raw sequencing. To predict serovars based on sequencing data, WGSBAC used SISTR v. 1.0.2 [40] and SeqSero2 [41]. For the detection of genes and point mutations potentially leading to antimicrobial resistance (AMR), AMRFinderPlus (v. 3.6.10) [42] was used within WGS- BAC. ABRicate (v. 0.8.10) [43] together with the data- bases Virulence Factor Database (VFDB) [44] and PlasmidFinder [45] were used to detect virulence fac- tors and plasmids, respectively. For genotyping, WGS- BAC first performed 7-gene multi-locus sequence typing (MLST) on assembled genomes using the software mlst v. 2.16.1 [46]. High-resolution genotyping was performed using both an SNP-based approach and an allele-based approach. Snippy v. 4.3.6 to identify core-genome single nucleotide polymorphisms (cgSNPs) was utilized within WGSBAC in standard settings [47]. As a reference genome, the complete genome sequence of Salmonella enterica subsp. enterica serovar Typhimurium strain LT2 (GenBank accession GCA_000006945.2) was used. To cal- culate pairwise SNP distances, SNPs-dists (v 0.63) were applied. Hierarchical clustering was performed using the hierClust function v.5.1 of the statistical language R. A cut-off of 10 cgSNPs was used to define closely related strains and 0 cgSNPs to define identical strains. For the allele-based approach, core-genome multi- locus sequence typing (cgMLST) was performed by applying Ridom Seqsphere + v. 5.1.0 [48] with default Akinyemi et al. BMC Microbiology (2023) 23:164 Page 5 of 17 Table 1 Sample summary and Salmonella enterica positive isolates distributed across human, animal and environmental sources Human samples Blood 1042 163 10 153 Stool 960 308 9 299 Animal samples Cattle Poultry Dung 150 54 11 43 Feaces 270 72 6 66 Environmental samples Effluent 100 70 12 58 Total 2,522 667 48 619 Number of samples Number of + ve cultures Salmonellaisolates (n) Other isolates *+ve: positive bacterial culture Fig. 1 Distribution of Salmonella enterica serovars according to sample source from Lagos, Nigeria settings together with the specific core-genome scheme (cgMLST v2) for Salmonella enterica developed by EnteroBase [49]. Again, a cut-off of 10 alleles was used to define clusters. Results Identification, distribution, and serotyping of NigerianSalmonella enterica isolates Out of 2,522 samples analyzed in this study, 667 sam- ples showed bacterial growth on selective media. From these positive bacterial culture samples, 51 presumptive Salmonella isolates were identified by Microbact 24E (Oxoid, England). Of the 51 presumptive Salmonella iso- lates, 48 isolates were identified as Salmonella enterica using MALDI-TOF MS (Table  1), and the remain- ing three isolates were Citrobacter spp. The serotyping results of the 48 Salmonella isolates revealed six different serovars from human, animal, and environmental sources (Fig.  1). The serovars with their predicted antigenic profiles included S. Cotham (n = 17), S. Give (n = 16), S. Mokola (n = 6), S. Abony (n = 4), S. Typhimurium (n = 4) and S. Senftenberg (n = 1) and are presented in Table S1. The distribution of the serovars is as follows: S. Cotham (2 from human, 10 from animal, and 5 from sewage Akinyemi et al. BMC Microbiology (2023) 23:164 samples), S. Give (12 from human, 2 from animal, and 2 from sewage samples), S. Mokola (5 from animal samples and 1 from a sewage sample), S. Abony (1 from a clinical sample and 3 from sewage samples), S. Typhimurium (3 from human samples and 1 from a sewage sample), and S. Senftenberg ( 1 from a human sample), In this study, seven strains of S. Give, one strain of S. Typhimurium, and one strain of S. Senftenberg were iNTS. Whole genome sequencing data Genome sequencing of the 48 Salmonella enterica isolates analyzed in this study yielded an average of 1,513,603 reads per isolate (range 465,448-3,046,296; Table S1). The mean coverage of the 48 Salmonella iso- lates was 52-fold (ranging from 23-fold to 148-fold in Table S2). To check for putative contamination, the software Kraken2 was used, which classified each read (or contig). On the species level, the top hit for all 48 isolates was always “Salmonella”. On average, 96% of the reads were classified as “Salmonella”. Table S3. The N50 of the 48 assembled genomes ranges from 153,458 bp to 708,946 bp. (Table S4) Page 6 of 17 Resistance profiling and AMR genes in 48 Nigerian Salmonella enterica serovar isolates from different sources All forty-eight Salmonella enterica isolates were 100% susceptible to ampicillin, piperacillin-tazobactam, cefo- taxime, ceftazidime, cefepime, azithronam, ertapenem, imipenem, meropenem, tigecycline, fosfomycin, colis- tin, trimethoprim, and trimethoprim-sulfamethoxazole. Meanwhile, 16 (33.33%) isolates were resistant to moxi- floxacin and 14 (29.2%) isolates showed intermediate resistance to Ciprofloxacin. There was no phenotypic expression of extended-spectrum β-lactamase (ESβL), inducible AmpC, Metallo- β-Lactamase, or blaOXA-48 among the isolates (Fig. 2). All isolates contained intrinsic chromosomal encoded aminoglycoside acetyltransferase aac(6)-Iaa resistance genes as well as mdf(A) genes coding for a multidrug efflux pump. Acquired quinolone resistance gene qnrB was detected in four strains of Salmonella Give 8.3% (4/48), while 12 strains of Salmonella Give harbored qnrB19. All Salmonella serovars harbor efflux mecha- nism genes sinH, mdsB, and mdsA and genes golT and golS coding for resistance to copper/gold and gold, respectively. However, genes pcoE, pcoS, pocR, pcoD, pcoC, pcoB, pcoA, silP, silA, silB, silF, silC, silR, silS, and silE coding for resistance to copper, silver, and copper/ silver were found in only one Salmonella Senftenberg Fig. 2 Antibiogram of Salmonella enterica serovars isolated from different sources in Nigeria in accordance with EUCAST Expert Rules x 3.2 Akinyemi et al. BMC Microbiology (2023) 23:164 strain. Plasmid replicons were detected in 20 of the 48 isolates. All 16 Salmonella Give strains harbored plasmid Col440I_1, while plasmid incFIB.B/incFII.S was detected in four Salmonella Typhimurium strains (Table 2). An average of 100 to 118 virulence gene markers dis- tributed across several Salmonella pathogenicity islands (SPIs), clusters, and plasmid operons were detected in all 48 Salmonella isolates. A total of 95 virulence genes were common to all Salmonella isolates. The virulence genes ctdB, fae, faeD, and faeE were found only in Sal- monella Give and S. Cotham strains. IpfA, IpfB, IpfC, IpfD, IpfE, pipB2, and sopD2 genes were detected in Sal- monella Senftenberg, S. Abony and S. Typhimurium. The virulence genes spvB, spvC, spvR, ssel, srfH, sseK2, and sspH2 were found only in S. Typhimurium strains, while the entE gene was detected in only S. Mokola strains (Table 3). The result of the raw sequence data for the serovar prediction using SISTR and SeqSero indi- cated a pass QC status for 42 isolates. Only six isolates of serovar Mokola had a warning QC status (Table S1), as only 186 cgMLST330 loci matched the number of cgMLST330 loci found (n = 330). The reason for this warning might be the relatively low number of publicly available genome sequence of serovar Mokola which have beend used for training these tools. In fact, Enterobase, the largest collection of Salmonella genome sequences with 403.715 strains (accessed May 2023), contained only three Mokola strains. The multi-locus sequence typing (cMLST), allelic profiles, and sequence types (ST) of 48 Salmonella enterica isolates were assigned by comparing the sequences with those in the MLST profile database. The 48 Salmonella isolates yielded six unique serovars, and the 7-gene MLST identified one sequence type (ST) for each serovar. Sequence type ST617 is shared by all 17 Salmonella Cotham strains while ST516 is shared by all 16 Salmonella Give strains. Similar findings were made with the ST19 sequence type for the four Salmonella Typhimurium strains and the ST1483 sequence type for the four Salmonella Abony strains. The only strain of Salmonella enterica subsp. enterica serovar Senftenberg belongs to the ST14 sequence type. In the MLST profile database, no sequence type for Salmonella Mokola was found (Table S5). The sequence core-genome multi-locus typing (cMLST)clustering of 48 Salmonella isolates based on the source of the isolate, type of samples, and clinical diag- nosis revealed five distinct clusters of 46 Salmonella iso- lates. Two of the Salmonella isolates (S. Senftenberg and one S. Typhimurium) were not assigned to any cluster. Cluster 1 with ST617 consisted of 17 Salmonella strains distributed among human isolates (2 strains), animal iso- lates (7 strains from cattle and 3 strains from poultry), and five environmental isolates. Cluster 2 with ST516 included 16 Salmonella strains, i.e., 12 human isolates, 2 Page 7 of 17 animal isolates (one cattle and one poultry), and 2 envi- ronmental isolates. Cluster 3 with unknown ST consisted of 6 strains, of which 5 strains were from animals (three strains from poultry and two from cattle) and one strain from an environmental source. Cluster 4 with ST1483 comprised 4 strains, of which 3 were from environmen- tal samples and one was from a human patient. Cluster 5 with ST19 was made up of 3 strains, of which 2 were from human patients and one from animal sources (Table S6 and Fig. 3. The distribution of the isolates within the local government areas is shown in Fig. 4. Discussion The intensity with which non-typhoidal Salmonella (NTS), responsible for foodborne diseases, acquires anti- microbial resistance genes over the years is worrisome and of public health concern [50]. Using WGS and bio- informatics tools, this study examined clonal relation- ships among NTS isolates from humans, animals, and the environment, their virulence potential, and the presence of antimicrobial resistance genes that may pose a pub- lic health threat if spread to other bacterial agents. Two thousand five hundred twenty-two samples yielded 48 different Salmonella. The infection was present in 0.9% (19/2002) of human samples, 4% (17/420) of animal sam- ples, and 12% (12/100) of environmental sample (sewage and wastewater). The 4% prevalence rate from animal sources (poultry: 7.3% (11/150) and cattle: 2.2% (6/270)) reported in this study is lower than that reported by Jibril et al. [51] with 14.3% for animal faecal samples. A similar high preva- lence was documented in other African countries, such as Ethiopia with 14.9% [52] and Ghana with 44% [53]. This difference could be caused by sample size, as a much lower number of samples was collected in this study. In Denmark, an annual report on Salmonella enterica zoo- nosis revealed a prevalence of 0–1.8% [54]. The low prev- alence recorded in Denmark has been associated with effective surveillance and control programs, a situation that is not well-footed in Nigeria. The dominant serotype in this study was S. Cotham (35.4%), followed by S. Give (33.3%), S. Mokola (12.5%), S. Typhimurium (8.3%), S. Abony (8.3%), and S. Senftenberg (2.1%). Except for S. Typhimurium, the serovars found in this study had not been characterized in Nigeria before, either from animals (poultry and cattle), humans, or the environment. Besides, little is known about their poten- tial to cause human disease. Interestingly, serovars S. Give and S. Cotham were cultured from patients, animals, and wastewater from abattoirs and the hospital environ- ment. Thus, both serovars (S. Give and S. Cotham) may be emerging Salmonella serovars in Nigeria. Although some of the serovars isolated in this study are not com- monly associated with human salmonellosis, they may be Akinyemi et al. BMC Microbiology (2023) 23:164 Table 2 Summary of the intrinsic and acquired resistance genes, transporter genes and plasmid replicons in 48 Salmonella enterica strains from Nigeria Strain ID Acquired genes/Transporters Intrinsic genes Serotype Source 19CS0255 Senftenberg Animal (cattle) aac.6…Iaa/mdf(A) 19CS0257 19CS0290 19CS0294 19CS0295 19CS0250 19CS0263 19CS0267 19CS0269 19CS0271 19CS0272 19CS0274 19CS0275 19CS0277 19CS0278 19CS0279 19CS0283 19CS0284 19CS0285 19CS0288 19CS0291 19CS0292 19CS0245 19CS0246 19CS0247 19CS0248 19CS0249 19CS0251 19CS0259 19CS0260 19CS0261 19CS0262 19CS0264 19CS0266 19CS0268 19CS0280 19CS0289 19CS0293 19CS0270 19CS0273 19CS0276 19CS0281 19CS0282 19CS0287 19CS0253 19CS0256 19CS0258 19CS0286 Abony Abony Abony Abony Cotham Cotham Cotham Cotham Cotham Cotham Cotham Cotham Cotham Cotham Cotham Cotham Cotham Cotham Cotham Cotham Cotham Give Give Give Give Give Give Give Give Give Give Give Give Give Give Give Give Mokola Mokola Mokola Mokola Mokola Mokola Typhimurium Typhimurium Typhimurium Typhimurium aac.6…Iaa/mdf(A) Sewage/wastewater aac.6…Iaa/mdf(A) Sewage/wastewater aac.6…Iaa/mdf(A) Human aac.6…Iaa/mdf(A) Human aac.6…Iaa/mdf(A) Animal (poultry) aac.6…Iaa/mdf(A) Sewage/wastewater aac.6…Iaa/mdf(A) Human aac.6…Iaa/mdf(A) Human aac.6…Iaa/mdf(A) Human aac.6…Iaa/mdf(A) Animal (cattle) aac.6…Iaa/mdf(A) Sewage/wastewater aac.6…Iaa/mdf(A) Human aac.6…Iaa/mdf(A) Sewage/wastewater aac.6…Iaa/mdf(A) Human aac.6…Iaa/mdf(A) Human aac.6…Iaa/mdf(A) Animal (poultry) aac.6…Iaa/mdf(A) Sewage/wastewater aac.6…Iaa/mdf(A) Animal (poultry) aac.6…Iaa/mdf(A) Sewage/wastewater aac.6…Iaa/mdf(A) Human Sewage/wastewater aac.6…Iaa/mdf(A) anderes Resistogramm aac.6…Iaa/mdf(A) aac.6…Iaa/mdf(A) Animal (cattle) aac.6…Iaa/mdf(A) Animal (poultry) aac.6…Iaa/mdf(A) Sewage/wastewater aac.6…Iaa/mdf(A) Sewage/wastewater aac.6…Iaa/mdf(A) Animal (poultry) aac.6…Iaa/mdf(A) Animal (cattle) aac.6…Iaa/mdf(A) Animal (cattle) aac.6…Iaa/mdf(A) Animal (cattle) aac.6…Iaa/mdf(A) Animal (cattle) aac.6…Iaa/mdf(A) Human aac.6…Iaa/mdf(A) Human aac.6…Iaa/mdf(A) Human aac.6…Iaa/mdf(A) Human aac.6…Iaa/mdf(A) Human aac.6…Iaa/mdf(A) Human aac.6…Iaa/mdf(A) Human aac.6…Iaa/mdf(A) Animal (poultry) aac.6…Iaa/mdf(A) Human aac.6…Iaa/mdf(A) Animal (cattle) aac.6…Iaa/mdf(A) Animal (cattle) aac.6…Iaa/mdf(A) Sewage/wastewater aac.6…Iaa/mdf(A) Sewage/wastewater aac.6…Iaa/mdf(A) Animal (cattle) aac.6…Iaa/mdf(A) Human aac.6…Iaa/mdf(A) Animal (poultry) sinH, mdsB. mdsA, golT, golS, pcoE,pcoS, pocR, pcoD, pcoC, pcoB, pcoA,silP,silA,silB, silF,silC,silR,silS,silE sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS qnrB, sinH, mdsB, mdsA, golT, golS qnrB19, sinH, mdsB, mdsA, golT, golS qnrB19, sinH, mdsB, mdsA, golT, golS qnrB19, sinH, mdsB, mdsA, golT, golS qnrB19, sinH, mdsB, mdsA, golT, golS qnrB, sinH, mdsB, mdsA, golT, golS qnrB19, sinH, mdsB, mdsA, golT, golS qnrB19, sinH, mdsB, mdsA, golT, golS qnrB, sinH, mdsB, mdsA, golT, golS qnrB19, sinH, mdsB, mdsA, golT, golS qnrB19, sinH, mdsB, mdsA, golT, golS qnrB19, sinH, mdsB, mdsA, golT, golS qnrB19, sinH, mdsB, mdsA, golT, golS qnrB, sinH, mdsB, mdsA, golT, golS qnrB19, sinH, mdsB, mdsA, golT, golS qnrB19, sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS sinH, mdsB, mdsA, golT, golS Page 8 of 17 Plasmid replicon - - - - - - - - - - - - - - - - - - - - - - Col440I_1 Col440I_1 Col440I_1 Col440I_1 Col440I_1 Col440I_1 Col440I_1 Col440I_1 Col440I_1 Col440I_1 Col440I_1 Col440I_1 Col440I_1 Col440I_1 Col440I_1 Col440I_1 - - - - - - incFIB.B/incFII.S incFIB.B/incFII.S incFIB.B/incFII.S incFIB.B/incFII.S Akinyemi et al. BMC Microbiology (2023) 23:164 Table 3 Distribution of Salmonella virulence genes clustered within several Salmonella pathogenicity Islands (SPIs) and plasmid operons from different source from Nigeria Page 9 of 17 Virulence loci Virulence genes Strain ID 19CS0245 19CS0246 19CS0247 19CS0248 19CS0249 19CS0250 19CS0251 19CS0253 19CS0255 19CS0256 Serovar Num of virulence gene 100 Give 100 Give 100 Give 100 Give 100 Give 100 Cotham Give 100 Typhimurium 116 SPI-1 SPI-2, SPI-3 SPI-11SPI-5, SPI-24/CS54 fimbriae operon SPI-1 SPI-11 SPI-2 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-11 SPI-2 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-11 SPI-2 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-11 SPI-2 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-11 SPI-2 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-11 SPI-2 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 Long polar fimbriae cluster, Plasmid encoded fim- briae cluster, Plasmid virulence Operon SPI-2 SPI-3 SPI-5 SPI-12 SPI-24/CS54 fimbriae operon Senftenberg 101 SPI-1 Long polar fimbriae cluster SPI-3 SPI-5 Typhimurium 116 SPI-1 SPI-2 SPI-3 Long polar fimbriae cluster, Plasmid encoded fimbriae cluster, Plasmid virulence Operon SPI-5 SPI-12 SPI-24/CS54 fimbriae operon 19CS0257 Abony 106 SPI-1 SPI-2 SPI-3 Long polar fimbriae cluster SPI-5 SPI-12 SPI-24/CS54 fimbriae operon 19CS0258 Typhimurium 118 SPI-1 SPI-2 SPI-12 SPI-24/CS54 Long polar fimbriae cluster, Plasmid encoded fimbriae SPI-5iae cluster, Plasmid viru- lence Operon fimbriae operon 19CS0259 19CS0260 19CS0261 19CS0262 19CS0263 19CS0264 19CS0266 19CS0267 19CS0268 19CS0269 19CS0270 19CS0271 19CS0272 19CS0273 19CS0274 19CS0275 19CS0276 19CS0277 19CS0278 19CS0279 19CS0280 19CS0281 Give Give Give Give Cotham Give Give Cotham Give Cotham Mokola Cotham Cotham Mokola Cotham Cotham Mokola Cotham Cotham Cotham Give Mokola 100 100 100 100 100 100 100 100 100 100 102 100 100 102 100 100 102 100 100 100 100 100 SPI-1 SPI-11 SPI-2 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-11 SPI-2 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-11 SPI-2 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-11 SPI-2 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-2 SPI-11 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-11 SPI-2 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-11 SPI-2 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-2 SPI-11 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-11 SPI-2 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-2 SPI-11 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-2 SPI-3 SPI-5 SPI-12 SPI-24/CS54 fimbiae operon SPI-1 SPI-2 SPI-11 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-2 SPI-11 SPI-3 SP1-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-2 SPI-3 SPI-5 SPI-12 SPI-24/CS54 fimbriae operon SPI-1 SPI-2SPI-11 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-2SPI-11 SPI-3 SPI-5 f SPI-24/CS54 imbriae operon SPI-1 SPI-2 SPI-3 SPI-5 SPI-12 SPI-24/CS54 fimbriae operon SPI-1 SPI-2SPI-11 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-2 SPI-11 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-2 SPI-11 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-2SPI-11 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-2 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon ctdB, fae. faeD faeE sspH1* CtdB fae. faeD faeE sspH1* CtdB fae. faeD faeE sspH1* CtdB fae. faeD faeE sspH1* CtdB fae. faeD faeE sspH1* CtdB fae. faeD faeE pipB2* ctdB fae. faeD faeE sspH1* gogB grvA ipfA ipfB ipfC ipfD ipfE pefA pefB pefC pefD pipB2 rck sodC1 sopD2 spvB spvC spvR ssel. srfH sseK2 sspH2* ipfA ipfB ipfC ipfD ipfE pipB2 sopD2* gogB grvA ipfA ipfB ipfC ipfD ipfE pefA pefB pefC pefD pipB2 rck sodC1 sopD2 spvB spvC spvR ssel. srfH sseK2 sspH2* ipfA ipfB ipfC ipfD ipfE pipB2 shdA sodC1 sopD2 sseK2 sspH2* gogB grvA ipfA ipfB ipfC ipfD ipfE pefA pefB pefC pefD pipB2 rck shdA sodC1 sopD2 spvB spvC spvR ssel. srfH sseK2 sspH2 sspH1* ctdB fae. faeD faeE sspH1* ctdB fae. faeD faeE sspH1* ctdB fae. faeD faeE sspH1* CtdB fae. faeD faeE sspH1* ctdB fae. faeD faeE pipB2* CtdB fae. faeD faeE sspH1* ctdB fae. faeD faeE sspH1* ctdB fae. faeD faeE pipB2* ctdB fae. faeD faeE sspH1* ctdB fae. faeD faeE pipB2* entE pipB2 shdA sodC1 sopD2 sseK2 sspH2* ctdB fae. faeD faeE pipB2* ctdB fae. faeD faeE pipB2* entE pipB2 shdA sodC1 sopD2 sseK2 sspH2* ctdB fae. faeD faeE pipB2* ctdB fae. faeD faeE pipB2* entE pipB2 shdA sodC1 sopD2 sseK2 sspH2* ctdB fae. faeD faeE pipB2* ctdB fae. faeD faeE pipB2* ctdB fae. faeD faeE pipB2* ctdB fae. faeD faeE sspH1* entE pipB2 shdA sodC1 sopD2* Akinyemi et al. BMC Microbiology (2023) 23:164 entE pipB2 shdA sodC1 sopD2 sseK2 sspH2* ctdB fae. faeD faeE pipB2* ctdB fae. faeD faeE pipB2* ctdB fae. faeD faeE pipB2* gogB grvA ipfA ipfB ipfC ipfD ipfE pefA pefB pefC pefD pipB2 rck sodC1 sopD2 spvB spvC spvR ssel. srfH sseK2 sspH2* entE pipB2 shdA sodC1 sopD2 sspH2* ctdB fae. faeD faeE pipB2* ctdB fae. faeD faeE sodC1 sopD2 sspH1 sspH1* ipfA ipfB ipfC ipfD ipfE pipB2 shdA sseK2 sspH2* ctdB fae. faeD faeE pipB2* ctdB fae. faeD faeE pipB2* ctdB fae. faeD faeE sspH1* ipfA ipfB ipfC ipfD ipfE pipB2 shdA sodC1 sopD2 sseK2 sspH2* ipfA ipfB ipfC ipfD ipfE pipB2 shdA sodC1 sopD2 sseK2 sspH2* gogB grvA ipfA ipfB ipfC ipfD ipfE pefA pefB pefC pefD pipB2 rck sodC1 sopD2 SshdA spvB spvC spvR ssel.srfH sseK2 sspH2* Table 3 (continued) Strain ID Serovar Page 10 of 17 Virulence loci Virulence genes 19CS0282 19CS0283 19CS0284 19CS0285 19CS0286 19CS0287 19CS0288 19CS0289 19CS0290 19CS0291 19CS0292 19CS0293 19CS0294 Num of virulence gene 102 Mokola 100 Cotham 100 Cotham Cotham 100 Typhimurium 116 Mokola Cotham Give Abony Cotham Cotham Give Abony 101 100 100 106 100 100 100 106 SPI-1 SPI-2 SPI-3 SPI-5 SPI-12 SPI-24/CS54 fimbriae operon SPI-1 SPI-2 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-2 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-2 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI- SPI-3 2 Long polar fimbriae cluster, Plasmid encoded fimbriae cluster, Plasmid virulence Operon SPI-5 SPI-12 SPI-24/CS54 fimbriae operon SPI-1 SPI-2 SPI-3 SPI-5 SPI-12 SPI-24/CS54 fimbriae operon SPI-1 SPI-2 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-2 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-2 SPI-3 SPI-24/CS54 Long polar fimbriae cluster SPI-5 fimbriae operon SPI-1 SPI-2 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-2 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-2 SPI-3 SPI-5 SPI-24/CS54 fimbriae operon SPI-1 SPI-2 SPI-3 SPI-12 SPI-24/CS54 Long polar fimbriae cluster SPI-5 fimbriae operon 19CS0295 Abony 106 SPI-1 SPI-2 SPI-3 SPI-12 SPI-24/CS54 Long polar fimbriae cluster SPI-5 fimbriae operon GCF0000069452ASM694v2 LT2 117 SPI-1 SPI-2 SPI-3 SPI-5 SPI-12 SPI-24/CS54 Long polar fimbriae cluster, Plasmid encoded fimbriae cluster, Plasmid virulence Operon fimbriae operon *Represents the following 95 virulence genes: avrA csgA csgB csgC csgD csgE csgF csgG entA entB fepC fepG fimC fimD fimF fimH fimI invA invB invC invE invF invG invH invI invJ mgtB mgtC mig.14 misL ompA orgA orgB orgC pipB prgH prgI prgJ prgK ratB sicA sicP sifA sifB sinH sipA.sspA sipB.sspB sipC.sspC sipD slrP sopA sopB. sigD sopD sopE2 spaO spaP spaQ spaR spaS spiC.ssaB sptP ssaC ssaD ssaE ssaG ssaH ssaI ssaJ ssaK ssaL ssaM ssaN ssaO ssaP ssaQ ssaR ssaS ssaT ssaU ssaV sscA sscB sseA sseB sseC sseD sseE sseF sseG sseJ sseK1 sseL steA steB steC 3° 24’ 23.2128’’ E. (Longitude) of special importance in the Nigerian setting. S. Give, for example, was reported in a multistate outbreak in Ger- many in 2004 that resulted in severe gastroenteritis and hospitalization of those infected [55].S. Give has become widespread recently in poultry in Nigeria [51] and Burkina Faso [56], and serovars S. Give and S. Cotham were reported repeatedly from animal sources in Nigeria in the past [11, 57]. This finding needs further research for a reliable risk assessment. In sub-Saharan Africa, S. Typhimurium and S. Enteritidis strains were the pre- vailing iNTS serovars associated with invasive systemic infections in children and adults [58, 59], with an esti- mated mortality rate among in-patients ranging from 4.4 to 27% for children [60] and 22 to 47% for adults [61]. In Burkina Faso, Salmonella Enteritidis-associated bactere- mia was also documented [56]. In this study, seven strains of S. Give, one each of S. Typhimurium, and S. Senften- berg strain, were iNTS. Although a previous study in Nigeria documented S. Typhimurium and S. Enteritidis- associated bacteremia [62], these are the first reported iNTS Give and Senftenberg-associated bacteremia cases in Nigeria. The emerging iNTS S. Typhimurium and S Enteritidis seem to be already prominent in Nigeria, and their potential to cause human disease and spread to other neighboring countries is not in doubt. All forty-eight Salmonella enterica isolates were 100% susceptible to piperacillin-tazobactam, cefotaxime, ceftazidime, cefepime, azithronam, ertapenem, imipe- nem, meropenem, tigecycline, fosfomycin, colistin, trim- ethoprim, and trimethoprim-sulfamethoxazole. The results also revealed that none of the Salmonella isolates produced extended-spectrum β-lactamases Akinyemi et al. BMC Microbiology (2023) 23:164 Page 11 of 17 Fig. 3 The minimum spanning tree (MST) of Salmonella strains (nodes) was used in this study. The node colour corresponds to the source of the strains (see legend). The allelic distances from cgMLST analysis are denoted at branches. Clusters of genetically similar strains were defined using a cut-off of 10 alleles and are visualized in grey gentamicin, which are known to bind the bacterial 50  S subunit of the ribosome and prevent protein synthesis. The presence of this gene has made the use of amino- glycoside ineffective in vivo especially for the treatment of invasive Salmonellosis and as such EUCAST expert rule 3.2 of 2019 reported all aminoglycoside should be reported as resistant irrespective of the result of the sus- ceptibility results. Acquired quinolone resistance genes were detected in all 16 (100%) strains of S. Give. Plasmid-mediated qui- nolone resistance (PMQR) qnrB gene was detected in 4, while the qnrB19 gene was found in 12 S. Give strains. All S. Give strains harbored plasmid Col440I. Fifteen S. Give strains (93.75%) were resistant to ciprofloxacin and 16 (100%) to moxifloxacin. These strains were also phe- notypically resistant to this antibiotic family but suscep- tible to all other tested antibiotics. This result agrees with the results of a recent report on phenotypic and geno- typic antimicrobial resistance and the presence of resis- tance genes in Salmonella isolates from poultry, where neither phenotypic expression of ESBL nor ESBL genes were present in the Salmonella strains studied [51]. This observation corresponds well with the fact that β-lactams are reported to be used rarely or not often in poultry production in Nigeria [63, 64], possibly due to their high price. However, fluoroquinolones have been reported to be used as a growth promoter and possibly contributed to the emergence of Salmonella and other bacteria resistant to ofloxacin and ciprofloxacin. Predictions by ResFinder indicated that these S. Give strains carried either qnrB19 or qnrB genes with or without the presence of point Fig. 4 Map of Nigeria showing the location of Lagos State and the distri- bution of Salmonella enterica serovars across different local government areas of Lagos State (ESβL), Metallo-β-lactamases, AmpC, or OXA-48 phenotypically. Chromosomally encoded aminoglycoside acetyltrans- ferase aac.(6)-Iaa resistance genes were present in all 48 Salmonella isolates (100%) and conferred resistance like tobramycin, amikacin, and to aminoglycosides Akinyemi et al. BMC Microbiology (2023) 23:164 Page 12 of 17 mutations in DNA gyrase and topoisomerase. It has been documented that the qnrB19 and qnrB genes encode transferable fluoroquinolone resistance mechanisms that are responsible for reduced susceptibility to quinolones [65]. Also, a point mutation in the quinolone resistance- determining region (QRDR) of the DNA gyrase-A (gyrA) and topoisomerase C (parC) genes is known to cause clinical resistance in members of Enterobacteriaceae [65]. Similar reports have been documented in Salmonella isolates from Nigeria [57, 66, 67]. The high-level detec- tion of qnr genes in S. Give may contribute not only to the stepwise development of high-level fluoroquinolone resistance but also to its spread between bacteria species [68]. The first report of a plasmid-mediated quinolone resistance (PMQR) mechanism, qnrA, was described in the late 1990s, and since then, several variants of the qnr gene have been discovered [5]. Among the 48 Salmonella isolates in this study, 20 isolates belonging to two serovars contained plasmid replicons. In total, three different plasmid replicas were detected, with Col440I being the most predominant rep- licon found in the “16” “S”. Give was predicted to carry qnrB, qnrB-19, sinH, mdsB, mdsA, golT, and golS, respec- tively. The Col440II-like plasmid detected in S. Schwar- zengrund isolates in Chile highlighted the fact that a small pPAB19-4-like plasmid plays an important role in the dissemination of qnrB19 [69]. Incompatible plasmids incFIB.B and incFII.S were found in all four strains of S. Typhimurium predicted to carry the sinH, mdsB, mdsA, golT, and golS genes. The plasmids incFIB.B and incFII.S from S. Typhimurium were aligned with the reference sequence GCF0000069452ASM694v2 (LT2). There was 100% identity with 100.0% query cover between the sequences in this study and the reference sequence. In this study, acquired golT genes coding for resistance to copper or gold and golS genes coding for resistance to gold were detected in all 48 isolates, with a percent- age of similarity ranging from 96.71 to 100% when com- pared to the GCF0000069452ASM694v2 (LT2) reference strain. Furthermore, the only iNTS S. Senftenberg strain detected in this study harbored several genes that confer resistance to heavy metals such as copper (pcoE, pcoS, pocR, pcoD, pcoC, pcoB, and pcoA), copper/silver (silA, silB, silF, silC, silR, and silS), and silver (silP and silE). The presence of these heavy metal resistance genes in this strain and its role as an invasive pathogen remain sub- jects of concern. Several virulence gene markers of Salmonella enterica distributed across several Salmonella pathogenicity islands (SPIs) have been found responsible for systemic transmission leading to severe infections [70, 71]. Viru- lence genes found in all 48 Salmonella isolates were pre- dicted by all-vfbd.xls. 100 to 118 virulence gene markers distributed across several Salmonella pathogenicity islands (SPIs), clusters, prophages, and plasmid oper- ons were found in each isolate. A total of 95 virulence genes were predicted to be common to all six Salmo- nella enterica serovars. The virulence genes ctdB, fae, faeD, and faeE were common to Salmonella Give, and S. Cotham. While ipfA, ipfB, ipfC, ipfD, ipfE, pipB2, and sopD2 were detected in S. Senftenberg, S. Abony, and S. Typhimurium. The virulence genes spvB, spvC, spvR, ssel.srfH, sseK2, and sspH2 were found only in S. Typhimurium, while the entE was unique to S. Mokola. However, rck genes that confer resistance to or protect against complement-mediated immune response were detected in all S. Typhimurium strains with 100% homol- ogy to the gene of strain GCF0000069452ASM694v2 (LT2). In Salmonella pathogenicity, the type 3 secretion sys- tem (T3SS), encoded by SPI-1 and SPI-2, contains major virulence determinants. The presence of the major viru- lence factors avrA, mgtC, sopB, ssaQ, and invA in all 48 isolates is an indication of their ability to colonize the liver of the host [50] and therefore may cause serious human disease. The ssp.H1 and steB genes coding for effectors are found in all 16 (100%) S. Give strains and one (25%) S. Typhimurium strain. These effector genes are mediators of cell invasion and modifications, which are major contributing factors to intracellular growth [72]. The cytolethal distending toxin islet gene (cdtB) was detected in all the S. Give and S. Cotham in this study. This gene has been reported to play a vital role in disease pathogenesis. This toxin islet has been known to cause DNA damage and cell cycle arrest in impaired cells [73]. The islet gene (cdtB) was detected in Salmonella Telelke- bir in a similar study conducted in Southwestern Nigeria [27]. The sopA and sopE2 pseudogenes were detected in all 48 isolates. SopD2 pseudogenes were found in all S. Abony, S. Typhimurium, S. Senftenberg, and S. Mokola isolates, while shdA pseudogenes were confined to S. Abony and S. Mokola isolates. Langridge et al. [74] defined a pseudogene as a “gene with a mutation” (i.e., a premature stop codon, frameshift, truncation, or syntenic deletion) compared to an intact version of that gene. It is easier for Salmonella to enter epithelial cells when the effector genes found in Salmonella pathogenicity island 1 (SPI-1) are pseudogenized. This enhances Salmonella adaptability to systemic infection in humans [75, 76]. As part of the limitations of the study, it was not pos- sible to determine if the number of pseudogenes pres- ent in each of the 48 Salmonella enterica strains was due to identical or non-identical mutations. The Salmonella plasmid virulence (spv) locus harbors five genes desig- nated spv RABCD. The Salmonella virulence plasmid (spv)-RBC was found in the four S. Typhimurium strains. The expression of the spv genes has been reported to Akinyemi et al. BMC Microbiology (2023) 23:164 Page 13 of 17 play a role in the intracellular multiplication of Salmo- nellae by distorting the cytoskeleton of the eukaryotic cells using its spvB ADP-ribosylates actin [77, 78]. The two operons of the chaperone–usher class pef and ipf (plasmid-encoded fimbriae) known to mediate adhesion of S. Typhimurium to the small intestine in mice were detected and confined to four S. Typhimurium strains (pefA, pefB, pefC, and pefD), while the long polar fimbriae operons (ipfA, ipfB, ipfC, and ipfD) were detected in four S. Abony and four S. Typhimurium. Other fimbriae oper- ons of the chaperon class detected contain fimC, fimD, fimF, fimH, fimI, steA, steB, and steC, which are well con- served in all 48 Salmonella isolates. The four Salmonella invasion proteins (SipsABCD), which have been shown to play essential roles in the secretion and translocation of SPI-1 effectors, are present in all isolates [79]. The in silico serotype prediction with Seqsero in com- parison to the serotyping scheme passed the Fast-QC threshold, with the QC status indicating “PASS” for 5 serotypes except for serotype S. Mokola (6 isolates) show- ing “WARN QC” status for the quality of the sequences. There was no ST number available for S. Mokola from the MLST profile database. This is an indication that the complete genomic sequence of serovar S. Mokola has not been deposited in the global Salmonella database yet. This result represents the first complete genomic sequence of S. Mokola serovar from Nigeria. Multilocus sequence typing (MLST), allelic profiles, and sequence types (ST) of the 48 Salmonella enterica isolates were assigned by comparing the sequences with those in the MLST profile database. Only one sequence type (ST) each was found for the 17 S. Cotham (ST617), 16 S. Give (ST516), 4 S. Typhimurium (ST19), and 4 S. Abony (ST1483), while S. Senftenberg (ST14) is the only ST generated from segments of seven housekeeping genes (aroC, dnaN, hemD, hisD, purE, sucA, and thrA). All seven MLST loci were successfully recovered for the 48 Salmonella enterica strains isolated from different sources. The presence of such clones in the samples of human, animal, and environmental origin indicates epi- demiological links between STs and reservoir isolates. In Sub-Saharan Africa, NTS and iNTS serovars have been a challenge. The MLST of S. Typhimurium is known to be prevalent in Burkina Faso in humans and poultry [80]. Also, S. Typhimurium ST313 strain D23580 isolated from a patient with an iNTS infection has been reported from Malawi [81]. S. Typhimurium ST19 detected in this study belongs to the globally circulating lineage [82]. It is the dominant ST of the MLST database, which, however, consists of sequences of strains isolated from Europe and Northern America. S. Typhimurium ST19 has already caused gastroenteritis in humans [83], which is in accordance with this study as S. Typhimurium ST19 was detected in stool samples of patients with diarrheal disease from two study centers 20  km apart. The isola- tion of S. Typhimurium ST19 in hospital wastewater in this study is an indication of its dissemination from clinical sources into the environment. In a related study of WGS analysis of Salmonella serovars from animals in North-Central Nigeria, eight diverse sequence types (STs) were detected and the most common STs were ST-321 and ST-19 (n = 4) exhibited by S. Muenster and S. Typhimurium, respectively [26]. The detection of inva- sive S. Typhimurium ST19 clones in the blood of febrile patients with systematic infection has also been reported in Iran [84]. S. Typhimurium ST19 has been documented to colonize the gut and cause inflammation by a Sal- monella pathogenicity island (SPI)-1-mediated process when ingested orally [83]. The MLST of the 48 Salmo- nella isolates based on the source of isolates, type of sam- ple, and clinical prognosis revealed five distinct clusters in the 46 Salmonella isolates. Two of the Salmonella iso- lates could not be assigned to any cluster. WGS revealed that strains in each MLST Salmonella cluster from this study were closely related and likely shared a common ancestor. The distribution of these clusters within the study areas showed that all MLST clusters, including S. Senftenberg, that were not assigned to any cluster were found in the Alimosho local government area (LGA). All of the LGAs chosen for this investigation had isolates in clusters 1 and 2. Blood from a feverish patient at the Lagos State University teaching hospital and stool from a 3-5-year-old with diarrhea at Messy Street Children Hospital both contained invasive iNTS Cotham ST617 in cluster 1. While a genetically identical strain was discov- ered in wastewater taken from Gbagada general hospital, a few kilometers from Badagry LGA and Alimosho LGA, Salmonella Cotham ST617 from the same cluster was also detected in stool samples from poultry birds in Bada- gry and cattle dung along the Governor Road. Similar to this, S. Give (ST516) strains in cluster 2 were found in the Alimosho LGA from human, animal, and environmental samples. A strain of this serovar was also identified from cattle at Odo-eran abattoir in the same LGA, around 1.6  km distant from Alimosho general hospital. It was noted that the Salmonella Give strains that were isolated from wastewater and human samples (stool and blood) were from Alimosho general hospital. Additionally, S. Give (ST516) strains were found in the Lagos Island LGA (Messy Street Children hospital), Badagry LGA (Oko- Afor poultry facility), and Ikeja LGA (place of Lagos Uni- versity teaching hospital). The discovery of this disease in these three LGAs illustrates the spread and circulation of ST516 within Lagos environments because they are geo- graphically apart. In Nigeria, S. Give and S. Cotham sero- types have recently been reported in water, fecal samples feed, dust, and boot swabs from different poultry farms within the same and different localities [85]. However, Akinyemi et al. BMC Microbiology (2023) 23:164 Page 14 of 17 pathogens in food chains, proper disposal of refuse, treat- ment of wastewater from hospitals and food production industries, and control of dump sites near hospitals are essential to preventing outbreaks. Abbreviations NTS INTS MLST CgMLST CgSNP MDR EsβL KPC MBL ST MT NAFDAC PMQR SPI LGA RFLP VNTR MLVA MALDI-TOF MS Non-typhoidal Salmonella Invasive non-typhoidal Salmonella Multi-locus sequence typing Core genome multi-locus sequence typing Core genome single nucleotide polymorphism Multiple-drug resistant Extended spectrum beta-lactamases Klebsiella pneumoniae producing carbapenemase Metallo-beta-lactamase sequence type Metric tons National Agency for Food and Drug Administration and Control. plasmid-mediated quinolone resistance Salmonella Pathogenicity Island Local Government Area Restriction fragment length polymorphism Variable number of tandem repeats Multiple Locus Variable-Number Tandem Repeat Analysis matrix-assisted laser desorption/ionization-time of flight mass spectrometry Supplementary Information The online version contains supplementary material available at https://doi. org/10.1186/s12866-023-02901-1. Supplementary Material 1 Acknowledgements We are grateful to the staff of the Department of Microbiology and the management of Lagos State University (LASU). We are equally grateful to the management and staff of the Ministry of Health and the staff of the various hospitals, abattoirs, and animal farms for their contributions to this research. We are grateful to the Alexander von Humboldt (AvH) Foundation, Germany, for sponsoring and funding this research work. We are sincerely grateful to the management of the Friedrich-Loeffler-Institute (FLI), Jena, Germany, for financial and technical support. Authors’ contributions K.O.A conceived the study. C.O.F., K.O.A., J.L., and U.M. were responsible for the methodology. C.O.F. J.L., and U.M, were responsible for the software, the validation of the study was done by K.O.A., C.O.F., J.L., U.M., G.W., H.T., and H.N. The formal analysis was done by K.O.A., C.O.F., J.L., and U.M., the investigation was done by K.O.A., C.O.F., J.L., U.M., G.W., and H.T. writing—original draft manuscript was done by C.O.F., critical review of original draft by K.O.A., review, and editing of the manuscript by K.O.A., J.L., G.W., H.T., and H.N., and supervision was done by K.O.A. and H.N. Funding This study was funded by the Alexander von Humboldt (AvH) Foundation, Germany, and the Friedrich-Loeffler Institute (FLI), Jena, Germany. AvH has no role in the design of the study, the collection, analysis, and interpretation of data, or in writing the manuscript. Data Availability Raw sequences from this study are available and were deposited in the European Nucleotide Archive (ENA) with bio project accession PRJEB56537 in the ENA bio project database: https://www.ebi.ac.uk/ena/browser/view/ PRJEB56537. local environmental conditions and a lack of biosecurity measures may have been responsible for the dissemina- tion of similar strains [85]. The presence of the two domi- nant clones (S. Give ST516 and S. Cotham ST617) in samples from human, animal, and environmental sources demonstrated the need for further epidemiologic stud- ies to identify the infection and to tailor countermea- sures for the local setting. This study points to the fact that transmission of S. Give ST516 and S. Cotham ST617 may occur not only from person to person but also from animal to human. Controlling environmental contami- nation and potential control methods that could serve as a guide for appropriate waste management require special consideration. Near Alimosho General Hospital are two large, overburdened garbage dump sites, one of which is only 300 m away and the other is 1.3 km distant. The area is occupied by low-income, lower, and upper- middle-class residents, but no functional waste disposal treatment unit is available. These dump sites have grossly polluted the environment, including the groundwater [86]. This may have contributed to the high prevalence of Salmonella infection within that area and continuous spread to other areas of the state. Conclusion The characterization of Salmonella isolates from hos- pital, environmental, and animal production sources using different molecular tools was conducted. The study revealed six serotypes, with S. Give and S. Cotham as the most predominant ones. Closely related Salmonella clones were detected in these samples, pointing to an epidemiological link between serotypes and sequence types. The study also showed that the isolates were resis- tant to multiple antibiotics, as reflected by the presence of intrinsic and acquired genes conferring resistance to antibiotics and heavy metals. A wide range of virulence genes that help the organism become pathogenic in the host were detected from the whole genomic sequences. It becomes obvious that it is time to act and develop a strat- egy for Nigeria against the spread of some emerging NTS Salmonella serovars, such as S. Give and S. Cotham. This strategy must take into consideration food-producing animals, i.e., occupational animal contacts and food con- tamination, spill-back risk of the already heavily polluted environment, which arises from indiscriminate disposal of refuse and untreated wastewater discharge (effluent) from the hospital environment, and/or other sources, and consumer behavior and food availability. For the One Health context, these results are alarming, and the risk of further spread of plasmid replicon Col440I_1 and viru- lence genes is imminent if no immediate action is taken. The first step must be the control and prevention of the evolution of new and more virulent NTS in hospitals and veterinary clinics. Prompt surveillance of emerging Akinyemi et al. BMC Microbiology (2023) 23:164 Declarations Ethics approval and consent to participate Ethical approval from the ethics committee of the following institutions was obtained before patients’ enrolment: The Human Research and Ethics Committee of the Lagos State University Teaching Hospital with reference number LREC/06/10/1012 and the Lagos State Health Service Commission with reference number LSHSC/2222/VOL.VC/352. All methods were carried out in accordance with relevant guidelines and regulations, and informed consent was obtained from all subjects and/or their legal guardian(s). Consent for publication Not applicable. Competing interests The authors declare no competing interests. Author details 1Department of Microbiology, Faculty of Science, Lagos State University, Ojo, Lagos, Nigeria 2Institute of Bacterial Infections and Zoonoses, Friedrich-Loeffler-Institute, Jena, Germany 3Department of Bacteriology, Immunology, and Mycology, Faculty of Veterinary Medicine, Benha University, PO Box 13736, Toukh, Moshtohor, Egypt 4Institute of Infectious Diseases and Infection Control, Jena University Hospital, Am Klinikum 1, 07747 Jena, Germany Received: 27 October 2022 / Accepted: 17 May 2023 References 1. Crump JA, Luby SP, Mintz ED. The global burden of typhoid fever. Bull World Health Organ. 2004;82(5):346–53. Braden CR. Salmonella enterica serotype Enteritidis and eggs: a national epidemic in the United States. Clin Infect Dis. 2006;43:512–7. Rotimi V, Jamal W, Pal T, Sonnevend A, Dimitrov T, Albert MJ. Emergence of multidrug-resistant Salmonella spp. and isolates with reduced susceptibility to ciprofloxacin in Kuwait and the United Arab Emirates. Diagnostic microbi- ology and infectious disease 2008, 60(1):71–7. Razzaque M, Bedair M, Abbas S, Al-Mutawa T. Economic impact of calf mor- tality on dairy farms in Kuwait. Pakistan Vet J. 2009;29(3):97–101. Parry CM, Thomas S, Aspinall EJ, Cooke RP, Rogerson SJ, Harries AD, Beeching NJ. A retrospective study of secondary bacteraemia in hospitalised adults with community acquired non-typhoidal Salmonella gastroenteritis. BMC Infect Dis. 2013;13(1):1–10. Akinyemi K, Iwalokun BA, Oyefolu AOB, Fakorede C. Occurrence of extended- spectrum and AmpC β-lactamases in multiple drug resistant Salmonella isolates from clinical samples in Lagos, Nigeria. Infection and drug resistance 2017, 10:19. Crump JA, Sjölund-Karlsson M, Gordon MA, Parry CM. Epidemiology, clinical presentation, laboratory diagnosis, antimicrobial resistance, and antimi- crobial management of invasive Salmonella infections. Clin Microbiol Rev. 2015;28:901–37. Crump JA, Heyderman RS. A perspective on invasive Salmonella disease in Africa. Clin Infect Dis. 2015;61(suppl4):235–S240. 2. 3. 4. 5. 6. 7. 8. 9. Majowicz SE, Musto J, Scallan E, Angulo FJ, Kirk M, O’Brien SJ, Jones TF, Fazil A, Hoekstra RM. Studies ICoEDBoI: the global burden of nontyphoidal Salmo- nella gastroenteritis. Clin Infect Dis. 2010;50(6):882–9. 10. Feasey N, Wansbrough-Jones M, Mabey DC, Solomon AW. Neglected tropical diseases. Br Med Bull. 2010;93(1):179–200. 11. Organization WH. Country profiles of environmental burden of disease. Geneva: WHO 2009:35. 12. Fagbamila IO, Barco L, Mancin M, Kwaga J, Ngulukun SS, Zavagnin P, Lettini AA, Lorenzetto M, Abdu PA, Kabir J. Salmonella serovars and their distribution in nigerian commercial chicken layer farms. PLoS ONE. 2017;12(3):e0173097. Page 15 of 17 13. Omololu-Aso J, Oluwatoyin O, Omololu-Aso M, Atiene AA, Owolabi A, Shesha A. Salmonellosis and shigellosis associated with cattle dung contaminant from indigenous abattoirs, Osun State, Nigeria. Br J Res 2017, 4(1). 14. Sahel. An assessment of the nigerian poultry sector. In : www fafinnigeria- com; 2015: 23. 15. Sati NM, Okolocha EC, Kazeem HM, Kabir J, Fagbamila IO, Muhammad M, AA L. Seminar Presentation on the Sources of Salmonella infections in some selected poultry farms in Jos, northern Nigeria.). Sources of Salmonella infec- tions in some selected poultry farms in Jos, northern Nigeria.). Sources of Salmonella infections in some selected poultry farms in Jos, northern Nigeria. In. Jos Nigeria:. In.: National veterinary research institute, Jos Nigeria; 2015. 16. Obaro SK, Hassan-Hanga F, Olateju EK, Umoru D, Lawson L, Olanipekun G, Ibrahim S, Munir H, Ihesiolor G, Maduekwe A. Salmonella bacteremia among children in central and northwest Nigeria, 2008–2015. Clin Infect Dis. 2015;61(suppl4):325–S331. 17. Manyi-Loh C, Mamphweli S, Meyer E, Okoh A. Antibiotic use in agriculture and its consequential resistance in environmental sources: potential public health implications. Molecules 2018, 23(4):795. 18. Woerther P-L, Burdet C, Chachaty E, Andremont A. Trends in human fecal carriage of extended-spectrum β-lactamases in the community: toward the globalization of CTX-M. Clinical microbiology reviews 2013, 26(4):744–58. 19. Akinyemi KO, Iwalokun BA, Alafe OO, Mudashiru SA, Fakorede C. blaCTX-MI 20. group extended spectrum beta lactamase-producing Salmonella typhi from hospitalized patients in Lagos, Nigeria. Infection and drug resistance 2015, 8:99. Iroha IR, Esimone CO, Neumann S, Marlinghaus L, Korte M, Szabados F, Gatermann S, Kaase M. First description of Escherichia coli producing CTX- M-15-extended spectrum beta lactamase (ESBL) in out-patients from south eastern Nigeria. Ann Clin Microbiol Antimicrob. 2012;11(1):1–5. 21. Dangel A, Berger A, Messelhäußer U, Konrad R, Hörmansdorfer S, Ackermann N, Sing A. Genetic diversity and delineation of Salmonella Agona outbreak strains by next generation sequencing, Bavaria, Germany, 1993 to 2018. Eurosurveillance. 2019;24(18):1800303. 22. Maiden MC, Van Rensburg MJJ, Bray JE, Earle SG, Ford SA, Jolley KA, McCarthy ND. MLST revisited: the gene-by-gene approach to bacterial genomics. Nat Rev Microbiol. 2013;11(10):728–36. 23. Ramisse V, Houssu P, Hernandez E, Denoeud F, Hilaire V, Lisanti O, Ramisse F, Cavallo JD, Vergnaud G. Variable number of tandem repeats in Salmo- nella enterica subsp. enterica for typing purposes. J Clin Microbiol. 2004 Dec;42(12):5722–30. 24. Parkhill J, Dougan G, James K, Thomson N, Pickard D, Wain J, Churcher C, Mungall K, Bentley S, Holden M. Complete genome sequence of a multiple drug resistant Salmonella enterica serovar Typhi CT18. Nature. 2001;413(6858):848–52. 25. Leopold SR, Goering RV, Witten A, Harmsen D, Mellmann A. Bacterial whole- genome sequencing revisited: portable, scalable, and standardized analysis for typing and detection of virulence and antibiotic resistance genes. J Clin Microbiol. 2014;52(7):2365–70. 27. 26. Raufu IA, Ahmed OA, Aremu A, Ameh JA, Timme RE, Hendriksen RS, Ambali AG. Occurrence, antimicrobial resistance and whole genome sequence analysis of Salmonella serovars from pig farms in Ilorin, North-central Nigeria. Int J Food Microbiol 2021 Jul 16;350:109245. Ikhimiukor OO, Oaikhena AO, Afolayan AO, Fadeyi A, Kehinde A, Ogunleye VO, Aboderin AO, Oduyebo OO, Elikwu CJ, Odih EE, Komolafe I, Argimón S, Egwuenu A, Adebiyi I, Sadare OA, Okwor T, Kekre M, Underwood A, Ihekweazu C, Aanensen DM, Okeke IN. Genomic characterization of invasive typhoidal and non-typhoidal Salmonella in southwestern Nigeria. PLoS Negl Trop Dis. 2022 Aug;26(8):e0010716. 28. England PH. Detection of Salmonella species National Infection Service ood Water and Environmental Microbiology Standard Method. In., vol. 4. Waterloo Road Wellington House London SE1 8UG; 2017. 29. Collee JG, Mackie TJ, McCartney JE. Mackie & McCartney practical medical microbiology. Harcourt Health Sciences; 1996. 30. Bizzini A, Greub G. Matrix-assisted laser desorption ionization time-of-flight mass spectrometry, a revolution in clinical microbial identification. Clin Microbiol Infect. 2010;16(11):1614–9. 31. Lüthje P, Pranada AB, Carruthers-Lay D, Desjardins M, Gaillot O, Wareham D, Ciesielczuk H, Özenci V. Identification of microorganisms grown on chromo- genic media by MALDI-TOF MS. J Microbiol methods 2017, 136:17–20. 32. Grimont PA, Weill F-X. Antigenic formulae of the Salmonella serovars. WHO collaborating centre for reference and research on Salmonella. 2007;9:1–166. 33. Clinical breakpoints and dosing of antibiotics. Akinyemi et al. BMC Microbiology (2023) 23:164 Page 16 of 17 34. García-Soto S, Abdel-Glil MY, Tomaso H, Linde J, Methner U. Emergence of multidrug-resistant Salmonella enterica subspecies enterica serovar infantis of multilocus sequence type 2283 in german broiler farms. Front Microbiol. 2020;11:1741. 35. García-Soto S, Tomaso H, Linde J, Methner U. Epidemiological analysis of Salmonella enterica subsp. enterica Serovar Dublin in german cattle herds using whole-genome sequencing. Microbiol Spectr. 2021;9(2):e00332–00321. 36. Andrews S. FastQC: a quality control tool for high throughput sequence data. In.: Babraham Bioinformatics, Babraham Institute, Cambridge, United Kingdom; 2010. 37. Seemann T. Shovill—Assemble Bacterial Isolate Genomes from Illumina Paired-End Reads. GitHub: San Francisco, CA, USA 2020. 38. Gurevich A, Saveliev V, Vyahhi N, Tesler G. QUAST: quality assessment tool for genome assemblies. Bioinformatics. 2013;29(8):1072–5. 39. Wood DE, Lu J, Langmead B. Improved metagenomic analysis with Kraken 2. Genome Biol. 2019;20(1):1–13. 40. Yoshida CE, Kruczkiewicz P, Laing CR, Lingohr EJ, Gannon VP, Nash JH, Taboada EN. The Salmonella in silico typing resource (SISTR): an open web- accessible tool for rapidly typing and subtyping draft Salmonella genome assemblies. PLoS ONE. 2016;11(1):e0147101. 41. Zhang S, Yin Y, Jones MB, Zhang Z, Deatherage Kaiser BL, Dinsmore BA, Fitzgerald C, Fields PI, Deng X. Salmonella serotype determination utiliz- ing high-throughput genome sequencing data. J Clin Microbiol 2015, 53(5):1685–92. 42. Feldgarden M, Brover V, Haft DH, Prasad AB, Slotta DJ, Tolstoy I, Tyson GH, Zhao S, Hsu C-H. McDermott PF: Using the NCBI AMRFinder tool to deter- mine antimicrobial resistance genotype-phenotype correlations within a collection of NARMS isolates. BioRxiv 2019:550707. 43. Abricate, Github. 44. Liu B, Zheng D, Zhou S, Chen L, Yang J. (2022) VFDB 2022: a general clas- sification scheme for bacterial virulence factors. Nucleic Acids Res 50: D912-D917.S. 45. Carattoli A, Zankari E, García-Fernández A, Voldby Larsen M, Lund O, Villa L, Møller Aarestrup F, Hasman H. In silico detection and typing of plasmids using PlasmidFinder and plasmid multilocus sequence typing. Antimicrob Agents Chemother. 2014;58(7):3895–903. 46. mlst GitHub. 47. Snippy. GitHub. 48. Jünemann S, Sedlazeck FJ, Prior K, Albersmeier A, John U, Kalinowski J, Mellmann A, Goesmann A, Von Haeseler A, Stoye J. Updating benchtop sequencing performance comparison. Nat Biotechnol. 2013;31(4):294–6. 49. Alikhan N-F, Zhou Z, Sergeant MJ, Achtman M. A genomic overview of the population structure of Salmonella. PLoS Genet. 2018;14(4):e1007261. 50. Veeraraghavan B, Jacob JJ, Prakash JAJ, Pragasam AK, Neeravi A, Narasim- man V, Anandan S. Extensive drug resistant Salmonella enterica serovar senftenberg carrying bla NDM encoding plasmid p5558 (IncA/C) from India. Pathogens and Global Health. 2019;113(1):20–6. Jibril AH, Okeke IN, Dalsgaard A, Menéndez VG, Olsen JE. Genomic analysis of antimicrobial resistance and resistance plasmids in Salmonella serovars from poultry in Nigeria. Antibiotics. 2021;10(2):99. 51. 52. Eguale T. Non-typhoidal Salmonella serovars in poultry farms in central Ethio- pia: prevalence and antimicrobial resistance. BMC Vet Res. 2018;14(1):1–8. 53. Andoh L, Dalsgaard A, Obiri-Danso K, Newman M, Barco L, Olsen J. Prevalence and antimicrobial resistance of Salmonella serovars isolated from poultry in Ghana. Epidemiol Infect. 2016;144(15):3288–99. 54. Helwigh B, Müller L. Annual Report on Zoonoses in Denmark 2017. Annual 55. Report on Zoonoses in Denmark 2017 2018. Jansen A, Frank C, Prager R, Oppermann H, Stark K. Nation-wide outbreak of Salmonella give in Germany, 2004. Z Gastroenterol. 2005;43(8):707–13. 56. Post AS, Diallo SN, Guiraud I, Lompo P, Tahita MC, Maltha J, Van Puyvelde S, Mattheus W, Ley B, Thriemer K. Supporting evidence for a human reservoir of invasive non-typhoidal Salmonella from household samples in Burkina Faso. PLoS Negl Trop Dis. 2019;13(10):e0007782. 57. Fashae K, Hendriksen RS. Diversity and antimicrobial susceptibility of Salmonella enterica serovars isolated from pig farms in Ibadan, Nigeria. Folia microbiologica 2014, 59(1):69–77. 58. Reddy EA, Shaw AV, Crump JA. Community-acquired bloodstream infec- tions in Africa: a systematic review and meta-analysis. Lancet Infect Dis. 2010;10(6):417–32. 59. Sigaúque B, Roca A, Mandomando I, Morais L, Quintó L, Sacarlal J, Macete E, Nhamposa T, Machevo S, Aide P. Community-acquired bacteremia among children admitted to a rural hospital in Mozambique. Pediatr Infect Dis J. 2009;28(2):108–13. 60. Brent AJ, Oundo JO, Mwangi I, Ochola L, Lowe B, Berkley JA. Salmonella bacteremia in kenyan children. Pediatr Infect Dis J. 2006;25(3):230–6. 61. Gordon MA, Graham SM, Walsh AL, Wilson L, Phiri A, Molyneux E, Zijlstra EE, Heyderman RS, Hart CA, Molyneux ME. Epidemics of invasive Salmonella enterica serovar enteritidis and S. enterica Serovar typhimurium infection associated with multidrug resistance among adults and children in Malawi. Clin Infect Dis. 2008;46(7):963–9. 62. Akinyemi KO, Bamiro BS, Coker AO. Salmonellosis in Lagos, Nigeria: incidence of Plasmodium falciparum-associated co-infection, patterns of antimicrobial resistance, and emergence of reduced susceptibility to fluoroquinolones. J Health Popul Nutr. 2007;25(3):351. 63. Oluwasile B, Agbaje M, Ojo O, Dipeolu M. Antibiotic usage pattern in selected poultry farms in Ogun state. Sokoto J Veterinary Sci. 2014;12(1):45–50. 64. Al-Mustapha AI, Adetunji VO, Heikinheimo A. Risk perceptions of antibiotic usage and resistance: a cross-sectional survey of poultry farmers in Kwara State, Nigeria. Antibiotics. 2020;9(7):378. 65. Veldman K, Cavaco LM, Mevius D, Battisti A, Franco A, Botteldoorn N, Bruneau M, Perrin-Guyomard A, Cerny T, De Frutos Escobar C. International collabora- tive study on the occurrence of plasmid-mediated quinolone resistance in Salmonella enterica and Escherichia coli isolated from animals, humans, food and the environment in 13 european countries. J Antimicrob Chemother. 2011;66(6):1278–86. 66. Fashae K, Folasade O, Frank M, Rene S. Antimicrobial susceptibility and serovars of Salmonella from chickens and humans in Ibadan, Nigeria. J Infect Dev Ctries. 2010;4(8):484–94. 67. Fortini D, Fashae K, García-Fernández A, Villa L, Carattoli A. Plasmid-mediated quinolone resistance and β-lactamases in Escherichia coli from healthy animals from Nigeria. J Antimicrob Chemother 2011, 66(6):1269–72. 68. Robicsek A, Sahm D, Strahilevitz J, Jacoby G, Hooper D. Broader distribution of plasmid-mediated quinolone resistance in the United States. Antimicrob Agents Chemother. 2005;49(7):3001–3. 69. Moreno-Switt AI, Pezoa D, Sepúlveda V, González I, Rivera D, Retamal P, Navarrete P, Reyes-Jara A, Toro M. Transduction as a potential dissemination mechanism of a clonal qnrB19-carrying plasmid isolated from Salmonella of multiple serotypes and isolation sources. Front Microbiol. 2019;10:2503. 70. Kombade S, Kaur N. Pathogenicity Island in < em > Salmonella. In A. Lamas, P. 71. Regal, & C. M. Franco, editors, Salmonella spp. - A Global Challenge. IntechO- pen. 2021. https://doi.org/10.5772/intechopen.96443. Jajere SM. A review of Salmonella enterica with particular focus on the patho- genicity and virulence factors, host specificity and antimicrobial resistance including multidrug resistance. Veterinary world. 2019;12(4):504. 72. Sabbagh SC, Forest CG, Lepage C, Leclerc J-M, Daigle F. So similar, yet so different: uncovering distinctive features in the genomes of Salmonella enterica serovars Typhimurium and Typhi. FEMS microbiology letters 2010, 305(1):1–13. 73. Chang SJ, Jin SC, Jiao X, Galan JE. Unique features in the intracellular trans- port of typhoid toxin revealed by a genome-wide screen. PLOS Pathogens. 2019; 15(4):e1007704. https://doi.org/10.1371/journal.ppat.1007704 PMID: 30951565. 74. Langridge GC, Fookes M, Connor TR, Feltwell T, Feasey N, Parsons BN, Seth- Smith HM, Barquist L, Stedman A, Humphrey T. Patterns of genome evolution that have accompanied host adaptation in Salmonella. Proceedings of the National Academy of Sciences 2015, 112(3):863–868. 75. Trombert AN, Berrocal L, Fuentes JA, Mora GC. Typhimurium sseJ gene decreases the S. Typhi cytotoxicity toward cultured epithelial cells. BMC Microbiol. 2010;10:1–10. 76. Valenzuela LM, Hidalgo AA, Rodríguez L, Urrutia IM, Ortega AP, Villagra NA, Paredes-Sabja D, Calderón IL, Gil F, Saavedra CP, Mora GC. Pseudogenization of sopA and sopE2 is functionally linked and contributes to virulence of Salmonella enterica serovar Typhi. Infect Genet Evol. 2015;33:131–42. 77. Lesnick ML, Reiner NE, Fierer J, Guiney DG. The Salmonella spvB virulence gene encodes an enzyme that ADP-ribosylates actin and destabilizes the cytoskeleton of eukaryotic cells. Mol Microbiol. 2001;39(6):1464–70. 78. Matsui H, Bacot CM, Garlington WA, Doyle TJ, Roberts S, Gulig PA. Virulence plasmid-borne spvB and spvC genes can replace the 90-kilobase plasmid in conferring virulence to Salmonella enterica serovar Typhimurium in subcuta- neously inoculated mice. J Bacteriol 2001, 183(15):4652–8. 79. Lou L, Zhang P, Piao R, Wang Y. Salmonella pathogenicity island 1 (SPI-1) and its complex regulatory network. Front Cell Infect Microbiol 2019, 9:270. Akinyemi et al. BMC Microbiology (2023) 23:164 Page 17 of 17 80. Kagambèga A, Lienemann T, Frye JG, Barro N, Haukka K. Whole genome 84. Ranjbar R, Elhaghi P, Shokoohizadeh L. Multilocus sequence typing of the sequencing of multidrug-resistant Salmonella enterica serovar typhimurium isolated from humans and poultry in Burkina Faso. Trop Med health. 2018;46(1):1–5. 81. Kingsley RA, Msefula CL, Thomson NR, Kariuki S, Holt KE, Gordon MA, Harris D, Clarke L, Whitehead S, Sangal V. Epidemic multiple drug resistant Salmonella Typhimurium causing invasive disease in sub-saharan Africa have a distinct genotype. Genome Res. 2009;19(12):2279–87. 82. Canals R, Hammarlöf DL, Kröger C, Owen SV, Fong WY, Lacharme-Lora L, Zhu X, Wenner N, Carden SE, Honeycutt J. Adding function to the genome of african Salmonella Typhimurium ST313 strain D23580. PLoS Biol. 2019;17(1):e3000059. 83. Hughes L, Wigley P, Bennett M, Chantrey J, Williams N. Multi-locus sequence typing of Salmonella enterica serovar Typhimurium isolates from wild birds in northern England suggests host‐adapted strain. Lett Appl Microbiol. 2010;51(4):477–9. clinical isolates of Salmonella enterica serovar Typhimurium in Tehran Hospi- tals. Iran J Med Sci. 2017;42(5):443. 85. Fagbamila IO, Barco L, Mancin M, Kwaga J, Ngulukun SS, Zavagnin P, Lettini AA, Lorenzetto M, Abdu PA, Kabir J, Umoh J. Salmonella serovars and their distribution in Nigerian commercial chicken layer farms. PLoS One. 2017 Mar 9;12(3):e0173097. 86. Salami L, Susu AA. 2019. A comprehensive study of leachate characteristics from three soluos dumpsites in Igando Area of Lagos State, Nigeria. Greener Journal of Environmental Management and Public Safety, 8(1), pp.1–14. Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Akinyemi et al. BMC Microbiology (2023) 23:164
null
10.1371_journal.pone.0244053.pdf
Data Availability Statement: The data are accessible via Dataverse (https://doi.org/10.7910/ DVN/8ZVOKW).
The data are accessible via Dataverse ( https://doi.org/10.7910/ DVN/8ZVOKW ).
RESEARCH ARTICLE Gender specific differences in COVID-19 knowledge, behavior and health effects among adolescents and young adults in Uttar Pradesh and Bihar, India Jessie PinchoffID D. Ngo1 1*, KG Santhya2, Corinne White1, Shilpi Rampal2, Rajib Acharya2, Thoai 1 Population Council, One Dag Hammarskjold Plaza, New York, NY, United States of America, 2 Population Council, India Habitat Centre, New Delhi, Delhi, India * jpinchoff@popcouncil.org Abstract On March 24, 2020 India implemented a national lockdown to prevent spread of the novel Coronavirus disease (COVID-19) among its 1.3 billion people. As the pandemic may dispro- portionately impact women and girls, this study examines gender differences in knowledge of COVID-19 symptoms and preventive behaviors, as well as the adverse effects of the lock- down among adolescents and young adults. A mobile phone-based survey was imple- mented from April 3–22, 2020 in Uttar Pradesh and Bihar among respondents randomly selected from an existing cohort study. Respondents answered questions related to demo- graphics, COVID-19 knowledge, attitudes, and preventive behaviors practiced, and impacts on social, economic and health outcomes. Descriptive analyses and linear probability regression models were performed for all participants and separately for men and women. A total of 1,666 adolescents and young adults (18–24 years old) were surveyed; 70% were women. While most participants had high awareness of disease symptoms and preventive behaviors, there was variation by gender. Compared to men, women were seven percent- age points (pp) less likely to know the main symptoms of COVID-19 (coeff = -0.071; 95% confidence interval: -0.122 - -0.021). Among women, there was variation in knowledge by education level, urban residence, and household wealth. Women were 22 pp less likely to practice key preventive behaviors compared to men (coeff = -0.222; 95% CIL -0.263, -0.181). Women were also more likely to report recent depressive symptoms than men (coeff = 0.057; 95% CI: 0.004, 0.109). Our findings underscore that COVID-19 is already disproportionately impacting adolescent girls and young women and that they may require additional targeted, gender-sensitive messaging to foster behavior change. Gender-sensi- tive information campaigns and provision of health services must be accessible and provide women and girls with needed resources and support during the pandemic to ensure gains in public health and gender equity are not lost. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Pinchoff J, Santhya K, White C, Rampal S, Acharya R, Ngo TD (2020) Gender specific differences in COVID-19 knowledge, behavior and health effects among adolescents and young adults in Uttar Pradesh and Bihar, India. PLoS ONE 15(12): e0244053. https://doi.org/10.1371/journal. pone.0244053 Editor: Kannan Navaneetham, University of Botswana, BOTSWANA Received: August 24, 2020 Accepted: November 25, 2020 Published: December 17, 2020 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0244053 Copyright: © 2020 Pinchoff et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The data are accessible via Dataverse (https://doi.org/10.7910/ DVN/8ZVOKW). PLOS ONE | https://doi.org/10.1371/journal.pone.0244053 December 17, 2020 1 / 13 PLOS ONE Funding: The initial UDAYA cohort was funded by the Bill and Melinda Gates Foundation and Packard Foundation. No additional funds were received for the COVID-19 survey. Competing interests: The authors have declared that no competing interests exist. Gender specific variation in COVID-19 knowledge, behavior and health effects among young adults Introduction To control the spread of the novel Coronavirus disease (COVID-19), the Indian government swiftly instituted a shutdown of international borders and a stay-at-home order on March 24, 2020 [1]. Such ‘lockdown’ policies to prevent the spread of COVID-19 originated in high- income countries and China; little primary research has explored the potential unintended consequences in countries including India characterized by densely populated urban slums, a highly mobile population, high proportions of informal sector workers, and stark variation in poverty levels [2]. Despite rising cases, the lockdown was lifted on June 8, 2020 to begin a phased reopening. As of August 2020, India surpassed 2.3 million cases of COVID-19, the third highest case load after the United States and Brazil [3]. Historically, epidemics and humanitarian crises have disproportionately impacted the most vulnerable, including women and girls [4]. Entrenched inequalities in access to education, job opportunities, and healthcare often leave women inadequately equipped to effectively protect themselves and their families against infection during an outbreak, and they are also more likely to bear secondary negative effects of prolonged crises, such as economic insecurity or challenges accessing essential health services [5]. Existing gender disparities in India may be exacerbated or reinforced by the pandemic and are likely to affect women’s ability to make informed decisions about adopting behaviors that mitigate risk of COVID-19. Prevention campaigns and behavior change communication interventions across various media, including a government-run mobile app (“Aarogya Setu”) that sends automated mes- sages, are informing the public about COVID-19 symptoms, risk factors, and promoting pre- ventive behaviors such as handwashing, social distancing, and wearing masks in India. To date, there is little to no research tracing how COVID-19 messages are reaching men and women or which sub-groups are adopting these behavioral recommendations. However, a rapid situational assessment in the South Asia region (not including India) suggests that women are less likely than men to have received COVID-19 information [6]. Moreover, liter- acy, internet usage and smartphone ownership is lower among women compared to men in India [7–9]. Accessing and understanding health promotion messages increases knowledge, which needs to be accompanied with structural facilitators and access to resources to adopt promoted preventive behaviors (e.g., making soap and water available for handwashing) [10– 12]. These gender gaps may result in lower adoption of promoted health behaviors and increased risk of infection for women and girls. The worsening COVID-19 pandemic in India is causing prolonged social and economic disruptions that are yielding unintended consequences including economic and food insecu- rity, and challenges in accessing healthcare. Challenges in accessing essential health services may lead to increases in other adverse health outcomes, from vaccine preventable diseases to poor birth outcomes and malnutrition [13,14]. This often disproportionately harms women who may require healthcare themselves and are also often responsible for taking care of their family’s health needs. Potential reasons for these challenges may include inability to pay clinic fees as COVID-19 related economic insecurity persists, mobility challenges, or fear of seeking care due to stigma or concerns about COVID-19 infection at the facility. Indeed, compared to March 2019, March 2020 data from the Indian National Health Mission showed marked reductions in indicators of regular health system usage [2]. In addition to physical health, lockdowns may exacerbate household stress, contributing to increases in sexual and gender-based violence (SGBV) and poor mental health symptoms [15,16]. While psychological distress increases generally during crises, experience of depressive symptoms is more common among women compared to men [17]. In addition to gender, a recent study also found that adolescents and younger adults (<25 years), those that had lost PLOS ONE | https://doi.org/10.1371/journal.pone.0244053 December 17, 2020 2 / 13 PLOS ONE Gender specific variation in COVID-19 knowledge, behavior and health effects among young adults employment, and/or lacked formal education were more likely to experience depressive symp- toms as a result of the pandemic’s effects [18]. Relatedly, stress and ongoing lockdowns have been linked with violence against women, as in past humanitarian crises [19]. Some countries reported increases in SGBV during COVID-19 lockdowns [15,20]. Concerns around these sec- ondary health and well-being effects are significant. As India is home to the largest population of adolescents and young adults of any country worldwide, understanding the impact of the pandemic on this important age cohort will also be critical. In the age- and gender-stratified settings of India, prevailing gender disparities and traditional gender norms affect health and well-being of adolescents and young people dispro- portionately. However, little is known regarding the experience during the COVID-19 pan- demic of Indian adolescent girls and young women compared to men. A cross-sectional mobile phone-based survey of households in Uttar Pradesh (UP) and Bihar was carried out four to six weeks after lockdown was imposed. This analysis highlights the gender specific vari- ation in COVID-19 knowledge and practice of preventive behaviors, and mental health effects among a cohort of adolescent and young adults. Findings from this study can inform the development of social service programs and education campaigns to ensure that adolescent and young women have access to tailored information and resources during this protracted crisis to ensure development and equity gains are not lost. Methods Sampling strategy A rapid telephone survey was conducted with a sample of participants drawn from an existing Population Council cohort study of adolescents and young adults. Understanding the Lives of Adolescents and Young Adults (UDAYA) is a state-level representative longitudinal study of adolescent girls and boys (aged 10–19) in rural and urban settings in Bihar (n = 10,433) and UP (n = 10,161), with baseline conducted in 2015–2016 and endline in 2018–19. The original UDAYA study objectives were to better understand adolescents’ acquisition of assets and their transition from adolescence to adulthood [21,22]. UDAYA researchers used the 2011 Indian Census to create a systematic, multi-stage sampling frame for the selection of 150 primary sampling units (PSU) in each state, with an equal breakdown between urban and rural areas. UDAYA was designed to provide estimates for five categories of adolescents, namely unmar- ried younger boys and girls aged 10–14, unmarried older boys and girls aged 15–19, and mar- ried older girls aged 15–19 that represent each state [21,22]. UDAYA households eligible for inclusion in the COVID-19 survey were those in which we interviewed a 15-19-year-old boy or girl in 2015–16. Phone numbers were available for 9,771 of such UDAYA participants– 2,437 boys and men and 7,334 girls and women. We randomly sampled households for the mobile phone survey from this list of telephone num- bers, stratified by gender. The enumerators contacted telephone numbers belonging to 5,520 UDAYA participants– 1,512 boys and men and 4,008 girls and women–attempting each number up to 3 times and completing about 10 interviews per day. Of those attempted, 51% of telephone numbers were no longer functional (of UDAYA participants, 44% of boys and men and 53% of girls and women). Of numbers we successfully reached, 5% of respon- dents refused to participate in the study. Overall, participants in the COVID-19 study had slightly higher educational attainment, were slightly more urban, and had slightly higher household wealth compared to the source cohort. The characteristics of the UDAYA base- line cohort compared to those who were enrolled in the COVID-19 mobile-phone survey is summarized in a S1 Table. PLOS ONE | https://doi.org/10.1371/journal.pone.0244053 December 17, 2020 3 / 13 PLOS ONE Gender specific variation in COVID-19 knowledge, behavior and health effects among young adults Mobile phone questionnaire Participants were contacted via mobile phone to remove the risk to field staff and participants of COVID-19 infection. After verbal consent for participation, a short questionnaire lasting no longer than 30 minutes was administered. The questionnaire included questions regarding basic demographics, awareness of COVID-19 or coronavirus, knowledge of symptoms, risk groups and transmission, perceived risk, COVID-19 prevention behaviors, and fears or con- cerns regarding the outbreak. Questions assessing household and individual needs under the government lockdown were also included. In the survey participants self-reported their sex as male or female; throughout this paper we will refer to respondents as men and women to illus- trate that our analysis reports how the pandemic impacts gender (the socially constructed characteristics of men and women) not biological sex. Ethical review We received expedited ethical approval from the Population Council’s Institutional Review Board (IRB) by meeting criteria for research conducted during COVID-19. The IRB permitted data collection with participants with previous consent from existing cohort studies, provided the research is aligned with national mitigation efforts. The UDAYA study protocol originally received IRB approval in 2015 for longitudinal data collection. Participants were told they could terminate the study at any time or skip any sections. No incentives were offered for tak- ing part in the study. Data management and analysis The survey responses were entered in mini laptops using instruments developed with CSPro 7.1 and exported to Stata v15 for analysis. Each household had a unique ID number, and all personally identifiable information was removed to ensure confidentiality. Two summary outcome variables were created. First, participants who correctly identified all three COVID-19 symptoms (fever, cough and difficulty in breathing) were considered to have correct knowledge (dichotomous variable). Participants who reported implementing all four preventive behaviors (staying home more, wearing a mask, washing hands/using sanitizer, and staying 2m apart) were categorized as implementing the four main preventive behaviors (dichotomous variable). Depressive symptoms, as measured by reporting feeling lonely, depressed or irritable during the lockdown, was collected as a dichotomous variable. To con- trol for household wealth, we created a proxy variable constructed from the presence of four basic amenities: safe drinking water, electricity, toilet facility and safe cooking fuel. Educational attainment was categorized into three levels, with grade 8 indicating completion of primary education and grade 10 indicating completion of secondary education. Religion was catego- rized as Hindu or Muslim (dichotomous variable), with 9 indicating ‘other’ and excluded from models. Lastly, caste was categorized as scheduled caste/tribe (SC/ST), other backward castes (OBC) and general (neither SC/ST nor OBC); these designations, as provisioned in the Indian constitution, are used to identify marginalized groups in the population. Only women were asked if they had experienced any violence in the home in the last 15 days under lockdown. All survey responses were tabulated by gender and tested for statistical significance (p<0.05) using chi-square tests. We implemented linear probability regression models based on three outcomes of interest. First, knowledge of all three key symptoms of COVID-19. Sec- ond, practicing all four of the key preventive behaviors. The third outcome was self-reported experience of loneliness, depression, or irritability (dichotomous variable) in the previous seven days used to define experience of depressive symptoms. Three separate linear probability PLOS ONE | https://doi.org/10.1371/journal.pone.0244053 December 17, 2020 4 / 13 PLOS ONE Gender specific variation in COVID-19 knowledge, behavior and health effects among young adults regression models were constructed for each of the three outcome variables, first for the full set of respondents and then stratified by gender. Results A total of 1,666 adolescents and young adults (18–24 years) previously enrolled in the UDAYA study were surveyed. Of these, 70% were women, over half had completed 10+ years of educa- tion (72%) and nearly half resided in urban areas (47%) (Table 1). Fewer women (40%) than Table 1. Demographics and COVID-19 related outcomes of interest tabulated by gender. Demographic/Household characteristics Age group 18–19 years 20–24 years Religion Hindu Muslim Other Completed years of education 0–7 years 8–9 years 10 and above years Caste General caste Other backward caste (OBC) Scheduled caste/tribe Current place of residence Urban (vs Rural) Have four key amenities1 Yes (vs No) COVID-19 Outcomes of Interest Mental Health: have you felt depressed, lonely or irritable under lockdown? Never Sometimes Most of the time Knowledge and behaviors Knows all 3 top symptoms2 Reports practicing all 4 main preventive measures3 Economic and health access effects Self or household member lost job/income source due to COVID-19 Among women who required each health service but could not access it: Antenatal care Family planning Child immunization Nutrition Men N = 506 Women N = 1,160 Total N = 1,666 88 (17%) 418 (83%) 432 (85%) 68 (13%) 6 (1%) 25 (5%) 59 (12%) 422 (83%) 108 (21%) 268 (53%) 130 (26%) 160 (14%) 1,000 (86%) 911 (79%) 246 (21%) 3 (0%) 195 (17%) 191 (16%) 774 (67%) 262 (23%) 615 (53%) 283 (24%) 248 (15%) 1,418 (85%) 1,343 (81%) 314 (18%) 9 (1%) 220 (13%) 250 (15%) 1,196 (72%) 370 (22%) 883 (53%) 413 (25%) 274 (54%) 502 (43%) 776 (47%) 180 (36%) 342 (29%) 522 (31%) 972 (58%) 578 (35%) 116 (7%) 266 (53%) 199 (39%) 321 (63%) 159 (31%) 26 (5%) 463 (40%) 158 (14%) 651 (56%) 419 (36%) 90 (8%) 729 (44%) 357 (21%) p-value 0.058 <0.001 <0.001 0.785 <0.001 0.014 0.011 <0.001 <0.001 274 (54%) 788 (68%) 1,062 (64%) <0.001 - - - - 138 (12%) 239 (21%) 433 (37%) 595 (51%) - - - - - - - - Notes 1 Includes source of light i.e. electricity, source of water i.e., improved water, source of clean fuel i.e. LPG/bio-gas and type of toilet facility i.e., own/public flush toilet 2 Three main symptoms are fever, cough, and difficulty breathing 3 Four main behaviors are stay home unless urgent, keep 2m apart from others, wear a mask, and wash hands/use sanitizer https://doi.org/10.1371/journal.pone.0244053.t001 PLOS ONE | https://doi.org/10.1371/journal.pone.0244053 December 17, 2020 5 / 13 PLOS ONE Gender specific variation in COVID-19 knowledge, behavior and health effects among young adults Table 2. Linear probability model of factors associated with knowledge of all 3 main COVID-19 symptoms (fever, cough, difficulty breathing), stratified by gender. VARIABLES Model 1: All Model 2: Men Model 3: Women (1) (2) (3) Women (vs men) Muslim (vs Hindu) Educational attainment (0–7 years = REF) 8–9 years 10+ years Caste (OBC = REF) General Category Scheduled Caste/ Tribe -0.069�� (-0.122 - -0.021) -0.047 NA -0.067 NA -0.049 (-0.109–0.015) (-0.198–0.064) (-0.120–0.023) 0.059 -0.102 0.088 (-0.028–0.146) 0.242�� (-0.336–0.132) 0.136 (-0.006–0.182) 0.250�� (0.171–0.314) (-0.070–0.342) (0.173–0.328) 0.069� (0.010–0.128) 0.022 0.138� (0.026–0.251) 0.044 0.040 (-0.029–0.110) 0.014 (-0.035–0.080) (-0.061–0.149) (-0.056–0.083) Age 20–24 (vs 18–19 years) 0.008 0.020 0.001 (-0.056–0.073) -0.067� (-0.123 - -0.011) 0.111�� (-0.093–0.133) -0.045 (-0.146–0.055) 0.104 (-0.078–0.080) -0.077� (-0.146 - -0.009) 0.109�� (0.049–0.172) (-0.004–0.211) (0.033–0.185) -0.039 0.017 -0.061 (-0.091–0.013) (-0.082–0.116) (-0.123–0.001) 1,666 0.095 506 0.073 1,160 0.092 Rural (vs urban) Household has all 4 amenities Bihar (vs UP) Observations R-squared CI in parentheses �� p<0.01 � p<0.05 https://doi.org/10.1371/journal.pone.0244053.t002 men (53%) knew the main symptoms of COVID-19 and fewer women than men practiced key preventive behaviors such as staying home unless it is urgent and wearing a mask (Table 1). Fewer women reported doing all prevention behaviors (14% vs 39% of men). A greater propor- tion of women respondents reported experience of depressive symptoms. In the full model, women were less likely than men to know COVID-19 symptoms (coeff = -0.069; 95% CI: -0.122 - -0.021) (Table 2). The model was then stratified by gender (men- and women-only models). For the men-only model, there were no key characteristics associated with more or less knowledge of symptoms, except that those in the general caste category were 14 pp more likely to know the symptoms compared with those in the OBC category (coeff = 0.138; 95% CI: 0.026–0.251). In the women-only model, several characteristics were associated with having more knowledge of key symptoms. Women who had completed 10 + years of education were 25 pp more likely to know the symptoms compared with those only having zero to seven years of education (coeff = 0.250; 95% CI: 0.173–0.328); relatedly, women residing in households with key amenities were much more likely to know the symptoms PLOS ONE | https://doi.org/10.1371/journal.pone.0244053 December 17, 2020 6 / 13 PLOS ONE Gender specific variation in COVID-19 knowledge, behavior and health effects among young adults Fig 1. Proportion of respondents that practice all four the key preventive behaviors, by gender and educational attainment. https://doi.org/10.1371/journal.pone.0244053.g001 (coeff = 0.109; 95% CI: 0.033–0.185). Women living in rural areas had lower knowledge of the symptoms. Fig 1 highlights the education and gender differences in reportedly practicing all four main preventive behaviors; this proportion increases across categories of educational attainment for both men and women (Fig 1). Findings also show that women respondents with secondary education (10+ years) were less likely than men respondents with less than primary education (0–7 years) to report practicing all four prevention measures. In the full model exploring characteristics associated with doing all four prevention behav- iors, women were 22 pp less likely than men to report doing all behaviors (coeff = -0.221; 95% CI: -0.263 - -0.180) (Table 3). The full model was re-run stratified by gender. Among men, sev- eral characteristics contributed to reportedly practicing all four prevention behaviors. Men who knew the top three symptoms were more likely to practice the four key preventive behav- iors (coeff = 0.107; 95% CI: 0.020–0.194). Men in rural areas and in Bihar were much less likely to carry out the four behaviors. For the women-only model, the only characteristic that was associated with conducting the four behaviors was knowledge of the three main symptoms (coeff = 0.160; 95% CI: 0.119, 0.201) (Table 3). The last model explored characteristics associated with self-reported experience of depres- sive symptoms. In the full model, women were 5 pp more likely to report that they were experiencing depressive symptoms compared to men (coeff = 0.052; 95% CI: -0.001, 0.104) (Table 4). When stratified by gender, among men only, household loss of employment was the only factor associated with depressive symptoms (coeff = 0.169; 95% CI 0.083, 0.254). Among women only, household loss of employment, religion, and experience of violence were signifi- cantly associated with depressive symptoms. Women belonging to the Muslim religion com- pared to those who identified as Hindu, were more likely to report experience of depressive symptoms (coeff = 0.084; 95% CI:0.012, 0.156). Women who reported violence in the home in the last 15 days were 30 pp more likely to report experience of depressive symptoms (coeff = 0.304; 95% CI: 0.133; 0.475). Women reported whether they had required health services in the previous week, and if so, if they were able to access them (this question was not included for men). Most women had not required health services in the previous week. Of the types of services that were required, nutrition services and child immunization services were the most reported. Among women who sought nutrition services, 51% required but could not access them, 1% required and were able to access them. For child immunization services 37% were unable to access them, none who needed child immunization services could access them. For family planning, 76% stated PLOS ONE | https://doi.org/10.1371/journal.pone.0244053 December 17, 2020 7 / 13 PLOS ONE Gender specific variation in COVID-19 knowledge, behavior and health effects among young adults Table 3. Linear probability model of factors associated with reporting all four main preventive behaviors are being implemented, by gender. VARIABLES Women (vs men) Knowledge of 3 key COVID symptoms (1) Model 1 -0.221�� (-0.263 - -0.180) 0.134�� (2) (3) Model 2: Men Model 3: Women NA 0.107� NA 0.160�� Muslim (vs Hindu) -0.045 -0.090 -0.018 (0.09–0.173) (0.020–0.194) (0.119–0.201) Educational attainment (0–7 years REF) 8–9 years 10+ years Caste (OBC = REF) General Category (-0.096–0.005) (-0.219–0.039) (-0.069–0.033) REF 0.009 REF 0.126 REF -0.008 (-0.062–0.079) (-0.105–0.357) (-0.075–0.059) 0.037 0.187 0.022 (-0.022–0.096) (-0.015–0.390) (-0.033–0.077) REF -0.022 REF -0.106 REF 0.020 (-0.070–0.026) (-0.217–0.004) (-0.029–0.069) Scheduled Caste/Tribe -0.002 0.011 -0.007 (-0.049–0.045) (-0.092–0.115) (-0.056–0.042) -0.004 -0.053 (-0.056–0.049) -0.019�� (-0.074 - -0.017) -0.056�� (-0.165–0.059) -0.147�� (-0.234 - -0.060) -0.103� 0.018 (-0.039–0.074) -0.011 (-0.052–0.029) -0.036 (-0.099 - -0.014) (-0.201 - -0.006) (-0.081–0.008) 1,666 0.128 506 0.059 1,160 0.066 Age group Rural (vs Urban) Bihar (vs UP) Observations R-squared CI in parentheses �� p<0.01 � p<0.05 https://doi.org/10.1371/journal.pone.0244053.t003 they did not require this service in the previous week, of those that did, 21% could not access family planning services (84% of those with a family planning service need) (Fig 2). Discussion Conducted early in the pandemic, our study identifies gender disparities in COVID-19 related knowledge and uptake of promoted preventive behaviors among young people in two states in India. Overall, women were less likely to be able to identify all three of the main COVID-19 symptoms correctly, potentially due to challenges in accessing information or receiving less accurate information of COVID-19 symptoms. Women were also less likely to be practicing the most effective prevention behaviors and they were also more likely to report symptoms of depression. Access to health services is also reportedly affected by the pandemic, with most women in need of services unable to access them, including nutrition, child immunization, family planning and antenatal care services. As of Fall 2020, the pandemic is still not under control globally, and the threat of continued infections remains; therefore, understanding the PLOS ONE | https://doi.org/10.1371/journal.pone.0244053 December 17, 2020 8 / 13 PLOS ONE Gender specific variation in COVID-19 knowledge, behavior and health effects among young adults Table 4. Linear probability model of factors associated with self-reported experience of depressive symptoms dur- ing lockdown, by gender. VARIABLES Women (vs men) Household lost employment Educational attainment (0–7 years REF) 8–9 years 10+ years Age 20–24 (vs 18–19 years) Rural (vs urban) Bihar (vs UP) Muslim (vs Hindu) Under lockdown, experienced any violence in the home in the last 15 days (women only) Observations R-squared CI in parentheses �� p<0.01 � p<0.05 (1) Model 1 0.052� (0.000– 0.104) 0.133�� (0.083– 0.183) REF -0.006 (-0.095– 0.083) 0.018 (-0.055– 0.091) 0.021 (-0.046– 0.088) -0.013 (-0.061– 0.035) 0.038 (-0.015– 0.092) 0.073�� (0.011– 0.135) NA (2) Model 2: Men (3) Model 3: Women NA NA 0.169�� (0.083– 0.254) REF 0.019 (-0.211– 0.250) 0.026 (-0.174– 0.226) -0.001 (-0.113– 0.110) -0.036 (-0.121– 0.049) 0.059 (-0.038– 0.156) 0.041 (-0.085– 0.168) NA 0.117�� (0.055–0.179) REF -0.022 (-0.121–0.076) 0.013 (-0.067–0.092) 0.033 (-0.050–0.115) 0.002 (-0.057–0.060) 0.021 (-0.044–0.086) 0.084�� (0.012–0.156) 0.304�� (0.133–0.475) 1,658 0.027 501 0.038 1,157 0.028 https://doi.org/10.1371/journal.pone.0244053.t004 needs and experiences of adolescents and young adults is critical to offering resources and social support, with attention to gender. Gender differences in accurate knowledge of key COVID-19 symptoms likely reflect young women’s lower levels of educational attainment and lower media exposure, as well as lower access to mobile phones [21,22]. Among women, there was significant variation in the charac- teristics of who had COVID-19 information, such as higher educational attainment, urban res- idence, and higher economic status. These factors likely reflect higher literacy and access to information among some young women. Interestingly, no variation was observed within men, and overall, their knowledge was higher than for women. This finding is supported by available literature on past pandemics. During an outbreak of influenza A (H1N1) in India, a small study found that men had more knowledge of H1N1; this was attributed to men having more PLOS ONE | https://doi.org/10.1371/journal.pone.0244053 December 17, 2020 9 / 13 PLOS ONE Gender specific variation in COVID-19 knowledge, behavior and health effects among young adults Fig 2. Among women, number requiring health services and of those, number unable to obtain them by type of service. https://doi.org/10.1371/journal.pone.0244053.g002 social interactions through employment and having higher literacy rates than women [23]. Higher knowledge among men may be influenced by their greater exposure to risk outside the home for work and socializing shaped by gendered social norms. A recent study from India found differential COVID-19 risk and mortality by gender, reporting that most infections are among men [24]. Our study also suggests that men have higher potential exposure but also higher knowledge of COVID-19 symptoms and prevention; gender dynamics and social norms may increase both knowledge and infection risk among men. Among women, lower adoption of promoted behaviors may also reflect the gender roles and the fact that women spend more time indoors. If women are not going outside, they may not be wearing masks or keeping 2m distance from others because they are not interacting outside the household. Knowledge was the only factor associated adoption of promoted behaviors among women; potentially there are other unmeasured characteristics that are associated with observed varia- tion among women. To bridge this knowledge gender gap, additional research on whether and how the pandemic is reinforcing gender roles may help inform gender sensitive education campaigns via media that women can access and understand even with limited literacy. Mental health and healthcare-seeking behavior for young people are also affected. Our find- ings suggest that loss of employment among household members due to the lockdown was associated with depressive symptoms among both men and women. Approximately 400 mil- lion informal sector workers in India have lost their livelihood due to COVID-19 and related lockdowns [25]; interviews with informal sector workers describe impending poverty, evic- tions and hunger as incomes and work opportunities are sharply curtailed [26]. Previous research has also found a link between loss of employment and SGBV, both of which likely relate to depressive symptoms during lockdown [15,16]. A recent study conducted prior to COVID-19 of mental health in India found being a woman, younger age, loss of employment, and other characteristics were associated with symptoms of depression, anxiety and stress [18]. Many women reported that they had forgone necessary medical services, which may lead to adverse secondary health outcomes and outbreaks of other diseases. Among women surveyed, most of those who did require a health service could not access them. Public transit commonly used to visit clinics was closed during lockdown, which may have affected access [2]. Chal- lenges in accessing health services must be carefully monitored to avoid unintended secondary health crises, including outbreaks of vaccine preventable disease, stunting/undernutrition, and unintended pregnancy or poor birth outcomes [27]. While most women reported they did not require any health services, this study was conducted early in the pandemic. If lockdowns PLOS ONE | https://doi.org/10.1371/journal.pone.0244053 December 17, 2020 10 / 13 PLOS ONE Gender specific variation in COVID-19 knowledge, behavior and health effects among young adults resume or access continues to be disrupted, utilization of essential services should be moni- tored, and steps taken to ensure accessibility. This study has several limitations. First there are inherent challenges in conducting surveys that are not face-to-face; mobile phone-based data collection relies on self-reported informa- tion conveyed by participants who may have challenges understanding questions, and we can- not guarantee protections for participants who may be vulnerable in their households [28]. Secondly, the representativeness of the sample may be compromised as we could only inter- view those with working phone numbers from the 2015–16 UDAYA survey. Our survey respondents had slightly higher educational attainment and household wealth compared to the full UDAYA cohort, suggesting that the most vulnerable from the original sample were not reachable. Third, we asked questions regarding knowledge of COVID-19 prevention behav- iors, then later asked about behaviors respondents were doing. Potentially, question order nudged recall, which could explain why the proportion aware of certain behaviors was lower than those who reported implementing them. However, both the knowledge and behavior questions were based on spontaneous responses, not a list read by the interviewer, so this effect should be minimal. Lastly, our measure of mental health was very simple and self-reported, validated depression measures are necessary but challenging to collect via mobile phone interview. Our findings suggest that early in the pandemic lockdown, there were significant knowl- edge gaps and secondary health effects disproportionately impacting adolescent girls and young women. To increase knowledge of symptoms and preventive behaviors, gender-sensi- tive behavior change campaigns should be developed, and adapted for cultural context, liter- acy, and accessibility. Improved access to information may lead to adoption of promoted behaviors, reducing risk of infection. Relatedly, steps to address mental health and the unin- tended secondary health impacts of the pandemic are required. To date, the Government of India has introduced several initiatives to address these issues, for example activating a toll- free helpline for those requiring psychosocial counseling and issuing guidelines for the sus- tained provision of essential health services. Government agencies are also launching special social protection initiatives. It is critical that these measures reach the most vulnerable popula- tions, including messaging targeted to women. Longer term efforts may also be necessary to address the prolonged and potentially gendered effects of COVID-19 and ensure that health and development gains are not lost due to the pandemic, especially as India’s case load has grown to one of the highest worldwide. Supporting information S1 Table. Differences in key background characteristics between respondents aged 15–19 whose number was not available, who were interviewed in the COVID-19 survey and who were not interviewed in COVID-19 survey. (TIF) S1 File. COVID-19 study questionnaire. (PDF) Acknowledgments The authors would like to acknowledge the dedicated team at Population Council Inc. in India that collected all of these surveys and made this research happen. Author Contributions Conceptualization: Jessie Pinchoff, KG Santhya, Rajib Acharya, Thoai D. Ngo. PLOS ONE | https://doi.org/10.1371/journal.pone.0244053 December 17, 2020 11 / 13 PLOS ONE Gender specific variation in COVID-19 knowledge, behavior and health effects among young adults Data curation: Shilpi Rampal. Formal analysis: Jessie Pinchoff, Shilpi Rampal. Investigation: Rajib Acharya, Thoai D. Ngo. Methodology: KG Santhya, Corinne White, Shilpi Rampal, Rajib Acharya. Project administration: KG Santhya, Corinne White, Rajib Acharya. Resources: KG Santhya. Supervision: Jessie Pinchoff, KG Santhya, Rajib Acharya, Thoai D. Ngo. Writing – original draft: Jessie Pinchoff, Corinne White. Writing – review & editing: KG Santhya, Shilpi Rampal, Rajib Acharya, Thoai D. Ngo. References 1. Lancet The. India under COVID-19 lockdown. The Lancet. 2020; 395: 1315. https://doi.org/10.1016/ S0140-6736(20)30938-7 PMID: 32334687 2. Cash R, Patel V. Has COVID-19 subverted global health? The Lancet. 2020; 395: 1687–1688. https:// doi.org/10.1016/S0140-6736(20)31089-8 PMID: 32539939 3. COVID-19 Map. In: Johns Hopkins Coronavirus Resource Center [Internet]. [cited 20 Jul 2020]. Avail- able: https://coronavirus.jhu.edu/map.html 4. Davies SE, Bennett B. A gendered human rights analysis of Ebola and Zika: locating gender in global health emergencies. Int Aff. 2016; 92: 1041–1060. https://doi.org/10.1111/1468-2346.12704 5. Kapoor M, Agrawal D, Ravi S, Roy A, Subramanian SV, Guleria R. Missing female patients: an observa- tional analysis of sex ratio among outpatients in a referral tertiary care public hospital in India. BMJ Open. 2019; 9: e026850. https://doi.org/10.1136/bmjopen-2018-026850 PMID: 31391189 6. Surveys show that COVID-19 has gendered effects in Asia and the Pacific | UN Women Data Hub. [cited 12 Jun 2020]. Available: https://data.unwomen.org/resources/surveys-show-covid-19-has- gendered-effects-asia-and-pacific 7. India. 27 Nov 2016 [cited 21 Jul 2020]. Available: http://uis.unesco.org/en/country/in 8. UNICEF, editor. Children in a digital world. New York, NY: UNICEF; 2017. 9. GSM Association. Connected Women: The Mobile Gender Gap Report 2020. London, United Kingdom: GSM Association; 2020. Available: https://www.gsma.com/mobilefordevelopment/wp-content/uploads/ 2020/05/GSMA-The-Mobile-Gender-Gap-Report-2020.pdf 10. Prochaska JO, Velicer WF. The transtheoretical model of health behavior change. Am J Health Promot AJHP. 1997; 12: 38–48. https://doi.org/10.4278/0890-1171-12.1.38 PMID: 10170434 11. Parker W. Rethinking conceptual approaches to behaviour change: The importance of context.: 7. 12. Ajzen I. The theory of planned behavior. Organ Behav Hum Decis Process. 1991; 50: 179–211. https:// doi.org/10.1016/0749-5978(91)90020-T 13. Roberton T, Carter ED, Chou VB, Stegmuller AR, Jackson BD, Tam Y, et al. Early estimates of the indi- rect effects of the COVID-19 pandemic on maternal and child mortality in low-income and middle- income countries: a modelling study. Lancet Glob Health. 2020; 8: e901–e908. https://doi.org/10.1016/ S2214-109X(20)30229-1 PMID: 32405459 14. Akseer N, Kandru G, Keats EC, Bhutta ZA. COVID-19 pandemic and mitigation strategies: implications for maternal and child health and nutrition. Am J Clin Nutr. 2020; 112: 251–256. https://doi.org/10.1093/ ajcn/nqaa171 PMID: 32559276 15. Wanqing Z. Domestic Violence Cases Surge During COVID-19 Epidemic. In: Sixth Tone [Internet]. 2 Mar 2020 [cited 24 Jun 2020]. Available: https://www.sixthtone.com/news/1005253/domestic-violence- cases-surge-during-covid19-epidemic 16. Liu N, Zhang F, Wei C, Jia Y, Shang Z, Sun L, et al. Prevalence and predictors of PTSS during COVID- 19 outbreak in China hardest-hit areas: Gender differences matter. Psychiatry Res. 2020; 287: 112921. https://doi.org/10.1016/j.psychres.2020.112921 PMID: 32240896 17. World Health Organization. Mental health in emergencies. 11 Jun 2019 [cited 20 Jul 2020]. Available: https://www.who.int/news-room/fact-sheets/detail/mental-health-in-emergencies PLOS ONE | https://doi.org/10.1371/journal.pone.0244053 December 17, 2020 12 / 13 PLOS ONE Gender specific variation in COVID-19 knowledge, behavior and health effects among young adults 18. Saikarthik J, Saraswathi I, Siva T. Risk factors and protective factors of mental health during COVID-19 outbreak and lockdown in adult Indian population- A cross-sectional study. medRxiv. 2020; 2020.06.13.20130153. https://doi.org/10.1101/2020.06.13.20130153 19. John N, Casey SE, Carino G, McGovern T. Lessons Never Learned: Crisis and gender-based violence. Dev World Bioeth. 2020; 20: 65–68. https://doi.org/10.1111/dewb.12261 PMID: 32267607 20. Organization WH. COVID-19 and violence against women: what the health sector/system can do, 7 April 2020. 2020 [cited 21 Oct 2020]. Available: https://apps.who.int/iris/handle/10665/331699 21. Santhya KG, Acharya R, Pandey N, Singh S, Rampal S, Zavier AJ, et al. Understanding the lives of ado- lescents and young adults (UDAYA) in Bihar, India. Population Council; 2017. https://doi.org/10.1017/ S0021932017000360 PMID: 29160192 22. Santhya KG, Acharya R, Pandey N, Gupta A, Rampal S, Singh S, et al. Understanding the lives of ado- lescents and young adults (UDAYA) in Uttar Pradesh, India (2015–16). Population Council; 2017. https://doi.org/10.1017/S0021932017000360 PMID: 29160192 23. Kamate SK, Agrawal A, Chaudhary H, Singh K, Mishra P, Asawa K. Public knowledge, attitude and behavioural changes in an Indian population during the Influenza A (H1N1) outbreak. J Infect Dev Ctries. 2010; 4: 007–014. https://doi.org/10.3855/jidc.501 PMID: 20130372 24. Joe W, Kumar A, Rajpal S, Mishra US, Subramanian SV. Equal risk, unequal burden? Gender differen- tials in COVID-19 mortality in India. J Glob Health Sci. 2020; 2. https://doi.org/10.35500/jghs.2020.2. e17 25. Kelley M, Ferrand RA, Muraya K, Chigudu S, Molyneux S, Pai M, et al. An appeal for practical social jus- tice in the COVID-19 global response in low-income and middle-income countries. Lancet Glob Health. 2020; 8: E888–E889. https://doi.org/10.1016/S2214-109X(20)30249-7 PMID: 32416766 26. Mobarak AM, Barnett-Howell Z. Poor Countries Need to Think Twice About Social Distancing. In: For- eign Policy [Internet]. [cited 28 May 2020]. Available: https://foreignpolicy.com/2020/04/10/poor- countries-social-distancing-coronavirus/ 27. Ministry of Health and Family Welfare. Guidance Note on Provision of Reproductive, Maternal, New- born, Child, Adolescent Health Plus Nutrition (RMNCAH+N) services during & post COVID-19 Pan- demic. Ministry of Health and Family Welfare; 2020 pp. 1–9. Available: https://www.mohfw.gov.in/pdf/ GuidanceNoteonProvisionofessentialRMNCAHNServices24052020.pdf 28. Undie C-C, Mathur S, Haberland N, Vieitez I, Pulerwitz J. Opportunities for SGBV Data Collection in the Time of COVID-19: The Value of Implementation Science. In: Sexual Violence Research Initiative [Internet]. 26 Jun 2020. Available: https://svri.org/blog/opportunities-sgbv-data-collection-time-covid- 19-value-implementation-science PLOS ONE | https://doi.org/10.1371/journal.pone.0244053 December 17, 2020 13 / 13 PLOS ONE
null
10.1371_journal.pone.0250044.pdf
Data Availability Statement: Data are available here: Wall, Kristin, 2021, "Replication Data for: "Etiologies of genital inflammation and ulceration in symptomatic Rwandan men and women responding to radio promotions of free screening and treatment services"", https://doi.org/10.7910/ DVN/CFX6UU, Harvard Dataverse.
Data are available here: Wall, Kristin,
RESEARCH ARTICLE Etiologies of genital inflammation and ulceration in symptomatic Rwandan men and women responding to radio promotions of free screening and treatment services 1*, Julien Nyombayire2, Rachel Parker1, Rosine Ingabire2, Kristin M. WallID Jean Bizimana2, Jeannine Mukamuyango2, Amelia Mazzei2, Matt A. Price3, Marie Aimee UnyuzimanaID 2, Amanda Tichacek1, Susan Allen1, Etienne Karita2 a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 1 Rwanda Zambia HIV Research Group, Department of Pathology & Laboratory Medicine, School of Medicine and Hubert Department of Global Health and Department of Epidemiology, Rollins School of Public Health, Laney Graduate School, Emory University, Atlanta, Georgia, United States of America, 2 Project San Francisco, Rwanda Zambia HIV Research Group, Kigali, Rwanda, 3 IAVI, NY, NY, University of California San Francisco, San Francisco, CA, United States of America OPEN ACCESS Citation: Wall KM, Nyombayire J, Parker R, Ingabire R, Bizimana J, Mukamuyango J, et al. (2021) Etiologies of genital inflammation and ulceration in symptomatic Rwandan men and women responding to radio promotions of free screening and treatment services. PLoS ONE 16(4): e0250044. https://doi.org/10.1371/journal. pone.0250044 Editor: Antonella Marangoni, Universita degli Studi di Bologna Scuola di Medicina e Chirurgia, ITALY Received: January 13, 2021 Accepted: March 30, 2021 Published: April 20, 2021 Copyright: © 2021 Wall et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: Data are available here: Wall, Kristin, 2021, "Replication Data for: "Etiologies of genital inflammation and ulceration in symptomatic Rwandan men and women responding to radio promotions of free screening and treatment services"", https://doi.org/10.7910/ DVN/CFX6UU, Harvard Dataverse. Funding: This work was funded by the National Institutes of Health (NIH) (NIAID R01 AI51231), the * kmwall@emory.edu Abstract Introduction The longstanding inadequacies of syndromic management for genital ulceration and inflam- mation are well-described. The Rwanda National Guidelines for sexually transmitted infec- tion (STI) syndromic management are not yet informed by the local prevalence and correlates of STI etiologies, a component World Health Organization guidelines stress as critical to optimize locally relevant algorithms. Methods Radio announcements and pharmacists recruited symptomatic patients to seek free STI services in Kigali. Clients who sought services were asked to refer sexual partners and symptomatic friends. Demographic, behavioral risk factor, medical history, and symptom data were collected. Genital exams were performed by trained research nurses and physi- cians. We conducted phlebotomy for rapid HIV and rapid plasma reagin (RPR) serologies and vaginal pool swab for microscopy of wet preparation to diagnose Trichomonas vaginalis (TV), bacterial vaginosis (BV), and vaginal Candida albicans (VCA). GeneXpert testing for Neisseria gonorrhoeae (NG) and Chlamydia trachomatis (CT) were conducted. Here we assess factors associated with diagnosis of NG and CT in men and women. We also explore factors associated with TV, BV and VCA in women. Finally, we describe genital ulcer and RPR results by HIV status, gender, and circumcision in men. Results Among 974 men (with 1013 visits), 20% were positive for CT and 74% were positive for NG. Among 569 women (with 579 visits), 17% were positive for CT and 27% were positive for NG. In multivariate analyses, factors associated with CT in men included younger age, PLOS ONE | https://doi.org/10.1371/journal.pone.0250044 April 20, 2021 1 / 21 PLOS ONE NIH AIDS International Training and Research Program Fogarty International Center (D43 TW001042); and the NIH-funded Emory Center for AIDS Research (P30 AI050409). This work was partially funded by IAVI with the generous support of USAID and other donors; a full list of IAVI donors is available at https://www.iavi.org. The contents of this manuscript are the responsibility of the authors and do not necessarily reflect the views of USAID or the US Government. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: No authors have competing interests. Etiologies of genital abnormalities in Rwandan men and women responding to radio advertisements, <17 days since suspected exposure, and not having dysuria. Factors associated with NG in men included not having higher education or full- time employment, <17 days since suspected exposure, not reporting a genital ulcer, and having urethral discharge on physical exam. Factors associated with CT in women included younger age and < = 10 days with symptoms. Factors associated with NG in women included younger age, lower education and lack of full-time employment, sometimes using condoms vs. never, using hormonal vs. non-hormonal contraception, not having genital ulcer or itching, having symptoms < = 10 days, HIV+ status, having BV, endocervical dis- charge noted on speculum exam, and negative vaginal wet mount for VCA. In multivariate analyses, only reporting >1 partner was associated with BV; being single and RPR+ was associated with TV; and having < = 1 partner in the last month, being pregnant, genital itch- ing, discharge, and being HIV and RPR negative were associated with VCA. Genital ulcers and positive RPR were associated with being HIV+ and lack of circumcision among men. HIV+ women were more likely to be RPR+. In HIV+ men and women, ulcers were more likely to be herpetic rather than syphilitic compared with their HIV- counterparts. Conclusions Syndromic management guidelines in Rwanda can be improved with consideration of the prevalence of confirmed infections from this study of symptomatic men and women repre- sentative of those who would seek care at government health centers. Inclusion of demo- graphic and risk factor measures shown to be predictive of STI and non-STI dysbioses may also increase diagnostic accuracy. Introduction Globally, over 1 million new sexually transmitted infections (STI) occur each day [1]. The prevalence of STI increased an estimated 59% in sub Saharan Africa between 1999 and 2005 and has continued to rise [2]. The World Health Organization (WHO) 2016–2021 Global Health Sector Strategy on Sexually Transmitted Infections aims to reduce STI 90% by 2030 using “[epidemiologic] information for focused action” [3]. The association between genital ulceration and inflammation (GUI) due to STI and non- STI etiologies and heterosexual HIV transmission and acquisition has been extensively studied in Africa [4–12]. Broadly, in observational studies GUI is associated with both transmitting and acquiring HIV in both men and women, and with transmission of more than one virion, an otherwise rare event, in cohabiting heterosexual discordant couples which comprise one of the largest HIV risk groups [6, 13–17]. Ulcerative STI that may facilitate HIV transmission include syphilis (Treponema pallidum, TP), Herpes simplex virus (HSV), and chancroid (Haemophilus ducreyi, HD) [18–20]. Inflam- matory STI that increase HIV transmission include gonorrhea (Neisseria gonorrhoeae, NG), chlamydia (Chlamydia trachomatis, CT), and Trichomonas vaginalis (TV) [21–24]. Common non-STI dysbioses associated with genital inflammation include bacterial vaginosis (BV) and vaginal Candida albicans (VCA) [25–29]. Untreated TP, HD, HSV, NG, CT and TV can cause severe morbidity and, along with BV and VCA (which are troublesome but non-invasive), can contribute to HIV transmission. In our studies in African HIV discordant heterosexual couples, GUI contribute a substantial pop- ulation attributable fraction of HIV transmission in both donor and recipient [15]. PLOS ONE | https://doi.org/10.1371/journal.pone.0250044 April 20, 2021 2 / 21 PLOS ONE Etiologies of genital abnormalities in Rwandan men and women The longstanding inadequacies of syndromic management for GUI are well-described [30– 37] but this approach remains the default in many resource-limited settings in Africa due to the high cost of molecular and culture-based diagnostics. The Rwanda National Guidelines for HIV and STI syndromic management were last updated in 2019 but these guidelines are not yet informed by the local prevalence and correlates of STI etiologies, a component WHO guidelines stress as critical to optimize locally relevant algorithms. We have previously pub- lished results of a survey of GUI among Female Sex Workers (FSW) in Kigali, but that study lacked molecular diagnostics for NG and CT [38]. Here we contribute to the epidemiologic data needed to inform improved diagnostic and treatment algorithms in Rwanda by exploring demographic, behavioral, medical history, symptom, genital exam, and laboratory factors associated with molecular diagnosis of NG and CT in men and women. We also explore factors associated with vaginal pathogens TV, BV and VCA in women. Finally, we describe genital ulcer and rapid plasma reagin (RPR) results strati- fied by gender, HIV status, and among men, by male circumcision status. Methods Ethics This program was approved as non-research by the Rwandan National Ethics Committee. This program was determined to be non-research by the Emory Institutional Review Board criteria. Diagnostic and treatment were provided anonymously as free services. Setting Kigali, the capital of Rwanda, has a population of over 1 million people and an adult HIV prev- alence of 4.3% [39]. Between January 2016 and August 2019, The Center for Family Health Research (CFHR), a research site established in Kigali in 1986 and affiliated with Emory Uni- versity in Atlanta, GA, USA, implemented a program for diagnosis and treatment of symptom- atic GUI residents of Kigali. CFHR has worked closely with the Rwanda Ministry of Health (MoH) on research for improved HIV and reproductive health care in government-run health centers for many years [25, 40–43]. Patient recruitment Patients were residents of Kigali, Rwanda and were recruited in three ways: radio announce- ments, partner/friend referral, and pharmacist referral. Radio announcements were made in Kinyarwanda, Rwanda’s vernacular, encouraging men and women with symptoms suggestive of GUI (e.g., genital discharge, discomfort, ulcer) to seek free services at CFHR clinic and were broadcast throughout Kigali. Clients who sought services were then asked to refer sexual part- ners and symptomatic friends. Local pharmacists were alerted to the program and asked to refer individuals seeking treatments for suggestive symptoms. There were no inclusion/exclu- sion criteria applied to participant recruitment. Participants are representative of residents of Kigali with genital symptoms who self-selected to receive care. Data collection and diagnostic procedures Demographics, behavioral risk factors, medical histories, and symptoms were collected using a standard instrument (S1 Fig). This information was obtained during interviews conducted by nurses who recorded data on paper and entered it into MS Access. Similarly, findings from genital exams performed by trained physicians and nurses were recorded on paper and entered into MS Access. Samples for laboratory testing were taken from all patients and included PLOS ONE | https://doi.org/10.1371/journal.pone.0250044 April 20, 2021 3 / 21 PLOS ONE Etiologies of genital abnormalities in Rwandan men and women phlebotomy for rapid HIV and RPR serologies and vaginal pool swab for microscopy of wet preparation to diagnose TV, BV and VCA. GeneXpert testing for NG and CT (Cepheid, Sun- nyvale USA) was conducted for all patients using endocervical swabs obtained from women and either urethral swabs (when discharge was reported or noted on physical exam) or urine samples from men. In collaboration with the MoH, CFHR developed a uniform alphanumeric identifier to allow anonymous data recording. Data analysis Analyses were conducted with Statistical Analysis Software (SAS, Cary, NC). Frequencies of single and multiple infections were stratified by gender and HIV status. Demographic, behavioral, medi- cal history, physical exam, microscopy and serology results were tabulated by gender and by NG and CT results. Bivariate and multivariate analyses of factors associated with NG or CT are pre- sented in tables. Multivariable logistic regression models included variables associated with each outcome at p<0.05 in bivariate analysis and then backward selection was applied. Prevalence odds ratios (crude and adjusted, cPOR and aPOR, respectively) and 95% confidence intervals (CIs) and 2-sided p-values are presented. Variable multi-collinearity was assessed. Repeated visits by STI clients with new complaints were accounted for using the GENMOD procedure. Bivariate and multivariate factors associated with vaginal pathogens TV, BV and VCA in women were analyzed in analogous fashion with results summarized in text. Demographic, behavioral, medical history, and HIV and RPR serology results were considered for model inclusion. Finally, genital ulcer and RPR results were described by gender, HIV status, and among men, by male circumcision status. Results Unless specified in text, p-values are <0.05 for comparisons with details presented in Tables. Summary of GUI diagnosed in men and women (Table 1) GeneXpert for NG and CT were provided to men during 1013 visits (974 unique men) between March 2017 and February 2019. Men tested HIV+ during 5% of these visits. Preva- lence of NG was 74% and prevalence of CT was 20%, with no differences by HIV status. In the 975 visits with RPR results, TP prevalence was significantly higher among HIV+ (13%) com- pared with HIV- (5%) men. Nineteen percent of visits were negative for all pathogens, and 17% of visits had more than one infection identified. GeneXpert for NG and CT were provided to women during 579 visits (569 unique women) between March 2017 and February 2019. Women tested HIV+ during 13% of these visits. Prevalence of NG was 26% and prevalence of CT was 17%, with higher prevalence of NG among HIV+ women. The prevalence of TV (overall 13%) was higher in HIV+ women, whereas the prevalence of VCA (overall 21%) was higher in HIV- women. In the 568 visits with RPR results, TP prevalence was significantly higher among HIV+ (22%) compared with HIV- (6%) women and having multiple pathogens identified was more prevalent among HIV + (36%) compared with HIV- (24%) women’s visits. Conversely, having no pathogen identi- fied was more prevalent in HIV- (31%) versus HIV+ (18%) women’s visits. Demographics and factors associated with CT and NG in men (Tables 2 and 3) Men averaged 30.8 years of age, 77% were single, 64% had at least a secondary education, 55% were employed full time, 22% reported more than one partner in the last 30 days and 57% reported never using condoms in the past three months. The most common symptoms PLOS ONE | https://doi.org/10.1371/journal.pone.0250044 April 20, 2021 4 / 21 PLOS ONE Etiologies of genital abnormalities in Rwandan men and women Table 1. Distribution of pathogens identified in symptomatic men and women in Kigali, Rwanda. Total HIV+ (N = 54) HIV- (N = 958) p-value Among all men (N = 1013 visits)� None identified CT NG CT and NG Among men with RPR results (N = 975 visits) None identified TP Any multiple infection Among all women (N = 579 visits) None identified CT NG CT and NG BV TV VCA Among women with RPR results (N = 568 visits) None identified TP Any multiple infection N 196 204 751 138 184 52 164 N 176 98 152 45 113 72 118 169 46 146 Col % 19% 20% 74% 14% 19% 5% 17% N 14 7 40 7 14 7 11 Col % 26% 13% 74% 13% 26% 13% 21% N 182 196 711 131 170 45 153 Col % 19% 20% 74% 14% 18% 5% 17% Total HIV+ (N = 75) HIV- (N = 504) Col % 30% 17% 26% 8% 21% 13% 21% 30% 8% 26% N 13 8 34 5 20 15 6 13 16 26 Col % 17% 11% 45% 7% 28% 20% 8% 18% 22% 36% N 163 90 118 40 93 57 112 156 30 120 Col % 32% 18% 23% 8% 19% 12% 23% 31% 6% 24% 0.210 0.181 0.981 0.882 0.150 0.019 0.433 0.008 0.121 <0.0001 0.702 0.087 0.039 0.004 0.020 <0.0001 0.031 TP: Treponema pallidum, NG: Neisseria gonorrhoeae, CT: Chlamydia trachomatis, TV: Trichomonas vaginalis, BV: bacterial vaginosis, VCA: vaginal Candida albicans; RPR: rapid plasma regain �One man missing HIV status https://doi.org/10.1371/journal.pone.0250044.t001 reported were urethral discharge (89%) and dysuria (80%). Physical findings included urethral discharge in 91% and genital ulcer in 5% of men (Table 2). Multivariate analyses (Table 3) showed younger age, responding to radio advertisements, <17 days since suspected exposure, and not having dysuria as independent factors associated with CT. Multivariate analyses (Table 3) showed not having higher education or full-time employ- ment, <17 days since suspected exposure, not reporting a genital ulcer, and urethral discharge on physical exam as independent factors associated with NG. HIV, RPR serologic results, and circumcision status were not associated with either CT or NG. Demographics and factors associated with CT and NG in women (Tables 4 and 5) The mean age women was 28.7, they had 1.3 children and desired 1.4 more on average, 54% were single, 53% had a secondary education or more, 34% had full-time employment, 83% reported < = 1 partner in the last 30 days and 63% reported never using condoms in the past three months. Vaginal discharge was the most common presenting symptom (82%) and endo- cervical inflammation or discharge was noted on 49% of speculum exams. (Table 4) Multivariate analyses (Table 5) showed younger age and having symptoms < = 10 days as independent factors associated with CT. PLOS ONE | https://doi.org/10.1371/journal.pone.0250044 April 20, 2021 5 / 21 PLOS ONE Etiologies of genital abnormalities in Rwandan men and women Table 2. Factors associated with CT or NG infection in men in Kigali, Rwanda (N = 1013). Demographics Age, continuous (years) Referrer Radio Advert Friends/Walk-in/Pharmacy/Contact Partner/ Internet Living and Marital Status Married and Cohabiting Single or Divorced/Separated/Widow Education Level None Primary Secondary Higher Employment Status Full-time employment Part-time/Student/Jobless Sexual behaviors Number of partners in last 30 days None or one partner More than one partner Condom use during vaginal sex in the last three months No partners or always used condoms Sometimes Never Number of days since sexual contact you suspect STI was acquired from < = 8 9–16 > = 17 Self-reported symptoms Urethral discharge Yes No Dysuria Yes No Genital itching Yes No Genital ulcer Yes No Total (N = 1013) CT-infected (N = 204) CT-uninfected (N = 809) p- value NG-infected (N = 751) NG-uninfected (N = 262) p-value n /mean 30.8 Col% /SD 7.1 n /mean 29.4 Row% /SD 5.6 n /mean 31.1 Row% /SD n /mean Row% /SD 7.4 0.001 30.5 7.0 n /mean 31.6 Row% /SD 7.3 0.029 688 325 232 781 25 339 454 193 552 459 704 203 27 363 517 331 288 323 895 114 810 199 67 854 41 878 68% 32% 23% 77% 2% 34% 45% 19% 55% 45% 78% 22% 3% 40% 57% 35% 31% 34% 89% 11% 80% 20% 7% 93% 4% 96% 151 53 33 171 1 66 89 47 122 82 138 45 4 69 110 83 58 50 188 16 153 51 14 171 6 183 22% 16% 14% 22% 4% 19% 20% 24% 22% 18% 20% 22% 15% 19% 21% 25% 20% 15% 21% 14% 19% 26% 21% 20% 15% 21% 537 272 199 610 24 273 365 146 430 377 566 158 23 294 407 248 230 273 707 98 657 148 53 683 35 695 0.037 0.011 0.095 0.095 0.422 0.555 0.010 0.081 0.034 0.864 0.336 78% 84% 86% 78% 96% 81% 80% 76% 78% 82% 80% 78% 85% 81% 79% 75% 80% 85% 79% 86% 81% 74% 79% 80% 85% 79% 488 263 156 595 16 267 343 123 392 357 518 163 14 279 387 292 235 177 717 32 599 150 39 649 13 681 71% 81% 67% 76% 64% 79% 76% 64% 71% 78% 74% 80% 52% 77% 75% 88% 82% 55% 80% 28% 74% 75% 58% 76% 32% 78% 200 62 76 186 9 72 111 70 160 102 186 40 13 84 130 39 53 146 178 82 211 49 28 205 28 197 0.001 0.006 0.001 0.015 0.051 0.015 <0.0001 <0.0001 0.680 0.001 <0.0001 29% 19% 33% 24% 36% 21% 24% 36% 29% 22% 26% 20% 48% 23% 25% 12% 18% 45% 20% 72% 26% 25% 42% 24% 68% 22% Number of days with symptoms 1–5 385 41% 100 26% 285 74% 0.004 332 86% 53 14% <0.0001 PLOS ONE | https://doi.org/10.1371/journal.pone.0250044 April 20, 2021 (Continued ) 6 / 21 PLOS ONE Etiologies of genital abnormalities in Rwandan men and women Table 2. (Continued) Total (N = 1013) CT-infected (N = 204) CT-uninfected (N = 809) p- value NG-infected (N = 751) NG-uninfected (N = 262) p-value Demographics n /mean Col% /SD n /mean Row% /SD n /mean Row% /SD n /mean Row% /SD n /mean Row% /SD 6–10 11–21 >21 Laboratory and physical exam HIV Status Positive Negative RPR Result Positive Negative Urethral discharge Yes No Genital ulcer Yes No Circumcision status Circumcised Uncircumcised 254 192 105 54 958 52 923 858 87 46 898 524 259 27% 21% 11% 5% 95% 5% 95% 91% 9% 5% 95% 67% 33% 40 33 17 7 196 13 182 178 13 8 183 122 45 16% 17% 16% 13% 20% 25% 20% 21% 15% 17% 20% 23% 17% 214 159 88 47 762 39 741 680 74 38 715 402 214 84% 83% 84% 87% 80% 75% 80% 79% 85% 83% 80% 77% 83% 0.181 0.354 0.199 0.623 0.058 191 121 56 40 711 43 677 692 15 17 686 416 195 75% 63% 53% 74% 74% 83% 73% 81% 17% 37% 76% 79% 75% 63 71 49 14 247 9 246 166 72 29 212 108 64 25% 37% 47% 26% 26% 17% 27% 19% 83% 63% 24% 21% 25% 0.981 0.136 <0.0001 <0.0001 0.192 Not significant not shown include: Self-reported symptoms dyspareunia, unpleasant odor, abdominal pain, anal discharge, anal ulcer, anal warts, and sore throat; genital exam results white accumulation, condyloma/warts, inguinal adenopathy >1cm unilateral and bilateral, inflammation, and testicular mass/tenderness RPR: Rapid plasma reagin; STI: Sexually transmitted disease; NG: Neisseria gonorrhoeae, CT: Chlamydia trachomatis https://doi.org/10.1371/journal.pone.0250044.t002 Multivariate analyses (Table 5) showed younger age, lower education and lack of full-time employment, sometimes using condoms vs. never, using hormonal contraception vs. other or no contraception, not having a genital ulcer or itching, having symptoms for < = 10 days, HIV + status, endocervical discharge noted on speculum exam, BV, and negative VCA as indepen- dent factors associated with NG. Factors associated with of BV, TV and VCA in women (not tabled) Only reporting >1 partner remained independently associated with BV in multivariate analy- ses (POR 2.21, p = 0.003). Factors associated with TV in multivariate analyses were being sin- gle and RPR+ (aPOR 2.05, p = 0.009 and aPOR 2.37, p = 0.023, respectively). Factors associated with VCA were having < = 1 partner in the last month (aPOR 4.26, p = 0.005), being pregnant (aPOR 3.05, p = 0.002), always using condoms or not having sex in the last three months vs. never using condoms (aPOR 2.42, p = 0.023), genital itching (aPOR 1.69, p = 0.034), genital discharge (aPOR 2.56, p = 0.011), and being HIV and RPR negative (aPOR 2.93, p = 0.025 and aPOR 4.94, p = 0.031, respectively). Genital ulcers in men and women (not tabled) Reported and/or observed genital ulcers were more common among HIV+ (20%) compared with HIV- (5%) men (p<0.001). Genital ulcers were noted during physical examination in PLOS ONE | https://doi.org/10.1371/journal.pone.0250044 April 20, 2021 7 / 21 PLOS ONE Etiologies of genital abnormalities in Rwandan men and women Table 3. Univariate and multivariate analysis of factors associated with CT or NG infection in men in Kigali, Rwanda (N = 1013). Demographics cPOR 95% CI CT infection aPOR 95% CI p- value p- value cPOR 95% CI p-value aPOR 95% CI p-value NG infection Age (per year increase) 0.96 0.94 0.99 0.001 0.96 0.94 0.98 0.001 0.98 0.96 1.00 0.029 Referrer Radio Advert 1.44 1.02 2.04 0.038 1.44 1.01 2.07 0.046 ref --- --- --- Friends/Walk-in/Pharmacy/Contact ref ref 1.76 1.28 2.43 0.001 Partner/Internet Living and Marital Status Married and Cohabiting ref Single or Divorced/Separated/Widow 1.69 1.13 2.54 0.011 Education Level ref --- --- --- 1.56 1.14 2.15 0.006 None/Primary/Secondary ref 1.84 1.32 2.58 0.000 2.39 1.57 3.63 <0.0001 Higher Employment Status Full-time employment Part-time/Student/Jobless Sexual behaviors Number of partners in last 30 days None or one partner More than one partner Condom use during vaginal sex in the last 3 months No partners or always used condoms Sometimes Never Number of days since sexual contact you suspect STI was acquired from 0–16 > = 17 Self-reported symptoms Urethral discharge Yes No Dysuria Yes No Genital itching Yes No Genital ulcer Yes No 1.37 0.95 1.99 0.092 ref --- --- --- ref --- --- --- 1.30 0.96 1.78 0.094 ref --- --- --- ref --- --- --- ref ref 1.17 0.80 1.71 0.424 0.22 1.90 0.426 0.62 1.21 0.411 0.64 0.87 ref 1.45 1.09 1.92 0.011 1.51 1.05 2.17 0.028 ref 1.5 --- 1.02 --- 2.2 --- 0.040 0.37 1.13 ref 0.17 0.83 --- 0.8 1.54 --- 0.012 0.450 --- 1.61 1.13 2.30 0.009 1.64 1.15 2.35 0.007 4.68 3.43 3.37 <0.0001 3.29 2.30 4.7 <0.0001 ref 1.63 0.94 2.83 0.084 ref ref ref ref ref --- --- --- ref --- --- --- 10.00 6.41 15.61 <0.0001 ref --- --- --- ref --- --- --- 1.48 1.03 2.13 0.034 1.51 1.03 2.22 0.035 1.05 0.74 1.49 0.792 1.05 0.57 1.93 0.872 ref ref ref --- --- --- 2.24 1.35 3.72 0.002 ref --- --- --- ref --- --- --- 1.53 0.63 3.70 0.345 7.50 3.79 14.85 <0.0001 4.50 2.22 9.13 <0.0001 Number of days with symptoms 1–10 > = 11 Laboratory and physical exam HIV Status Positive 1.39 0.97 1.98 0.075 ref 3.06 2.26 4.15 <0.0001 ref --- --- --- 0.58 0.26 1.31 0.189 0.99 0.53 1.86 0.980 PLOS ONE | https://doi.org/10.1371/journal.pone.0250044 April 20, 2021 (Continued ) 8 / 21 PLOS ONE Etiologies of genital abnormalities in Rwandan men and women Table 3. (Continued) Demographics cPOR 95% CI CT infection aPOR 95% CI p- value p- value cPOR 95% CI p-value aPOR 95% CI p-value NG infection Negative RPR Result Positive Negative Urethral discharge Yes No Genital ulcer Yes No ref ref --- --- --- 1.30 0.68 2.52 0.429 ref 1.65 ref 0.82 --- 3.3 --- 0.158 --- 1.49 0.81 2.75 0.204 19.94 11.12 35.76 <0.0001 16.38 7.28 36.89 <0.0001 ref ref --- --- --- ref --- --- --- 0.82 0.38 1.80 0.626 ref ref --- --- --- 5.52 2.96 10.28 <0.0001 aPOR: Adjusted prevalence odds ratio; cPOR: Crude prevalence odds ratio; RPR: Rapid plasma reagin; CI: Confidence interval; STI: Sexually transmitted disease; NG: Neisseria gonorrhoeae, CT: Chlamydia trachomatis Not significant not shown include: Self-reported symptoms dyspareunia, unpleasant odor, abdominal pain, anal discharge, anal ulcer, anal warts, and sore throat; genital exam results white accumulation, condyloma/warts, inguinal adenopathy >1cm https://doi.org/10.1371/journal.pone.0250044.t003 19% of RPR+ and 4% of RPR- men and conversely 20% of men with ulcers were RPR+ com- pared to 4% of men without ulcers (p<0.001). Among HIV+ men, none of the seven who were RPR+ had reported and/or observed ulcers while 23% of 43 HIV+ RPR- men had ulcers (p = 0.319). In contrast, among HIV- RPR+ men 21% had reported or observed ulcers com- pared to only 4% of HIV-RPR- men (p<0.001). This suggests that ulcers among HIV+ men were more likely herpetic while among HIV- men at least one fifth were syphilitic. Although HIV- men were more likely to be circumcised than HIV+ men (67% vs. 58%) in our program, this difference was not significant (p = 0.196). Among circumcised men, those who were HIV+ were more likely to have ulcers (13% vs. 4%, p = 0.074) and to be RPR+ (20% vs. 4%, p = 0.003). Among uncircumcised men, those who were HIV+ were also more likely to have ulcers (27% vs. 7%, p = 0.001) while the difference in RPR+ results was not significant (12% vs. 6%, p = 0.324). Among women, the prevalence of reported or observed ulcers was not significantly differ- ent by HIV serostatus (20% in HIV+ vs.14% p = 0.196). Genital ulcers were noted during phys- ical examination for 28% of RPR+ women compared with 14% of RPR- women (p<0.001). As with men, the association between RPR results and reported and/or observed ulcers differed in HIV+ and HIV- women: 25% of HIV+RPR+ vs. 20% of HIV+RPR- had ulcers, p = 0.729, com- pared with 37% of HIV-RPR+ vs. 13% of HIV-RPR- women having ulcers (p = 0.001). Discussion We found a high prevalence of NG and CT among symptomatic men and women in Kigali. Among men, urethral discharge was strongly associated with a diagnosis of NG while dysuria was not associated with either infection. Specific symptoms were less helpful in identifying NG and CT among women. Physical exam findings, demographic variables and reported risk behaviors were independently predictive of NG and/or CT in both men and women, as were vaginal wet mount findings and HIV serologies among women. Among women, TV and BV were associated with sexual risk behaviors but not with symptoms while VCA was associated with vaginal itching and discharge and with low-risk profiles. There were complex inter- PLOS ONE | https://doi.org/10.1371/journal.pone.0250044 April 20, 2021 9 / 21 PLOS ONE Etiologies of genital abnormalities in Rwandan men and women Table 4. Factors associated with CT or NG infection in women in Kigali, Rwanda (N = 579). Demographics n /mean Col %/SD n /mean Row %/SD n /mean Row% /SD n /mean Row %/SD n /mean Row %/SD Age, continuous (years) 28.7 7.2 25.6 6.1 29.3 7.2 <0.0001 26.8 6.3 29.4 7.4 <0.0001 Total (N = 579) CT-infected (N = 98) CT-uninfected (N = 481) p-value NG-infected (N = 152) NG-uninfected (N = 427) p-value Referrer Radio Advert Friends/Walk-in/Pharmacy/Contact Partner/ Internet Living and Marital Status Married and Cohabiting Single or Divorced/Separated/Widow Education Level None Primary Secondary Higher Employment Status Full-time employment Part-time/Student/Jobless Sexual behaviors Number of partners in last 30 days None or one partner More than one partner Condom use during vaginal sex in the last 3 months No partners or always used condoms Sometimes Never Number of days since sexual contact you suspect STI was acquired from < = 8 9–16 > = 17 Number of children under 18, continuous Number of additional children desired, continuous Pregnant Yes No Want more children in next two years Yes No Family planning method among women not pregnant and do not want more children in next two years 284 295 268 311 25 242 246 66 199 379 444 88 35 163 334 46 78 409 1.3 1.4 48 528 125 419 49% 51% 46% 54% 4% 42% 42% 11% 34% 66% 83% 17% 7% 31% 63% 9% 15% 77% 1.3 1.1 8% 92% 23% 77% 37 61 34 64 2 38 46 12 30 68 70 20 6 34 50 8 20 61 1.0 1.6 8 90 20 74 13% 21% 13% 21% 8% 16% 19% 18% 15% 18% 16% 23% 17% 21% 15% 17% 26% 15% 1.1 1.1 17% 17% 16% 18% 247 234 234 247 23 204 200 54 169 311 374 68 29 129 284 38 58 348 1.3 1.4 40 438 105 345 87% 79% 87% 79% 92% 84% 81% 82% 85% 82% 84% 77% 83% 79% 85% 83% 74% 85% 1.3 1.2 83% 83% 84% 82% Non-Hormonal Method (IUD/Condom/Tubal 268 66% 47 18% 221 82% 0.498 Ligation/Natural Method) or No Method Hormonal Implant Injectable 50 48 12% 12% 9 5 18% 10% 41 43 82% 90% 0.014 0.012 0.513 67 85 60 92 9 81 54 8 0.383 39 113 0.112 0.259 0.066 0.026 0.040 0.947 0.666 95 43 4 66 68 15 31 92 1.2 1.3 11 141 33 106 56 24 16 24% 29% 22% 30% 36% 33% 22% 12% 20% 30% 21% 49% 11% 40% 20% 33% 40% 22% 1.1 1.0 23% 27% 27% 26% 217 210 208 219 16 161 192 58 160 266 349 45 31 97 266 31 47 317 1.3 1.4 37 387 88 298 0.153 0.050 0.001 0.008 76% 71% 78% 70% 64% 67% 78% 88% 80% 70% 79% <0.0001 51% 89% <0.0001 60% 80% 67% 60% 78% 1.3 1.2 77% 73% 73% 74% 0.003 0.600 0.279 0.569 0.821 21% 212 79% 0.001 48% 33% 26 32 52% 67% (Continued ) PLOS ONE | https://doi.org/10.1371/journal.pone.0250044 April 20, 2021 10 / 21 PLOS ONE Etiologies of genital abnormalities in Rwandan men and women Table 4. (Continued) Demographics Pills Family planning method and pregnancy composite Pregnant Hormonal method (implant, injectable, pills) Non-Hormonal (IUD/Condom/ Tubal Ligation/ Natural Method) or No Method Self-reported symptoms Vaginal discharge Yes No Genital itching Yes No Dysuria Yes No Genital ulcer Yes No Number of days with symptoms 1–5 6–10 11–21 >21 Laboratory and physical exam HIV Status Positive Negative RPR Result Positive Negative Trichomonas Positive Negative Candida Positive Negative Bacterial vaginosis Positive Negative Vaginal Inflammation or Discharge Yes No Endocervical Inflammation or Discharge Total (N = 579) CT-infected (N = 98) n /mean 40 Col %/SD 10% n /mean 9 Row %/SD 23% CT-uninfected (N = 481) n /mean 31 Row% /SD 78% p-value NG-infected (N = 152) NG-uninfected (N = 427) p-value n /mean 12 Row %/SD 30% n /mean 28 Row %/SD 70% 48 139 388 475 101 320 254 266 311 64 508 72 77 131 257 75 504 46 522 72 491 118 437 113 438 469 69 8% 24% 67% 82% 18% 56% 44% 46% 54% 11% 89% 13% 14% 24% 48% 13% 87% 8% 92% 13% 87% 21% 79% 21% 79% 87% 13% 8 23 67 78 20 52 44 44 54 9 89 16 17 18 37 8 90 10 88 18 75 12 80 25 65 75 15 17% 17% 17% 16% 20% 16% 17% 17% 17% 14% 18% 22% 22% 14% 14% 11% 18% 22% 17% 25% 15% 10% 18% 22% 15% 16% 22% 40 116 321 397 81 268 210 222 257 55 419 56 60 113 220 67 414 36 434 54 416 106 357 88 373 394 54 0.979 0.412 0.732 0.793 0.489 0.170 0.121 0.401 0.038 0.035 0.062 0.232 83% 83% 83% 84% 80% 84% 83% 83% 83% 86% 82% 78% 78% 86% 86% 89% 82% 78% 83% 75% 85% 90% 82% 78% 85% 84% 78% 11 52 89 123 28 57 92 75 76 9 140 24 27 40 48 34 118 21 128 18 129 13 132 47 96 116 24 23% 37% 23% 26% 28% 18% 36% 28% 24% 14% 28% 33% 35% 31% 19% 45% 23% 46% 25% 25% 26% 11% 30% 42% 22% 25% 35% 37 87 299 352 73 263 162 191 235 55 368 48 50 91 209 41 386 25 394 54 362 105 305 66 342 353 45 0.003 0.704 77% 63% 77% 74% 72% 82% <0.0001 0.306 0.020 0.003 64% 72% 76% 86% 72% 67% 65% 69% 81% 55% <0.0001 77% 54% 75% 75% 74% 0.002 0.818 89% <0.0001 70% 58% <0.0001 78% 75% 65% 0.076 (Continued ) PLOS ONE | https://doi.org/10.1371/journal.pone.0250044 April 20, 2021 11 / 21 PLOS ONE Etiologies of genital abnormalities in Rwandan men and women Table 4. (Continued) Demographics Yes No Genital Ulcer Yes No Total (N = 579) CT-infected (N = 98) n /mean Col %/SD n /mean 262 275 48 481 49% 51% 9% 91% 55 35 9 78 Row %/SD 21% 13% 19% 16% CT-uninfected (N = 481) n /mean Row% /SD 207 240 39 403 79% 87% 81% 84% p-value NG-infected (N = 152) NG-uninfected (N = 427) p-value 0.010 0.652 n /mean 88 52 12 126 Row %/SD 34% 19% 25% 26% n /mean 174 223 36 355 Row %/SD 66% 81% 75% 74% <0.001 0.857 Not significant not shown include: Self-reported symptoms anal discharge, anal ulcer, anal warts, and sore throat; genital exam results non-menstrual bleeding (cervix and vagina), condyloma/warts (cervix and vagina), inguinal adenopathy >1cm unilateral and bilateral, adnexal tenderness and adnexal mass. IUD: intrauterine device; RPR: Rapid plasma reagin; CI: Confidence interval; STI: Sexually transmitted disease; NG: Neisseria gonorrhoeae, CT: Chlamydia trachomatis https://doi.org/10.1371/journal.pone.0250044.t004 relationships between HIV and RPR serologies and genital ulcers, and these were further influ- enced by circumcision status among men. These findings exemplify the locally relevant data that can inform approaches to diagnosis and treatment in Rwanda as called for by WHO. Our models had good discrimination and use of these data may offer improvement over the current algorithm recommended by the Rwandan National Guidelines. As in other studies, syndromic management may perform better among men compared to women due to the ease of detecting abnormalities on external genitalia and the high likelihood of NG among men reporting urethral discharge [44]. Surprisingly, dysuria was as common as discharge in men but contrary to conventional wisdom we did not find an association between dysuria and NG or CT [45]. The most common presenting symptom among women was vaginal discharge which was only associated with VCA and not with NG, CT, BV or TV. Genital itching was reported by over half of patients and was also predictive of VCA. Itching was also useful in pointing away from NG, as was reported ulcer. Gynecologic exam, specifically endocervical discharge, was helpful in the diagnosis of NG. Interestingly, wet mount results were predictive NG (BV+, VCA-), suggesting that these inexpensive and simple tests should be included in any workup of symptomatic women. Despite extensive laboratory testing, we failed to find an etiology for a substantial proportion of women seeking care. This may reflect poor sensitivity of microscopy as well as non-infectious causes of symptoms. As has been noted elsewhere, factors associated with NG were more useful in predicting infections than those for CT [46, 47]. For both men and women, younger age was predictive of both NG and CT and lower edu- cation level and jobless or part-time employment status were predictive of NG. Interestingly, number of partners was not independently associated with CT or NG. Most men and women reported never using condoms and very few reported always using condoms. Women who sometimes used condoms were at higher risk of NG than those who never used them. This may be due to increased condom use in women with higher risk partners. Genital ulcers were not a common presenting symptom and were not associated with RPR results among HIV+ patients. RPR provided a diagnosis for 20% of ulcers among HIV- men and 15% among HIV- women. As others in Africa have reported, HSV is the most likely diag- nosis for RPR- ulcers which was more common among HIV+ patients [48]. Non-circumcision among men is associated with HIV acquisition and with increased prevalence and incidence of ulcerative STI [49–52]. We have previously shown a relationship between ulcers, smegma and HIV acquisition in uncircumcised men [15]. Among HIV- men, those who were uncircum- cised were not more likely to report ulcers or to be RPR+ than their circumcised counterparts. PLOS ONE | https://doi.org/10.1371/journal.pone.0250044 April 20, 2021 12 / 21 PLOS ONE Etiologies of genital abnormalities in Rwandan men and women Table 5. Univariate and multivariate analysis of factors associated with CT or NG infection in women in Kigali, Rwanda (N = 579). cPOR 95% CI p-value aPOR 95% CI p-value cPOR 95% CI p-value aPOR 95% CI p-value CT infection NG infection Demographics Age (per year increase) 0.91 0.88 0.95 <0.0001 0.90 0.86 0.94 <0.0001 0.95 0.92 0.97 < .001 0.93 0.89 0.97 <0.001 Referrer Radio Advert ref ref Friends/Walk-in/Pharmacy/Contact 1.74 1.11 2.72 0.015 1.31 0.91 1.90 0.150 Partner/Internet Living and Marital Status Married and Cohabiting Other Education Level None/Primary Secondary/Higher Employment Status Full-time employment Part-time/Student/Jobless Sexual behaviors Number of partners in last 30 days None or one partner More than one partner Condom use during vaginal sex in the last 3 months No partners or always used condoms Sometimes Never Number of days since sexual contact you suspect STI was acquired from 0–8 9–16 > = 17 Number of children under 18 (per child increase) Number of additional children desired (per child increase) Family planning method and pregnancy composite Pregnant Hormonal method (implant, injectable, pills) Non-Hormonal (IUD/Condom/Tubal Ligation/Natural Method) or No Method Self-reported symptoms Vaginal discharge Yes No Genital itching Yes No Dysuria ref ref 1.78 1.13 2.80 0.012 1.46 1.00 2.13 0.048 ref 1.30 0.83 2.01 0.248 ref 2.09 1.44 3.04 0.000 2.13 1.30 3.48 0.003 ref ref ref ref 1.23 0.77 1.96 0.383 1.76 1.16 2.66 0.008 1.95 1.12 3.39 0.019 ref ref 1.56 0.89 2.75 0.119 3.53 2.19 5.69 <0.0001 0.46 2.97 0.92 2.42 0.741 0.107 0.53 2.69 1.1 3.49 0.666 0.022 1.17 1.49 ref 1.20 1.96 ref 0.15 1.5 0.207 1.81 4.18 <0.0001 0.22 2.41 1.07 2.98 0.611 0.025 0.74 1.79 ref 0.87 3.16 1.36 3.87 0.126 0.002 0.48 2.75 ref 1.66 2.29 ref 0.82 0.69 0.99 0.037 0.96 0.84 1.10 0.594 1.22 1.02 1.45 0.027 0.92 0.79 1.07 0.274 0.96 0.95 0.43 2.14 0.56 1.59 0.915 0.837 1.00 2.01 0.49 2.03 1.32 3.05 0.999 0.001 1.30 1.73 0.57 2.99 1.02 2.94 0.532 0.040 ref ref ref ref 1.26 0.73 2.18 0.408 1.10 0.68 1.78 0.692 ref ref ref ref 1.08 0.69 1.68 0.738 2.62 1.79 3.84 <0.0001 2.54 1.55 4.17 0.0002 PLOS ONE | https://doi.org/10.1371/journal.pone.0250044 April 20, 2021 (Continued ) 13 / 21 PLOS ONE Etiologies of genital abnormalities in Rwandan men and women Table 5. (Continued) Yes No Genital ulcer Yes No Number of days with symptoms 1–10 11 or more Laboratory and physical exam HIV Status Positive Negative RPR Result Positive Negative Trichomonas Positive Negative Candida Positive Negative Bacterial vaginosis Positive Negative Vaginal Inflammation OR Discharge Yes No Endocervical Inflammation or Discharge Yes No Genital Ulcer Yes No cPOR 95% CI p-value aPOR 95% CI p-value cPOR 95% CI p-value aPOR 95% CI p-value CT infection NG infection ref 1.06 0.68 1.64 0.796 ref 1.21 0.84 1.75 0.303 ref ref ref 1.30 0.62 2.73 0.489 2.33 1.12 4.84 0.024 2.52 1.09 5.80 0.030 1.72 1.06 2.78 0.027 1.76 1.07 2.88 0.026 1.76 1.16 2.68 0.008 1.78 1.05 3.00 0.032 ref ref ref ref ref 2.73 1.66 4.47 <0.0001 2.05 1.10 3.83 0.024 1.83 0.85 3.96 0.124 ref ref 1.37 0.66 2.88 0.401 2.58 1.41 4.7 0.002 ref 1.85 1.03 3.32 0.041 ref ref ref ref 1.06 0.60 1.88 0.838 ref ref 1.98 1.04 3.77 0.038 3.56 1.89 6.69 <0.0001 2.20 1.11 4.36 0.024 1.63 0.98 2.72 0.063 2.63 1.67 4.15 <0.0001 1.89 1.07 3.34 0.028 ref ref ref ref ref 1.47 0.79 2.75 0.2201 1.67 0.97 2.86 0.063 1.83 1.15 2.91 0.010 2.17 1.46 3.23 0.000 1.80 1.11 2.93 0.018 ref 1.19 0.56 2.56 0.649 ref ref ref ref 1.06 0.53 2.10 0.875 IUD: intrauterine device; aPOR: Adjusted prevalence odds ratio; cPOR: Crude prevalence odds ratio; RPR: Rapid plasma reagin; CI: Confidence interval; STI: Sexually transmitted disease; NG: Neisseria gonorrhoeae, CT: Chlamydia trachomatis https://doi.org/10.1371/journal.pone.0250044.t005 In contrast, among HIV+ men, those who were uncircumcised were more likely to have an ulcer and less likely to be RPR+ than circumcised men. Circumcision is widely promoted in Rwanda and available at no cost in most government health centers as part of HIV prevention services. Though the focus is on protecting HIV- men, our results here suggest that circumci- sion can benefit HIV+ men by reducing ulcer incidence [53]. It is likely that we missed other less common ulcer etiologies including HD, lymphogranu- loma venereum (LGV), and granuloma inguinale (Klebsiella granulomatis) [54]. Our clinicians did suspect chancroid in a few cases, but the service program did not record detailed descrip- tions or photographs of ulcers and we lacked laboratory diagnostics. The most recent PLOS ONE | https://doi.org/10.1371/journal.pone.0250044 April 20, 2021 14 / 21 PLOS ONE Etiologies of genital abnormalities in Rwandan men and women publication presenting confirmed chancroid diagnoses in Rwanda was based on data collected in 1992, which found 27% of ulcers in men and 20% in women had culture-confirmed HD [55–59]. For many years the prevalence of HD had been decreasing in much of Africa [48, 54], but recent publications indicate HD may be staging a comeback [21]. More investigations are needed in Rwanda. Physical exam findings made important contributions in our program. Examination of male genitalia does not require specialized equipment, but speculum exam requires a skilled cli- nician, a gynecologic exam table and light which are in limited supply in low resource settings. While genital exams would not be feasible for all symptomatic patients, targeted genital exams in specific circumstances would be feasible and potentially very useful. Distinguishing between vaginal and endocervical discharges would greatly improve diagnostic accuracy and bi-manual exam would identify pelvic inflammatory disease. Similarly, in our setting where less than one in five ulcer patients are RPR+, assessing ulcer characteristics may be worthwhile. Visual exam has traditionally been viewed as unreliable as many ulcers do not have a paradigmatic presenta- tion (e.g. painless ‘clean’ TP ulcer, painful ‘dirty’ HD with inguinal adenopathy, multiple chronic or recurrent shallow vesicular HSV lesions). However, a recent study in Jamaica com- pared clinical diagnosis with M-PCR and found visual diagnoses of TP, HSV, and HD were 67.7%, 53.8%, and 75% sensitive and 91.2%, 83.6%, and 75.4% specific, respectively [60]. The advent of point-of-care diagnostics for NG and CT has transformed STI diagnosis, but given relatively expensive equipment and reagents, this remains out of reach in many low resource settings. We have used pooling to reduce the per-patient cost in Zambia and this could be explored in other settings [61]. GeneXpert kits are also available for TV and they are more sensitive than microscopy. The US CDC has in-house multiplex PCR (M-PCR) for ulcer etiologies including syphilis, HSV and chancroid. A focused study would provide prevalence information that could inform the next update of national guidelines. Our program has several limitations. Social desirability bias may have led to under-report- ing of risky sexual behaviors. We focused on symptomatic men and women and thus missed the many people who are asymptomatically infected [62, 63]. We did not screen for active viral hepatitis as recent unpublished surveys have shown a low prevalence of both hepatitis B and C (4% and 3%, respectively reported nationally, 4% and 5% among female sex workers tested in our laboratory). We did not have funding or resources to perform any direct method of detec- tion for TP using ulcer material, and thus may have misclassified some recently infected people who were negative by RPR test. While we did treat TV in male partners referred by TV + women, we did not systematically test for TV in men. Microscopy for TV detection in men is extremely insensitive, and we did not have resources to conduct GeneXpert testing for TV. TV could therefore be the reason for a portion of the symptomatic men with unknown etiol- ogy. We did not include HSV serologies because adult seroprevalence is high [64]. Assessment of cervical intraepithelial neoplasia requires more resources than would be achievable on a large scale in Rwandan health centers so we did not address this important problem. Fortu- nately, 93% of Rwandan girls now receive the human papillomavirus vaccine and future gener- ations will be protected [65]. Lastly, we and others have published an association between female genital schistosomiasis and HIV [66, 67], but this is most commonly seen with S.Hae- matobium while only S.Mansoni is endemic in Rwanda, thus we did not screen for genital schistosomiasis [68]. Conclusions Syndromic management guidelines in Rwanda can be improved with consideration of the prevalence of confirmed infections from this program offering services to symptomatic men PLOS ONE | https://doi.org/10.1371/journal.pone.0250044 April 20, 2021 15 / 21 PLOS ONE Etiologies of genital abnormalities in Rwandan men and women and women representative of those who would seek care at government health centers. Our findings indicate that syndromic management performs better among men but is poor among women. Inclusion of demographic and risk factor measures shown to be predictive of STI and non-STI dysbioses may also increase diagnostic accuracy. In symptomatic women, wet mount results for BV and VCA may help diagnose NG and are inexpensive and could be offered for management of women. Targeted genital exams for women in specific circumstances (e.g., in women without genital itching) may also be useful to diagnose NG. More data is needed on how often local prevalence and epidemiology should be reassessed to maintain improved syn- dromic management. Supporting information S1 Fig. STI baseline clinical form. (DOCX) Author Contributions Conceptualization: Julien Nyombayire, Rosine Ingabire, Susan Allen, Etienne Karita. Data curation: Kristin M. Wall, Julien Nyombayire, Rachel Parker, Susan Allen. Formal analysis: Kristin M. Wall, Rachel Parker. Funding acquisition: Susan Allen. Investigation: Julien Nyombayire, Rosine Ingabire, Jean Bizimana, Jeannine Mukamuyango, Amelia Mazzei, Matt A. Price, Marie Aimee Unyuzimana, Susan Allen, Etienne Karita. Methodology: Kristin M. Wall, Jean Bizimana, Matt A. Price, Marie Aimee Unyuzimana, Amanda Tichacek, Susan Allen, Etienne Karita. Project administration: Julien Nyombayire, Rosine Ingabire, Jean Bizimana, Jeannine Muka- muyango, Amelia Mazzei, Amanda Tichacek, Susan Allen, Etienne Karita. Resources: Susan Allen, Etienne Karita. Supervision: Julien Nyombayire, Rosine Ingabire, Jean Bizimana, Jeannine Mukamuyango, Amelia Mazzei, Marie Aimee Unyuzimana, Amanda Tichacek, Susan Allen, Etienne Karita. Validation: Julien Nyombayire, Rachel Parker, Rosine Ingabire, Jeannine Mukamuyango, Amelia Mazzei, Matt A. Price, Marie Aimee Unyuzimana, Amanda Tichacek, Susan Allen. Writing – original draft: Kristin M. Wall, Susan Allen. Writing – review & editing: Julien Nyombayire, Rachel Parker, Rosine Ingabire, Jean Bizi- mana, Jeannine Mukamuyango, Amelia Mazzei, Matt A. Price, Marie Aimee Unyuzimana, Amanda Tichacek, Susan Allen, Etienne Karita. References 1. World_Health_Organization. Sexually transmitted infections (STIs) Geneva: WHO; 2019 [updated 14 June 2019]. Available from: https://www.who.int/news-room/fact-sheets/detail/sexually-transmitted- infections-(stis). 2. World_Health_Organization. Sexually Transmiited Infections Geneva: World_Health_Organization; 2019 [updated July 2019; cited 2020 Jan 30]. Available from: https://www.who.int/news-room/fact- sheets/detail/sexually-transmitted-infections-(stis). PLOS ONE | https://doi.org/10.1371/journal.pone.0250044 April 20, 2021 16 / 21 PLOS ONE Etiologies of genital abnormalities in Rwandan men and women 3. World_Health_Organization. Global health sector strategy on Sexually Transmitted Infections, 2016– 2021: WHO; 2016 [updated July 2016]. Available from: http://www.who.int/reproductivehealth/ publications/rtis/ghss-stis/en/. 4. Carlson JM, Schaefer M, Monaco DC, Batorsky R, Claiborne DT, Prince J, et al. HIV transmission. Selection bias at the heterosexual HIV-1 transmission bottleneck. Science (New York, NY). 2014; 345 (6193):1254031. Epub 2014/07/12. https://doi.org/10.1126/science.1254031 PMID: 25013080; PubMed Central PMCID: PMC4289910. 5. Mujugira A, Magaret AS, Baeten JM, Celum C, Lingappa J. Risk Factors for HSV-2 Infection among Sexual Partners of HSV-2/HIV-1 Co-Infected Persons. BMC Res Notes. 2011; 4:64. Epub 2011/03/17. https://doi.org/10.1186/1756-0500-4-64 PMID: 21406077; PubMed Central PMCID: PMC3064615. 6. Haaland RE, Hawkins PA, Salazar-Gonzalez J, Johnson A, Tichacek A, Karita E, et al. Inflammatory genital infections mitigate a severe genetic bottleneck in heterosexual transmission of subtype A and C HIV-1. PLoS pathogens. 2009; 5(1):e1000274. Epub 2009/01/24. https://doi.org/10.1371/journal.ppat. 1000274 PMID: 19165325; PubMed Central PMCID: PMC2621345. 7. Auvert B, Buve A, Ferry B, Carael M, Morison L, Lagarde E, et al. Ecological and individual level analy- sis of risk factors for HIV infection in four urban populations in sub-Saharan Africa with different levels of HIV infection. AIDS. 2001; 15 Suppl 4:S15–30. Epub 2001/11/01. https://doi.org/10.1097/00002030- 200108004-00003 PMID: 11686462. 8. Piot P. AIDS: the impact of other sexually transmitted diseases. Netw Res Triangle Park N C. 1988; 9 (2):4. Epub 1988/01/01. PMID: 12280952. 9. Dhana A, Luchters S, Moore L, Lafort Y, Roy A, Scorgie F, et al. Systematic review of facility-based sex- ual and reproductive health services for female sex workers in Africa. Global Health. 2014; 10:46. Epub 2014/06/12. https://doi.org/10.1186/1744-8603-10-46 PMID: 24916010; PubMed Central PMCID: PMC4070634. 10. Vandenhoudt HM, Langat L, Menten J, Odongo F, Oswago S, Luttah G, et al. Prevalence of HIV and other sexually transmitted infections among female sex workers in Kisumu, Western Kenya, 1997 and 2008. PLoS One. 2013; 8(1):e54953. Epub 2013/02/02. https://doi.org/10.1371/journal.pone.0054953 PMID: 23372801; PubMed Central PMCID: PMC3553007. 11. Vickerman P, Ndowa F, O’Farrell N, Steen R, Alary M, Delany-Moretlwe S. Using mathematical model- ling to estimate the impact of periodic presumptive treatment on the transmission of sexually transmitted infections and HIV among female sex workers. Sex Transm Infect. 2010; 86(3):163–8. Epub 2009/10/ 27. https://doi.org/10.1136/sti.2008.034678 PMID: 19854700. 12. Seck K, Samb N, Tempesta S, Mulanga-Kabeya C, Henzel D, Sow PS, et al. Prevalence and risk fac- tors of cervicovaginal HIV shedding among HIV-1 and HIV-2 infected women in Dakar, Senegal. Sex Transm Infect. 2001; 77(3):190–3. Epub 2001/06/13. https://doi.org/10.1136/sti.77.3.190 PMID: 11402227; PubMed Central PMCID: PMC1744303. 13. Celum C, Wald A, Lingappa JR, Magaret AS, Wang RS, Mugo N, et al. Acyclovir and transmission of HIV-1 from persons infected with HIV-1 and HSV-2. The New England journal of medicine. 2010; 362 (5):427–39. Epub 2010/01/22. https://doi.org/10.1056/NEJMoa0904849 PMID: 20089951; PubMed Central PMCID: PMC2838503. 14. Daniels B, Wand H, Ramjee G, Team MDP. Prevalence of Herpes Simplex Virus 2 (HSV-2) infection and associated risk factors in a cohort of HIV negative women in Durban, South Africa. BMC Res Notes. 2016; 9(1):510. Epub 2016/12/14. https://doi.org/10.1186/s13104-016-2319-5 PMID: 27955706; PubMed Central PMCID: PMC5154041. 15. Wall KM, Kilembe W, Vwalika B, Haddad LB, Hunter E, Lakhi S, et al. Risk of heterosexual HIV trans- mission attributable to sexually transmitted infections and non-specific genital inflammation in Zambian discordant couples, 1994–2012. Int J Epidemiol. 2017; 46(5):1593–606. Epub 2017/04/13. https://doi. org/10.1093/ije/dyx045 PMID: 28402442; PubMed Central PMCID: PMC5837621. 16. Joseph Davey DL, Wall KM, Kilembe W, Naw HK, Brill I, Vwalika B, et al. HIV Incidence and Predictors of HIV Acquisition From an Outside Partner in Serodiscordant Couples in Lusaka, Zambia. J Acquir Immune Defic Syndr. 2017; 76(2):123–31. Epub 2017/07/25. https://doi.org/10.1097/QAI. 0000000000001494 PMID: 28737591; PubMed Central PMCID: PMC5597474. 17. Boeras DI, Hraber PT, Hurlston M, Evans-Strickfaden T, Bhattacharya T, Giorgi EE, et al. Role of donor genital tract HIV-1 diversity in the transmission bottleneck. Proceedings of the National Academy of Sci- ences of the United States of America. 2011; 108(46):E1156–63. Epub 2011/11/09. https://doi.org/10. 1073/pnas.1103764108 PMID: 22065783; PubMed Central PMCID: PMC3219102. 18. Looker KJ, Welton NJ, Sabin KM, Dalal S, Vickerman P, Turner KME, et al. Global and regional esti- mates of the contribution of herpes simplex virus type 2 infection to HIV incidence: a population attribut- able fraction analysis using published epidemiological data. Lancet Infect Dis. 2020; 20(2):240–9. Epub 2019/11/23. https://doi.org/10.1016/S1473-3099(19)30470-0 PMID: 31753763; PubMed Central PMCID: PMC6990396. PLOS ONE | https://doi.org/10.1371/journal.pone.0250044 April 20, 2021 17 / 21 PLOS ONE Etiologies of genital abnormalities in Rwandan men and women 19. Rietmeijer CA, Mungati M, Kilmarx PH, Barr BT, Gonese E, Kularatne RS, et al. Serological Markers for Syphilis Among Persons Presenting With Syndromes Associated With Sexually Transmitted Infections: Results From the Zimbabwe STI Etiology Study. Sex Transm Dis. 2019; 46(9):579–83. Epub 2019/04/ 23. https://doi.org/10.1097/OLQ.0000000000001006 PMID: 31008842; PubMed Central PMCID: PMC6885999. 20. Hayes RJ, Schulz KF, Plummer FA. The cofactor effect of genital ulcers on the per-exposure risk of HIV transmission in sub-Saharan Africa. J Trop Med Hyg. 1995; 98(1):1–8. Epub 1995/02/01. PMID: 7861474. 21. Phiri S, Zadrozny S, Weiss HA, Martinson F, Nyirenda N, Chen CY, et al. Etiology of genital ulcer dis- ease and association with HIV infection in Malawi. Sex Transm Dis. 2013; 40(12):923–8. Epub 2013/ 11/14. https://doi.org/10.1097/OLQ.0000000000000051 PMID: 24220352. 22. Takuva S, Mugurungi O, Mutsvangwa J, Machiha A, Mupambo AC, Maseko V, et al. Etiology and anti- microbial susceptibility of pathogens responsible for urethral discharge among men in Harare, Zimba- bwe. Sex Transm Dis. 2014; 41(12):713–7. Epub 2015/01/13. https://doi.org/10.1097/OLQ. 0000000000000204 PMID: 25581806. 23. Sylverken AA, Owusu-Dabo E, Yar DD, Salifu SP, Awua-Boateng NY, Amuasi JH, et al. Bacterial etiol- ogy of sexually transmitted infections at a STI clinic in Ghana; use of multiplex real time PCR. Ghana Med J. 2016; 50(3):142–8. Epub 2016/10/19. PMID: 27752188; PubMed Central PMCID: PMC5044789. 24. Chirenje ZM, Dhibi N, Handsfield HH, Gonese E, Tippett Barr B, Gwanzura L, et al. The Etiology of Vag- inal Discharge Syndrome in Zimbabwe: Results from the Zimbabwe STI Etiology Study. Sex Transm Dis. 2018; 45(6):422–8. Epub 2018/02/22. https://doi.org/10.1097/OLQ.0000000000000771 PMID: 29465674; PubMed Central PMCID: PMC6879447. 25. Haddad LB, Wall KM, Kilembe W, Vwalika B, Khu NH, Brill I, et al. Bacterial vaginosis modifies the asso- ciation between hormonal contraception and HIV acquisition. AIDS. 2018; 32(5):595–604. Epub 2018/ 01/16. https://doi.org/10.1097/QAD.0000000000001741 PMID: 29334545; PubMed Central PMCID: PMC5832628. 26. Barnabas SL, Dabee S, Passmore JS, Jaspan HB, Lewis DA, Jaumdally SZ, et al. Converging epidem- ics of sexually transmitted infections and bacterial vaginosis in southern African female adolescents at risk of HIV. Int J STD AIDS. 2018; 29(6):531–9. Epub 2017/12/05. https://doi.org/10.1177/ 0956462417740487 PMID: 29198180. 27. Masha SC, Cools P, Descheemaeker P, Reynders M, Sanders EJ, Vaneechoutte M. Urogenital patho- gens, associated with Trichomonas vaginalis, among pregnant women in Kilifi, Kenya: a nested case- control study. BMC Infect Dis. 2018; 18(1):549. Epub 2018/11/08. https://doi.org/10.1186/s12879-018- 3455-4 PMID: 30400890; PubMed Central PMCID: PMC6219184. 28. Kerubo E, Laserson KF, Otecko N, Odhiambo C, Mason L, Nyothach E, et al. Prevalence of reproduc- tive tract infections and the predictive value of girls’ symptom-based reporting: findings from a cross- sectional survey in rural western Kenya. Sex Transm Infect. 2016; 92(4):251–6. Epub 2016/01/29. https://doi.org/10.1136/sextrans-2015-052371 PMID: 26819339; PubMed Central PMCID: PMC4893088. 29. Lewis DA, Marsh K, Radebe F, Maseko V, Hughes G. Trends and associations of Trichomonas vagina- lis infection in men and women with genital discharge syndromes in Johannesburg, South Africa. Sex Transm Infect. 2013; 89(6):523–7. Epub 2013/04/23. https://doi.org/10.1136/sextrans-2013-051049 PMID: 23605850. 30. Morikawa E, Mudau M, Olivier D, de Vos L, Joseph Davey D, Price C, et al. Acceptability and Feasibility of Integrating Point-of-Care Diagnostic Testing of Sexually Transmitted Infections into a South African Antenatal Care Program for HIV-Infected Pregnant Women. Infect Dis Obstet Gynecol. 2018; 2018:3946862. Epub 2018/06/05. https://doi.org/10.1155/2018/3946862 PMID: 29861622; PubMed Central PMCID: PMC5971359. 31. Garrett NJ, Osman F, Maharaj B, Naicker N, Gibbs A, Norman E, et al. Beyond syndromic manage- ment: Opportunities for diagnosis-based treatment of sexually transmitted infections in low- and middle- income countries. PLoS One. 2018; 13(4):e0196209. https://doi.org/10.1371/journal.pone.0196209 PMID: 29689080; PubMed Central PMCID: PMC5918163. 32. Francis SC, Ao TT, Vanobberghen FM, Chilongani J, Hashim R, Andreasen A, et al. Epidemiology of curable sexually transmitted infections among women at increased risk for HIV in northwestern Tanza- nia: inadequacy of syndromic management. PLoS One. 2014; 9(7):e101221. Epub 2014/07/16. https:// doi.org/10.1371/journal.pone.0101221 PMID: 25025338; PubMed Central PMCID: PMC4099080. 33. Guimaraes H, Castro R, Tavora Tavira L, da LEF. Assessing therapeutic management of vaginal and urethral symptoms in an anonymous HIV testing centre in Luanda, Angola. J Infect Dev Ctries. 2013; 7 (10):720–5. Epub 2013/10/17. https://doi.org/10.3855/jidc.2752 PMID: 24129624. PLOS ONE | https://doi.org/10.1371/journal.pone.0250044 April 20, 2021 18 / 21 PLOS ONE Etiologies of genital abnormalities in Rwandan men and women 34. Marx G, John-Stewart G, Bosire R, Wamalwa D, Otieno P, Farquhar C. Diagnosis of sexually transmit- ted infections and bacterial vaginosis among HIV-1-infected pregnant women in Nairobi. Int J STD AIDS. 2010; 21(8):549–52. Epub 2010/10/27. https://doi.org/10.1258/ijsa.2010.010005 PMID: 20975086; PubMed Central PMCID: PMC3050991. 35. Black V, Magooa P, Radebe F, Myers M, Pillay C, Lewis DA. The detection of urethritis pathogens among patients with the male urethritis syndrome, genital ulcer syndrome and HIV voluntary counselling and testing clients: should South Africa’s syndromic management approach be revised? Sex Transm Infect. 2008; 84(4):254–8. Epub 2008/01/15. https://doi.org/10.1136/sti.2007.028464 PMID: 18192290. 36. 37. 38. Frohlich JA, Abdool Karim Q, Mashego MM, Sturm AW, Abdool Karim SS. Opportunities for treating sexually transmitted infections and reducing HIV risk in rural South Africa. J Adv Nurs. 2007; 60(4):377– 83. Epub 2007/09/08. https://doi.org/10.1111/j.1365-2648.2007.04405.x PMID: 17822425. Tann CJ, Mpairwe H, Morison L, Nassimu K, Hughes P, Omara M, et al. Lack of effectiveness of syn- dromic management in targeting vaginal infections in pregnancy in Entebbe, Uganda. Sex Transm Infect. 2006; 82(4):285–9. Epub 2006/08/01. https://doi.org/10.1136/sti.2005.014845 PMID: 16877576; PubMed Central PMCID: PMC2564710. Ingabire R, Parker R, Nyombayire J, Ko JE, Mukamuyango J, Bizimana J, et al. Female sex workers in Kigali, Rwanda: a key population at risk of HIV, sexually transmitted infections, and unplanned preg- nancy. Int J STD AIDS. 2019; 30(6):557–68. Epub 2019/02/08. https://doi.org/10.1177/ 0956462418817050 PMID: 30727831; PubMed Central PMCID: PMC6512058. 39. RWANDA POPULATION-BASED HIV IMPACT ASSESSMENT RPHIA 2018–2019 2019 [cited 2020 Nov 20]. Available from: https://phia.icap.columbia.edu/wp-content/uploads/2019/10/RPHIA-Summary- Sheet_Oct-2019.pdf. 40. Conkling M, Shutes EL, Karita E, Chomba E, Tichacek A, Sinkala M, et al. Couples’ voluntary counsel- ling and testing and nevirapine use in antenatal clinics in two African capitals: a prospective cohort study. Journal of the International AIDS Society. 2010; 13:10. Epub 2010/03/17. https://doi.org/10. 1186/1758-2652-13-10 PMID: 20230628; PubMed Central PMCID: PMC2851580. 41. Karita E, Nsanzimana S, Ndagije F, Wall KM, Mukamuyango J, Mugwaneza P, et al. Implementation and Operational Research: Evolution of Couples’ Voluntary Counseling and Testing for HIV in Rwanda: From Research to Public Health Practice. J Acquir Immune Defic Syndr. 2016; 73(3):e51–e8. Epub 2016/10/16. https://doi.org/10.1097/QAI.0000000000001138 PMID: 27741033; PubMed Central PMCID: PMC5367509. 42. Mazzei A, Ingabire R, Mukamuyango J, Nyombayire J, Sinabamenye R, Bayingana R, et al. Community health worker promotions increase uptake of long-acting reversible contraception in Rwanda. Reprod Health. 2019; 16(1):75. Epub 2019/06/06. https://doi.org/10.1186/s12978-019-0739-0 PMID: 31164155; PubMed Central PMCID: PMC6549304. 43. Ingabire R, Nyombayire J, Hoagland A, Da Costa V, Mazzei A, Haddad L, et al. Evaluation of a multi- level intervention to improve postpartum intrauterine device services in Rwanda. Gates Open Res. 2018; 2(38):38. Epub 2019/09/25. https://doi.org/10.12688/gatesopenres.12854.3 PMID: 30569036; PubMed Central PMCID: PMC6266741.3. 44. Rietmeijer CA, Mungati M, Machiha A, Mugurungi O, Kupara V, Rodgers L, et al. The Etiology of Male Urethral Discharge in Zimbabwe: Results from the Zimbabwe STI Etiology Study. Sex Transm Dis. 2018; 45(1):56–60. Epub 2017/12/15. https://doi.org/10.1097/OLQ.0000000000000696 PMID: 29240635. 45. Gonorrhea—CDC Fact Sheet (Detailed Version): U.S. Centers for Disease Control; 2020 [cited 2020 6/ 29/2020]. Available from: https://www.cdc.gov/std/gonorrhea/stdfact-gonorrhea-detailed.htm. 46. Dela H, Attram N, Behene E, Kumordjie S, Addo KK, Nyarko EO, et al. Risk factors associated with gon- orrhea and chlamydia transmission in selected health facilities in Ghana. BMC Infect Dis. 2019; 19 (1):425. Epub 2019/05/18. https://doi.org/10.1186/s12879-019-4035-y PMID: 31096920; PubMed Cen- tral PMCID: PMC6524331. 47. Connolly S, Wall KM, Parker R, Kilembe W, Inambao M, Visoiu AM, et al. Sociodemographic factors and STIs associated with Chlamydia trachomatis and Neisseria gonorrhoeae infections in Zambian female sex workers and single mothers. Int J STD AIDS. 2020; 31(4):364–74. Epub 2020/03/05. https:// doi.org/10.1177/0956462419894453 PMID: 32126947. 48. Mungati M, Machiha A, Mugurungi O, Tshimanga M, Kilmarx PH, Nyakura J, et al. The Etiology of Geni- tal Ulcer Disease and Coinfections With Chlamydia trachomatis and Neisseria gonorrhoeae in Zimba- bwe: Results From the Zimbabwe STI Etiology Study. Sex Transm Dis. 2018; 45(1):61–8. Epub 2017/ 12/15. https://doi.org/10.1097/OLQ.0000000000000694 PMID: 29240636; PubMed Central PMCID: PMC5994235. 49. MacNeily AE. Editorial comment. J Urol. 2011; 185(6):2306–7. Epub 2011/04/23. https://doi.org/10. 1016/j.juro.2011.02.2702 PMID: 21511302. PLOS ONE | https://doi.org/10.1371/journal.pone.0250044 April 20, 2021 19 / 21 PLOS ONE Etiologies of genital abnormalities in Rwandan men and women 50. Weiss HA, Thomas SL, Munabi SK, Hayes RJ. Male circumcision and risk of syphilis, chancroid, and genital herpes: a systematic review and meta-analysis. Sex Transm Infect. 2006; 82(2):101–9; discus- sion 10. Epub 2006/04/04. https://doi.org/10.1136/sti.2005.017442 PMID: 16581731; PubMed Central PMCID: PMC2653870. 51. Morris BJ, Hankins CA. Effect of male circumcision on risk of sexually transmitted infections and cervi- cal cancer in women. Lancet Glob Health. 2017; 5(11):e1054–e5. Epub 2017/10/14. https://doi.org/10. 1016/S2214-109X(17)30386-8 PMID: 29025620. 52. Davis S, Toledo C, Lewis L, Maughan-Brown B, Ayalew K, Kharsany ABM. Does voluntary medical male circumcision protect against sexually transmitted infections among men and women in real-world scale-up settings? Findings of a household survey in KwaZulu-Natal, South Africa. BMJ Glob Health. 2019; 4(3):e001389. Epub 2019/07/03. https://doi.org/10.1136/bmjgh-2019-001389 PMID: 31263584; PubMed Central PMCID: PMC6570991. 53. Mehta SD, Moses S, Parker CB, Agot K, Maclean I, Bailey RC. Circumcision status and incident herpes simplex virus type 2 infection, genital ulcer disease, and HIV infection. AIDS. 2012; 26(9):1141–9. Epub 2012/03/03. https://doi.org/10.1097/QAD.0b013e328352d116 PMID: 22382150; PubMed Central PMCID: PMC3668787. 54. Kularatne RS, Muller EE, Maseko DV, Kufa-Chakezha T, Lewis DA. Trends in the relative prevalence of genital ulcer disease pathogens and association with HIV infection in Johannesburg, South Africa, 2007–2015. PLoS One. 2018; 13(4):e0194125. Epub 2018/04/05. https://doi.org/10.1371/journal.pone. 0194125 PMID: 29617372; PubMed Central PMCID: PMC5884493. 55. Bogaerts J, Kestens L, van Dyck E, Tello WM, Akingeneye J, Mukantabana V. Genital ulcers in a pri- mary health clinic in Rwanda: impact of HIV infection on diagnosis and ulcer healing (1986–1992). Int J STD AIDS. 1998; 9(11):706–10. Epub 1998/12/24. https://doi.org/10.1258/0956462981921242 PMID: 9863586. 56. Bogaerts J, Vuylsteke B, Martinez Tello W, Mukantabana V, Akingeneye J, Laga M, et al. Simple algo- rithms for the management of genital ulcers: evaluation in a primary health care centre in Kigali, Rwanda. Bull World Health Organ. 1995; 73(6):761–7. Epub 1995/01/01. PMID: 8907769; PubMed Central PMCID: PMC2486690. 57. Van Dyck E, Bogaerts J, Smet H, Tello WM, Mukantabana V, Piot P. Emergence of Haemophilus ducreyi resistance to trimethoprim-sulfamethoxazole in Rwanda. Antimicrob Agents Chemother. 1994; 38(7):1647–8. Epub 1994/07/01. https://doi.org/10.1128/aac.38.7.1647 PMID: 7979300; PubMed Cen- tral PMCID: PMC284607. 58. Roggen EL, Hoofd G, Van Dyck E, Piot P. Enzyme immunoassays (EIAs) for the detection of anti-Hae- mophilus ducreyi serum IgA, IgG, and IgM antibodies. Sex Transm Dis. 1994; 21(1):36–42. Epub 1994/ 01/01. https://doi.org/10.1097/00007435-199401000-00008 PMID: 8140487. 59. Bogaerts J, Ricart CA, Van Dyck E, Piot P. The etiology of genital ulceration in Rwanda. Sex Transm Dis. 1989; 16(3):123–6. Epub 1989/07/01. https://doi.org/10.1097/00007435-198907000-00001 PMID: 2510325. 60. Behets FM, Brathwaite AR, Hylton-Kong T, Chen CY, Hoffman I, Weiss JB, et al. Genital ulcers: setiol- ogy, clinical diagnosis, and associated human immunodeficiency virus infection in Kingston, Jamaica. Clin Infect Dis. 1999; 28(5):1086–90. Epub 1999/08/19. https://doi.org/10.1086/514751 PMID: 10452639. 61. Connolly S, Kilembe W, Inambao M, Visoiu AM, Sharkey T, Parker R, et al. A population-specific opti- mized GeneXpert pooling algorithm for Chlamydia trachomatis and Neisseria gonorrhoeae to reduce cost of molecular STI screening in resource-limited settings. J Clin Microbiol. 2020. Epub 2020/06/12. https://doi.org/10.1128/JCM.00176-20 PMID: 32522828. 62. Johnson LF, Dorrington RE, Bradshaw D, Coetzee DJ. The effect of syndromic management interven- tions on the prevalence of sexually transmitted infections in South Africa. Sex Reprod Healthc. 2011; 2 (1):13–20. Epub 2010/12/15. https://doi.org/10.1016/j.srhc.2010.08.006 PMID: 21147454. 63. Garrett N, Mitchev N, Osman F, Naidoo J, Dorward J, Singh R, et al. Diagnostic accuracy of the Xpert CT/NG and OSOM Trichomonas Rapid assays for point-of-care STI testing among young women in South Africa: a cross-sectional study. BMJ Open. 2019; 9(2):e026888. Epub 2019/02/21. https://doi. org/10.1136/bmjopen-2018-026888 PMID: 30782948; PubMed Central PMCID: PMC6367982. 64. Lingappa JR, Kahle E, Mugo N, Mujugira A, Magaret A, Baeten J, et al. Characteristics of HIV-1 discor- dant couples enrolled in a trial of HSV-2 suppression to reduce HIV-1 transmission: the partners study. PLoS One. 2009; 4(4):e5272. Epub 2009/05/01. https://doi.org/10.1371/journal.pone.0005272 PMID: 19404392; PubMed Central PMCID: PMC2671170. 65. Cousins S. How Rwanda could be the first country to wipe out cervical cancer: Mosaicscience.com; 2019 [updated 7May2019]. Available from: https://mosaicscience.com/story/rwanda-cervical-cancer- hpv-vaccine-gardasil-cervarix/. PLOS ONE | https://doi.org/10.1371/journal.pone.0250044 April 20, 2021 20 / 21 PLOS ONE Etiologies of genital abnormalities in Rwandan men and women 66. Wall KM, Kilembe W, Vwalika B, Dinh C, Livingston P, Lee YM, et al. Schistosomiasis is associated with incident HIV transmission and death in Zambia. PLoS Negl Trop Dis. 2018; 12(12):e0006902. Epub 2018/12/14. https://doi.org/10.1371/journal.pntd.0006902 PMID: 30543654; PubMed Central PMCID: PMC6292564. 67. Secor WE. The effects of schistosomiasis on HIV/AIDS infection, progression and transmission. Cur- rent opinion in HIV and AIDS. 2012; 7(3):254–9. Epub 2012/02/14. https://doi.org/10.1097/COH. 0b013e328351b9e3 PMID: 22327410. 68. Rujeni N, Morona D, Ruberanziza E, Mazigo HD. Schistosomiasis and soil-transmitted helminthiasis in Rwanda: an update on their epidemiology and control. Infect Dis Poverty. 2017; 6(1):8. Epub 2017/03/ 02. https://doi.org/10.1186/s40249-016-0212-z PMID: 28245883; PubMed Central PMCID: PMC5331630. PLOS ONE | https://doi.org/10.1371/journal.pone.0250044 April 20, 2021 21 / 21 PLOS ONE
null
10.1126_sciadv.abn5709.pdf
null
null
Supplementary Materials for A cooperative network at the nuclear envelope counteracts LINC-mediated forces during oogenesis in C. elegans Chenshu Liu et al. Corresponding author: Chenshu Liu, chenshu.liu@berkeley.edu; Abby F. Dernburg, afdernburg@berkeley.edu Sci. Adv. 9, eabn5709 (2023) DOI: 10.1126/sciadv.abn5709 The PDF file includes: Figs. S1 to S18 Tables S1 to S7 Legends for movies S1 to S5 Legend for Data S1 References Other Supplementary Material for this manuscript includes the following: Movies S1 to S5 Data S1 Fig. S1. Fig. S1. (Related to Figs. 1 and 3) CRISPR tagging and auxin-inducible degradation of LMN-1 in C. elegans germline. (A) Multiple sequence alignments of nematode LMN-1 proteins were generated using Clustal Omega and visualized using Jalview, showing the position of the degron/V5 insertion in C. elegans LMN-1. (B) Morphology of eggs/embryos laid by hermaphrodite worms treated ± auxin for 48 hrs (from young adulthood). Images acquired using DIC. Scale bar, 20 µm. Fig. S2. Fig. S2. (Related to Fig. 1) Representative images showing co-staining of DAPI and nucleoporins NPP-7 (A) or NPP-10 (B). Composite images are maximum-intensity projections. Meiosis progresses from left to right. Scale bars, 10 µm. Fig. S3. Fig. S3. (Related to Fig. 1) Quantifying nuclear collapse. (A) Composite images (maximum- intensity projections) showing nuclei in late meiotic prophase stained for DAPI (cyan), NPP-7 (red) or SYP-1 (red). Meiosis progresses from top left to bottom right. Scale bars, 10 µm. (B) Segmentation of chromosome volume based on DAPI fluorescence. Meiosis progresses from top left to bottom right. Scale bar, 10 µm. (C) Quantification of nuclear radii based on NPP-7 or SYP-1 staining (with maximum Z projection), nuclear volume based on radius3, nuclear size based on DAPI volume and integrated DAPI fluorescence intensity per nucleus. Nuclear sizes in late pachytene were normalized to one. Unpaired two-sample two-sided t-test was used to calculate p-values. At least 58 nuclei from three animals were analyzed per stage. Fig. S4. Fig. S4. (Related to Figs. 1 and 3) LMN-1 depletion causes SPO-11 independent DNA damage. (A) LMN-1 depletion causes persistent DNA damage marked by RAD-51 foci. Scale bar, 10 µm. (B) RAD-51 staining in mid-pachytene nuclei from control and hermaphrodites depleted of SPO-11, LMN-1, LMN-1 and SPO-11 both, or LMN-1 and SPO-11 and SUN-1 simultaneously. Control hermaphrodite has TIR1 but no AID-tagged genes. Animals of indicated genotypes (all homozygous for Psun-1::TIR1) were exposed to auxin for 24 hours from the L4 stage to young adulthood prior to dissection. Scale bar, 5 µm. (C) Quantification of RAD-51 foci per nucleus as a function of meiotic progression. Gonads were divided into five zones of equal length spanning the premeiotic region to early diplotene (as in Fig. S6C). Control, N = 964 nuclei (4 animals); SPO-11 depletion, N = 1480 nuclei (4 animals); LMN-1 depletion, N = 1781 nuclei (6 animals); LMN-1, SPO-11 double depletion, N = 1329 nuclei (5 animals); LMN-1, SPO-11, SUN-1 triple depletion, N = 844 nuclei (5 animals); SUN-1 depletion, N = 1201 nuclei (4 animals). Pairwise comparisons for proportions were performed to compute the p-values (adjusted by the Benjamini-Hochberg method, see Data S1). Experimental conditions were the same as in (B). (D) Designated crossover sites marked by GFP::COSA-1 foci in late prophase nuclei. Following LMN-1 depletion, nuclei still showed six designated crossover sites, whereas GFP::COSA-1 foci were absent from apoptotic nuclei. Animals of indicated genotypes (all homozygous for Psun-1::TIR1) were exposed to auxin for 24 hours from the L4 stage to young adulthood prior to dissection. Dashed arrows indicate meiotic progression. Scale bar, 5 µm. Fig. S5. Fig. S5. ATG-7 depletion does not rescue nuclear collapse. (A) Raw (non-deconvolved) maximum-intensity Z-projection showing the cytoplasmic localization of ATG-7 in the germline and the effectiveness of depletion by 12hr auxin treatment. Gray-scale images were scaled identically. Scale bar, 10 µm. (B) Nuclear morphology at later stages of meiotic prophase. Colored dashed lines mark meiotic stages based on the anatomical positions of gonads from control animals. Scale bar, 10 μm. Fig. S6. Fig. S6. (Related to Fig. 2) Chromosome pairing and synapsis occur normally following LMN-1 depletion. (A) Dynamics of synapsis, based on immunostaining of SYP-1 (synaptonemal complex) and HTP-3 (chromosome axes). Yellow asterisks indicate the distal end of gonads. Scale bar, 10 µm. (B) Normal pairing of X chromosomes is revealed by immunofluorescence of HIM-8, which localizes to the X-chromosome pairing centers. Scale bar, 2 µm. (C) Quantification of homolog pairing and synapsis. Diagram of distal gonad divided into five zones of equal length. Three gonads were measured per condition. Mean ± SD are plotted. Two-sided two-proportions z-test was used for computing the p-values. ***, p < 0.001; **, p < 0.01; ns, p ≥ 0.05 (see Data S1). Fig. S7. Fig. S7. (Related to Fig. 2) Prolonged ZYG-12::GFP clustering following LMN-1 depletion. (A) Mean intensities of ZYG-12::GFP at the nuclear envelope per nucleus, normalized against intensity levels at the transition zone. Medians (black crossbars) and means (black boxes) are shown. Fluorescence intensity surrounding each nucleus was manually segmented and quantified from additive projection images after background subtraction. 30 nuclei per stage were measured per condition (without or with auxin). TZ, transition zone; MP, mid-pachytene; LP, late pachytene; Dip, diplotene. Two-way ANOVA was used to calculate the p-value. (B) Clustering of ZYG-12::GFP, defined as the relative standard deviation (ratio of the standard deviation to the mean value) of fluorescence intensity at the NE in each nucleus. Medians (black crossbars) and means (black boxes) are shown. Fluorescence intensity was measured in the same way as in (A). 30 nuclei per stage were measured per condition. TZ, transition zone; MP, mid-pachytene; LP, late pachytene; Dip, diplotene. p-value was calculated by two-way ANOVA. Fig. S8. Fig. S8. (Related to Figs. 2 and 3) Quantifying LINC complex distribution at the NE. (A) Quantifying SUN-1 distribution at the NE. Representative line-scan profiles of SUN-1::mRuby fluorescence intensity at the circumference of individual nuclei. Grayscale images are additive projections and are scaled using the same look up table (LUT). Scale bar, 5 µm. (B) Quantifying asymmetric ZYG-12 distribution at the NE. Top panel shows grayscale images of additive projections with background subtraction showing polarized ZYG-12::GFP distribution at the NE in a LMN-1 depleted diplotene nucleus. The right panel shows line scan profile along the circumference of the nucleus and approximate locations of angles mapped subsequently during data alignment. Scale bar, 5 µm. Bottom panels show data alignment and normalization. Raw intensity measurement from line-scan was aligned and mapped such that 180° corresponds to the coordinate along the NE’s circumference with the maximum-intensity, and the intensity at 0° was normalized to one (red); or in cases where one nucleus has multiple ZYG-12 intensity peaks on the NE, 180° corresponds to the coordinate along the NE’s circumference with the minimum- intensity, and the intensity at 180° was normalized to one (green). Fig. S9. Fig. S9. (Related to Fig. 3) Co-depletion of DHC-1 or DLI-1, but not DYLT-1, also rescues nuclear collapse despite marked nuclear mispositioning. Nuclear morphology upon depleting DHC-1, DLI-1 or DYLT-1 in lmn-1::AID::V5 worms ±auxin treatment. Mitotic defects are seen in the proliferative region, and meiotic nuclei are mispositioned throughout the gonad following depletion of DHC-1 or DLI-1. Scale bar, 10 µm. Fig. S10. Fig. S10. (Related to Figs. 3 and 4) Efficacy of auxin-induced degradation or RNAi. (A) Nuclear morphology in hermaphrodites depleted of ZYG-12, SUN-1 or LEM-2 with auxin- induced degradation, or depleted of LEM-2 or LMN-1 using RNAi. The duration of auxin treatment was at least 8 hours and the duration of feeding RNAi was 48hr. Scale bars, 10 µm. All worms were homozygous for Psun-1::TIR1 or Pgld-1::TIR1. ZYG-12 depletion results in mispositioning of meiotic nuclei. (B) DNC-1 at NE can be effectively depleted using RNAi. Live imaging stills of DIC and DNC-1::GFP in the late pachytene region of the germline in worms fed with Control RNAi or dnc-1(RNAi). Arrows indicate meiotic progression. Scale bar, 10 µm. Fig. S11. Fig. S11. (Related to Fig. 3) CRISPR tagging and auxin-inducible degradation of SUN-1 in C. elegans germline. (A) Prediction of transmembrane region in C. elegans SUN-1 using TMHMM (http://www.cbs.dtu.dk/services/TMHMM/). Probability of amino acids inside the INM (green), outside the ONM (blue) and in the transmembrane region (magenta) is plotted. (B) Multiple sequence alignments of nematode SUN-1 proteins were generated using Clustal Omega and visualized using Jalview, showing position of degron/V5 insertion in C. elegans SUN-1. (C) X-chromosome pairing (HIM-8) and SC assembly (SYP-1) in sun-1::AID::V5 worms without or with auxin treatment. All worms were homozygous for Psun-1::TIR1. Scale bar, 10 µm. (D) SUN- 1 is required for ZYG-12 localization at the NE of meiotic cells. Composite images showing live meiotic nuclei from mid/late pachytene in worms of indicated genotypes or treatments. All worms were homozygous for Psun-1::TIR1 or Pgld-1::TIR1. ZYG-12::GFP in green and mRuby::SYP-3 in magenta. All images are maximum-intensity projections and scaled identically. Scale bar, 5 µm. Fig. S12. Fig. S12. (Related to Fig. 3) LMN-1 depletion is equally efficient upon auxin-induced co- degradation. (A) Four strains carrying lmn-1::AID::V5 alone or in combination with other AID- tagged genes were immunostained for V5 after the same 12hr treatment from young adult (-/+ Auxin). All strains have Psun-1 or Pgld-1 driven TIR1::mRuby. All images are maximum-intensity projections and scaled identically between -/+ Auxin. Scale bar, 5 µm. (B) Quantification of mean intensity of V5 staining per nucleus. Fluorescence intensity was measured from additive projection images after background subtraction. 30 nuclei in late pachytene/early diplotene were measured per condition. Medians (black crossbars) and means (black boxes) are shown. p-values were calculated using one-way ANOVA and post hoc pairwise t-tests (two-sided; adjusted by the Benjamini-Hochberg method). p > 0.05 between all + Auxin groups. A.U., arbitrary unit. Fig. S13. Fig. S13. (Related to Fig. 3) SUN-1, but not ZYG-12, is required for damage-induced apoptosis during meiosis. (A) Germline apoptosis is increased upon depleting ZYG-12. Apoptosis was quantified using CED-1::GFP. syp-1(me17)/nT1 heterozygous and syp-1(me17) homozygous animals were used as controls. Mean ± SD are plotted. Pairwise Mann-Whitney test was used for computing the p-values. (B) Germline apoptosis does not change upon SUN-1 depletion. Red dashed arrows indicate the direction of meiotic progression. Scale bar, 10 µm. Unpaired two-sample two-sided Mann-Whitney test was used for computing the p-value. (C) X- chromosome pairing (HIM-8) and SC assembly (SYP-1) in HA::AID::zyg-12 worms without or with auxin treatment. Yellow arrow heads indicate nuclei with unpaired HIM-8 foci, white arrow heads indicate nuclei with incomplete synapsis (HTP-3 staining devoid of SYP-1). Scale bar, 10 μm. (D) RAD-51 staining in mid-pachytene nuclei in HA::AID::zyg-12 worms without or with auxin treatment. Scale bar, 10 µm. (E) Single Z-section of non-deconvolved immunofluorescence images showing CED-1::GFP positive meiotic nuclei stain positive for SUN-1 but negative for ZYG-12 (orange arrow heads). One worm per row is shown. Scale bar, 5 µm. Fig. S14. Fig. S14. (Related to Fig. 3) The absence of the meiotic NE protein MJL-1 does not rescue nuclear collapse. Nuclear morphology in wild-type (N2) or mjl-1 null (tm1651) worms following Control RNAi or lmn-1(RNAi). Scale bar, 10 µm. Fig. S15. Fig. S15. (Related to Fig. 4) Pairing and synapsis upon depleting SAMP-1, EMR-1 or LEM- 2. X-chromosome pairing (HIM-8) and SC assembly (SYP-1 and HTP-3) upon RNAi-mediated depletion of SAMP-1 or of LEM-2 in emr-1(gk119) mutants. Insets showing X-chromosome pairing in early pachytene and bivalents formation in diakinesis under each condition. Scale bars, 10 µm. Fig. S16. Fig. S16. (Related to Fig. 4) Immunostaining of nuclear pores in germlines depleted of EMR-1, LEM-2 and LMN-1. Composite images showing a representative gonad from an emr- 1(gk119), lmn-1::AID animal with lem-2(RNAi) and Auxin treatment, DAPI in cyan and NPP-7 (marking nuclear pore complex) in red. Insets show that nuclei undergoing exacerbated collapse can still be surrounded by nuclear pore complexes marked by NPP-7. All scale bars, 10 µm. Fig. S17. Fig. S17. (Related to Fig. 4) Asymmetric distribution of ZYG-12::GFP at early pachytene NE upon co-depleting LMN-1 and SAMP-1 or EMR-1/LEM-2. (A and C) Composite images showing mRuby::SYP-3 (magenta) and ZYG-12::GFP (green) at the NE of early pachytene nuclei of indicated genotypes or treatments. All images are maximum-intensity projections and scaled identically per experiment. Scale bar, 5 µm. (B and D) Quantification of ZYG-12::GFP fluorescence as a function of angle along the circumference of NE in early pachytene nuclei. Intensity measurement was performed as in Fig. S8B. Because of the multiple bright foci/patches of LINC complexes at the NE in each early pachytene nucleus, data had to be aligned and normalized differently than that in Fig. 3B: data were aligned and mapped so that 180° corresponds to the coordinate along the NE’s circumference with the minimum-intensity of ZYG-12::GFP, which was normalized as one. In (B), N = 32 early pachytene nuclei pooled from four animals were measured for each condition. p < 2.2e-16 between the control RNAi and samp-1(RNAi) with 2-way ANOVA. In (D), N = 51 (Control RNAi) and 35 (lem-2(RNAi)) early pachytene nuclei pooled from seven or four animals were measured. p < 2.2e-16 between the control RNAi and lem-2(RNAi) with 2-way ANOVA. See Data S1. All worms were homozygous for Pgld-1::TIR1. Fig. S18. Fig. S18. (Related to Fig. 5) Isolating meiotic nuclei for stiffness measurement using mechano-NPS. (A) Schematic of the workflow of microdissecting C. elegans gonads to isolate meiotic nuclei. (B) Nuclear size measured with mechano-NPS. The nuclear size is calculated using supplemental equation 1. Welch ANOVA with Games-Howell multiple comparisons test revealed no significant differences between nuclear diameter in each group, p>0.05. Sample sizes for the Control RNAi, -Auxin, the Control RNAi, +Auxin, and the samp-1(RNAi), +Auxin were n=79, n=94, n=113 nuclei respectively. The darker region of the histogram to the left of the dotted line contains nuclei which were excluded from wCDI analysis because they were too small to experience any strain, i.e. their size was smaller than or equal to the width of the contraction channel employed. All nuclei to the right of the dotted line were included in wCDI analysis. Auxin n Embryos laid (±SD) Embryonic viability (±SD) %* Male progeny (±SD) % Table S1. Strains Psun-1::TIR1IV Psun-1::TIR1 Psun-1::TIR1 IV - + (from L1) + (from L4) lmn-1::AID::V5 I; Psun-1::TIR1 IV - lmn-1::AID::V5 I; Psun-1::TIR1 IV + (from L1) lmn-1::AID::V5 I; Psun-1::TIR1 IV + (from L4) HA::AID::zyg-12 II; Psun-1::TIR1 IV - HA::AID::zyg-12 II; Psun-1::TIR1 IV + (from L1) HA::AID::zyg-12 II; Psun-1::TIR1 IV + (from L4) sun-1::AID::V5 V; Psun-1::TIR1 IV - sun-1::AID::V5 V; Psun-1::TIR1 IV + (from L1) sun-1::AID::V5 V; Psun-1::TIR1 IV + (from L4) emr-1(gk119) I; lem-2::HA::AID II; Psun- 1::TIR1 IV - 6 4 3 7 6 3 3 3 3 6 5 3 6 225.5 ± 18.9 106.1 ± 3.2 228.0 ± 26.5 106.8 ± 3.8 245.7 ± 43.8 108.0 ± 2.3 252.1 ± 30.4 106.4 ± 5.1 83.3 ± 45.8 39.7 ± 44.0 9.6 ± 23.5 0.0 ± 0.0 254.3 ± 76.4 104.1 ± 4.7 0.0 ± 0.0 0.0 ± 0.0 0.0 ± 0.0 0.0 ± 0.0 225.5 ± 53.5 81.3 ± 21.7 31.0 ± 26.5 33.3 ± 35.3 0.0 ± 0.0 0.0 ± 0.0 216.3 ± 22.5 106.4 ± 2.4 0.0 ± 0.0 0.0 ± 0.0 0.1 ± 0.2 0.1 ± 0.2 0.0 ± 0.0 0.0 ± 0.0 0.0 ± 0.0 0.0 ± 0.0 0.0 ± 0.0 0.7 ± 0.7 0.0 ± 0.0 0.0 ± 0.0 0.1 ± 0.2 * Embryonic viability of >100% reflects the fact that some embryos are overlooked when counting, but adult worms hatched from these embryos are easier to count accurately. Table S1. Quantification of brood size, embryonic viability, and male self-progeny of worm strains with AID alleles generated in this study. Sample size n indicates the number of broods examined. Note that high levels of uncertainty are usually associated with counting deformed eggs resulting from auxin treatment. Table S2. Allele Genotype ie137 lmn-1(ie137[lmn-1::AID::V5]) ie138 zyg-12(ie138[HA::AID::zyg-12]) ie139 sun-1(ie139[sun-1::AID::V5]) ie140 lem-2(ie140[lem-2::HA::AID]) ie203 atg-7(ie203[atg-7::AID::HA]) Table S2. Alleles generated in this study. Information about mutagenesis Internally tagged; generated using dpy-10 Co-CRISPR in ieSi38[sun-1p::TIR1::mRuby::sun-1 3'UTR, Cbr-unc-119(+)] IV generated using dpy-10 Co-CRISPR in ieSi38[sun- 1p::TIR1::mRuby::sun-1 3'UTR, Cbr-unc-119(+)] IV Internally tagged; generated using dpy-10 Co-CRISPR in ieSi38[sun-1p::TIR1::mRuby::sun-1 3'UTR, Cbr-unc-119(+)] IV generated using dpy-10 Co-CRISPR in ieSi38[sun- 1p::TIR1::mRuby::sun-1 3'UTR, Cbr-unc-119(+)] IV generated using dpy-10 Co-CRISPR in ieSi38[sun- 1p::TIR1::mRuby::sun-1 3'UTR, Cbr-unc-119(+)] IV Table S3. Strains C. elegans: N2 Bristol, wild isolate C.elegans: bcIs39(ced-1::GFP) V C. elegans: dnc-1::GFP Source Caenorhabditis Genetics Center Caenorhabditis Genetics Center Zhang, Skop and White (89) C.elegans: syp-1 (me17) bcIs39(ced-1::GFP) V/ nT1 [qIs51] (IV;V) Bhalla et al. (43) C.elegans: ieDf2/mIs11 IV C. elegans: ieSi38[sun-1p::TIR1::mRuby::sun-1 3’ UTR, Cbr-unc-119 (+)] IV C.elegans: ieSi64[gld-1p::TIR1::mRuby::gld-1 3'UTR, Cbr-unc-119(+)] II; unc- 119(ed3) III C.elegans: ieSi65[sun-1p::TIR1::sun-1 3'UTR, Cbr-unc-119(+)] II; unc-119(ed3) III C. elegans: meIs8[pie-1p::GFP::cosa-1, unc-119(+)] II; spo-11(ie59[spo- 11::AID::3xFLAG]), ieSi38[sun-1p::TIR1::mRuby::sun-1 3'UTR, Cbr-unc-119(+)] IV C. elegans: lmn-1(ie137[lmn-1::AID::V5]) I; unc-119 (ed3) III; ieSi38 [Psun- 1::TIR1::mRuby::sun-1 3’UTR, cb-unc-119(+)] IV C. elegans: lmn-1(ie137[lmn-1::AID::V5]) I; unc-119 (ed3) III; ieSi38 [Psun- 1::TIR1::mRuby::sun-1 3’UTR, cb-unc-119(+)] IV; bcIs39 (ced-1::GFP) V C. elegans: lmn-1(ie137[lmn-1::AID::V5]) I; ;ced-4(n1162) III; ieSi38 [Psun- 1::TIR1::mRuby::sun-1 3’UTR, cb-unc-119(+)] IV C. elegans: lmn-1(ie137[lmn-1::AID::V5]) I; ieSi65[sun-1p::TIR1::sun-1 3'UTR, Cbr-unc-119(+)] II; unc-119(ed3) III; ieSi21 (sun-1::mRuby) IV C. elegans: unc-119 (ed3) III; ieSi38 [Psun-1::TIR1::mRuby::sun-1 3’UTR, cb-unc- 119(+)] IV; sun-1(ie139[sun-1::AID::V5]) V C. elegans: lmn-1(ie137[lmn-1::AID::V5]) I; ieSi64[gld-1p::TIR1::mRuby::gld-1 3'UTR, Cbr-unc-119(+)] II; unc-119(ed3) III; sun-1(ie139[sun-1::AID::V5]) V C. elegans: zyg-12(ie138[HA::AID::zyg-12]) II; unc-119 (ed3) III; ieSi38 [Psun- 1::TIR1::mRuby::sun-1 3’UTR, cb-unc-119(+)] IV C. elegans: lmn-1(ie137[lmn-1::AID::V5]) I; zyg-12(ie138[HA::AID::zyg-12]) II; unc- 119 (ed3) III; ieSi38 [Psun-1::TIR1::mRuby::sun-1 3’UTR, cb-unc-119(+)] IV C. elegans: lmn-1(ie137[lmn-1::AID::V5]), emr-1(gk119) I; unc-119 (ed3) III; ieSi38 [Psun-1::TIR1::mRuby::sun-1 3’UTR, cb-unc-119(+)] IV C. elegans: lem-2(ie140[lem-2::HA::AID]) II; ieSi64[gld-1p::TIR1::mRuby::gld-1 3'UTR, Cbr-unc-119(+)] II; unc-119(ed3) III; C. elegans: lmn-1(ie137[lmn-1::AID::V5]) I; lem-2(ie140[lem-2::HA::AID]), ieSi64[gld-1p::TIR1::mRuby::gld-1 3'UTR, Cbr-unc-119(+)] II; unc-119(ed3) III; C. elegans: lmn-1(ie137[lmn-1::AID::V5]), emr-1(gk119) I; unc-119 (ed3) III; ieSi38 [Psun-1::TIR1::mRuby::sun-1 3’UTR, cb-unc-119(+)] IV; sun-1(ie139[sun- 1::AID::V5]) V C. elegans: syp-3 (ok857) I; ieSi64[gld-1p::TIR1::mRuby::gld-1 3'UTR, Cbr-unc- 119(+)] II; unc-119(ed3) III; ieSi19 (mRuby::SYP-3),ojIs9 [zyg-12(all)::GFP + unc- 119(+)] IV; sun-1(ie139[sun-1::AID::V5]) V C. elegans: lmn-1(ie137[lmn-1::AID::V5]) I; ieSi64[gld-1p::TIR1::mRuby::gld-1 3'UTR, Cbr-unc-119(+)] II; unc-119(ed3) III; ieSi19 (mRuby::SYP-3),ojIs9 [zyg- 12(all)::GFP + unc-119(+)] IV; sun-1(ie139[sun-1::AID::V5]) V C. elegans: lmn-1(ie137[lmn-1::AID::V5]) I; ieSi64[gld-1p::TIR1::mRuby::gld-1 3'UTR, Cbr-unc-119(+)] II; unc-119(ed3) III; ieSi19 (mRuby::SYP-3),ojIs9 [zyg- 12(all)::GFP + unc-119(+)] IV; C. elegans: lmn-1(ie137[lmn-1::AID::V5]), emr-1(gk119) I; ieSi64[gld- 1p::TIR1::mRuby::gld-1 3'UTR, Cbr-unc-119(+)] II; unc-119(ed3) III; ieSi19 (mRuby::SYP-3),ojIs9 [zyg-12(all)::GFP + unc-119(+)] IV; C. elegans: lmn-1(ie137[lmn-1::AID::V5]) I; meIs8[pie-1p::GFP::cosa-1, unc- 119(+)] II; unc-119 (ed3) III; ieSi38 [Psun-1::TIR1::mRuby::sun-1 3’UTR, cb-unc- 119(+)] IV C. elegans: lmn-1(ie137[lmn-1::AID::V5]) I; meIs8[pie-1p::GFP::cosa-1, unc- 119(+)] II; unc-119 (ed3) III; spo-11(ie59[spo-11::AID::3xFLAG]), ieSi38 [Psun- 1::TIR1::mRuby::sun-1 3’UTR, cb-unc-119(+)] IV Identifier N2 CA195 (MD701) CA846 CA885 CA998 CA1199 CA1352 CA1353 CA1423 CA1532 CA1561 CA1562 CA1563 CA1564 CA1565 CA1566 CA1567 CA1568 CA1569 CA1570 CA1571 Harper et al. (57); Caenorhabditis Genetics Center Zhang et al. (35); Caenorhabditis Genetics Center Zhang et al. (35); Caenorhabditis Genetics Center Zhang et al. (35); Caenorhabditis Genetics Center Zhang et al. (82); Caenorhabditis Genetics Center This paper This paper This paper This paper This paper This paper This paper This paper This paper This paper This paper This paper This paper CA1572 This paper CA1573 This paper CA1574 This paper CA1575 This paper CA1576 This paper CA1577 40 C. elegans: lmn-1(ie137[lmn-1::AID::V5]) I; unc-119 (ed3) III; spo-11(ie59[spo- 11::AID::3xFLAG]), ieSi38 [Psun-1::TIR1::mRuby::sun-1 3’UTR, cb-unc-119(+)] IV; sun-1(ie139[sun-1::AID::V5]) V C. elegans: zyg-12(ie138[HA::AID::zyg-12]) II; unc-119 (ed3) III; ieSi38 [Psun- 1::TIR1::mRuby::sun-1 3’UTR, cb-unc-119(+)] IV; bcIs39 (ced-1::GFP) V C. elegans: emr-1(gk119) I; lem-2(ie140[lem-2::HA::AID]) II; ieSi64[gld- 1p::TIR1::mRuby::gld-1 3'UTR, Cbr-unc-119(+)] II; unc-119(ed3) III; C. elegans: mjl-1(tm1651) I; hT2 [bli-4(e937) let-?(q782) qIs48] (I,III) C. elegans: unc-119 (ed3) III; atg-7(ie203[atg-7::AID::HA]) IV; ieSi38 [Psun- 1::TIR1::mRuby::sun-1 3’UTR, cb-unc-119(+)] IV This paper CA1578 This paper This paper CA1579 CA1677 National Bioresource Project CA1728 This paper CA1729 Table S3. Genotypes of worm strains generated and used in this study. Table S4. Transgenes crRNAs and repair templates (mostly gBlock) lmn-1::AID::V5 in ie137 HA::AID::zyg-12 in ie138 sun-1::AID::V5 in ie139 lem-2::HA::AID in ie140 atg-7::AID::HA in ie203 dpy-10 5’ – AGAAGTTCGTCACAAGAGAC – 3’; 5’ - TCTGGAAGAAGATCTCGCTTTTGCTCTTCAA CAGCACAAGGGAGAACTTGAAGAAGTTCGcC ACAAGAGgCAGGTCGACATGACAACCTACG GCGGCGGAGGATCCatgcctaaagatccagccaaac ctccggccaaggcacaagttgtgggatggccaccggtgagatc ataccggaagaacgtgatggtttcctgccaaaaatcaagcggtg gcccggaggcggcggcgttcgtgaagggaggatccggaGG AAAGCCAATTCCAAACCCACTTCTTGGACTC GACTCCACCGCCAAGCAGATTAATGATGAGT ATCAATCTAAGCTT -3’ 5’ – GAATCTGAGTCGTCAGACAA – 3’; 5’ – aaaatctatcaatttcttttttttcagaacaaaatcatgTACCCAT ACGATGTTCCAGATTACGCTggaggatccggaatg cctaaagatccagccaaacctccggccaaggcacaagttgtgg gatggccaccggtgagatcataccggaagaacgtgatggtttcct gccaaaaatcaagcggtggcccggaggcggcggcgttcgtga agGGCGGCGGAGGATCCGGAGGAGGAGGC AGTGGAGGCGGCGGTTCTGGCGGTGGCGG CTCAGGCGGAGGTGGATCGTTAGACCTGAC AAACAAAGAGTCCGAGTCTTCAGACAACGGA AATAGCAAGTACGAAGATTCCATAGACGGAC GA – 3’ 5’ – GCTGGAATATCGCATTCGCA – 3’; 5’ – TACAAGGAGCATTTTAGCTACAAAGAAATCA CTTCGATGAAGAAGGAAATGTGGTATGACTG GCTGGAATATCGCATcCGtGGCGGCGGAGGA TCCatgcctaaagatccagccaaacctccggccaaggcaca agttgtgggatggccaccggtgagatcataccggaagaacgtg atggtttcctgccaaaaatcaagcggtggcccggaggcggcgg cgttcgtgaagggaggatccggaGGAAAGCCAATTCC AAACCCACTTCTTGGACTCGACTCCACCATG GTTCGGCGTCGTTTTGTTCCAACGTGGGCC CAGTTTAAACGTACTCTT – 3’ 5’ – TGTGCCGTGTGGAAGTGGAT – 3’; 5’ – CTACCGATGTTCTTGTGCTTCCGTCTGGAAA TGAGTGcGCtGTcTGGAAaTGGATCGGAAATC AGTCTCAGAAGAGATGGTACCCATACGATGT TCCAGATTACGCTggaggatccggaatgcctaaagatc cagccaaacctccggccaaggcacaagttgtgggatggccacc ggtgagatcataccggaagaacgtgatggtttcctgccaaaaat caagcggtggcccggaggcggcggcgttcgtgaagTAGatc attgttttgctgtataatttttcgatttt– 3’ 5’ – GATGATGAAGATTTCTGAat – 3’; 5’ – CAGAACTCTGTTAATGCTATTGATATCGATTT TGAGGATGATGAAGATTTCGGCGGCGGAGG ATCCGGAGGAGGAGGCAGTGGAGGCGGCG GTTCTGGCGGTGGCGGCTCAatgcctaaagatcca gccaaacctccggccaaggcacaagttgtgggatggccaccg gtgagatcataccggaagaacgtgatggtttcctgccaaaaatc aagcggtggcccggaggcggcggcgttcgtgaagggaggatc cggaTACCCATACGATGTTCCAGATTACGCTT GAattggtcgcctcaaatttttaccttttctgtataattg – 3’ 5’ – GCTACCATAGGCACCACGAG – 3’; 5’ – ATACGGCAAGATGAGAATGACTGGAAACCGT ACCGCATGCGGTGCCTATGGTAGCGGAGCT TCACATGGCTTCAGA – 3’ (ssDNA repair template) Genotyping primer names oCL41 (F) Primer sequences 5' - AAA GCA GAA CAT CAC TCT TCG TGA CAC CGT AGA AG -3’; Fragment sizes oCL42 (R) 5' - TTT GAT GCA AAT TGT TCT TGA ACT GAG CAC GCA TCT C -3’ WT, 277bp; inserted, 481bp oCL95 (F) 5’ – TTGTAAACTCTACCAGCC T -3’; oCL96 (R) 5’ – TCAGAGGTAGTTTAGTGG C -3’; WT, 401bp; inserted, 650bp oCL101 (F) 5'- CTT CGA TGA AGA AGG AAA TGT GGT ATG ACT GGC -3'; oCL102 (R) 5'- CTC TTC GAT TGC CGA CTC TTT CCA TCC TTT -3'; WT, 456bp; inserted, 660bp oCL137 (F) 5’ – GAAGCTCTACGAGCTCAT C – 3’; oCL138 (R) 5’ – gtcattgtgataccttaggc – 3’; WT, 379bp; inserted, 553bp Atg-7-Fw (F) 5’ – GTCGTCTCGAAGAAGTCA C – 3’; WT, 186bp; inserted, 420bp Atg-7-Rw (R) 5’ – ggaggcaaaatagaatcac – 3’; N.A. N.A. N.A. Table S4. Sequences of crRNAs, repair templates, and DNA primers used to genotype edited progeny. Table S5. Target gene GenePairs Name Plate dnc-1 dlc-1 lis-1 dhc-1 dli-1 dylt-1 lem-2 samp-1 lmn-1 ZK593.5 T26A5.9 T03F6.5 T21E12.4 C39E9.14 F13G3.4 W01G7.5 T24F1.2 DY3.2 111 75 88 4 116 11 63 58 14 Table S5. RNAi clones used in this study. Well B12 H6 F4 H2 A3 E6 B2 A6 D12 Table S6. Channel A Channel B 5.00 x 10-5 ∆𝐼𝐼𝑠𝑠/𝐼𝐼 4.23 x 10-5 0.768 x 10-5 0.719 x 10-5 𝜎𝜎𝐼𝐼 De 10.2 11.1 n 25 21 Table S6. Measuring channel effective diameters. Polystyrene microspheres (Sigma-Aldrich) with a diameter of 2 µm ± 0.05 µm suspended in the 1x Wash Buffer were measured with the mechano-NPS platform. One platform consisted of two independent microfluidic channels for measurement (Channel A and B), which acted as two independent devices. Polystyrene beads were measured in both devices to calculate their effective diameters. corresponds to the average ratio of current pulse amplitude to baseline current produced as the polystyrene beads transited the sizing segment and diameter, as calculated from Equation S1 using beads. n corresponds to the number of beads measured in each device. corresponds to the standard deviation. De is the effective and the known diameter of the polystyrene ∆𝑰𝑰𝒔𝒔/𝑰𝑰 𝝈𝝈𝑰𝑰 ∆𝑰𝑰𝒔𝒔/𝑰𝑰 Table S7. ( m/msec) Channel A Channel B 10.84 flow 10.15 U µ m/msec) ( 0.80 𝜎𝜎𝑈𝑈 1.23 µ n 109 68 Table S7. Measuring fluid velocity. Uflow is the approximate fluid velocity ( m/msec), its standard deviation ( m/msec), and n is the number of nuclei measured in each device (Channel A and B). The Uflow in a device is the average nuclear velocity in the sizing segment of 𝜎𝜎𝑈𝑈 is all nuclei measured in that device (Equation S3). One platform consisted of two independent microfluidic channels for measurement, Channel A and B, which acted as independent devices, therefore Uflow was calculated for each device. Three replicas of both Channel A and B were used to calculate the Uflow values reported. Only nuclei whose wCDI was measured (nuclei that underwent > 0% strain) were included in the Uflow calculation for each device. µ µ Equations Particle size was calculated using, 3 𝑑𝑑 is the magnitude of the current drop produced by the particle as it transits the sizing 𝐷𝐷𝑒𝑒� where 1−0.8� segment (Fig. 5B), I is the baseline current, d is the diameter of the particle, L is the overall channel length, and De is the effective diameter (Table S6) (96). ∆𝐼𝐼𝑠𝑠 𝐿𝐿 � � (Eq. S1) ∆𝐼𝐼𝑠𝑠 𝐼𝐼 = 3 𝑑𝑑 2 𝐷𝐷𝑒𝑒 1 The whole-cell deformability index (wCDI) was calculated using, (Eq. S2) 𝑣𝑣𝑐𝑐 where dn is the nuclear diameter, h is the channel height, vc is the velocity of the nucleus in the 𝑈𝑈𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓 � contraction segment, and Uflow is the fluid velocity (67). The contraction segment velocity is , where Lc is the contraction segment length and tc is the nuclear transit defined as time in the contraction segment. The fluid velocity, Uflow, can be approximated as the average nuclear velocity in the sizing segment, 𝑣𝑣𝑐𝑐 = 𝐿𝐿𝑐𝑐 𝑡𝑡𝑐𝑐⁄ 𝑤𝑤𝑤𝑤𝐷𝐷𝐼𝐼 = 𝑑𝑑𝑛𝑛 ℎ � (Eq. S3) ∑ 𝑣𝑣𝑠𝑠 𝑈𝑈𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓~𝑣𝑣𝑠𝑠_𝑎𝑎𝑣𝑣𝑎𝑎 = where vs is the nuclear velocity in the sizing segment and n is the number of nuclei measured. The sizing segment velocity is defined as , where Ls is the sizing segment length and ts is the nuclear transit time through the sizing segment. An average of all nuclei’s vs is used to 𝑣𝑣𝑠𝑠 = 𝐿𝐿𝑠𝑠 𝑡𝑡𝑠𝑠⁄ calculate Uflow to take into account variations in the fluid velocity due to off-axis hydrodynamic effects (97). For the experiments reported here, we calculated Uflow for each device (Channel A and B) utilized (Table S7). 𝑛𝑛 Movie S1. (separate file) Nuclear envelope dynamics marked by SUN-1::mRuby in late prophase after LMN-1 depletion. Two representative time-lapse recordings of SUN-1::mRuby at the NE of meiotic nuclei in late meiotic prophase after LMN-1 depletion are shown. Arrow points to a collapsing nucleus. Note many meiotic nuclei with asymmetrically distributed SUN-1::mRuby at the NE. Time stamp is hr:min:sec. Scale bars, 10 µm. Movie S2. (separate file) Tracking SUN-1::mRuby patches in early meiosis. An example of drift-corrected time-lapse recordings of SUN-1::mRuby patch movement on the NE of transition zone nuclei, using reference frame followed by 3D particle tracking in Imaris. Time stamp is hr:min:sec. Scale bar, 5 µm. Movie S3. (separate file) LMN-1 depletion changes the mobility of SUN-1::mRuby patches throughout prophase. Side-by-side comparison of time-lapse recordings of the movement of SUN-1::mRuby patches or foci during different stages of meiosis, with or without LMN-1 depletion. Time stamp is hr:min:sec. Scale bars, 5 µm. Movie S4. (separate file) Dynamics of diplotene nuclear collapse. Time-lapse recording of a dual-color labeled oocyte nucleus during diplotene collapse. Green, ZYG-12::GFP; Magenta, mRuby::SYP-3. Scale bar, 5 µm. Movie S5. (separate file) Contact between NE and SC happens during diplotene nuclear collapse. Time-lapse recording of another dual-color labeled diplotene nucleus during collapse. The frame showing initial contact between NE and SC is annotated. Green, ZYG-12::GFP; Magenta, mRuby::SYP- 3. Time stamp is hr:min:sec. Scale bar, 2 µm. Data S1. (separate file) Statistical source data. Additional output of statistical tests performed in this study (with related figures or supplementary figure numbers). REFERENCES AND NOTES 1. N. Bhalla, A. F. Dernburg, Prelude to a division. Annu. Rev. Cell Dev. Biol. 24, 397–424 (2008). 2. Z. Yu, Y. Kim, A. F. Dernburg, Meiotic recombination and the crossover assurance checkpoint in Caenorhabditis elegans. Semin. Cell Dev. Biol. 54, 106–116 (2016). 3. A. Sato, B. Isaac, C. M. Phillips, R. Rillo, P. M. Carlton, D. J. Wynne, R. A. Kasad, A. F. Dernburg, Cytoskeletal forces span the nuclear envelope to coordinate meiotic chromosome pairing and synapsis. Cell 139, 907–919 (2009). 4. D. J. Wynne, O. Rog, P. M. Carlton, A. F. Dernburg, Dynein-dependent processive chromosome motions promote homologous pairing in C. elegans meiosis. J. Cell Biol. 196, 47–64 (2012). 5. C.-Y. Lee, H. F. Horn, C. L. Stewart, B. Burke, E. Bolcun-Filas, J. C. Schimenti, M. E. Dresser, R. J. Pezza, Mechanism and regulation of rapid telomere prophase movements in mouse meiotic chromosomes. Cell Rep. 11, 551–563 (2015). 6. J. Link, M. Leubner, J. Schmitt, E. Göb, R. Benavente, K. T. Jeang, R. Xu, M. Alsheimer, Analysis of meiosis in SUN1 deficient mice reveals a distinct role of SUN2 in mammalian meiotic LINC complex formation and function. PLOS Genet. 10, e1004099 (2014). 7. Y. Luo, I.-W. Lee, Y.-J. Jo, S. Namgoong, N.-H. Kim, Depletion of the LINC complex disrupts cytoskeleton dynamics and meiotic resumption in mouse oocytes. Sci. Rep. 6, 20408 (2016). 8. B. Burke, LINC complexes as regulators of meiosis. Curr. Opin. Cell Biol. 52, 22–29 (2018). 9. W. Chang, H. J. Worman, G. G. Gundersen, Accessorizing and anchoring the LINC complex for multifunctionality. J. Cell Biol. 208, 11–22 (2015). 10. J. Fan, H. Jin, B. A. Koch, H.-G. Yu, Mps2 links Csm4 and Mps3 to form a telomere-associated LINC complex in budding yeast. Life Sci. Alliance 3, e202000824 (2020). 11. H. J. Kim, C. Liu, A. F. Dernburg, How and why chromosomes interact with the cytoskeleton during meiosis. Genes 2022;13:901. 12. K. Zhou, M. M. Rolls, D. H. Hall, C. J. Malone, W. Hanna-Rose, A ZYG-12–dynein interaction at the nuclear envelope defines cytoskeletal architecture in the C. elegans gonad. J. Cell Biol. 186, 229– 241 (2009). 13. O. Rog, Abby F. Dernburg, Direct visualization reveals kinetics of meiotic chromosome synapsis. Cell Rep. 10, 1639–1645 (2015). 14. M. Zhang, in Encyclopedia of Reproduction, M. K. Skinner, Ed. (Academic Press, ed. 2, 2018), pp. 153–158. 15. G. Nagamatsu, S. Shimamoto, N. Hamazaki, Y. Nishimura, K. Hayashi, Mechanical stress accompanied with nuclear rotation is involved in the dormant state of mouse oocytes Sci. Adv. 5, eaav9960 (2019). 16. Y. Tsatskis, R. Rosenfeld, J. D. Pearson, C. Boswell, Y. Qu, K. Kim, L. Fabian, A. Mohammad, X. Wang, M. I. Robson, K. Krchma, J. Wu, J. Gonçalves, D. Hodzic, S. Wu, D. Potter, L. Pelletier, W. H. Dunham, A. C. Gingras, Y. Sun, J. Meng, D. Godt, T. Schedl, B. Ciruna, K. Choi, J. R. B. Perry, R. Bremner, E. C. Schirmer, J. A. Brill, A. Jurisicova, H. McNeill, The NEMP family supports metazoan fertility and nuclear envelope stiffness Sci. Adv. 6, eabb4591 (2020). 17. A. E. Goldman, G. Maul, P. M. Steinert, H. Y. Yang, R. D. Goldman, Keratin-like proteins that coisolate with intermediate filaments of BHK-21 cells are nuclear lamins. Proc. Natl. Acad. Sci. U.S.A. 83, 3839–3843 (1986). 18. U. Aebi, J. Cohn, L. Buhle, L. Gerace, The nuclear lamina is a meshwork of intermediate-type filaments. Nature 323, 560–564 (1986). 19. F. D. McKeon, M. W. Kirschner, D. Caput, Homologies in both primary and secondary structure between nuclear envelope and intermediate filament proteins. Nature 319, 463–468 (1986). 20. P. M. Davidson, J. Lammerding, Broken nuclei—Lamins, nuclear mechanics, and disease. Trends Cell Biol. 24, 247–256 (2014). 21. J. Swift, I. L. Ivanovska, A. Buxboim, T. Harada, P. C. D. P. Dingal, J. Pinter, J. D. Pajerowski, K. R. Spinler, J. W. Shin, M. Tewari, F. Rehfeldt, D. W. Speicher, D. E. Discher, Nuclear lamin-A scales with tissue stiffness and enhances matrix-directed differentiation. Science 341, 1240104 (2013). 22. J. L. V. Broers, F. C. S. Ramaekers, G. Bonne, R. B. Yaou, C. J. Hutchison, Nuclear lamins: Laminopathies and their role in premature ageing. Physiol. Rev. 86, 967–1008 (2006). 23. H. J. Worman, Nuclear lamins and laminopathies. J. Pathol. 226, 316–325 (2012). 24. C. M. Denais, R. M. Gilbert, P. Isermann, A. L. McGregor, M. te Lindert, B. Weigelin, P. M. Davidson, P. Friedl, K. Wolf, J. Lammerding, Nuclear envelope rupture and repair during cancer cell migration. Science 352, 353–358 (2016). 25. A. J. Earle, T. J. Kirby, G. R. Fedorchak, P. Isermann, J. Patel, S. Iruvanti, S. A. Moore, G. Bonne, L. L. Wallrath, J. Lammerding, Mutant lamins cause nuclear envelope rupture and DNA damage in skeletal muscle cells. Nat. Mater. 19, 464–473 (2020). 26. E. M. Hatch, A. H. Fischer, T. J. Deerinck, M. W. Hetzer, Catastrophic nuclear envelope collapse in cancer cell micronuclei. Cell 154, 47–60 (2013). 27. M. Kneissig, K. Keuper, M. S. de Pagter, M. J. van Roosmalen, J. Martin, H. Otto, V. Passerini, A. Campos Sparr, I. Renkens, F. Kropveld, A. Vasudevan, J. M. Sheltzer, W. P. Kloosterman, Z. Storchova, Micronuclei-based model system reveals functional consequences of chromothripsis in human cells. eLife 8, e50292 (2019). 28. J. Link, D. Paouneskou, M. Velkova, A. Daryabeigi, T. Laos, S. Labella, C. Barroso, S. Pacheco Piñol, A. Montoya, H. Kramer, A. Woglar, A. Baudrimont, S. M. Markert, C. Stigloher, E. Martinez- Perez, A. Dammermann, M. Alsheimer, M. Zetka, V. Jantsch, Transient and partial nuclear lamina disruption promotes chromosome movement in early meiotic prophase. Dev. Cell 45, 212–225.e7 (2018). 29. J. Link, D. Jahn, J. Schmitt, E. Göb, J. Baar, S. Ortega, R. Benavente, M. Alsheimer, The meiotic nuclear lamina regulates chromosome dynamics and promotes efficient homologous recombination in the mouse. PLOS Genet. 9, e1003261 (2013). 30. M. C. Vantyghem, D. Vincent-Desplanques, F. Defrance-Faivre, J. Capeau, C. Fermon, A. S. Valat, O. Lascols, A. C. Hecart, P. Pigny, B. Delemer, C. Vigouroux, J. L. Wemeau, Fertility and obstetrical complications in women with LMNA-related familial partial lipodystrophy. J. Clin. Endocrinol. Metabol. 93, 2223–2229 (2008). 31. J. Liu, T. R. Ben-Shahar, D. Riemer, M. Treinin, P. Spann, K. Weber, A. Fire, Y. Gruenbaum, Essential roles for Caenorhabditis elegans lamin gene in nuclear organization, cell cycle progression, and spatial organization of nuclear pore complexes. Mol. Biol. Cell 11, 3937–3947 (2000). 32. R. A. Green, H. L. Kao, A. Audhya, S. Arur, J. R. Mayers, H. N. Fridolfsson, M. Schulman, S. Schloissnig, S. Niessen, K. Laband, S. Wang, D. A. Starr, A. A. Hyman, T. Schedl, A. Desai, F. Piano, K. C. Gunsalus, K. Oegema, A high-resolution C. elegans essential gene network based on phenotypic profiling of a complex tissue. Cell 145, 470–482 (2011). 33. G. Huelgas-Morales, M. Sanders, G. Mekonnen, T. Tsukamoto, D. Greenstein, Decreased mechanotransduction prevents nuclear collapse in a Caenorhabditis elegans laminopathy. Proc. Natl. Acad. Sci. U.S.A. 117, 31301–31308 (2020). 34. E. Haithcock, Y. Dayani, E. Neufeld, A. J. Zahand, N. Feinstein, A. Mattout, Y. Gruenbaum, J. Liu, Age-related changes of nuclear architecture in Caenorhabditis elegans. Proc. Natl. Acad. Sci. U.S.A. 102, 16690–16695 (2005). 35. L. Zhang, J. D. Ward, Z. Cheng, A. F. Dernburg, The auxin-inducible degradation (AID) system enables versatile conditional protein depletion in C. elegans. Development 142, 4374–4384 (2015). 36. G. Velez-Aguilera, S. Nkombo Nkoula, B. Ossareh-Nazari, J. Link, D. Paouneskou, L. van Hove, N. Joly, N. Tavernier, J. M. Verbavatz, V. Jantsch, L. Pintard, PLK-1 promotes the merger of the parental genome into a single nucleus by triggering lamina disassembly. eLife 9, e59510 (2020). 37. T. Dechat, S. A. Adam, P. Taimen, T. Shimi, R. D. Goldman, Nuclear lamins. Cold Spring Harb. Perspect. Biol. 2, a000547 (2010). 38. C. Rinaldo, P. Bazzicalupo, S. Ederle, M. Hilliard, A. La Volpe, Roles for Caenorhabditis elegans rad-51 in meiosis and in resistance to ionizing radiation during development. Genetics 160, 471–479 (2002). 39. R. Yokoo, K. A. Zawadzki, K. Nabeshima, M. Drake, S. Arur, A. M. Villeneuve, COSA-1 reveals robust homeostasis and separable licensing and reinforcement steps governing meiotic crossovers. Cell 149, 75–87 (2012). 40. A. Penkner, L. Tang, M. Novatchkova, M. Ladurner, A. Fridkin, Y. Gruenbaum, D. Schweizer, J. Loidl, V. Jantsch, The nuclear envelope protein matefin/SUN-1 is required for homologous pairing in C. elegans meiosis. Dev. Cell 12, 873–885 (2007). 41. T. Bohr, G. Ashley, E. Eggleston, K. Firestone, N. Bhalla, Synaptonemal complex components are required for meiotic checkpoint function in Caenorhabditis elegans. Genetics 204, 987–997 (2016). 42. S. A. Raiders, M. D. Eastwood, M. Bacher, J. R. Priess, Binucleate germ cells in Caenorhabditis elegans are removed by physiological apoptosis. PLOS Genet. 14, e1007417 (2018). 43. N. Bhalla, A. F. Dernburg, A conserved checkpoint monitors meiotic chromosome synapsis in Caenorhabditis elegans. Science 310, 1683–1686 (2005). 44. T. Bohr, C. R. Nelson, E. Klee, N. Bhalla, Spindle assembly checkpoint proteins regulate and monitor meiotic synapsis in C. elegans. J. Cell Biol. 211, 233–242 (2015). 45. A. J. MacQueen, M. P. Colaiácovo, K. McDonald, A. M. Villeneuve, Synapsis-dependent and - independent mechanisms stabilize homolog pairing during meiotic prophase in C. elegans. Genes Dev. 16, 2428–2442 (2002). 46. A. Gartner, S. Milstein, S. Ahmed, J. Hodgkin, M. O. Hengartner, A conserved checkpoint pathway mediates DNA damage–induced apoptosis and cell cycle arrest in C. elegans. Mol. Cell 5, 435–443 (2000). 47. Y. Zhang, L. Yan, Z. Zhou, P. Yang, E. Tian, K. Zhang, Y. Zhao, Z. Li, B. Song, J. Han, L. Miao, H. Zhang, SEPA-1 mediates the specific recognition and degradation of P granule components by autophagy in C. elegans. Cell 136, 308–321 (2009). 48. Y. Tian, Z. Li, W. Hu, H. Ren, E. Tian, Y. Zhao, Q. Lu, X. Huang, P. Yang, X. Li, X. Wang, A. L. Kovács, L. Yu, H. Zhang, C. elegans screen identifies autophagy genes specific to multicellular organisms. Cell 141, 1042–1055 (2010). 49. O. Rog, S. Köhler, A. F. Dernburg, The synaptonemal complex has liquid crystalline properties and spatially regulates meiotic recombination factors. eLife 6, e21455 (2017). 50. I. L. Minn, M. M. Rolls, W. Hanna-Rose, C. J. Malone, SUN-1 and ZYG-12, mediators of centrosome–nucleus attachment, are a functional SUN/KASH pair in Caenorhabditis elegans. Mol. Biol. Cell 20, 4586–4595 (2009). 51. M. Terasawa, M. Toya, F. Motegi, M. Mana, K. Nakamura, A. Sugimoto, Caenorhabditis elegans ortholog of the p24/p22 subunit, DNC-3, is essential for the formation of the dynactin complex by bridging DNC‐1/p150Glued and DNC-2/dynamitin. Genes Cells 15, 1145–1157 (2010). 52. R. H. Harders, T. H. Morthorst, A. D. Lande, M. O. Hesselager, O. A. Mandrup, E. Bendixen, A. Stensballe, A. Olsen, Dynein links engulfment and execution of apoptosis via CED-4/Apaf1 in C. elegans. Cell Death Dis. 9, 1012 (2018). 53. C. Liu, J.-Z. Chuang, C.-H. Sung, Y. Mao, A dynein independent role of Tctex-1 at the kinetochore. Cell Cycle 14, 1379–1388 (2015). 54. S. M. O'Rourke, M. D. Dorfman, J. C. Carter, B. Bowerman, Dynein modifiers in C. elegans: Light chains suppress conditional heavy chain mutants. PLOS Genet. 3, e128 (2007). 55. Y. B. Tzur, A. Margalit, N. Melamed-Book, Y. Gruenbaum, Matefin/SUN-1 is a nuclear envelope receptor for CED-4 during Caenorhabditis elegans apoptosis. Proc. Natl. Acad. Sci. U.S.A. 103, 13397–13402 (2006). 56. C. M. Phillips, A. F. Dernburg, A family of zinc-finger proteins is required for chromosome-specific pairing and synapsis during meiosis in C. elegans. Dev. Cell 11, 817–829 (2006). 57. N. C. Harper, R. Rillo, S. Jover-Gil, Z. J. Assaf, N. Bhalla, A. F. Dernburg, Pairing centers recruit a polo-like kinase to orchestrate meiotic chromosome dynamics in C. elegans. Dev. Cell 21, 934–947 (2011). 58. H. J. Kim, C. Liu, L. Zhang, A. F. Dernburg, MJL-1 is a nuclear envelope protein required for homologous chromosome pairing and regulation of synapsis during meiosis in C. elegans Sci. Adv. 9, eadd1453 (2023). 59. M. B. Gerstein, Z. J. Lu, E. L. van Nostrand, C. Cheng, B. I. Arshinoff, T. Liu, K. Y. Yip, R. Robilotto, A. Rechtsteiner, K. Ikegami, P. Alves, A. Chateigner, M. Perry, M. Morris, R. K. Auerbach, X. Feng, J. Leng, A. Vielle, W. Niu, K. Rhrissorrakrai, A. Agarwal, R. P. Alexander, G. Barber, C. M. Brdlik, J. Brennan, J. J. Brouillet, A. Carr, M. S. Cheung, H. Clawson, S. Contrino, L. O. Dannenberg, A. F. Dernburg, A. Desai, L. Dick, A. C. Dosé, J. du, T. Egelhofer, S. Ercan, G. Euskirchen, B. Ewing, E. A. Feingold, R. Gassmann, P. J. Good, P. Green, F. Gullier, M. Gutwein, M. S. Guyer, L. Habegger, T. Han, J. G. Henikoff, S. R. Henz, A. Hinrichs, H. Holster, T. Hyman, A. L. Iniguez, J. Janette, M. Jensen, M. Kato, W. J. Kent, E. Kephart, V. Khivansara, E. Khurana, J. K. Kim, P. Kolasinska-Zwierz, E. C. Lai, I. Latorre, A. Leahey, S. Lewis, P. Lloyd, L. Lochovsky, R. F. Lowdon, Y. Lubling, R. Lyne, M. MacCoss, S. D. Mackowiak, M. Mangone, S. McKay, D. Mecenas, G. Merrihew, D. M. Miller III, A. Muroyama, J. I. Murray, S. L. Ooi, H. Pham, T. Phippen, E. A. Preston, N. Rajewsky, G. Rätsch, H. Rosenbaum, J. Rozowsky, K. Rutherford, P. Ruzanov, M. Sarov, R. Sasidharan, A. Sboner, P. Scheid, E. Segal, H. Shin, C. Shou, F. J. Slack, C. Slightam, R. Smith, W. C. Spencer, E. O. Stinson, S. Taing, T. Takasaki, D. Vafeados, K. Voronina, G. Wang, N. L. Washington, C. M. Whittle, B. Wu, K. K. Yan, G. Zeller, Z. Zha, M. Zhong, X. Zhou; modENCODE Consortium, J. Ahringer, S. Strome, K. C. Gunsalus, G. Micklem, X. S. Liu, V. Reinke, S. K. Kim, L. D. W. Hillier, S. Henikoff, F. Piano, M. Snyder, L. Stein, J. D. Lieb, R. H. Waterston, Integrative analysis of the Caenorhabditis elegans genome by the modENCODE Project. Science 330, 1775– 1787 (2010). 60. K. Ikegami, T. A. Egelhofer, S. Strome, J. D. Lieb, Caenorhabditis elegans chromosome arms are anchored to the nuclear membrane via discontinuous association with LEM-2. Genome Biol. 11, R120 (2010). 61. T. R. Mandigo, B. D. Turcich, A. J. Anderson, M. R. Hussey, E. S. Folker, Drosophila emerins control LINC complex localization and transcription to regulate myonuclear position. J. Cell Sci. 132, jcs235580 (2019). 62. Y. Hiraoka, H. Maekawa, H. Asakawa, Y. Chikashige, T. Kojidani, H. Osakada, A. Matsuda, T. Haraguchi, Inner nuclear membrane protein Ima1 is dispensable for intranuclear positioning of centromeres. Genes Cells 16, 1000–1011 (2011). 63. S. Gudise, R. A. Figueroa, R. Lindberg, V. Larsson, E. Hallberg, Samp1 is functionally associated with the LINC complex and A-type lamina networks. J. Cell Sci. 124, 2077–2085 (2011). 64. J. Borrego-Pinto, T. Jegou, D. S. Osorio, F. Auradé, M. Gorjánácz, B. Koch, I. W. Mattaj, E. R. Gomes, Samp1 is a component of TAN lines and is required for nuclear movement. J. Cell Sci. 125, 1099–1105 (2012). 65. C. R. Bone, E. C. Tapley, M. Gorjánácz, D. A. Starr, The Caenorhabditis elegans SUN protein UNC-84 interacts with lamin to transfer forces from the cytoplasm to the nucleoskeleton during nuclear migration. Mol. Biol. Cell 25, 2853–2865 (2014). 66. A. Morales-Martínez, A. Dobrzynska, P. Askjaer, Inner nuclear membrane protein LEM-2 is required for correct nuclear separation and morphology in C. elegans. J. Cell Sci. 128, 1090–1096 (2015). 67. J. Kim, S. Han, A. Lei, M. Miyano, J. Bloom, V. Srivastava, M. R. Stampfer, Z. J. Gartner, M. A. LaBarge, L. L. Sohn, Characterizing cellular mechanical phenotypes with mechano-node-pore sensing. Microsyst. Nanoeng. 4, 17091 (2018). 68. B. Li, A. Maslan, S. E. Kitayama, C. Pierce, A. M. Streets, L. L. Sohn, Mechanical phenotyping reveals unique biomechanical responses in retinoic acid-resistant acute promyelocytic leukemia. iScience 25, 103772 (2022). 69. K. G. Geles, S. A. Adam, Germline and developmental roles of the nuclear transport factor importin α3 in C. elegans. Development 128, 1817–1830 (2001). 70. N. T. Chartier, A. Mukherjee, J. Pfanzelter, S. Fürthauer, B. T. Larson, A. W. Fritsch, R. Amini, M. Kreysing, F. Jülicher, S. W. Grill, A hydraulic instability drives the cell death decision in the nematode germline. Nat. Phys. 17, 920–925 (2021). 71. A. F. Dernburg, Pushing the (nuclear) envelope into meiosis. Genome Biol. 14, 110 (2013). 72. E. C. Schirmer, R. Foisner, Proteins that associate with lamins: Many faces, many functions. Exp. Cell Res. 313, 2167–2179 (2007). 73. A. Brachner, S. Reipert, R. Foisner, J. Gotzmann, LEM2 is a novel MAN1-related inner nuclear membrane protein associated with A-type lamins. J. Cell Sci. 118, 5797–5810 (2005). 74. L. Clements, S. Manilal, D. R. Love, G. E. Morris, Direct interaction between emerin and lamin A. Biochem. Biophys. Res. Commun. 267, 709–714 (2000). 75. M. Sakaki, H. Koike, N. Takahashi, N. Sasagawa, S. Tomioka, K. Arahata, S. Ishiura, Interaction between emerin and nuclear lamins. J. Biochem. 129, 321–327 (2001). 76. R. Zhu, C. Liu, G. G. Gundersen, Nuclear positioning in migrating fibroblasts. Semin. Cell Dev. Biol. 82, 41–50 (2018). 77. E. S. Folker, C. Östlund, G. W. G. Luxton, H. J. Worman, G. G. Gundersen, Lamin A variants that cause striated muscle disease are defective in anchoring transmembrane actin-associated nuclear lines for nuclear movement. Proc. Natl. Acad. Sci. U.S.A. 108, 131–136 (2011). 78. L. Penfield, B. Wysolmerski, M. Mauro, R. Farhadifar, M. A. Martinez, R. Biggs, H. Y. Wu, C. Broberg, D. Needleman, S. Bahmanyar, Dynein pulling forces counteract lamin-mediated nuclear stability during nuclear envelope repair. Mol. Biol. Cell 29, 852–868 (2018). 79. W. Chang, Y. Wang, G. W. G. Luxton, C. Östlund, H. J. Worman, G. G. Gundersen, Imbalanced nucleocytoskeletal connections create common polarity defects in progeria and physiological aging. Proc. Natl. Acad. Sci. U.S.A. 116, 3578–3583 (2019). 80. Y. K. Wu, H. Umeshima, J. Kurisu, M. Kengaku, Nesprins and opposing microtubule motors generate a point force that drives directional nuclear motion in migrating neurons. Development 145, dev158782 (2018). 81. S. Brenner, The genetics of Caenorhabditis elegans. Genetics 77, 71–94 (1974). 82. L. Zhang, S. Köhler, R. Rillo-Bohn, A. F. Dernburg, A compartmentalized signaling network mediates crossover control in meiosis. eLife 7, e30789 (2018). 83. J. A. Arribere, R. T. Bell, B. X. H. Fu, K. L. Artiles, P. S. Hartman, A. Z. Fire, Efficient marker-free recovery of custom genetic modifications with CRISPR/Cas9 in Caenorhabditis elegans. Genetics 198, 837–846 (2014). 84. A. Paix, A. Folkmann, D. Rasoloson, G. Seydoux, High efficiency, homology-directed genome editing in Caenorhabditis elegans using CRISPR-Cas9 ribonucleoprotein complexes. Genetics 201, 47–54 (2015). 85. L. Timmons, D. L. Court, A. Fire, Ingestion of bacterially expressed dsRNAs can produce specific and potent genetic interference in Caenorhabditis elegans. Gene 263, 103–112 (2001). 86. T. Davies, H. X. Kim, N. Romano Spica, B. J. Lesea-Pringle, J. Dumont, M. Shirasu-Hiza, J. C. Canman, Cell-intrinsic and -extrinsic mechanisms promote cell-type-specific cytokinetic diversity. eLife 7, e36204 (2018). 87. R. S. Kamath, J. Ahringer, Genome-wide RNAi screening in Caenorhabditis elegans. Methods 30, 313–321 (2003). 88. R. S. Kamath, A. G. Fraser, Y. Dong, G. Poulin, R. Durbin, M. Gotta, A. Kanapin, N. le Bot, S. Moreno, M. Sohrmann, D. P. Welchman, P. Zipperlen, J. Ahringer, Systematic functional analysis of the Caenorhabditis elegans genome using RNAi. Nature 421, 231–237 (2003). 89. H. Zhang, A. R. Skop, J. G. White, Src and Wnt signaling regulate dynactin accumulation to the P2- EMS cell border in C. elegans embryos. J. Cell Sci. 121, 155–161 (2008). 90. C. M. Phillips, K. L. McDonald, A. F. Dernburg, in Meiosis: Volume 2, Cytological Methods, S. Keeney, Ed. (Humana Press, 2009), pp. 171–195. 91. A. J. MacQueen, C. M. Phillips, N. Bhalla, P. Weiser, A. M. Villeneuve, A. F. Dernburg, Chromosome sites play dual roles to establish homologous synapsis during meiosis in C. elegans. Cell 123, 1037–1050 (2005). 92. C. M. Phillips, C. Wong, N. Bhalla, P. M. Carlton, P. Weiser, P. M. Meneely, A. F. Dernburg, HIM- 8 binds to the X chromosome pairing center and mediates chromosome-specific meiotic synapsis. Cell 123, 1051–1063 (2005). 93. A. Gartner, A. J. MacQueen, A. M. Villeneuve, in Checkpoint Controls and Cancer: Volume 1: Reviews and Model Systems, A. H. Schönthal, Ed. (Humana Press, 2004), pp. 257–274. 94. M. Han, G. Wei, C. E. McManus, L. W. Hillier, V. Reinke, Isolated C. elegans germ nuclei exhibit distinct genomic profiles of histone modification and gene expression. BMC Genomics 20, 500 (2019). 95. A. Lai, R. Rex, K. L. Cotner, A. Dong, M. Lustig, L. L. Sohn, Mechano-node-pore sensing: A rapid, label-free platform for multi-parameter single-cell viscoelastic measurements. J. Vis. Exp., e64665 (2022). 96. R. W. DeBlois, C. P. Bean, Counting and sizing of submicron particles by the resistive pulse technique. Rev. Sci. Instrum. 41, 909–916 (1970). 97. O. A. Saleh, L. L. Sohn, Correcting off-axis effects in an on-chip resistive-pulse analyzer. Rev. Sci. Instrum. 73, 4396–4398 (2002).
Could not heal snippet
10.1126_sciadv.ade1817.pdf
null
null
Supplementary Materials for Starvation-induced changes in somatic insulin/IGF-1R signaling drive metabolic programming across generations Merly C. Vogt et al. Corresponding author: Merly C. Vogt, merly.vogt@helmholtz-munich.de Sci. Adv. 9, eade1817 (2023) DOI: 10.1126/sciadv.ade1817 The PDF file includes: Figs. S1 to S7 Legends for data S1 to S12 Other Supplementary Material for this manuscript includes the following: Data S1 to S12 Fig. S1. Inter- and transgenerational increase in exploration can be observed after one round of ancestral starvation, related to Fig. 1 (A) Schematic overview of exploratory behavior assay. (B) Exploratory behavior was assessed over a 16h period starting with animals at the young L4 stage. Data was obtained from 2-4 independent biological replicates (indicated by different colors of each data point) with n=15-20 animals/replicate. Data is presented as mean +/-SEM. Significance was determined by Mann Whitney U test. Fig. S2, Early life starvation results in differential expression of genes involved in lipid metabolism transgenerationally, but does not increase lifespan, related to Fig. 2 (A) Transcriptome analysis was performed across 3 independent biological replicates with ~10.000 L1 animals/replicate. Samples were collected under fed and starved conditions. Adjusted p-values were determined by Wald test with subsequent Benjamini and Hochberg correction using the DESeq2 package (97). (B) In our hands, early life starvation does not increase lifespan transgenerationally in either P0L1starved+3fed (n control= 139; n 1xAS=130; data combined from two independent biological replicates) or F4L1starved+3fed (n Control=381; n 5xAS=399; data combined from four independent biological replicates) as previously reported (42, 43). Fig. S3, L1 starvation affects fertility and transcriptome of individuals within a population to different extents, related to Fig. 4 (A) Transcriptome analysis was performed across 3 independent biological replicates with ~100 manually picked animals/group at day 1 of adulthood under fed conditions. Animals were separated into fertile and sterile F4L1starved. Differential gene expression between control, fertile and sterile F4L1starved animals and overlap between groups is displayed as Venn-Diagram (not drawn to scale). (B) Downstream DAF-16/FoxO target genes that showed an overall relative upregulation in fertile F4L1starved animals compared to control animals (Fig. 3B), showed an even further upregulation in sterile F4L1starved compared to fertile F4L1starved animals. Values in heatmap were z-scored normalized and plotted using heatmap3 in RStudio. Each row represents a single gene and each column represents a single RNA-seq replicate. Blue=relative downregulation, red=relative upregulation in sterile F4L1starved animals. Adjusted p-values were determined by Wald test with subsequent Benjamini and Hochberg correction using the DESeq2 package (97). Significant enrichment was determined by hypergeometric distribution. Fig. S4, DAF-16/FoxO does not act directly in the germline to mediate metabolic programming across generations, related to Fig. 5 (A) Acute starvation in control animals results in overall upregulation of class 1 IIS downstream target genes. Values of differentially expressed genes (adj.p<0.1) between fed and starved L1 control animals were z-score normalized and plotted using heatmap3 using RStudio. Each row represents a single gene and each column represents a single RNA-Seq replicate. Blue=relative downregulation, red=relative upregulation in starved L1 animals. Imaging of fed and acutely starved L1 animals with an endogenously-tagged mNeonGreen DAF-16/FoxO allele (ot853) confirms nuclear accumulation, and thus activation, of DAF-16/FoxO::mNeonGreen across the animal, including germline precursor cells, epidermis, neurons and intestines upon starvation. Scale bar =15µm. (B) DAF-16/FoxO::mNG also localizes (peri)nuclearly in the mitotic germline upon acute starvation in control fed adult animals. However, the severity and subcellular distribution of DAF-16/FoxO::mNG is different in acutely starved adult compared to L1 starved, currently fed adult animals. Scale bar = 15 µm. (C) Nearly 100% of adult fed sterile F4L1starved animals display nuclear accumulation of DAF-16/FoxO::mNG in mitotic germline. (D) Relative nuclear accumulation of DAF-16/FoxO::mNG in mitotic germline is still significantly increased in immediate progeny, but lost in subsequent descendants of F4L1starved. Data was collected from n=10 animals/group and 3 biological replicates and is presented as mean+/-SEM. Significance was determined by Chi-square test. Color indicates sub-cellular localization of DAF-16/FoxO in mitotic germline: White= cytoplasm; Green= Nucleus. Control and ancestrally starved animals were imaged in presence of acute auxin unlike animals displayed in Fig. 4A and S4B, which could explain slight increase in nuclear accumulation in mitotic germline even in control animals. (E) Impaired fertility of F4L1starved is not mediated across generations to F4L1starved +1fed animals. (F) One round of L1 starvation is sufficient to impair fertility later in life. Data is displayed as mean +/- SEM and was collected from n=16/group across two biological replicates. Significance was determined by Mann Whitney U test. (G) One round of L1 starvation is sufficient to induce DAF-16/FoxO nuclear accumulation in mitotic germline in adult animals. Data was collected from n=10 animals/group and 3 biological replicates and is presented as mean+/-SEM. Significance was determined by Chi-square test. (H-K) Results for exploratory behavior and oxidative stress resistance in F4L1starved +1fed upon auxin-induced DAF-16/FoxO depletion specifically in the germline and only during indicated generation and time-points (same outline as displayed in Fig. 5C-5F for F4L1starved +3fed). Data for exploratory behavior was collected from n=20/group across 3 biological replicates per condition and displayed as mean +/-SEM. Statistical significance was determined using One Way ANOVA with posthoc Tukey. For oxidative stress resistance, the combined data of 3 independent biological replicates per group and condition with n=70-103/replicate is displayed. Statistical significance was determined by Log-rank (Mantel-Cox test). Fig. S5, Starvation-induced increase of DAF-16/FoxO activity in somatic tissues causes metabolic programming across generation, related to Fig. 6 (A-D) AID/TIR1 system efficiently and specifically depletes DAF-16/FoxO::mNG::AID from tissues of interest. DAF-16/FoxO::mNG::AID is efficiently depleted from (A) germline precursor cells in animals expressing TIR1 under the control of a germline- specific promoter (ot853;ieSi38), (B) pan-somatically in animals expressing TIR1 under control of a pan-somatic promoter (ot853; ieSi57), (C) the intestine in animals expressing TIR1 under control of intestine-specific promoter (ot853, ieSi61), and (D) pan-neuronally in animals expressing TIR1 under the control of a panneuronal-specific promoter (ot853; reSi7) in the presence of auxin only. Scale bar =15 µm. Animals are imaged at L2 stage under fed conditions. (E) Auxin exposure does not efficiently deplete DAF- 16/FoxO::mNG::AID in embryos in our paradigm. Displayed are embryos expressing DAF-16/FoxO::mNG::AID and pan-somatic TIR1 (ot853; ieSi57) at various stages in the absence or presence of auxin. Scale bar =15 µm. F-H) Results for oxidative stress resistance upon pan-somatic depletion of DAF-16/FoxO during distinct time points are displayed as combined data from three independent biological replicates with n=80- 114/replicate are displayed. Statistical significance was determined by Log-rank (Mantel- Cox test). (F) Acute pan-somatic DAF-16/FoxO depletion decreased oxidative stress in control and F4L1starved +1fed animals. (G) Pan-somatic DAF-16/FoxO depletion during L1 starvation in the parental generation reverted decreased oxidative stress resistance in F4L1starved +1fed (ot853; ieSi57 control vs ot853; ieSi57 F4L1starved +1fed animals not significant). (H) Pan-somatic DAF-16/FoxO depletion during recovery phase in the parental generation reverted decreased oxidative stress resistance in F4L1starved +1fed (ot853; ieSi57 control vs ot853; ieSi57 F4L1starved +1fed animals not significant). (I) Acute pan-somatic depletion of DAF-16/FoxO was insufficient to increase exploration in either control or ancestrally starved animals. (J) Pan-somatic DAF-16/FoxO depletion during L1 starvation or (K) recovery phase in parental generation does not revert increased exploratory behavior inter- (F4L1starved +1fed) or transgenerationally (F4L1starved +3fed). Data was obtained from n=20/group and 3 independent biological replicates and is displayed as mean+/-SEM. Statistical significance was determined using One Way ANOVA with posthoc Tukey. Fig. S6, Intestinal TIR1- and DAF-16/FoxO::mNG::AID expressing strain displays slight increase in oxidative stress resistance even in absence of auxin, related to Fig. 7 (A) Oxidative stress resistance is significantly increased in animals expressing DAF- 16/FoxO::mNG::AID and intestine-specific TIR1 (ot853;ieSi61) compared to animals expressing DAF-16/FoxO::mNG::AID (ot853) animals even in the absence of auxin. Data is displayed as combined data from three independent biological replicates with n=80-100/replicate. Statistical significance was determined by Log-rank (Mantel-Cox test). Fig. S7, A large subset of genes encoding for critical factors involved in the biogenesis and function of small RNAs are “class 2” downstream targets of IIS, related to Fig. 8 (A) Transcriptome analysis between control and fertile F4L1starved revealed that genes encoding for critical factors involved in biogenesis and function of small RNAs displayed relative downregulation in fertile F4L1starved compared to control animals at day 1 of adulthood under fed conditions. Values of genes between fed and starved L1 control animals were z-score normalized and plotted using heatmap3 using RStudio. Each row represents a single gene and each column represents a single RNA-Seq replicate. Blue=relative downregulation, red=relative upregulation. Adjusted p-values were determined by Wald test with subsequent Benjamini and Hochberg correction using the DESeq2 package (97). Data S1. (Separate file) Data S1: Transcriptome Analysis Control vs F4+3 – All Genes and Conditions Related to Fig. 2 Data S2: Significantly Expressed Genes Control vs F4+3 under fed conditions Related to Fig. 2A, 2B Data S3: Significantly Expressed Genes Control vs F4+3 under starved conditions Related to Fig. 2A, 2B Data S4: Enrichment Differentially Expressed Genes Control vs F4+3 Fed for class1 IIS genes Related to Fig. 2E Data S5: Enrichment Differentially Expressed Genes Control vs F4+3 Fed for class2 IIS genes Related to Fig. 2E Data S6: Transcriptome Analysis Parental Control vs F4 – All Genes and Groups Related to Fig. 4 Data S7: Differentially expressed Genes Parental Control vs F4 fertile Related to Fig. 4C Data S8: Enrichment Analysis Parental Control vs F4 fertile– DAF16/FoxO class1 Related to Fig. 4D Data S9: Enrichment Analysis Parental Control vs F4 fertile– TGFb Related to Fig. 4D Data S10: Enrichment Analysis Parental Control vs F4 fertile– 5HT_tph1mut Related to Fig. 4D Data S11: Enrichment Analysis Parental Control vs F4 fertile– AMPK_aak2mutdown Related to Fig. 4D Data S12: Enrichment Analysis F4 fertile vs F4 sterile – DAF16/FoxO class1 Related to Fig. S3
Could not heal snippet
10.1073_pnas.2301985120.pdf
Data, Materials, and Software Availability. Cryo-EM density maps of the KCNQ1 channel with the voltage sensor in the up, intermediate, and down conformation have been deposited in the electron microscopy data bank under accession codes EMD-40508 (70), EMD-40509 (71), and EMD-40510 (72), respectively. Atomic coordinates of the KCNQ1 channel with the voltage sen- sor in the up, intermediate, and down conformation have been deposited in the protein data bank under accession codes 8SIK (73), 8SIM (74), and 8SIN (75), respectively.
null
RESEARCH ARTICLE | BIOCHEMISTRY OPEN ACCESS The membrane electric field regulates the PIP2-binding site to gate the KCNQ1 channel Venkata Shiva Mandalaa,b and Roderick MacKinnona,b,1 Edited by David Clapham, HHMI, Ashburn, VA; received February 3, 2023; accepted April 13, 2023 Voltage-dependent ion channels underlie the propagation of action potentials and other forms of electrical activity in cells. In these proteins, voltage sensor domains (VSDs) regulate opening and closing of the pore through the displacement of their positive-charged S4 helix in response to the membrane voltage. The movement of S4 at hyperpolarizing membrane voltages in some channels is thought to directly clamp the pore shut through the S4–S5 linker helix. The KCNQ1 channel (also known as Kv7.1), which is important for heart rhythm, is regulated not only by membrane voltage but also by the signaling lipid phosphatidylinositol 4,5-bisphosphate (PIP2). KCNQ1 requires PIP2 to open and to couple the movement of S4 in the VSD to the pore. To understand the mechanism of this voltage regulation, we use cryogenic electron microscopy to visualize the movement of S4 in the human KCNQ1 channel in lipid membrane vesicles with a voltage difference across the membrane, i.e., an applied electric field in the membrane. Hyperpolarizing voltages displace S4 in such a manner as to sterically occlude the PIP2-binding site. Thus, in KCNQ1, the voltage sensor acts primarily as a regulator of PIP2 binding. The voltage sensors’ influence on the channel’s gate is indirect through the reaction sequence: voltage sensor movement → alter PIP2 ligand affinity → alter pore opening. KCNQ1 channel | Kv7.1 channel | voltage sensor | cryo-EM | membrane potential Voltage sensor domains (VSDs) are integral membrane proteins that undergo conforma- tional changes in response to voltage differences across the cell membrane. These domains regulate pore opening and closing in voltage-dependent ion channels (1) and enzymatic activity in voltage-dependent phosphatases (2). Voltage sensors have a conserved structure consisting of four transmembrane (TM) helices (S1 to S4) that form a helical bundle (3–6). The fourth helix, S4, contains a repeated sequence of positive-charged amino acids (typically arginines), every third residue that confers sensitivity to voltage. Inside the lipid bilayer, a gating charge transfer center, composed of aspartate, glutamate, and phenylala- nine residues, stabilizes the arginines one at a time as they traverse the hydrophobic core of the membrane (7, 8). The movement of S4 in response to the TM voltage difference is ultimately responsible for the regulation of protein activity. This mechanism underlies the action potential in neurons (1, 9) and the initiation of muscle contraction (4, 10), among other cellular processes. While the structure of a VSD is highly conserved across all voltage-dependent ion channels, there are two configurations for VSD attachment to the pore of the channel (formed by the S5 and S6 helices). In the so-called domain-swapped channels (Fig. 1A), which include voltage-dependent K+ (Kv) channels 1 to 9, Na+ (Nav) channels, Ca2+ (Cav) channels, and most transient receptor potential channels, the VSD of one subunit interacts with the pore domain of an adjacent subunit, connected through a long interfacial helix— the S4–S5 linker (7, 11–16). Meanwhile, in nondomain-swapped channels (Fig. 1A) such as Kv10-12, Slo1, and hyperpolarization-activated cyclic nucleotide-gated (HCN) chan- nels, the VSD contacts the pore domain of the same subunit through a short S4–S5 loop (17–19). This naturally raises the question: how do the conserved VSDs mediate voltage-dependent gating in these two sets of channels with different structures? We have shown recently that in a nondomain-swapped channel Eag (Kv10.1) (20), the S4 helix on the cytoplasmic side forms an interfacial helix in the hyperpolarized (i.e., negative voltage inside) conformation, which functions as a constrictive cuff around the pore, prevent- ing it from opening. Domain-swapped channels already have an interfacial S4–S5 linker helix that contacts the S6 helix in the depolarized (i.e., no applied or positive inside voltage) con- formation (Fig. 1A) (7, 16), suggesting a gating mechanism that is distinct from that in nondomain-swapped channels. In domain-swapped channels, it has been proposed that the displacement of S4 in response to a hyperpolarizing potential moves the S4–S5 linker helix into a position that clamps the pore shut by pushing down on the S6 helical bundle (7, 16). Structures of domain-swapped channels in detergent micelles at zero mV with chemical Significance Voltage-gated ion channels underlie electrical signaling in cells. The structures and functions of voltage-dependent K+, Na+, and Ca2+ and transient receptor potential ion channels have been studied extensively since their discovery. Despite these efforts, it is still not well understood how the voltage sensors in these different ion channels change their conformation in response to membrane voltage changes, and how these movements regulate the opening or closing of the channel’s gate. This study presents structures of the human KCNQ1 (Kv7.1) voltage– dependent and phosphatidylinositol 4,5-bisphosphate (PIP2)- dependent K+ channel in electrically polarized lipid vesicles using cryogenic electron microscopy, showing how the voltage sensors influence gating indirectly by regulating the ability of PIP2 to bind to the channel. Author affiliations: aLaboratory of Molecular Neuro- biology and Biophysics, The Rockefeller University, New York, NY 10065; and bHHMI, The Rockefeller University, New York, NY 10065 Author contributions: V.S.M. and R.M. designed research; V.S.M. performed research; V.S.M. and R.M. analyzed data; and V.S.M. and R.M. wrote the paper. The authors declare no competing interest. This article is a PNAS Direct Submission. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1To whom correspondence may be addressed. Email: mackinn@rockefeller.edu. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2301985120/-/DCSupplemental. Published May 16, 2023. PNAS  2023  Vol. 120  No. 21  e2301985120 https://doi.org/10.1073/pnas.2301985120   1 of 12 cross-links, toxins, mutations, and metal affinity bridges thought to mimic the hyperpolarized condition are supportive of this mecha- nism (21–27). Here, we present a cryo-EM analysis of the domain-swapped human KCNQ1 (Kv7.1) channel in lipid mem- brane vesicles with a hyperpolarizing voltage generated across the membrane, illustrating—at least in some domain-swapped chan- nels—a different gating mechanism than previously thought. Results The Rationale for Polarizing KCNQ1. The KCNQ1 Kv channel, also known as Kv7.1, is the pore-forming subunit of the slow delayed rectifier potassium channel (IKS) (28, 29) that plays an important role in the repolarization phase of cardiac action potentials (30, 31). Mutations in the kcnq1 gene are associated with several congenital cardiac diseases, including long and short QT syndromes as well as familial atrial fibrillation (32). Importantly, KCNQ1 and other Kv7 members are regulated both by membrane voltage and the signaling lipid phosphatidylinositol 4,5-bisphosphate (PIP2) (33–36). The voltage sensors close the channel at hyperpolarizing membrane voltages, while PIP2 is required for the channel to open. When PIP2 is depleted in the membrane, such as when phospholipase C is activated through stimulation of Gq-coupled receptors (34, 37), the voltage sensors undergo voltage-dependent conformational changes, but the pore does not open at depolarizing voltages (38, 39). PIP2 is thus thought to be required for the coupling of voltage sensor movements to pore opening. In other words, KCNQ1 is thought to act as a ligand-regulated voltage-dependent channel, where the binding of PIP2 allows the channel to be gated by the membrane potential. In the absence of PIP2 at zero mV, we would expect a closed pore and depolarized voltage sensors, which is exactly the KCNQ1 struc- ture observed in detergent micelles (12, 40). If we now apply a hyperpolarizing voltage across the membrane, the pore should remain closed, but the voltage sensors should adopt the polarized conformation. Because in this circumstance the voltage sensors do not have to perform mechanical work to close the pore, it should be easier to move the voltage sensors when the pore is already closed (due to the absence of PIP2). We note that we exclude the KCNE beta subunits (41) in this study because they are known to modify the voltage sensitivity of KCNQ1 and thus could trap the voltage sensor in a specific conformation (for instance, KCNE3 appears to stabilize the depolarized conformation) (40, 42). KCNQ1 Reconstitution and Polarization. The human KCNQ1 channel was purified as a complex with the structurally obligate subunit calmodulin (40, 43) in the presence of Ca2+ and reconstituted into liposomes composed of 90: 5: 5 1-palmitoyl-2- Fig. 1. Structures of voltage-dependent ion channels and the preparation of unpolarized and polarized KCNQ1 (Kv7.1) proteoliposomes. (A) The two domain arrangements in voltage-dependent ion channels. Channels are shown with α-helix cylinders and one of the four subunits colored blue and the S4–S5 linker colored red. In domain-swapped channels (Left), the VSD of one subunit interacts with the pore domain of an adjacent subunit and is connected to the pore domain through a long interfacial helix––the S4–S5 linker (red). The structure of Kv1.2 paddle chimera (PDB ID: 2R9R) (7) is shown as an example. In nondomain-swapped channels (Right), the VSD interacts with the pore domain of the same subunit through a short S4–S5 loop. The Eag channel (PDB ID: 8EOW) (20) is shown here as an example. (B) Schematic of the protocol used to obtain polarized vesicles for cryo-EM analysis. Kv7.1 is reconstituted into liposomes with symmetrical KCl, and valinomycin (val.) is added to mediate K+-flux. The external KCl is exchanged for NaCl using a buffer-exchange column. Potassium efflux through valinomycin generates a potential difference across the membrane such that the inside of the vesicle is negative with respect to the outside. Unpolarized and polarized vesicles containing Kv7.1 were frozen on a holey carbon grid for structure determination. (C) Two-dimensional class-averages of membrane-embedded Kv7.1 in unpolarized (Left) and polarized (Right) vesicles from cryo-EM. (D) Liposome-based flux assay to test polarization of vesicles. Recordings (n = 5, mean ± SD) were made using empty vesicles (black) or vesicles with Kv7.1 (blue). Addition of the H+-ionophore CCCP allows entry of protons, which is detected by quenching of the fluorescent reporter ACMA. Protons enter when the K+ ionophore valinomycin is added. 2 of 12   https://doi.org/10.1073/pnas.2301985120 pnas.org oleoyl-sn-glycero-3-phosphocholine (POPC) to 1-palmitoyl-2- oleoyl- sn-glycero-3-phosphoglycerol (POPG) to cholesterol [wt: wt: wt] with 300 mM KCl (SI Appendix, Fig. S1 A and B). Following our work on Eag (20), valinomycin was added to the vesicles and the extravesicular solution was exchanged to 300 mM NaCl using a buffer-exchange column (Fig.  1B). The valinomycin-mediated K+-efflux generates a membrane voltage with an upper limit of about −145 mV, such that the inside is negative with respect to the outside. These polarized vesicles were immediately applied to a holey carbon grid and frozen for cryogenic electron microscopy (cryo-EM) analysis (Fig. 1C and SI Appendix, Fig. S1C). Grids for an unpolarized control lacking the buffer exchange step (i.e., with symmetric 300 mM KCl) were also prepared. The permeability of these KCNQ1-containing liposomes to small ions was tested using a liposome flux assay (Fig. 1D) (44). The vesicles prepared in 300 mM KCl (without valinomycin) were diluted into a buffer with a fluorescent dye, 9-amino-6-chloro-2- methoxyacridine (ACMA), and isotonic NaCl to generate a K+ gradient. The proton ionophore carbonyl cyanide m-chlorophenylhydrazone (CCCP) was added to allow H+ influx, which leads to quenched ACMA fluorescence. Without valinomycin, no flux was detected, consistent with the channel being tightly closed under these con- ditions. Subsequent addition of the K+-selective ionophore valino- mycin gave rise to rapid quenching of ACMA in both KCNQ1 proteoliposomes and in control liposomes without protein (Fig. 1D), indicating that the valinomycin-generated membrane potential is stable for at least a few minutes. Identification of Three Structural Classes in the Polarized Dataset. We collected large cryo-EM datasets on polarized and unpolarized vesicles using the same microscope, and the structures of KCNQ1 in both were determined using single- particle analysis (SI  Appendix, Figs.  S2–S4 and Table  S1). As we found for the Kv channel Eag, channels were reconstituted exclusively in an inside-in orientation and thus, when polarized, experience hyperpolarizing (i.e., negative inside) potentials under the applied electric field (Fig. 1C). After two rounds of three- dimensional (3D) classification to select for the best subset of particles in each dataset, we carried out 3D classification without alignment with a mask on the TM domain while imposing C4 symmetry (SI Appendix, Fig. S2). The unpolarized dataset showed little heterogeneity: 88% of the particles were in a homogeneously “up” (detailed below) conformation that closely resembled the detergent structure of KCNQ1, while the remaining particles were in an indeterminate state. Meanwhile, the polarized dataset was noticeably more heterogeneous, with only 34% of particles in a homogeneously up conformation, 19% consistent with an “intermediate” conformation, 10% consistent with a “down” conformation, and the remaining indeterminate. Classification without symmetry on a symmetry-expanded particle set showed that these 34% of particles had all four voltage sensors in an up conformation—suggesting that these channels are in vesicles that have lost the ion gradient, and likely do not reflect the distribution of voltage sensor states under the applied potential. Similar classification on the remaining 66% of particles showed classes with voltage sensors in different states, indicating that the higher proportion of indeterminate particles in the polarized dataset is due to a mixture of conformations. In summary, the observation of distinct structural classes for the voltage sensor in the polarized but not the unpolarized dataset indicates that these conformational changes are likely caused by the application of an electric field (the alternative being due to Na+ in the external solution). From the unpolarized dataset, the best up structure (C4-symmetric; SI Appendix, Fig. S3) had an overall resolution of 2.9 Å (Fig. 2A and SI Appendix, Fig. S4). We solved three structures from the polarized dataset (SI Appendix, Figs. S3 and S4): C4-symmetric up and inter- mediate structures with overall resolutions of 3.4 Å and 6.2 Å, respec- tively, and a C1-symmetric down structure from a symmetry-expanded particle set with an overall resolution of 6.8 Å. The up structures from the unpolarized and polarized datasets are nearly identical (SI Appendix, Fig. S5 A and B), so we focus on the better-resolved former structure. We note that one interesting difference between the two up structures regards the occupancy of K+ ions in the selectivity filter (SI Appendix, Fig. S5 C and D). In the polarized sample, due to the low extravesic- ular concentration of K+, density is only visible at the first and third positions in the selectivity filter, while density is present at all four positions in the unpolarized sample. Similar differences were observed in our previous study on Eag (20) and are qualitatively consistent with crystal structures of KcsA solved under symmetrical high and low K+ concentrations (45). The Up Conformation of the Voltage Sensor. The up map is best defined in the TM domain, with local resolution estimates of ~2.4 to 2.8 Å for much of S1 through S6 (SI Appendix, Fig. S4D). Density for individual hydrogen-bonded water molecules is visible in the voltage sensor (SI  Appendix, Fig.  S6A). These water molecules do not represent a bulk water-filled crevice, but nevertheless undoubtedly contribute to the stabilization of positive-charged residues (20). Tightly bound phospholipid and sterol molecules (SI Appendix, Fig. S6B) are also visible at both the outer and inner leaflets of the membrane. These features are not discussed further in this paper but are highlighted to demonstrate the feasibility of obtaining high-quality cryo-EM reconstructions in lipid bilayers. A structural model was built by fitting the detergent structure of KCNQ1 (40) and making adjustments where needed (Fig. 2 B–D). The up structure in lipid bilayers is very similar to the depolarized structure in detergent micelles. The S4–S5 linker is an α-helix from I257 to G245 and a short loop (Q244 to D242) connects the S4–S5 linker to S4 (Fig. 2D). The S4 is a 310 helix from V241 to R237 and an α-helix from L236 to T224 (Fig. 2D). The S3–S4 loop is partially flexible—with the four residues (GQVF) in between K218 (top of S3) and A223 (top of S4) not well defined—a point we shall return to later. The six positive-charged residues in S4 are positioned as such (Fig. 2C): R6 (R243) lies below the gating charge transfer center. H5 (H240) occupies the gating charge transfer center consisting of F167 from S2 and the negative-charged E170 and D202 from S2 and S3, respectively. R4 (R237) is directly above the gating charge transfer center and interacts with E160 in S2. Q3 (Q234), R2 (R231), and R1 (R228) lie further toward the extracellular side of the membrane. Q3 lies within the voltage sensor helical bundle, R2 is at the periphery, and R1 is pointed toward the headgroups of the phospholipid bilayer (Figs. 2C and 3 A–C). The Down and Intermediate Conformations of the Voltage Sensor. The intermediate and down maps are less well defined due to heterogeneity, but clear differences in the main chain compared to the up map (modeled in Fig. 3A) were used to build partial models. We compare the down (Fig. 3 E and F) and intermediate (Fig. 3D) maps to an up map that is filtered to a comparable resolution (Fig. 3 B and C). Compared to the up map, the down map shows a dramatic change in the bottom half of S4, near the intracellular surface (Fig. 3 C and F and Movie S1). At the intracellular surface, the loop connecting PNAS  2023  Vol. 120  No. 21  e2301985120 https://doi.org/10.1073/pnas.2301985120   3 of 12 Fig. 2. Structure of KCNQ1 in lipid vesicles with the voltage sensor in the up conformation. (A) Cryo-EM density map of the up structure of the KCNQ1 channel from the unpolarized dataset. Each channel subunit is shown in a different color and calmodulin (CaM) is shown in magenta. Bound lipids or sterols are shown as gray density. (B) Structure of the KCNQ1–CaM complex (cartoon representation) showing domains within one monomer (blue) from the N- to C-terminus: voltage sensor, pore domain, and the cytosolic domain. The other monomers are colored gray for clarity and the bound calmodulin is colored magenta. (C) Stereoview of the KCNQ1 voltage sensor (Cα trace) in the up (depolarized) conformation. The six positive charges in S4 (α carbon marked by blue spheres), three negative charges in S2 and S3 (E160, E170, and D202), and the hydrophobic Phe in S2 (F167, green sticks) are shown in stick-and-ball representation. (D) Stereoview of the main chain in S4 and the S4–S5 linker (stick representation) in the up conformation. The α carbons of the six positive charges in S4 are marked by blue spheres. Regions with different secondary structures are indicated: α-helix (green), 310 helix (cyan), and loop (magenta). S4 to the S4–S5 linker helix becomes lengthened by eight or nine amino acids. The lengthening occurs while the S4–S5 linker helix on the C-terminal side of W248, whose side chain density is apparent even at the lower resolution of the down map, remains unchanged in its position. Given that the S4–S5 linker helix does not move, the lengthened loop must result from amino acids originating in the downward displacement of the S4 helix and the four residues in the S3–S4 loop. The density in the newly formed extended loop suggests that as S4 moves downward, it forms a broken helix ~30° relative to the bilayer normal, and an extended loop (Fig. 3F and Movie S1). On the extracellular side, the top of helical densities for both S3 and S4 appears embedded about one turn below the expected plane of the extracellular membrane surface, while a short loop connecting them reaches to the extracellular surface (Fig. 3 B and E). 4 of 12   https://doi.org/10.1073/pnas.2301985120 pnas.org Fig. 3. Characterization of electric field–induced movements in the KCNQ1 voltage sensor. (A) Stereoview (Cα trace) of the voltage sensor in the up model. For reference, the positions of the alpha carbons of the six positive charges in S4 are marked by blue spheres, that of K218 at the top of S3 is marked by a red sphere, and that of W248 at the end of the S4–S5 linker is marked by a green sphere. F167 in S2, which is part of the gating charge transfer center, is shown in magenta stick representation. (B and E) Stereoviews of the top part of S3 and S4 in the lowpass-filtered up model and map (unpolarized dataset, B) and in the down model and map (polarized dataset, E). (C, D and F) Stereoview of the bottom part of S4 and the S4–S5 linker in the lowpass-filtered up model and map (unpolarized dataset, C), intermediate (inter.) model and map (polarized dataset, D), and in the down model and map (polarized dataset, F). The up map was lowpass filtered to 6.5 Å to facilitate comparison to the down and intermediate maps. In the absence of side chain density, given the large conforma- tional change in the position of S4 required to form the large loop on the intracellular side, we could not build a model of this region with certainty in the polypeptide register. We built two tentative polyalanine models into the continuous main chain density, one invoking a three helical turn displacement of S4 and the other invoking a two helical turn displacement (SI Appendix, Fig. S7). One and four helical turn models are incompatible with the observed density. The three helical turn displacement, which would place Q3-R6 below, R2 in, and R1 above the gating charge transfer, more reliably accounts for density, but additional data will be needed to establish this conclusion. We note at this point that the mechanism presented in the current study (to be dis- cussed) does not rely on modeling side chains or the detailed register of the S4 helix, because the main chain movements we do observe clearly interfere with the PIP2-binding site and thus explain the basic mechanism of this channel’s gating. In contrast to the down structure, the intermediate structure largely preserves the secondary structure of the up conformation but displays a ~4 Å downward displacement of the loop connecting S4 to the S4–S5 linker (Fig. 3 C and D). As in the down structure, the S4–S5 linker helix does not move appreciably. Given that the motion is likely to be a rigid body movement of S4, we included S4 sidechains in the structural model of the intermediate conformation. The PNAS  2023  Vol. 120  No. 21  e2301985120 https://doi.org/10.1073/pnas.2301985120   5 of 12 intermediate structure places R6 and H5 below the gating charge transfer center; R4 in the gating charge transfer center; and Q3, R2, and R1 above the gating charge transfer center. In all three structures, the pore appears tightly closed, as expected in the absence of PIP2 (SI Appendix, Fig. S8C). The pore radius is ~1 Å at S349 in the up structure, which is notably smaller than the radius of a hydrated K+ ion (~4 Å). While the side chain of S349 is not visible in the intermediate and down maps, the position of S6 is the same as in the up structure with a closed pore and different from the PIP2-bound structure with an open pore (SI Appendix, Fig. S8C), thus being consistent with a closed pore. Discussion The Relationship between Voltage Sensor Movements and PIP2 Binding. The three structures presented in this study delineate the movement of the S4 helix in the KCNQ1 channel in response to polarization, while the pore of the channel is closed in all the three cases due to the absence of PIP2 in our preparation. The structure of PIP2-bound KCNQ1 with an open pore and the voltage sensor in the up conformation was already determined (40, 46, 47). By comparing these structures, we can deduce how the voltage sensor movements relate to PIP2 binding. The PIP2-binding site in KCNQ1 comprises positive-charged and polar residues in the S4–S5 linker, S4 helix, the S2–S3 foot, and the S0 helix (Fig. 4A, see also Fig. 2C). This structure of the pocket is maintained when the pore of the channel is open or closed as long as the voltage sensor is in the up conformation (Fig. 4B). In other words, when the voltage sensor is up, PIP2 can bind to this pocket and promote channel opening, as described previously (33–36, 40). We note that all the structures of KCNQ1 solved in the presence of PIP2 show an open pore, but only some also show a large conformational change in the cytoplasmic domain (SI Appendix, Fig. S8 C–E) (40, 46). The relationship between the two is not clear, but it is apparent that PIP2-binding causes the pore to open, which is what we focus on here. Overlays of the intermediate (Fig. 4C) and down (Fig. 4D) volt- age sensor conformations with the PIP2-bound, voltage sensor up conformation show that the PIP2-binding site is reshaped when S4 moves. The position of S4 in the down conformation sterically occludes the PIP2-binding site altogether (Fig. 4D). Thus, while the voltage sensor is in the down conformation, PIP2 cannot bind to the channel and open the pore. In the intermediate conformation, the residues that bind PIP2 are displaced relative to one another due to the movement of S4 (Fig. 4C). This intermediate conformational change would likely alter the affinity of the PIP2-binding site, but it might not definitively preclude the binding of PIP2. Voltage-Dependent Regulation of PIP2 Binding in KCNQ1. A mechanism for voltage-dependent regulation of KCNQ1 channel activity thus follows (Fig. 5E). We have made a movie to visualize the sequence of events (Movie S2). At hyperpolarized membrane voltages (i.e., at the resting potential of a cell, corresponding to our polarized vesicles), the voltage sensor is in the down conformation, which prevents PIP2 from binding because the site is occluded. Depolarization drives the S4 helix up, which is coupled to the formation of the PIP2 binding site. PIP2 can then bind, which causes the pore to open through an allosteric mechanism (40, 46, 47). In other words, KCNQ1 activity is modulated by a ligand (PIP2), the binding of which is regulated by the voltage sensor. This is different in detail from a mechanism in which the binding of PIP2 permits voltage sensor conformational changes to regulate the pore through direct mechanical coupling (Fig. 5E). This voltage-dependent regulation of PIP2-binding mechanism is compatible with electrophysiological studies of KCNQ channels, Fig. 4. The relationship between voltage sensor movements and PIP2 binding. (A) Stereoview (gray Cα trace) of the PIP2-bound structure of KCNQ1 (PDB ID: 6V01) (40) with the voltage sensor in the up conformation and an open pore. Positive-charged and polar residues that interact with PIP2 are labeled and shown as gray sticks (α carbon marked by gray spheres) and PIP2 is shown as yellow sticks. (B) Stereoview of the PIP2-free structure of KCNQ1 with the voltage sensor in the up conformation and a closed pore (blue Cα trace) and the PIP2-bound structure shown in panel A (gray Cα trace). (C) Stereoview of the PIP2-free structure of KCNQ1 with the voltage sensor in the intermediate conformation and a closed pore (magenta Cα trace) and the PIP2-bound structure shown in panel A (gray Cα trace). (D) Stereoview of the PIP2-free structure of KCNQ1 with the voltage sensor in the down conformation and a closed pore (green Cα trace) and the PIP2-bound structure shown in panel A (gray Cα trace). In panels (B–D), the α carbon positions of the PIP2-interacting residues are shown as spheres in the same color as the α carbon trace. which show that voltage sensor movements slightly precede pore opening (48, 49), that PIP2 is required for activity (33–36), and that in the absence of PIP2, the voltage sensors move but the pore does not open (38). Moreover, if the voltage sensors were to perform work directly on the pore to open it, and if PIP2 was required for this coupling, one would expect a shift in the voltage activation midpoint (i.e., as measured by the movement of S4) depending on whether PIP2 is bound or not. But, the movement of S4 happens at the same membrane voltage whether PIP2 is present in the mem- brane or not (38), suggesting that the voltage sensors do not perform 6 of 12   https://doi.org/10.1073/pnas.2301985120 pnas.org Fig. 5. The structure of S4 and the S4–S5 linker of Kv7.1 (KCNQ1) and Kv2.1 determined in lipid vesicles. (A) Sequence alignment of S4 and the S4–S5 linker for all domain-swapped Kv channel families. All members of the Kv7 family are included for comparison to Kv7.1. Residues conserved across all families are highlighted in blue. (B) Cryo-EM density map of the human Kv2.1 channel determined in lipid vesicles. Each channel subunit is shown in a different color and associated lipids or sterols are shown as gray density. (C and D) Stereoviews of the connection between S4 and the S4–S5 linker (stick representation) in the depolarized conformations of Kv7.1 (C) and Kv2.1 (D) overlaid with cryo-EM density (blue mesh). (E) Cartoons depicting the new gating model (Top) and the old gating model (Bottom) for KCNQ1. The pore domain is colored orange, the voltage sensor is gray, the S4 helix is blue, the S4–S5 linker is green, and PIP2 is depicted as a magenta hexagon. In the new model, the voltage sensor regulates the binding of PIP2 by occluding the binding site in the down conformation (Left). When membrane depolarization occurs, the voltage sensor moves to the up conformation (Middle), which then allows PIP2 to bind to the channel and open the pore (Right). In the old model that is inconsistent with our data, a PIP2-binding site is present in the down conformation, allowing PIP2 to bind to the channel. work on the pore at depolarized potentials to open it. Finally, voltage clamp fluorometry using a reporter on the S3–S4 linker shows two components: a larger fluorescence change that has a midpoint of ~−60 mV and a smaller change in fluorescence with a midpoint of ~30 mV (50, 51). The pore begins to open during the first (more negative voltage) fluorescence change, which has led to the proposal that there are two open states of the channel. These observations could be related to the intermediate and up voltage sensor confor- mations that we observe. Unique Features of the KCNQ1 Voltage Sensor. KCNQ1 is unique among domain-swapped Kv channels in its requirement of PIP2 to open. To this point, a comparison of the KCNQ1 structure to that of a different domain-swapped Kv channel is informative. A primary sequence alignment from S4 through the S4–S5 linker for the Shaker channel, one member from each of the domain-swapped Kv channel families (Kv1-9), and other members of the Kv7 (KCNQ) family, is given in Fig. 5A. Stereoviews of the S4 to S4–S5 helix linker connection along with cryo-EM density PNAS  2023  Vol. 120  No. 21  e2301985120 https://doi.org/10.1073/pnas.2301985120   7 of 12 are shown for the depolarized structures of Kv7.1 (Fig. 5C, see also Fig. 2D) and human Kv2.1 (52), both in lipid vesicles (Fig. 5D). The Kv2.1 structure was determined to an overall resolution of 3.0 Å (Fig. 5B and SI Appendix, Fig. S9). An important difference between Kv7.1 and Kv2.1 becomes apparent at the junction of S4 and the S4–S5 linker. In KCNQ1, these residues form a helix– loop–helix motif, with three flexible amino acids (G245, G246, and T247) in or adjacent to the loop. Meanwhile, in Kv2.1, the junction is a helix–turn–helix motif. In other words, Kv7.1 has a natural propensity to form a loop in this region, which is not shared by the domain-swapped channel Kv2.1. The S4 movement that we observe in the intermediate and down conformations is centered exactly at this flexible “GGT” motif. Moreover, this motif (and in fact, most of the S4–S5 linker) is conserved among Kv7 family members but is absent in other domain-swapped Kv channels (Fig. 5A), indicating that it is a hallmark of the PIP2- gated KCNQ channels. Whether Shaker-related channels under an applied electric field undergo similar or distinct voltage sensor movements compared to KCNQ1 remains to be seen. Implications for Other Voltage-Dependent Channels. In KCNQ1, membrane polarization causes the S4 helix to displace by one helical turn (~5  Å) in the intermediate structure and most likely three helical turns (~15 Å) in the down structure, but the S4–S5 linker helix does not move appreciably (SI Appendix, Fig. S8A). One might argue that this is because the pore is closed due to the lack of PIP2. But, structures of KCNQ1 with an open pore are known (40, 46), and the S4–S5 linker helix occupies a similar position in those as well (SI Appendix, Fig. S8B). This finding suggests that the position of the S4–S5 linker helix is not strictly coupled to pore opening and closing in the KCNQ1 channel. It is still possible that small movements in the S4–S5 linker bias the conformational state of the pore, but S4–S5 helix movements are minimal when the S4 helix is displaced. What does this static S4–S5 helix in KCNQ1, if anything, suggest for other domain-swapped channels like Shaker-related Kv channels, Nav, or Cav channels? When the first molecular structure of a eukar- yotic voltage-dependent ion channel—the Shaker-related Kv1.2— was determined (16), the S4–S5 linker was found to contact S6 directly. A simple mechanical model for voltage-dependent regula- tion of the pore was proposed: When S4 moves in response to an electric field, the amino terminal end of the S4–S5 linker is displaced, applying a force on S6 and causing pore closure through straighten- ing of the S6 helix at a conserved “PxP” motif (53). Many years later, structures of chemically cross-linked or trapped Nav channels (21, 24, 25) and metal bridge–linked Kv4.2 channels (26) showed that it is indeed possible to trap channels in conformations consistent with the simple mechanical model. We observe in the present study, however, that KCNQ1 does not function according to this model. While KCNQ1 is an outlier among domain-swapped voltage-dependent channels for the reasons discussed above, we remain open minded to the possibility that the simple mechanical model assumed for other domain-swapped voltage-dependent ion channels (16, 21, 24, 25), despite support from mutational and chemical crossbridge data, could be incorrect. Mutations and chem- ical crossbridges likely do not replicate the forces applied to a polar- ized voltage sensor in membranes because an electric force field acts on all charged atoms spread throughout the protein. Ultimately, to know whether the simple mechanical model is correct for other domain-swapped channels, we will need to determine their structures in lipid bilayers with an applied electrostatic force field. Comparison of Voltage Sensor Movements in EAG and KCNQ1. We now know how the voltage sensors in two potassium channels, KCNQ1 (domain-swapped) and Eag (nondomain-swapped) (20), undergo conformational changes in response to an applied voltage difference across the membrane. For comparison, side views of the voltage sensors in these channels are shown in Fig. 6. In both channels, as S4 displaces downward (i.e., toward the cytoplasm), an extended interfacial segment is formed through a break in S4, which is accompanied by a remodeling of the connection between S4 and the S4–S5 linker helix (KCNQ1; Fig. 6A) or S4 and S5 (Eag; Fig. 6B) (20). Apparently, because S4 is both charged and hydrophobic, an interfacial location is energetically more favorable than an aqueous location. The extra amino acids that account for the downward displacement of S4 originate from the S3–S4 linker and the top of S3 in both channels. While the S4 displacement and interfacial helix formation are similar in KCNQ1 and Eag, the helices bend in opposite direc- tions with respect to the pore. In Eag, the polarized S4 bends toward the pore axis, causing it to clamp down on the pore-lining S6 helix, which prevents pore opening (Fig. 6B) (20). In KCNQ1, the polarized S4 bends away from the pore axis so that it occludes the PIP2-binding site. These variations show how the same struc- tural element—a voltage sensor—confers conformational sensi- tivity to an electric field in two Kv channels that differ both in their voltage sensor configuration (i.e., domain-swapped versus nondomain-swapped) and in their modulation by other effectors. Future studies of other voltage-dependent ion channels might uncover other interesting mechanisms for coupling the movement of S4 to gating the pore. On the Magnitude of S4 Displacement in KCNQ1. As we state above, our inability to define the register of the S4 helix main chain (SI Appendix, Fig. S7) prevents us from distinguishing with certainty whether the down map, with its occluded PIP2-binding site, corresponds to a two or three helical turn displacement of S4. A two-turn displacement was anticipated because that is what we observed in a polarized Eag channel (20), and what has been seen in cross-linked Nav and HCN voltage sensors (21, 23–25). Moreover, if S4 can displace three helical turns, it must pass through a two-turn-displaced intermediate. Why then would we not observe this intermediate? A possible answer lies in the unique S4 sequence of KCNQ1, which contains a neutral glutamine at “charged” position 3 (Fig. 5A). A two helical turn displacement would place the neutral glutamine into the highly negative- charged gating charge transfer center (Fig. 2C). For this reason, conformations with one or three helical turn displacements (which both place an arginine in the gating charge transfer center) may be energetically more stable in KCNQ1 than a conformation with two. If this is the case, a two helical turn displacement would function as a transient energy barrier in the conformational change of the voltage sensor. The notion that different residues are stable to varying degrees when they occupy the gating charge transfer center is apparent from functional measurements in other voltage-dependent ion channels. For instance, in the Shaker channel, it has been shown that it is more favorable for a lysine than an arginine to occupy the gating charge transfer center (8). Depending on the position of the substitution within the S4 helix, either the open or the closed state of the channel can be stabilized (corresponding to an up or down conformation of the voltage sensor). Given that even two positive-charged residues, arginine or lysine, can differ in their relative stability, it seems quite possible that a neutral glutamine behaves differently than an arginine. It is also useful to look at another example: Consider the Shaker channel and the domain-swapped channel Kv2.1 (Fig. 5A). Both have lysine at the fifth position (K5) in the gating 8 of 12   https://doi.org/10.1073/pnas.2301985120 pnas.org Fig. 6. Comparison of voltage sensor movements in KCNQ1 and Eag. (A and B) Stereoviews of KCNQ1 (A) and Eag (B) showing the up conformation (Left, red S4) and the down conformation (Right, blue S4). The view looks through one voltage sensor in each channel toward the pore axis. The approximate position of the lipid membrane bilayer is marked by yellow lines and the protein is shown in Cα trace representation. charge transfer center. In Shaker, all four residues above the gating charge transfer center are arginines, while Kv2.1 has a glutamine at position one (Q1) followed by three arginines. The gating charge estimated by nonlinear membrane capacitance for Shaker is ~12 to 14 elementary charges per channel (~3 to 3.5 per voltage sensor) and that for Kv2.1 is only ~6 to 7 per channel (~1.5 to 2 per voltage sensor) (54, 55). The gating charge esti- mates for Shaker indicate that the first residue (R1) does not traverse the membrane potential. Yet the presence of a neutral glutamine at the first position in Kv2.1 reduces the apparent gating charge in half. We suppose that in the down conformation of Kv2.1, there is a tendency for R2 to neutralize the extracellular negative-charged residue while R3 occupies the gating charge transfer center, consistent with the net movement of ~2 gating charges. These observations are consistent with the idea that it is more favorable for an arginine (compared to a glutamine) to interact with negative-charged residues in the voltage sensor. In summary, this study provides the structural description of a domain-swapped Kv channel in a lipid bilayer under the influence of a polarizing electric field. The structures reveal a mechanism in which the voltage sensor regulates the affinity of PIP2, thus con- trolling its ability to gate the pore. Materials and Methods Cell Lines. Sf9 (Spodoptera frugiperda Sf21) cells were used for production of baculovirus and were cultured in Sf-900 II SFM medium (GIBCO) supplemented with 100 U/mL penicillin and 100 U/mL streptomycin at 27 °C under atmospheric CO2. HEK293S GnTl− cells were used for protein expression and were cultured in Freestyle 293 medium (GIBCO) supplemented with 2% fetal bovine serum, 100 U/mL penicillin, and 100 U/mL streptomycin at 37 °C in 8% CO2. Expression and Purification of the KCNQ1–Calmodulin Complex. The KCNQ1(Kv7.1)–calmodulin complex (hereby referred to as KCNQ1) was expressed and purified as described before (40), with slight modifications. We used a con- struct corresponding to human KCNQ1 with N-terminal and C-terminal trunca- tions, leaving residues 76 to 620. The construct was cloned into the BacMan expression vector with a C-terminal green fluorescent protein (GFP)-His6 tag linked by a preScission protease (PPX) site (56). A separate BacMan expression vector without a tag was used for vertebrate calmodulin (CaM). Bacmids were generated for KCNQ1 and CaM using DH10Bac Escherichia coli cells. Baculoviruses for KCNQ1 and CaM were produced in SF9 cells transfected with bacmid DNA using the Cellfectin II reagent (Invitrogen). Baculovirus was amplified three times in suspension cultures of SF9 cells grown at 27 °C. Four liters of suspension cultures of HEK293S GnTI- at ~3 × 106 cells/mL were infected with 12% (v/v) of 5:1 KCNQ1:CaM baculovirus at 37 °C for ~8 h. Protein expres- sion was induced by adding 10 µM sodium butyrate, and the incubation tem- perature was changed to 30 °C for the duration of expression. Cell pellets were harvested ~48 h after induction and flash frozen in liquid nitrogen for later use. Four liters of cell pellet were resuspended in ~100 mL of lysis buffer (25 mM Tris pH 8.0, 300 mM KCl, 1 mM MgCl2, 5 mM CaCl2, 2 mM dithiothreitol (DTT), 1 µg/mL leupeptin, 1 µg/mL pepstatin, 1 mM benzamidine, 1 µg/mL aprotinin, 1  mM phenylmethylsulfonyl fluoride, 1  mM 4-(2-aminoethyl) benzenesulfo- nyl fluoride, and 0.1  mg/mL DNase), stirred for 10 min at 4 °C, and Dounce homogenized with a loose pestle till homogenous. The resultant suspension was clarified by centrifugation at 39,800 × g for 15 min at 4 °C. The pellet was resuspended in ~100 mL of lysis buffer and Dounce homogenized with a tight PNAS  2023  Vol. 120  No. 21  e2301985120 https://doi.org/10.1073/pnas.2301985120   9 of 12 pestle. To extract the KCNQ1–CaM complex, we added 15 mL of a 10%:2% n-no- decyl-β-D-maltopyranoside (DDM):cholesteryl hemisuccinate (CHS) mixture and stirred for 1 h at 4 °C. The mixture was clarified by centrifugation at 39,800 × g for 30 min at 4 °C. The supernatant was bound to ~2.5 mL GFP nanobody-coupled Sepharose resin (prepared in-house) (57) in a 250-mL conical centrifuge tube (Corning) by gentle rotation for 1 h at 4 °C. The resin was transferred to a glass gravity flow column (Bio-Rad) and washed with ~30 column volumes of wash buffer (10 mM Tris pH 8.0, 300 mM KCl, 0.05%:0.01% DDM:CHS, 1 mM CaCl2, and 2 mM DTT). The resin was resuspended in five column volumes of wash buffer, PPX (prepared in-house) was added at a concentration of 0.05 mg/mL to remove the GFP tag, and the solution was rotated for 1 h at 4 °C. The cleaved protein was collected in the flow through and a subsequent wash step with five column volumes of wash buffer. The protein was concentrated to ~500 µL at 3,000 × g and 4 °C using a 15-mL Amicon spin concentrator with a 100-kDa molecular weight cutoff mem- brane. The concentrated protein was filtered through a Corning 0.2 µm spin filter and then purified by size-exclusion chromatography (SEC) using a Superose 6 Increase column (10/300 GL) preequilibrated with SEC buffer (10 mM Tris pH 8.0, 300 mM KCl, 0.025%:0.005% DDM:CHS, 1 mM CaCl2, and 5 mM DTT). Fractions containing hKCNQ1 and calmodulin (SI Appendix, Fig. S1 A and B) were pooled and concentrated to an A280 of 3.8 mg/mL at 3,000 × g and 4 °C using a 4-mL Amicon spin concentrator with a 100-kDa molecular weight cutoff membrane. Purified protein was immediately used for reconstitution into liposomes. Reconstitution of the KCNQ1–Calmodulin Complex and Cryo-EM Grid Preparation. The purified KCNQ1 complex was reconstituted into liposomes con- sisting of 90%: 5%: 5% POPC:POPG:cholesterol (wt: wt: wt, Avanti Polar Lipids) (20). The phospholipids and sterol were mixed together in chloroform at a concentration of 10 mg/mL. Ten milligrams of the lipid mixture were dried to a thin film in a screw-top glass tube under a gentle stream of argon. The lipid film was further dried for ~3 h in a room-temperature vacuum desiccator, and then resuspended at a concentra- tion of 10 mg/mL by gentle vortexing in reconstitution buffer (10 mM Tris pH 8.0, 300 mM KCl and 1 mM DTT). Small unilamellar vesicles (SUVs) were formed by bath sonication (Branson Ultrasonics M1800) at room temperature till the solution was mostly transparent (A400 ~ 0.2), which typically took ~40 min. To permeabilize but not solubilize the lipid vesicles, the detergent C12E10 was added to the 10 mg/mL lipid stock solution to a final concentration of 2 mg/mL (5:1 lipid:detergent, wt/wt) and incubated on ice for ~15 min. Two hundred microliters of this permeabilized vesicle solution was mixed with 27 µL of the purified KCNQ1 complex (3.8 mg/ mL) and 173 µL of reconstitution buffer, giving a total reaction volume of 400 µL (chosen to ensure proper mixing in a 1.5 mL Eppendorf tube), a protein:lipid ratio of 1:20 (wt/wt), and a final lipid concentration of 5 mg/mL. The lipid–protein–detergent mixture was incubated on ice for ~1.5 h. Detergent was removed using adsorbent Bio-Beads SM-2 resin (Bio-Rad) by adding 20 mg of a 50% (wt/vol) Bio-Beads slurry in reconstitution buffer and rotating at 4 °C for ~14 h. The biobeads procedure was repeated twice again for 3 h each at 4 °C to ensure complete removal of detergent. The suspension was bath sonicated briefly (twice for 10 s each) after the biobeads step to minimize vesicle clumping. Polarized and unpolarized vesicles were prepared from the same batch of proteoliposomes. From an 8 mM stock in dimethyl sulfoxide, 2 µM valinomycin was added to the proteoliposomes and incubated for 30 min on ice. Polarized vesicles were prepared as follows: 70 µL of the above solution was added to a 0.5-mL Zeba spin desalting column (40 kDa cutoff, Thermo Scientific), preequili- brated with sodium reconstitution buffer (10 mM Tris pH 8.0 and 300 mM NaCl), to exchange the external K+ for Na+. The sample was centrifuged for ~20 to 30 s at room temperature at 1,500 × g and ~20 µL of flow-through containing vesicles was collected. The residual external K+ concentration is about 1 mM (20). Onto a glow-discharged Quantifoil R1.2/1.3 400 mesh Au grid, 3.5 µL of the polarized vesicle solution was immediately applied. The vesicle solution was incubated on the grid for 3 min at 20 °C under a humidity of 100%. The grid was then manually blotted from the edge of the grid using a piece of filter paper. Another 3.5 µL of the polarized vesicle solution was applied to the same grid for 20 s (58), and then the grid was blotted for 3 s with a blotting force of 0 and flash frozen in liquid ethane using a FEI Vitrobot Mark IV (FEI). Each grid with polarized vesicles used a freshly buffer exchanged sample. Grids for the unpolarized vesicles were frozen by skipping the buffer exchange step, i.e., directly applying the proteoliposomes (with valinomycin) on the Quantifoil grids. Expression and Purification of Kv2.1. Full-length human Kv2.1 (NP_004966.1) with a C-terminal GFP-His6 tag linked by a PPX site and full-length 14-3-3 protein epsilon (empirically found to increase the yield of Kv2.1, XP_040497056.1) were both cloned into a pBig1a vector from the biGBac system (59). Bacmids and baculovirus were generated, and protein was expressed in HEK293S GnTI- cells as described above for KCNQ1. The channel (hKv2.1) was purified following essentially the same protocol as KCNQ1 except that 150 mM KCl (instead of 300 mM KCl) was used for the wash buffer and calcium chloride was not included after the lysis buffer step. The final purification step entailed SEC on a Superose 6 Increase column (10/300 GL) preequilibrated with SEC buffer (10 mM Tris pH 8.0, 150 mM KCl, 0.03%:0.006% DDM:CHS, and 5 mM DTT). Fractions containing hKv2.1 (SI Appendix, Fig. S9A) were pooled and concentrated to an A280 of 1.4 mg/mL at 3,000 × g and 4 °C. Reconstitution of Kv2.1 and Cryo-EM Grid Preparation. Purified protein was reconstituted into liposomes of 90%:5%:5% POPC:POPG:cholesterol prepared in 150 mM KCl using a protein:lipid ratio of 1:20 (wt/wt), following the same protocol as for KCNQ1. A fourfold-molar excess of hanatoxin (compared to hKv2.1 monomers) isolated from Chilean rose tarantula (Grammostola rosea) venom (60) was incubated with the proteoliposomes before freezing grids, but the toxin was not visible in the cryo-EM reconstructions. Grids were frozen exactly as described for unpolarized vesicles containing KCNQ1 (but without added valinomycin). Liposome Flux Assay. The flux assay was carried out as described before (44), with minor modifications. The proteoliposome vesicles or control vesicles without protein (subjected to a mock reconstitution) prepared in 300 mM KCl were diluted 10-fold in isotonic sodium buffer (10 mM Tris pH 8.0 and 300 mM NaCl) imme- diately prior to the assay. Six microliters of the diluted vesicle solution was mixed with 6 µL ACMA solution (10 mM Tris pH 8.0, 300 mM NaCl, and 5 mM ACMA) and 12 µL buffer (10 mM Tris pH 8.0 and 300 mM NaCl). ACMA fluorescence was recorded every 5 s (excitation wavelength = 410 nm, emission wavelength = 490 nm) using a 384-well plate (Grainger) on a fluorescence plate reader (Tecan Infinite M1000). After the ACMA fluorescence stabilized, 6 µL of CCCP solution (10 mM Tris pH 8.0, 300 mM NaCl, and 15 mM CCCP) was added. The resultant KCNQ1-dependent flux, or in this case, the lack thereof because of the absence of PIP2, was measured. At the end of the assay, 2 µL of a 1.2-µM valinomycin solution (in 10 mM Tris pH 8.0 and trace dimethyl sulfoxide) was added to initiate K+ efflux from all the vesicles and determine the minimum ACMA fluorescence. The fluorescence data for each run were normalized by the fluorescence value right before the addition of CCCP (i.e., at 90 s). The normalized data were averaged across five independent measurements, and the mean and SDs are reported. Cryo-EM Data Acquisition and Processing. Data for the polarized and unpo- larized KCNQ1 liposomes were collected on the same microscope—a 300-keV FEI Titan Krios2 microscope located at the HHMI Janelia Research Campus. The microscope was equipped with a Gatan Image Filter (GIF) BioQuantum energy filter and a Gatan K3 camera. A total of 33,057 movies (polarized sample) or 19,998 movies (unpolarized sample) were recorded on Quantifoil grids in super- resolution mode using SerialEM (61). The movies were recorded with a physical pixel size of 0.839 Å (superresolution pixel size of 0.4195 Å) and a target defocus range of −1.0 to −2.0 µm. The total exposure time was ~2 s (fractionated into 50 frames) with a cumulative dose of ~60 e−/Å2. The data-processing workflow is detailed in SI Appendix, Figs. S2 and S3, and followed the same strategy we previously reported for Eag (20). Data processing was carried out using cryoSPARC v3.3.1 (62) and RELION 4.0 (63). The super- resolution movies were gain-normalized, binned by a factor of 2 with Fourier cropping, and corrected for full-frame and sample motion using the Patch motion correction tool (grid = 15 × 10). Contrast transfer function parameters were esti- mated from the motion-corrected micrographs using the Patch CTF estimation tool, which uses micrographs without dose weighting. All subsequent processing was performed on motion-corrected micrographs with dose weighting. Particle picking was initially carried out using the Blob picker. 2D classes with clear protein density were used to train a TOPAZ picking model (64), which was used to pick additional particles. Particles with clear protein density after 2D classification were pooled and duplicate picks were removed. An ab initio model was generated from 2D classes with clear secondary structure features, and 3D classification and refinement was carried out either in cryoSPARC or RELION as detailed in SI Appendix, Figs. S2 and S3. 10 of 12   https://doi.org/10.1073/pnas.2301985120 pnas.org Data for the hKv2.1 liposomes were collected on a 300-keV FEI Titan Krios microscope located at the HHMI Janelia Research Campus. The microscope was equipped with a spherical aberration corrector (Cs corrector), a GIF BioQuantum energy filter, and a Gatan K3 camera. A total of 17,007 movies were recorded on a single Quantifoil grid in superresolution mode using SerialEM. The movies were recorded with a physical pixel size of 0.844 Å (superresolution pixel size of 0.422 Å) and a target defocus range of −1.0 to −2.0 µm. The total exposure time was ~2 s (fractionated into 50 frames) with a cumulative dose of ~60 e−/Å2. Data processing was carried out as described for KCNQ1 liposomes. Model Building and Refinement. A structural model for the up conformation was built by docking the structure of KCNQ1–CaM in detergent micelles (PDB ID: 6UZZ) (40) into the up map and making adjustments where needed. The model was edited and refined using the ISOLDE (65) plugin in ChimeraX v1.2.0 (66) or WinCoot v0.98.1 (67) followed by real-space refinement in Phenix (68). The down and intermediate models were built starting from the up model. The up model was initially fit in the intermediate or down maps as a rigid body using Phenix. The S4 helix and the surrounding regions were manually adjusted and then a final step of real-space refinement was carried out in Phenix. The quality of the final models was evaluated using the MolProbity plugin in Phenix (SI Appendix, Table S1). Graphical representations of models and cryo-EM density maps were prepared using PyMOL (69) and ChimeraX. Data, Materials, and Software Availability. Cryo-EM density maps of the KCNQ1 channel with the voltage sensor in the up, intermediate, and down conformation have been deposited in the electron microscopy data bank under accession codes EMD-40508 (70), EMD-40509 (71), and EMD-40510 (72), respectively. Atomic coordinates of the KCNQ1 channel with the voltage sen- sor in the up, intermediate, and down conformation have been deposited in the protein data bank under accession codes 8SIK (73), 8SIM (74), and 8SIN (75), respectively. ACKNOWLEDGMENTS. We thank Rui Yan, Zhiheng Yu, and the team at the Howard Hughes Medical Institute Janelia CryoEM Facility for their effort in cryo-EM microscope operation and data collection; Mark Ebrahim, Johanna Sotiris, and Honkit Ng at the Evelyn Gruss Lipper Cryo-EM Resource Center for assistance with cryo-EM grid screening; Yi Chun Hsiung for assistance with insect and mammalian cell cultures; Dr. Chen Zhao and other members of the MacKinnon Lab for helpful discussions; and Dr. Jue Chen and her group for their input. Dr. Chia-Hseuh Lee carried out the cloning and initial biochemical characterization of the Kv2.1 channel. V.S.M. is supported by the Jane Coffin Childs Memorial Fund Fellowship. R.M. is an investigator in the HHMI. 1. 2. B. Hille, Ionic Channels of Excitable Membranes (Sinauer Associates, ed. 3, 2001) (January 27, 2023). Y. Murata, H. Iwasaki, M. Sasaki, K. Inaba, Y. Okamura, Phosphoinositide phosphatase activity coupled to an intrinsic voltage sensor. Nature 435, 1239–1243 (2005). 30. T. Jespersen, M. Grunnet, S.-P. Olesen, The KCNQ1 potassium channel: From gene to physiological function. Physiology 20, 408–416 (2005). 31. J. Robbins, KCNQ potassium channels: Physiology, pathophysiology, and pharmacology. Pharmacol. Ther. 90, 1–19 (2001). 3. M. Noda et al., Primary structure of electrophorus electricus sodium channel deduced from cDNA 32. P. L. Hedley et al., The genetic basis of long QT and short QT syndromes: A mutation update. Hum. 4. 5. 6. 7. 8. 9. sequence. Nature 312, 121–127 (1984). T. Tanabe et al., Primary structure of the receptor for calcium channel blockers from skeletal muscle. Nature 328, 313–318 (1987). B. L. Tempel, D. M. Papazian, T. L. Schwarz, Y. N. Jan, L. Y. Jan, Sequence of a probable potassium channel component encoded at Shaker locus of Drosophila. Science 237, 770–775 (1987). Y. Jiang et al., X-ray structure of a voltage-dependent K+ channel Nature 423, 33–41 (2003). S. B. Long, X. Tao, E. B. Campbell, R. MacKinnon, Atomic structure of a voltage-dependent K+ channel in a lipid membrane-like environment Nature 450, 376–382 (2007). X. Tao, A. Lee, W. Limapichat, D. A. Dougherty, R. MacKinnon, A gating charge transfer center in voltage sensors. Science 328, 67–73 (2010). A. L. Hodgkin, A. F. Huxley, A quantitative description of membrane current and its application to conduction and excitation in nerve. J. Physiol. 117, 500–544 (1952). Mutat. 30, 1486–1511 (2009). 33. H. Zhang et al., PIP2 activates KCNQ channels, and its hydrolysis underlies receptor-mediated inhibition of M currents. Neuron 37, 963–975 (2003). 34. B.-C. Suh, B. Hille, Recovery from muscarinic modulation of M current channels requires phosphatidylinositol 4,5-bisphosphate synthesis. Neuron 35, 507–520 (2002). 35. G. Loussouarn et al., Phosphatidylinositol-4,5-bisphosphate, PIP2, controls KCNQ1/KCNE1 voltage- gated potassium channels: A functional homology between voltage-gated and inward rectifier K+ channels EMBO J. 22, 5412–5421 (2003). 36. C. P. Ford, P. L. Stemkowski, P. A. Smith, Possible role of phosphatidylinositol 4,5, bisphosphate in luteinizing hormone releasing hormone-mediated M-current inhibition in bullfrog sympathetic neurons. Eur. J. Neurosci. 20, 2990–2998 (2004). 37. P. Delmas, D. A. Brown, Pathways modulating neural KCNQ/M (Kv7) potassium channels. Nat. Rev. 10. K. G. Beam, C. M. Knudson, J. A. Powell, A lethal mutation in mice eliminates the slow calcium Neurosci. 6, 850–862 (2005). current in skeletal muscle cells. Nature 320, 168–170 (1986). 38. M. A. Zaydman et al., Kv7.1 ion channels require a lipid to couple voltage sensing to pore opening. 11. J. Payandeh, T. Scheuer, N. Zheng, W. A. Catterall, The crystal structure of a voltage-gated sodium Proc. Natl. Acad. Sci. U.S.A. 110, 13180–13185 (2013). channel. Nature 475, 353–358 (2011). 39. J. Cui, Voltage-dependent gating: Novel insights from KCNQ1 channels. Biophys. J. 110, 14–25 12. J. Sun, R. MacKinnon, Cryo-EM structure of a KCNQ1/CaM complex reveals insights into congenital (2016). long QT syndrome. Cell 169, 1042–1050.e9 (2017). 40. J. Sun, R. MacKinnon, Structural basis of human KCNQ1 modulation and gating. Cell 180, 340–347. 13. J. Wu et al., Structure of the voltage-gated calcium channel Cav1.1 at 3.6 Å resolution. Nature 537, e9 (2020). 191–196 (2016). 14. H. Shen et al., Structure of a eukaryotic voltage-gated sodium channel at near-atomic resolution. Science 355, eaal4326 (2017). 41. Z. A. McCrossan, G. W. Abbott, The MinK-related peptides. Neuropharmacology 47, 787–821 (2004). 42. R. Barro-Soria, M. E. Perez, H. P. Larsson, KCNE3 acts by promoting voltage sensor activation in KCNQ1. Proc. Natl. Acad. Sci. U.S.A. 112, E7286–E292 (2015). 15. M. Liao, E. Cao, D. Julius, Y. Cheng, Structure of the TRPV1 ion channel determined by electron cryo- 43. L. Shamgar et al., Calmodulin is essential for cardiac IKS channel gating and assembly: Impaired microscopy. Nature 504, 107–112 (2013). function in long-QT mutations. Circ. Res. 98, 1055–1063 (2006). 16. S. B. Long, E. B. Campbell, R. MacKinnon, Voltage sensor of Kv1.2: Structural basis of electromechanical coupling. Science 309, 903–908 (2005). 44. Z. Su, E. C. Brown, W. Wang, R. MacKinnon, Novel cell-free high-throughput screening method for pharmacological tools targeting K + channels Proc. Natl. Acad. Sci. U.S.A. 113, 5748–5753 (2016). 17. C.-H. Lee, R. MacKinnon, Structures of the human HCN1 hyperpolarization-activated channel. Cell 45. J. H. Morais-Cabral, Y. Zhou, R. MacKinnon, Energetic optimization of ion conduction rate by the K+ 168, 111–120.e11 (2017). selectivity filter Nature 414, 37–42 (2001). 18. X. Tao, R. K. Hite, R. MacKinnon, Cryo-EM structure of the open high-conductance Ca2+-activated 46. D. Ma et al., Structural mechanisms for the activation of human cardiac KCNQ1 channel by electro- K+ channel Nature 541, 46–51 (2017). mechanical coupling enhancers. Proc. Natl. Acad. Sci. U.S.A. 119, e2207067119 (2022). 19. J. R. Whicher, R. MacKinnon, Structure of the voltage-gated K+ channel Eag1 reveals an alternative 47. Y. Zheng et al., Structural insights into the lipid and ligand regulation of a human neuronal KCNQ voltage sensing mechanism Science 353, 664–669 (2016). channel. Neuron 110, 237–247.e4 (2022). 20. V. S. Mandala, R. MacKinnon, Voltage-sensor movements in the Eag Kv channel under an applied electric field. Proc. Natl. Acad. Sci. U.S.A. 119, e2214151119 (2022). 21. T. Clairfeuille et al., Structural basis of α-scorpion toxin action on Nav channels. Science 363, eaav8573 (2019). 22. J. Guo et al., Structure of the voltage-gated two-pore channel TPC1 from Arabidopsis thaliana. Nature 531, 196–201 (2016). 48. K. J. Ruscic et al., IKs channels open slowly because KCNE1 accessory subunits slow the movement of S4 voltage sensors in KCNQ1 pore-forming subunits. Proc. Natl. Acad. Sci. U.S.A. 110, E559–E566 (2013). 49. F. Miceli, E. Vargas, F. Bezanilla, M. Taglialatela, Gating currents from Kv7 channels carrying neuronal hyperexcitability mutations in the voltage-sensing domain. Biophys. J. 102, 1372–1382 (2012). 50. P. Hou et al., Inactivation of KCNQ1 potassium channels reveals dynamic coupling between voltage 23. C.-H. Lee, R. MacKinnon, Voltage sensor movements during hyperpolarization in the HCN channel. sensing and pore opening. Nat. Commun. 8, 1730 (2017). Cell 179, 1582–1589.e7 (2019). 24. G. Wisedchaisri et al., Resting-state structure and gating mechanism of a voltage-gated sodium channel. Cell 178, 993–1003.e12 (2019). 51. M. A. Zaydman et al., Domain–domain interactions determine the gating, permeation, pharmacology, and subunit modulation of the IKs ion channel. Elife 3, e03606 (2014). 52. H. Misonou, D. P. Mohapatra, J. S. Trimmer, Kv2.1: A voltage-gated K+ channel critical to dynamic 25. H. Xu et al., Structural basis of Nav1.7 inhibition by a gating-modifier spider toxin. Cell 176, control of neuronal excitability NeuroToxicology 26, 743–752 (2005). 702–715.e14 (2019). 26. W. Ye et al., Activation and closed-state inactivation mechanisms of the human voltage-gated KV4 53. Y. Jiang et al., The open pore conformation of potassium channels. Nature 417, 523–526 (2002). 54. S. K. Aggarwal, R. MacKinnon, Contribution of the S4 segment to gating charge in the shaker K+ channel complexes. Mol. Cell 82, 2427–2442.e4 (2022). channel Neuron 16, 1169–1177 (1996). 27. G. Huang et al., Unwinding and spiral sliding of S4 and domain rotation of VSD during the 55. S. K. Aggarwal, Analysis of the Voltage Sensor in a Voltage-Activated Potassium Channel (Harvard electromechanical coupling in Nav1.7. Proc. Natl. Acad. Sci. U.S.A. 119, e2209164119 (2022). 28. J. Barhanin et al., KvLQT1 and IsK (minK) proteins associate to form the IKS cardiac potassium current. Nature 384, 78–80 (1996). University, Cambridge, MA, 1996) (January 27, 2023). 56. A. Goehring et al., Screening and large-scale expression of membrane proteins in mammalian cells for structural studies. Nat. Protoc. 9, 2574–2585 (2014). 29. M. C. Sanguinetti et al., Coassembly of KVLQT1 and minK (IsK) proteins to form cardiac IKS 57. A. Kirchhofer et al., Modulation of protein properties in living cells using nanobodies. Nat. Struct. potassium channel. Nature 384, 80–83 (1996). Mol. Biol. 17, 133–138 (2010). PNAS  2023  Vol. 120  No. 21  e2301985120 https://doi.org/10.1073/pnas.2301985120   11 of 12 58. L. Tonggu, L. Wang, Cryo-EM sample preparation method for extremely low concentration 67. P. Emsley, B. Lohkamp, W. G. Scott, K. Cowtan, Features and development of Coot. Acta Crystallogr. D liposomes. Ultramicroscopy 208, 112849 (2020). Biol. Crystallogr. 66, 486–501 (2010). 59. F. Weissmann et al., biGBac enables rapid gene assembly for the expression of large multisubunit 68. D. Liebschner et al., Macromolecular structure determination using X-rays, neutrons and protein complexes. Proc. Natl. Acad. Sci. U.S.A. 113, E2564–E2569 (2016). 60. K. J. Swartz, R. MacKinnon, An inhibitor of the Kv2.1 potassium channel isolated from the venom of a Chilean tarantula. Neuron 15, 941–949 (1995). 61. D. N. Mastronarde, Automated electron microscope tomography using robust prediction of electrons: Recent developments in Phenix. Acta Crystallogr. Sect. Struct. Biol. 75, 861–877 (2019). 69. L. L. C. Schrödinger, The PyMOL Molecular Graphics System (Ver. 2, Schrödinger, 2015). 70. V. S. Mandala, R. MacKinnon, KCNQ1 with voltage sensor in the up conformation. EMDB. https:// specimen movements. J. Struct. Biol. 152, 36–51 (2005). www.ebi.ac.uk/emdb/EMD-40508. Deposited 14 April 2023. 62. A. Punjani, J. L. Rubinstein, D. J. Fleet, M. A. Brubaker, cryoSPARC: Algorithms for rapid unsupervised 71. V. S. Mandala, R. MacKinnon, KCNQ1 with voltage sensor in the intermediate conformation. EMDB. cryo-EM structure determination. Nat. Methods 14, 290–296 (2017). https://www.ebi.ac.uk/emdb/EMD-40509. Deposited 14 April 2023. 63. D. Kimanius, L. Dong, G. Sharov, T. Nakane, S. H. W. Scheres, New tools for automated cryo-EM 72. V. S. Mandala, R. MacKinnon, KCNQ1 with voltage sensor in the down conformation. EMDB. https:// single-particle analysis in RELION-4.0. Biochem. J. 478, 4169–4185 (2021). www.ebi.ac.uk/emdb/EMD-40510. Deposited 14 April 2023. 64. T. Bepler et al., Positive-unlabeled convolutional neural networks for particle picking in cryo-electron 73. V. S. Mandala, R. MacKinnon, KCNQ1 with voltage sensor in the up conformation. PDB. https://www. micrographs. Nat. Methods 16, 1153–1160 (2019). rcsb.org/structure/8SIK. Deposited 14 April 2023. 65. T. I. Croll, ISOLDE: A physically realistic environment for model building into low-resolution electron- 74. V. S. Mandala, R. MacKinnon, KCNQ1 with voltage sensor in the intermediate conformation. PDB. density maps. Acta Crystallogr. Sect. Struct. Biol. 74, 519–530 (2018). https://www.rcsb.org/structure/8SIM. Deposited 14 April 2023. 66. E. F. Pettersen et al., UCSF ChimeraX: Structure visualization for researchers, educators, and 75. V. S. Mandala, R. MacKinnon, KCNQ1 with voltage sensor in the down conformation. PDB. https:// developers. Protein Sci. 30, 70–82 (2021). www.rcsb.org/structure/8SIN. Deposited 14 April 2023. 12 of 12   https://doi.org/10.1073/pnas.2301985120 pnas.org
null
10.1002_advs.202200181.pdf
Data Availability Statement The data that support the findings of this study are available from the cor- responding authors upon reasonable request.
Data Availability Statement The data that support the findings of this study are available from the corresponding authors upon reasonable request.
RESEARCH ARTICLE www.advancedscience.com Moiré-Driven Topological Transitions and Extreme Anisotropy in Elastic Metasurfaces Simon Yves, Matheus Inguaggiato Nora Rosa, Yuning Guo, Mohit Gupta, Massimo Ruzzene,* and Andrea Alù* The twist angle between a pair of stacked 2D materials has been recently shown to control remarkable phenomena, including the emergence of flat-band superconductivity in twisted graphene bilayers, of higher-order topological phases in twisted moiré superlattices, and of topological polaritons in twisted hyperbolic metasurfaces. These discoveries, at the foundations of the emergent field of twistronics, have so far been mostly limited to explorations in atomically thin condensed matter and photonic systems, with limitations on the degree of control over geometry and twist angle, and inherent challenges in the fabrication of carefully engineered stacked multilayers. Here, this work extends twistronics to widely reconfigurable macroscopic elastic metasurfaces consisting of LEGO pillar resonators. This work demonstrates highly tailored anisotropy over a single-layer metasurface driven by variations in the twist angle between a pair of interleaved spatially modulated pillar lattices. The resulting quasi-periodic moiré patterns support topological transitions in the isofrequency contours, leading to strong tunability of highly directional waves. The findings illustrate how the rich phenomena enabled by twistronics and moiré physics can be translated over a single-layer metasurface platform, introducing a practical route toward the observation of extreme phenomena in a variety of wave systems, potentially applicable to both quantum and classical settings without multilayered fabrication requirements. S. Yves, A. Alù Photonics Initiative Advanced Science Research Center City University of New York New York, NY 10031, USA E-mail: aalu@gc.cuny.edu M. I. N. Rosa, Y. Guo, M. Gupta, M. Ruzzene Department of Mechanical Engineering University of Colorado Boulder Boulder, CO 80309, USA E-mail: massimo.ruzzene@colorado.edu A. Alù Physics Program Graduate Center City University of New York New York, NY 10026, USA The ORCID identification number(s) for the author(s) of this article can be found under https://doi.org/10.1002/advs.202200181 © 2022 The Authors. Advanced Science published by Wiley-VCH GmbH. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. DOI: 10.1002/advs.202200181 1. Introduction New discoveries in condensed matter physics have recently shown how a twist in pairs of 2D stacked layers can produce highly unexpected emergent phenomena. Notably, the fine-tuning of such twist allows the emergence of a magic angle at which a plethora of new phenom- ena can be observed, including flat-band superconductivity,[1] the quantum Hall effect,[2] the creation of moiré excitons,[3–8] as well as interlayer magnetism.[9] Based on these concepts, atomic photonic crystals in twisted bilayer graphene have shown the ability to route solitons[10,11] and pro- duce quasi-crystalline phases,[12] higher- topology,[13] non-Abelian gauge order potential,[14] and helical topological state mosaics.[15,16] These phenomena, at the heart of the thriving field of twistronics,[17] arise from the hybridization of the band structures associated with the two isolated monolayers, and the associated formation of moiré superlattices. Macroscopic-scale implementations of these concepts using phononic and photonic metamaterials[18,19] have demonstrated flat bands in macroscopic analogues of bi- layer graphene,[20–23] field localization within moiré lattices,[24–26] the destruction of valley topological protection,[27] artificial gauge fields,[28] and broadband tunable bianisotropy for biosensing applications.[29–31] These concepts have also been recently transposed to optical metamaterials, based on extreme anisotropic responses over hy- perbolic metasurfaces (HMTs).[32] Their iso-frequency contours (IFCs) support an open, hyperbolic topology,[33–37] featuring wave propagation with enhanced local density of states, and enabling subwavelength imaging, as well as negative refraction and canal- ization, inherently broadband in nature. By stacking two hyper- bolic metasurfaces and rotating one with respect to the other, it is possible to largely modify the IFCs, inducing transitions be- tween different topologies, from hyperbolic to elliptical.[38] Such effect is the wave analogue of a Lifshitz transition in electronic band structures,[39] which is known to play a crucial role in the physics of Weyl and Dirac semimetals.[40] These exciting phe- nomena have also been recently demonstrated in polaritonic systems.[41–43] The remarkable features of twisted bilayers exploit the in- terplay between two distinct layers with exotic wave responses, Adv. Sci. 2022, 9, 2200181 2200181 (1 of 8) © 2022 The Authors. Advanced Science published by Wiley-VCH GmbH www.advancedsciencenews.com www.advancedscience.com especially for field localization effects, and generally require a precise control over their coupling, alignment and twist angle. Hence, experimental setups, however reconfigurable, are quite challenging.[44] In an attempt to circumvent such difficulties, a few recent studies have theoretically explored the emergence of analogous responses in single-layer systems, with interac- tions or properties modulated by a second virtual layer. For in- stance, quasi-flat bands, Dirac cones, and quantum anomalous phases have been predicted in modulated optical lattices,[45,46] while topological spectral gaps characterized by second Chern numbers akin to the 4D quantum Hall effect were illustrated in phononic lattices.[47] Lifting the requirement of two stacked layers opens new prospects for the implementation of twistronics across several electronic, photonic, and phononic platforms. Toward this goal, in this Letter we explore the effects of emergent moiré patterns in monolayer pillared metasurfaces formed by the relative twist of two 2D spatial features: the lattice defined by the position of the pillars and the one defined by the anisotropic modula- tion profile of their height, which defines their resonant fea- tures. We first demonstrate that aligning these two lattices, re- sulting in an untwisted metasurface, and controlling their fea- tures, can produce a wide range of elliptical and hyperbolic IFCs. Next, we show that introducing a relative rotation between these two 2D lattices generates quasi-periodic moiré patterns govern- ing topological transitions between open and closed IFCs for spe- cific twist angles. Such transitions inherently occur in a differ- ent way from those emerging in twisted bilayers,[32,41,42,43] for which the interplay between two material hyperbolic surfaces de- fines the transition instead of the emerging moiré patterns.We demonstrate the extreme wave phenomena in such twisted inter- leaved lattices with a highly reconfigurable metasurface formed by LEGO pillar-cone resonators over an elastic plate, which is a 2D extension of previously employed implementations used to study the role of disorder[48] and quasi-periodicity.[49] Our re- sults show the great potential of this platform to study analogues of condensed matter phenomena at the macroscopic scale and in classical settings, and open the door to applications harness- ing both strong anisotropy and moiré physics for enhanced wave manipulation. 2. Results and Discussion Our metasurface consists of a thin elastic plate featuring an ar- ray of pillars in a square lattice of period a (Figure 1a). The pillars can be modeled as mechanical dipolar resonators cou- pled to the transverse motion of the plate, and they are char- acterized by two bending resonant modes (along x and y) of equal frequency due to their symmetric cross section. Pillar-type resonators have been employed in the design of metamateri- als and metasurfaces, notably in the context of bandgaps,[50–54] cloaking,[55] and for seismic mitigation.[56] Nonsymmetric em- bedded resonators have previously been used to implement elas- tic hyperbolic metamaterials,[57–61] whereby large asymmetric couplings within the plate can generate anisotropic effective properties, which can be exploited in the context of waveguiding and subdiffraction imaging. Rather than breaking the resonator symmetry, here we induce strong anisotropy through lattice ef- fects, by spatially modulating the resonant features of the array with a wavelength 𝜆 = Na. This effect introduces a spatial mis- match within the lattice, effectively creating a resonant macrocell including N distinct resonators responsible for asymmetric cou- plings across the metasurface. More specifically, we modify the pillar heights according to the modulation profile S (x, y, 𝜃) = cos [2𝜋/𝜆(cos 𝜃x + sin 𝜃y)], where 𝜃 is a twist angle measured with re- spect to the x axis (Figure 1a). The height hn of each pillar defines the resonant frequency of the dominant mode of interest, and it is assigned by sampling the modulation surface at the lattice sites xn, yn, i.e., hn = h0 [1 + 𝛼S(xn,yn,𝜃)], where h0 is the mean height and 𝛼 is the modulation amplitude. This scheme generates two interleaved spatial features, consisting of the underlying square lattice of period a and of the sampled height distribution at the lattice sites. We begin by highlighting the wave propagation features of the plate in the untwisted (𝜃 = 0◦ , 𝛼 = 0.1) configuration, with pe- riod a along y and Na along x. We consider N = 2, resulting in a diatomic lattice of resonators (Figure 1b), for which the band structure is shown in Figure 1c (the complete band structure can be found in Figure S1, Supporting Information). All band struc- ture computations and response simulations here are calculated with COMSOL Multiphysics, with details provided in the Sec- tion S1 (Supporting Information). The interleaving of the two lattices corresponding to the position of the resonators on the plate and to the resonance modulation, introduces an asymme- try and therefore a mismatch in IFCs along x and y, which re- sults in hyperbolic IFCs around the resonance, two of which are highlighted by black lines in the figure, as well as elliptical ones. The existence of hyperbolic and elliptical IFCs is confirmed by simulating the harmonic response due to a point source exci- tation applied at the center of a finite sample comprising 80 × 80 unit cells. The resulting out-of-plane displacement field, and its Fourier transform (FT) displayed in Figure 1d, illustrate the emergence of hyperbolic and elliptic bands for the three frequen- cies marked in Figure 1c, namely 485 Hz, 625 Hz and 670 Hz. Modifying the height modulation, quantified by the parameter 𝛼, can dramatically change the coupling asymmetry within the sur- face, and correspondingly tailor the IFC shape. Two examples are displayed in Figure 1e at the frequencies of the hyperbolic con- tours in Figure 1c. In the top panel, at 485 Hz, an increase in 𝛼 results in an inversion of IFC curvature, which changes from hyperbolic, to flat, to open-elliptical, to finally close into an ellip- tical shape, demonstrating a topological transition. In the bottom panel, at 625 Hz, another topological transition from hyperbolic to elliptical phases occurs, this time as 𝛼 decreases. In this case, the presence of elliptical IFCs at neighboring frequencies (Fig- ure 1c) facilitates the transition between the two regimes, requir- ing smaller variations of 𝛼 to drive the process. We confirm these phenomena experimentally using our elastic metasurface platform, exploiting the fact that underlying sampling lattice is square. Our metasurface comprises 44 × 44 resonators whose heights are modulated with 𝜆 = 2a (Figure 1f), and we choose the maximum value of 𝛼 allowed by the LEGO pillar geometry, as shown in the inset. The resonances are tuned by sliding the cones along the pillars, following the modulation of hn defined above (Figure 1f, inset). Our LEGO platform provides straightforward tunability and reconfigurability, which we harness to demonstrate extreme wave phenomena and topological transitions. The plate is excited at its center by an Adv. Sci. 2022, 9, 2200181 2200181 (2 of 8) © 2022 The Authors. Advanced Science published by Wiley-VCH GmbH www.advancedsciencenews.com www.advancedscience.com Figure 1. a) Moiré interleaved metasurface: A square lattice of pillars whose heights are modulated according to a rotating profile, here for 𝛼 = 0.1. b) Schematic of the periodic system with 𝜃 = 0◦ , and corresponding unit cell (inset). c) Numerical band structure of (b) with three contours corresponding to hyperbolic (at 485 Hz) along y, hyperbolic along x (at 625 Hz) and elliptical (at 670 Hz) highlighted with black lines. d) Simulated displacement field and corresponding spatial FT zoomed in at the center of the Brillouin zone for the three frequencies highlighted in (c). e) Modification of the IFCs as a function of height modulation at 485 Hz (top) and 625 Hz (bottom). f) Corresponding sample made of LEGO elements with cones at alternating heights (inset). g) Experimentally measured out-of-plane displacement field map and corresponding spatial FT at 345 Hz (left), 470 Hz (center), and 510 Hz (left). electrodynamic shaker, which applies a pseudo-random excita- tion in the 200 − 700 Hz range, and the resulting out-of-plane wave fields are recorded by a scanning Doppler vibrometer (see Figure S2, Supporting Information for the experimental setup). While some hybridization exists between symmetric and asym- metric Lamb waves in the close vicinity of the resonances due to out-of-plane breaking of mirror symmetry, the modes of interest are mainly polarized along the out-of-plane direction (see Figure S3, Supporting Information for the out-of-plane polarization). Figure 1g displays the real and reciprocal space maps of the mea- sured fields at three selected frequencies (345, 470, and 510 Hz): the measured hyperbolic and elliptical propagation are consis- tent with those predicted in simulations, with a small frequency shift attributed to minor differences between experimental and numerical models (see Figure S1, Supporting Information for the simulation of the LEGO lattice band structure). These results clearly show that a spatially anisotropic resonance frequency modulation can generate broadband hyperbolic mechanical Lamb waves, easily implemented over our platform. Moreover, the straightforward tuning of the height modulation amplitude enables a precise control and drastic variations of the supported IFCs. Next, we explore the effect of rotating the interleaved lattices, by twisting the modulation profile relative to the underlying square lattice of resonators (Figure 1a). The misalignment be- tween the lattice and modulation profile produces moiré patterns associated with complex spatial arrangements of the couplings. As illustrated in Figure 2a, 2D modulation patterns with a strong angular dependence appear for 0◦ < 𝜃 < 45◦ (after 45◦, the be- havior is simply inverted because of symmetry). For a generic twist angle, the resulting pattern is quasiperiodic, and the peri- odicity is only restored for specific angles 𝜃 = cos −1(p2/q), where {p2,q} are integers belonging to a Pythagorean triple satisfying p2 1 + p2 2 = q2.[24] These periodic configurations are characterized by unit cells that are typically very large: for instance, the two smallest super-cells are obtained for 𝜃 = cos −1(4/5)≅36.87°, re- sulting in a 5 × 10 super-cell, and 𝜃 = cos −1(12/13) ≅22.62°, re- sulting in a 13 × 26 super-cell. The complexity of the periodic angles and the increasing size of the super-cells makes the analy- sis through Bloch procedures very challenging, if not prohibitive. Adv. Sci. 2022, 9, 2200181 2200181 (3 of 8) © 2022 The Authors. Advanced Science published by Wiley-VCH GmbH www.advancedsciencenews.com www.advancedscience.com Figure 2. a) Pillar height modulation profile as a function of the rotation angle (zoomed detail in inset). b) Simulated out-of-plane displacement field maps (top) and spatial FT (bottom) as a function of the rotation angle for hyperbolicity along the y axis at 485 Hz. c) Same as (b) for the hyperbolicity along the x axis at 625 Hz. Instead, we observe the proposed moiré phenomena by analyz- ing the out-of-plane displacement in real and reciprocal spaces for fixed frequency as a function of the twist angle. Overall, we find evidence of a very rich behavior of the result- ing metasurfaces, whereby different IFC transitions occur at dif- ferent frequencies. We focus on the two hyperbolic regimes pre- sented in Figure 1c. The first example at 485 Hz is illustrated in Figure 2b, at which the wave directionality rotates in the opposite direction compared to the twist angle, until 𝜃 = 30°. The effec- tive wavelength of the guided waves drastically increases in the small angle regime (𝜃 < 10◦), as displayed on Figure 2b, evidence of the progressive emergence of a moiré pattern introducing a super-lattice with long spatial wavelengths (Figure 2a). We note that the metasurface response is strongly affected by the twist an- gle, as long as the wavelength of the moiré pattern is larger than the wavelength of the propagating waves. In reciprocal space, we correspondingly observe the presence of spatial harmonics that move away from the center of the Brillouin zone toward larger wavenumbers as the twist angle increases, in line with a decrease in moiré periodicity. When 𝜃 gets closer to 45◦, an inverted phe- nomenon arises, albeit less noticeable in the 2D modulation pro- file, and some spatial harmonics move closer to the center, caus- ing a distortion of the IFCs. Similar to the case at 𝜃 = 4◦ , this effect hampers surface wave propagation, which is linked to the emergence of partial bandgaps and band flattening caused by the interaction of different spatial harmonics. The rigorous analysis of this phenomenon is inherently complex due to the quasiperi- odic nature of the system and it goes beyond the scope of this work. A different evolution of the supported band structure as a func- tion of the twist angle can be observed in Figure 2c, correspond- ing to excitation at 625 Hz. At 𝜃 = 0◦ the wave propagation is hyperbolic but with opposite orientation. As 𝜃 increases, the field progressively loses directionality, and becomes completely delocalized above 10◦. In reciprocal space (bottom row), this field evolution manifests itself as a topological transition of the asso- ciated IFCs, which evolve from open hyperbolic to closed ellip- tical for increasing twist angle. Similar to Figure 2b, this is ex- plained by the distortion of the original contours due to emerging quasi-periodic modulation and super-lattice patterns. Although the transition here is driven by the twist angle, its emergence is inherently different from the ones observed in previous studies of twisted hyperbolic metasurface bilayers:[32,41,42,43] here the sys- tem consists of a single layer whose interleaved spatial modu- lation lattices and emerging moiré patterns directly control the coupling between resonators, governing the IFC features. Adv. Sci. 2022, 9, 2200181 2200181 (4 of 8) © 2022 The Authors. Advanced Science published by Wiley-VCH GmbH www.advancedsciencenews.com www.advancedscience.com Figure 3. a) Doubling the modulation period enables a better sampling of the resulting modulation profile. b) The modulation profile is better preserved during the twist. c) Pillar height modulation as a function of rotation angle, with a zoom-in inset for four angles. d) Simulated out-of-plane displacement field maps (top) and spatial FT (bottom) as a function of the rotation angle for hyperbolicity along y at 470 Hz. e–g) Same as (d) in the case of topological transitions at 595, 620, and 645 Hz, respectively. Next, we explore the possibility of precisely tuning the dis- persion profile across smoother topological transitions. As noted above, the spatial features emerging from twisting the two inter- leaved lattices are characterized by a large change in their peri- odicities as the twist angle is varied. Increasing the twist angle can quickly degrade the nature of the modulation, as a function of how coarse the height modulation is sampled by the period of the square lattice. The associated angular sensitivity of this phe- nomenon can be expectedly reduced by increasing the periodicity of the modulation profile. For example, Figure 3a considers the untwisted scenario when the modulation period is doubled to 𝜆 = 4a. This change translates into a smoother correlation between twist angle and resulting anisotropic contours, with considerably smaller distortions (Figure 3b,c). As a consequence, the original spatially anisotropic distribution of couplings within the mono- layer, which is responsible for the hyperbolic features, can be bet- ter preserved as the twist angle changes. The resulting wave propagation features are summarized in Figure 3. Figure 3d considers the frequency for which the original untwisted structure supports directional hyperbolic waves ori- ented along the y axis, for excitation at 470 Hz. Although super- lattice phenomena occur at small angles, their impact on the IFCs Adv. Sci. 2022, 9, 2200181 2200181 (5 of 8) © 2022 The Authors. Advanced Science published by Wiley-VCH GmbH www.advancedsciencenews.com www.advancedscience.com Figure 4. a) LEGO metasurfaces as a function of rotation angle for 𝜆 = 4a. b) Experimentally measured field maps (top) and spatial FT (bottom) as a function of the rotation angle in the case of hyperbolicity along the y axis at 311 Hz. c–e) Same as (b) in the case of topological transitions at 430, 452.5, and 462.5 Hz, respectively. is less pronounced now, compared to Figure 2a. Indeed, as the angle increases, the propagation of directional waves smoothly follows the modulation rotation. In reciprocal space, the corre- sponding hyperbolic contours, albeit progressively flatter due to small changes in the couplings induced by moiré effects, undergo a similar rotation, indicating an effective twist of the metasurface properties occurring over a large angular range. Next, we focus on the frequency range associated with hyper- bolicity along the x direction, displayed on Figure 3e–g (for 595, 620, and at 645 Hz, respectively). In the untwisted case (𝜃 = 0◦ ), these frequencies are related to different anisotropic phases: the contour in (e) is an ellipse that progressively opens as the fre- quency is increased to become flat in (f). A further frequency in- crease results in a curvature inversion, leading to a hyperbolic IFC (Figure 3g). As we increase the twist angle, Figure 3e demon- strates an opening of the IFC, and correspondingly a topological transition from elliptical to hyperbolic. The resulting canalized waves follow the rotation of the modulation profile until 𝜃 = 45◦ . In the case of Figure 3f, an overall rotation of the flat contour, as well as its progressive curvature inversion, is observed as a func- tion of the twist angle. Finally, Figure 3g shows a complete topo- logical transition from hyperbolic to elliptical contours, driven by the twist. These findings clearly show that the moiré patterns in- duced by the twist between the interleaved lattices are responsi- ble for topological transitions and canalized waves. The increased modulation wavelength (𝜆 = 4a) results in a better preservation of the untwisted anisotropic coupling distribution. This smoothens the transitions compared to the results of Figure 2b and allows to observe these moiré phenomena over larger angular ranges. These results suggest a straightforward experimental imple- mentation and observation of these phenomena on our reconfig- urable LEGO platform comprising 44 × 44 resonators. We im- plemented several configurations for 𝜃 = 0◦ , 15◦, 30◦, 45◦, as shown in Figure 4a. The sample snapshots illustrate the rotation of the modulation profile, as well as the distortion caused by the sampling as it is twisted relative to the interleaved metasurface lattice (see the pattern formed by the black and blue stripes in the insets). Figure 4b shows the experimentally measured field Adv. Sci. 2022, 9, 2200181 2200181 (6 of 8) © 2022 The Authors. Advanced Science published by Wiley-VCH GmbH www.advancedsciencenews.com www.advancedscience.com profile (top) and corresponding spatial FT (bottom) for hyperbol- icity along y (at 311 Hz) as a function of the twist angle. As 𝜃 in- creases, the wave directionality accordingly rotates, reflected into the corresponding IFCs, which also become flatter and closer to the origin in Fourier space. This behavior, albeit less clear than in Figure 3d because of the smaller size of the sample, follows the trend seen in simulations. Figure 4c–e consider frequencies that support hyperbolic waves along x. Panel (c), for excitation at 430 Hz, starts from delocalized fields for 𝜃 = 0◦ , and the prop- agation becomes strongly canalized as the angle increases, with a directionality following the modulation twist, consistent with Figure 3e. Our measurements confirm a moiré-driven topologi- cal phase transition between closed and open contours. Next, Fig- ure 4d, for excitation at 452.5 Hz, shows a rotation in wave direc- tionality from 𝜃 = 0◦ to 45◦. Although less evident due to the smaller size of the plate, this transition is consistent with the re- sults in Figure 3f. Finally, Figure 4e presents results at 462.5 Hz, at which the waves are canalized and twisted from 𝜃 = 0◦ to 30◦, and then become delocalized at 45◦, experimentally confirming a reverse topological transition, from open to closed contours, as the twist angle increases, similar to Figure 3f. Overall, these ex- perimental results clearly illustrate that tailoring the modulation parameters of twisted interleaved lattices over a metasurface pro- duces topological transitions between delocalized and canaliza- tion regimes, as well as an effective rotation of the guided wave directionality, with the frequency being a key parameter that de- fines the type of observed transition. Overall, these results showcase the rich behavior associated with moiré physics and hyperbolic dispersion within a single- layer metasurface. We note that not all moiré physics associated to bilayer systems can be easily transposed to single layers. No- tably, the interlayer coupling parameter is important for some applications, and it can be challenging to find its equivalent in single-layer systems.[45,46] Moreover, while bilayers may be mod- eled based on the properties of the individual layers,[32,41] addi- tional moiré effects and quasi-periodicity are inevitable in our sys- tem, which makes their modeling more complex. Finally, while inducing dynamical reconfigurability in our single layer moiré system requires a more complex implementation of active de- vices, it does not rely on any physical displacement between the layers, making it more robust and fully controllable. 3. Conclusion In this paper, we have investigated the effect of twisting inter- leaved lattices over a single-layer pillared metasurface. We first explored the case where the two governing spatial features, the position and height modulation profile of the pillars, are aligned but feature a mismatch in their spatial period along one direc- tion. The resulting periodic metasurface supports hyperbolic fea- tures over a broad range of frequencies, whose emergence has been observed both numerically and experimentally over a LEGO platform. Next, we introduced a relative rotation between the in- terleaved lattices, which induces moiré patterns that generates emerging wave phenomena, resulting in drastic modifications of the IFCs that undergo a transition from open to closed contours as the twist angle varies. A coarser sampling of the modulation patterns causes these transitions to be abrupt and to occur over a limited range of twist angles. The effect of sampling is expounded by increasing the modulation wavelength, which results in a bet- ter preservation of the spatial features as the twist angle changes, and produces smoother topological transitions. Such transitions are associated with extreme anisotropic features, inducing wave canalization along specific directions that are controlled by the twist angle within a range of angles and frequencies. In stark contrast to the case of twisted hyperbolic bilayers,[32,41,42,43] these transitions are driven by the moiré patterns emerging within the sample as the rotation angle changes. Moreover, albeit of topo- logical nature, they differ from topological phases in chiral hy- perbolic metamaterials, which are related to pseudo-spin propa- gation at the edge of the system.[62] We have observed these phe- nomena over a simple, practical, and highly reconfigurable LEGO platform, which allowed us to observe with flexibility the various regimes discovered in our study. As such and considering ad- ditional practical implementation challenges, they can be trans- lated over a broad range of physical domains, including quan- tum and nanophotonic systems or microphononics in the con- text of pillared media,[63,64] opening opportunities for twistronic- induced phenomena that do not require multilayered fabrication and careful control alignment, interlayer couplings, and twisting. The reconfigurability of our approach, in contrast with previously investigated twisted bilayer configurations, opens new opportu- nities for single-layer moiré metasurfaces featuring high tunabil- ity of anisotropic responses. Such tunability emerges as a func- tion of the twist angle, which is a single parameter defining the considered modulation. This suggests new opportunities stem- ming from the rich physics of twistronics and moiré phenomena, which may also open the door to dynamically reconfigurable de- vices capable of real-time enhanced wave manipulation. Supporting Information Supporting Information is available from the Wiley Online Library or from the author. Acknowledgements S.Y. and M.I.N.R. contributed equally to this work. S.Y. and A.A. acknowl- edge funding from the Simons Foundation, the National Science Founda- tion EFRI program, and the Air Force Office of Scientific Research MURI program. M. I. N.R. and M.R. gratefully acknowledge the support from the National Science Foundation (NSF) through the EFRI 1741685 grant and from the Army Research Office through grant W911NF-18-1-0036. LEGO is a trademark of the LEGO Group, which does not sponsor, authorize, or endorse this paper. Conflict of Interest The authors declare no conflict of interest. Data Availability Statement The data that support the findings of this study are available from the cor- responding authors upon reasonable request. Keywords hyperbolic, metasurface, moiré materials, phononics, quasi-periodicity, topological transitions, wave steering Adv. Sci. 2022, 9, 2200181 2200181 (7 of 8) © 2022 The Authors. Advanced Science published by Wiley-VCH GmbH www.advancedsciencenews.com www.advancedscience.com Received: January 10, 2022 Revised: February 9, 2022 Published online: March 6, 2022 [30] Y. Zhao, A. N. Askarpour, L. Sun, J. Shi, X. Li, A. Alù, Nat. Commun. 2017, 8, 14180. [31] H. X. Xu, G. Hu, L. Han, M. Jiang, Y. Huang, Y. Li, X. Yang, X. Ling, L. Chen, J. Zhao, C. W. Qiu, Adv. Opt. Mater. 2019, 7, 1801479. [32] G. Hu, A. Krasnok, Y. Mazor, C. W. Qiu, A. Alù, Nano Lett. 2020, 20, 3217. [1] Y. Cao, V. Fatemi, S. Fang, K. Watanabe, T. Taniguchi, E. Kaxiras, P. Jarillo-Herrero, Nature 2018, 556, 43. [2] C. R. Dean, L. Wang, P. Maher, C. Forsythe, F. Ghahari, Y. Gao, J. Ka- toch, M. Ishigami, P. Moon, M. Koshino, T. Taniguchi, Nature 2013, 497, 598. [33] A. Poddubny, I. Iorsh, P. Belov, Y. Kivshar, Nat. Photonics 2013, 7, 948. [34] A. A. High, R. C. Devlin, A. Dibos, M. Polking, D. S. Wild, J. Perczel, [35] N. P. De Leon, M. D. Lukin, H. Park, Nature 2015, 522, 192. J. S. Gomez-Diaz, M. Tymchenko, A. Alù, Phys. Rev. Lett. 2015, 114, 233901. [3] K. Tran, G. Moody, F. Wu, X. Lu, J. Choi, K. Kim, A. Rai, D. A. Sanchez, [36] D. Correas-Serrano, J. S. Gomez-Diaz, A. A. Melcon, A. Alù, J. Opt. J. Quan, A. Singh, J. Embley, Nature 2019, 567, 71. [4] K. L. Seyler, P. Rivera, H. Yu, N. P. Wilson, E. L. Ray, D. G. Mandrus, J. Yan, W. Yao, X. Xu, Nature 2019, 567, 66. [5] C. Jin, E. C. Regan, A. Yan, M. I. B. Utama, D. Wang, S. Zhao, Y. Qin, S. Yang, Z. Zheng, S. Shi, K. Watanabe, Nature 2019, 567, 76. [6] E. M. Alexeev, D. A. Ruiz-Tijerina, M. Danovich, M. J. Hamer, D. J. Terry, P. K. Nayak, S. Ahn, S. Pak, J. Lee, J. I. Sohn, M. R. Molas, Nature 2019, 567, 81. [7] H. Yu, G. B. Liu, J. Tang, X. Xu, W. Yao, Sci. Adv. 2017, 3, e1701696. [8] F. Wu, T. Lovorn, A. H. MacDonald, Phys. Rev. Lett. 2017, 118, 147401. [9] W. Chen, Z. Sun, Z. Wang, L. Gu, X. Xu, S. Wu, C. Gao, Science 2019, 366, 983. 2016, 18, 104006. J. S. Gomez-Diaz, A. Alù, ACS Photonics 2016, 3, 2211. [37] [38] H. N. S. Krishnamoorthy, Z. Jacob, E. Narimanov, I. Kretzschmar, V. M. Menon, Science 2012, 336, 205. I. M. Lifshitz, Sov. Phys. - JETP 1960, 11, 1130. [39] [40] N. P. Armitage, E. J. Mele, A. Vishwanath, Rev. Mod. Phys. 2018, 90, 015001. [41] G. Hu, Q. Ou, G. Si, Y. Wu, J. Wu, Z. Dai, A. Krasnok, Y. Mazor, Q. [42] Zhang, Q. Bao, C. W. Qiu, A. Alù, Nature 2020, 582, 209. J. Duan, N. Capote-Robayna, J. Taboada-Gutiérrez, G. Álvarez-Pérez, I. Prieto, J. Martín-Sánchez, A. Y. Nikitin, P. Alonso-González, Nano Lett. 2020, 20, 5323. [10] L. Jiang, Z. Shi, B. Zeng, S. Wang, J. H. Kang, T. Joshi, C. Jin, L. Ju, J. [43] Z. Zheng, F. Sun, W. Huang, J. Jiang, R. Zhan, Y. Ke, H. Chen, S. Deng, Kim, T. Lyu, Y. R. Shen, Nat. Mater. 2016, 15, 840. Nano Lett. 2020, 20, 5301. [11] S. S. Sunku, G. Ni, B. Y. Jiang, H. Yoo, A. Sternbach, A. S. McLeod, T. Stauber, L. Xiong, T. Taniguchi, K. Watanabe, P. Kim, Science 2018, 362, 1153. [12] S. J. Ahn, P. Moon, T. H. Kim, H. W. Kim, H. C. Shin, E. H. Kim, H. W. Cha, S. J. Kahng, P. Kim, M. Koshino, Y. W. Son, Science 2018, 361, 782. [13] B. Liu, L. Xian, H. Mu, G. Zhao, Z. Liu, A. Rubio, Z. F. Wang, Phys. Rev. Lett. 2021, 126, 066401. [14] P. San-Jose, J. Gonzalez, F. Guinea, Phys. Rev. Lett. 2012, 108, 216802. [15] S. Huang, K. Kim, D. K. Efimkin, T. Lovorn, T. Taniguchi, K. Watanabe, A. H. MacDonald, E. Tutuc, B. J. LeRoy, Phys. Rev. Lett. 2018, 121, 037702. [44] R. Ribeiro-Palau, C. Zhang, K. Watanabe, T. Taniguchi, J. Hone, C. R. Dean, Science 2018, 361, 690. [45] T. Salamon, A. Celi, R. W. Chhajlany, I. Frérot, M. Lewenstein, L. Tar- ruell, D. Rakshit, Phys. Rev. Lett. 2020, 125, 030504. [46] T. Salamon, R. W. Chhajlany, A. Dauphin, M. Lewenstein, D. Rakshit, Phys. Rev B 2020, 102, 235126. [47] M. I. N. Rosa, M. Ruzzene, E. Prodan, Commun. Phys. 2021, 4, 130. [48] P. Celli, B. Yousefzadeh, C. Daraio, S. Gonella, Appl. Phys. Lett. 2019, 114, 091903. [49] M. I. N. Rosa, Y. Guo, M. Ruzzene, Appl. Phys. Lett. 2021, 118, 131901. [50] Y. Achaoui, A. Khelif, S. Benchabane, L. Robert, V. Laude, Phys. Rev. B 2011, 83, 10401. [16] Q. Tong, H. Yu, Q. Zhu, Y. Wang, X. Xu, W. Yao, Nat. Phys. 2017, 13, [51] M. Rupin, F. Lemoult, G. Lerosey, P. Roux, Phys. Rev. Lett. 2014, 112, 356. 234301. [17] S. Carr, D. Massatt, S. Fang, P. Cazeaux, M. Luskin, E. Kaxiras, Phys. Rev. B 2017, 95, 075420. [18] G. Hu, C. W. Qiu, A. Alù, Opt. Mater. Express 2021, 11, 1377. [19] B. Lou, N. Zhao, M. Minkov, C. Guo, M. Orenstein, S. Fan, Phys. Rev. Lett. 2021, 126, 136101. [52] L. Cao, Z, Y., Y. Xu, B. Assouar, Smart Mater. Struct. 2018, 27, 075051. [53] W. Wang, B. Bonello, B. Djafari-Rouhani, Y. Pennec, J. Zhao, Phys. Rev. Appl. 2018, 10, 064011. [54] O. R. Bilal, A. Foehr, C. Daraio, Extreme Mech. Lett. 2017, 15, 103. [55] A. Colombi, P. Roux, S. Guenneau, M. Rupin, J. Acoust. Soc. Am. 2015, [20] M. R. López, F. Peñaranda, J. Christensen, P. San-Jose, Phys. Rev. Lett. 137, 1783. 2020, 125, 214301. [21] Y. Deng, M. Oudich, N. J. Gerard, J. Ji, M. Lu, Y. Jing, Phys. Rev. B 2020, 102, 180304. [22] S. M. Gardezi, H. Pirie, S. Carr, W. Dorrell, J. E. Hoffman, 2D Mater. 2021, 8, 031002. [23] K. Dong, T. Zhang, J. Li, Q. Wang, F. Yang, Y. Rho, D. Wang, C. P. Grigoropoulos, J. Wu, J. Yao, Phys. Rev. Lett. 2021, 126, 223601. [24] P. Wang, Y. Zheng, X. Chen, C. Huang, Y. V. Kartashov, L. Torner, V. V. [56] P. Roux, D. Bindi, T. Boxberger, A. Colombi, F. Cotton, I. Douste- Bacque, S. Garambois, P. Gueguen, G. Hillers, D. Hollis, T. Lecocq, Seismol. Res. Lett. 2018, 89, 582. J. H. Oh, H. M. Seung, Y. Y. Kim, Appl. Phys. Lett. 2014, 104, 073503. J. H. Oh, Y. K. Ahn, Y. Y. Kim, Struct. Multidiscip. Optim. 2015, 52, 1023. [57] [58] [59] H. Lee, J. H. Oh, H. M. Seung, S. H. Cho, Y. Y. Kim, Sci. Rep. 2016, 6, 24026. Konotop, F. Ye, Nature 2020, 577, 42. [60] R. Zhu, Y. Y. Chen, Y. S. Wang, G. K. Hu, G. L. Huang, J. Acoust. Soc. [25] M. Martí-Sabaté, D. Torrent, Phys. Rev. Appl. 2021, 15, L011001. [26] Q. Fu, P. Wang, C. Huang, Y. V. Kartashov, L. Torner, V. V. Konotop, F. Ye, Nat. Photonics 2020, 14, 663. Am. 2016, 139, 3303. [61] H. W. Dong, S. D. Zhao, Y. S. Wang, C. Zhang, Sci. Rep. 2018, 8, 2247. [62] W. Gao, M. Lawrence, B. Yang, F. Liu, F. Fang, B. Béri, J. Li, S. Zhang, [27] Y. Jin, W. Wang, Z. Wen, D. Torrent, B. Djafari-Rouhani, Extreme Mech. Phys. Rev. Lett. 2015, 114, 037402. Lett. 2020, 39, 100777. [63] M. Yan, J. Lu, F. Li, W. Deng, X. Huang, J. Ma, Z. Liu, Nat. Mater. 2018, [28] M. I. Cohen, C. Jörg, Y. Lumer, Y. Plotnik, E. H. Waller, J. Schulz, G. 17, 993. von Freymann, M. Segev, Light: Sci. Appl. 2020, 9, 200. [29] Y. Zhao, M. Belkin, A. Alù, Nat. Commun. 2012, 3, 870. [64] M. Yan, W. Deng, X. Huang, Y. Wu, Y. Yang, J. Lu, F. Li, Z. Liu, Phys. Rev. Lett. 2021, 127, 136401. Adv. Sci. 2022, 9, 2200181 2200181 (8 of 8) © 2022 The Authors. Advanced Science published by Wiley-VCH GmbH
null
10.1371_journal.pone.0257328.pdf
Data Availability Statement: LANYERO, HINDUM (2021), Validity of caregivers’ reports on prior use of antibacterials in children under five years presenting to health facilities in Gulu, northern Uganda, Dryad, Dataset, https://doi.org/10.5061/ dryad.sj3tx9642.
null
RESEARCH ARTICLE Validity of caregivers’ reports on prior use of antibacterials in children under five years presenting to health facilities in Gulu, northern Uganda Hindum LanyeroID Katureebe Agaba4, Joan N. Kalyango5,6, Jaran Eriksen3,7, Sarah Nanzigu1* 1, Moses Ocan1, Celestino Obua2, Cecilia Stålsby Lundborg3, 1 Department of Pharmacology and Therapeutics, Makerere University College of Health Sciences, Kampala, Uganda, 2 Mbarara University of Science and Technology, Mbarara, Uganda, 3 Department of Global Public Health, Karolinska Institutet, Stockholm, Sweden, 4 Infectious Diseases Research Collaboration, Kampala, Uganda, 5 Department of Pharmacy, Makerere University College of Health Sciences, Kampala, Uganda, 6 Clinical Epidemiology Unit, Makerere University College of Health Sciences, Kampala, Uganda, 7 Department of Infectious Diseases, South General Hospital, Stockholm, Sweden * snanzigu@yahoo.com Abstract Introduction Given the frequent initiation of antibacterial treatment at home by caregivers of children under five years in low-income countries, there is a need to find out whether caregivers’ reports of prior antibacterial intake by their children before being brought to the healthcare facility are accurate. The aim of this study was to describe and validate caregivers’ reported use of antibacterials by their children prior to seeking care at the healthcare facility. Methods A cross sectional study was conducted among children under five years seeking care at healthcare facilities in Gulu district, northern Uganda. Using a researcher administered questionnaire, data were obtained from caregivers regarding reported prior antibacterial intake in their children. These reports were validated by comparing them to common anti- bacterial agents detected in blood and urine samples from the children using liquid chroma- tography with tandem mass spectrometry (LC-MS/MS) methods. Results A total of 355 study participants had a complete set of data on prior antibacterial use col- lected using both self-report and LC-MS/MS. Of the caregivers, 14.4% (51/355, CI: 10.9– 18.5%) reported giving children antibacterials prior to visiting the healthcare facility. How- ever, LC-MS/MS detected antibacterials in blood and urine samples in 63.7% (226/355, CI: 58.4–68.7%) of the children. The most common antibacterials detected from the laboratory analysis were cotrimoxazole (29%, 103/355), ciprofloxacin (13%, 46/355), and metronida- zole (9.9%, 35/355). The sensitivity, specificity, positive predictive value (PPV), negative predictive value and agreement of self-reported antibacterial intake prior to healthcare a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Lanyero H, Ocan M, Obua C, Stålsby Lundborg C, Agaba K, Kalyango JN, et al. (2021) Validity of caregivers’ reports on prior use of antibacterials in children under five years presenting to health facilities in Gulu, northern Uganda. PLoS ONE 16(9): e0257328. https://doi. org/10.1371/journal.pone.0257328 Editor: Orvalho Augusto, University of Washington, UNITED STATES Received: February 12, 2021 Accepted: August 28, 2021 Published: September 16, 2021 Copyright: © 2021 Lanyero et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: LANYERO, HINDUM (2021), Validity of caregivers’ reports on prior use of antibacterials in children under five years presenting to health facilities in Gulu, northern Uganda, Dryad, Dataset, https://doi.org/10.5061/ dryad.sj3tx9642. Funding: Makerere University -SIDA collaboration (Sida PI0010) The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. PLOS ONE | https://doi.org/10.1371/journal.pone.0257328 September 16, 2021 1 / 14 PLOS ONE Validity of caregivers’ reports on prior use of antibacterials in children under five years Competing interests: The authors have declared that no competing interests exist. facility visit were 17.3% (12.6–22.8), 90.7% (84.3–95.1), 76.5% (62.5–87.2), 38.5% (33.0– 44.2) and 43.9% (k 0.06) respectively. Conclusion There is low validity of caregivers’ reports on prior intake of antibacterials by these children. There is need for further research to understand the factors associated with under reporting of prior antibacterial use. Introduction Antibacterial agents are used to treat a wide range of bacterial infections and are essential life- saving medicines. They are the most commonly used medicines in Sub-Saharan Africa due to the high prevalence of infectious diseases [1]. Used correctly, they deliver enormous benefits to the health of the population worldwide [2]. Antibacterials are, according to the national drug policy of Uganda, prescription only medi- cines [3]. However, they are readily accessible and affordable to most patients within the com- munities in Uganda, not only as prescription medicines as they can often also be obtained over-the-counter especially in private medicine outlets [4]. The relative ease with which com- munities access these medicines poses several challenges for antibacterial stewardship [4]. The majority of caregivers in low-income countries initiate treatment of their children at home [5]. The use of antibacterials prior to hospital visits is common, especially in low-income countries, and may influence patient treatment outcomes. According to a study in Nigeria, 85% of patients reported to have self-medicated before coming to the health facility and antibacterials were among the most common medicines used [6]. A study in Uganda reported that 62.2% of patients had used antibacterial agents prior to coming to health facility [4]. Another study done in Haiti to assess self-medication among patients presenting at an out-patient depart- ment found that 45.5% practiced self-medication with antibacterials [7]. Caregivers’ ability to report antibacterial intake prior to coming to a health facility is crucial for appropriate prescription of medicines at the health facility. Self-reports have been shown to have low validity as they are prone to recall bias and social desirability bias. Respondents nor- mally provide information that conforms to their perceived expectations of the health workers or researchers [5, 8]. A study carried out in Uganda in 2009 reported a limited validity of care- givers’ reports of use of sulfamethoxazole, chloroquine and sulfadoxine in their children prior to arrival to the hospital [5]. Similarly, a study from Tanzania reported that 97% of the children without history of prior chloroquine treatment had detectable levels of chloroquine in blood [9]. Another study in Ghana reported a high prevalence (64%) of antibacterials detected in urine samples of patients compared to the self-reported use (13%) [10]. To our knowledge no study has validated caregivers’ reports of intake of antibacterials in children under five years in rural communities in low resource settings. In this study we describe and validate caregivers’ reported use of antibacterials by their children under five years for treatment prior to seeking care at the healthcare facility. Materials and methods Ethics statement The protocol was reviewed and approved by the Makerere University School of Biomedical Sciences Research and Ethics Committee (reference SBS-570) and the Uganda National PLOS ONE | https://doi.org/10.1371/journal.pone.0257328 September 16, 2021 2 / 14 PLOS ONE Validity of caregivers’ reports on prior use of antibacterials in children under five years Council of Science and Technology (reference HS235ES) (S1 Appendix). Administrative clear- ance was obtained from the healthcare facilities where the study was conducted. Written informed consent was obtained from caregivers of children under five years prior to data col- lection (S2 Appendix). Study design and setting A cross-sectional study was conducted among children under five years and their caregivers in healthcare facilities in Gulu district, northern Uganda. Gulu is located about 360 km from the capital city Kampala. In Uganda, the lowest level of the district-based healthcare system con- sists of the village health teams/community medicine distributors, which constitute level 1 of health care. This is operated by members of the community who can read and write at least in the local language of the community. The next level is health centre II which is operated by a professionally trained nurse with a diploma and is intended to serve 5000 patients. This is fol- lowed by health centre level III which is operated by a professionally trained clinical officer with a diploma in clinical medicine and intended to serve 10,000 patients. Above health centre level III is health centre level IV and then district hospitals headed by medical officers with a basic degree in medicine and surgery and intended to serve about 100,000 patients. Regionally there are regional referral hospitals where patients are referred to from the district hospitals. The regional referral hospitals are expected to have specialist health professionals covering the major disciplines such as surgery, internal medicine, and paediatrics. At the top of the health care system are the national referral hospitals [11]. Gulu district has a total of 19 health centre level II, 10 health centre level III, one health centre level IV, 31 registered pharmacies and 135 licensed drug shops [12–14]. This study was carried out in three health centre level III and one health centre level IV. These healthcare centers were purposively selected because they serve the greatest number of patients in the out-patient departments in Gulu district. The most com- mon diseases in children under five years seeking care at healthcare facilities in this area include; malaria, diarrhea, pneumonia, acute childhood malnutrition and HIV/AIDS [15–17]. Study population Sick children under five years and their caregivers seeking care at the four healthcare facilities were included in the study after caregivers’ consent. Children who were brought to the health center by caregivers who did not take care of the children from the onset of the current illness were excluded from the study. Children who had come for review or continuation of treatment for current illness were also excluded from the study. Sample size The sample size was computed based on formula for estimation of sample size for a single pro- portion [18]. Assuming that the proportion of children getting antibacterial treatment prior to health facility visit was 50%, in order to have a 95% confidence interval and a 5% margin of error, the minimum sample size needed was set to 385. The number of children sampled from each facility was determined from the volume of patients at the health facility using propor- tionate sampling. Sampling procedures The patients were selected by systematic random sampling. On each of the data collection dates the first patient to be recruited into the study was randomly selected by having a blind- folded data collector walk around in the waiting area and point at a random patient among the PLOS ONE | https://doi.org/10.1371/journal.pone.0257328 September 16, 2021 3 / 14 PLOS ONE Validity of caregivers’ reports on prior use of antibacterials in children under five years patients waiting in line to be seen by the healthcare worker in the outpatient department. Thereafter, every fourth patient in line towards the entrance of the healthcare workers room was selected for recruitment. In the event that the selected patient was above five years of age, they were skipped and the next patient recruited while maintaining the sampling interval. Approximately 10 days were spent collecting data in each healthcare facility. Data collection An interviewer administered questionnaire was used for data collection. The questionnaire was pre-tested on caregivers of 30 children in outpatient departments of Gulu regional referral hospital. This tool was adapted from a tool used to collect data on prevalence and predictors of prior antibacterial use among patients presenting to hospitals in northern Uganda in a previ- ous study [4], it was written in English and translated to Acholi (the most common local lan- guage spoken in the study area). The data collection team was divided into four groups each comprising of two people, one pharmacy technician (health professional with diploma in pharmacy) and a laboratory techni- cian. The pharmacy technician conducted interviews while the laboratory technician collected the blood and urine samples. Information on the following variables was collected; sub-county of residence, age of child, age of care-giver, sex of child, sex of caregiver, whether medication was given to child before coming to the healthcare facility since the onset of this current illness, the type and source of the medicine, and the person who recommended the medicine. In case the caregiver did not know the name of the medicine, the interviewer asked them to describe it or show the packing material if at all they had come with it to the health center. Each interview lasted about 20 min- utes per patient. Sample collection and transportation. Two hundred microlitres (200μL) of blood was collected from the fingertips of children under five years using a 200μL micro-pipette with ethylenediamine tetra-acetic acid (EDTA), and spotted on a filter paper and left to dry for 3 hours in room temperature. After the blood had dried on the filter paper, each filter paper was put in a separate plastic zip bag with a desiccant and transported to the laboratory for analysis. Urine samples were collected in sterile wide mouth containers. In the very young children who couldn’t void in the wide mouth containers, urine samples were collected by placing a thick layer of cotton wool inside the child’s nappy and squeezing the urine in the urine sample bottles. Two hundred microlitres (200μL) of urine was collected from the wide mouth contain- ers using a plastic pipette and spotted on a filter paper and left to dry for 3 hours at room tem- perature. After the urine had dried on the filter paper, each filter paper was put in a separate plastic zip bag with a desiccant and transported to the laboratory for analysis. The dried blood spot (DBS) and dried urine spot (DUS) samples obtained from patients were stored at -20˚C and -80˚C respectively until analysis. Extraction and analysis of antibacterials in dry blood spot and dry urine spot sam- ples. The whole diameter disk (containing 200μl of blood or urine) was cut out from each DBS and DUS. The cut disc was placed in an Eppendorf tube (1.5 mL capacity) and mixed with 1000 μL of methanol (20%) and acetonitrile (80%). The sample was vortex-mixed twice for 20 s at 10-min intervals and then centrifuged at 3500 revolutions per minute (RPM) for 5 minutes. After the extraction period, the filter paper was removed, and 500 μL of the extract was transferred into an auto-sampler vial to be injected onto the LC-MS/MS system for analysis. A simple, fast, sensitive and selective qualitative LC-MS/MS method for identification of fifteen (15) antibacterials in DBS and DUS was used for analysis (S3 Appendix). The limit PLOS ONE | https://doi.org/10.1371/journal.pone.0257328 September 16, 2021 4 / 14 PLOS ONE Validity of caregivers’ reports on prior use of antibacterials in children under five years of detection for the different antibacterials were: amoxicillin (1.34 ng/mL), ampicillin (0.001 ng/mL), penicillin G (0.005 ng/mL), penicillin V (0.03 ng/mL), cloxacillin (0.2 ng/mL), cephalexin (0.22 ng/mL), sulfamethoxazole (0.95 ng/mL), trimethoprim (0.52 ng/mL), eryth- romycin (1.1 ng/mL), ciprofloxacin (0.1 ng/mL), tetracycline (0.14 ng/mL), clarithromycin (1.4 ng/mL), metronidazole (0.0004 ng/mL), chloramphenicol (0.0001ng/mL) and azithromy- cin (0.22 ng/mL). Data on key pharmacokinetics properties that may have affected the interpretation of our results, have been presented in the supporting information section (S1 Table), and these include: clearance, terminal half-life, percentage of medicine excreted in urine, time to peak plasma concentrations and volume of distribution. Data management Double data entry was done using Epi-Data 3.1 software for both the questionnaire and labora- tory data. The two datasets were reconciled by comparing them for each field in the question- naire and laboratory result, in case of any discrepancies, the corresponding questionnaire or patient laboratory record was checked to establish the correct entry. Data were then imported into Stata 14/IC (Stata Inc., Texas USA) for analysis. Statistical analysis Descriptive statistics were presented using median and interquartile range (IQR) for continu- ous variables or frequencies and proportions for categorical variables. The dependent vari- ables, treatment of child with antibacterials prior to healthcare facility visit as reported by their caregiver and detectable antibacterials in DBS or DUS samples, were summarized as propor- tions. In order to adjust for potential biases associated with point estimates from the sampling design, we used svy commands in stata to compute proportions and respective 95% confidence intervals. Pearson’s chi-square test was used to assess associations for the categorical variables. In order to validate caregivers’ reported use of antibacterials, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), prevalence, agreement and kappa coefficient were calculated. Laboratory results for detection of antibacterials in dry blood spot or dry urine spot samples were considered as the gold standard and caregivers’ reports of use of antibacterials prior to health facility visit were considered as the test results. Results Socio-demographic characteristics of the caregivers and children under five years Of the 385 sampled children, 355 (92.2%) had data on both caregiver’s report on antibacterial use prior to health facility visit and results from urine and blood analysis and were thus included in the analysis. The 30 (7.8%) observations were dropped because they were missing blood analysis data. Over half (53.2%, n = 189) of the children were female. The median age of the children was 29 (IQR: 16–46) months. The majority (96.1%, n = 341) of the caregivers were female. The median age of the caregivers was 25 (IQR: 21–31) years. About half (53.2%, n = 189) of the children attended a healthcare facility located in a rural area. (Table 1). Prevalence of antibacterial use prior to coming to the health facility as reported by caregivers of children under five years Out of the 355 children under five years who were included in the analysis, 51 (14.4%, CI: 10.9–18.5) were reported by the caregivers to have been treated with antibacterials prior to PLOS ONE | https://doi.org/10.1371/journal.pone.0257328 September 16, 2021 5 / 14 PLOS ONE Validity of caregivers’ reports on prior use of antibacterials in children under five years Table 1. Socio-demographic characteristics and prevalence of antibacterial use in children under five years prior to health facility visit as reported by caregivers of children under five years in rural communities of Gulu district, northern Uganda (August, 2019). Characteristics Description Respondent’s Frequency (%) Proportion of reported antibacterial use, n (%) 95% CI P-value (Pearson’s chi-square test) Overall Sex of child Location of health facility Age of child (months) Age of child caregiver (years) Sex of child caregiver Source of antibacterials 355 (100) 166 (46.8) 189 (53.2) 166 (46.8) 189 (53.2) 64 (18.0) 176 (49.6) 115 (32.4) 121 (34.1) 168 (47.3) 51 (14.4) 7 (2.0) 8 (2.2) 14 (3.9) 341 (96.1) Male Female Urban Rural 1–12 13–36 37–59 13–22 23–32 33–42 43–52 � 53 Male Female Home cabinet Public health facility Private clinics Drug shops Retail shops Traditional healers Antibacterials recommended by Caregiver Other household member Friend/neighbor Doctor/nurse Drug seller/pharmacist Traditional healer Age of child (months), median (IQR) Age of child caregiver (years), median (IQR) 29 (16.46) 25 (21.31) n: Sample size; CI: Confidence Interval; %: Percentage; IQR: Interquartile range https://doi.org/10.1371/journal.pone.0257328.t001 51 (14.4) 22 (13.3) 29 (15.3) 36 (21.7) 15 (7.9) 6 (9.4) 32 (18.2) 13 (11.3) 15 (12.4) 27 (16.1) 7 (13.7) 1 (14.3) 1 (12.5) 1 (7.1) 50 (14.7) 11 (23.9) 9 (37.5) 8 (30.8) 18 (35.3) 4 (30.8) 1 (50) 11 (31.4) 3 (37.5) 1 (25.0) 17 (34.0) 18 (29.0) 1 (33.3) 0.575 <0.001 0.119 0.936 0.432 <0.001 10.9–18.5 8.9–19.4 10.9–21.3 16.0–28.6 4.8–12.8 4.2–19.5 13.1–24.6 6.7–18.6 7.6–19.6 11.2–22.5 6.6–26.3 1.7–62.3 1.5–57.5 0.9–39.0 11.3–18.9 12.5–38.8 18.8–59.4 14.3–51.8 22.4–49.9 9.1–61.4 1.3–98.7 16.9–49.3 0.253 8.5–75.5 0.6–80.6 21.2–48.8 18.2–41.9 0.8–90.6 coming to the healthcare facility. Of these 51 children, the prevalence of antibacterial use was higher in those from urban areas (21.7%, CI: 16.0–28.6) and in those who got antibacterials from public health facilities (37.5%, CI: 18.8–59.4) (Table 1). Prevalence of antibacterials detected in blood and urine samples of children under five years Of the 355 children under five years who were included in the analysis, 226 (63.7%, CI: 58.4– 68.7) had detectable levels of antibacterials in urine or blood in the samples taken upon arrival to the healthcare facility (Table 2). Most commonly used antibacterials. The most commonly used antibacterials as reported by the care givers were amoxicillin (6.2%, 22/355), cotrimoxazole (2.8%, 10/355), and metroni- dazole (2.3%, 8/355). The most common antibacterials detected from the laboratory analysis were cotrimoxazole (29%, 103/355), ciprofloxacin (13%, 46/355), and metronidazole (9.9%, 35/355) (Fig 1) PLOS ONE | https://doi.org/10.1371/journal.pone.0257328 September 16, 2021 6 / 14 PLOS ONE Validity of caregivers’ reports on prior use of antibacterials in children under five years Table 2. Prevalence of antibacterials detected in blood or urine samples of children under five years in rural communities of Gulu district, northern Uganda (August, 2019). Characteristics Description Respondent’s Frequency (%) Proportion of antibacterial detected, n (%) Overall Sex of child Location of health facility Age of child (months) Age of child caregiver (years) Sex of child caregiver Male Female Urban Rural 1–12 13–36 37–59 13–22 23–32 33–42 43–52 � 53 Male 355 (100) 166 (46.8) 189 (53.2) 166 (46.8) 189 (53.2) 64 (18.0) 176 (49.6) 115 (32.4) 121 (34.1) 168 (47.3) 51 (14.4) 7 (2.0) 8 (2.2) 14 (3.9) 226 (63.7) 108 (65.1) 118 (62.4) 103 (62.0) 123 (65.1) 45 (70.3) 112 (63.6) 69 (60.0) 83 (68.6) 99 (58.9) 36 (70.6) 5 (71.4) 3 (37.5) 8 (57.1) Female 341 (96.1) 218 (63.9) n: Sample size; CI: Confidence Interval; %: Percentage https://doi.org/10.1371/journal.pone.0257328.t002 95% CI 58.4–68.7 57.5–71.9 55.3–69.1 54.4–69.1 57.9–71.6 57.9–80.3 56.2–70.4 50.7–68.6 59.7–76.3 51.3–66.2 56.6–81.5 29.7–93.7 11.4–73.6 30.7–80.1 58.7–68.9 P-value (Pearson’s chi-square test) 0.608 0.554 0.389 0.164 0.605 Validity of caregivers’ reports of antibacterial intake in children under five years The sensitivity, specificity, PPV, NPV, agreement and kappa coefficient of the caregivers’ reports of use of antibacterials for treatment of children prior to healthcare facilities visit were 17.3% (12.6–22.8), 90.7% (84.3–95.1), 76.5% (62.5–87.2), 38.5% (33.0–44.2), 43.9% (38.7– 49.3%) and 0.06 (0.01–0.12) respectively. The sensitivity, specificity, PPV,NPV, agreement and kappa coefficient varied between the different antibacterials (see Table 3). Discussion In this study we demonstrated that the prevalence of antibacterial use prior to health facility visit was high and that caregivers under reported the use of antibacterials in the children under five years prior to coming to the health facility. Antibacterial use prior to healthcare facility visit is a common practice in many resource limited settings globally. Caregivers’ ability to report antibacterial use before coming to the health facility is crucial for appropriate prescrip- tion of antibacterial upon reaching health facilities [5]. Appropriate prescription of antibacter- ials is important because it reduces the emergence of antibacterial resistance, poor clinical outcomes, increased mortality and wastage of financial resources [19]. In the current study, almost two thirds (63.7%) of the samples (blood and/or urine) tested positive for antibacterials. This implies that the prevalence of antibacterial use prior to health facility visit is much higher than what was self-reported (14.4%). This finding is similar to those from other low and middle income countries (LMIC) [4, 5, 10], a study carried out in Ghana reported a prevalence of self-reported antibacterial use prior to health facility visit of 13%, however, analysis of urine samples reported a much higher prevalence of 64% [10]. In Uganda, self-medication with antibacterials is a common practice [1, 4, 20] which is reflected in the high prevalence of antibacterials found in the samples (blood and/or urine) in the cur- rent study [1]. Another reason for the high prevalence of antibacterial use in our study is the PLOS ONE | https://doi.org/10.1371/journal.pone.0257328 September 16, 2021 7 / 14 PLOS ONE Validity of caregivers’ reports on prior use of antibacterials in children under five years Fig 1. Commonly used antibacterials according to the laboratory analysis. https://doi.org/10.1371/journal.pone.0257328.g001 high prevalence of infectious diseases in these communities. In Uganda, 71% of children under five years attending healthcare facilities do so due to acute respiratory infections [21], however, in the community in this study, the most common diseases in children under five years seeking care at healthcare facilities include; malaria, diarrhea, pneumonia, acute child- hood malnutrition and HIV/AIDS [15–17]. High prevalence of antibacterials found in the samples in the current study could also be due to exposure to antibacterials through Table 3. Validity of caregivers’ reports of antibacterial intake in children under five years in rural communities of Gulu district, northern Uganda (August, 2019). Parameters Sensitivity (95% CI) Specificity (95% CI) PPV (95% CI) NPV (95% CI) Prevalence (95% CI) Agreement (95% CI) κ (95% CI) Overall 17.3 (12.6–22.8) 90.7 (84.3–95.1) 76.5 (62.5–87.2) 38.5 (33.0–44.2) 63.7 (58.4–68.7) 43.9 (38.7–49.3) 0.06 (0.01–0.12) Amoxicillin 5.6 (0.1–27.3) 93.8 (90.6–96.1) 4.5 (0.1–22.8) 94.9 (92.0–97.0) 5.1 (3.0–7.9) 89.3 (85.6–92.3) Cotrimoxazole 5.8 (2.3–12.2) 98.4 (96.0–99.6) 60.0 (26.3–87.8) 71.9 (66.8–76.6) 29.0 (24.3–34.0) 71.5 (66.5–76.2) -0.01 (-0.11–0.09) 0.06 (-0.01–0.12) Metronidazole 2.9 (0.1–14.9) 97.8 (95.5–99.1) 12.5 (0.3–52.7) 90.2 (86.6–93.1) 9.9 (7.0–13.4) 88.5 (84.7–91.6) 0.01 (-0.08–0.1) Ciprofloxacin 4.3 (0.5–14.8) 99.7 (98.2–100) 66.7 (9.4–99.2) 87.5 (83.6–90.8) 13.0 (9.6–16.9) 87.3 (83.4–90.6) 0.07 (-0.03–0.16) %: percentage; CI: Confidence interval; κ: Kappa coefficient https://doi.org/10.1371/journal.pone.0257328.t003 PLOS ONE | https://doi.org/10.1371/journal.pone.0257328 September 16, 2021 8 / 14 PLOS ONE Validity of caregivers’ reports on prior use of antibacterials in children under five years consumption of water, vegetables and animal products [22, 23]. In the Hong Kong survey to determine the presence of veterinary antibiotics in food, drinking water, and the urine of pre- school children, it was found that 13 veterinary antibiotics were detectable in the urine of 77.4% of primary school children with norfloxacin and penicillin having the highest detection rates. Enrofloxacin, penicillin, and erythromycin were the most detected veterinary antibiotics in raw and cooked food [21]. Studies in Uganda, report a high prevalence of veterinary use of antibacterials. The most commonly used antibacterials in veterinary medicine in Uganda include; procaine penicillin, trimethoprim/sulfadiazine, erythromycin sulphate, tylosin tar- trate, oxytetracycline hydrochloride [24, 25]. This high prevalence of antibacterial use can lead to increased risk of resistance within the community [26]. A study was carried out in Uganda to determine the epidemiology and antibiotic susceptibility of Vibrio cholerae associated with the 2017 outbreak in Kasese district, and it reported that V. cholerae was highly resistant to the commonly used antibiotics [27]. Most caregivers reported to have given their children amoxicillin, cotrimoxazole and met- ronidazole. This is consistent with reports from a study in northern Uganda where metronida- zole, amoxicillin, ciprofloxacin, doxycycline or cotrimoxazole were reported as the most commonly used antibacterials by patients prior to hospital visit [4]. Metronidazole is com- monly used for bacterial gastroenteritis, amoxicillin is used for bacterial chest infections, and cotrimoxazole is used to treat pneumonia, bronchitis, infections of the urinary tract, ears intes- tines and as prophylaxis against opportunistic infections in HIV [28, 29]. In our study the most commonly detected antibacterials in the laboratory analysis results were cotrimoxazole, ciprofloxacin and metronidazole, similar to findings from a study carried out in Ghana which reported ciprofloxacin, trimethoprim or metronidazole as the most common antibacterials detected in urine samples [30]. Ciprofloxacin is commonly used to treat pneumonia, typhoid fever, infectious diarrhea, skin and bone infections [28, 29]. Amoxicillin was the most com- monly reported antibacterial used and yet it was not among the most commonly detected anti- bacterials from the laboratory analysis. This could be explained by the pharmacokinetics of amoxicillin, which has a very short half-life of about 1 hour and will usually be out of the sys- tem within 5 hours. Thus, meaning that for it to be detected in the blood or urine samples, it should have been taken within a few hours before healthcare facility visit [28, 29]. We also observed that the number of children who had cotrimoxazole in their biological samples was higher than those who reported the use. It is possible that some of these children may have tested positive for cotrimoxazole since they could have been receiving it as prophylaxis against opportunistic infections in HIV [31, 32]. The prevalence of HIV/AIDS in northern Uganda as of 2019 when data for this study was collected, was 7.2% in adults and 0.5% in children under five years [33]. Since we were interested in antibacterial use for current illness, for which the children were brought to the healthcare facility, caregivers might not have found it not neces- sary to report the use of cotrimoxazole as prophylaxis against opportunistic infections in HIV. The positive predicative value we found for reported use of antibacterials is not high enough to allow caregivers reports to guide treatment. The high specificity values indicate under reporting but the negative predictive value indicate that many children were given drugs that were not reported by caregivers. This study was carried out in rural communities of Gulu dis- trict in Uganda where the adult literacy levels are low [1, 20], and the inconsistencies in care- givers’ response to interview questions and laboratory findings, could have been because of caregivers inability to identify medicines taken as antibacterials. Another reason for the incon- sistencies in self-reported antibacterial use and laboratory findings could have been due to social desirability bias [34]. The caregivers could have been aware that self-medication is not a good practice, and therefore feared to tell the interviewers the truth. Another reason for the inconsistencies could have been due to consumption of these medicines from diffuse sources PLOS ONE | https://doi.org/10.1371/journal.pone.0257328 September 16, 2021 9 / 14 PLOS ONE Validity of caregivers’ reports on prior use of antibacterials in children under five years such as milk, water or food, studies in Uganda have reported veterinary use of antibacterials [24, 25]. Another worrying explanation for the inconsistencies could be the quality of antibac- terial medicines, some of these antibacterials may not contain the actual quantity of the active medicine the manufacturers claim they contain. Although we did not set out to study the qual- ity of antibacterials in this study, high prevalence of substandard antibacterial medicines has been previously reported in developing countries [35]. Furthermore, inaccuracies in self- reports may lead to duplication of therapy, incorrect management of the ill child, failure to appreciate non-compliance leading to exacerbation of chronic medical conditions, or inaccu- rate research conclusions [36]. We observed a strong association between high self-reported prior antibacterial use and the source of antibacterials being from public health facilities. This could be attributed to the low financial status of the people in these communities [1] forcing them to seek free healthcare from public healthcare facilities. The district-based healthcare system in Uganda consists of level I, II, III, IV and district hospitals [11]. This therefore means that by the time these chil- dren were brought to health care level III and IV, they could have already sought care from the lower levels and were referred to these higher levels for further management. There is need for further research to understand the reasons for caregivers’ poor reports on their children’s prior intake of antibacterials before coming to the health facility. Improved validity could be promoted by encouraging health care workers to carefully explain to the care- givers the medicines they administer to these children when they fall sick. Proper documenta- tion of the medicines given to these children when they are sick could also improve the validity of self-reported medicine use. There is need for the healthcare workers to educate the caregiv- ers about the dangers of using antibacterials without consulting a healthcare worker, and also further research is required to better understand why caregivers initiate antibacterial use at home without consulting a healthcare service provider. This all will allow policy makers to be better informed when planning interventions to reduce the large amount of incorrect antibac- terial use in the community. The results of our study should be considered in light of some limitations. This study could have been affected by recall bias, where antibacterials given may have been forgotten. The study could have also been affected by social desirability bias since the study was carried out in a hospital setting and probably caregivers feared telling the truth because they thought it could affect patient care. Under reporting could have been affected by how the questions were understood by the caregivers. Failure to detect some of the antibacterials in the samples could have been due to the pharmacokinetics of the antibacterials. Factors such as education level of the caregivers could have contributed to the under reporting of antibacterial use prior to healthcare facility visit, however, we didn’t collect this information. This is because adult liter- acy levels in this community are low [1, 20] and to our knowledge previous studies have not reported associations between self-report and education level [10, 37]. However, there is need for further research to determine if there’s an association between caregivers’ education level and reporting of prior antibacterial use in this setting. The discrepancy between the reported use and the detected antibacterials in blood/urine samples could have also been because the antibacterial could have been given for the management of another condition, such as cotri- moxazole for prophylaxis against HIV related opportunistic infections, but we did not collect this information. Our tool was designed to capture only antibacterial use for the illness for which the children were brought to the healthcare facility. We were unable to report the levels of antibacterials in relation to how far back the antibacterials were taken, this is because we used a qualitative LC-MS/MS method which was developed to report the presence or absence of antibacterials and not to quantify them. PLOS ONE | https://doi.org/10.1371/journal.pone.0257328 September 16, 2021 10 / 14 PLOS ONE Validity of caregivers’ reports on prior use of antibacterials in children under five years Conclusion A high proportion of children under five years take antibacterials prior to visiting a healthcare facility in northern Uganda. However, there is low validity of caregivers’ reports on prior intake of antibacterials by these children. There is need for further research to understand the factors associated with under reporting of prior antibacterial medicine use by caregivers of children under five years. In addition, we suggest that health care workers should endeavor to explain the role and names of medicines during dispensing, as well as the importance of reporting correctly on prior medication intake. There is also need to educate the caregivers about the dangers of using antibacterials without consulting a healthcare worker, and also fur- ther research is required to better understand why caregivers initiate antibacterial use at home without consulting a healthcare service provider. This all will allow policy makers to be better informed when planning interventions to reduce the large amount of incorrect antibacterial use in the community. Supporting information S1 Table. Summary of key pharmacokinetic properties of some of the antibacterials that are commonly used among children under five years in rural communities of Gulu district, northern Uganda (August, 2019). (DOCX) S1 Appendix. Ethical approval letters. (DOCX) S2 Appendix. Consent form. (DOCX) S3 Appendix. LC-MS/MS method. (DOCX) S4 Appendix. Questionnaire. (DOCX) Acknowledgments We appreciate the tireless effort of the data collection team; Apio Patricia, Ojok Albert, Kagood Francis, Hassan Chakaal and the village local chair persons for their guidance. Author Contributions Conceptualization: Hindum Lanyero, Moses Ocan, Celestino Obua, Cecilia Stålsby Lund- borg, Joan N. Kalyango, Jaran Eriksen. Data curation: Hindum Lanyero, Joan N. Kalyango, Jaran Eriksen. Formal analysis: Hindum Lanyero, Katureebe Agaba. Funding acquisition: Celestino Obua, Cecilia Stålsby Lundborg, Joan N. Kalyango, Jaran Eriksen. Investigation: Hindum Lanyero, Sarah Nanzigu. Methodology: Hindum Lanyero, Moses Ocan, Katureebe Agaba, Joan N. Kalyango, Sarah Nanzigu. PLOS ONE | https://doi.org/10.1371/journal.pone.0257328 September 16, 2021 11 / 14 PLOS ONE Validity of caregivers’ reports on prior use of antibacterials in children under five years Project administration: Moses Ocan, Celestino Obua, Cecilia Stålsby Lundborg, Joan N. Kalyango, Jaran Eriksen. Resources: Hindum Lanyero, Jaran Eriksen. Software: Hindum Lanyero. Supervision: Moses Ocan, Celestino Obua, Cecilia Stålsby Lundborg, Joan N. Kalyango, Jaran Eriksen, Sarah Nanzigu. Validation: Hindum Lanyero, Jaran Eriksen, Sarah Nanzigu. Visualization: Hindum Lanyero. Writing – original draft: Hindum Lanyero. Writing – review & editing: Hindum Lanyero, Moses Ocan, Celestino Obua, Cecilia Stålsby Lundborg, Katureebe Agaba, Joan N. Kalyango, Jaran Eriksen, Sarah Nanzigu. References 1. Ocan M, Bwanga F, Bbosa GS, Bagenda D, Waako P, Ogwal-Ogeng J, et al. Patterns and predictors of self-medication in Northern Uganda. PLoS ONE. 2014; 9(3):e92323. https://doi.org/10.1371/journal. pone.0092323 PMID: 24658124 2. Wellcome Trust, UK Government. Safe, secure and controlled:managing the supply chain of antimicro- bials. Review on antimicrobial Resistance. 2015. 3. Ministry of Health Uganda. Uganda National Drug Policy. 2002. Available from: http://library.health.go. ug/publications/policy-documents/uganda-national-drug-policy-2002. 4. Ocan M, Manabe YC, Baluku H, Atukwase E, Ogwal-Okeng J, Obua C. Prevalence and predictors of prior antibacterial use among patients presenting to hospitals in Northern Uganda. BMC Pharmacology and Toxicology. 2015; 16(1):26. https://doi.org/10.1186/s40360-015-0027-8 PMID: 26407973 5. Hildenwall H, Lindkvist J, Tumwine JK, Bergqvist Y, Pariyo G, Tomson G, et al. Low validity of caretak- ers’ reports on use of selected antimalarials and antibiotics in children with severe pneumonia at an urban hospital in Uganda. Transactions of the Royal Society of Tropical Medicine and Hygiene. 2009; 103(1):95–101. https://doi.org/10.1016/j.trstmh.2008.04.046 PMID: 18678381 6. Omolase C, Adeleke O, Afolabi A, Ofolabi O. Self medication amongst general outpatients in a Nigerian community hospital. Annals of Ibadan postgraduate medicine. 2007; 5(2):64–7. https://doi.org/10.4314/ aipm.v5i2.64032 PMID: 25161435 7. Moise K, Bernard JJ, Henrys JH. Evaluation of antibiotic self-medication among outpatients of the state university hospital of Port-Au-Prince, Haiti: a cross-sectional study. Pan African Medical Journal. 2017; 28(1). https://doi.org/10.11604/pamj.2017.28.4.12589 PMID: 29138650 8. Stirratt MJ, Dunbar-Jacob J, Crane HM, Simoni JM, Czajkowski S, Hilliard ME, et al. Self-report mea- sures of medication adherence behavior: recommendations on optimal use. Transl Behav Med. 2015; 5 (4):480–2. https://doi.org/10.1007/s13142-015-0315-2 PMID: 26622919 9. Nsimba SE, Massele AY, Eriksen J, Gustafsson LL, Tomson G, Warsame M. Case management of malaria in under-fives at primary health care facilities in a Tanzanian district. Trop Med Int Health. 2002; 7(3):201–9. https://doi.org/10.1046/j.1365-3156.2002.00847.x PMID: 11903982 10. Lerbech AM, Opintan JA, Bekoe SO, Ahiabu M-A, Tersbøl BP, Hansen M, et al. Antibiotic exposure in a low-income country: screening urine samples for presence of antibiotics and antibiotic resistance in coagulase negative staphylococcal contaminants. PLoS One. 2014; 9(12):e113055. https://doi.org/10. 1371/journal.pone.0113055 PMID: 25462162 11. Kamwesiga J. Uganda Health Care System. Kampala, Uganda: Makerere University. 2011. https://doi. org/10.1186/1745-6215-12-192 PMID: 21838913 12. Ministry of Health Uganda. National Health Facilty Master List. 2017. Available from: http://library. health.go.ug/sites/default/files/resources/National%20Health%Facility%MasterLlist%202017.pdf. 13. National Drug Authority. Licensed Out-lets. 2021.Available from: https://www.nda.or.ug/licensed- outlets/. 14. National Drug Authority. Drug Shops Licensed. 2021. Available from: https://www.nda.or.ug/drug- shops-licensed-in-2020/#1539347133280-41e0655f-3957. PLOS ONE | https://doi.org/10.1371/journal.pone.0257328 September 16, 2021 12 / 14 PLOS ONE Validity of caregivers’ reports on prior use of antibacterials in children under five years 15. Simple O, Mindra A, Obai G, Ovuga E, Odongo-Aginya EI. Influence of Climatic Factors on Malaria Epi- demic in Gulu District, Northern Uganda: A 10-Year Retrospective Study. Malar Res Treat. 2018; 2018:5482136. https://doi.org/10.1155/2018/5482136 PMID: 30186590 16. Omona S, Malinga GM, Opoke R, Openy G, Opiro R. Prevalence of diarrhoea and associated risk fac- tors among children under five years old in Pader District, northern Uganda. BMC Infectious Diseases. 2020; 20(1):1–9. https://doi.org/10.1186/s12879-020-4770-0 PMID: 31931735 17. Namakula J, Ssengooba F, Ssali S. Country situation analysis: northern Uganda. ReBUILD Consotium Kampala: Makerere University School of Public Health. 2011. Available from: https://www. rebuildconsortium.com/media/1024/country-situation-analysis-uganda.pdf 18. Kish Leslie. Survey Sampling. New York, United States: John Wiley & Sons Inc; 1995. 19. O’Connor R, O’Doherty J, O’Regan A, Dunne C. Antibiotic use for acute respiratory tract infections (ARTI) in primary care; what factors affect prescribing and why is it important? A narrative review. Irish Journal of Medical Science (1971-). 2018; 187(4):969–86. 20. Lanyero H, Eriksen J, Obua C, Stålsby Lundborg C, Nanzigu S, Katureebe A, et al. Use of antibacterials in the management of symptoms of acute respiratory tract infections among children under five years in Gulu, northern Uganda: Prevalence and determinants. PloS one. 2020; 15(6):e0235164. https://doi. org/10.1371/journal.pone.0235164 PMID: 32574206 21. UNICEF. Uganda (UGA)-Demographics, Health & Infant mortality-UNICEF DATA. 2020. Available from: https://data.unicef.org/country/uga/. 22. Radyowijati A, Haak H. Improving antibiotic use in low-income countries: an overview of evidence on determinants. Social science & medicine. 2003; 57(4):733–44. https://doi.org/10.1016/s0277-9536(02) 00422-7 PMID: 12821020 23. Li N, Ho KW, Ying G-G, Deng W-J. Veterinary antibiotics in food, drinking water, and the urine of pre- school children in Hong Kong. Environment international. 2017; 108:246–52. https://doi.org/10.1016/j. envint.2017.08.014 PMID: 28889029 24. Nayiga S, Kayendeke M, Nabirye C, Willis LD, Chandler CI, Staedke SG. Use of antibiotics to treat humans and animals in Uganda: a cross-sectional survey of households and farmers in rural, urban and peri-urban settings. JAC-Antimicrobial Resistance. 2020; 2(4):dlaa082. https://doi.org/10.1093/jacamr/ dlaa082 PMID: 34223037 25. Bashahun D, Odoch T. Assessment of antibiotic usage in intensive poultry farms in Wakiso District, Uganda. Livestock Research for Rural Development. 2015; 27(12). 26. Austin DJ, Kristinsson KG, Anderson RM. The relationship between the volume of antimicrobial con- sumption in human communities and the frequency of resistance. Proceedings of the National Academy of Sciences. 1999; 96(3):1152–6. https://doi.org/10.1073/pnas.96.3.1152 PMID: 9927709 27. Iramiot JS, Rwego IB, Kansiime C, Asiimwe BB. Epidemiology and antibiotic susceptibility of Vibrio cho- lerae associated with the 2017 outbreak in Kasese district, Uganda. BMC public health. 2019; 19(1):1– 9. https://doi.org/10.1186/s12889-018-6343-3 PMID: 30606151 28. Rang HP, Ritter JM, Flower R, Henderson G. Rang & Dale’s pharmacology E-book. Elsevier Health Sciences; 2014. Available from: https://www.elsevier.com/books/rang-andampamp-dales- pharmacology/ritter/978-0-7020-5362-7 29. 30. Flower RJ, Henderson G, Loke YK, MacEwan D, Rang HP. Rang & Dale’s Pharmacology E-Book: Elsevier Health Sciences; 2018. Available from: https://www.elsevier.com/books/rang-and-dales- pharmacology/ritter/978-0-7020-7448-6 Lerbech AM, Opintan JA, Bekoe SO, Ahiabu M-A, Tersbøl BP, Hansen M, et al. Antibiotic exposure in a low-income country: screening urine samples for presence of antibiotics and antibiotic resistance in coagulase negative staphylococcal contaminants. PloS one. 2014; 9(12):e113055. https://doi.org/10. 1371/journal.pone.0113055 PMID: 25462162 31. Badri M, Ehrlich R, Wood R, Maartens G. Initiating co-trimoxazole prophylaxis in HIV-infected patients in Africa: an evaluation of the provisional WHO/UNAIDS recommendations. AIDS (London, England). 2001; 15(9):1143–8. https://doi.org/10.1097/00002030-200106150-00009 PMID: 11416716 32. Mermin J, Lule J, Ekwaru JP, Downing R, Hughes P, Bunnell R, et al. Cotrimoxazole prophylaxis by HIV-infected persons in Uganda reduces morbidity and mortality among HIV-uninfected family mem- bers. AIDS (London, England). 2005; 19(10):1035–42. https://doi.org/10.1097/01.aids.0000174449. 32756.c7 PMID: 15958834 33. Ministry of Health Uganda. THE 2019 HIV EPIDEMIOLOGICAL SURVEILLANCE REPORT FOR UGANDA. 2020. Available from: https://www.health.go.ug/cause/the-2019-hiv-epidemiological- surveillance-report-for-uganda/. 34. Johnson LC, Beaton R, Murphy S, Pike K. Sampling bias and other methodological threats to the valid- ity of health survey research. International Journal of Stress Management. 2000; 7(4):247–67. PLOS ONE | https://doi.org/10.1371/journal.pone.0257328 September 16, 2021 13 / 14 PLOS ONE Validity of caregivers’ reports on prior use of antibacterials in children under five years 35. Caudron JM, Ford N, Henkens M, Mace C, Kiddle-Monroe R, Pinel J. Substandard medicines in resource-poor settings: a problem that can no longer be ignored. Tropical Medicine & International Health. 2008; 13(8):1062–72. 36. Monte AA, Heard KJ, Hoppe JA, Vasiliou V, Gonzalez FJ. The accuracy of self-reported drug ingestion histories in emergency department patients. The Journal of Clinical Pharmacology. 2015; 55(1):33–8. https://doi.org/10.1002/jcph.368 PMID: 25052325 37. Hafferty JD, Campbell AI, Navrady LB, Adams MJ, MacIntyre D, Lawrie SM, et al. Self-reported medica- tion use validated through record linkage to national prescribing data. Journal of clinical epidemiology. 2018; 94:132–42. https://doi.org/10.1016/j.jclinepi.2017.10.013 PMID: 29097340 PLOS ONE | https://doi.org/10.1371/journal.pone.0257328 September 16, 2021 14 / 14 PLOS ONE
null
10.3389_fpsyt.2022.1086038.pdf
null
Data availability statement The datasets presented in this article are not readily available because ethics approval did not include public data sharing. Requests to access the datasets should be directed to the corresponding author. organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
fpsyt-13-1086038 January 13, 2023 Time: 17:35 # 1 OPEN ACCESS EDITED BY Nuno Madeira, University of Coimbra, Portugal REVIEWED BY Sandra Vieira, Institute of Psychiatry, Psychology and Neuroscience, King’s College London, United Kingdom Marco Simoes, University of Coimbra, Portugal João Valente Duarte, University of Coimbra, Portugal *CORRESPONDENCE Diana Prata diana.prata@kcl.ac.uk SPECIALTY SECTION This article was submitted to Neuroimaging, a section of the journal Frontiers in Psychiatry RECEIVED 31 October 2022 ACCEPTED 29 December 2022 PUBLISHED 19 January 2023 CITATION Tavares V, Vassos E, Marquand A, Stone J, Valli I, Barker GJ, Ferreira H and Prata D (2023) Prediction of transition to psychosis from an at-risk mental state using structural neuroimaging, genetic, and environmental data. Front. Psychiatry 13:1086038. doi: 10.3389/fpsyt.2022.1086038 COPYRIGHT © 2023 Tavares, Vassos, Marquand, Stone, Valli, Barker, Ferreira and Prata. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms. TYPE Original Research PUBLISHED 19 January 2023 DOI 10.3389/fpsyt.2022.1086038 Prediction of transition to psychosis from an at-risk mental state using structural neuroimaging, genetic, and environmental data Vânia Tavares1,2, Evangelos Vassos3,4, Andre Marquand5,6, James Stone7, Isabel Valli8,9, Gareth J. Barker10, Hugo Ferreira1 and Diana Prata1,11* 1Instituto de Biofísica e Engenharia Biomédica, Faculdade de Ciências, Universidade de Lisboa, Lisbon, Portugal, 2Faculdade de Medicina, Universidade de Lisboa, Lisbon, Portugal, 3Social, Genetic and Developmental Psychiatry Centre, Institute of Psychiatry, Psychology and Neuroscience, King’s College London, London, United Kingdom, 4National Institute for Health Research Maudsley Biomedical Research Centre, South London and Maudsley National Health System Trust, London, United Kingdom, 5Donders Centre for Cognitive Neuroimaging, Donders Institute for Brain, Cognition and Behaviour, Radboud University, Nijmegen, Netherlands, 6Department of Cognitive Neuroscience, Radboud University Medical Centre, Nijmegen, Netherlands, 7Brighton and Sussex Medical School, University of Sussex, Brighton, United Kingdom, 8Department of Psychosis Studies, Institute of Psychiatry, Psychology and Neuroscience, King’s College London, London, United Kingdom, 9Institut d’Investigacions Biomèdiques August Pi i Sunyer, University of Barcelona, Barcelona, Spain, 10Department of Neuroimaging, Institute of Psychiatry, Psychology and Neuroscience, King’s College London, London, United Kingdom, 11Department of Old Age Psychiatry, Institute of Psychiatry, Psychology and Neuroscience, King’s College London, London, United Kingdom Introduction: Psychosis is usually preceded by a prodromal phase in which patients are clinically identified as being at in an “At Risk Mental State” (ARMS). A few studies have demonstrated the feasibility of predicting psychosis transition from an ARMS using structural magnetic resonance imaging (sMRI) data and machine learning (ML) methods. However, the reliability of these findings is unclear due to possible sampling bias. Moreover, the value of genetic and environmental data in predicting transition to psychosis from an ARMS is yet to be explored. Methods: In this study we aimed to predict transition to psychosis from an ARMS using a combination of ML, sMRI, genome-wide genotypes, and environmental risk factors as predictors, in a sample drawn from a pool of 246 ARMS subjects (60 of whom later transitioned to psychosis). First, the modality-specific values in predicting transition to psychosis were evaluated using several: (a) feature types; (b) feature manipulation strategies; (c) ML algorithms; (d) cross-validation strategies, as well as sample balancing and bootstrapping. Subsequently, the modalities whose at least 60% of the classification models showed an balanced accuracy (BAC) statistically better than chance level were included in a multimodal classification model. Results and discussion: Results showed that none of the modalities alone, i.e., neuroimaging, genetic or environmental data, could predict psychosis from an ARMS statistically better than chance and, as such, no multimodal classification model was trained/tested. These results suggest that the value of structural MRI data and genome-wide genotypes in predicting psychosis from an ARMS, which has been fostered by previous evidence, should be reconsidered. KEYWORDS machine learning, biomarker, schizophrenia, ARMS, prognosis Frontiers in Psychiatry 01 frontiersin.org fpsyt-13-1086038 January 13, 2023 Time: 17:35 # 2 Tavares et al. 10.3389/fpsyt.2022.1086038 1. Introduction Psychosis is a severe condition usually within the context of a mental disorder such as a schizophrenia, some neurological disorders (e.g., Alzheimer’s disease) or other medical conditions (e.g., induced by drugs or illicit substances), characterized by a disconnection from reality (1). The onset of psychosis, when in the context of a mental disorder, is typically preceded by a prodromal phase that lasts months to years (2); and usually starts early during adolescence and precedes the onset of psychotic symptoms by 10 or more years (3). In this prodromal phase, subtle and subjectively experienced disturbances in mental processes emerge (basic symptoms). These are the first manifestations of the neurobiological processes underlying psychosis and are mainly distinguished from other symptoms (i.e., positive or negative symptoms) by their self-experience nature (4). As the course of the psychotic illness evolves, increasingly disabling behavioral symptoms start to emerge, generally called negative symptoms, in particular a reduction of motivation and/or expressiveness (5). Additionally, cognitive deficits in attention, memory, reasoning, lack of concentration and executive functioning appear (6). Lastly, positive symptoms emerge, such as hallucinations, delusions, disorganized speech, and behavior (1). A patient may be clinically identified as being at a late prodromal phase of psychosis or having an “At Risk Mental State” (hereinafter: ARMS) if they present a functional decline in association with one or more of the following commonly used criteria (2, 7): (1) attenuated psychotic symptoms (APS), such as delusions, hallucinations, or disorganized speech with a frequency of at least once per week in the past month; (2) a brief limited intermittent psychotic (BLIP) episode lasting less than 1 week which resolves without antipsychotic medication; or (3) a genetic liability to psychosis or schizotypal traits, i.e., having either a first-degree relative with psychosis or a schizotypal personality disorder. Transition to psychosis from an ARMS may be evaluated based on the severity, frequency, and total duration of the psychotic symptoms, i.e., when the subject experiences a first episode of psychosis (FEP). Subjects with an ARMS and seeking help have a transition rate to psychosis of about 9% in the first 6 months and 25% in the first 3 years (8) and, in particular, an increased risk of transition to schizophrenia of 15.7% within an average period of 2.35 years, as shown by a meta-analysis (9). Thus, most of the people with an ARMS who later develop a psychotic illness will be diagnosed with schizophrenia. Furthermore, since about 70% of subjects diagnosed with an ARMS never develop a full- blown psychotic illness (9), these people may benefit from a less intensive treatment to ameliorate symptoms or need no treatment at all. Such increase in treatment cost-effectiveness would represent a substantial decrease in healthcare costs, and treatment burden to patients, including pharmacological side effects. However, there is no method for distinguishing between individuals with an ARMS who will subsequently develop a psychotic illness from those who will not (i.e., before a FEP onset). Given the above need, an effective, precise, and quantitative tool for the prediction of transition to psychosis from an ARMS has been sought by several studies employing machine learning (ML) methods and structural magnetic resonance imaging (sMRI). Indeed, several studies have consistently showed prediction of transition to psychosis from as ARMS with accuracies ranging between 74 and 84% (10– 15). Transition to psychosis from an ARMS using only sMRI and ML was first predicted using whole-brain gray matter volume metrics with an accuracy of 82% [(15 ARMS who transitioned to psychosis (ARMS-T) and 18 who did not (ARMS-NT)] (10). This finding was later replicated: (a) in an independent sample by the same group [balanced accuracy (BAC) = 84%, 16 ARMS-T and 21 ARMS-NT] (11); (b) combining both these samples (BAC = 80%, 33 ARMS-T and 33 AMRS-NT) (12); (c) using also one of the above samples for graph- extracted network metrics from cortical gyrification (BAC = 81%, 16 ARMS-T and 63 ARMS-NT) (15), and regional gray matter metrics (BAC = 74%, 16 ARMS-T and 19 ARMS-NT) (14); and (d) using regional gray matter metrics in an independent sample (BAC = 77%, 17 ARMS-T and 17 ARMS-NT; specificity of a replication sample of individuals with an ARMS who did not develop psychosis = 68%, 40 ARMS-NT) (13). To date, only two, relatively small, ARMS samples have been used for sMRI and ML analysis: FETZ (10, 12, 15) and FePsy (11, 12, 14). Thus, the robustness and generalizability of the above findings are still unclear due to possible specific sample characteristics, i.e., small sample sizes (from 33 individuals to at most 79 individuals with ARMS), with several studies stemming from a single site (10, 11, 13–15) or a combination of previously studied sites (12), which makes them not actual replications, with one exception (13). Interestingly, to the best of our knowledge, genetic data has been explored for the prediction of the transition to psychosis from an ARMS only once (16). In this study, a schizophrenia polygenic risk score (PRS) was able to predict transition to psychosis in individuals with an European [area under the curve (AUC) = 0.65; 32 ARMS- T and 92 ARMS-NT] and with a Non-European (AUC = 0.59; 48 ARMS-T and 156 ARMS-NT) ancestry, respectively. This is despite there being several classification studies showing that genetic markers can predict schizophrenia (17–22), FEP (23) or ARMS (23), both of individual polymorphisms (18, 19, 21, 23) or, composite polygenic scores (20–22), and gene expression profiles (24). From an environmental exposure perspective, and to the best of our knowledge, environmental data have never been explored for predicting individual transition to psychosis from an ARMS. The combination of neuroimaging measures and genetics or environmental measures, using ML, has, to the best of our knowledge, been explored once to predict ARMS prognosis (i.e., transition to psychosis from an ARMS) in a study running in parallel to ours (25). Therein, a large sample from the PRONIA project (26 ARMS-T and 308 ARMS-NT from 7 sites) was used to build a sequential stacked multimodal model using clinical-neurocognitive (including environmental data), genetic (in the form of a PRS for schizophrenia) and neuroimaging (in the form of voxel-based gray matter volume maps) data and - unlike the present study–human prognostic ratings, showing a final balanced accuracy in predicting transition to psychosis of 86%. In the present longitudinal prognostic biomarker study, we aimed to explore the use of ML models trained with sMRI, genetic, and environmental baseline data to predict the individual-level transition to psychosis from an ARMS within a 2-year follow up. While providing such preliminary (given the unprecedented data combinations/features and a limited sample size) evidence at the multimodal level, we took the opportunity to attempt to replicate previous promising sMRI-ML findings of studies using similar or smaller sample size (10–15). Methods-wise, we used naturalistically diverse samples but balanced them for demographic (age and sex) and imaging (scan acquisition sMRI protocol) variables. We set out to train and test modality-specific models first and then, provided Frontiers in Psychiatry 02 frontiersin.org F r o n t i e r s TABLE 1 Socio-demographic and clinical information of the At Risk Mental State (ARMS) sample with structural MRI data. Protocol 1 Protocol 2 Protocol 3 Group comparison i n P s y c h a t r y i Age at baseline (years) Age at follow-up or transition (years) Age at scan (years) Interval between baseline and scan age (years) ARMS-T (n = 14) 23.2 ± 3.4 [15.6 26.9] 25.6 ± 4.2 [17.3 33.4] 23.0 ± 3.6 [17.5 27.8] –0.2 ± 1.4 [–2.3 1.9] ARMS-NT (n = 19) 24.5 ± 4.8 [19.2 34.5] 32.7 ± 5.2 [22.6 43.9] 23.9 ± 4.8 [18.5 34.8] –0.5 ± 1.1 [–2.3 2.1] ARMS-T (n = 3) 26.2 ± 7.0 [20.1 33.8] 29.2 ± 5.4 [20.2 38.8] 27.0 ± 8.2 [20.2 36.1] 0.9 ± 1.3 [0.1 2.4] ARMS-NT (n = 16) 24.5 ± 5.2 [17.8 35.3] 28.8 ± 5.6 [22.9 43.1] 25.1 ± 5.4 [18.6 37.4] 0.5 ± 0.5 [0.1 2.1] ARMS-T (n = 6) 23.4 ± 4.5 [17.5 29.2] 25.2 ± 4.8 [18.3 31.0] 24.1 ± 4.8 [18.3 30.8] 0.6 ± 0.5 [0.2 1.6] ARMS-NT (n = 41) 21.8 ± 4.3 [17.1 33.1] 25.6 ± 4.8 [20.3 41.2] 22.4 ± 4.6 [17.7 38.3] 0.6 ± 1.0 [0.0 5.1] Sex (male/female) 11/3 9/10 2/1 14/2 3/3 19/22 Handednessa (right/left/ambidextrous) 12/0/1 16/0/0 3/0/0 13/1/0 4/0/0 36/4/0 0 3 Self-reported ethnicity (White/Black/Asian/other) 7/5/1/1 11/5/1/2 2/1/0/0 13/1/1/1 4/1/1/0 19/19/1/2 Years of education IQ (z-standardized)b GAF at baseline GAF at follow-upc CAARMS at baselined CAARMS at follow-upe 13.4 ± 2.1 [10 18] –1.1 ± 1.1 [–2.1 1.0] 52.9 ± 16.0 [35 90] 49.3 ± 18.6 [10 69] 33.2 ± 13.0 [9 56] 19.6 ± 23.0 [0 63] 13.1 ± 1.9 [10 17] 0.0 ± 1.1 [–2.1 1.8] 57.8 ± 11.4 [35 75] 58.5 ± 17.9 [20 94] 28.4 ± 15.3 [8 58] 11.6 ± 10.9 [0 31] 11.7 ± 2.3 [9 13] 0.1 ± 0.1 [0.0 0.2] 58.7 ± 3.2 [55 61] 27.3 ± 6.8 [22 35] 29.3 ± 21.9 [12 54] 42.0 ± 43.3 [6 90] 14.1 ± 2.6 [11 20] 0.5 ± 0.9 [–1.3 1.6] 58.6 ± 9.9 [41 75] 62.3 ± 13.5 [46 93] 23.2 ± 14.3 [0 51] 14.7 ± 18.4 [0 54] 15.2 ± 2.5 [11 18] −0.1 ± 1.3 [–2.1 1.6] 50.3 ± 11.4 [35 65] 52.5 ± 20.0 [30 78] 39.7 ± 24.1 [0 69] 42.7 ± 42.1 [0 102] 13.0 ± 2.2 [9 20] 0.1 ± 1.1 [–2.1 3.5] 53.6 ± 14.8 [0 75] 66.2 ± 13.6 [33 87] 28.5 ± 16.7 [0 81] 15.5 ± 17.2 [0 60] Protocol: p = 0.271 Transition: p = 0.592 Protocol × Transition: p = 0.447 Protocol: p = 0.027* Transition: p = 0.099 Protocol × Transition: p = 0.025* Protocol: p = 0.261 Transition: p = 0.499 Protocol × Transition: p = 0.541 Protocol: p < 0.001*** Transition: p = 0.419 Protocol × Transition: p = 0.795 Protocol × Transition: Protocol 1: p = 0.070 Protocol 2: p = 0.422 Protocol 3: p = 1 Protocol × Transition: Protocol 1: p = 0.448 Protocol 2: p = 1 Protocol 3: p = 1 Protocol × Transition: Protocol 1: p = 0.933 Protocol 2: p = 0.530 Protocol 3: p = 0.212 Protocol: p = 0.298 Transition: p = 0.966 Protocol × Transition: p = 0.024* Protocol: p = 0.427 Transition: p = 0.252 Protocol × Transition: p = 0.923 Protocol: p = 0.402 Transition: p = 0.475 Protocol × Transition: p = 0.877 Protocol: p = 0.064 Transition: p < 0.001*** Protocol × Transition: p = 0.095 Protocol: p = 0.505 Transition: p = 0.153 Protocol × Transition: p = 0.824 Protocol: p = 0.082 Transition: p = 0.001*** Protocol × Transition: p = 0.262 f r o n t i e r s i n o r g . Data format: mean ± standard deviation [min max]. Information not available for a1 ARMS-T and 3 ARMS-NT (Protocol 1), 2 ARMS-NT (Protocol 2), 2 ARMS-T and 1 ARMS-NT (Protocol 3); b1 ARMS-T and 1 ARMS-NT (Protocol 2), 1 ARMS-NT (Protocol 3); c2 ARMS and 5 ARMS-NT (Protocol 1), 4 ARMS-NT (Protocol 2), 8 ARMS-NT (Protocol 3); d2 ARMS-T and 7 ARMS-NT (Protocol 1), 3 ARMS-NT (Protocol 2), 2 ARMS-NT (Protocol 3); e3 ARMS-T and 6 ARMS-NT (Protocol 1), 3 ARMS-NT (Protocol 2), 8 ARMS-NT (Protocol 3). ARMS, at-risk mental state; ARMS-T, individuals at ARMS that did not transition to psychosis; ARMS-NT, individuals at ARMS that did not transitioned to psychosis. *p < 0.05; ***p < 0.001. f p s y t - 1 3 - 1 0 8 6 0 3 8 J a n u a r y 1 3 , 2 0 2 3 T i m e : 1 7 : 3 5 # 3 T a v a r e s e t a l . . 1 0 3 3 8 9 / f p s y t . 2 0 2 2 . 1 0 8 6 0 3 8 fpsyt-13-1086038 January 13, 2023 Time: 17:35 # 4 Tavares et al. 10.3389/fpsyt.2022.1086038 these performed above chance-level, a multimodal one. For the sMRI data, we used state-of-the-art preprocessing and ML pipelines; and explored several unprecedented combinations of brain structural measures, feature manipulation and cross-validation (CV) strategies. For the genetic data, we explored several approaches: a schizophrenia individual GWA-implicated SNPs (27), and a brain- PRS (26), specific expression Quantitative Trait Loci (eQTL) score. For the environmental data, we employed a schizophrenia environmental risk score (ERS) (28), and individual risk factors. 2. Materials and methods 2.1. Sample description The total sample consisted of 246 individuals with an ARMS, recruited at first presentation from consecutive referrals to the Outreach and Support in South London (OASIS) high-risk service, South London and Maudsley NHS Foundation Trust (29). The presence of ARMS was assessed using the CAARMS, a detailed clinical assessment (30). When the subjects were first diagnosed as having an ARMS (i.e., baseline) a set of data were acquired: (a) a sMRI scan; (b) genome-wide genotypes; and (c) assessment of environmental risk exposures. Subjects were labeled as transitioned to psychosis (ARMS-T) if they later presented a FEP or as not- transitioned to psychosis (ARMS-NT) if they did not present a FEP within a period of at least 2 years. For a detailed description of the recruitment, inclusion and exclusion criteria please refer to the Supplementary material. Additional socio-demographic including: and clinical measures were also assessed at baseline, age; sex; handedness; self-reported ethnicity; full scale intelligence TABLE 2 Socio-demographic and clinical information of the At Risk Mental State (ARMS) sample with genetic data and an European ancestry. ARMS-T (n = 21) ARMS-NT (n = 54) Group comparison Age at baseline (years) Age at follow-up or transition (years) 23.8 ± 5.3 [15.6 33.8] 25.3 ± 5.9 [17.3 38.8] 22.5 ± 4.0 [14.6 34.5] 27.9 ± 5.1 [18.5 43.9] Sex (male/female) 14/7 30/24 Years of education IQ (z-standardized)a GAF at baseline GAF at follow-upb CAARMS at baselinec CAARMS at follow-upd 13.0 ± 2.2 [10.0 18.0] 0.1 ± 1.0 [–2.1 2.2] 54.0 ± 15.7 [0 80] 47.8 ± 24.3 [0 79] 37.6 ± 17.5 [6 69] 24.4 ± 27.9 [0 90] 12.0 ± 4.4 [0 18.0] 0.2 ± 1.0 [–2.1 1.8] 53.6 ± 16.0 [0 78] 59.2 ± 21.0 [0 94] 29.9 ± 16.2 [0 81] 12.4 ± 14.0 [0 60] p = 0.284 p = 0.069 p = 0.380 p = 0.292 p = 0.678 p = 0.923 p = 0.050 p = 0.097 p = 0.019* quotient measured by the National Adult Reading Test (31); years of education; and global assessment of function using the GAF instrument tool at baseline and at follow-up (32), and CAARMS (at baseline and follow-up) (30). Regarding the sMRI, genetic and environmental sub-samples: 99, 135 and all the 246 individuals with an ARMS had a baseline sMRI scan (Table 1), genome-wide genotyped data (Table 2), and environmental risk factors assessment data (Table 3), respectively (more details in the Supplementary material). Over the 2-years follow-up period, 23, 41, and 60 individuals at an ARMS from each of the previous sub-samples developed psychosis (AMRS-T) and the remaining 15, 94, and 186 did not (ARMS-NT), respectively. Moreover, part of the study’s data collection occurred under the Genetic and Psychosis (GAP) umbrella project (33). Ethics approval was obtained by the NHS South East London Research Ethics Committee (Project GAP; Ref. 047/04), consistent with the Helsinki Declaration of 1975 (as revised in 2008) and all subjects gave written informed consent. Socio-demographic and clinical variables were analyzed using a two-tailed independent t-test or a Univariate Analysis of Variance (ANOVA) for continuous data and a chi-square test or Fisher’s exact test (if there were less than 5 subjects in one group) for ordinal data (Tables 1–3). These statistical analyses were performed using the Statistical Package for the Social Sciences 26 (SPSS 26 for Windows, Chicago, IL, USA). 2.2. Structural neuroimaging data 2.2.1. Structural magnetic resonance imaging acquisition Structural magnetic resonance imaging (sMRI) scans were acquired with one of two scanners (one with a field strength of 1.5T, three 3-Dimensional enhanced fast gradient echo protocols (detailed description in Supplementary material). the other 3T) using one of 2.2.2. Image processing High spatial resolution volumetric T1-weighted images were processed with the Computational Anatomy Toolbox for Statistical Parametric Mapping (SPM) –12 (CAT12; v10921), an SPM12 add- on (v69092) using default settings and MATLAB (9.3) as we have described elsewhere (34) (detailed description in Supplementary material). In summary, gray and white matter volumes for 64 regions-of-interest is in the Supplementary Table 1) were extracted using the Hammers atlas (35). Additionally, regional-based cortical thickness and surface measures (i.e., folding measures)–gyrification index, the depth of sulci and the measurement of local surface complexity were extracted for 68 ROIs (description of each ROI is in the Supplementary Table 2) defined by the Desikan–Killiany atlas (36). (ROIs; description of each ROI 2.2.3. Image quality control The quality of each processed image was empirically assessed using the quality assurance framework of CAT12 (detailed description in the Supplementary material). We set the subject’s image inclusion threshold at D (sufficient), i.e., only subjects whose Data format: mean ± standard deviation [min max]. Information not available for a2 ARMS-T and 9 ARMS-NT; b4 ARMS-NT; c1 ARMS and 9 ARMS-NT; d1 ARMS-T and 3 ARMS-NT. ARMS, at-risk mental state; ARMS-T, individuals at ARMS that did not transition to psychosis; ARMS-NT, individuals at ARMS that did not transitioned to psychosis. *p < 0.05. 1 http://www.neuro.uni-jena.de/cat/ 2 http://www.fil.ion.ucl.ac.uk/spm/ Frontiers in Psychiatry 04 frontiersin.org fpsyt-13-1086038 January 13, 2023 Time: 17:35 # 5 Tavares et al. 10.3389/fpsyt.2022.1086038 images had an image quality rate of A (excellent) to D (sufficient) (in a scale that goes up to F–unacceptable/failed) were included in the final sample, as it has been shown that typical scientific (clinical) data get good-to-satisfactory ratings (37). All this study’s images passed the above criteria and thus were included in all analyses (see Supplementary material for more details). migrant; (4) belonging to an ethnic minority; (5) the upbringing urbanicity level; (6) the parental age at birth; (7) the presence of childhood trauma; and (8) the season of birth (detailed description of how the risk for psychosis was assessed in each factor is in Supplementary material). 2.3. Genetic data Genotyping procedures have been previously described (26, 38). In summary, samples were genotyped at two different sites with two distinct chips (Illumina HumanCore Exome BeadChip and Genome-wide Human SNP Array 6.0). A standard quality control screening (exclusion of SNPs with low minor allele frequency, high genotypic failure and not in Hardy Weinberg equilibrium) followed by imputation procedures were conducted. Then, samples from both sites were merged by keeping only the overlapped imputed SNPs followed by a second quality control screening. Finally, a population stratification analysis was conducted with principal component analysis (PCA) to select only subjects with a European ancestry (the number of subjects per self-reported ethnicity is in the Supplementary Table 3). For a detailed description see the Supplementary material. 2.4. Environmental data Each subject was assessed on at least one of eight environmental risk factors: (1) tobacco and (2) cannabis consumption; (3) being TABLE 3 Socio-demographic and clinical information of the At Risk Mental State (ARMS) sample with environmental data (with less than 20% of the environmental risk factors missing). ARMS-T (n = 37) ARMS-NT (n = 97) Group comparison Age at baseline (years) Age at follow-up or transition (years)a 23.6 ± 4.8 [15.6 33.6] 25.6 ± 5.6 [17.3 39.2] 21.9 ± 3.7 [14.6 33.1] 27.1 ± 4.7 [18.5 41.2] Sex (male/female) 22/15 50/47 Years of educationb IQ (z-standardized)c GAF at baselined GAF at follow-upe CAARMS at baselinef CAARMS at follow-upg 13.2 ± 2.7 [8 18] −0.3 ± 1.0 [–2.1 2.2] 55 ± 12.5 [35 90] 50.4 ± 19.9 [10 88] 30.9 ± 19.4 [0 78] 29.7 ± 31.2 [0 102] 13.3 ± 2.0 [9 18] 0.1 ± 1.0 [–2.1 3.5] 56.7 ± 8.6 [40 80] 63.2 ± 14.2 [20 94] 28.3 ± 16.0 [0 81] 13.3 ± 14.2 [0 60] p = 0.027* p = 0.131 p = 0.411 p = 0.686 p = 0.049* p = 0.523 p =< 0.001* p = 0.478 p =<0.001* Data format: mean ± standard deviation [min max]. Information not available for a1 ARMS-T; b5 ARMS-T and 6 ARMS-NT; c7 ARMS-T and 13 ARMS-NT; d5 ARMS-T and 4 ARMS-NT; e5 ARMS-T and 8 ARMS-NT; f6 ARMS-T and 13 ARMS-NT; g4 ARMS-T and 8 ARMS-NT; subject. ARMS, at-risk mental state; ARMS-T, individuals at ARMS that did not transition to psychosis; ARMS-NT, individuals at ARMS that did not transition to psychosis. *p < 0.05. 2.5. Machine learning approach Several ML strategies to generate prediction models for transition to psychosis from sMRI data using our ARMS sample were investigated (Figures 1, 2). These include: (a) sample balancing and bootstrapping; and testing several: (b) feature types; (c) feature manipulation approaches; and (d) CV approaches. The analyses were conducted using the neuroimaging ML tool NeuroMiner v1.0 ELESSAR3 for sMRI data, chosen given that it was used in the previous above-mentioned ARMS prognosis studies and provided therein high accuracy results (12, 39, 40), and R software 4.0.5 (41) for genetic (16) and environmental data. As detailed below, we have used SVM on the neuroimaging data since that is the approach which not only is more often employed with sMRI data but also that which has shown higher accuracies in psychiatric diagnostic classifications using sMRI data including in the ARMS population (10–14) which we herein attempt to replicate. We have used elastic-net algorithm for the genetic data (SNPs and eQTL scores) and environmental risk factors as it a well-suited method for dealing with high-dimensional data and possibly correlated data; and it performs an embedded feature selection and model fitting at once. The PRS and the environmental risk score were analyzed with logistic regression, given that only one feature was used. 2.5.1. Sample balancing and bootstrapping The final sample used in the ML analyses was defined by all the ARMS-T subjects available (23 subjects for the sMRI predictors, 19 for the PRS predictor, 21 for the SNP’s alleles predictors, 21 for eQTL scores predictors, 37 for the ERS predictor, and 17 for the individual environmental predictors), and the same number of ARMS-NT subjects randomly selected to match the ARMS-T for age and sex (for each data modality), and for scan acquisition protocol (for sMRI data). The matching criteria for age and sex were based on the non-rejection of the null hypothesis (i.e., p > 0.05) that the ARMS-T and ARMS-NT groups had the same median age (tested with a two-sided Mann–Whitney U-test) and sex proportions (tested with a two-sided chi-square statistic). The matching for the scan acquisition protocol was done in a one-to-one manner, i.e., the number of ARMS-NT subjects is the same as the number of ARMS-T. within each protocol Of note, we have considered the approach of applying a class- weighted support vector machine for our neuroimaging measures and have detected that differences in terms of accuracies between a model with weights vs. no-weights (considering the full unbalanced samples) were practically null (results not shown)–and therefore we did not pursue that approach. Then, each subsampling was repeated five times, i.e., 5 bootstrapped samples were created, and the subsequent ML analyses were conducted for each of the bootstrapped sample. 3 http://proniapredictors.eu/neurominer/index.html Frontiers in Psychiatry 05 frontiersin.org fpsyt-13-1086038 January 13, 2023 Time: 17:35 # 6 Tavares et al. 10.3389/fpsyt.2022.1086038 FIGURE 1 Overall machine learning approach taken for assessing the predictive value, i.e., the accuracy, of each type of extracted neuroimaging, genetic or environmental feature in predicting transition to psychosis from an At Risk Mental State (ARMS). ERS, environmental risk score; eQTL score, expression quantitative trait loci; PRS, polygenic risk score; ROIGM, regional-based gray matter volumes; ROISurface, surface-based regional cortical thickness, and gyrification, sulci, and complexity indexes; ROIWM, regional-based white matter volumes; SNP, single nucleotide polymorphism; VMGM, voxel-based gray matter volume maps; VMWM, voxel-based white matter volume maps. 2.5.2. Feature types 2.5.2.1. Structural magnetic resonance imaging data Individual ML models were trained and validated for each of the following brain measures: (a) voxel-based gray matter (VBGM) maps (297,811 initial features); (b) voxel-based white matter (VBWM) maps (204,706 initial features); (c) regional-based gray (ROIGM) and (d) white (ROIWM) matter volumes (each with 64 initial features) scaled to the total intracranial volume (TIV); and (e) surface-based regional cortical thickness, and gyrification, sulci, and complexity indexes (ROISurface; 272 initial features). Each feature is scaled between 0 and 1 before entering a support vector machine (SVM) classification algorithm. 2.5.2.2. Genetic data We tested whether a PRS which we have previously found to predict (R2 = 0.94) a cross-sectional diagnosis of FEP (vs. healthy controls) would be a good longitudinal predictor for ARMS prognosis. Following the same methodology (26), this PRS was computed as the sum of SNPs alleles statistically associated with schizophrenia in a GWAS meta-analysis study (42) weighted by the effect size of that association (more details in Supplementary material). In addition, two other novel prediction models using the present ARMS sample were trained and tested. One used SNPs’ alleles (79,247 SNPs) as predictors and the other used eQTL scores of genes expressed in brain tissue (141 genes across several brain tissues). Both SNPs and genes’ eQTL scores were chosen as the ones most associated with psychosis as ascertained in a recent meta-analysis (27). The eQTL score of each gene was extracted with the eGenScore which we developed and published previously (43) and it was computed as the sum of the alleles of SNPs showing a statistically significant association with the brain gene expression in a standard genomic and transcriptomic sample weighted by the size of that effect (further details available in Supplementary material). 2.5.2.3. Environmental data We tested whether an ERS for psychosis which we have previously developed (28) would be a good longitudinal predictor for ARMS prognosis. Only subjects with less than 20% of missing information (i.e., missing data for less than 2 environmental risk factors) were considered for the ERS-based ML analysis. Therefore, the final sample included 37 ARMS-T subjects and 97 ARMs-NT subjects. Then, each environmental risk factor (see Section “2.4. Environmental data”) was used as an individual feature in the model. For this ML analysis only subjects with information for all the environmental risk factors (i.e., with no missing information) were considered (i.e., 17 ARMS-T and 49 ARMS-NT subjects). Further details available in Supplementary material. 2.5.3. Feature manipulation Feature manipulation was performed only in ML analyses using sMRI data. In particular, feature dimensionality reduction was performed for VBGM and VBWM features using robust PCA (44, Frontiers in Psychiatry 06 frontiersin.org fpsyt-13-1086038 January 13, 2023 Time: 17:35 # 7 Tavares et al. 10.3389/fpsyt.2022.1086038 FIGURE 2 Scheme of the cross-validation (CV) approach taken to train, test, and validate classification models trained with (A) neuroimaging data and support vector machines (SVM); genetic (single nucleotide polymorphisms or expression quantitative trait loci) or environmental (environmental risk factors) data and elastic-net; or (B) genetic (polygenic risk score) or environmental (environmental risk score) data and logistic regression. 45). Here the robust PCA was applied during the inner CV cycle (see Section “2.5.5. Cross-validation”). The number of principal components that were retained explained up to 80% of the variance in the data and were limited by the inner CV cycle’s sample size, n, i.e., a maximum of only n/2 components could indeed be extracted. Supplementary Table 5 shows the maximum number of principal components that can be extracted for each inner CV cycle in each CV scheme that was used (see also Section “2.5.5. Cross-validation”) (for detailed description see the Supplementary material). Feature selection was performed on regional brain features (i.e., ROIGM, ROIWM, and ROISurface) using a greedy forward search feature selection algorithm. This is a stepwise algorithm that starts with an empty set of features and then tests the predictive value of every single feature, selecting the ones improving the overall accuracy across the inner CV cycle folds (see Section “2.5.5. Cross- validation”). The final set of features is, then, composed by the 10% most predictive variables. Additionally, no feature selection, i.e., using the total number regional brain features, was also tested. 2.5.4. Machine learning algorithm Binary classification of transition to psychosis from an ARMS (i.e., ARMS-T vs. ARMS-NT) was performed using linear SVM for sMRI data, and logistic regression and elastic net for both genetic and environmental data. 2.5.4.1. Support vector machine classification Binary classification of transition to psychosis from an ARMS (i.e., ARMS-T vs. ARMS-NT) using sMRI data was performed using linear SVM (46, 47). In this study we exclusively used a linear kernel SVM to reduce the risk of overfitting the data (given our final sample size being relatively small). Furthermore, the linear SVM classifier has a penalty parameter C that controls the trade- off between having zero training error and allowing misclassification. Herein, a parameter search was carried out to identify the optimal C value (i.e., 2l, l [−5 : 1 : 4]) in the inner CV cycle (see Section “2.5.5. Cross-validation”). (ERS) data was performed using simple logistic regression. A threshold of 0.5 was applied to the probability of observing i.e., an ARMS-T (see Supplementary material the outcome, for more details). 2.5.4.3. Elastic net for classification transition to psychosis Binary classification of from an ARMS (i.e., ARMS-T vs. ARMS-NT) using genetic (psychosis- associated SNPs or eQTL scores of psychosis-associated genes) or environmental (environmental risk factors) data was performed using logistic regularized regression with elastic net (48) using l1 and λ values hyperparameters search to identify the optimal (regression weights shrinkage) (i.e., l1 0 : 0.1 : 1; λ 0.01 : 0.01 : 1) in the inner CV cycle (see Section “2.5.5. Cross-validation”) (for detailed description see the Supplementary material). The elastic net was implemented using the “glmnet” v4.0 R package. 2.5.5. Cross-validation Each model (trained with sMRI, psychosis-associated SNPs or eQTL scores of psychosis-associated genes and environmental risk factors) was trained in a nested-CV scheme for hyperparameter tuning (in the inner CV cycle) and to estimate the generalizability of the trained prediction model and its performance (in the outer CV cycle) (Figure 2A). For more details see the Supplementary material. For sMRI models, we tested three different nested-CV schemes: (a) leave-one scan acquisition protocol-out (LSO); (b) leave-one per group from the same scan acquisition protocol-out (LPO); and (c) classic 5-fold CV. For the remaining sMRI, genetic (trained with psychosis-associated SNPs or eQTL scores of psychosis-associated genes data) and environmental (trained with environmental risk factors data) models, nested-CV was defined with an inner 5- fold and an outer leave-one per group-out (LPO) CV schemes. Furthermore, the logistic regression (trained with genetic–PRS–and environmental–ERS–data) was trained and tested in a simple LPO CV scheme (Figure 2B). 2.5.6. Performance measures 2.5.4.2. Logistic regression for classification Binary classification of transition to psychosis from an ARMS (i.e., ARMS-T vs. ARMS-NT) using genetic (PRS) or environmental Each model’s performance was evaluated using measures derived from the confusion matrix: sensitivity; specificity; BAC; positive likelihood ratio; negative likelihood ratio; and diagnostic odds ratio Frontiers in Psychiatry 07 frontiersin.org fpsyt-13-1086038 January 13, 2023 Time: 17:35 # 8 Tavares et al. 10.3389/fpsyt.2022.1086038 (DOR). Moreover, permutation testing was used to test if the BAC was higher than chance–50%–with a statistical significance of 5% (For a detailed description of each measure see the Supplementary material). The prediction ability of each tested combination of feature type, feature manipulation, and CV scheme was defined as significant if the BAC was higher than chance–50% in at least 3 out of 5 bootstrapped samples. evaluated by testing the statistical significance of the median BAC across bootstrapped samples using a one-tailed Wilcoxon signed rank test (i.e., to test if the median BAC across bootstrapped samples is higher than chance– 50%, with a statistical significance level of 5%). P-values were not adjusted for multiple comparisons due to non-independence of the samples used in each statistical test. 3. Results Overall, the BAC of the classification models trained and validated using each combination of feature type (i.e., ROIGM, ROIWM, ROISurface, VBGM, or VBWM–for sMRI data; PRS, psychosis-associated SNPs or psychosis-associated brain eQTL score genes scores–for genetic data; or ERS or individual environmental risk factors–for environmental data), feature manipulation (i.e., feature dimensionality reduction through PCA; no feature selection; or forward feature selection), CV scheme (i.e., LSO CV; LPO CV; or 5-fold CV), and bootstrapped sample (i.e., one of the 5 samples) ranged from 37 to 67% for the classification models trained with sMRI (Tables 4, 5 and Figures 3, 4), from 26 to 62% for the models trained with genetic data (Table 6 and Figure 5) and from 38 to 61% for models trained with environmental data (Table 6 and Figure 6). The prediction ability of each model was not significant as less than 3 bootstrapped samples per each feature type showed a BAC statistically higher than chance–50%. 4. Discussion This study aimed to predict transition to psychosis from an ARMS using ML applied to quantitative data across modalities– i.e., neuroimaging (sMRI), genetics (genome-wide genotypes), and environment–collected when subjects first sought clinical help (i.e., at baseline) and were identified with an ARMS. This is, to the best of our knowledge, the first study: (1) of longitudinal design exploring sMRI, genetic and environmental data to predict the development of a psychotic disorder from a prodromal stage; and (2) when considering each modality individually, exploring a range of approaches (for genetics and environmental data) and/or feature combinations (for sMRI data). 4.1. Prediction of transition to psychosis using structural neuroimaging data In this study we applied ML to structural neuroimaging data using a relatively larger sample and an ML approach, improved to the best of our ability, to detect transition to psychosis from an ARMS and to replicate previous positive findings of accuracies 74 to 84% of six studies, which together used 3 independent samples (10–15). For this, we decided: to: (1) use only the most recent versions of the image processing tools (i.e., CAT12) and ML tools (i.e., NeuroMiner); (2) replicate as accurately as possible the methods that were described in the abovementioned MRI papers since it was not possible to access their processing and ML pipelines; (3) add a layer of ML generalizability by bootstrapping and fitting a model to each subsample; and (4) overcome previous studies’ limitations (e.g., sample unbalancing for demographics). Furthermore, we explored, for the first time, the use of whole brain white matter volume and regional white matter volume, cortical thickness, and surface-based brain gyrification, sulci depth, and complexity indexes with ML to predict transition to psychosis. Unexpectedly, we did not replicate previous findings. After balancing the samples for binary classification of transition to psychosis accounting for age, sex, and the three different scan acquisition protocols to avoid overoptimistic results, the performance of all tested combinations (i.e., of feature type–ROIGM, ROIWM, ROISurface, VBGM, or VBWM; feature manipulation–feature dimensionality reduction through PCA, no feature selection, or forward feature selection; and CV scheme–LSO CV, LPO CV, or 5-fold CV) were not significantly better than chance level. Compared to the previous studies reporting high balanced accuracies (74 to 84%) in predicting transition to psychosis from sMRI maps (10–15), the current study has some advantages. First, this study’s sample is drawn from a more naturalistic ARMS population as it includes subjects whose sMRI images were acquired using three different scan acquisition protocols. Training a classification model with data from different centers potentially increases its generalizability. Only one of the previous transition to psychosis prediction studies used a two-site group balanced sample (12), combining the samples reported in two previous studies by the same authors (10, 11). The main differences between this report and our study are the following: (a) Their sample was larger than our balanced bootstrapped samples (i.e., 36% larger than ours, measured as the absolute value of the change in sample size, divided by the average of the size of the two samples). However, we tested our ML models on five balanced subsamples (i.e., through bootstrapping), allowing us to obtain a measure of generalizability of these models’ performance. Moreover, they do not present a measure of the statistical significance of the model’s BAC, which we do herein. (b) They controlled the effect of site on the classification using partial correlations during the training phase of the CV cycle, whereas we controlled it by keeping the same proportion of subjects at an ARMS that transitioned to psychosis and those who did not in each scan protocol during the training phase of the CV cycle (i.e., when using the LPO CV scheme as the previous study did). Additionally, we also guaranteed that the pair of subjects left out for testing/validation were from the same site. This potentially increases the generalizability of the classification model by training it with a more heterogeneous sample (and, as explained above, more naturalistic) and diminishing the effect of site on the testing/validation classification accuracy, which is not taken into account in the previous report (12). Second, we trained our classification models with samples balanced for group (subjects at an ARMS who later transitioned to psychosis and who did not), age at scan and sex. Balancing for group is important to avoid biasing the classification model to the most represented group and it was not taken into account by three out of six previous reports (10, 11, 14). Moreover, the effects of age (49) and sex (50) on brain structure, rate of transition to psychosis from ARMS (2), and prevalence of psychosis (3, 51), have been consistently reported and, therefore, should be taken into Frontiers in Psychiatry 08 frontiersin.org fpsyt-13-1086038 January 13, 2023 Time: 17:35 # 9 Tavares et al. 10.3389/fpsyt.2022.1086038 TABLE 4 Performance measures of each structural magnetic resonance imaging (sMRI) classification model based on brain regional features across bootstrapped samples. ROIGM No-FS 55.7 ± 6.4 [47.8, 65.2] 55.7 ± 10.4 [43.5, 69.6] 55.7 ± 5.5 [47.8, 63.0] 1.3 ± 0.3 [0.9, 1.9] 0.8 ± 0.2 [0.6, 1.1] 1.7 ± 0.8 [0.8, 3.0] 1 47.8 ± 6.1 [43.5, 56.5] 54.8 ± 6.6 [47.8, 60.9] 51.3 ± 4.8 [45.7, 58.7] 1.1 ± 0.2 [0.8, 1.4] 1.0 ± 0.2 [0.7, 1.2] 1.2 ± 0.5 [0.7, 2] 0 42.6 ± 3.6 [39.1, 47.8] 54.8 ± 12.5 [34.8, 65.2] 48.7 ± 6.6 [39.1, 56.5] 1.0 ± 0.3 [0.7, 1.4] 1.1 ± 0.3 [0.8, 1.6] 1.0 ± 0.5 [0.4, 1.7] 0 LSO CV scheme SE (%) SP (%) BAC (%) PLR NLR DOR Significant models LPO CV scheme SE (%) SP (%) BAC (%) PLR NLR DOR Significant models 5-fold CV scheme SE (%) SP (%) BAC (%) PLR NLR DOR Significant models FFS 59.1 ± 11.7 [47.8, 78.3] 40.9 ± 10.5 [26.1, 52.2] 50.0 ± 3.8 [43.5, 52.2] 1.0 ± 0.1 [0.8, 1.1] 1.0 ± 0.2 [0.8, 1.4] 1.0 ± 0.3 [0.6, 1.3] 0 67.0 ± 7.9 [56.5, 73.9] 44.3 ± 10.8 [34.8, 60.9] 55.7 ± 7.1 [45.7, 65.2] 1.3 ± 0.3 [0.9, 1.8] 0.8 ± 0.3 [0.5, 1.3] 1.9 ± 1.1 [0.7, 3.6] 0 40.9 ± 5.8 [34.8, 47.8] 40.9 ± 7.3 [30.4, 47.8] 40.9 ± 2.4 [37.0, 43.5] 0.7 ± 0.1 [0.6, 0.8] 1.5 ± 0.2 [1.3, 1.9] 0.5 ± 0.1 [0.3, 0.6] 0 ROIWM No-FS 57.4 ± 19.1 [30.4, 82.6] 46.1 ± 14 [21.7, 56.5] 51.7 ± 5.6 [43.5, 58.7] 1.1 ± 0.2 [0.7, 1.4] 0.9 ± 0.2 [0.7, 1.2] 1.3 ± 0.5 [0.6, 2.0] 0 49.6 ± 8.5 [39.1, 60.9] 52.2 ± 5.3 [43.5, 56.5] 50.9 ± 5.2 [43.5, 56.5] 1.0 ± 0.2 [0.8, 1.3] 1.0 ± 0.2 [0.8, 1.3] 1.1 ± 0.4 [0.6, 1.7] 0 59.1 ± 6.6 [52.2, 69.6] 40.9 ± 8.5 [30.4, 52.2] 50.0 ± 5.1 [45.7, 56.5] 1.0 ± 0.2 [0.9, 1.2] 1.0 ± 0.3 [0.7, 1.3] 1.1 ± 0.5 [0.7, 1.8] 0 ROISurface FFS No-FS FFS 62.6 ± 13.3 [52.2, 82.6] 27.8 ± 22.3 [0.0, 56.5] 45.2 ± 6.6 [37.0, 54.3] 0.9 ± 0.2 [0.7, 1.2] 1.5 ± 0.7 [0.8, 2.5] 0.6 ± 0.5 [0.0, 1.4] 0 39.1 ± 9.2 [26.1, 47.8] 50.4 ± 12.5 [34.8, 69.6] 44.8 ± 6.6 [37.0, 54.3] 0.8 ± 0.3 [0.5, 1.3] 1.3 ± 0.3 [0.9, 1.5] 0.7 ± 0.4 [0.3, 1.5] 0 45.2 ± 8.5 [34.8, 56.5] 53 ± 11.3 [34.8, 65.2] 49.1 ± 2.5 [45.7, 52.2] 1.0 ± 0.1 [0.9, 1.1] 1.1 ± 0.1 [0.9, 1.3] 0.9 ± 0.2 [0.7, 1.2] 0 41.7 ± 14.6 [26.1, 60.9] 61.7 ± 17.8 [34.8, 82.6] 51.7 ± 7.4 [43.5, 63.0] 1.2 ± 0.4 [0.8, 1.8] 1.0 ± 0.3 [0.6, 1.4] 1.4 ± 0.9 [0.6, 2.9] 1 53.9 ± 6.6 [43.5, 60.9] 54.8 ± 5.0 [47.8, 60.9] 54.3 ± 4.3 [50.0, 60.9] 1.2 ± 0.2 [1.0, 1.6] 0.8 ± 0.1 [0.6, 1.0] 1.5 ± 0.6 [1.0, 2.4] 1 53.0 ± 10.4 [39.1, 65.2] 54.8 ± 6.6 [47.8, 60.9] 53.9 ± 5.2 [47.8, 60.9] 1.2 ± 0.2 [0.9, 1.6] 0.9 ± 0.2 [0.6, 1.1] 1.5 ± 0.6 [0.8, 2.4] 0 39.1 ± 20.6 [17.4, 65.2] 63.5 ± 12.1 [43.5, 73.9] 51.3 ± 9.8 [43.5, 67.4] 1.1 ± 0.6 [0.6, 2.1] 1.0 ± 0.3 [0.5, 1.2] 1.5 ± 1.6 [0.5, 4.3] 1 52.2 ± 6.9 [43.5, 60.9] 54.8 ± 12.9 [39.1, 69.6] 53.5 ± 7.5 [43.5, 60.9] 1.2 ± 0.3 [0.8, 1.6] 0.9 ± 0.3 [0.6, 1.3] 1.5 ± 0.8 [0.6, 2.4] 0 57.4 ± 8.4 [47.8, 69.6] 53 ± 11.7 [39.1, 65.2] 55.2 ± 9.3 [47.8, 67.4] 1.3 ± 0.5 [0.9, 2.0] 0.9 ± 0.3 [0.5, 1.1] 2.0 ± 1.6 [0.8, 4.3] 1 Measures for each tested combination of brain regional feature type [i.e., regional-based gray (ROIGM) and white (ROIWM) matter volume; and surface-based regional cortical thickness, gyrification, sulci, and complexity indexes (ROISurface)], feature selection [i.e., no feature selection (No-FS); and forward feature selection (FFS)], and cross-validation (CV) scheme [i.e., leave-one scan acquisition protocol-out (LSO) CV; leave-one per group-out (LPO) CV; and 5-fold CV] are presented. Statistical significance of the balanced accuracy (BAC) for each bootstrapped sample was tested using permutation testing with a significance level of 5%. Data format: mean ± standard deviation [min max]. DOR, diagnostic odds ratio; NLR, negative likelihood ratio; PLR, positive likelihood ratio; SE, sensitivity; SP, specificity. Significant models: number of models with statistically significant BAC higher than 50%. account in these studies. All previous reports (and the current study) matched transition proportion for age and sex (10–14), except for one (15). Das and colleagues reported a statistically significant and better than chance level BAC in predicting transition to psychosis using a sample unbalanced for both group and sex. Although they used a ML algorithm with class (i.e., group) weighing–which in summary increases the influence of the minority class when training the model by assigning higher weights to rare cases, the authors Frontiers in Psychiatry 09 frontiersin.org fpsyt-13-1086038 January 13, 2023 Time: 17:35 # 10 Tavares et al. 10.3389/fpsyt.2022.1086038 TABLE 5 Performance measures of each structural magnetic resonance imaging (SMRI) classification model based on voxel-wise features across bootstrapped samples. LSO CV scheme LPO CV scheme 5-fold CV scheme SE (%) SP (%) BAC (%) PLR NLR DOR Significant models VBGM 20.9 ± 34.8 [0, 82.6] 72.2 ± 35.7 [8.7, 91.3] 46.5 ± 3.6 [43.5, 52.2] 0.6 ± 0.6 [0.0, 1.5] 1.3 ± 0.4 [1.0, 2.0] 0.7 ± 0.9 [0.0, 2.2] 0 VBWM 46.1 ± 38.2 [4.3, 78.3] 53 ± 35.4 [21.7, 95.7] 49.6 ± 2.4 [45.7, 52.2] 0.9 ± 0.3 [0.3, 1.1] 1.0 ± 0.1 [0.9, 1.1] 0.8 ± 0.4 [0.1, 1.1] 1 VBGM 47.0 ± 10.4 [34.8, 60.9] 55.7 ± 8.4 [43.5, 65.2] 51.3 ± 7.5 [45.7, 63.0] 1.1 ± 0.4 [0.8, 1.8] 1.0 ± 0.3 [0.6, 1.2] 1.3 ± 1.0 [0.7, 3.1] 0 VBWM 50.4 ± 11.3 [34.8, 60.9] 53.0 ± 7.1 [47.8, 65.2] 51.7 ± 2.8 [47.8, 54.3] 1.1 ± 0.1 [0.9, 1.2] 0.9 ± 0.1 [0.8, 1.1] 1.1 ± 0.2 [0.8, 1.4] 0 VBGM 30.4 ± 10.2 [21.7, 43.5] 51.3 ± 7.8 [43.5, 60.9] 40.9 ± 2.4 [37.0, 43.5] 0.6 ± 0.1 [0.5, 0.8] 1.4 ± 0.1 [1.3, 1.5] 0.4 ± 0.1 [0.2, 0.6] 0 VBWM 41.7 ± 8.5 [34.8, 56.5] 52.2 ± 6.1 [43.5, 60.9] 47.0 ± 6.8 [41.3, 58.7] 0.9 ± 0.3 [0.7, 1.4] 1.1 ± 0.3 [0.7, 1.4] 0.9 ± 0.7 [0.5, 2.1] 0 Measures for each tested combination of voxel-wise feature type [i.e., voxel-based gray (VBGM) and white (VBWM) matter volume maps], feature dimensionality reduction through principal component analysis and cross-validation (CV) scheme [i.e., leave-one scan acquisition protocol-out (LSO) CV; leave-one per group-out (LPO) CV; and 5-fold CV] are presented. Statistical significance of the balanced accuracy (BAC) for each bootstrapped sample was tested using permutation testing with a significance level of 5%. Data format: mean ± standard deviation [min max]. DOR, diagnostic odds ratio; NLR, negative likelihood ratio; PLR, positive likelihood ratio; SE, sensitivity; SP, specificity. Significant models: number of models with statistically significant BAC higher than 50%. FIGURE 3 Balanced accuracy across bootstrapped samples for each tested combination of regional feature type [i.e., regional-based gray and white matter volume; and surface-based regional cortical thickness, gyrification, sulci, and complexity indexes (surface-based regional measures)], feature selection [i.e., no feature selection; and forward feature selection (FFS)], and cross-validation (CV) scheme [i.e., leave-one scan acquisition protocol-out (LSO) CV; leave-one per group-out (LPO) CV; and 5-fold CV]. Dots represent the balanced accuracy value in each of the five bootstrapped samples and are red colored if the balanced accuracy is statistically significant (i.e., p < 0.05) or blue colored if it is not (i.e., p > 0.05). The statistical significance of the balanced accuracy in each bootstrapped sample was evaluated through permutation testing. performed an unspecified correction for sex effect (as well as for age and TIV effects) to the data during the training CV cycle. This approach may not be the most appropriate given the known effect of sex on brain structure (50) and the, abovementioned, empirically tested association between sex and group (i.e., transition to psychosis from an ARMS vs. no transition) (15), which makes sex a potential confounder in this analysis. Furthermore, in three of the six previous reports, the effects of age and sex were corrected before entering the ML analysis (10), and during the training CV cycle (11, 15) using partial correlations (10, 11) or an unspecified method (15)– which we did not perform. Correction for age effects in ML analysis has been previously shown to increase classification accuracy in Alzheimer’s disease, when it is estimated from healthy subjects (52). Correction for effects of no interest in ML analyses should be done with extreme caution as it can easily remove relevant subject-specific information (53). This is especially important when the correction Frontiers in Psychiatry 10 frontiersin.org fpsyt-13-1086038 January 13, 2023 Time: 17:35 # 11 Tavares et al. 10.3389/fpsyt.2022.1086038 FIGURE 4 Balanced accuracy across bootstrapped samples for each tested combination of voxel-wise feature type [i.e., voxel-based gray (VBGM) and white (VBWM) matter volume maps], feature dimensionality reduction through principal component analysis and cross-validation (CV) scheme [i.e., leave-one scan acquisition protocol-out (LSO) CV; leave-one per group-out (LPO) CV; and 5-fold CV. Dots represent the balanced accuracy value in each of the five bootstrapped samples and are red colored if the balanced accuracy is statistically significant (i.e., p < 0.05) or blue colored if it is not (i.e., p > 0.05). The statistical significance of the balanced accuracy in each bootstrapped sample was evaluated through permutation testing. TABLE 6 Performance measures of: (1) a genetic schizophrenia polygenic risk score (PRS), (2) a list of psychosis-associated single nucleotide polymorphisms (SNPs), (3) expression quantitative trait loci (eQTL) scores (43) of a list of psychosis-associated genes expressed in the brain; (4) an environmental schizophrenia risk score (ERS), and (5) a list of schizophrenia-associated environmental risk factors, classification models across bootstrapped samples. PRS SNP eQTL score ERS Environmental risk factors SE (%) SP (%) BAC (%) PLR NLR DOR 42.1 ± 20.0 [21.1, 63.2] 46.3 ± 11.4 [31.6, 57.9] 44.2 ± 15.3 [26.3, 60.5] 0.9 ± 0.5 [0.3, 1.5] 1.4 ± 0.8 [63.6, 2.5] 1.0 ± 1.0 [0.1, 2.4] Significant models 0 41.9 ± 13.6 [23.8, 61.9] 50.5 ± 16.4 [28.6, 66.7] 46.2 ± 10.7 [33.3, 61.9] 0.9 ± 0.4 [0.5, 1.6] 1.3 ± 0.6 [0.6, 2.2] 1.0 ± 1.0 [0.2, 2.6] 0 61.0 ± 17.0 [47.6, 85.7] 31.4 ± 23.2 [4.8, 57.1] 46.2 ± 4.9 [40.5, 52.4] 0.9 ± 0.1 [0.8, 1.1] 1.9 ± 1.1 [0.9, 3.0] 0.7 ± 0.4 [0.3, 1.2] 1 44.9 ± 5.1 [29.7, 56.8] 50.8 ± 8.2 [45.9, 64.9] 47.8 ± 8.8 [37.8, 60.8] 1.0 ± 0.4 [0.6, 1.6] 1.1 ± 0.3 [0.7, 1.5] 1.0 ± 0.8 [0.4, 2.4] 0 10.6 ± 4.9 [5.9, 17.6] 70.6 ± 7.2 [64.7, 82.4] 40.6 ± 2.5 [38.2, 44.1] 0.4 ± 0.1 [0.2, 0.5] 1.3 ± 0.1 [1.1, 1.4] 0.3 ± 0.1 [0.2, 0.4] 0 Statistical significance of the balanced accuracy (BAC) for each bootstrapped sample was tested using permutation testing with a significance level of 5%. Data format: mean ± standard deviation [min max]. DOR, diagnostic odds ratio; NLR, negative likelihood ratio; PLR, positive likelihood ratio; SE, sensitivity; SP, specificity. Significant models: number of models with statistically significant BAC higher than 50%. is being performed in a non-healthy (i.e., non-standard) population, because the effect of external variables such as age and sex might be modulated by the presence of the disease (e.g., being at ARMS or having schizophrenia). Third, this study’s sample is composed of subjects whose clinical diagnosis of an ARMS was based on having a schizotypal personality disorder or on the subject’s familial-high risk coupled with functioning decline and on the CAARMS (54), which mainly evaluates positive symptoms. These were not the same criteria as those used in the previous studies predicting transition to psychosis from an ARMS. These previous studies all used samples of subjects clinically assessed with tools that evaluate not only positive symptoms, but also basic and negative symptoms (10–12, 14, 15), except one (13), which included only familial-high risk subjects in its sample. This potentially increases the inclusion of subjects in the early phase of the psychosis prodrome (characterized by the presence of basic and negative symptoms), whereas our sample includes mainly subjects in the late prodromal phase of psychosis (characterized mainly by the presence of positive symptoms) (2). Therefore, our results suggest that previously reported accuracies in predicting transition to psychosis may be population-specific, poorly generalizable to differently clinically characterized populations (as ours herein). 4.2. Prediction of transition to psychosis using genetic data In this study we applied ML to genetic data and used three types of genetic features to detect transition to psychosis from an ARMS: (a) a schizophrenia PRS that we have previously shown to distinguish Frontiers in Psychiatry 11 frontiersin.org fpsyt-13-1086038 January 13, 2023 Time: 17:35 # 12 Tavares et al. 10.3389/fpsyt.2022.1086038 FIGURE 5 Balanced accuracy across bootstrapped samples for each model trained with the polygenic risk score, the list of psychosis-associated single nucleotide polymorphism (SNPs) or with the list of psychosis-associated genes for which an expression quantitative trait loci (eQTL) score was extracted. Dots represent the balanced accuracy value in each of the 5 bootstrapped samples and are red colored if the balanced accuracy is statistically significant (i.e., p < 0.05) or blue colored if it is not (i.e., p > 0.05). The statistical significance of the balanced accuracy in each bootstrapped sample was evaluated through permutation testing. FIGURE 6 Balanced accuracy across bootstrapped samples for each model trained with the environmental risk score or with each environmental risk factors as features. Dots represent the balanced accuracy value in each of the 5 bootstrapped samples and are red colored if the balanced accuracy is statistically significant (i.e., p < 0.05) or blue colored if it is not (i.e., p > 0.05). The statistical significance of the balanced accuracy in each bootstrapped sample was evaluated through permutation testing. FEP patients from healthy controls (26) and ARMS-T from ARMS- NT (16), (b) a set of psychosis-associated SNPs previously associated with schizophrenia in a recent GWAS meta-analysis (27), and (c) a brain-specific expression Quantitative Trait Loci (eQTL) score including the latter genes. Genetic data showed a poor performance in predicting transition to psychosis from an ARMS. SNPs-based classification models have been previously shown to classify schizophrenia (18, 19, 21), and FEP patients (23) (vs. healthy controls) better than chance level, but not subjects at an ARMS vs. healthy controls or FEP patients (23). Furthermore, one of these studies has selected a list of SNPs from the Psychiatric Genomics Consortium 2 (PGC2) (21, 42), which potentially overlaps with the ones selected in this study (27). Despite the (scarce) evidence of the potential of PRS for schizophrenia (20–22) to classify schizophrenia patients (vs. healthy controls) and the one report showing the schizophrenia PRS’s ability to predict transition to psychosis (16) we were not able to predict transition to psychosis from an ARMS using this type of genetic feature. Although the latter study (16) used a larger sample (i.e., 106% higher than ours, measured as the absolute value of the change in sample size, divided by the average of the size of the two samples) to train the PRS-based model, sample balancing in terms of group and age or sex were not taken into account or that was unclear, respectively. Furthermore, herein we applied a bootstrapped sample approach to estimate generalizability of the PRS-based model by assuring that each bootstrapped sample met the balancing conditions for group, age, and sex–which does not seem to be the case in that study (16). Furthermore, another possible explanation for the PRS negative results is that although the genetic architecture, conveyed through a PRS, has been shown to differ between patients with schizophrenia and healthy controls, one cannot exclude the possibility that it is specific to schizophrenia (a fully developed psychotic disorder), and might even be present in all subjects at an ARMS, i.e., those who later transition to psychosis and those who do not. The constellation of genetic variations (i.e., SNPs) that might confer susceptibility to transition to psychosis already from a prodromal stage is not necessarily the same as the one for schizophrenia (when drawn in comparison to healthy controls). This Frontiers in Psychiatry 12 frontiersin.org fpsyt-13-1086038 January 13, 2023 Time: 17:35 # 13 Tavares et al. 10.3389/fpsyt.2022.1086038 may justify the advantage of using a less hypothesis-based approach for the selection of genetic features (as we did by pre-selecting a large list of SNPs and performing an embedded feature selection using elastic net regression). Lastly, using a PRS formula made specifically for transition to psychosis from an ARMS would require a larger and independent sample to estimate SNP effect sizes, which might be better provided by multicenter projects, such as NAPLS 2 (55) and PRONIA4 over the next years. Expression Quantitative Trait Loci (eQTL) scores for psychosis associated genes expressed in the brain were also not able to predict transition to psychosis from an ARMS. Only one previous study has shown the predictive value of gene expression profiling in the frontal brain region in classifying schizophrenia patients (vs. healthy controls) (17). In the present study, instead of actual gene expression measures we used a proxy for a-genetically regulated component of the expression of genes, the eQTL scores. Although we have computed eQTL scores only for the genes having a validated eQTL score model (43), this does not guarantee that the estimated gene expression represents (or correlates perfectly with) the real levels of the expression. Furthermore, although we have selected the initial list of genes as the ones most associated with schizophrenia (vs. healthy controls), this selection did not take into account the expression profile of these genes in the brain, and we have computed an eQTL score for several brain tissues. A future improvement of this step would be to test an eQTL scores-based model with a selection of genes that: (a) are highly expressed in the brain in healthy subjects, and (b) their expression is associated to a schizophrenia diagnosis, or even better with the transition to psychosis from an ARMS. 4.3. Prediction of transition to psychosis using environmental data In this study we applied, for the first time, ML to environmental data using two types of features to detect transition to psychosis from an ARMS: (a) a schizophrenia ERS which we have previously reported (28), and (b) a set of environmental risk factors as predictors. Overall, neither environmental risk assessment, could predict transition to psychosis from an ARMS with an averaged accuracy, i.e., across bootstrapped samples, better than chance level. Although we know of no similar longitudinal ARMS transition study, the closest other report using ML and environmental data to diagnose schizophrenia (vs. healthy controls) (22) also found a BAC not statistically better than chance level, even having included features such as the presence of obstetric complications and of developmental anomalies, the parental socio-economic status; and –without feature selection– trained and tested the model in a 13 times larger, albeit age, sex, and group -unbalanced, sample (103 patients and 337 controls) than ours (22). However, due to the still poorly understood environmental risk mechanisms one cannot exclude the lack of statistical power as a potential explanation for these negative findings including ours. The ML model trained with the ERS for schizophrenia, which we have tested as an (admittedly limited) exploratory predictor of the transition to psychosis from an ARMS, showed a poor performance, i.e., a BAC similar to chance level. Indeed, ERS is a composite score of individual risk factors computed under the assumption that the risk factors are completely independent (28), which has been shown 4 http://pronia.eu not to be the case (56)–i.e., intercorrelated risk factors may inflate the ERS estimation. This crude approach may limit the ability of the ERS to capture the detailed environmental architecture underlying psychosis. Moreover, just as for a PRS, an ERS for schizophrenia may not be a good substitute of a potential ERS for transition to psychosis from an ARMS (57). Lastly, our criterion for training and testing a fully multimodal ML model with modalities that would show an ML model performance statistically better than chance (i.e., 50%) predicting transition to psychosis from an ARMS in at least 3 of 5 bootstrapped samples was not fulfilled given that none of the modality-based ML models survived that threshold. This conservative criterion was chosen given the already small sample size available for the training of the multimodal ML model, i.e., only 6 ARMS-T and 23 ARMS-NT (only this subset of subjects had data for the three data modalities, simultaneously). The decrease in sample size, remarkably impairs the prediction power of the model, i.e., its accuracy. Without previous evidence of the ability to predict transition to psychosis from an ARMS by modality supporting its integration in a multimodal ML model, negative results from this multimodal model would be highly difficult to explain, as they could theoretically be explained by the increase of noise in the model due to the inclusion of features that did show previous predictive ability or by the lack of predictive power due to the very small sample size. Moreover, the parallel-to-ours, multi-site study, albeit very group-unbalanced (only 26 ARMS-T patients vs. 308 ARMS-NT), from the PRONIA project, showed that a stacked model combining similar data to our study’s plus human prognostic ratings could predict transition to psychosis with a balanced accuracy of 86% and a good geographical generalizability (25). This multimodal approach was showed to improve biological-based unimodal models by 15% (VBGM volume maps-based model) and 20% (PRS for schizophrenia-based model). As such, the replication of this promising finding, following the same multimodal approach as that study, using in our study’s sample and data features co-existing in both samples, would be interesting as an additional method to ascertain whether our negative findings are due to lack of power or to no discriminability with our feature sets. 4.4. Limitations This study was limited by several factors. First, and foremost, the small sample size may have limited the performance of classification models, even though our sample size was informed by previous ML studies showing 74–84% accuracies in predicting transition to psychosis from an ARMS (10–15). Indeed, this is a critical limitation when dealing with high dimensional data, such as neuroimaging and genetics–which we have used herein. Although we have taken measures to avoid overfitting and an overestimation of the classification models’ performance such as artificially increasing the sampling through bootstrapping and employing CV strategies, this might not be enough to overcome this limitation. Indeed, our complementary analysis comparing the models’ training and testing performance (results in the Supplementary material) is indicative that some of the tested classification models (mainly trained with neuroimaging or with SNPs) might suffer from some degree of overfitting. Ultimately, we cannot determine whether our negative findings were due to lack of power to obtain a good performance or due to a true lack of association between the predictors and the transition to psychosis from an ARMS (and hence inflated findings Frontiers in Psychiatry 13 frontiersin.org fpsyt-13-1086038 January 13, 2023 Time: 17:35 # 14 Tavares et al. 10.3389/fpsyt.2022.1086038 from previous studies). This is one of the reasons why replication studies in independent datasets are essential in ML literature. As a final note, a power analysis for this study design would have been the most informative way to define the sample size needed to achieve an accuracy in predicting transition to psychosis from an ARMS better than chance level. However, this is not a trivial task in ML analysis and there is no established method to perform this analysis as there is for univariate analysis [for examples of studies exploring innovative ways of computing sample size for classification problems see Refs. (58, 59)] and, therefore, it was not performed. Second, in order to dilute possible confounding effects in the developed classification models we have restricted the samples used to train the models to: (a) be class-balanced, i.e., with the same number of ARMS-T and ARMS-NT subjects; (b) be matched for age, sex, scanning acquisition protocols for neuroimaging data; (c) include subjects with European ancestry only for genetic data; and (d) limit the proportion of missing data for the environment data. Although this has artificially homogenized the study sample thus avoiding the presence of statistical confounders, it has deemed the sample to be less representative of the ARMS population. Third, overall, the findings of this study are only valid to young help-seeking individuals, i.e., that are clinically screened for ARMS criteria, and whose ARMS diagnosis was based on having a schizotypal personality disorder or on the subject’s familial-high risk coupled with functioning decline and on the CAARMS (54), which mainly evaluates positive symptoms. 5. Conclusion and future directions In this study, we explored the value of using exclusively quantitative and multimodal data (i.e., as predictors) to predict transition to psychosis from an ARMS. Overall, we found that, contrary to what has been previously reported, sMRI could not predict transition to psychosis from an ARMS. We have employed several ML strategies aiming to replicate the highly promising previous positive sMRI findings (74–84%) (10–15). This is even though our sample was larger than four of the above 6 studies (10, 11, 13, 14), respectively (Conversely, our sample was smaller than two of the above studies [Das et al. (15); Koutsouleris et al. (12), respectively]. This points to the need for a cautious interpretation of small sample size studies. Also, we could not replicate the one previous evidence of the value of the schizophrenia PRS in predicting transition to psychosis. Moreover, and to the best of our knowledge, we explored for the first time the value of environment in the prediction of psychosis already from a prodromal stage. Lastly, the genetic and the environmental data used could not predict transition to psychosis from an ARMS. In summary, the present study should serve as a call for caution and skepticism regarding the currently achievable prognostic and diagnostic biomarker development goals, with the existing modeling tools and data measurement tools. Additionally, our study’s methodological approaches tailored to each data modality, may serve as suggestive proofs-of-concept for the exploration of future multimodal datasets, either for novel discovery or replication of previous promising findings, across psychiatric disorders, not exclusive to ARMS. We further suggest larger samples (in the several hundreds) should be employed for both model training and testing, given the inherent high data dimensionality (specially of neuroimaging and genetics) and the little established relevance of individual features. Although still heterogeneity in phenotypic measurements is increased in larger they bring not only statistical power but ecological samples, generalizability, and thus carry a higher potential to be clinically useful. This is best achieved with consortia multi-center studies which are increasingly common albeit not without challenges (60). Alternatively, methods for synthetic generation of data such as the Generative Adversarial Networks (GAN)-based are also a promising avenue for sample size augmentation, now starting to be applied in the clinical research field (61). Last, but not least, we recommend the use of objective and quantitative criteria-based tools for the assessment of a ML biomarker’s clinical applicability, once high effect size and accuracy estimates are achieved, such as one we have previously proposed (62). Data availability statement The datasets presented in this article are not readily available include public data sharing. should be directed to the because ethics approval did not Requests the datasets corresponding author. to access Ethics statement The studies involving human participants were reviewed and approved by NHS South East London Research Ethics Committee. The patients/participants provided their written informed consent to participate in this study. Author contributions VT ran most data preprocessing and statistical analyses and drafted the manuscript. EV coordinated genotyping and advised the genetic and environmental data analysis. AM and HF provided advise on imaging data processing and machine learning analysis. JS and IV collected imaging data. GB provided advise on imaging data processing. DP collected genetic and environmental data, co- designed the study, ran preliminary data preprocessing and machine learning analyses, and supervised the study. All authors revised the manuscript and agreed with its final version. Funding This study, VT received support from Fundação para a Ciência e a Tecnologia (FCT) Ph.D. fellowship PD/BD/114460/2016 and DSAIPA/DS/0065/2018 grants; DP received primary support from National Institute for Health Research (NIHR) PDF-2010-03-047 grant, and additionally from FCT FCT-IF/00787/2014, LISBOA- 01–0145-FEDER-030907, and DSAIPA/DS/0065/2018 grants, and a European Commission (EC) Marie Curie Career Integration Grant (FP7-PEOPLE-2013-CIG 631952). EV was part-funded by the NIHR Maudsley Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and King’s College London. IV was supported by EC’s Horizon 2020 Marie Skłodowska-Curie grant (Ref. 754550, project BITRECS) and “La Caixa” Foundation (LCF/PR/GN18/50310006). Frontiers in Psychiatry 14 frontiersin.org fpsyt-13-1086038 January 13, 2023 Time: 17:35 # 15 Tavares et al. 10.3389/fpsyt.2022.1086038 Acknowledgments We thank Prof. Philip McGuire for his invaluable guidance during data design and collection, the OASIS team, and all volunteers with an ARMS who made this study possible. organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. Conflict of interest Author disclaimer The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer SV declared a shared affiliation with the authors EV, IV, GB, and DP to the handling editor at the time of review. Publisher’s note The views expressed were those of the authors and not necessarily those of any of the above sponsors. Supplementary material All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpsyt.2022.1086038/ full#supplementary-material References 1. Schrimpf L, Aggarwal A, Lauriello J. Psychosis. Contin Lifelong Learn Neurol. (2018) 24:845–60. doi: 10.1212/CON.0000000000000602 2. Fusar-Poli P, Borgwardt S, Bechdolf A, Addington J, Riecher-Rössler A, Schultze- Lutter F, et al. The psychosis high-risk state. JAMA Psychiatry. (2013) 70:107. doi: 10.1001/ jamapsychiatry.2013.269 3. Kahn R, Sommer I, Murray R, Meyer-Lindenberg A, Weinberger D, Cannon T, et al. Schizophrenia. Nat Rev Dis Prim. (2015) 1:15067. doi: 10.1038/nrdp.2015.67 4. Schultze-Lutter F, Ruhrmann S, Fusar-Poli P, Bechdolf AG, Schimmelmann B, Klosterkotter J. Basic symptoms and the prediction of first-episode psychosis. Curr Pharm Des. (2012) 18:351–7. doi: 10.2174/138161212799316064 5. Correll C, Schooler N. Negative symptoms in schizophrenia: A review and clinical guide for recognition, assessment, and treatment. Neuropsychiatr Dis Treat. (2020) 16:519–34. doi: 10.2147/NDT.S225643 6. Sheffield J, Karcher N, Barch D. Cognitive deficits in psychotic disorders: A lifespan perspective. Neuropsychol Rev. (2018) 28:509–33. doi: 10.1007/s11065-018-9388-2 7. Yung A, McGorry P, McFarlane C, Jackson H, Patton G, Rakkar A. Monitoring and care of young people at incipient risk of psychosis. Schizophr Bull. (1996) 22:283–303. doi: 10.1093/schbul/22.2.283 8. Salazar De Pablo G, Radua J, Pereira J, Bonoldi I, Arienti V, Besana F, et al. Probability of transition to psychosis in individuals at clinical high risk: An updated meta-analysis. JAMA Psychiatry. (2021) 78:970–8. doi: 10.1001/jamapsychiatry.2021.0830 9. Fusar-Poli P, Bechdolf A, Taylor M, Bonoldi I, Carpenter W, Yung A, et al. At risk for schizophrenic or affective psychoses? A meta-analysis of DSM/ICD diagnostic outcomes in individuals at high clinical risk. Schizophr Bull. (2013) 39:923–32. doi: 10.1093/schbul/ sbs060 10. Koutsouleris N, Meisenzahl E, Davatzikos C, Bottlender R, Frodl T, Scheuerecker J, et al. Use of neuroanatomical pattern classification to identify subjects in at-risk mental states of psychosis and predict disease transition. Arch Gen Psychiatry. (2009) 66:700–12. doi: 10.1001/archgenpsychiatry.2009.62 11. Koutsouleris N, Borgwardt S, Meisenzahl E, Bottlender R, Möller H, Riecher-Rössler A. Disease prediction in the at-risk mental state for psychosis using neuroanatomical biomarkers: Results from the fepsy study. Schizophr Bull. (2012) 38:1234–46. doi: 10.1093/ schbul/sbr145 12. Koutsouleris N, Riecher-Rössler A, Meisenzahl E, Smieskova R, Studerus E, Kambeitz-Ilankovic L, et al. Detecting the psychosis prodrome across high-risk populations using neuroanatomical biomarkers. Schizophr Bull. (2015) 41:471–82. doi: 10.1093/schbul/sbu078 13. Zarogianni E, Storkey A, Johnstone E, Owens D, Lawrie S. Improved individualized prediction of schizophrenia in subjects at familial high risk, based on neuroanatomical data, schizotypal and neurocognitive features. Schizophr Res. (2017) 181:6–12. doi: 10. 1016/j.schres.2016.08.027 15. Das T, Borgwardt S, Hauke D, Harrisberger F, Lang U, Riecher-Rössler A, et al. Disorganized gyrification network properties during the transition to psychosis. JAMA Psychiatry. (2018) 75:613–22. doi: 10.1001/jamapsychiatry.2018.0391 16. Perkins D, Loohuis L, Barbee J, Ford J, Jeffries C, Addington J, et al. Polygenic risk score contribution to psychosis prediction in a target population of persons at clinical high risk. Am J Psychiatry. (2020) 177:155–63. doi: 10.1176/appi.ajp.2019.18060721 17. Struyf J, Dobrin S, Page D. Combining gene expression, demographic and clinical data in modeling disease: A case study of bipolar disorder and schizophrenia. BMC Genomics. (2008) 9:531. doi: 10.1186/1471-2164-9-531 18. Yang H, Liu J, Sui J, Pearlson G, Calhoun V. A hybrid machine learning method for fusing fMRI and genetic data: Combining both improves classification of schizophrenia. Front Hum Neurosci. (2010) 4:192. doi: 10.3389/fnhum.2010.00192 19. Aguiar-Pulido V, Seoane J, Rabuñal J, Dorado J, Pazos A, Munteanu C. Machine learning techniques for single nucleotide polymorphism - disease classification models in schizophrenia. Molecules. (2010) 15:4875–89. doi: 10.3390/molecules1507 4875 20. Chen J, Wu J, Mize T, Shui D, Chen X. Prediction of schizophrenia diagnosis by integration of genetically correlated conditions and traits. J Neuroimmune Pharmacol. (2018) 13:532–40. doi: 10.1007/s11481-018-9811-8 21. Vivian-Griffiths T, Baker E, Schmidt K, Bracher-Smith M, Walters J, Artemiou A, et al. Predictive modeling of schizophrenia from genomic data: Comparison of polygenic risk score with kernel support vector machines approach. Am J Med Genet Part B Neuropsychiatr Genet. (2019) 180:80–5. doi: 10.1002/ajmg.b.32705 22. Antonucci L, Pergola G, Pigoni A, Dwyer D, Kambeitz-Ilankovic L, Penzel N, et al. A pattern of cognitive deficits stratified for genetic and environmental risk reliably classifies patients with schizophrenia from healthy control subjects. Biol Psychiatry. (2020) 87:697–707. doi: 10.1016/j.biopsych.2019.11.007 23. Pettersson-Yeo W, Benetti S, Marquand A, Dell’Acqua F, Williams S, Allen P, et al. Using genetic, cognitive and multi-modal neuroimaging data to identify ultra-high-risk and first-episode psychosis at the individual level. Psychol Med. (2013) 43:2547–62. doi: 10.1017/S003329171300024X 24. Fromer M, Roussos P, Sieberts S, Johnson J, Kavanagh D, Perumal T, et al. Gene expression elucidates functional impact of polygenic risk for schizophrenia. Nat Neurosci. (2016) 19:1442–53. 25. Koutsouleris N, Dwyer D, Degenhardt F, Maj C, Urquijo-Castro M, Sanfelici R, et al. Multimodal machine learning workflows for prediction of psychosis in patients with clinical high-risk syndromes and recent-onset depression. JAMA Psychiatry. (2021) 78:195–209. 26. Vassos E, Di Forti M, Coleman J, Iyegbe C, Prata D, Euesden J, et al. An examination of polygenic score risk prediction in individuals with first-episode psychosis. Biol Psychiatry. (2017) 81:470–7. doi: 10.1016/j.biopsych.2016.06.028 14. Zarogianni E, Storkey A, Borgwardt S, Smieskova R, Studerus E, Riecher-Rössler A, et al. Individualized prediction of psychosis in subjects with an at-risk mental state. Schizophr Res. (2019) 214:18–23. doi: 10.1016/j.schres.2017.08.061 27. Pardiñas A, Holmans P, Pocklington A, Escott-Price V, Ripke S, Carrera N, et al. Common schizophrenia alleles are enriched in mutation-intolerant genes and in regions under strong background selection. Nat Genet. (2018) 50:381–9. Frontiers in Psychiatry 15 frontiersin.org fpsyt-13-1086038 January 13, 2023 Time: 17:35 # 16 Tavares et al. 10.3389/fpsyt.2022.1086038 28. Vassos E, Sham P, Kempton M, Trotta A, Stilo S, Gayer-Anderson C, et al. The Maudsley environmental risk score for psychosis. Psychol Med. (2020) 50:2213–20. doi: 10.1017/S0033291719002319 45. Hubert M, Rousseeuw P, Vanden Branden K. ROBPCA: A new approach to robust principal component analysis. Technometrics. (2005) 47:64–79. doi: 10.1198/ 004017004000000563 29. Broome M, Woolley J, Johns L, Valmaggia L, Tabraham P, Gafoor R, et al. Outreach and support in south London (OASIS): implementation of a clinical service for prodromal psychosis and the at risk mental state. Eur Psychiatry. (2005) 20:372–8. doi: 10.1016/j. eurpsy.2005.03.001 30. Phillips L, Yung A, McGorry P. Identification of young people at risk of psychosis: validation of Personal Assessment and Crisis Evaluation Clinic intake criteria. Aust N Z J Psychiatry. (2000) 34Suppl:S164–9. doi: 10.1046/j.1440-1614.2000.00798.x 31. Nelson H. The National Adult Reading Test (NART): Test Manual. Wind UK NFER-Nelson. (1982) 124:0–25. 32. American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders, Text Revision (DSM-IV-TR). Fourth ed. (Vol. 1). Arlington, VA: American Psychiatric Association (2000). 33. Murray R, Mondelli V, Stilo S, Trotta A, Sideli L, Ajnakina O, et al. The influence of risk factors on the onset and outcome of psychosis: What we learned from the GAP study. Schizophr Res. (2020) 225:63–8. doi: 10.1016/j.schres.2020.01.011 34. Tavares V, Prata D, Ferreira H. Comparing SPM12 and CAT12 segmentation pipelines: a brain tissue volume-based age and Alzheimer’s disease study. J Neurosci Methods. (2020) 334:108565. doi: 10.1016/j.jneumeth.2019.108565 35. Hammers A, Allom R, Koepp M, Free S, Myers R, Lemieux L, et al. Three- dimensional maximum probability atlas of the human brain, with particular reference to the temporal lobe. Hum Brain Mapp. (2003) 19:224–47. doi: 10.1002/hbm.10123 36. Desikan R, Ségonne F, Fischl B, Quinn B, Dickerson B, Blacker D, et al. An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest. Neuroimage. (2006) 31:968–80. doi: 10.1016/j.neuroimage. 2006.01.021 37. Dahnke, R, Ziegler G, Grosskreutz J, Gaser C. Retrospective Quality Assurance of MR Images. Seattle, WA: Human Brain Mapping (2013). Availale online at: http://dbm.neuro. uni-jena.de/HBM2013/Dahnke01.pdf 38. Bramon E, Pirinen M, Strange A, Lin K, Freeman C, Bellenguez C, et al. A genome-wide association analysis of a broad psychosis phenotype identifies three loci for further investigation. Biol Psychiatry. (2014) 75:386–97. doi: 10.1016/j.biopsych.2013.03 .033 39. Kambeitz-Ilankovic L, Meisenzahl E, Cabral C, von Saldern S, Kambeitz J, Falkai P, et al. Prediction of outcome in the psychosis prodrome using neuroanatomical pattern classification. Schizophr Res. (2016) 173:159–65. doi: 10.1016/j.schres.2015.03 .005 40. Koutsouleris N, Kambeitz-Ilankovic L, Ruhrmann S, Rosen M, Ruef A, Dwyer D, et al. Prediction models of functional outcomes for individuals in the clinical high-risk state for psychosis or with recent-onset depression: A multimodal, multisite machine learning analysis. JAMA Psychiatry. (2018) 75:1156–72. doi: 10.1001/jamapsychiatry.2018. 2165 41. R Core Team. R: A Language and Environment for Statistical Computing. Vienna: R Foundation for Statistical Computing (2014). 42. Ripke S, Neale B, Corvin A, Walters J, Farh K, Holmans P, et al. Biological insights from 108 schizophrenia-associated genetic loci. Nature. (2014) 511:421–7. doi: 10.1038/ nature13595 43. Tavares V, Monteiro J, Vassos E, Coleman J, Prata D. Evaluation of genotype-based gene expression model performance: A cross-framework and cross-dataset study. Genes (Basel). (2021) 12:1531. doi: 10.3390/genes12101531 44. Hubert M, Rousseeuw P, Verdonck T. Robust PCA for skewed data and its outlier map. Comput Stat Data Anal. (2009) 53:2264–74. doi: 10.1016/j.csda.2008.05.027 46. Cortes C, Vapnik V. Support-vector networks. Mach Learn. (1995) 20:273–97. doi: 10.1007/BF00994018 47. Burges C. A tutorial on support vector machines for pattern recognition. Data Min Knowl Discov. (1998) 2:121–67. 48. Zou H, Hastie T. Regularization and variable selection via the elastic net. J R Stat Soc Ser B Stat Methodol. (2005) 67:301–20. doi: 10.1111/j.1467-9868.2005.00503.x 49. Fjell A, Walhovd K. Structural brain changes in aging: courses, causes and cognitive consequences. Rev Neurosci. (2010) 21:187–221. doi: 10.1515/REVNEURO.2010.21.3.187 50. Ruigrok A, Salimi-Khorshidi G, Lai M, Baron-Cohen S, Lombardo MV, Tait R, et al. A meta-analysis of sex differences in human brain structure. Neurosci Biobehav Rev. (2014) 39:34–50. doi: 10.1016/j.neubiorev.2013.12.004 51. Castillejos M, Martín-Pérez C, Moreno-Küstner B. Incidence of psychotic disorders and its association with methodological issues. A systematic review and meta-analyses. Schizophr Res. (2019) 204:458–9. doi: 10.1016/j.schres.2018.07.031 52. Falahati F, Ferreira D, Soininen H, Mecocci P, Vellas B, Tsolaki M, et al. The effect of age correction on multivariate classification in Alzheimer’s Disease, with a focus on the characteristics of incorrectly and correctly classified subjects. Brain Topogr. (2016) 29:296–307. doi: 10.1007/s10548-015-0455-1 53. Wachinger C, Rieckmann A, Pölsterl S. Detect and correct bias in multi-site neuroimaging datasets. Med Image Anal. (2021) 67:101879. doi: 10.1016/j.media.2020. 101879 54. Yung A, Yung A, Pan Yuen H, Mcgorry P, Phillips L, Kelly D, et al. Mapping the onset of psychosis: The comprehensive assessment of at-risk mental states. Aust New Zeal J Psychiatry. (2005) 39:964–71. doi: 10.1080/j.1440-1614.2005.01714.x 55. Addington J, Cadenhead K, Cornblatt B, Mathalon D, McGlashan T, Perkins D, et al. North American Prodrome Longitudinal Study (NAPLS 2): Overview and recruitment. Schizophr Res. (2012) 142:77–82. doi: 10.1016/j.schres.2012.09.012 56. Guloksuz S, Rutten B, Pries L, Ten Have M, De Graaf R, Van Dorsselaer S, et al. The complexities of evaluating the exposome in psychiatry: A data-driven illustration of challenges and some propositions for amendments. Schizophr Bull. (2018) 44:1175–9. doi: 10.1093/schbul/sby118 57. Padmanabhan J, Shah J, Tandon N, Keshavan M. The “polyenviromic risk score”: Aggregating environmental risk factors predicts conversion to psychosis in familial high-risk subjects. Schizophr Res. (2017) 181:17–22. doi: 10.1016/j.schres.2016. 10.014 58. Figueroa R, Zeng-Treitler Q, Kandula S, Ngo L. Predicting sample size required for classification performance. BMC Med Inform Decis Mak. (2012) 12:8. doi: 10.1186/1472- 6947-12-8 59. Dobbin K, Simon R. Sample size planning for developing classifiers using high-dimensional DNA microarray data. Biostatistics. (2007) 8:101–17. doi: 10.1093/ biostatistics/kxj036 60. Tognin S, Van Hell H, Merritt K, Winter-Van Rossum I, Bossong M, Kempton M, et al. Towards precision medicine in psychosis: Benefits and challenges of multimodal multicenter studies - PSYSCAN: Translating neuroimaging findings from research into clinical practice. Schizophr Bull. (2020) 46:432–41. 61. Laino M, Cancian P, Politi L, Della Porta M, Saba L, Savevski V. Generative adversarial networks in brain imaging: A narrative review. J Imaging. (2022) 8:83. doi: 10.3390/jimaging8040083 62. Prata D, Mechelli A, Kapur S. Clinically meaningful biomarkers for psychosis: A systematic and quantitative review. Neurosci Biobehav Rev. (2014) 45:134–41. doi: 10. 1016/j.neubiorev.2014.05.010 Frontiers in Psychiatry 16 frontiersin.org
Could not heal snippet
10.3390/ijerph19127064
Data Availability Statement: The code and data underlying this article will be shared on reasonable request to the corresponding author.
Data Availability Statement: The code and data underlying this article will be shared on reasonable request to the corresponding author.
Article The Impact of Influencers on Cigar Promotions: A Content Analysis of Large Cigar and Swisher Sweets Videos on TikTok Jiaxi Wu 1 Traci Hong 1 and Jessica L. Fetterman 9,* , Alyssa F. Harlow 2, Derry Wijaya 3, Micah Berman 4, Emelia J. Benjamin 5,6,7 , Ziming Xuan 8, 1 College of Communication, Boston University, Boston, MA 02215, USA; jiaxiw@bu.edu (J.W.); tjhong@bu.edu (T.H.) 2 Department of Population and Public Health Sciences, Keck School of Medicine, University of Southern California, Los Angeles, CA 90032, USA; afharlow@usc.edu 3 Department of Computer Science, Boston University, Boston, MA 02215, USA; wijaya@bu.edu 4 College of Public Health & Moritz College of Law, The Ohio State University, Columbus, OH 43210, USA; berman.31@osu.edu 5 National Heart, Lung, and Blood Institute’s Framingham Heart Study, Framingham, MA 20892, USA; 6 emelia@bu.edu Section of Cardiovascular Medicine, Boston Medical Center, Department of Medicine, School of Medicine, Boston University, Boston, MA 02118, USA 7 Department of Epidemiology, School of Public Health, Boston University, Boston, MA 02118, USA 8 Department of Community Health Sciences, School of Public Health, Boston University, 9 Boston, MA 02118, USA; zxuan@bu.edu Evans Department of Medicine, Whitaker Cardiovascular Institute, School of Medicine, Boston University, Boston, MA 02118, USA * Correspondence: jefetter@bu.edu; Tel.: +1-617-358-7544 Citation: Wu, J.; Harlow, A.F.; Wijaya, D.; Berman, M.; Benjamin, E.J.; Xuan, Z.; Hong, T.; Fetterman, J.L. The Impact of Influencers on Cigar Promotions: A Content Analysis of Large Cigar and Swisher Sweets Videos on TikTok. Int. J. Environ. Res. Public Health 2022, 19, 7064. https://doi.org/10.3390/ ijerph19127064 Academic Editor: David Berrigan Received: 16 March 2022 Accepted: 7 June 2022 Published: 9 June 2022 Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affil- iations. Abstract: Little is known about the content, promotions, and individuals in cigar-related videos on TikTok. TikTok videos with large cigar and Swisher Sweets-related hashtags between July 2016 and September 2020 were analyzed. Follower count was used to identify influencers. We compared con- tent characteristics and demographics of featured individuals between cigar types, and by influencer status. We also examined the association between content characteristics and video engagement. Compared to large cigar videos, Swisher Sweets videos were more likely to feature arts and crafts with cigar packages, cannabis use, and flavored products. In addition, Swisher Sweets videos were also more likely to feature females, Black individuals, and younger individuals. Both Swisher Sweets and large cigar influencers posted more videos of cigar purchasing behaviors than non-influencers, which was associated with more video views. None of the videos disclosed sponsorship with #ad or #sponsored. Videos containing the use of cigar packages for arts and crafts, and flavored products highlight the importance of colorful packaging and flavors in the appeal of Swisher Sweets cigars, lending support for plain packaging requirements and the prohibition of flavors in cigar products to decrease the appeal of cigars. The presence and broad reach of cigar promotions on TikTok requires stricter enforcement of anti-tobacco promotion policies. Keywords: promotion; TikTok cigars; little cigars; flavored cigars; Swisher Sweets; social media; influencer Copyright: © 2022 by the authors. 1. Introduction Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/). In 2020, cigars were the most used combustible tobacco product with a higher preva- lence of use than cigarettes among US youth [1]. Between 2009 and 2020, the increase in US cigar sales was primarily driven by the sale of flavored cigars [2]. Swisher Sweets, a leading flavored cigar brand, accounted for over 22% of market shares in US Convenience Stores in 2020 [3]. Factors that contribute to the increased cigar use among youth and young adults include the availability of flavors [4], small pack sizes [5], the industry’s targeted Int. J. Environ. Res. Public Health 2022, 19, 7064. https://doi.org/10.3390/ijerph19127064 https://www.mdpi.com/journal/ijerph International Journal ofEnvironmental Researchand Public Health Int. J. Environ. Res. Public Health 2022, 19, 7064 2 of 14 marketing [6], features that facilitate cannabis use [7], psychosocial factors [8], and reduced health risk perceptions of cigar smoking compared to cigarette smoking [9]. Cigars are broadly categorized into three types: large cigars, cigarillos, and little cigars [10]. The large cigar category includes both premium hand-rolled and machine- made large cigars. Cigarillos are short and narrow cigars that usually do not contain a filter and are available in a wide array of flavors. Little cigars are similar in size and shape to cigarettes and typically contain a filter. Historically, cigars were primarily premium cigars and were used mainly by older, predominantly white men who reported infrequent use and no inhalation [11]. As a result, cigars are not as heavily regulated and taxed as cigarettes [12]. Cigar companies promote product features prohibited in cigarettes, such as flavors and small pack sizes [13]. Regulatory loopholes have resulted in the generation of cigar products, specifically flavored little cigars and cigarillos (LCCs), designed to appeal to youth, young adults, and individuals of low-socioeconomic status [12]. As a result, the demographic characteristics of users, cigar product use patterns, purchasing behaviors, and reasons for use vary by cigar type [14]. Compared to users of premium large cigars, users of LCCs tend to be younger, non-Hispanic Black, have low educational attainment, and have low income. People who smoked LCCs are also more likely to report the use of a flavored, commonly used brand cigar than individuals who used premium cigars. Thus, it is critical to distinguish between people who use premium large cigars from people who use flavored LCCs, and to develop targeted public health and intervention efforts toward users of different cigar products [15]. Cigar smoking in the US also presents a critical health equity issue. Cigar companies use targeted strategies to promote products to communities of color [16]. As a result, the prevalence of cigar use is higher among non-Hispanic Black individuals than among other racial/ethnic groups [15]. In addition, Black youth are more likely to initiate tobacco use with cigars compared to White youth [17]. Recent surveillance data also indicate a higher prevalence of cigar use among high school students who are Black, compared to White and Hispanic youth [1]. Social media use is pervasive among youth, with 85% of youth using at least one social media site, and 45% say they are constantly online [18]. Social media has given rise to a class of non-traditional celebrities called influencers, who have large online followings and are valued as opinion leaders [19]. Influencers promoting tobacco products can potentially affect their followers’ attitudes and use of tobacco products. Followers of tobacco influencers are more likely to be an especially vulnerable group because they tend to be younger, have lower education, and are more likely to report past month tobacco use than those who do not follow tobacco influencers [20]. The positive association between exposure and engagement with tobacco-related social media content and tobacco use among youth warrants investigation into the prevalence of tobacco-related influencer promotions on social media [21]. Social learning theory provides a basis for explaining the effects of exposure to tobacco- related influencer posts on youth tobacco use [22]. Social learning theory posits that people acquire behaviors through observing, modeling, and imitating the behaviors of others [22]. Individuals who are observed are referred to as models. In real life, young people are surrounded by different types of influential models, from parents and teachers to peers. In contrast, on social media, models for behaviors are more limited, primarily to influencers who are often paid for sponsorships of products and services. Thus, by observing examples of behavior through social media, people, especially youth, are more likely to adopt the attitudes and behaviors exhibited by the influencer [23]. TikTok is the fastest-growing social media platform in the world and was the most downloaded mobile app in 2021 [24]. TikTok is especially popular among youth, with those 14 years and younger accounting for more than a third of TikTok’s 49 million daily users in the US [25]. On TikTok, users are continuously provided with new content, as videos start automatically one after another, unprompted [26]. On average, users spend Int. J. Environ. Res. Public Health 2022, 19, 7064 3 of 14 52 min on the platform and consume more than 200 videos per day, including carefully targeted ads [27]. TikTok’s user guidelines prohibit the posting of “content that depicts minors consuming, possessing, or suspected of consuming alcoholic beverages, drugs, or tobacco”, and the advertising or trade of tobacco products is also prohibited [28,29]; however, previous research found widespread tobacco promotions on Facebook, despite the platform’s policies prohibiting tobacco advertising [30]. It is unknown if TikTok’s anti-tobacco promotion policies are enforced, warranting further examination. The goal of the current study was to examine the portrayals and promotions of large cigars and LCCs on TikTok, which is largely unexplored. Specifically, we compared the content features and individuals in videos of the different cigar types, and between influencer and non-influencer cigar videos. Our analysis of cigar-related videos on TikTok adds to the literature describing the differences in the users and use patterns of different types of cigar products. We also examined the association between content features and video popularity on TikTok. 2. Materials and Methods 2.1. Data Collection We searched for different large cigar and LCC-related hashtags on TikTok to identify the most representative hashtags for cigar-related videos. Our initial observation on TikTok revealed that the hashtags “#cigar” and “#cigars” were commonly used in videos of traditional large cigars (see Supplemental Table S1 for details on the number of videos identified by hash- tag). We also searched hashtags #littlecigar, #littlecigars, #cigarillo, and #cigarillos to identify LCC-related videos; however, these LCC-related hashtags were rarely used by TikTok users. To identify content on TikTok related to LCCs, we extended our search to include Swisher Sweets-related hashtags (“#swisher”, “#swishers”, “#swishersweet”, “#swishersweets”), given that LCC users are more familiar with brand names rather than the terms “cigarillos” or “little cigars” [31] and are more likely to mention specific brands when posting about LCCs [32]. We found that Swisher Sweets-related hashtags were viewed over 16 million times on TikTok. Because Swisher Sweets is the leading cigar brand in the US for LCCs [33], is preferred among the US youth and young adults [34], and is commonly used as a keyword in previous social media research of LCCs [35–38], we only used Swisher Sweets-related hashtags to retrieve LCC-related TikTok videos for the current study. TikTok’s Terms of Service [39] prohibits the use of public videos for commercial purposes, which was not the intent of this study. On September 17, 2020, using an open- source TikTok scraping tool [40], we scraped all 4361 publicly available videos with cigar and Swisher Sweets-related hashtags ever posted on TikTok. Pre-determined large cigar and Swisher Sweets hashtags were used to identify all publicly available TikTok videos that contained those hashtags up to the scraping date. We scraped: (1) 3456 videos with cigar-related hashtags and (2) 905 videos with Swisher Sweets-related hashtags. We also collected the associated metadata, including numbers of video views, likes, shares, and follower counts. Data for this study were stored in a password-protected computer and were only accessible to the authors. Research procedures were deemed to not meet the definition of human subjects research by the Authors’ Institutional Review Board due to the use of publicly available data. 2.2. Sample Previous research has defined influencers as individuals with a minimum of 1000 followers [37]. For our study, we defined influencers as those with the top 75th percentile of total followers within each hashtag dataset because the follower counts varied between the large cigar and Swisher Sweets videos. Because people who smoke large cigars and Swisher Sweets differ in demographic features, cigar-smoking patterns, and purchasing behaviors [14], we sampled influencers for large cigar and Swisher Sweets videos separately; thus, instead of using a fixed number, which may cause biased sampling, Int. J. Environ. Res. Public Health 2022, 19, 7064 4 of 14 we used the 75th percentile of followers within each hashtag to ensure we identified the top influential users for each of the two cigar types. The minimum number of followers for an influencer was 15,000 and 1759 for large cigar videos and Swisher Sweets videos, respectively. We included all videos from influencers and randomly selected the matched numbers of videos from non-influencers for each of the respective hashtag categories. We also excluded non-English and non-relevant videos, resulting in a final sample of N = 1700 videos (1333 cigar videos and 367 Swisher Sweets videos, Figure 1). Figure 1. Data sampling procedure. Influencers’ videos were identified by defining influencers as those individuals with the top 75th quantile of number of followers within each hashtag category. The same number of non-influencers’ videos were randomly sampled for each hashtag category. The final sample of large cigar and Swisher Sweets videos was N = 1700 after excluding non-English videos. 2.3. Video Coding and Inter-Coder Reliability We created a coding scheme of fifteen content features identified from previous studies and that emerged from the current dataset. Specifically, we identified nine themes from previous social media analyses of cigar and LCC-related posts including: (1) Product promotion [35]; (2) Smoking cigars [32,35]; (3) Cannabis use (e.g., removing some, or all the tobacco from the cigar and replacing it with cannabis, Figure 2A) [35]; (4) Smoke trick [35]; (5) Prevention [32]; (6) Flavor [38]; (7) Purchasing behaviors (Figure 2B) [38]; (8) Product review (Figure 2C) [41]; and (9) Cigar-related marketing events (Figure 2D) [36]. We further identified six themes observed from the current dataset including: (10) Arts and crafts with cigar packages (Figure 2E); (11) Individuals dancing with background music referring to Int. J. Environ. Res. Public Health 2022, 19, x 4 of 15 behaviors [14], we sampled influencers for large cigar and Swisher Sweets videos separately; thus, instead of using a fixed number, which may cause biased sampling, we used the 75th percentile of followers within each hashtag to ensure we identified the top influential users for each of the two cigar types. The minimum number of followers for an influencer was 15,000 and 1759 for large cigar videos and Swisher Sweets videos, respectively. We included all videos from influencers and randomly selected the matched numbers of videos from non-influencers for each of the respective hashtag categories. We also excluded non-English and non-relevant videos, resulting in a final sample of N = 1700 videos (1333 cigar videos and 367 Swisher Sweets videos, Figure 1). Figure 1. Data sampling procedure. Influencers’ videos were identified by defining influencers as those individuals with the top 75th quantile of number of followers within each hashtag category. The same number of non-influencers’ videos were randomly sampled for each hashtag category. The final sample of large cigar and Swisher Sweets videos was N = 1700 after excluding non-English videos. 2.3. Video Coding and Inter-Coder Reliability We created a coding scheme of fifteen content features identified from previous studies and that emerged from the current dataset. Specifically, we identified nine themes from previous social media analyses of cigar and LCC-related posts including: (1) Product promotion [35]; (2) Smoking cigars [32,35]; (3) Cannabis use (e.g., removing some, or all the tobacco from the cigar and replacing it with cannabis, Figure 2A) [35]; (4) Smoke trick [35]; (5) Prevention [32]; (6) Flavor [38]; (7) Purchasing behaviors (Figure 2B) [38]; (8) Int. J. Environ. Res. Public Health 2022, 19, 7064 5 of 14 cigar smoking (making smoking gestures); (12) Use of the Cardi B Swisher Sweets song as video background music; (13) Showing off multiple Swisher Sweets packages (Figure 2F). We additionally coded: (14) Written health warnings; and (15) Audio health warnings based upon previous research suggesting that health warnings in tobacco-related social media posts results in a more negative tobacco brand perception [42]. Table 1 displays descriptions of the coded video content features. Video content features were not mutually exclusive, meaning that a video could contain multiple content features simultaneously. Figure 2. Representative images of select video themes. (A) Cannabis use: a cigar that has been hollowed out and filled with cannabis; (B) Purchasing behavior: a video of a user purchasing cigar products in a convenience store; (C) Product review: a video review of the “double corona” large cigar; (D) Cigar-related marketing events: a video of a Swisher Sweets marketing event; (E) Arts and crafts: a video of paraphernalia emblazoned with the Swisher Sweets logo through arts and crafts; (F) Showing off multiple Swisher Sweets packages: a video of one individual showing off all of the Swisher Sweets packages the person has smoked. Int. J. Environ. Res. Public Health 2022, 19, x 5 of 15 Product review (Figure 2C) [41]; and (9) Cigar-related marketing events (Figure 2D) [36]. We further identified six themes observed from the current dataset including: (10) Arts and crafts with cigar packages (Figure 2E); (11) Individuals dancing with background music referring to cigar smoking (making smoking gestures); (12) Use of the Cardi B Swisher Sweets song as video background music; (13) Showing off multiple Swisher Sweets packages (Figure 2F). We additionally coded: (14) Written health warnings; and (15) Audio health warnings based upon previous research suggesting that health warnings in tobacco-related social media posts results in a more negative tobacco brand perception [42]. Table 1 displays descriptions of the coded video content features. Video content features were not mutually exclusive, meaning that a video could contain multiple content features simultaneously. Figure 2. Representative images of select video themes. (A) Cannabis use: a cigar that has been hollowed out and filled with cannabis; (B) Purchasing behavior: a video of a user purchasing cigar products in a convenience store; (C) Product review: a video review of the “double corona” large cigar; (D) Cigar-related marketing events: a video of a Swisher Sweets marketing event; (E) Arts and crafts: a video of paraphernalia emblazoned with the Swisher Sweets logo through arts and crafts; (F) Showing off multiple Swisher Sweets packages: a video of one individual showing off all of the Swisher Sweets packages the person has smoked. Int. J. Environ. Res. Public Health 2022, 19, 7064 6 of 14 Table 1. Descriptions of content features in English and relevant cigar/Swisher Sweets videos. Video Content Features Product Promotion Smoking Cigars Cannabis Use Smoke Trick Prevention Flavor Purchasing Behavior Product Review Cigar-related Marketing Events Arts and Crafts with Cigar Packages Individual Dancing Cardi B Swisher Sweets Music Showing off multiple Swisher Sweets packages Written Health Warnings Audio Health Warnings Sex Presence of Males Presence of Females Race Black White Spanish/Hispanic Asian Presence of Young Individuals Descriptions A video selling cigar products or promoting cigar stores, and professional ads A video showing individuals smoking featured cigar products A video of cannabis use (e.g., blunt: a cigar that has been hollowed out and filled with cannabis) A video of smoke tricks such as making smoke rings A video with a main theme of the negative effects and prevention of cigar and LCC products A video showing or referring to flavored cigar products A video of accessing and purchasing cigar or LCC products A video commenting on or reviewing flavors, tastes, or features of cigar or LCC products A video promoting cigar companies’ marketing events (e.g., musical events) A video of using cigar/LCC products’ packages to create arts and crafts such as a rolling tray A video of people dancing and making smoking gestures without smoking real cigar products A video using Cardi B’s “Swisher music” as background music A video showing more than five Swisher Sweets packages simultaneously A video containing written form of health warnings or disclaimers of cigar/LCC smoking superimposed on the video A video containing oral form of health warnings or disclaimers of cigar/LCC smoking A video featuring males A video featuring females A video featuring Black individuals A video featuring White individuals A video featuring Spanish/Hispanic individuals A video featuring Asian individuals A video featuring individuals who look like teens in middle/high school to people who are under the age of 21 For videos containing people, we also coded for the following demographic features: (1) perceived sex—the presence of males and females; (2) perceived race—the presence of Black, White, Asian, and Hispanic or Latino individuals; and (3) perceived age—the presence of younger or older individuals. Coders used all available visual and audial cues (e.g., skin color, background voices, appearances) to inform their coding choices. Consistent with a previous study that coded social media users’ age from their profile pictures [32], we assigned “younger” to individuals who look like they were under the age of 21 (i.e., individuals who look like teens in middle or high school to young adults under the age of 21) and older individuals (≥21 years of age) based on visual and audio cues in the videos. In the event demographic features were difficult to determine, coders could select “unknown” for the sex/race/age of featured individuals if none of the visual and audial cues were available to determine the demographics. To attain high coding reliability, three coders were first trained on 20 videos together to ensure the visual/audio cues used to code the content features and demographic categories were consistent across all coders. Next, three coders independently coded 50 videos, after which discrepancies were discussed to resolve coding disagreements. Coders first determined whether the video was in English and relevant to cigars and Swisher Sweets. A total of 280 videos (273 large cigar and 7 Swisher Sweets videos) were not in the English language and were excluded from the analyses. A video was considered relevant only when the video showed or referred to cigar smoking. An example of non-relevant videos included videos of Swisher brand lawn mowers. Next, following the coding scheme, coders Int. J. Environ. Res. Public Health 2022, 19, 7064 7 of 14 identified the content features and demographics of individuals in the videos. The inter- coder reliability was calculated using 10% (N = 221) of the sample. Coding agreements were assessed with Cohen’s Kappa values, which were above 0.7 across all content variables, indicating a high level of intercoder reliability [43]. Three coders independently coded the remaining videos in the sample. 2.4. Statistical Analyses We performed chi-square tests to compare the content features and individual char- acteristics for large cigar and Swisher Sweets videos, between influencer’s and non- influencer’s videos within each of the cigar types, and between large cigars and Swisher Sweets influencers’ videos. When comparing large cigars and Swisher Sweets videos, we excluded the content features (1) use of the Cardi B Swisher Sweets song as video back- ground music” and (2) showing off multiple Swisher Sweets packages from the analysis because these features were unique to Swisher Sweets videos. We additionally calculated odds ratios and 95% confidence intervals. Chi-square analyses were conducted using SPSS (Version 26) with an alpha level of 0.05 (2-tailed) with Bonferroni correction to account for multiple testing. 2.5. Modeling Video Engagement with Video Content Features To identify the video content features associated with engagement (i.e., number of views, likes, and shares) of a large cigar or Swisher Sweets videos, we formulated negative binomial models for views and likes and negative binomial hurdle models for shares within each of the large cigar and Swisher Sweets datasets (see SI Section 2 for modeling details). For each of the main models predicting video popularity (i.e., views, likes, shares), we used video content features as predictors. Given that some content features were rare in the data sets, we only included content features that appeared more than ten times within each dataset. We also adjusted for follower counts, which can potentially affect the engagement of social media posts, including videos on TikTok. Negative binomial and hurdle models were fitted using the glmmTMB package in R (version 4.1.0). A p-value less than 0.05 (two-tailed) was considered statistically significant. The p-values of each negative binomial and hurdle model were adjusted with the Bonferroni correction method. 2.6. Hashtag Analysis of Video Descriptions To prohibit misleading or deceptive advertising, the Federal Trade Commission (FTC) requires that any type of online sponsored content must clearly disclose sponsorship [44]. For each of the video post descriptions, we analyzed whether the video description con- tained the FTC recommended sponsorship disclosure hashtags #ad and #sponsored [44]. The FTC requires disclosure of any financial relationship to the brand (including the provi- sion of free products) and suggests the use of hashtags as one method of disclosure [45]. We used string matching techniques in R (Version 4.1.0) to determine if the description of a post for a video contained either of the two FTC recommended hashtags #ad and #sponsored. 3. Results At the time of the study, the 1700 videos with large cigar and Swisher Sweets-related hashtags had been viewed over 159 million times on TikTok. The median follower counts for large cigar and Swisher Sweets hashtag videos were 14,900 and 1740, respec- tively. With respect to video engagement, influencers’ cigar videos received an average of 88,749 views, 4455 likes, and 41 shares; non-influencers’ cigar videos attracted an average of 8718 views, 304 likes, and 7 shares. Influencers’ Swisher Sweets videos received an average of 45,625 views, 4962 likes, and 116 shares; non-influencers’ Swisher Sweets videos elicited an average of 2434 views, 150 likes, and 10 shares (See Supplemental Table S2 for details of video engagement for the two cigar types). For large cigar videos, the top two prevalent video themes were smoking cigars (59.2%) and product review (13.5%). For Swisher Sweets videos, the top two themes were flavor (48.8%) and cannabis use (28.9%) Int. J. Environ. Res. Public Health 2022, 19, 7064 8 of 14 (See Supplemental Table S3). Coders also determined if Swisher Sweets LCCs or packaging were included in large cigar videos. Coders found that none of the 1333 sampled large cigar videos contained a Swisher Sweets LCC product or packaging, which is consistent with previous research that suggests users of LLC often mentioned specific brands when posting about LCCs, instead of using general tobacco product terms such as “cigar” and “cigars” [32]. 3.1. Comparisons of Video Content Features and Featured Individual Demographics Chi-square analyses comparing large cigar and Swisher Sweets videos indicated that large cigar videos contained more product promotions (p = 0.005), product reviews (p < 0.001), and individuals smoking cigars (p < 0.001) compared to Swisher Sweets videos. Swisher Sweets videos contained more videos of purchasing behaviors (p < 0.001), arts and crafts with cigar packages (p < 0.001), cannabis use (p < 0.001), dancing individuals making smoking gestures (p < 0.001), and flavored products (p < 0.001) compared to large cigar videos (See Supplemental Table S3). Large cigar videos were more likely to feature males (p < 0.001), while Swisher Sweets videos were more likely to show females (p < 0.001). In addition, large cigar videos contained more White individuals than Swisher Sweets videos (p < 0.001). In contrast, Swisher Sweets videos were associated with Black (p = 0.017), Asian (p = 0.001), and younger individuals (p < 0.001) (See Supplemental Table S4). Compared to large cigar non-influencers, large cigar influencers were more likely to post about purchasing behaviors (p = 0.025) and product reviews (p < 0.001). For Swisher Sweets videos, influencers were also more likely to post purchasing behaviors content (p < 0.001), dancing individuals (p < 0.001), and flavored products (p = 0.041) compared to Swisher Sweets non-influencers (See Supplemental Table S5). Interestingly, large cigar influencers were less likely to be younger compared to the non-influencers (p = 0.009), while the opposite association was observed for Swisher Sweets influencers, who were more likely to be younger than non-influencers (p = 0.002) (See Supplemental Table S6). Lastly, Chi-square analyses comparing large cigar influencer posts and Swisher Sweets influencer posts suggested that large cigar influencers were more likely to post product promotions (p = 0.020), product reviews (p < 0.001), and smoking individuals (p < 0.001); however, Swisher Sweets influencers were more likely to post videos of purchasing behav- iors (p < 0.001), arts and crafts with cigar packages (p < 0.001), cannabis use (p < 0.001), dancing individuals making smoking gestures (p < 0.001), and flavored products (p < 0.001) (See Supplemental Table S7). Lastly, we found large cigar influencers were more likely to be males (p < 0.001) and White individuals than Swisher Sweets influencers (p < 0.001). On the contrary, when compared to large cigar influencers, Swisher Sweets influencers were more likely to be females (p < 0.001), Asian (p < 0.001), and young individuals (p < 0.001) (See Supplemental Table S8). 3.2. Predicting Video Popularity with Video Content Features When interpreting results of a predictor from negative binomial and Hurdle models, all other predictors are held constant. For large cigar videos, negative binomial models showed that after adjusting for the number of account followers and other content feature predictors, video content of purchasing behaviors (IRR = 29.03, p < 0.001, CI = 17.03, 49.47) and product reviews (IRR = 2.50, p < 0.001, CI = 1.98, 3.17) elicited more likes. Purchasing behaviors (IRR = 20.92, p < 0.001, CI = 11.98, 36.53) and product reviews (IRR = 2.03, p < 0.001, CI = 1.58, 2.61) also had a greater number of views compared to videos without these content features. We found that product promotions (IRR = 0.30, p < 0.001, CI = 0.21, 0.44) and content of an individual smoking cigars (IRR = 0.65, p < 0.001, CI = 0.54, 0.77) was associated with fewer likes. Product promotions (IRR = 0.32, p < 0.001, CI = 0.22, 0.47) and content of individuals smoking cigars (IRR = 0.56, p < 0.001, CI = 0.46, 0.67) were also associated with fewer views (See Supplemental Table S9). As for shares, the zero portion of Hurdle models revealed no significant results for any of the content features. The positive portion of Hurdle model suggested that videos of purchasing behaviors (IRR = 44.70, p < 0.001, CI = 9.13, 218.86) Int. J. Environ. Res. Public Health 2022, 19, 7064 9 of 14 predicted more shares. Consistent with likes and views, product promotions (IRR = 0.29, p = 0.019, CI = 0.12, 0.74) and content of individuals smoking cigars (IRR = 0.47, p = 0.007, CI = 0.29, 0.77) were associated with fewer shares (See Supplemental Table S10). For Swisher Sweets videos, negative binomial analyses showed that, after adjusting for the number of account followers and other content feature predictors, videos that contained content on purchasing behavior received more video views (IRR = 2.96, p = 0.036, CI = 1.26, 7.00); however, video content of an individual smoking cigars was associated with fewer video likes (IRR = 0.37, p = 0.005, CI = 0.21, 0.68) and fewer video views (IRR = 0.32, p = 0.001, CI = 0.18, 0.59) (See Supplemental Table S11). When predicting video shares, none of the content features were associated with video shares in the zero portion of the model. The non-zero portion Hurdle model showed that smoking cigars also led to fewer shares of a Swisher Sweets post on TikTok (IRR = 0.01, p = 0.008, CI = 0.00, 0.17) (See Supplemental Table S12). 3.3. Disclosure of Sponsorship in Video Description Text analyses of all of the 1700 cigars and Swisher Sweets video descriptions showed that none of the video descriptions contained hashtags that disclosed sponsorship, including #ad and #sponsored. Only three large cigar videos contained oral health warnings or disclaimers. 4. Discussion The demographic characteristics, including age, sex, and race/ethnicity, differ be- tween users of large cigars and LCCs [14]. Research suggests that socially disadvantaged communities, especially non-Hispanic Black individuals, are more likely to smoke cigars and to develop established cigar-smoking behaviors compared with non-Hispanic White individuals [46]. Our findings confirmed the literature on the demographic differences between users posting about large cigars or Swisher Sweets on TikTok. We found that Swisher Sweets TikTok videos were more likely to feature females, younger, Black, and Asian individuals compared to large cigar TikTok videos. When comparing between in- fluencers and non-influencers’ videos, we found that compared to large cigar influencers, Swisher Sweets influencers tend to be younger and were more likely to be female. Future prevention campaigns for cigar products (i.e., large cigars and LCCs) should consider the different use patterns and demographics of users. Swisher Sweets is the leading LCC brand popular among youth and young adults in the US [3]. Our study sheds light on the Swisher Sweets-related content that youth and young adults may be exposed to on TikTok, the fastest growing social media platform in the world that appeals to a younger population [24,25]. Future studies are needed to evaluate the content of videos of additional cigar brands to gain a more comprehensive understanding of LCC influencer promotions on TikTok. Consistent with previous research across other social media platforms [35,37], our study noted that Swisher Sweets videos on TikTok were more likely to feature cannabis use and flavored products compared to videos featuring large cigars [35,38]. In addition, our study found that both large cigar and Swisher Sweets influencers were more likely to post about purchasing behaviors than non-influencer users. Large cigar influencers also posted product review videos more frequently than non-influencer users—a common promotional strategy of large cigars observed in traditional media such as magazines [47]. Our findings suggest that content of purchasing behavior and product reviews lead to more views of large cigar and Swisher Sweets videos. Future research is needed to investigate how exposure to short-form videos of cigar-related content on TikTok affect youth’s attitudes towards cigars and their susceptibilities to initiate and use cigars. Exposure to celebrity-endorsed LCC promotions correlates with brand-specific cigar smoking among young adult smokers [48]. We found that one out of ten Swisher Sweets videos on TikTok used Cardi B’s “Swisher Sweets” song as the background music, suggesting that celebrity celebration of tobacco products can echo widely on social media. Moreover, our study also revealed that TikTok users, through arts and crafts, are re-purposing Swisher Int. J. Environ. Res. Public Health 2022, 19, 7064 10 of 14 Sweets packages to streetwear and paraphernalia emblazoned with the Swisher Sweets logo. Many LCC brands, including Swisher Sweets, use colorful and shiny packages that boldly communicate the flavor appeal to youth and young adults. Visual cues from the cigar packaging impact young adults’ affect and increase their susceptibility to cigar smoking [49]. Regulations on the packaging of flavored cigars, including the use of pictorial warning labels, plain packaging, and restrictions on celebrity endorsement and other youth appealing promotional content may decrease the appeal and use of cigar products among youth. To prohibit misleading or deceptive advertising, the FTC requires that any type of online sponsored content must clearly disclose sponsorship. For sponsored social media posts, the FTC recommends the use of clear hashtags such as “#ad” and “#sponsored”, or other similarly clear methods to disclose sponsorship [44]. None of the videos in our analyzed data set contained hashtags indicating sponsorship. We acknowledge that we do not have access to data regarding a financial transaction between the manufacturers and the influencers; however, given that many of the influencers in our dataset were retailers and cigar manufacturers, the promotional regulations could also be applied to those users. Future research may examine the prevalence and potential effects of exposure to promotions from retailers and cigar manufacturers on purchasing intentions and behaviors. Research has indicated that labeling commercially sponsored tobacco content on social media with #ad and #sponsored attracts the attention of youth and young adults, making it a viable strategy to inform audiences of the promotional nature of the posts [50]. Sponsorship disclosure promotes critical evaluation of the promotional post and ultimately decreases advertising effectiveness [51]. A study found that clear sponsorship disclosure decreased young adult participants’ perceptions of influencer credibility and their intentions to engage with e-cigarette Instagram posts [52]. Examining the effects of influencer posts and regulatory interventions in influencer promotions on youth tobacco experimentation is an important direction for future research. Even though TikTok prohibits the promotion of tobacco products and the posting of minors consuming tobacco products [29], we still observed such content on TikTok. Exposure and engagement with tobacco-related social media content are associated with tobacco use among youth [21]. Because more than a third of TikTok’s daily users are under the age of 14 [25], it is crucial to restrict the promotion of youth-appealing tobacco content on TikTok to reduce the effects of such promotions on tobacco use among youth. Similar to a study that found that few, if any, influencers’ cigar-related posts on Twitter contained health warnings [37], we observed that only three of all large cigar and Swisher Sweets promotional videos contained audio health warnings or disclaimers, and no written health warning labels were observed. Prior research has reported that the inclusion of health warning statements in tobacco-related social media posts results in more negative brand perceptions [52]; therefore, from a public health perspective, it is important to enforce disclosures of sponsorship and health warnings on all content promoting tobacco products and to remove any content featuring the use of tobacco by underage users. Our study has several limitations. Our study was cross-sectional and observational and hence, we cannot rule out residual confounding or establish causation. We also cannot exclude misclassification of the features we coded. For instance, identification of an individual’s demographic features was by external appearance and auditory cues, which may not be as accurate as self-reported demographic data, especially for age; however, we attempted to limit misclassification by including the option of coders to select unknown for the demographic identification. In addition, coders were instructed to code the individuals as over 21-years old when they were in doubt about the age determination. It is possible that we may be under-estimating the number of videos featuring younger individuals. We also do not know the tobacco use status of the individuals who engaged with the content; thus, we cannot determine causation or whether engagement with influencers’ cigar videos leads to cigar experimentation. In violation of FTC guidance [44], many influencers do not use the recommended methods of disclosure, such as the use of the hashtags #ad or #sponsored to indicate that Int. J. Environ. Res. Public Health 2022, 19, 7064 11 of 14 they have a “material connection” with a brand. As a result, paid influencer posts can be difficult to identify and study. We identified influencers as those in the top 75th percentile for the number of followers but acknowledge that we may be missing influencers who do not fall within the 75th percentile or include individuals with large numbers of followers who have no material connections to the industry. Previous research has identified LCC influencers with follower counts by categorizing users with 1000 and more followers as influencers [10]. Our study utilized a novel approach that considers the distribution of follower counts; however, the selection of the 75th percentile was arbitrary. To address this limitation, we conducted a sensitivity analysis identifying influencers as those individuals in the top 90th percentile for followers (Supplemental Table S13) and compared the results against those within the top 75th percentile. The sensitivity analysis resulted in similar findings when comparing the individual demographics and content features in videos of the two cigar types (see Supplemental Table S14, Supplemental Table S15, and Supplemental Table S16 for details on the sensitivity analysis), lending validation to our methodlogy. Our study focused on the TikTok platform and a single brand, Swisher Sweets; hence, our findings may not be generalizable to other social media platforms or other LCC brands. We studied the English language content on TikTok; the generalizability to other languages is unknown. Despite these limitations, the content and user characteristics identified in this study could inform the design of media campaigns and the development of tobacco control efforts. 5. Conclusions In summary, our study found that the demographics of the featured individuals and content features differ between videos of large cigar and Swisher Sweets on TikTok. Specifically, compared to large cigar videos, Swisher Sweets videos were more likely to contain content of arts and crafts with cigar packages, cannabis use, and flavored products. Swisher Sweets videos were also more likely to feature individuals who are female, younger, Black, and Asian compared to large cigar TikTok videos. In addition, Swisher Sweets influencers’ videos tend to feature younger, and female individuals than large cigar influencers’ videos. We also identified specific content features that may facilitate the engagement of large cigar and Swisher Sweets videos on TikTok. Both Swisher Sweets and large cigar influencers posted more videos of cigar purchasing behaviors than non- influencers, which was associated with more video views. For large cigar videos, the content of product reviews was associated with more video views and likes on TikTok. Our findings may inform the design and implementation of cigar prevention cam- paigns with targeted demographic populations. Our findings lend support for enforcement of disclosures of sponsorship and health warnings on TikTok and related regulatory actions to restrict the promotions of youth-appealing tobacco content on social media. Our findings also suggest an urgent need for the prohibition of flavors in cigars and extension of plain packaging to cigar products. Supplementary Materials: The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/ijerph19127064/s1, Supplemental Table S1. Views of videos with different cigar-related hashtags on TikTok. Supplemental Table S2. Average followers and engagement with large cigar and Swisher Sweets TikTok Videos. Supplemental Table S3. Multiple comparisons of video features between large cigar and Swisher Sweets videos. Supplemental Table S4. Multiple comparisons of featured individual demographics for videos with people between large cigar and Swisher Sweets videos. Supplemental Table S5. Multiple comparisons of video content features between influencer and non-influencer videos within each hashtag category. Supplemental Table S6. Multiple comparisons of featured individual demographics for videos with recognizable individuals between influencer and non-influencer videos within each hashtag category. Supplemental Table S7. Multiple comparisons of content features between large cigar and Swisher Sweets influencers’ videos. Supplemental Table S8. Multiple comparisons of featured individual demographics between large cigar and Swisher Sweets influencers’ videos. Supplemental Table S9. Predicting likes and views of large cigar videos with video content features. Supplemental Table S10. Predicting shares of large cigar Int. J. Environ. Res. Public Health 2022, 19, 7064 12 of 14 videos with video content features. Supplemental Table S11. Predicting likes and views of Swisher Sweets videos with video content features. Supplemental Table S12. Predicting shares of Swisher Sweets videos with video content features. Supplemental Table S13. Different cutoffs of follower count in large cigar and Swisher Sweets videos. Supplemental Table S14. Multiple comparisons of video features between large cigar and Swisher Sweets videos using the 90th percentile of followers to define influencers. Supplemental Table S15. Multiple comparisons of featured individual demographics for videos with people between large cigar and Swisher Sweets videos using the top 90th percentile of followers to define influencers. Supplemental Table S16. Multiple comparisons of featured individual demographics between large cigar and Swisher Sweets influencers’ videos using the top 90th percentile of followers to define influencers. Author Contributions: Conceptualization, J.W., T.H., J.L.F., M.B. and E.J.B.; data curation, J.W., T.H. and J.L.F.; methodology, J.W., T.H. and J.L.F.; writing—Original draft, J.W., T.H. and J.L.F.; writing— Review and editing, J.W., A.F.H., D.W., M.B., E.J.B., Z.X., T.H. and J.L.F. All authors have read and agreed to the published version of the manuscript. Funding: Research reported in this publication was supported, in part, by the National Heart, Lung, And Blood Institute of the National Institutes of Health under Award Number U54HL120163. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. J. Wu reports grants from American Heart Association. J.L. Fetterman reports grants from American Heart Association and a National Heart, Lung, and Blood Institute K01 HL143142. A.F. Harlow reports grants from American Heart Association. E.J. Benjamin reports grants from NIH R01HL092577 and American Heart Association AHA_18SFRN34110082. Institutional Review Board Statement: Research procedures were deemed to not meet the definition of human subjects by the Authors’ Institutional Review Board due to the use of publicly available data. Informed Consent Statement: Not applicable. Data Availability Statement: The code and data underlying this article will be shared on reasonable request to the corresponding author. Conflicts of Interest: The authors declare no conflict of interest. References 1. 2. Gentzke, A.S.; Wang, T.W.; Jamal, A.; Park-Lee, E.; Ren, C.; Cullen, K.A.; Neff, L. Tobacco Product Use Among Middle and High School Students—United States, 2020. MMWR Morb. Mortal. Wkly. Rep. 2020, 69, 1881–1888. [CrossRef] [PubMed] Delnevo, C.D.; Miller Lo, E.; Giovenco, D.P.; Cornacchione Ross, J.; Hrywna, M.; Strasser, A.A. Cigar Sales in Convenience Stores in the US, 2009–2020. JAMA 2021, 326, 2429–2432. [CrossRef] [PubMed] 3. Wang, X.; Kim, Y.; Borowiecki, M.; Tynan, M.A.; Emery, S.; King, B.A. Trends in Cigar Sales and Prices, by Product and Flavor 4. 5. 6. 7. 8. 9. Type—the United States, 2016–2020. Nicotine Tob. Res. 2021, 24, ntab238. [CrossRef] Cullen, K.A.; Liu, S.T.; Bernat, J.K.; Slavit, W.I.; Tynan, M.A.; King, B.A.; Neff, L.J. Flavored Tobacco Product Use Among Middle and High School Students—United States, 2014–2018. MMWR Morb. Mortal. Wkly. Rep. 2019, 68, 839–844. [CrossRef] Giovenco, D.P.; Spillane, T.E.; Talbot, E.; Wackowski, O.A.; Audrain-McGovern, J.; Ganz, O.; Delnevo, C.D. Packaging Characteris- tics of Top-Selling Cigars in the United States, 2018. Nicotine Tob. Res. 2022, ntac070. [CrossRef] Kong, A.Y.; Queen, T.L.; Golden, S.D.; Ribisl, K.M. Neighborhood Disparities in the Availability, Advertising, Promotion, and Youth Appeal of Little Cigars and Cigarillos, United States, 2015. Nicotine Tob. Res. 2020, 22, 2170–2177. [CrossRef] Giovenco, D.P.; Miller Lo, E.J.; Lewis, M.J.; Delnevo, C.D. “They’re Pretty Much Made for Blunts”: Product Features That Facilitate Marijuana Use Among Young Adult Cigarillo Users in the United States. Nicotine Tob. Res. 2017, 19, 1359–1364. [CrossRef] Bierhoff, J.; Haardörfer, R.; Windle, M.; Berg, C.J. Psychological Risk Factors for Alcohol, Cannabis, and Various Tobacco Use among Young Adults: A Longitudinal Analysis. Subst. Use Misuse 2019, 54, 1365–1375. [CrossRef] Sterling, K.L.; Fryer, C.S.; Fagan, P. The Most Natural Tobacco Used: A Qualitative Investigation of Young Adult Smokers’ Risk Perceptions of Flavored Little Cigars and Cigarillos. Nicotine Tob. Res. 2016, 18, 827–833. [CrossRef] 10. CDC. Cigars; CDC: Atlanta, GA, USA, 2022. Available online: https://www.cdc.gov/tobacco/data_statistics/fact_sheets/ tobacco_industry/cigars/index.htm (accessed on 16 May 2022). 11. National Cancer Institute. Cigars: Health Effects and Trends; Smoking and Tobacco Control Monographs; No 98-4302; NIH Pub: Bethesda, MD, USA, 1998. 12. Delnevo, C.D.; Hrywna, M.; Giovenco, D.P.; Miller Lo, E.J.; O’Connor, R.J. Close, but no cigar: Certain cigars are pseudo-cigarettes designed to evade regulation. Tob. Control 2017, 26, 349–354. [CrossRef] 13. Delnevo, C.D.; Giovenco, D.P.; Miller Lo, E.J. Changes in the Mass-merchandise Cigar Market since the Tobacco Control Act. Tob. Regul. Sci. 2017, 3 (Suppl. S1), S8–S16. [CrossRef] Int. J. Environ. Res. Public Health 2022, 19, 7064 13 of 14 14. Corey, C.G.; Holder-Hayes, E.; Nguyen, A.B.; Delnevo, C.D.; Rostron, B.L.; Bansal-Travers, M.; Kimmel, H.L.; Koblitz, A.; Lambert, E.; Pearson, J.L.; et al. US Adult Cigar Smoking Patterns, Purchasing Behaviors, and Reasons for Use According to Cigar Type: Findings From the Population Assessment of Tobacco and Health (PATH) Study, 2013–2014. Nicotine Tob. Res. 2018, 20, 1457–1466. [CrossRef] [PubMed] 15. Weinberger, A.H.; Delnevo, C.D.; Zhu, J.; Gbedemah, M.; Lee, J.; Cruz, L.N.; Kashan, R.S.; Goodwin, R.D. Trends in Cigar Use in the United States, 2002–2016: Diverging Trends by Race/Ethnicity. Nicotine Tob. Res. 2020, 22, 583–587. [CrossRef] [PubMed] 16. Ribisl, K.M.; D’Angelo, H.; Feld, A.L.; Schleicher, N.C.; Golden, S.D.; Luke, D.A.; Henriksen, L. Disparities in tobacco marketing and product availability at the point of sale: Results of a national study. Prev. Med. 2017, 105, 381–388. [CrossRef] [PubMed] 17. Kowitt, S.D.; Goldstein, A.O.; Sutfin, E.L.; Osman, A.; Meernik, C.; Heck, C.; Ranney, L.M. Adolescents’ first tobacco products: Associations with current multiple tobacco product use. PLoS ONE 2019, 14, e0217244. [CrossRef] [PubMed] 18. Pew Research Center. Teens, Social Media and Technology; Pew Research Center: Washington, DC, USA, 2018. Available online: https://www.pewresearch.org/internet/2018/05/31/teens-social-media-technology-2018/ (accessed on 3 July 2021). 19. De Veirman, M.; Cauberghe, V.; Hudders, L. Marketing through Instagram influencers: The impact of number of followers and product divergence on brand attitude. Int. J. Advert. 2017, 36, 798–828. [CrossRef] 20. Chu, K.H.; Majmundar, A.; Allem, J.P.; Soto, D.W.; Cruz, T.B.; Unger, J.B. Tobacco Use Behaviors, Attitudes, and Demographic Characteristics of Tobacco Opinion Leaders and Their Followers: Twitter Analysis. J. Med. Internet Res. 2019, 21, e12676. [CrossRef] 21. Cavazos-Rehg, P.; Li, X.; Kasson, E.; Kaiser, N.; Borodovsky, J.T.; Grucza, R.; Chen, L.-S.; Bierut, L.J. Exploring How Social Media Exposure and Interactions Are Associated With ENDS and Tobacco Use in Adolescents From the PATH Study. Nicotine Tob. Res. 2021, 23, 487–494. [CrossRef] 22. Bandura, A.; Walters, R.H. Social Learning Theory; Prentice Hall: Englewood Cliffs, NJ, USA, 1977. 23. De Veirman, M.; Hudders, L.; Nelson, M.R. What Is Influencer Marketing and How Does It Target Children? A Review and 24. Direction for Future Research. Front. Psychol. 2019, 10, 2685. [CrossRef] Forbes. Top 10 Most Downloaded Apps and Games of 2021: TikTok, Telegram Big Winners; Forbes: Jersey City, NJ, USA, 2021. Available online: https://www.forbes.com/sites/johnkoetsier/2021/12/27/top-10-most-downloaded-apps-and-games-of-2021-tiktok- telegram-big-winners/?sh=38e68673a1fe (accessed on 16 May 2022). 25. Zhong, R.; Frenkel, S. A Third of TikTok’s U.S. Users May Be 14 or Under, Raising Safety Questions. The New York Times. 14 August 2020. Available online: https://www.nytimes.com/2020/08/14/technology/tiktok-underage-users-ftc.html (accessed on 16 May 2022). 26. Guarda, T.; Augusto, M.F.; Victor, J.A.; Mazón, L.M.; Lopes, I.; Oliveira, P. The Impact of TikTok on Digital Marketing. In Marketing and Smart Technologies; Rocha, Á., Reis, J.L., Peter, M.K., Cayolla, R., Loureiro, S., Bogdanovi´c, Z., Eds.; Springer: Singapore, 2021; pp. 35–44. [CrossRef] 27. Rach, M.; Peter, M.K. How TikTok’s Algorithm Beats Facebook & Co. for Attention Under the Theory of Escapism: A Network Sample Analysis of Austrian, German and Swiss Users. In Advances in Digital Marketing and eCommerce; Martínez-López, F.J., López López, D., Eds.; Springer International Publishing: Cham, Switzerland, 2021; pp. 137–143. 28. TikTok. TikTok Advertising Policies—Ad Creatives & Landing Page. 2021. Available online: https://ads.tiktok.com/help/ article?aid=9552 (accessed on 3 July 2021). 29. TikTok. Community Guidelines. 2021. Available online: https://www.tiktok.com/community-guidelines?lang=en (accessed on 30. 3 July 2021). Jackler, R.K.; Li, V.Y.; Cardiff, R.A.L.; Ramamurthi, D. Promotion of tobacco products on Facebook: Policy versus practice. Tob. Control 2019, 28, 67. [CrossRef] 31. Dickinson, D.M.; Johnson, S.E.; Coleman, B.N.; Tworek, C.; Tessman, G.K.; Alexander, J. The Language of Cigar Use: Focus Group 32. Findings on Cigar Product Terminology. Nicotine Tob. Res. 2016, 18, 850–856. [CrossRef] [PubMed] Step, M.M.; Bracken, C.C.; Trapl, E.S.; Flocke, S.A. User and Content Characteristics of Public Tweets Referencing Little Cigars. Am. J. Health Behav. 2016, 40, 38–47. [CrossRef] [PubMed] 33. CDC. Tobacco Brand Preferences; CDC: Atlanta, GA, USA, 2021. Available online: https://www.cdc.gov/tobacco/data_statistics/ fact_sheets/tobacco_industry/brand_preference/index.htm (accessed on 6 December 2021). 34. Delnevo, C.D.; Giovenco, D.P.; Ambrose, B.K.; Corey, C.G.; Conway, K.P. Preference for flavoured cigar brands among youth, young adults and adults in the USA. Tob. Control 2015, 24, 389. [CrossRef] [PubMed] 35. Allem, J.P.; Escobedo, P.; Chu, K.H.; Boley Cruz, T.; Unger, J.B. Images of Little Cigars and Cigarillos on Instagram Identified by the Hashtag #swisher: Thematic Analysis. J. Med. Internet Res. 2017, 19, e255. [CrossRef] 36. Ganz, O.; Rose, S.W.; Cantrell, J. Swisher Sweets ‘Artist Project’: Using musical events to promote cigars. Tob. Control 2018, 27, e93–e95. [CrossRef] 37. Kostygina, G.; Tran, H.; Shi, Y.; Kim, Y.; Emery, S. ‘Sweeter Than a Swisher’: Amount and themes of little cigar and cigarillo content on Twitter. Tob. Control 2016, 25, i75–i82. [CrossRef] 38. Allem, J.-P.; Uppu, S.P.; Boley Cruz, T.; Unger, J.B. Characterizing Swisher Little Cigar–Related Posts on Twitter in 2018: Text Analysis. J. Med. Internet Res. 2019, 21, e14398. [CrossRef] 39. TikTok. Terms of Service. 2021. Available online: https://www.tiktok.com/legal/terms-of-service?lang=en (accessed on 23 October 2021). Int. J. Environ. Res. Public Health 2022, 19, 7064 14 of 14 40. Nord, A. TikTok Scraper & Downloader. 2021. Available online: https://www.npmjs.com/package/tiktok-scraper (accessed on 17 September 2020). 41. Montgomery, L.; Yockey, A. Rolling and scrolling: The portrayal of marijuana cigars (blunts) on YouTube. J. Subst. Use 2018, 23, 436–440. [CrossRef] 42. Guillory, J.; Kim, A.E.; Fiacco, L.; Cress, M.; Pepper, J.; Nonnemaker, J. An Experimental Study of Nicotine Warning Statements in E-cigarette Tweets. Nicotine Tob. Res. 2020, 22, 814–821. [CrossRef] 43. Landis, J.R.; Koch, G.G. The measurement of observer agreement for categorical data. Biometrics 1977, 33, 159–174. [CrossRef] 44. Federal Trade Commission. Disclosures 101: New FTC Resources for Social Media Influencers; Federal Trade Commission: Washington, DC, USA, 2019. Available online: https://www.ftc.gov/news-events/blogs/business-blog/2019/11/disclosures-101-new-ftc- resources-social-media-influencers (accessed on 3 July 2021). Federal Trade Commission. Advertising and Marketing; Federal Trade Commission: Washington, DC, USA, 2021. Available online: https://www.ftc.gov/tips-advice/business-center/advertising-and-marketing (accessed on 3 July 2021). 45. 46. Chen-Sankey, J.C.; Mead-Morse, E.L.; Le, D.; Rose, S.W.; Quisenberry, A.J.; Delnevo, C.D.; Choi, K. Cigar-Smoking Patterns by Race/Ethnicity and Cigar Type: A Nationally Representative Survey Among U.S. Adults. Am. J. Prev. Med. 2021, 60, 87–94. [CrossRef] [PubMed] 47. Wenger, L.D.; Malone, R.E.; George, A.; Bero, L.A. Cigar magazines: Using tobacco to sell a lifestyle. Tob. Control 2001, 10, 279–284. 48. 49. [CrossRef] [PubMed] Sterling, K.L.; Moore, R.S.; Pitts, N.; Duong, M.; Ford, K.H.; Eriksen, M.P. Exposure to celebrity-endorsed small cigar promotions and susceptibility to use among young adult cigarette smokers. J. Environ. Public Health 2013, 2013, 520286. [CrossRef] [PubMed] Sterling, K.L.; Fryer, C.S.; Nix, M.; Fagan, P. Appeal and Impact of Characterizing Flavors on Young Adult Small Cigar Use. Tob. Regul. Sci. 2015, 1, 42–53. [CrossRef] 50. Klein, E.G.; Czaplicki, L.; Berman, M.; Emery, S.; Schillo, B. Visual Attention to the Use of #ad versus #sponsored on e-Cigarette Influencer Posts on Social Media: A Randomized Experiment. J. Health Commun. 2020, 25, 925–930. [CrossRef] 51. Lee, S.; Kim, E. Influencer marketing on Instagram: How sponsorship disclosure, influencer credibility, and brand credibility impact the effectiveness of Instagram promotional post. J. Glob. Fash. Mark. 2020, 11, 232–249. [CrossRef] 52. Vogel, E.A.; Guillory, J.; Ling, P.M. Sponsorship Disclosures and Perceptions of E-cigarette Instagram Posts. Tob. Regul. Sci. 2020, 6, 355–368. [CrossRef]
null
10.1103_physrevb.107.094206.pdf
null
null
PHYSICAL REVIEW B 107, 094206 (2023) Probability transport on the Fock space of a disordered quantum spin chain Isabel Creed ,1,* David E. Logan ,1,2,† and Sthitadhi Roy 3,‡ 1Physical and Theoretical Chemistry, Oxford University, South Parks Road, Oxford OX1 3QZ, United Kingdom 2Department of Physics, Indian Institute of Science, Bengaluru 560012, India 3International Centre for Theoretical Sciences, Tata Institute of Fundamental Research, Bengaluru 560089, India (Received 12 January 2023; revised 9 February 2023; accepted 8 March 2023; published 29 March 2023) Within the broad theme of understanding the dynamics of disordered quantum many-body systems, one of the simplest questions one can ask is, given an initial state, how does it evolve in time on the associated Fock-space graph, in terms of the distribution of probabilities thereon? A detailed quantitative description of the temporal evolution of out-of-equilibrium disordered quantum states and probability transport on the Fock space is our central aim here. We investigate it in the context of a disordered quantum spin chain, which hosts a disorder-driven many-body localization transition. Real-time dynamics/probability transport is shown to exhibit a rich phenomenology, which is markedly different between the ergodic and many-body localized phases. The dynamics is, for example, found to be strongly inhomogeneous at intermediate times in both phases, but while it gives way to homogeneity at long times in the ergodic phase, the dynamics remain inhomogeneous and multifractal in nature for arbitrarily long times in the localized phase. Similarly, we show that an appropriately defined dynamical lengthscale on the Fock-space graph is directly related to the local spin autocorrelation, and as such sheds light on the (anomalous) decay of the autocorrelation in the ergodic phase, and lack of it in the localized phase. DOI: 10.1103/PhysRevB.107.094206 I. INTRODUCTION The out-of-equilibrium dynamics of isolated quantum many-body systems can show a rich range of behavior in the presence of disorder. One of the most striking examples is the driving of such a system from the default ergodic phase into a many-body localized (MBL) phase at sufficiently strong disorder [1–8], via a dynamical phase transition [9,10]. In contrast to the ergodic phase, the system in the MBL phase fails to thermalize under its own dynamics, and memory of the initial state survives locally for arbitrarily long times. Standard signatures of these include the absence of transport of conserved quantities, and autocorrelations of local observ- ables saturating to finite values at long times [4,11,12], rather than vanishing. As such behavior falls outside the paradigm of conventional statistical mechanics, the dynamics in the MBL phase is naturally of fundamental interest. At the same time, even within the ergodic phase but at disorder strengths preceding the MBL transition, the dynamics is anomalously slow. This is commonly manifest in subdiffusive transport of conserved quantities, and autocorrelations of local observ- ables decaying in time with anomalous power-law exponents *isabel.creed@chem.ox.ac.uk †david.logan@chem.ox.ac.uk ‡sthitadhi.roy@icts.res.in Published by the American Physical Society under the terms of the Creative Commons Attribution 4.0 International license. Further distribution of this work must maintain attribution to the author(s) and the published article’s title, journal citation, and DOI. [13–18]. This behavior attests to the fact that the out-of- equilibrium dynamics of disordered quantum systems across a range of disorder strengths straddling the MBL transition is an interesting question. From a phenomenological point of view, there has been substantial progress in understanding the dynamics, both in the MBL phase as well as in the anomalous ergodic regime. The absence of transport in the MBL phase can be ex- plained via the presence of an extensive number of emergent local integrals of motion (or equivalently, local conserved charges), such that an effective model for the MBL phase involves interactions only between these entities [19–21]. More recently, resonances between configurations of these charges have been shown to further explain several features of the MBL phase [22–27]. In the anomalous ergodic regime, progress in understanding the slow dynamics has centered on phenomenological theories based on rare Griffiths regions, as well as anomalous spectral properties of local observables [13,16,18]. It is nevertheless desirable to have a theoretical framework, rooted in microscopics, for understanding both the slow dynamics preceding the MBL phase and the arrested dynamics in the MBL phase. Mapping the dynamics of the many-body system to that of probabilities on its Fock-space graph provides such a framework. Indeed, understanding the physics of many-body localiza- tion from the perspective of the associated Fock space (FS) has emerged as a fruitful approach over the last few years [1,28–52]. This approach involves recasting the Hamiltonian of a disordered, interacting quantum system as a tight-binding Hamiltonian on the complex, correlated FS graph of the system [53]. The problem then becomes one of Anderson localization (AL) of a fictitious particle on the FS graph, albeit 2469-9950/2023/107(9)/094206(19) 094206-1 Published by the American Physical Society CREED, LOGAN, AND ROY PHYSICAL REVIEW B 107, 094206 (2023) a distinctly unconventional AL problem due to the strong correlations in effective disorder on the FS graph [43]. This mapping opens a new window into the connections between the anatomy of the eigenstates on the FS graph, and their manifestations in terms of real-space properties. For instance, the spread of the eigenstates on the FS graph, and an asso- ciated emergent correlation length, has been shown to carry information about eigenstate expectation values of local ob- servables [49] as well as that of the l-bit localization length [46]. Higher-point correlations of eigenstate amplitudes en- code their entanglement structure [51]. However, most of these studies have focused on eigenstate properties, and much less so in the context of dynamical, time-dependent properties. Motivated by this, we investigate here the dynamics of an out-of-equilibrium quantum state on the FS graph. Ar- guably, the most fundamental question one can ask in this regard is, given an initial out-of-equilibrium state, how do the probability densities of the state on the FS graph evolve in time and spread out on the graph? As will be shown, a detailed characterization of this probability transport carries a plethora of information providing insights into the dynamics of disordered quantum many-body systems. This is the central goal of the work. We begin with a brief overview of the paper. Overview As a concrete setting for our analysis, we consider a quantum Ising chain with disordered longitudinal fields and interactions, together with a constant transverse field (of strength (cid:2)). A description of the model and its associated FS graph is given in Sec. II. Classical spin configurations form a convenient set of basis states; they also form the nodes (or “sites”) of the FS graph, with the transverse field generating links between them. The Hamming distance [54] between two classical spin configurations endows the FS graph with a natural measure of distance. Initialising the state in a classical spin configuration corresponds to initializing it on a site on the FS graph. Consequently, the FS graph can be organized such that the given initial state sits at the apex and all sites at a fixed Hamming distance from the initial site are arranged row-wise (see Fig. 1). Although the FS graph for a chain of length L is an L- dimensional hypercube, the above organization of the graph gives rise to two natural “axes” along which the probability transport can be defined; we refer to them as longitudinal and lateral probability transport. The former quantifies how the probability flows down sites which are at increasing dis- tances from the initial FS site. Lateral probability transport on the other hand measures how the probability spreads across sites at the same Hamming distance from the initial site, i.e., on a given row. Section III formalizes these two no- tions of FS probability transport. We show in particular that a time-dependent lengthscale r(t ), which characterizes the longitudinal spread of the wavefunction, is directly related to the real-space spin autocorrelation function. We also quantify the extent to which the time-evolving state is (de)localized on the graph, via t-dependent inverse participation ratios (IPR) and their corresponding fractal exponents. These IPRs can be defined over the entire FS graph, or can be defined row-wise (which corresponds to the lateral transport). FIG. 1. Fock-space (FS) graph of the disordered TFI model Eq. (1) in the basis of σ z-product states, illustrated for L = 8. An arbitrary FS site I is placed at the apex. The graph has L + 1 rows, and the number of FS sites on row r is Nr = . Any FS site J on row r is a Hamming distance rIJ = r from I. Links/hoppings can connect only FS sites on adjacent rows; with each FS site connected to precisely L others. (cid:2) L r (cid:3) In Sec. IV we analyze the short-time dynamics, which is independent of whether the ultimate late-time behavior of the system is ergodic or MBL in character. For (cid:2)t (cid:2) 1, the probability of finding the system in a given FS site/state at distance r is shown to scale as ∼((cid:2)t )2r. An essential outcome of this is an emergent multifractality of the wavefunction over the full FS, with a fractal exponent growing ∝ t 2, independent of disorder strength. By contrast, the row-resolved IPRs on these timescales do not show fractal statistics, indicating that the short-time wavefunction is spread homogeneously across any given row of the FS graph. A further, rather striking consequence of the analysis, is that r(t ) becomes extensive in system size L at any finite O(1) time. This is mandated by the spin autocorrelation being strictly <1 at any finite O(1) time, and can be understood via the extensive connectivity of the FS graph. Section V is devoted to consideration of longitudinal prob- ability transport, notably for long times. A central result here is that, in the ergodic regime, the lengthscale r(t ) grows sub- diffusively, ∼t α with α < 1/2, until it reaches its maximal value of L/2 (modulo the role of mobility edges and finite-size effects, as explained later). This is shown to imply that the spin autocorrelation also decays as a power law with the same exponent. In the MBL regime by contrast, r(t ) saturates to an extensive but submaximal value, which in turn implies that the spin autocorrelation remains nonzero at arbitrarily long times. A further implication of these results is that the emergent fractality present at short to intermediate times gives way to fully delocalized states at long times in the ergodic regime, whereas the fractality persists for arbitrarily long times in the MBL regime. In Sec. VI we turn to the analysis of lateral probability transport, via row-resolved IPRs. The picture that emerges is that, following the short-time homogeneity, at intermediate times—and for any disorder strength—the time-dependent 094206-2 PROBABILITY TRANSPORT ON THE FOCK SPACE OF A … PHYSICAL REVIEW B 107, 094206 (2023) probabilities on any row develop strong inhomogeneities, re- flected in (multi)fractal scalings of the row-resolved IPRs. For sufficiently long times, however, this fractality gives way to complete homogeneity in the ergodic regime, while it persists in the MBL regime. Since the lateral transport in essence cap- tures inhomogeneity in the evolution of probabilities on the FS graph, it is also natural to study t-dependent distributions of probabilities over sites on a given row. Consistent with the above picture, we find that the inhomogeneities are accom- panied by heavy-tailed Lévy distributions, whereas temporal regimes in which probability spreads homogeneously are characterized by narrow distributions. We summarize our results in Sec. VII (see Fig. 16 for a visual summary), and close with concluding remarks and a future outlook. II. MODEL AND FOCK-SPACE GRAPH We consider a disordered transverse-field Ising (TFI) spin- 1/2 chain, specified by the Hamiltonian H = L−1(cid:4) (cid:5)=1 J(cid:5) ˆσ z (cid:5) ˆσ z (cid:5)+1 (cid:5) h(cid:5) ˆσ z (cid:5) + (cid:2) ˆσ x (cid:5) (cid:6) , (1) + L(cid:4) (cid:5)=1 where h(cid:5) and J(cid:5) are i.i.d. random variables, uniformly dis- tributed with h(cid:5) ∈ [−W, W ] and J(cid:5) ∈ [J − δJ , J + δJ ]. For numerical studies, we consider J = 1, δJ = 0.2, and (cid:2) = 1. With these parameters, and the range of system sizes accessible in practice to exact diagonalization (ED), the crit- ical disorder strength above which all eigenstates are MBL is estimated to be Wc (cid:6) 3.8 [55]. Some recent papers on standard disordered models [23,27,56–62] have, however, suggested that a genuine MBL phase, stable in the thermody- namic limit L → ∞, can arise only for much larger values of W , and that the apparent localization found for finite systems at W > Wc is indicative of a prethermal regime. Here we take the view that the MBL phenomenology clearly observed at W > Wc ∼ 4 for ED-accessible system sizes persists in the thermodynamic limit, albeit for larger W values. Fock space (FS) provides a natural framework for studying many-body localization [28–52], in part because a generic many-body Hamiltonian maps exactly onto a tight-binding model on the associated FS graph (or “lattice”), of form H = (cid:4) J E J |J(cid:9)(cid:10)J| + (cid:4) (cid:11) J,K TJK |J(cid:9)(cid:10)K| (2) (where (cid:11) means K (cid:12)= J). The FS graph of the TFI model in the basis of σ z-product states is an L-dimensional hypercube with NH = 2L vertices, or FS sites, as illustrated in Fig. 1. A FS site J represents a many-body quantum state |J(cid:9) of L spins, which is an eigenstate of each ˆσ z (cid:5) |J(cid:9) = S(cid:5),J |J(cid:9) (cid:5) operator, ˆσ z where S(cid:5),J = ±1. It is thus an eigenstate of H0 = (cid:5) + (cid:5)[h(cid:5) ˆσ z J(cid:5) ˆσ z (cid:5)+1], i.e., H0|J(cid:9) = EJ |J(cid:9), with EJ the corresponding site energy for the FS site (with the {EJ } maximally correlated [43], and not i.i.d.). Links, or hoppings, on the FS graph are generated by the term H1 = H − H0 = (cid:2) (cid:5) . Each FS site is thus connected to precisely L others, lying solely on adjacent rows of the graph, and each of which corresponds to flipping a spin on a particular real-space site. This generates (cid:5) ˆσ x (cid:5) ˆσ z (cid:7) (cid:7) the hopping contribution to Eq. (2), in which all nonvanishing hopping matrix elements are simply TJK = (cid:2). As illustrated in Fig. 1 the graph consists of L + 1 rows, r = 0 − L. A single FS site, denoted by I in Fig. 1 (and with arbitrary spin orientations for the real-space sites) lies at the apex of the graph, r = 0. The number of FS sites on (cid:3) (cid:2) L ; with the final site, r = L, row r of the graph is Nr := r corresponding to the state |I(cid:9) in which all real-space spins on |I(cid:9) have been flipped. As a measure of distance between two sites on the FS graph we use the Hamming distance, as mentioned in Sec. I. For any pair of FS sites J, I separated by a Hamming distance rIJ = r, then by definition r real-space sites (cid:5) have S(cid:5),J = −S(cid:5),I while L − r sites have S(cid:5),J = +S(cid:5),I . Hence L−1 (cid:4) (cid:5) S(cid:5),I S(cid:5),J = 1 − 2 . rIJ L (3) This connection between Hamming distance on the FS graph and the spin orientations will prove important in Sec. III A in relating the real-space spin autocorrelation function (or imbalance) to the first moment of the FS probabilities. III. DIAGNOSING PROBABILITY TRANSPORT The basic underlying quantities considered are the proba- bilities PIJ (t ) = |GIJ (t )|2 (cid:2) 0, given by PIJ (t ) = |(cid:10)J|(cid:7)(t )(cid:9)|2 = |(cid:10)J|e−iHt |I(cid:9)|2 (4) with |(cid:7)(t )(cid:9) the t-dependent wavefunction. We add here that, unless stated otherwise, time is shown in units of (cid:2)−1 in all figures (i.e., (cid:2) ≡ 1). Physically, PIJ (t ) gives the proba- bility that the system will be found on FS site J at time t, given its initiation on site I (and with PII (t ) the commonly studied return probability). As reflected in PIJ (t = 0) = δIJ , the initial state |I(cid:9) is site-localized on the FS graph, and as such wholly Anderson-localized thereon. On increasing t, the distribution of probabilities spreads in some fashion through the FS graph/lattice. Understanding at least some aspects of this many-sided process, both temporally and as a function of disorder strength, is the aim of this paper. In the following H is presumed real symmetric, as relevant to the TFI model considered explicitly, such that PIJ (t ) = PJI (t ) = PIJ (−t ). Expressed in terms of eigenstate amplitudes AnI = (cid:10)I|n(cid:9), with eigenstates |n(cid:9) and corresponding eigenval- ues En, note for later use that e−i(En PIJ (t ) = −Em )t AnI AnJ AmI AmJ (cid:4) (5) . n,m (cid:7) Probability is of course conserved, viz., J PIJ (t ) = 1 for all t and any initial FS site I. For any given I, PIJ (t ) can thus be regarded as the time-dependent distribution, over all FS sites J, of the conserved “mass” MI = J PIJ (t ) = 1. A natural way to quantify such a distribution is via its moments. To this end, first define (cid:7) PI (r; t ) = (cid:4) PIJ (t ), (cid:4) J:rIJ =r I P(r; t ) = N −1 H PI (r; t ), (6) 094206-3 CREED, LOGAN, AND ROY PHYSICAL REVIEW B 107, 094206 (2023) (cid:3) (cid:2) L r with the J sum over all Nr = FS sites on a given row r of the graph, for which the Hamming distance rIJ = r (I lying at the apex of the graph, see Fig. 1). PI (r; t ) gives the r=0 PI (r; t ) = 1 ∀t, I]; its total probability on row r [with sample average over initial FS sites I is denoted P(r; t ). An average over disorder realizations will be denoted, according to convenience, either by an overbar [e.g., PI (r; t )], or by angle brackets ((cid:10)· · · (cid:9)d). The r and t dependence of P(r; t ) in particular will be considered explicitly in Sec. V. (cid:7) L Moments of the {PIJ (t )} follow directly, e.g., the first mo- ment rI (t ) = (cid:4) J L(cid:4) rIJ PIJ (t ) = rPI (r; t ) , (7) r=0 (cid:7) and its sample average r(t ) = N −1 H consider the disorder-averaged moments I rI (t ). In Sec. V we will r(t ) = δr2(t ) = L(cid:4) r=0 L(cid:4) r=0 rP(r; t ), (8a) r2P(r; t ) − (cid:9)2 rP(r; t ) , (8b) (cid:8) L(cid:4) r=0 in particular the former. As now shown, for any disorder realization, rI (t ) and r(t ) are in fact directly related to the real- space spin autocorrelation function; providing thereby a direct connection between real-space and Fock-space perspectives. A. Longitudinal transport and spin autocorrelator Consider C(t ) defined by L(cid:4) C(t ) = 1 L (cid:5)=1 C(cid:5)(cid:5)(t ), C(cid:5)(cid:5)(t ) = 1 NH (cid:2) (cid:5) (t ) ˆσ z ˆσ z (cid:5) (cid:3) , Tr (9) (cid:7) with C(cid:5)(cid:5)(t ) the local real-space spin autocorrelator. The trace Tr can equivalently be either over FS sites, C(cid:5)(cid:5)(t ) = N −1 (cid:5)(cid:5)(t ) with C [I] (cid:5) |I(cid:9), or over eigen- (cid:5)(cid:5)(t ) = (cid:10)I| ˆσ z C [I] (cid:7) H states, C(cid:5)(cid:5)(t ) = N −1 C [n] (cid:5)(cid:5)(t ). A simple calculation then H relates C [I] n (cid:5)(cid:5)(t ) to the probabilities {PIJ (t )}, (cid:5) (t ) ˆσ z I C[I] (cid:5)(cid:5) (t ) = S(cid:5),I S(cid:5),J PIJ (t ), (10) J (cid:5) |I(cid:9) (= ±1). Using Eq. (3), together with where S(cid:5),I = (cid:10)I| ˆσ z conservation of probability, it follows directly that C[I](t ) := L−1 (cid:4) C[I] (cid:5)(cid:5) (t ) (cid:4) is given by (cid:4) C[I](t ) = 1 − 2 L J (cid:4) ⇒ C(t ) = N −1 H (cid:5) rIJ PIJ (t ) = 1 − 2 L rI (t ) C[I](t ) = 1 − 2 L I r(t ). (11) Equation (11) relates directly the real-space spin autocor- relation function to the first moment of the FS probabilities {PIJ (t )} (and is not confined to the TFI model, holding equally for XXZ or spinless fermion models). It is also interesting to note that experiments where MBL has been observed [63,64] essentially measure C[I] (cid:5)(cid:5) (t ) by employing a similar protocol– initializing the system in a specific σ z configuration |I(cid:9), and measuring the expectation value (cid:10) ˆσ z (cid:5) (t )|I(cid:9) such that C[I] (cid:5) (t )(cid:9) ≡ (cid:10)I| ˆσ z (cid:5)(cid:5) (t ) = (cid:10) ˆσ z (cid:5) (t )(cid:9) S(cid:5),I . A striking feature of the dynamics is that, on timescales for which C[I](t ) departs by merely a nonvanishing amount from its t = 0 value of 1, the first moment rI (t ) ∝ L is extensive in system size. Intuitively, one expects such timescales to be determined by the hopping energy scale (cid:2), which acts to de- phase the initially synchronized spins, and as such to be on the order (cid:2)t ∼ O(1). The resultant extensivity of rI (t ) means that an excitation, initially Anderson-localized on the single FS site I, spreads significantly throughout the Fock space on the shortest timescales of order (cid:2)t ∼ O(1)—and would appear to do so regardless of whether the system is ultimately ergodic or MBL. Understanding how this behavior arises, the essential characteristics of the Fock-space graph which it reflects, and the physical picture it gives rise to, is conceptually significant and considered in Sec. IV (see also Sec. V). |(cid:10)n| ˆσ z One can also bound r(t ). Since rIJ (cid:3) L, it follows triv- J PIJ (t ) = 1 ∀t] that r(t )/L (cid:3) 1 ially from Eq. (11) [using for all t. More useful is a bound in the t → ∞ limit. Re- solving C(cid:5)(cid:5)(t ) as an eigenstate trace, its infinite-time limit C(cid:5)(cid:5)(∞) = N −1 (cid:5) |n(cid:9)|2, so C(cid:5)(cid:5)(∞) and thus C(∞) can- H not be negative; whence [Eq. (11)] r(∞) (cid:3) L/2 necessarily. Sufficiently deep in an ergodic phase, with essentially all many-body eigenstates delocalized and no remnant memory of initial conditions, one expects C(cid:5)(cid:5)(∞) to vanish. Hence r(∞) = L/2—the midpoint of the FS graph—is characteristic of such “complete” ergodicity. In an MBL phase by contrast, persistent memory of initial conditions means C(cid:5)(cid:5)(∞) > 0. In that case, the long-time limit of r(t ) is perforce less than L/2. (cid:7) (cid:7) n B. Lateral transport For any disorder realization, PI (r; t ) gives [Eq. (6)] the total probability on row r of the graph/lattice. Study of its (r, t )-dependence thus reveals how probability flows in time “down” the FS graph, row by row. It does not, however, give information on the important issue of how the distribution of probabilities spreads out laterally, and in general inhomoge- neously, across the rows of the graph. One such measure of the latter, studied numerically in Sec. VI, is provided by RI (r; t ) (cid:2) 1 defined by RI (r; t ) = (cid:2) 1 Nr 1 Nr J:rIJ =r P2 IJ (t ) (cid:3) 2 J:rIJ =r PIJ (t ) (cid:7) (cid:7) . (12) For any given disorder realization, this is simply the ratio of the mean squared probability per FS site on row r, to the square of the corresponding mean probability, [N −1 r PI (r; t )]2. So it provides an obvious measure of fluctuations in the dis- tribution of PIJ ’s along a given row. In particular, RI (r; t ) = 1 in a limit of extreme homogeneity where all PIJ (t )’s on the row are the same. The latter behavior will in fact be shown in Sec. IV to arise at sufficiently short times, independently of disorder strength W ; before evolving in t to a distribution which is W dependent, and strongly inhomogeneous in the MBL regime (Sec. VI). 094206-4 PROBABILITY TRANSPORT ON THE FOCK SPACE OF A … PHYSICAL REVIEW B 107, 094206 (2023) The average of RI (r; t ) over disorder realizations and FS sites I will be denoted for brevity by (cid:10)R(cid:9) ≡ (cid:10)R(cid:9)(r; t ), (cid:4) (cid:10)R(cid:9) = N −1 H (cid:10)RI (r; t )(cid:9) d (13) I with (cid:10)· · · (cid:9)d the disorder average. More generally, we also study in Sec. VI A the full probability distribution of RI (r; t ), given by (cid:4) PR (x) = N −1 H (cid:10)δ(x − RI (r; t ))(cid:9) d (14) I (cid:10) dx xPR(x) = (cid:10)R(cid:9)). In the MBL (of which the first moment is regime in particular, PR(x) at sufficiently long times will be shown to be characterized by a heavy-tailed Lévy alpha-stable distribution. (cid:7) The quantity RI (r; t ) is directly related to another nat- ural measure of fluctuations in the distribution of PIJ (t )’s along a given FS row: the row-resolved, t-dependent in- verse participation ratio (IPR). To motivate this, consider the t-dependent wavefunction following the initial quench, |(cid:7)(t )(cid:9) = e−iHt |I(cid:9), expanded as |(cid:7)(t )(cid:9) = A(I ) J (t )|J(cid:9); such that, from Eq. (4), the squared amplitudes |A(I ) J (t )|2 = PIJ (t ) are just the probabilities of interest. Time-dependent wave- function densities, normalized on any given row r, are then given by |B(I ) J (t )|2 = |A(I ) J (t )|2/ (cid:7) J:r J:r J:rIJ =r); for which the associated generalized shorthand for (cid:7) IPR is II,q = J (t )|2q. Hence, for the standard case of J:r q = 2 on which we focus explicitly, the IPR is related simply (cid:2) L to RI (r; t ) [Eq. (12)] via Nr = r J (t )|2 (with |A(I ) |B(I ) (cid:3) , (cid:7) (cid:7) J (cid:7) I I,2(r; t ) = (cid:2) (cid:7) ⇒ (cid:10)I (cid:9) = N −1 H 2 J:rIJ =r P2 IJ (t ) (cid:3) 2 J:rIJ =r PIJ (t ) (cid:4) (cid:10)I I,2(r; t )(cid:9) d = N −1 r RI (r; t ) = N −1 r (cid:10)R(cid:9) (15) I [with the corresponding probability distribution of II,2 follow- ing trivially from that for RI (r; t ), Eq. (14)]. r (cid:7) |2 ∼ N −1 . Hence (cid:10)I2(cid:9) ∼ N −1 We can then reason physically as follows, consider- ing some particular time t. If the amplitudes |B(I ) J (t )|2 = PIJ (t )/ J:r PIJ (t )—and hence the probabilities PIJ (t )—are essentially uniformly distributed over the Nr FS sites on row r, and in turn (cid:10)R(cid:9) ∼ then each |B(I ) J O(1) should be of order unity (and as such L independent). If by contrast the wavefunction is strongly inhomogeneously distributed on the row, one might anticipate (cid:10)I2(cid:9) ∼ N −ν r with a fractal exponent ν ≡ ν(t ) < 1; and hence from Eq. (15) that (cid:10)R(cid:9) ∼ N 1−ν r —which thus grows with increasing system size L. These two behaviors will indeed be shown to arise in Sec. VI, the former characteristic at long time of the ergodic regime, and the latter characteristic of the MBL regime at larger disorder strengths. r Finally, as a complement to PR(x) [Eq. (14)], we also study in Sec. VI A the probability distribution Prel(x) = 1 NH (cid:4) I 1 Nr (cid:4) J:rIJ =r (cid:11) (cid:8) δ x − (cid:7) PIJ (t ) J:rIJ =r PIJ (t ) 1 Nr (cid:9)(cid:12) . d (16) For any given row r on the graph, this gives the distri- bution of PIJ (t ) relative to its mean value on the row, x = PIJ (t )/[N −1 r PI (r; t )]; its second moment being precisely (cid:10) dx x2Prel(x) = (cid:10)R(cid:9), see Eqs. (13) and (12) (and its first mo- ment is 1 by construction). IV. SHORT-TIME BEHAVIOR We turn now to the short-time behavior of probability transport for the disordered TFI model. While the underlying calculations are simple, the physical picture arising is rather rich; including the emergence at short times of multifractality in the t-dependent wavefunction |(cid:7)(t )(cid:9) = e−iHt |I(cid:9)—for any disorder strength W , and as such independent of whether the ultimate long-time behavior of the system is ergodic or MBL in nature. Consider PIJ (t ) = |GIJ (t )|2, where [Eq. (4)] GIJ (t ) = (cid:10)J|e−iHt |I(cid:9) = ∞(cid:4) n=0 (−i)n n! t n(cid:10)J|Hn|I(cid:9), (17) K [Eq. and (cid:7) separate H ≡ H0 + H1 (2)], with H0 = EK |K(cid:9)(cid:10)K| and H1 the hopping term. With Km denoting any FS site on row m, H|Km(cid:9) connects solely to FS sites in rows m ± 1 (and m), since nonzero hopping matrix elements ((cid:2)) connect only FS sites on adjacent rows of the graph. Now let J in Eq. (18) be some given FS site on row r, call it Jr. Obviously, (cid:10)Jr|Hn|I(cid:9) vanishes identically for all n < r. Hence GIJr (t ) = (−i)r r! t r(cid:10)Jr |(H1)r|I(cid:9) + O(t r+1). (18) The leading term here will clearly dominate GIJr (t ) for suf- ficiently small t. Importantly, it involves solely FS hoppings, consisting of “forward paths” from I to Jr, each containing precisely r hops (i.e., r factors of (cid:2)). For any given FS site Jr there are however r! identical contributions to (cid:10)Jr|(H1)n|I(cid:9), because there are r! distinct forward paths from I to Jr on the FS graph; and each such contribution has a value of (cid:2)r. This cancels the 1/r! factor in Eq. (18), from which the leading (t ) ∼ (−i)r ((cid:2)t )r, and that of PIJr (t ) small-t behavior is GIJr thus PIJr (t ) ∼ ((cid:2)t )2r. (19) Note the following points about behavior: this leading short-time (i) It holds for any r, and for all FS sites on row r. By virtue of the latter, the distribution of probabilities along any given row is fully homogeneous in the time window over which Eq. (19) holds. In consequence, RI (r; t ) = 1 [Eq. (12)], the distribution PR(x) = δ(x − 1) [Eq. (14)] is δ distributed, and the row-resolved IPR I2(r; t ) = N −1 [Eq. (15)]. By it- self the above calculation does not of course prescribe the timescale over which such behavior occurs, but we ascertain it below. (ii) Relatedly, since solely the disorder-independent hoppings (cid:2) generate Eq. (19), the result is independent of disorder strength W . (iii) Although PIJr (t ) ∼ ((cid:2)t )2r decreases exponentially rapidly with r, the number Nr = of FS sites on row r grows exponentially with r. Hence, even for short times, one cannot neglect the contribution of sites on any row r to, e.g., the first moment of the probability distribution, (cid:2) L r (cid:3) r 094206-5 CREED, LOGAN, AND ROY PHYSICAL REVIEW B 107, 094206 (2023) are thus embodied in δr2(t )/L2, direct evaluation of which using Eq. (20) gives δr2(t )/L2 = ((cid:2)t )2[1 − ((cid:2)t )2]/L. Since this is ∝1/L, such fluctuations vanish in the thermodynamic limit, with r(t )/L distributed as a Dirac-delta function at its mean. Finally, although by itself a somewhat limited diagnostic of probability transport on Fock space, we comment parentheti- cally on the commonly studied [14,65,66] return probability, PII (t ). This corresponds to r = 0 in Eq. (20), which for (cid:2)t (cid:2) 1 recovers the known behavior [66] PII (t ) ∼ exp(−L((cid:2)t )2), whereby for any nonzero (cid:2)t, even if small, the return proba- bility is exponentially suppressed in system size L. FIG. 2. ED results with L = 13 for ¯r(t )/L vs t ((cid:2) ≡ 1), shown over the indicated range of disorder strengths W . For times t (cid:2) 0.1, the W -independent behavior of Eq. (21) (dashed line) is seen to arise. Inset: Same results, on smaller t scale. (cid:7) (cid:7) L r=0 J rIJ PIJ (t ) ≡ (cid:3) (cid:2) L rI (t ) = rPIJr (t ), as considered be- r low. (iv) The calculation above naturally reflects the intrinsic structure of the FS graph (Fig. 1) for the disordered TFI model. We simply remark that the result arising would be quite different if one considered a tree graph (Cayley tree/Bethe lattice); for while in that case Eq. (18) holds for any given site Jr on generation r of the tree, there is just a single path connecting the root site I to the given Jr. As it stands, direct use of Eq. (19) for each r fails to conserve total probability. This, however, is readily taken into account by writing PIJr (t ) = g(r; t )((cid:2)t )2r where, for all r, g(r; t ) must satisfy (a) g(r; t = 0) = 1, such that the leading low-t behavior of PIJr (t ) is Eq. (19); (b) g(r; t ) > 0 for all times for which the calculation is valid; and (c) J PIJ (t ) = 1, i.e., overall probability must be conserved, (cid:3) (cid:7) g(r; t )((cid:2)t )2r = 1 ∀t. This has the solution g(r; t ) = [1 − ((cid:2)t )2](L−r). And g(r; t ) > 0 ∀r is satisfied provided (cid:2)t < 1, which upper bounds the time-window over which the cal- culation holds. (cid:2) L r L r=0 (cid:7) The essential result for the short-t behavior of PIJr (t ) is then (t ) = ((cid:2)t )2r[1 − ((cid:2)t )2](L−r). (20) P IJr As this is independent of both the initial FS site I and disorder strength W , the resultant first moments [Eqs. (7) and (8a)] rI (t ) ≡ r(t ) ≡ r(t ) coincide, and follow from Eq. (20) as (cid:13) (cid:14) rI (t ) ≡ r(t ) = r P IJr (t ) = L((cid:2)t )2. (21) L(cid:4) r=0 L r That the short-time behavior is indeed W independent is cor- roborated in Fig. 2, which shows ED results for r(t )/L vs (cid:2)t, over a range of disorder strengths W . In all cases, the asymptotic behavior Eq. (21) indeed arises at short times—in practice for (cid:2)t (cid:2) 0.1 or so, consistent with the bound above. The fact that r(t ) ∝ L is extensive for finite (cid:2)t means of course that it is r(t )/L, which remains finite in the thermody- namic limit L → ∞. The relevant fluctuations in this quantity Emergent multifractality As shown above, for (cid:2)t small compared to unity but fi- nite, the probability density has spread through Fock space to macroscopically large Hamming distances on the order of L. The probabilities PIJr (t ) are uniform on any given row of the FS graph [Eq. (20)], symptomatic of which the row-resolved IPR [Eq. (15)] is II,2(r; t ) = N −1 r One can however also ask for the behavior of the con- ventional IPR over the full Fock-space. For a wavefunction |(cid:7)(t )(cid:9) = J (t )|2 (= PIJ (t )) normalized to unity over all FS sites J, the generalized (q-dependent) IPR is defined by J (t )|J(cid:9), with squared amplitudes |A(I ) A(I ) (cid:7) . J L I,q(t ) = (cid:4) (cid:15) (cid:15)A(I ) J (t ) (cid:15) (cid:15)2q = (cid:4) Pq IJ (t ) , (22) J J where only q > 1 is considered henceforth (trivially, for all t, LI,0(t ) = NH and LI,1(t ) = 1). The L dependence of L I,q(t ) is embodied in the exponent τq ≡ τq(t ) defined by −τq LI,q(t ) ∼ N H . If τq = 0 for any specified t, then the wave- function |(cid:7)(t )(cid:9) is Anderson localized on O(1) FS sites of the graph/lattice, while if τq = q − 1 it is essentially uniformly spread over all FS sites on the graph, and as such ergodic. But if by contrast 0 < τq < q − 1, then the wavefunction is fractal; more specifically, if τq is a nonlinear function of q, then it is multifractal. To consider this in the present context, it is convenient to rewrite Eq. (20) in the binomial form (t ) = [z(t )]r[1 − z(t )](L−r) P IJr (23) with z(t ) = ((cid:2)t )2 for short times (cid:2)t (cid:2) 1. This in turn can be expressed as (t ) = P IJr (cid:5) 1 + e−1/ξF (t ) (cid:6)−L e−r/ξF (t ), (24) in terms of a correlation length ξF (t ) defined by ξ −1 F (t ) = − 1). Since the short-time PIJr (t )’s are the same for ln( 1 (cid:3) (cid:3) (cid:2) z(t ) Pq L sites on row r of the graph, LI,q(t ) ≡ (t ). all IJr r Hence from Eq. (24) (cid:2) L r L r=0 (cid:7) τ q (t ) = log2 (cid:16) (1 + e−1/ξF (t ))q (1 + e−q/ξF (t )) (cid:17) , (25) where e−1/ξF (t ) ∼ ((cid:2)t )2 for (cid:2)t (cid:2) 1. For t = 0 precisely, τq = 0. This is just as expected, reflecting the fact that |(cid:7)(t = 0)(cid:9) = |I(cid:9) is Anderson localized on the FS graph. 094206-6 PROBABILITY TRANSPORT ON THE FOCK SPACE OF A … PHYSICAL REVIEW B 107, 094206 (2023) FIG. 3. t dependence of the IPR exponent τ2(t ) obtained from ED calculations, with W = 1, 1.5, 2 exemplifying the ergodic phase and W = 6, 7 the MBL regime. The data is obtained by fitting the −τ2 (t ) instantaneous IPR, L2(t ) to c(t )N over the range L = 8 − 13 H for each t. Dashed line shows the low-t asymptotic behavior τ2(t ) ∼ 2((cid:2)t )2/ ln 2 [from Eq. (25)]. Full discussion in text. However, for any nonzero (cid:2)t (cid:2) 1, Eq. (25) is readily seen to be nonlinear in q and to satisfy 0 < τq(t ) (cid:2) q − 1. The wavefunction is thus multifractal. Moreover, this behavior arises for any disorder strength W . Emergent multifractality at short times is therefore common both to W ’s for which the system is ergodic in the long-time limit, as well as for W ’s for which it is MBL at long times. In the latter case, one anticipates continued persistence of multifractality beyond the short-time window. In the ergodic case by contrast, one expects multifractality to dissipate with further increasing t, as the distribution of probabilities homogenises over the entire graph and the long-time limit of τq(t = ∞) = q − 1 arises [67]. That the above behavior indeed arises is illustrated in Fig. 3, which, for the standard case q = 2, shows ED results for the t-dependent exponent τ2(t ). We define the latter in general via the averaged IPR, (cid:4) L2(t ) = N −1 H LI,2(t ) , written as I L2(t ) = c(t )N −τ2 (t ) H . (26) (27) Note that for short times (cid:2)t (cid:2) 1, this definition is the same as that arising from Eq. (25), since PIJr (t ) in Eq. (24) is independent of both disorder and the FS site I. For any chosen t, a plot of ln L2(t ) vs ln NH ∝ L then gives −τ2(t ) from the slope (c(t ) is assumed to be L independent); and very good linear fits are indeed found for the data shown. times (cid:2)t (cid:2) 0.1, As seen in Fig. 3, for short the W - independent result from Eq. (25) is indeed recovered, viz., τ2(t ) ∼ 2((cid:2)t )2/ ln 2, and the wavefunction is multifractal for all W . For W = 6, 7 illustrative of the MBL regime, τ2(t ) remains <1 on increasing t beyond the short-time regime and multifractality persists at all times. But for W = 1, 1.5, 2 illus- trating the ergodic regime, τ2(t ) grows with increasing t and ultimately plateaus to a long-time value of τ2 = 1 (≡ q − 1), indicating ergodic behavior. FIG. 4. t-dependent Fock-space distribution P(r; t ), for W=1.5 (ergodic phase) in the left column, and for W = 7 (MBL) in the right column. Top panels show P(r; t ) as a color map in the (r, t ) plane, with white denoting 0 and black denoting 1. Bottom panels show P(r; t ) as a function of r/L for different time slices as indicated in the legend. Data for L = 14, averaged over 2 − 3 × 103 disorder realizations. V. LONGITUDINAL PROBABILITY TRANSPORT In this section we consider how, following a t = 0 quench into some FS site, probability flows in time down the FS graph, row by row. To give an initial broad overview, Fig. 4 shows the r and t dependence of the disorder-averaged total probability on row r, P(r; t ) [Eq. (6)]; for W = 1.5 (left panels) as representative of the ergodic phase, and for W = 7 (right panels) as typical of the MBL regime. The top panels show P(r; t ) as a color map in the (r, t )-plane, while the bottom panels show it as function of r/L, for the (logarithmic) sequence of time slices indicated. The qualitative features arising are clear. At short times, (cid:2)t (cid:2) 0.1, P(r; t ) is the same for both W ’s, as expected from the considerations of Sec. IV. The distributions begin to spread out in an obvious sense for times (cid:2)t (cid:3) 0.5, and in prac- tice reach their long-time steady state by (cid:2)t ∼ 101 − 102. For the W = 1.5 example, the mode of the long-time P(r; t ) lies at r/L = 1/2, the midpoint of the FS graph; and its r-profile is Gaussian (with a width that decreases with increasing system size, a point to which we return later). Similar behavior is found for W = 7, but with the notable difference that in this case the mode of the long-time P(r; t ) occurs at an r/L that is markedly less than 1/2. (cid:7) Quite a bit of information is contained in plots such as Fig. 4. To interrogate it, we turn now to the first moment of L the probability distribution, r(t ) = r=0 rP(r; t ) [Eq. (8a)]. More specifically, we consider r(t )/L, since it is this quan- tity which necessarily remains finite in the thermodynamic limit L → ∞ (Secs. III and IV). We add here that in all figures shown in the paper, disorder averages are taken over a minimum of 103 realizations. 094206-7 CREED, LOGAN, AND ROY PHYSICAL REVIEW B 107, 094206 (2023) (cid:7) n with D(ω) = N −1 δ(ω − En) the (self-averaging) many- H body density of states. The behavior of ξF (ω) with disorder strength W is known from a detailed scaling analysis [49]. For W ’s greater than the critical Wc(ω) for which states at the chosen energy ω become MBL, ξF (ω) remains finite (includ- ing W = Wc(ω)+). For W < Wc(ω) by contrast, ξF (ω) ∝ L and thus diverges in the thermodynamic limit, as expected for delocalized states. The disorder strength denoted throughout as Wc is that above which all states in the band are MBL [i.e., Wc ≡ Wc(ω = 0), as band center states are the last to localize]. For W < Wc, some states in the band will be delocalized, and others MBL—the spectrum hosts mobility edges. For any such W then, from the above, delocalized states contribute a factor of 1/2 to the summand in Eq. (30) as L → ∞, while MBL states contribute a factor strictly <1/2. It is therefore only if all states in the band are delocalized—or in practice all but a tiny fraction—that the long-time limit r(∞)/L will be 1/2. From Fig. 5, this indeed appears to be the case for W = 1. On further increasing W , however, a non-negligible fraction of MBL states must arise, resulting in r(∞)/L < 1/2. The W = 2 case in Fig. 5 appears to provide an example of this (at least up to the largest L considered here). And the trend certainly becomes more pronounced with increasing W , e.g., for W = 3, r(∞)/L is (cid:2) 0.4 for the largest L studied. Equation (30) shows that r(t = ∞)/L can be resolved as a sum over contributions from all eigenstates in the band. This in fact is true for any t. As elaborated in Appendix A, it arises because PIJ (t ) can be eigenstate resolved in the form PIJ (t ) = N −1 IJ (t ), with P(n) IJ (t ) pertaining to a particular state n of H energy En, and given by n P(n) (cid:7) N −1 H P(n) IJ (t ) = cos[(En − Em)t]AnI AnJ AmI AmJ ; (31) (cid:4) m such that for times (cid:2)t (cid:16) 1, P(n) IJ (t ) is controlled by states m lying in a progressively narrowing window |En − Em| (cid:2) (cid:2) in the vicinity of the chosen energy En. We remark in passing that N −1 IJ (t ) can equally be expressed as an eigenstate ex- pectation value of an operator, see Eq. (A2). (cid:7) Since r(t ) is linear in the {PIJ (t )}, it too can be eigenstate H P(n) resolved, r(t ) = N −1 H n r (n)(t ), with (cid:4) r (n)(t ) = N −1 H rIJ P (n) IJ (t ) . (32) I,J In particular, from Eq. (30), r (n)(∞)/L = 1/[1 + e1/ξF,n ]. The lower panels in Fig. 5 show the t dependence of r (n)(t )/L for states n in the immediate vicinity of the band center, with W = 1, 2. Since W < Wc here, one expects the long- time limit of r (n)(t )/L for band center states to be 1/2, which appears consistent with the data. For the W = 2 example, it is also seen from Fig. 5 that the regime of slower dynamics mentioned above, setting in above (cid:2)t ∼ 1, is evident in both r(t )/L and r (n)(t )/L; suggesting that this behavior is associated with delocalized states in the spectrum. To examine further these slow dynamics at intermediate times, we consider equivalently the t dependence of the spin autocorrelation functions, C(t ) = 1 − 2r(t )/L [Eq. (11)] and its eigenstate-resolved counterpart C [n] (t ) = 1 − 2r (n)(t )/L. The former is shown on a log-log scale in the left panel of FIG. 5. For W = 1 and 2, upper panels show ED results for ¯r(t )/L vs t, for L = 8 − 14. Lower panels show corresponding r (n)(t )/L for band center eigenstates n. Full discussion in text. A. r(t ): Ergodic regime For disorder strengths W = 1, 2, the upper panels in Fig. 5 show the t dependence of r(t )/L, over a time window com- parable to or in excess of the associated Heisenberg times tH [the inverse of the mean level spacing, tH is discussed briefly in Appendix B and given for the model by Eq. (B1)]. For both W ’s, r(t )/L for (cid:2)t (cid:2) 0.1 is given by the W- and L-independent short-time result Eq. (21) (as shown in Fig. 2). For the W = 1 case, on further increasing t, r(t )/L rapidly increases towards a value, which, for practical purposes, is ∼1/2 for (cid:2)t (cid:3) 10 or so. For W = 2, the situation is rather different. In that case, while r(t )/L again grows rapidly up to around (cid:2)t ∼ 1, the “elbow” seen in Fig. 5 around this time is succeeded at longer, intermediate times by a regime of slower dynamics, and the long-time limit is discernibly <1/2 for the largest system size studied. To obtain some understanding here, it is first helpful to con- nI A2 nJ , I,J rIJ PIJ (∞) can be expressed as sider the infinite-t limit. From Eq. (5), PIJ (∞) = from which r(∞) = N −1 H n A2 (cid:7) (cid:7) (cid:4) L(cid:4) r(∞) = N −1 H rF n(r) (28a) with F n(r) = (cid:4) n r=0 nI A2 A2 nJ . I,J:rIJ =r (28b) F n(r) itself was studied in detail in [49], where it was shown to be of form F n(r) = (cid:14) (cid:13) L r (1 + e−1/ξ F,n )−Le−r/ξ F,n , (29) with ξF,n a FS correlation length for eigenstates n at the partic- ular energy ω considered (while band center states ω = 0 were considered explicitly in [49], there is nothing special about this energy). Equations (28a) and (29) give r(∞) L = N −1 H (cid:4) n 1 1 + e1/ξ F,n (cid:18) = dω D(ω) 1 + e1/ξ F (ω) (30) 094206-8 PROBABILITY TRANSPORT ON THE FOCK SPACE OF A … PHYSICAL REVIEW B 107, 094206 (2023) FIG. 6. For W = 1.5, 2, and 2.5, with L = 14, ED results for spin autocorrelation functions vs t, on a log-log scale. Left panel: C(t ) = 1 − 2r(t )/L. Right panel: C [n] (t ) = 1 − 2r (n)(t )/L for band center eigenstates n. In either panel, for each W , dashed lines show power-law fits to the intermediate-time behavior, with the power-law exponents found to decrease with increasing W . Fig. 6 for W = 1.5, 2, and 2.5, with L = 14. As seen from the figure, C(t ) and hence r(t )/L exhibits an intermediate-time power-law decay, C(t ) ∝ t −α with α < 1. With increasing W , the exponent α is found to decrease steadily and, subject to the usual caveat of modest system sizes, appears to vanish in the vicinity of W (cid:6) Wc ∼ 4. For eigenstates n in the vicinity of the band center, which are themselves ergodic for W < Wc, the corresponding behavior of C [n] (t ) is shown in the right panel of Fig. 6. It too shows an intermediate-time power-law with an exponent α(cid:11), which, while larger decay, C [n] than the corresponding α at the same W , likewise decreases steadily with increasing W and vanishes around Wc. The L- dependence of C [n] (t ) vs t for band center states is shown in Fig. 7, from which the data is seen to scale progressively further onto the power-law decay with increasing L. (t ) ∝ t −α(cid:11) Subdiffusive dynamics of the spin autocorrelation function, in the ergodic phase for a wide range of disorder strength preceding the MBL regime, has been extensively studied in models with conserved total magnetization, such as the disordered XXZ chain [13–18]. In the Ising spin chain (1) considered here, total magnetization is not by contrast con- served, so the appearance of subdiffusive dynamics warrants explanation. Indeed, for a Floquet version of the Ising chain (1), a previous numerical study raised the possibility that the spin autocorrelation decays as a stretched exponential in FIG. 7. For W = 1.5, 2, (t ) = 1 − 2r (n)(t )/L vs t, for band center states n. Dashed lines show power-law fits to the intermediate-time behavior, onto which the data scales progressively with increasing L. showing 2.5, and C [n] FIG. 8. For W = 7, ED results for the spin autocorrelation func- tion C(t ) = 1 − 2r(t )/L (left panel) and r(t )/L itself (right panel), vs t ((cid:2) ≡ 1) and for the system sizes L indicated. Solid black line shows for comparison the corresponding exact result for MBL0 Eqs. (36) and (34); red line in the left panel gives the asymptotic behavior Eq. (39). (cid:5) + 1 time [68]. However, the key point here is that although total magnetization is not conserved in our model, total energy is (trivially, the Hamiltonian being time independent). As a result, the autocorrelator of the local energy density shows subdiffusive dynamics; (cid:10) ˆH(cid:5)(t ) ˆH(cid:5)(cid:9) ∼ t −α(cid:11)(cid:11) (cid:5) + (cid:2) ˆσ x (cid:5)−1 ˆσ z (cid:5) is not, however, orthogonal to the local energy density operator, Tr[ ˆσ z (cid:5) ˆH(cid:5)] (cid:12)= 0. Therefore at intermediate to late times, the spin autocorrelation picks up the (sub)diffusive tails emerging from the autocorrrelator of the local energy density; explain- ing physically the origin of the power-law decay of the spin autocorrelator. where ˆH(cid:5) ≡ h(cid:5) ˆσ z (cid:5) ]. The spin operator ˆσ z + J(cid:5)−1 ˆσ z 2 [J(cid:5) ˆσ z (cid:5) ˆσ z (cid:5)+1 B. r(t ): MBL regime To illustrate results in the MBL regime, Fig. 8 shows the spin autocorrelation function C(t ), and r(t )/L itself, for dis- order strength W = 7. The behavior seen is representative of the MBL regime for W (cid:3) 4.5 or so, and qualitatively different from that characteristic of the ergodic regime. C(t ) = 1 − 2r(t )/L in Fig. 8 shows clear damped oscilla- tory behavior. It plateaus to a nonzero long-time value (∼0.7, well above zero), indicative of persistent memory of initial conditions; and is barely L dependent over the range studied. Equivalently, the long-time limit of r(t )/L is (cid:2) 1/2 (as seen also in Fig. 4 for the mode of P(r; t ) at long times). This in turn is consistent with Eq. (30) above, where, with all states n MBL for W > Wc, all correlation lengths ξF,n are finite and hence r(∞)/L < 1/2. Two further points about Fig. 8 should be made at this stage, each of which merits some understanding (Sec. V B 1 below). First, while the long-time behavior is seen to be reached in practice by (cid:2)t ∼ 102, damped oscillations about that limit set in at shorter times (cid:2)t ∼ O(1), above which the envelope of the oscillation is in fact rather well fit by a power-law decay ∝ t −β with β ≈ 1/2. Second, in paral- lel to Sec. V A for the ergodic phase, in the MBL regime one can equally consider the eigenstate-resolved C [n] (t ) = 1 − 2r (n)(t )/L, e.g., for states n in the vicinity of the band center. On doing that, one finds essentially no discernible difference from the results for C(t ) shown in Fig. 8. 094206-9 CREED, LOGAN, AND ROY PHYSICAL REVIEW B 107, 094206 (2023) 1. MBL0 To obtain an understanding of the above results, we con- sider now what we refer to as MBL0 [49]. Sufficiently deep in the MBL phase, the model (1) is perturbatively connected to the noninteracting limit J(cid:5) = 0 (MBL0). Here, although the system is “trivially” MBL—because H [Eq. (1)] is site- separable in real-space and the system a set of noninteracting spins—the behavior on the Fock space is known [49] to be nontrivial, and the Fock-space H Eq. (2) remains fully con- nected on the graph. As outlined in Appendix C, for MBL0 the exact disorder- averaged PIJ (t ) can be obtained, starting from the basic definition PIJ (t ) = |GIJ (t )|2, Eq. (4). With J, I any pair of FS sites separated by a Hamming distance rIJ = r, the result is PIJ (t ) = [z0(t )]r[1 − z0(t )](L−r) (33) with z0(t ) given by (cid:19) z0(t ) = = (cid:2)2 h2 + (cid:2)2 (cid:18) W dh W 0 (cid:20) sin2( h2 + (cid:2)2 t ) (cid:21) d (cid:2)2 h2 + (cid:2)2 sin2( (cid:20) h2 + (cid:2)2 t ). (34) Note that PIJ (t ) again has the binomial form Eq. (23), de- duced on general grounds in Sec. IV for short times; and Eq. (34) obviously recovers, as it ought, the known asymptotic behavior z0(t ) = ((cid:2)t )2 as (cid:2)t → 0. Equation (33) in fact depends solely on the Hamming distance r and is otherwise independent of the particular FS sites J, I (see Appendix C). Hence, from Eq. (6), PI (r; t ) is independent of I, and P(r; t ) ≡ PI (r; t ) is thus (cid:13) L r (cid:14) [z0(t )]r[1 − z0(t )](L−r). P(r; t ) = (35) From this follows the first moment r(t ) [Eq. (8a)] and hence C(t ), r(t ) L = z0(t ), C(t ) = 1 − 2z0(t ), (36) each of which is L independent for all t. Figure 8 compares this result for C(t ) and r(t )/L, to ED results for the interacting case. The strong qualitative parallels between the two are self evident. MBL0 is fully determined by z0(t ) [Eq. (34)], which, via the double-angle formula for sin2θ (and setting (cid:2) ≡ 1) is z0(t ) = p − 1 2 K (t ), p = tan−1(W ) 2W (37) √ with K (t ) = (cid:10)(h2 + 1)−1cos(2 h2 + 1 t )(cid:9)d. K (t ) vanishes as t → ∞ [see Eq. (38)]. The long-time limits are then r(∞)/L = p and C(∞) = 1 − 2p; with p < 1/2 necessarily such that C(∞) > 0 and r(∞)/L < 1/2, as characteristic of an MBL phase. For W = 7, as in Fig. 8, the MBL0 C(∞) (cid:6) 0.8, only slightly larger than its interacting counterpart of C(∞) (cid:6) 0.7. We add that in Appendix C we also point out the connec- tion P(r; ∞) ≡ F n(r) between the long-time limit of P(r; t ) and the eigenstate correlation function F n(r) [Eq. (28b), which for MBL0 is the same for all eigenstates n]. The t dependence of K (t ) is readily determined. Its asymp- totic behavior, formally for t (cid:16) 1, is given by (cid:22) K (t ) ∼ 1 2W π 2 [cos(2t ) − sin(2t )] √ t , (38) vanishing as a power law ∝ 1/ t superimposed on the os- cillating envelope of period π . The maxima of the oscillatory part occur at the discrete set of points t = 7 π + π n (n ∈ N0), 8 at which C(t ) ∼ (cid:13) 1 − tan−1(W ) W (cid:14) + √ π 2W . 1√ t (39) √ This is superimposed on the MBL0 result for C(t ) shown in Fig. 8, and in practice is seen to account very well for the behavior down to times t on the order of unity. For MBL0 one can also determine the eigenstate-resolved C [n] (t ) = 1 − 2r (n)(t )/L for an arbitrary eigenstate |n(cid:9). In this case, reflecting the real-space site-separability of H, it can be shown (although we do not prove it here) that the disorder- averaged C [n] (cid:5) |n(cid:9) is independent of both the site (cid:5) and the particular eigenstate |n(cid:9). In consequence, C [n] (t ) ≡ C(t ); providing a rationale for the fact, mentioned in Sec. V B above, that our ED calculations of C [n] (t ) in the interacting case are barely discernible from those for C(t ). (cid:5)(cid:5)(t ) = (cid:10)n| ˆσ z (cid:5) (t ) ˆσ z 2. Fluctuations While our primary focus has been the first moment of P(r; t ), higher central moments are also of course calculable. As previously mentioned, the fact that it is r(t )/L, which generically remains finite in the thermodynamic limit means that it is fluctuations in this quantity one should consider, as reflected in σ 2(t ) := δr2(t )/L2 [with δr2(t ) from Eq. (8b)]. As shown in Sec. IV for the short-t domain, which holds for all disorder/interaction strengths, σ 2(t ) ∝ 1/L. Fluctuations are thus entirely suppressed in the thermodynamic limit. Just the same situation arises for MBL0, which, from Eq. (35) for P(r; t ), gives σ 2(t ) = z0(t )[1 − z0(t )]/L. Indeed, employing steepest descents on Eq. (35) shows P(r; t ) as a function of y ≡ r/L to be Gaussian, with a mean of z0(t ) (= r(t )/L) and variance σ 2(t ) ∝ 1/L; such that it becomes δ distributed in the thermodynamic limit. More generally, across essentially the full range of disorder strengths, our ED calculations are also qualitatively consistent with the above conclusions: aside from a small W interval around Wc ∼ 4, with increasing L we find δr2(t )/L2 to pro- gressively decrease, and the P(r; t ) profile to narrow. VI. LATERAL PROBABILITY TRANSPORT We turn now to the substantive question of how the t- dependent wavefunction spreads out laterally, as reflected in the time-dependent distribution of probabilities across the rows of the FS graph. As explained in Sec. III B, the quantity RI (r; t ) [Eq. (12)] provides a natural measure of fluctuations in the distribution of PIJ ’s along any given row r, and is related directly to the row-resolved, t-dependent IPR by RI (r; t ) = NrII,2(r; t ) [Eq. (15)], with Nr = the number of FS sites on row r. We (cid:3) (cid:2) L r 094206-10 PROBABILITY TRANSPORT ON THE FOCK SPACE OF A … PHYSICAL REVIEW B 107, 094206 (2023) FIG. 9. t dependence of the average (cid:10)R(cid:9) (= Nr(cid:10)I2(cid:9)), see Eqs. (12), (13), and (15). Shown for r = L/2 with W = 1, 2 (top row) and W = 5, 7 (bottom row), for system sizes indicated. consider first the averages, (cid:10)R(cid:9) ≡ (cid:10)R(cid:9)(r; t ) or (cid:10)I2(cid:9) [Eqs. (13) and (15)], over disorder realizations and FS sites I, before turning to the full probability distribution PR(x) [Eq. (14)] of RI (r; t ). There is no a priori requirement here to average over all FS sites I, so in this section we choose (for numerical convenience) to average over FS sites I whose site energies EI lie close to their mean value of zero. Since our interest lies in dynamics, we also focus on a particular, representa- tive r throughout the section. We choose the midpoint of the FS graph, r = L/2 [or r = (L − 1)/2 for odd L], and have checked that the key results arising are not dependent on this choice. Figure 9 shows the t dependence of (cid:10)R(cid:9) for the system sizes indicated, with W = 1, 2 representative of the ergodic regime and W = 5, 7 of the MBL regime. r r Three notable points are evident in Fig. 9. First, in the short-time domain t (cid:2) 0.1, (cid:10)R(cid:9) = 1 independently of L, for all interaction strengths W . As pointed out in Sec. III B [under Eq. (12)], this is the limit of complete homogeneity, where all PIJ (t )’s on any given row of the graph are the same (itself shown in Sec. IV). In consequence, RI (r; t ) = 1 [Eq. (12)] and hence (cid:10)R(cid:9) = 1, as seen; equivalently the row-resolved IPR (cid:10)I2(cid:9) = N −1 (cid:10)R(cid:9) = N −1 . Second, consider now the opposite limit in Fig. 9, viz. the long-time behavior. For W = 1, 2, (cid:10)R(cid:9) here is O(1) and L independent, just as it is in the short-time domain; while for W = 5, 7 by contrast, (cid:10)R(cid:9) clearly grows with increasing L. As explained in Sec. III B [under Eq. (15)], the former behavior again reflects the essentially uniform distribution of probabilities PIJ (t ) over FS sites on the row, with (cid:10)I2(cid:9) = N −1 , as one expects for an ergodic regime at late r times. In the MBL regime by contrast, the growth of (cid:10)R(cid:9) with L reflects that probabilities, and hence the wavefunction, are strongly inhomogeneously distributed on the row. (cid:10)R(cid:9) ∝ N −1 r FIG. 10. Results here refer to long-time behavior (taken at t = 104). Left panel: (cid:10)R(cid:9) vs W for system sizes indicated. Right panel: ln(cid:10)R(cid:9) vs ln Nr, shown for W = 5, 6, 7 and W = 1; showing the scal- ≡ (cid:10)I2(cid:9) ∼ N −ν ing behavior (cid:10)R(cid:9) ∼ N 1−ν , with exponent ν < 1 in the MBL regime and ν = 1 in the ergodic regime. r r to decrease with increasing W (for W = 5, 7, ν (cid:6) 0.4, 0.3 respectively). This figure also shows the same plot for W = 1, confirming the L independence of (cid:10)R(cid:9) (corresponding to ν = 1). The left panel of Fig. 10 gives the late-time (cid:10)R(cid:9) as a function of W , confirming both the strong L dependence inside the MBL regime, and its corresponding absence in the ergodic phase. Equally, it shows a typical “crossover W win- dow”, whose presence is inevitable given accessible system sizes; and which, without further detailed scaling analysis, precludes substantive consideration of W ’s in the vicinity of Wc ∼ 3.8 [55] (which is not our aim here). Third, consider again Fig. 9 for the ergodic phase W ’s. Al- though as above (cid:10)R(cid:9) ≡ (cid:10)R(cid:9)(t ) is L independent at both short- and long-times, for times t on the order of unity (cid:10)R(cid:9)(t (cid:6) 1) shows a strong L dependence. This too is found to have the form (cid:10)R(cid:9)(t (cid:6) 1) ∼ N 1−ν , directly analogous to Fig. 10 (right panel), and with an exponent ν ≡ ν(t (cid:6) 1) that likewise de- creases with increasing W [for W = 1, 2, ν(t = 1) (cid:6) 0.8 and 0.7 respectively]. r The overall physical picture arising from the above is then as follows. Following the W -independent, short- time complete homogeneity of the squared wavefunction amplitudes/probabilities along the row of the graph, the L dependence arising by t ∼ 1—again for all W —indicates the dynamical emergence of multifractal behavior of the wave- function. The latter persists with increasing t in the MBL regime, until by t ∼ 10 or so the long-time multifractality is well established. For the ergodic W ’s by contrast, that evo- lution is arrested; and the system instead crosses over from incipient multifractality to the ergodic behavior reflected in (cid:10)R(cid:9) ∼ N 0 ), indicating an essentially r uniform distribution of probabilities along the row. This pic- ture, pertaining to the row-resolved IPR, provides a rather natural and consistent complement to that shown in Sec. IV to arise for the behavior of the conventional IPR over the full Fock space (see Fig. 3). ∼ O(1) (i.e., (cid:10)I2(cid:9) ∼ N −1 r As was conjectured on physical grounds in Sec. III B, the L dependence of (cid:10)R(cid:9) in the MBL regime is indeed found to be (cid:10)R(cid:9) ∼ N 1−ν r —or equivalently (cid:10)I2(cid:9) ∼ N −ν for the row- resolved IPR—with a long-time (multi)fractal exponent ν < 1. That this is so is demonstrated in Fig. 10, right panel, where in the MBL regime the long-time exponent ν is also seen r Probability distributions The discussion above has centered on the average value (cid:10)R(cid:9) of RI (r; t ) = NrII,2(r; t ) [Eq. (12)]. Now we consider the full probability distribution of RI (r; t ), given by [Eq. (14)] PR(x) = (cid:10)δ(x − RI (r; t ))(cid:9)d,I (with the I averaging over sites 094206-11 CREED, LOGAN, AND ROY PHYSICAL REVIEW B 107, 094206 (2023) FIG. 11. Probability distribution PR(x) of x ( ≡ RI (r; t )) for W = 1 (top row) and W = 7 (bottom), at the sequence of t’s specified, and for different system sizes L as indicated. Full discussion in text. whose FS-site energies are close to their mean, as mentioned above); and the first moment of which distribution is precisely (cid:10)R(cid:9) considered above. To illustrate the key points here, Fig. 11 shows PR(x) vs x for W = 1 (top row) and W = 7 (bottom), at the sequence of t’s indicated, and over the range of system sizes studied. Note that, for either W , PR(x) for t = 0.1 is Dirac-delta distributed at x = 1. This reflects the short-time regime (t (cid:2) 0.1) for which, as discussed above in relation to Fig. 9, RI (r; t ) = 1 (for any r, I and all W ), and hence PR(x) = δ(x − 1). First consider W = 1 in Fig. 11, illustrating the ergodic regime. For any given L, the PR(x) distribution evolves most significantly with t over the interval 0.1 (cid:2) t (cid:2) 1. On further increasing t, the mean (= (cid:10)R(cid:9)) of PR(x) decreases, as also evident from Fig. 9. By t = 100 the mean of the evidently symmetrical PR(x) appears rather well converged over the accessible L range; and the distribution is both narrow and sharpening with increasing L (indeed that behavior is evident by t ∼ 10). A simple fit to PR(x) for t = 100, shows it clearly to be normally distributed, with a variance decreasing with L. The situation is quite different in the MBL regime, illus- trated by W = 7 in Fig. 11. Here again, PR(x) evolves most significantly with t over the interval 0.1 (cid:2) t (cid:2) 1. For fixed L, the distributions are in fact practically converged to their long-time limit by t ∼ 1, above which little further tempo- ral evolution occurs. Clearly, however, the late-time PR(x) is much broader than its counterpart in the ergodic regime (note the the greatly increased x scale compared to W = 1), reflecting the substantial inhomogeneity arising in the MBL regime, as discussed above. With increasing L the mean and mode of PR(x) continue to increase, as discussed in regard to Fig. 9. And the distribution is not only visibly broad but appears to be heavy tailed. To obtain some understanding of the form of PR(x) in the MBL regime, note first that, in contrast to the ergodic phase, the mean (cid:10)R(cid:9) of PR(x) is itself increasing with L (as per Fig. 9). To distill this out from the large-x tail of PR(x), we thus consider the distribution P˜R (x) = (cid:10)δ(x − ˜R)(cid:9) d,I : ˜R = RI (r; t ) (cid:10)R(cid:9) (40) of ˜R = RI (r; t )/(cid:10)R(cid:9), which has a mean of unity for all t. This is shown in Fig. 12 (left panel) from which, given the modest accessible L range, reasonable scaling behavior is seen; and showing a power-law tail P˜R(x) ∼ x−α with α (cid:6) 2.5, such that the variance of the distribution, and all higher moments, are unbounded. It appears in fact that PR(x) itself is described by a general- ized Lévy distribution, PR,L´evy(x) = Aα−1 (cid:2)(α − 1) 1 xα exp(−A/x) x(cid:16)A∝ x−α (41) [with A an L-dependent constant and (cid:2)(z) the gamma func- tion]; and which heavy-tailed distribution is stable provided α < 3. The mode of PR,L´evy(x) is xmode = A/α and, provided FIG. 12. Left panel: For W = 7 at t = 100, showing P˜R(x) [Eq. (40)] vs x for L values indicated. Dashed line shows comparison to the corresponding Lévy distribution P˜R,L´evy(x) [Eq. (42)]. Inset: PR(x) for L = 13, compared to a two-parameter fit to PR,L´evy(x) [Eq. (41)]. Right panel: Now for W = 2 at t = 1, showing P˜R(x) vs x. Dashed line again compares to corresponding P˜R,L´evy(x). 094206-12 PROBABILITY TRANSPORT ON THE FOCK SPACE OF A … PHYSICAL REVIEW B 107, 094206 (2023) FIG. 13. For W = 1. Top panels: Distribution Prel(x) of x [Eq. (43)] vs x, at the t’s specified, and for system sizes L indicated. Bottom panels: Cumulative distribution Prel,c(x). Red dashed line in final panel shows half-Gaussian fit to data (see text). α > 2, its mean is finite and given by x = A(cid:2)(α − 2)/(cid:2)(α − 1) := A/ f (α). The corresponding Lévy distribution for ˜R is then P˜R,L´evy (x) = [ f (α)]α−1 (cid:2)(α − 1) x−α exp(− f (α)/x), (42) and depends solely on α and not on A. The inset to the left panel of Fig. 12 shows PR(x) itself (for L = 13 at t = 100), compared to a two-parameter fit (viz., α, A) to PR,L´evy(x) (leading to α (cid:6) 2.5); the agreement is rather good. The left panel of Fig. 12 shows P˜R(x) for increasing system sizes, com- pared to the corresponding P˜R,L´evy(x) Eq. (42) (dashed line). We add that the L dependence of the fit PR,L´evy(x) itself arises largely from the L dependence of A; by contrast, over the accessible L window, α varies relatively little (and for which reason P˜R,L´evy(x) in Fig. 12 shows reasonable convergence with increasing L). While we have latterly focused on the MBL regime, it was pointed out above that in the ergodic phase—e.g., W = 1, 2 in Fig. 9—the average (cid:10)R(cid:9)(t ) shows strong L depen- dence at times t (cid:6) 1; reflecting incipient multifractality in the wavefunction, which is arrested at later times as the system crosses over to characteristic ergodic behavior. Naturally, such behavior around t (cid:6) 1 is equally apparent in the full PR(x) dis- tributions for the ergodic phase shown in Fig. 11. Accordingly, the right panel in Fig. 12 shows (for W = 2) the distributions P˜R(x) for t = 1, in direct analogy to Fig. 12 left panel. Once again the Lévy form appears to describe the data rather well; now with a larger tail exponent (α (cid:6) 7) than for the MBL regime [such that the variance of P˜R(x) is finite]. Prel(x) distribution Complementary insight into the spread of probabilities across a row of the FS graph comes from the distribu- tion Prel(x) [Eq. (16)]. For a given row r, this gives the distribution—over disorder realizations, FS sites J on the row, and initial FS sites I—of PIJ (t ) relative to its mean value on the row, x ≡ (cid:7) PIJ (t ) J:rIJ =r PIJ (t ) 1 Nr = PIJ (t ) 1 Nr PI (r; t ) . (43) The first moment of Prel(x) is 1 by construction, while its second moment is the average (cid:10)R(cid:9) studied above. First, consider W = 1 (again choosing r = L/2). The top row of Fig. 13 show Prel(x) vs x at the sequence of t’s specified, and for different system sizes L. Corresponding cumulative distributions, (cid:18) Prel,c(x) = 0 x dy Prel(y) , (44) are shown in the bottom panels. While Prel(x) is not converged in L for t = 1—as expected from the preceding discussion— the distributions appear converged in L for the other t’s shown. The long-time Prel(x) is reached by t = 102 (indeed essentially so by t ∼ 10); and consistent with the convergence of Prel(x) with L, the long-time value of (cid:10)R(cid:9) = dx x2Prel(x) is seen from Fig. 9 to be O(1) and L independent. This long-time Prel(x) is in fact rather well captured by a half-Gaussian distri- bution, of form PG(x) = (2/π ) exp(−x2/π ) (for x (cid:2) 0), with a mean of unity and a corresponding cumulative distribution π ). The latter is compared to the ED data in Fig. 13 Erf (x/ (final panel, dashed line), and seen to agree well with it. √ (cid:10) The important physical point here is that the long-time Prel(x) (or PG(x)) has a mean of unity, and fluctuations that are also O(1). This means the probabilities PIJ are essentially uniformly distributed across the row; as evident, e.g., from Eq. (43) where, if all PIJ ’s on the row are comparable, then x ∼ O(1). This is symptomatic of the ergodic behavior one expects for weak disorder. But now consider the case of W = 7, representative of the MBL regime, for which corresponding results are shown in Fig. 14. The situation here is very different since—particularly 094206-13 CREED, LOGAN, AND ROY PHYSICAL REVIEW B 107, 094206 (2023) FIG. 14. For W = 7. Top panels: Distribution Prel(x) of x [Eq. (43)] vs x, at the sequence of t’s specified, and for system sizes L indicated. Bottom panels: Corresponding cumulative distribution Prel,c(x). in the wings of Prel(x)—the distribution is clearly not con- verging with L for any t (save for t (cid:2) 0.1, as known on general grounds, Sec. IV). This is to be expected because, as shown above, e.g., the long-time value of (cid:10)R(cid:9) = dx x2Prel(x) grows with increasing L, as (cid:10)R(cid:9) ∼ N 1−ν r with exponent ν < 1. The question then is, what features of the long-time Prel(x) distribution determine that behavior? It must surely arise from the large-x tails of Prel(x), which, from Fig. 14, are “filling out” in an obvious sense with increasing L. (cid:10) To examine this, consider the effective cumulative distribu- tion (cid:18) x dy y2Prel(y), (45) Fc (x) = 0 giving the contribution to (cid:10)R(cid:9) arising from different parts of the Prel distribution. This is shown in Fig. 15 for x > 1 (dashed lines and right axis), together with Prel(x) itself (solid lines, FIG. 15. For W = 7, with t = 100. Fc(x) [Eq. (45)] vs x (dashed lines, right-hand scale), shown for x > 1 and different L as indicated; and Prel(x) (solid lines, left-hand scale). Black arrowheads show Nr (= for odd L). Dashed black line shows fit to Prel(x) data (see text). for even L, and L (L±1)/2 L L/2 (cid:2) (cid:2) (cid:3) (cid:3) left axis). We add in passing that while the large-x behavior of Prel(x) does not appear to be a pure power law, it is quite well captured by Prel(x) ∼ ax−n ln x (with n (cid:6) 2.3), shown as the black dashed line in Fig. 15. As seen from the figure, Fc(x) for a given L tends to its saturation value at the x = xm(L) for which Prel(x) “crashes” in a self-evident sense. xm(L) grows strongly with L, and the Fc(x)’s for different L progressively collapse onto an essentially common curve. (cid:3) (cid:2) L r As discussed below, the maximum possible value of x [Eq. (43)] in Prel(x) is in fact Nr = (and thus exponentially large in L for any finite r/L). That this is indeed the xm(L) for which Prel(x) crashes and Fc(x) consequently plateaus is seen in Fig. 15, where black arrows show Nr. The fact that max(x) = Nr is evident from Eq. (43). Since all PIJ (cid:2) 0 then, over the set of probabilities PIJ for the Nr FS sites J on row r, it arises in the case where only a single PIJ —call it PIJ ∗— J:rIJ =r PIJ ≡ completely dominates the others (such that PIJ ∗). More generally, if the set of PIJ are correspondingly non-negligible for an O(1) number of FS sites J on the row, then the associated x is again O(Nr ). (cid:7) As shown, it is then the large-x behavior of Prel(x), which governs the second moment (cid:10)R(cid:9), and consequently all higher moments, of the distribution. Physically, this arises from FS sites J for which PIJ greatly exceeds the mean probabil- ity N −1 J:rIJ =r PIJ on the row. And that of course reflects the strong inhomogeneity in the distribution of PIJ ≡ PIJ (t ) across a row, which is symptomatic of the MBL regime for sufficiently long times. (cid:7) r VII. SUMMARY AND DISCUSSION The central question we posed at the outset was, given an initial spin configuration, how do the probability densities of the time-evolving quantum state spread out on the FS graph of a disordered quantum spin chain? In the course of investigat- ing this question, a rather rich phenomenology was uncovered for the anatomy of probability transport on FS. This can be 094206-14 PROBABILITY TRANSPORT ON THE FOCK SPACE OF A … PHYSICAL REVIEW B 107, 094206 (2023) quantum spin chain, it also motivates several further ques- tions of immanent interest. For conserved quantities, or local observables which have a finite overlap with the former it is worth asking if there exists a connection between potentially anomalous transport of the conserved quantity in real space, and FS probability transport. This naturally involves space- time correlations in real space, and not just autocorrelations. Going beyond systems with conserved quantities, one can also ask about the fate of probability transport in the absence of any conserved quantities, such that r(t ) is not restricted to be subdiffusive nor C(t ) to decay as a power-law in time. In this paper we focused on FS probability transport, which is clearly a two-point correlation function on FS. One can generalize the question to that of the dynamics of four-point correlations on FS, with the aim of understanding entangle- ment growth in disordered quantum systems [69–71], both in the ergodic as well as the MBL phase. The persistence of dynamical inhomogeneities in the MBL phase can also provide us with a starting point for under- standing and theorising about the role of resonances in the MBL phase from a FS point of view. Speculating that these resonances are caused by rare disorder fluctuations in real space, it is also interesting to ask similar questions for MBL phases with quasiperiodic potentials, which are devoid of such rare regions [72–74]. Finally, we add that understanding probability transport on Fock space is not solely of theoretical interest, but is also of direct experimental relevance; as evident, e.g., in a recent preprint [52] in which aspects of longitudinal FS probability transport were studied in an experimental realization of a two-dimensional disordered hard-core Bose-Hubbard model on a superconducting quantum processor. ACKNOWLEDGMENTS This work was supported in part by EPSRC, under Grant No. EP/L015722/1 for the TMCS Centre for Doctoral Train- ing, and Grant No. EP/S020527/1. S.R. also acknowledges support from an ICTS-Simons Early Career Faculty Fellow- ship, via a grant from the Simons Foundation (677895, R.G.). We are grateful to Qiujiang Guo for drawing our attention to the recent preprint [52]. APPENDIX A: PIJ (t ) EIGENSTATE RESOLUTION As mentioned in Sec. V, the probabilities PIJ (t ) = |(cid:10)J|e−iHt |I(cid:9)|2 can be eigenstate resolved in the form PIJ (t ) = N −1 H (cid:4) n P(n) IJ (t ) (A1) (cid:7) (cid:7) n r (n) I (t ) = with the sum over eigenstates n. Any quantity linear in the {PIJ (t )}’s can thus likewise be eigenstate resolved. For J rIJ PIJ (t ) is rI (t ) = example, the first moment rI (t ) = (cid:7) N −1 I (t ) with r (n) IJ (t ), and it is the dis- H order averaged r (n)(t )/L = N −1 I (t )/L shown in Fig. 5; (cid:7) H similarly [see Eqs. (9) and (11)], C(t ) = N −1 C [n](t ) with H C [n](t ) = 1 − 2 (cid:7) Since PIJ (t ) ≡ RePIJ (t ) is pure real, Eq. (5) gives PIJ (t ) = n,m cos[(En − Em)t]AnI AnJ AmI AmJ . Hence on comparison to J rIJ P(n) (cid:7) I r (n) L r (n)(t ). n FIG. 16. Schematic summary of the three main time windows in the dynamics, and their characteristic features in the ergodic and MBL phases. conveniently summarized by considering three time windows, as follows (see Fig. 16 for a visual summary). (i) Short times, t (cid:2) 1: In this regime, the system is agnos- tic to which phase it is in, ergodic or MBL. The dynamics on these scales is characterized by an emergent (multi)fractality of the time-evolving state, but homogeneous lateral proba- bility transport across rows of the FS graph. Another crucial feature of this regime is that the emergent lengthscale r(t ) ∼ O(L) for any finite t, however small; this ensures that the spin-autocorrelation C(t ) is necessarily less than unity. The fractal exponent τ2, as well as the lengthscale r(t ), grows as ∝ t 2 in this regime. (ii) Intermediate times, 1 (cid:2) t (cid:2) tH : This is arguably the most interesting dynamical regime. While the emergent (multi)fractality of the entire wavefunction persists, albeit with an increasing τ2, strong inhomogeneities in the lateral probability transport also set in, reflected in (multi)fractal scalings of the row-resolved IPRs. This is also manifest in the distributions PR and Prel not being converged with L. On these timescales, at intermediate disorder strengths preced- ing the MBL regime, r(t ) (or r (n)(t )) grows subdiffusively, ∼t β with β < 1/2, implying an anomalous t −β power-law decay of the real-space spin autocorrelation. In the MBL phase as well, there is a power-law envelope to the decay of the spin autocorrelation, but with clear signatures of the incipient saturation to a finite value, characteristic of that phase. (iii) Long times, t (cid:3) tH: This is the regime where the dynamics is essentially saturated and one sees the eigenstate properties. In the ergodic phase, the (multi)fractality gives way to a fully extended homogeneous state, both in terms of the IPR of the entire state as well as the row-resolved IPRs; and as also reflected in (cid:10)R(cid:9) saturating to an L-independent value, and similarly for the distributions PR and Prel. This is qualitatively different from the MBL regime, in which the (multi)fractality, for both the full state and at the row-resolved level, persists for arbitrarily long times. This is symptomatic of strongly inhomogeneous probability transport on the FS graph in the MBL phase, and is also manifest in PR exhibiting a heavy-tailed Lévy alpha-stable distribution. While the paper has presented quite a comprehensive pic- ture of probability transport on the FS graph of a disordered 094206-15 CREED, LOGAN, AND ROY PHYSICAL REVIEW B 107, 094206 (2023) Eq. (A1), Eq. (31) follows directly, expressed in terms of (real) eigenstate amplitudes (AmJ = (cid:10)J|m(cid:9)) and eigenvalues. Considering PIJ (t ) = Re[(cid:10)I|eiHt |J(cid:9)(cid:10)J|e−iHt |I(cid:9)], and in- (cid:7) serting the identity operator ˆ1 = |n(cid:9)(cid:10)n|, gives n N −1 H P(n) (A2) = |J(cid:9)(cid:10)J| |n(cid:9) + ˆOI ˆOJ : ˆOJ (t ))|n(cid:9) IJ (t ) = Re(cid:10)n| ˆOJ (t ) ˆOI = 1 (cid:10)n|( ˆOJ (t ) ˆOI 2 ˆOJ = |J(cid:9)(cid:10)J| thus defined (a so-called be- with the operator hemoth operator [75]); showing that P(n) IJ (t ) can equally be expressed as an eigenstate expectation value of the self-adjoint operator, ˆO(t ) = 1 2 ( ˆOJ (t ) ˆOI + ˆOI ˆOJ (t )). While any PIJ (t ) itself is non-negative for all t, we add that the same is not guaranteed for P(n) IJ (t ); although it is obvious from Eq. (31) that this does hold in the short- and long-time limits, for which N −1 nI and N −1 IJ (t = ∞) = A2 nI A2 nJ . In practice, this is, however, of little import, with quantities such as r(n)(t )/L shown in Fig. 5 found to be non-negative for all t, as expected physically. IJ (t = 0) = δIJ A2 H P(n) H P(n) APPENDIX B: HEISENBERG TIMES The mean level spacing at energy ω is [NHD(ω)]−1, with D(ω) the density of states/eigenvalues normalized to unity over ω. Reflecting the central limit theorem, D(ω) is known to be a Gaussian [34,53] with vanishing mean for the disor- ∝ L dered TFI model under consideration, and a variance μ2 E given exactly by [43] μ2 3W 2 + (cid:2)2]. E The Heisenberg time tH is the inverse of the mean level spacing. We consider it at the band center, ω = 0 (where it is largest), so tH = NHD(0) = NH/ 3 (δJ )2 + 1 = L[J 2 + 1 2π μ2 E and hence (cid:23) tH = (cid:23) 2π L (cid:5) J 2 + 1 2L 3 (δJ )2 + 1 3W 2 + (cid:2)2 (cid:6) . (B1) For all ED calculations, J = 1, δJ = 0.2 and (cid:2) ≡ 1 are fixed. tH obviously increases with L and decreases with disor- der strength W . For W = 1, 2, and L ∈ [8, 14], tH ranges from ∼20 to ∼103, while for W = 6, 7 it correspondingly ranges from ∼10 to ∼400. APPENDIX C: MBL0 We outline basic steps underlying the results given in Sec. V B 1 for MBL0, which corresponds to the noninteracting limit J(cid:5) = 0 of H, Eq. (1). The Hamiltonian in this case is site-separable, H = (cid:5) . The latter is diagonalized as H(cid:5), with H(cid:5) = h(cid:5) ˆσ z (cid:5) + (cid:2) ˆσ x L (cid:5)=1 (cid:7) (cid:23) H(cid:5) = φ(cid:5) ˆ˜σ z (cid:5) : φ(cid:5) = (cid:5) + (cid:2)2 h2 (C1) in terms of the spin-1/2 operator h(cid:5) ˆσ z (cid:23) (cid:5) = ˆ˜σ z (cid:5) + (cid:2) ˆσ x (cid:5) (cid:5) + (cid:2)2 h2 (C2) (such that [ ˆ˜σ z product state of the set of ˜σ spins, |n(cid:9) = |{ ˜σ z either +1 or −1. (cid:5) ]2 = 1). An eigenstate |n(cid:9) of H is simply a (cid:5) }(cid:9) with each ˜σ z (cid:5) Now consider the probability amplitude GIJ (t ) = (cid:10)J|e−iHt |I(cid:9) (with a general FS site |K(cid:9) ≡ |{S(cid:5),K }(cid:9) in the notation specified in Sec. II). Since H is site-separable, GIJ (t ) is a separable product, GIJ (t ) = (cid:10)J|e−iHt |I(cid:9) = L(cid:24) (cid:5)=1 (cid:10)S(cid:5),J |e−iH(cid:5)t |S(cid:5),I (cid:9), (C3) and e−iH(cid:5)t = cos(φ(cid:5)t ) − i ˆ˜σ z the product are readily evaluated, (cid:5) sin(φ(cid:5)t ). The matrix elements in (cid:10)S(cid:5),J ⎧ ⎨ (cid:9) = |e−iH(cid:5)t |S(cid:5),I cos(φ(cid:5)t ) − ih(cid:5)√ (cid:5) +(cid:2)2 h2 − i(cid:2)√ sin(φ(cid:5)t ) (cid:5) +(cid:2)2 h2 ⎩ S(cid:5),I sin(φ(cid:5)t ) : S(cid:5),J = S(cid:5),I (C4) : S(cid:5),J = −S(cid:5),I according to whether the local spin S(cid:5),J = ±S(cid:5),I . Let the FS sites J, I be separated by a Hamming distance rIJ = r. Then by definition r real-space sites have S(cid:5),J = −S(cid:5),I , while (L − r) sites have S(cid:5),J = +S(cid:5),I . Equations (C3) and (C4) then give ⎡ ⎤ GIJ (t ) = (cid:24) ⎢ ⎣ (cid:5)∈r (cid:23) −i(cid:2) (cid:5) + (cid:2)2 h2 ⎡ sin(φ(cid:5)t ) ⎥ ⎦ (cid:24) × (cid:5)∈(L−r) ⎢ ⎣cos(φ(cid:5)t ) − ih(cid:5)(cid:23) (cid:5) + (cid:2)2 h2 S(cid:5),I sin(φ(cid:5)t ) ⎤ ⎥ ⎦ in an obvious notation. From this (recalling [S(cid:5),I ]2 = 1) PIJ (t ) = |GIJ (t )|2 follows, (cid:16) (cid:24) PIJ (t ) = (cid:17) sin2(φ(cid:5)t ) (cid:2)2 (cid:5) + (cid:2)2 h2 (cid:16) 1 − (cid:24) (cid:5)∈r × (cid:5)∈(L−r) (cid:17) (C5) sin2(φ(cid:5)t ) . (cid:2)2 (cid:5) + (cid:2)2 h2 This can now be averaged over disorder realizations, and since the random fields {h(cid:5)} are i.i.d., (cid:19) PIJ (t ) = sin2(φ(cid:5)t ) (cid:21)r d (cid:2)2 (cid:5) + (cid:2)2 h2 (cid:19) × 1 − (cid:21)(L−r) (cid:2)2 (cid:5) + (cid:2)2 h2 sin2(φ(cid:5)t ) , (C6) d which is Eqs. (33) and (34) as required. Equation (C6) is indeed seen to depend solely on the Hamming distance rIJ = r between FS sites J, I; such that, from Eq. (6), P(r; t ) ≡ PI (r; t ) ≡ (cid:3) PIJ (t ), as given explicitly in Eq. (35). Equation (35) can obviously be cast in the form (cid:14) (cid:2) L r (cid:13) P(r; t ) = [1 + e−1/ξ 0 F (t )]−Le−r/ξ 0 F (t ) (C7) L r in terms of a correlation length ξ 0 ln( 1 z0 (t ) F (t ) = − 1). This is the MBL0 counterpart of the short-time F (t ) defined by 1/ξ 0 094206-16 PROBABILITY TRANSPORT ON THE FOCK SPACE OF A … PHYSICAL REVIEW B 107, 094206 (2023) (cid:3) result Eq. (24) [in the latter case, P(r; t ) ≡ PIJr (t )]. Equa- tion (24) itself is of course general—in the sense that it holds for all interaction and disorder strengths—and Eq. (C7) cor- rectly reduces to it for (cid:2)t (cid:2) 1. (cid:2) L r We also point out the connection between the long-time limit P(r; t = ∞) of Eq. (C7), and the eigenstate corre- lation function F n(r) defined generally by Eq. (28b) and given in terms of FS correlation lengths ξF,n for eigenstates n by Eq. (29). P(r; ∞) is given generally by P(r; ∞) = N −1 n F n(r). For MBL0 one can however show that the H (cid:7) nI A2 nJ disorder-averaged A2 is independent of the particular eigenstate n. From Eq. (28b), F n(r) is thus independent of n, whence P(r; ∞) ≡ F n(r) gives the connection sought. (cid:2) L r dynamical correlation length ξ 0 eigenstate correlation length, viz., ξ 0 explicitly [using Eq. (37)] by ξ 0 ξ 0 F (∞) ∝ 1/ ln W for W (cid:16) 1. Comparison of Eq. (C7) for t = ∞ to Eq. (29), F n(r) = (cid:3) (1 + e−1/ξF,n )−Le−r/ξF,n , then relates directly the infinite-t F (∞) to the (n-independent) F (t = ∞) ≡ ξF,n; given − 1)]−1, with F (∞) = [ln( 1 p [1] D. M. Basko, I. L. Aleiner, and B. L. Altshuler, Metal– insulator transition in a weakly interacting many-electron sys- tem with localized single-particle states, Ann. Phys. 321, 1126 (2006). [2] I. V. Gornyi, A. D. Mirlin, and D. G. Polyakov, Interacting Elec- trons in Disordered Wires: Anderson Localization and Low-T Transport, Phys. Rev. Lett. 95, 206603 (2005). [3] V. Oganesyan and D. A. Huse, Localization of interact- ing fermions at high temperature, Phys. Rev. B 75, 155111 (2007). [4] M. Žnidariˇc, T. Prosen, and P. Prelovšek, Many-body localiza- tion in the Heisenberg XXZ magnet in a random field, Phys. Rev. B 77, 064426 (2008). [5] Y. Bar Lev, G. Cohen, and D. R. Reichman, Absence of Diffusion in an Interacting System of Spinless Fermions on a One-Dimensional Disordered Lattice, Phys. Rev. Lett. 114, 100601 (2015). [6] R. Nandkishore and D. A. Huse, Many-body localization and thermalization in quantum statistical mechanics, Annu. Rev. Condens. Matter Phys. 6, 15 (2015). [7] D. A. Abanin and Z. Papi´c, Recent progress in many-body localization, Ann. Phys. 529, 1700169 (2017). [8] D. A. Abanin, E. Altman, I. Bloch, and M. Serbyn, Colloquium: thermalization, and entanglement, Many-body localization, Rev. Mod. Phys. 91, 021001 (2019). [9] A. Pal and D. A. Huse, Many-body localization phase transition, Phys. Rev. B 82, 174411 (2010). [17] D. J. Luitz and Y. Bar Lev, The ergodic side of the many-body localization transition, Ann. Phys. 529, 1600350 (2017). [18] S. Roy, Y. Bar Lev, and D. J. Luitz, Anomalous thermalization and transport in disordered interacting Floquet systems, Phys. Rev. B 98, 060201(R) (2018). [19] D. A. Huse, R. Nandkishore, and V. Oganesyan, Phenomenol- ogy of fully many-body-localized systems, Phys. Rev. B 90, 174202 (2014). [20] M. Serbyn, Z. Papi´c, and D. A. Abanin, Local Conservation Laws and the Structure of the Many-Body Localized States, Phys. Rev. Lett. 111, 127201 (2013). [21] V. Ros, M. Müller, and A. Scardicchio, Integrals of motion in the many-body localized phase, Nucl. Phys. B 891, 420 (2015). [22] S. J. Garratt, S. Roy, and J. T. Chalker, Local resonances and parametric level dynamics in the many-body localized phase, Phys. Rev. B 104, 184203 (2021). [23] A. Morningstar, L. Colmenarez, V. Khemani, D. J. Luitz, and D. A. Huse, Avalanches and many-body resonances in many-body localized systems, Phys. Rev. B 105, 174205 (2022). [24] R. Ghosh and M. Žnidariˇc, Resonance-induced growth of num- ber entropy in strongly disordered systems, Phys. Rev. B 105, 144203 (2022). [25] S. J. Garratt and S. Roy, Resonant energy scales and local observables in the many-body localized phase, Phys. Rev. B 106, 054309 (2022). [10] D. J. Luitz, N. Laflorencie, and F. Alet, Many-body localization edge in the random-field Heisenberg chain, Phys. Rev. B 91, 081103(R) (2015). [26] P. J. D. Crowley and A. Chandran, A constructive theory of the numerically accessible many-body localized to thermal crossover, SciPost Phys. 12, 201 (2022). [11] M. Serbyn, Z. Papi´c, and D. A. Abanin, Quantum quenches in the many-body localized phase, Phys. Rev. B 90, 174302 (2014). [27] D. M. Long, P. J. D. Crowley, V. Khemani, and A. Chandran, Phenomenology of the prethermal many-body localized regime, arXiv:2207.05761. [12] D. J. Luitz, N. Laflorencie, and F. Alet, Extended slow dynami- cal regime close to the many-body localization transition, Phys. Rev. B 93, 060201(R) (2016). [28] B. L. Altshuler, Y. Gefen, A. Kamenev, and L. S. Levitov, Quasiparticle Lifetime in a Finite System: A Nonperturbative Approach, Phys. Rev. Lett. 78, 2803 (1997). [13] K. Agarwal, S. Gopalakrishnan, M. Knap, M. Müller, and E. Demler, Anomalous Diffusion and Griffiths Effects Near the Many-Body Localization Transition, Phys. Rev. Lett. 114, 160401 (2015). [14] E. J. Torres-Herrera and L. F. Santos, Dynamics at the many- body localization transition, Phys. Rev. B 92, 014208 (2015). [15] D. J. Luitz, Long tail distributions near the many-body localiza- tion transition, Phys. Rev. B 93, 134201 (2016). [16] D. J. Luitz and Y. Bar Lev, Anomalous Thermalization in Er- godic Systems, Phys. Rev. Lett. 117, 170404 (2016). [29] C. Monthus and T. Garel, Many-body localization transition in a lattice model of interacting fermions: Statistics of renormal- ized hoppings in configuration space, Phys. Rev. B 81, 134202 (2010). [30] A. De Luca and A. Scardicchio, Ergodicity breaking in a model showing many-body localization, Europhys. Lett. 101, 37003 (2013). [31] M. Serbyn, Z. Papi´c, and D. A. Abanin, Criterion for Many- Body Localization-Delocalization Phase Transition, Phys. Rev. X 5, 041047 (2015). 094206-17 CREED, LOGAN, AND ROY PHYSICAL REVIEW B 107, 094206 (2023) [32] F. Pietracaprina, V. Ros, and A. Scardicchio, Forward approx- imation as a mean-field approximation for the Anderson and many-body localization transitions, Phys. Rev. B 93, 054201 (2016). [33] C. L. Baldwin, C. R. Laumann, A. Pal, and A. Scardicchio, The many-body localized phase of the quantum random energy model, Phys. Rev. B 93, 024202 (2016). [34] D. E. Logan and S. Welsh, Many-body localization in Fock space: A local perspective, Phys. Rev. B 99, 045131 (2019). [35] S. Roy, D. E. Logan, and J. T. Chalker, Exact solution of a percolation analog for the many-body localization transition, Phys. Rev. B 99, 220201(R) (2019). [36] S. Roy, J. T. Chalker, and D. E. Logan, Percolation in Fock space as a proxy for many-body localization, Phys. Rev. B 99, 104206 (2019). [37] S. Roy and D. E. Logan, Self-consistent theory of many-body localisation in a quantum spin chain with long-range interac- tions, SciPost Phys. 7, 042 (2019). multifractality across the many-body localization transition, Phys. Rev. B 106, 054203 (2022). [51] S. Roy, Hilbert-space correlations beyond multifractality and bipartite entanglement in many-body localized systems, Phys. Rev. B 106, L140204 (2022). [52] Y. Yao, L. Xiang, Z. Guo, Z. Bao, Y.-F. Yang, Z. Song, H. Shi, X. Zhu, F. Jin, J. Chen et al., Observation of many-body Fock space dynamics in two dimensions, arXiv:2211.05803. [53] S. Welsh and D. E. Logan, Simple probability distributions on a Fock-space lattice, J. Phys.: Condens. Matter 30, 405601 (2018). [54] The Hamming distance between two classical spin configura- tions is simply the number of spins that are different between the two. [55] D. Abanin, J. Bardarson, G. De Tomasi, S. Gopalakrishnan, V. Khemani, S. Parameswaran, F. Pollmann, A. Potter, M. Serbyn, and R. Vasseur, Distinguishing localization from chaos: Challenges in finite-size systems, Ann. Phys. 427, 168415 (2021). [38] N. Macé, F. Alet, and N. Laflorencie, Multifractal Scalings Across the Many-Body Localization Transition, Phys. Rev. Lett. 123, 180601 (2019). [56] J. Šuntajs, J. Bonˇca, T. Prosen, and L. Vidmar, Quantum chaos challenges many-body localization, Phys. Rev. E 102, 062144 (2020). [39] F. Pietracaprina and N. Laflorencie, Hilbert-space fragmenta- tion, multifractality, and many-body localization, Ann. Phys. 435, 168502 (2021). [57] J. Šuntajs, J. Bonˇca, T. Prosen, and L. Vidmar, Ergodicity break- ing transition in finite disordered spin chains, Phys. Rev. B 102, 064207 (2020). [40] G. De Tomasi, D. Hetterich, P. Sala, and F. Pollmann, Dynamics of strongly interacting systems: From Fock-space fragmenta- tion to many-body localization, Phys. Rev. B 100, 214313 (2019). [41] S. Ghosh, A. Acharya, S. Sahu, and S. Mukerjee, Many-body localization due to correlated disorder in Fock space, Phys. Rev. B 99, 165131 (2019). [58] P. Sierant, D. Delande, and J. Zakrzewski, Thouless Time Anal- ysis of Anderson and Many-Body Localization Transitions, Phys. Rev. Lett. 124, 186601 (2020). [59] M. Kiefer-Emmanouilidis, R. Unanyan, M. Fleischhauer, and J. Sirker, Evidence for Unbounded Growth of the Number En- tropy in Many-Body Localized Phases, Phys. Rev. Lett. 124, 243601 (2020). [42] S. Nag and A. Garg, Many-body localization in the presence of long-range interactions and long-range hopping, Phys. Rev. B 99, 224203 (2019). [60] D. Sels and A. Polkovnikov, Dynamical obstruction to local- ization in a disordered spin chain, Phys. Rev. E 104, 054105 (2021). [43] S. Roy and D. E. Logan, Fock-space correlations and the ori- gins of many-body localization, Phys. Rev. B 101, 134202 (2020). [61] M. Kiefer-Emmanouilidis, R. Unanyan, M. Fleischhauer, and J. Sirker, Slow delocalization of particles in many-body localized phases, Phys. Rev. B 103, 024203 (2021). [44] G. Biroli and M. Tarzia, Anomalous dynamics on the ergodic side of the many-body localization transition and the glassy phase of directed polymers in random media, Phys. Rev. B 102, 064211 (2020). [45] M. Tarzia, Many-body localization transition in Hilbert space, Phys. Rev. B 102, 014208 (2020). [46] G. De Tomasi, I. M. Khaymovich, F. Pollmann, and S. Warzel, Rare thermal bubbles at the many-body localization transition from the Fock space point of view, Phys. Rev. B 104, 024202 (2021). [47] M. Hopjan and F. Heidrich-Meisner, Many-body localiza- tion from a one-particle perspective in the disordered one- dimensional Bose-Hubbard model, Phys. Rev. A 101, 063617 (2020). [48] K. S. Tikhonov and A. D. Mirlin, Eigenstate correlations around the many-body localization transition, Phys. Rev. B 103, 064204 (2021). [49] S. Roy and D. E. Logan, Fock-space anatomy of eigenstates across the many-body localization transition, Phys. Rev. B 104, 174201 (2021). [50] J. Sutradhar, S. Ghosh, S. Roy, D. E. Logan, S. Mukerjee, and S. Banerjee, Scaling of the Fock-space propagator and [62] D. Sels, Bath-induced interacting disordered spin chains, Phys. Rev. B 106, L020202 (2022). delocalization in [63] M. Schreiber, S. S. Hodgman, P. Bordia, H. P. Lüschen, M. H. Fischer, R. Vosk, E. Altman, U. Schneider, and I. Bloch, Observation of many-body localization of interacting fermions in a quasirandom optical lattice, Science 349, 842 (2015). [64] J. Smith, A. Lee, P. Richerme, B. Neyenhuis, P. W. Hess, P. Hauke, M. Heyl, D. A. Huse, and C. Monroe, Many-body localization in a quantum simulator with programmable random disorder, Nat. Phys. 12, 907 (2016). [65] E. J. Torres-Herrera, A. M. García-García, and L. F. Santos, Generic dynamical features of quenched interacting quan- tum systems: Survival probability, density imbalance, and out-of-time-ordered correlator, Phys. Rev. B 97, 060303(R) (2018). [66] M. Schiulaz, E. J. Torres-Herrera, and L. F. Santos, Thouless and relaxation time scales in many-body quantum systems, Phys. Rev. B 99, 174313 (2019). [67] That τq(t = ∞) = q − 1 in this case is easily seen from an ar- gument applicable deep in an ergodic phase, where eigenstates 094206-18 PROBABILITY TRANSPORT ON THE FOCK SPACE OF A … PHYSICAL REVIEW B 107, 094206 (2023) (cid:7) are effectively Gaussian random vectors (grv’s). From Eq. (5), ∼ 1/NH for the PIJ (∞) = nJ quite generally, with A2 nI case of grv’s. Hence PIJ (∞) ∼ 1/NH, from which LI,q(∞) = (cid:7) . IJ (∞) ∼ N 1−q H = N −τq (∞) H nI A2 n A2 J Pq [68] T. L. M. Lezama, S. Bera, and J. H. Bardarson, Apparent slow dynamics in the ergodic phase of a driven many-body localized system without extensive conserved quantities, Phys. Rev. B 99, 161106(R) (2019). [69] J. H. Bardarson, F. Pollmann, J. E. Moore, Unbounded Growth of in Models Many-Body Localization, Phys. Rev. Lett. 109, 017202 (2012). and of Entanglement [70] M. Serbyn, Z. Papi´c, Slow Growth of Entanglement Disordered (2013). Systems, Phys. Rev. Lett. and D. A. Abanin, Universal in Interacting Strongly 110, 260601 [71] T. L. M. Lezama and D. J. Luitz, Power-law entanglement growth from typical product states, Phys. Rev. Res. 1, 033067 (2019). [72] V. Khemani, D. N. Sheng, and D. A. Huse, Two Universality Classes for the Many-Body Localization Transition, Phys. Rev. Lett. 119, 075702 (2017). [73] V. Khemani, S. P. Lim, D. N. Sheng, and D. A. Huse, Critical Properties of the Many-Body Localization Transition, Phys. Rev. X 7, 021013 (2017). [74] N. Roy, J. Sutradhar, and S. Banerjee, Diagnostics of non- ergodic extended states and many body localization prox- imity effect through real-space and Fock-space excitations, arXiv:2208.10714. [75] I. M. Khaymovich, M. Haque, and P. A. McClarty, Eigenstate Thermalization, Random Matrix Theory, and Behemoths, Phys. Rev. Lett. 122, 070601 (2019). 094206-19
null
10.1371_journal.ppat.1010103.pdf
Data Availability Statement: The RNA-seq datasets generated during this study are available at Bioproject accession number PRJNA742496 in the NCBI Bioproject database (http://www.ncbi. nlm.nih.gov/bioproject/742496).
The RNA-seq datasets generated during this study are available at Bioproject accession number PRJNA742496 in the NCBI Bioproject database ( http://www.ncbi. nlm.nih.gov/bioproject/742496 ).
RESEARCH ARTICLE γδ T cell IFNγ production is directly subverted by Yersinia pseudotuberculosis outer protein YopJ in mice and humans Timothy H. ChuID Yue ZhangID Vincent W. Yang3, James B. Bliska6, Brian S. SheridanID 1,2, Camille KhairallahID 1,2, Onur Eskiocak4, David G. Thanassi1,2, Mark H. KaplanID 1,2* 1,2, Jason Shieh3, Rhea Cho1,2, Zhijuan QiuID 1,2, 5, Semir Beyaz4, a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 1 Department of Microbiology and Immunology, Renaissance School of Medicine, Stony Brook University, Stony Brook, New York, United States of America, 2 Center for Infectious Diseases, Renaissance School of Medicine, Stony Brook University, Stony Brook, New York, United States of America, 3 Department of Medicine, Renaissance School of Medicine, Stony Brook University, Stony Brook, New York, United States of America, 4 Cold Spring Harbor Laboratory, Cold Spring Harbor, New York, United States of America, 5 Department of Microbiology and Immunology, School of Medicine, Indiana University, Indianapolis, Indiana, United States of America, 6 Department of Microbiology and Immunology, Geisel School of Medicine, Dartmouth College, Dartmouth, New Hampshire, United States of America OPEN ACCESS Citation: Chu TH, Khairallah C, Shieh J, Cho R, Qiu Z, Zhang Y, et al. (2021) γδ T cell IFNγ production is directly subverted by Yersinia pseudotuberculosis outer protein YopJ in mice and humans. PLoS Pathog 17(12): e1010103. https:// doi.org/10.1371/journal.ppat.1010103 Editor: Denise M. Monack, Stanford University School of Medicine, UNITED STATES Received: July 29, 2021 Accepted: November 9, 2021 Published: December 6, 2021 Copyright: © 2021 Chu et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The RNA-seq datasets generated during this study are available at Bioproject accession number PRJNA742496 in the NCBI Bioproject database (http://www.ncbi. nlm.nih.gov/bioproject/742496). Funding: This work was supported by The G. Harold and Leila Y. Mathers Foundation grant MF- 1901-00210 (B.S.S.), the NIH grants T32 AI007539 (T.H.C.), R01 AI099222 (J.B.B.), K12 GM102778 (Z.Q.), and R01 AI141633 (D.G.T.), and funds provided by The Research Foundation for the State * brian.sheridan@stonybrook.edu Abstract Yersinia pseudotuberculosis is a foodborne pathogen that subverts immune function by translocation of Yersinia outer protein (Yop) effectors into host cells. As adaptive γδ T cells protect the intestinal mucosa from pathogen invasion, we assessed whether Y. pseudotu- berculosis subverts these cells in mice and humans. Tracking Yop translocation revealed that the preferential delivery of Yop effectors directly into murine Vγ4 and human Vδ2+ T cells inhibited anti-microbial IFNγ production. Subversion was mediated by the adhesin YadA, injectisome component YopB, and translocated YopJ effector. A broad anti-pathogen gene signature and STAT4 phosphorylation levels were inhibited by translocated YopJ. Thus, Y. pseudotuberculosis attachment and translocation of YopJ directly into adaptive γδ T cells is a major mechanism of immune subversion in mice and humans. This study uncov- ered a conserved Y. pseudotuberculosis pathway that subverts adaptive γδ T cell function to promote pathogenicity. Author summary Unconventional γδ T cells are a dynamic immune population important for mucosal pro- tection of the intestine against invading pathogens. We determined that the foodborne pathogen Y. pseudotuberculosis preferentially targets an adaptive subset of these cells to subvert immune function. We found that direct injection of Yersinia outer proteins (Yop) into adaptive γδ T cells inhibited their anti-pathogen functions. We screened all Yop effec- tors and identified YopJ as the sole effector to inhibit adaptive γδ T cell production of IFNγ. We determined that adaptive γδ T cell subversion occurred by limiting activation of the transcription factor STAT4. When we infected mice with Y. pseudotuberculosis PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 1 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function University of New York (B.S.S.) and Stony Brook University (B.S.S.). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. expressing an inactive YopJ, this enhanced the adaptive γδ T cell response and led to greater cytokine production from this subset of cells to aid mouse recovery. This mecha- nism of immune evasion appears conserved in humans as direct injection of Y. pseudotu- berculosis YopJ into human γδ T cells inhibited cytokine production. This suggested to us that Y. pseudotuberculosis actively inhibits the adaptive γδ T cell response through YopJ as a mechanism to evade immune surveillance at the site of pathogen invasion. Introduction Pathogens in the genus Yersinia include three species (Y. pestis, Y. pseudotuberculosis, and Y. enterocolitica) that can cause human disease. Y. pseudotuberculosis and Y. enterocolitica cause enteric infections [1,2] while Y. pestis is the causative agent of bubonic, septicemic, and pneu- monic plague that has claimed over 200 million human lives [3,4]. Bubonic and septicemic plague is transmitted by blood sucking fleas while aerosols spread the pneumonic plague. Despite vaccine availability [5], and sensitivity to antibiotic treatment, pneumonic plague com- monly results in fatality in part due to the rapid course of the infection [6]. Pathogenic Yersinia spp. harbor a virulence plasmid that encodes numerous virulence fac- tors to subvert host immune responses, including IFNγ production [7–9]. Immune cell subver- sion requires Yersinia adherence to host cells through bacterial adhesins and translocation of Yersinia outer proteins (Yop) effectors into the host cell cytoplasm by a type III secretion sys- tem (T3SS). Yersinia spp. predominately target host phagocytes like macrophages, dendritic cells (DC), neutrophils, and B cells to subvert immune function during infection, but injection into other immune populations like conventional T cells has been reported, albeit to a lesser degree than their phagocytic counterparts [10–12]. Yersinia virulence factors include compo- nents of the T3SS (e.g., YopB) and translocated effectors (e.g., YopJ and YopH). YopB forms a pore in the host cell membrane necessary for translocation of Yop effectors [13,14]. Numerous Yop effectors translocate into host cells to inhibit immune responses and promote Yersinia spp. pathogenesis. One notable example is YopJ, an acetyl transferase and a possible cysteine protease that inhibits the mitogen-activated protein kinase (MAPK) pathway and tumor necrosis factor receptor-associated factor (TRAF) ubiquitination [15–19]. YopJ is the major Yop effector responsible for the induction of pyroptosis in macrophages during infection [20] and limits toll-like receptor 4 (TLR4) dependent signaling pathways [21]. While YopJ has no known direct effects on conventional T cell activation, YopP (a YopJ homolog in Y. enterocoli- tica) indirectly inhibits T cell priming via DC subversion [22]. YopH has been reported to have direct effects on conventional T cells in vitro. Transfection of a YopH expression plasmid into Jurkat or human T cells inhibited T cell receptor (TCR) signaling and promoted T cell apoptosis [23,24]. Additionally, stimulation of Jurkat cells with a YopH deficient Y. pseudotu- berculosis restored T cell signaling and IL-2 production [25,26]. Even in this context, it is nota- ble that many of the downstream targets in the αβ TCR signaling pathway were inhibited at an excessively high (>50) multiplicity of infection (MOI) and in vivo relevance is unclear [23,26]. Thus, the role of direct subversion of T cell function, especially among unconventional T cells, by Yersinia spp. remains largely unexplored. γδ T cells make up a large proportion of lymphocytes at barrier surfaces and mucosal tissues including the intestines of mice and humans [27,28]. This is particularly pertinent to infections caused by Y. pseudotuberculosis, which has evolved to invade the intestinal barrier. The activity of γδ T cells can be modulated by numerous cell-intrinsic and environmental factors like the γδ TCR, cytokines, and co-stimulatory or inhibitory receptors [29]. For example, IL-12 and PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 2 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function IL-18 may promote IFNγ production from some γδ T cell subsets whereas IL-1β and IL-23 predominantly drive IL-17A production from other γδ T cell subsets [30–35]. Vγ4Vδ1 (Gar- man nomenclature [36]) T cells have traditionally been considered an innate-like cell. How- ever, our group recently characterized a long-lived CD27- CD44hi Vγ4Vδ1 T cell memory population in the context of foodborne Listeria monocytogenes infection [37,38]. While Vγ4 T cells are typically programmed for IL-17A production, this subset has the multifunctional capacity to produce both IL-17A and IFNγ [37]. Similar observations of IFNγ production were made in clonally expanded Vγ4 T cells in response to Staphylococcus aureus in the skin [39]. IFNγ activates macrophages to kill intracellular pathogens or phagocytosed bacteria and induces chemokines that attract immune cells to the site of infection. IFNγ is a critical cytokine in protection from Y. enterocolitica infection [2,40], Y. pestis intranasal challenge [41], and associated with protection from Y. pseudotuberculosis [42]. Interestingly, IFNγ but not IL-17A production from type-3 innate lymphoid cells is critical for the control of foodborne Y. entero- colitica infection [43]. As such, unconventional T cells like Vγ4 T cells that are ideally placed to provide protection against pathogen invasion at mucosal sites may be particularly relevant to Yersinia infections that invade mucosal barriers of the lungs (pneumonic Y. pestis) and gut (Y. pseudotuberculosis and Y. enterocolitica). Despite a foundational understanding of Yersinia pathogenesis, physiologically robust evi- dence linking Yersinia pathogenesis to direct subversion of T cell function is lacking. Here, we uncovered a novel YopJ-dependent immunomodulatory pathway used by Y. pseudotuberculo- sis to directly subvert a murine Vγ4Vδ1 anti-microbial response to aid Y. pseudotuberculosis pathogenesis. Y. pseudotuberculosis also directly subverted a human Vδ2+ T cell IFNγ response, suggesting that this pathway may function similarly in human infection to aid Y. pseudotuberculosis pathogenesis. Results Viable Y. pseudotuberculosis inhibits IFNγ production by adaptive γδ T cells in a YopB- and YadA-dependent manner Initial experiments were carried out to determine if Y. pseudotuberculosis inhibits adaptive γδ T cell function ex vivo. To overcome the extremely low number of Vγ4 T cells in gut-associated lymphoid tissues of naïve specific pathogen free (SPF) mice, a previously established in vivo methodology was utilized to generate a sizable population of adaptive Vγ4 T cells for in vitro manipulation. As such, naïve Balb/c mice were exposed to foodborne L. monocytogenes and MLN enriched in adaptive γδ T cells were isolated 9 days after infection [37], several days after mice typically clear L. monocytogenes [44]. MLN single cell suspensions were infected directly ex vivo with heat-killed or live wild-type (WT) Y. pseudotuberculosis (Yptb) 32777 (Table 1) at a multiplicity of infection (MOI) of 10 for 2 hours. Antibiotics were then added to prevent overgrowth of the live bacteria, and the cultures were incubated an additional 22 hours. Flow cytometry in conjunction with intracellular cytokine staining was used to assess IFNγ produc- tion from Vγ1.1/2- CD44hi CD27- γδ T cells (identifying the adaptive Vγ4 T cell subset [37,38]). Heat-killed Y. pseudotuberculosis elicited a significantly higher IFNγ response from adaptive Vγ4 T cells than was detectable after stimulation with live Y. pseudotuberculosis (Fig 1A). This observation suggests that live Y. pseudotuberculosis subverts adaptive Vγ4 T cell function. The virulence activity of Y. pseudotuberculosis relies substantially on its T3SS and translocation of Yop effectors into host cells. To determine if the T3SS is required for live Y. pseudotuberculosis inhibition of γδ T cell function, MLN single cell suspensions were infected with WT Y. pseudotuberculosis or Y. pseudotuberculosis that were unable to translocate Yop effectors (ΔYopB) or lacked the virulence plasmid that encodes the T3SS (32777c) (Table 1) PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 3 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function Table 1. Y. pseudotuberculosis strains and mutants used in this study. Y. pseudotuberculosis Notation Relevant Characteristics 32777 32777c 32777 YopJC172A 32777 YopHR409A 32777 ΔYopB 32777 YopER144A 32777 YopTC139A 32777 ΔYopM 32777 ΔYpkA 32777 ΔYopK 32777 YopE/β-lac 32777 ΔYopB YopE/β-lac IP2666 IP40 IP2666 ΔInv IP2666 ΔYadA IP2666 ΔInv ΔYadA IP40 ΔInv ΔYadA WT WT2777c YopJC172A YopHR409A ΔYopB YopER144A YopTC139A ΔYopM ΔYpkA ΔYopK WT Yptb-βla ΔYopB Yptb-βla WT ΔYopB ΔInv ΔYadA ΔInv ΔYadA Yptb wild-type serogroup O:1 strain Virulence pYV-cured derivative of 32777 that lacks the T3SS Catalytically inactive YopJ Catalytically inactive YopH Deletion of YopB Catalytically inactive YopE Catalytically inactive YopT Deletion of YopM Deletion of YpkA Frameshift mutation in YopK YopE TME-1 β-lactamase fusion protein Deletion of YopB in the YopE/β-lac Yptb wild-type serogroup O:3 strain IP2666 yopB40 (a stop codon at codon 8 of YopB followed by a frameshift) Deletion of adhesin and invasin Deletion of adhesin and YadA Deletion of adhesin, invasion, and YadA ΔYopB ΔInv ΔYadA Deletion of invasion and YadA in IP40 and pMMB207 mCherry https://doi.org/10.1371/journal.ppat.1010103.t001 References [47] [47] [45] [45,48,49] [45,46] [45] [45] [50] [51] [52] [53–55] [53–55] [47] [56] [12,57] [12,57] [12,57] [12,57] [45–47]. γδ T cell function was assessed 24 hours later as described above. Stimulation with live Y. pseudotuberculosis strains ΔYopB or 32777c restored the IFNγ response of Vγ1.1/2- CD44hi CD27- γδ T cells (Fig 1B), similar to levels seen after stimulation with heat-killed WT Y. pseudotuberculosis (Fig 1A). These data indicate that Y. pseudotuberculosis inhibits IFNγ production by Vγ4 T cells in a manner that requires the T3SS and translocation of Yop effectors. Y. pseudotuberculosis adheres to host cells with the bacterial adhesins invasin (Inv) and YadA to translocate effectors through the T3SS [58–60]. For Y. enterocolitica, both Inv and YadA bind β1-integrin either directly or indirectly through the extracellular matrix, respec- tively [61]. Additionally, β1-integrin expressed on host cells is a known adhesion target for Y. pseudotuberculosis [60,62]. To evaluate the role of these adhesins in the inhibition of γδ T cell function, live Y. pseudotuberculosis with a deletion of Inv (ΔInv), YadA (ΔYadA), or both (ΔInv ΔYadA) (Table 1) were utilized to infect MLN cell suspensions. Vγ1.1/2- CD44hi CD27- γδ T cells stimulated with ΔInv bacteria produced only minimal IFNγ, comparable to unstimu- lated cells or cells stimulated with WT (Fig 1C). In contrast, ΔYadA or ΔInv ΔYadA stimula- tion led to partial restoration of IFNγ production, and stimulation with ΔYopB or ΔYopB ΔInv ΔYadA bacteria led to full restoration of IFNγ production (Fig 1C). Thus, YadA but not Inv contributes to translocation dependent inhibition of IFNγ production by Vγ4 T cells. Translocation of Yop effectors into adaptive γδ T cells by Y. pseudotuberculosis is associated with IFNγ inhibition To determine if Y. pseudotuberculosis can translocate Yop effectors into adaptive Vγ4 T cells, a WT strain expressing a YopE-β-lactamase fusion protein (Yptb-βla) in conjunction with a FRET-based β-lactamase reporter assay was used [63]. YopE translocation into target cells can be readily assessed by a change in fluorescence using flow cytometry. Thus, translocation of the YopE-β-lactamase fusion protein reports Yop effector translocation by emission in the blue range (Yop+) or lack thereof by emission in the green range (Yop-) [10,64,65]. A YopB PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 4 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function Fig 1. Y. pseudotuberculosis inhibition of Vγ1.1/2- CD44hi CD27- γδ T cell function is YopB- and YadA- dependent. MLN cell suspensions from L. monocytogenes infected Balb/c mice were left unstimulated or stimulated with 10 MOI of the indicated Y. pseudotuberculosis for 24 hours. Antibiotics were given 2 hours post-stimulation and brefeldin A was added for the last 5 hours of stimulation. Vγ1.1/2- CD44hi CD27- γδ T cells were analyzed for IFNγ production. Representative histograms are displayed. (A) Cells were stimulated with live or heat-killed (HK) wild-type (WT) Y. pseudotuberculosis. The graph depicts the mean ± SEM and represents at least two independent experiments with 4 mice/group/experiment. (B) Cells were stimulated with live WT, ΔYopB, or 32777c Y. pseudotuberculosis. The graph depicts the mean ± SEM and represents at least two independent experiments with 4 mice/group/experiment. (C) Cells were stimulated with WT, ΔYopB, ΔYadA, ΔInv, ΔInv ΔYadA, or ΔYopB ΔInv ΔYadA Y. pseudotuberculosis. The graph depicts the mean ± SEM pooled from two independent experiments with 3 mice/group/experiment. ���p < 0.0001, ��p < 0.01, and �p < 0.05. An unpaired t-test was used for (A) and a repeated measures one-way ANOVA was used for (B) and (C). Experimental groups were compared to live WT Y. pseudotuberculosis in (A) and WT Y. pseudotuberculosis in (B) and (C). https://doi.org/10.1371/journal.ppat.1010103.g001 deficient β-lactamase Y. pseudotuberculosis reporter (ΔYopB Yptb-βla) was used as a transloca- tion deficient control. Stimulation of MLN cell suspensions with WT or ΔYopB Yptb-βla con- firmed reporter activity at various MOI (S1A and S1B Fig). Two hours post stimulation with WT Yptb-βla, the majority of Vγ1.1/2- CD44hi CD27- γδ T cells were positive for Yop effector PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 5 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function translocation (Fig 2A). Yop translocation into Vγ1.1/2- CD44hi CD27- γδ T cells was compara- ble to known DC and macrophage targets (Fig 2A). Additionally, Yop translocation was more efficient into Vγ1.1/2- CD44hi CD27- γδ T cells than CD4 or CD8 T cells (Fig 2A). Y. pseudotu- berculosis also preferentially targeted Vγ1.1/2- CD44hi CD27- γδ T cells over CD44- γδ T cells and activated phenotype CD4 or CD8 T cells for Yop translocation (S1C Fig). YadA and Inv promote Yersinia adherence by direct or indirect interactions with the β1-integrin [66–69]. Analysis of β1-integrin expression on Vγ1.1/2- CD44hi CD27- γδ T cells, CD4 T cells, and CD8 T cell revealed that most Vγ1.1/2- CD44hi CD27- γδ T cells expressed the β1-integrin (S2A Fig). In contrast, most conventional CD4 and CD8 T cells did not express the β1-integrin. In addition, use of the WT Yptb-βla reporter for Yop translocation demonstrated that Yop trans- location was associated with higher β1-integrin expression among γδ T cells (S2B Fig). Thus, Y. pseudotuberculosis selectively targets adaptive γδ T cells for Yop translocation among a diverse group of immune populations assessed in an ex vivo culture system. As adaptive Vγ4 T cells were directly targeted with Yop effector translocation, WT Yptb-βla was utilized to determine whether Vγ1.1/2- CD44hi CD27- γδ T cells that contained Yop effec- tors were functionally impaired. An MOI of 1 was used as it provided similarly sized popula- tions of Vγ1.1/2- CD44hi CD27- γδ T cells that did or did not contain translocated effectors from the same culture conditions (S1B Fig). Among WT Yptb-βla stimulated cells, Yop+ Vγ1.1/2- CD44hi CD27- γδ T cells had reduced IFNγ production as compared to their Yop- counterparts (Fig 2B). To extend these results, the ability of Y. pseudotuberculosis to translocate Yop effectors into human γδ T cells and inhibit IFNγ production was assessed in peripheral blood mononuclear cells (PBMC) cultures stimulated with the WT Yptb-βla reporter. Approx- imately 8% of human Vδ2+ T cells were Yop+ and these cells had significantly reduced IFNγ production as compared to the Yop- counterparts (Fig 2C and 2D). These data indicate that Y. pseudotuberculosis is capable of translocating Yop effectors into γδ T cell subsets and inhibiting IFNγ production in mice and humans. YopJ is necessary for Y. pseudotuberculosis to inhibit IFNγ production in adaptive γδ T cells As multiple effectors are translocated into target cells, a panel of yop mutant Y. pseudotubercu- losis (Table 1) [45] was screened to determine if individual Yop effectors inhibit IFNγ produc- tion. Similar to the ΔYopB mutant, stimulation with a catalytically inactive YopJ (YopJC172A) mutant that lacks acetyl transferase activity, but not other mutant Y. pseudotuberculosis, restored IFNγ production in Vγ1.1/2- CD44hi CD27- γδ T cells (Fig 3A). The C172A mutation in YopJ prevents YopJ mediated inhibition of MAPK and NF-κB signaling pathways by abol- ishing its serine and threonine acetylation activity [70]. A similar restoration of IFNγ produc- tion was observed in human Vδ2+ T cells from PBMC of healthy donors upon YopJC172A Y. pseudotuberculosis stimulation (Fig 3B). Thus, the YopJ effector is responsible for inhibition of IFNγ production from murine Vγ4 and human Vδ2+ T cells. YopJ inhibits expression of multiple genes, including ifng, in adaptive γδ T cells To uncover mechanisms by which YopJ inhibits IFNγ production, the transcriptome of cell sorter-purified Vγ1.1/2- CD44hi CD27- γδ T cells after WT or YopJC172A Y. pseudotuberculosis stimulation of MLN cells was assessed by RNA-Seq. Principal component analysis revealed unique gene expression clustering, and approximately 900 genes were expressed at higher lev- els in the YopJC172A stimulation as compared to WT Y. pseudotuberculosis (Fig 4A and 4B). These differentially expressed genes may include genes that are directly inhibited by YopJ PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 6 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function Fig 2. Direct translocation of Yop effectors inhibits the function of murine Vγ4 and human Vδ2+ T cells. MLN suspensions from L. monocytogenes infected mice (A and B) or human PBMC (C and D) were left unstimulated or stimulated with WT or ΔYopB Yptb-βla as indicated. Cells were loaded with CCF4-AM dye prior to stimulation to measure β-lactamase activity. FITC indicates CCF4-AM loaded cells without translocation (Yop-) and BV421 indicates CCF4-AM loaded cells with Yop translocation (Yop+). (A) Adaptive γδ T cells (Vγ1.1/2- CD44hi CD27- γδ T cells), DC (CD11chi MHCIIhi), Macrophages (F4/80+ CD11b+), and CD4 and CD8 T cells were analyzed for Yop translocation 2 hours post stimulation at an MOI of 10. Representative contour plots are displayed. Yop translocation among the indicated populations is depicted as mean ± SEM and is pooled from 2 experiments with a total of 4–8 mice per group. (B) Antibiotics were given 2 hours post-stimulation and brefeldin A was added for the last 5–6 hours of stimulation. Vγ1.1/2- CD44hi CD27- γδ T cells were analyzed for Yop translocation and IFNγ production 24 hours after stimulation. Representative contour plots and histograms are shown. IFNγ production among the indicated populations is depicted as mean ± SEM and is pooled from 3 experiments with a total of 8 mice per group. (C and D) Antibiotics were given 2 hours post-stimulation and brefeldin A was added for the last 5–6 hours of stimulation. Vδ2+ T cells were analyzed for Yop translocation and IFNγ production post stimulation. Representative contour plots are displayed and IFNγ production is quantified among Yop+ or Yop- Vδ2+ T cells. The graph depicts mean ± SEM and is pooled from 3 experiments with 5 healthy donors per group. ����p < 0.0001, ���p < 0.001, ��p < 0.01, and �p < 0.05. An ordinary one-way ANOVA was used for (A), a repeated measures one-way ANOVA was used for (B), and a paired t-test was used for (D). Comparisons were performed to adaptive γδ T cells in (A), to unstimulated or as depicted in figure in (B), and to Yop+ in (C). https://doi.org/10.1371/journal.ppat.1010103.g002 activity in Vγ1.1/2- CD44hi CD27- γδ T cells or indirectly inhibited by YopJ activity in other cells such as DC or macrophages. To resolve this, the WT Yptb-βla reporter provided an opportunity to evaluate the molecular changes elicited by the activity of translocated Yop in adaptive γδ T cells. The transcriptome of sort purified Yop- and Yop+ Vγ1.1/2- CD44hi CD27- γδ T cells after WT Yptb-βla stimulation was assessed by RNA-Seq. Principal component anal- ysis revealed unique gene expression clustering, and approximately 900 genes were more highly expressed in Yop- vs Yop+ Vγ1.1/2- CD44hi CD27- γδ T cells after WT Yptb-βla stimula- tion (Fig 4C and 4D). Overlapping gene expression profiles from the two datasets were assessed to narrow the analysis to direct YopJ effects on Vγ1.1/2- CD44hi CD27- γδ T cells. This comparison revealed 130 genes that were differentially expressed in both datasets, sug- gesting they are regulated directly by translocated YopJ in adaptive Vγ4 T cells (Fig 4E). These genes were categorized into different groups depending on their known functions. Some dif- ferentially expressed genes play a particular role in anti-infection functions (3.9%), stress sens- ing (1.6%), and lymphocyte activation/regulation (7.9%), genes that may be important for protective T cell responses (Fig 4E and 4F). Among these genes, IFNγ was the single most sig- nificant differentially expressed gene suggesting it is a major target of direct YopJ-mediated inhibition of adaptive γδ T cell function (Fig 4F). Differentially expressed genes among those that promote antimicrobial function included several that are important in augmenting type-1 PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 7 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function Fig 3. YopJ is necessary for inhibition of IFNγ production in murine Vγ4 and human Vδ2+ T cells. (A) MLN from L. monocytogenes infected mice were left unstimulated or stimulated with 10 MOI of the indicated Y. pseudotuberculosis for 24 hours. Antibiotics were given 2 hours post- stimulation and brefeldin A was added for the last 5–6 hours. Vγ1/2- CD44hi CD27- γδ T cells were analyzed for IFNγ post stimulation. Representative histograms are displayed. The graph depicts mean ± SEM and represents at least two independent experiments with 2–4 mice per group. (B) Human PBMC were stimulated with 1 MOI of WT or YopJC172A Y. pseudotuberculosis for 24 hours. Antibiotics were given 2 hours post- stimulation. Brefeldin A was added for the last 5–6 hours of stimulation. Vδ2+ γδ T cells were analyzed for IFNγ production post stimulation. Representative flow plots gated on Vδ2+ T cells are displayed. The graph depicts mean ± SEM and is pooled from 2 experiments with 4 healthy donors. ��p < 0.01 and �p < 0.05. A repeated measures one-way ANOVA was used for (A) and a paired t-test was used for (B). Experimental groups were compared to WT Y. pseudotuberculosis. https://doi.org/10.1371/journal.ppat.1010103.g003 and -3 inflammation in T cells. For example, Ptgs2 (encodes cyclooxygenase-2, COX2), Nkg7 (natural killer cell granule protein 7), Prf1 (perforin-1), and Il17a (IL-17A) appear to be regu- lated directly by translocated YopJ in adaptive Vγ4 T cells (Fig 4F) [71–73]. However, analysis of IL-17A protein after stimulation of MLN cell suspensions with YopJC172A, WT, and ΔYopB Y. pseudotuberculosis demonstrated that YopJ did not regulate IL-17A production from Vγ1.1/ 2- CD44hi CD27- γδ T cells (S4 Fig). Some of the observed differentially expressed genes are important in the activation status of T cells (e.g., Il2ra, Ctla4, and Cd69) and suggest that trans- located YopJ may limit the activation of adaptive Vγ4 T cells. There was also a notable impact (48.0%) on genes associated with cell proliferation, metabolism and energy, mitosis and cell cycle, RNA/DNA processing, and ER/Golgi processing (Figs S3A and S3B and 4E), suggesting that YopJ influences the adaptive γδ T cell transcriptional profile more broadly than just tar- geting the IFNγ pathway. Genes that were differentially expressed upon WT Y. pseudotubercu- losis stimulation or among Yop+ cells also suggest that many of the processes associated with immune responses and cellular activity were regulated by YopJ (S3C–S3E Fig). Collectively, YopJ appears to regulate the expression of many genes associated with T cell function in Vγ1.1/2- CD44hi CD27- γδ T cells, suggesting that adaptive Vγ4 T cells are broadly constrained in their immune functions by Y. pseudotuberculosis. YopJ inhibits the IL-12p40-mediated STAT4 pathway in adaptive γδ T cells To gain potential mechanistic insights into YopJ inhibition of IFNγ production and other Vγ1.1/2- CD44hi CD27- γδ T cell functions, a motif discovery algorithm designed for regula- tory element analysis was utilized to assess our RNA sequencing results [74]. Several transcrip- tion factor binding motifs related to IFNγ signaling were differentially expressed after YopJC172A Y. pseudotuberculosis but not WT Y. pseudotuberculosis stimulation including mem- bers of the E twenty-six (ETS)-domain family, Kru¨ppel-like factor and specificity protein (KLF/SP) transcription factor gene family, and the interferon regulatory factors (IRF) family PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 8 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function Fig 4. YopJ translocation leads to the inhibition of a broad anti-microbial gene response from Vγ4 T cells. (A and B) MLN suspensions from L. monocytogenes infected mice were stimulated with 10 MOI of WT or YopJC172A Y. pseudotuberculosis for 24 hours. Antibiotics were given 2 hours post- stimulation. Five hundred Vγ1.1/2- CD44hi CD27- γδ T cells from each stimulation were flow sorted and processed for RNA sequencing. (A) PCA plots are depicted for similarity of groups YopJC172A and WT Y. pseudotuberculosis stimulated Vγ1.1/2- CD44hi CD27- γδ T cells. (B) Heat maps are depicted for differentially expressed genes of YopJC172A or WT Y. pseudotuberculosis stimulated Vγ1.1/2- CD44hi CD27- γδ T cells. (C and D) MLN suspension from L. monocytogenes infected mice were stimulated with 1 MOI of WT Yptb-βla. Five hundred Yop+ or Yop- Vγ1.1/2- CD44hi CD27- γδ T cells were flow sorted and processed for RNA sequencing. (C) PCA plots are depicted for similarity of Yop+ or Yop- Vγ1.1/2- CD44hi CD27- γδ T cells. (D) Heat maps are depicted for differentially expressed genes of Yop- or Yop+ stimulated Vγ1.1/2- CD44hi CD27- γδ T cells. (E-G) A Venn diagram of differentially expressed genes (higher) that overlapped between RNA sequencing analyses favoring YopJC172A Y. pseudotuberculosis stimulation or Yop- cells is displayed. Shared genes were categorized by gene function. (F) The heat map highlights differentially expressed genes among Vγ1.1/2- CD44hi CD27- γδ T cells from the indicated stimulations and categories. (G) Homer motif analysis was performed on the RNA sequencing dataset. Motifs and associated genes to YopJC172A stimulated Vγ1.1/2- CD44hi CD27- γδ T cells are highlighted. Each experiment was performed with 3 biologic samples per group. Cutoffs for significant genes are p < 0.05 and FDR < 0.10. https://doi.org/10.1371/journal.ppat.1010103.g004 of transcription factors (S5A Fig). IRF8 protein was validated after WT, ΔYopB, and YopJC172A Y. pseudotuberculosis stimulation. Indeed, a higher percentage of Vγ4 T cells expressed IRF8 PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 9 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function protein after stimulation with ΔYopB compared to WT Y. pseudotuberculosis stimulation (S5B Fig). Stimulation with YopJC172A Y. pseudotuberculosis was also able to partially restore IRF8 levels to those seen after ΔYopB Y. pseudotuberculosis stimulation (S5B Fig). IRF8 was also impacted by IL-12p40 blockade, which signals through signal transducer and activator of tran- scription 4 (STAT4) (S5B Fig). Interestingly, a number of transcription factor binding motifs downstream of STAT4 signaling were enriched including Etv5, Runx3, and Tead1 (Fig 4G) [75–78]. The RNAseq and homer motif analyses were also compared to an existing STAT4 ChIP-on-chip [79]. 7 genes identified from our main analyses (Figs 4F and 4G and S5B) were STAT4 target genes (S5C Fig). In summary, transcriptional profiling revealed a global subver- sion of anti-pathogen immune functions that may be associated with YopJ subversion of STAT4 activity. IL-12 signaling leads to STAT4 phosphorylation and formation of STAT4-STAT4 homodi- mers that re-localize to the nucleus where they directly bind to the Ifng promoter to induce IFNγ expression [79–81]. To determine whether YopJ inhibits IFNγ production by interfering with the STAT4 pathway, STAT4 protein and phosphorylation were assessed by flow cytome- try of MLN cells stimulated with Y. pseudotuberculosis. STAT4 phosphorylation was analyzed 6 hours after stimulation with WT, ΔYopB, or YopJC172A Y. pseudotuberculosis. Consistent with suppression of IFNγ production and the RNA-Seq analysis, WT Y. pseudotuberculosis sig- nificantly reduced the percentage of pSTAT4+ CD44hi CD27- γδ T cells as compared to ΔYopB and YopJC172A Y. pseudotuberculosis (Fig 5A). However, STAT4 protein levels were the same in all three infection conditions (WT, ΔYopB, and YopJC172A Y. pseudotuberculosis) (Fig 5B). Flow cytometry antibodies for STAT4 protein were validated by comparing STAT4 from WT and STAT4 KO splenocytes (S5D Fig). These data suggest that STAT4 phosphorylation but not protein is decreased upon YopJ translocation. STAT4 phosphorylation was also evaluated using the Yptb-βla reporter system described above. Among CD44hi CD27- γδ T cells, Yop- cells had a higher percentage of pSTAT4+ cells compared to Yop+ cells suggesting intrinsic Yop mediated inhibition of STAT4 phosphorylation levels (Fig 5C). Additionally, as STAT4 phosphorylation is downstream of IL-12 signaling, an anti-IL-12/23p40 subunit antibody (anti-p40) was used to determine whether IL-12 signals in the environment regulated STAT4 phosphorylation after Y. pseudotuberculosis stimulation. Indeed, IL-12/23p40 neutralization abrogated STAT4 phosphorylation levels regardless of Yop translocation (Fig 5C). As IL-12/ 23p40 was required to elicit IFNγ production from adaptive Vγ4 T cells in the culture condi- tions, we assessed whether YopJC172A Y. pseudotuberculosis stimulation modulated IL-12p70. The concentration of IL-12p70 was comparable between WT and YopJC172A Y. pseudotubercu- losis stimulated cultures (Fig 5D). Thus, changes in IL-12 were unlikely to contribute to adap- tive Vγ4 T cell subversion in vitro. To understand the role of YopJ and IL-12 on Vγ1.1/2- CD44hi CD27- γδ T cells in a more simplified system, purified γδ T cells were stimulated with YopJC172A Y. pseudotuberculosis in the presence of excessive IL-12p70. Adaptive Vγ4 T cells were unable to produce IFNγ in response to YopJC172A Y. pseudotuberculosis and IL-12p70 (S6A Fig). Finally, we assessed whether the addition of IL-12p70 could overcome the YopJ mediated inhibition of IFNγ production after WT Y. pseudotuberculosis stimulation. While a supraphysiologic level of IL-12p70 (50 ng/ml) was able to partially overcome YopJ mediated inhibition, lower levels of IL-12p70 addition (2 and 10 ng/ml) were unable to overcome YopJ mediated inhibition (S6B Fig). Importantly, these latter concentrations were orders of magni- tude higher than those detected in our culture conditions. Thus, IL-12 is not sufficient to induce IFNγ production from adaptive Vγ4 T cells. Collectively, these results suggest that Y. pseudotuberculosis stimulation elicits IL-12 production to promote adaptive Vγ4 T cell IFNγ responses, and that YopJ translocation into adaptive Vγ4 T cells inhibits IL-12 mediated STAT4 phosphorylation. PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 10 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function Fig 5. YopJ inhibits IL-12p40 mediated STAT4 phosphorylation. (A) MLN cell suspensions from L. monocytogenes infected mice were stimulated with 10 MOI of WT, YopJC172A, or ΔYopB Y. pseudotuberculosis for 6 hours. Antibiotics were given 2 hours after stimulation. Vγ1.1/2- CD44hi CD27- γδ T cells were analyzed for pSTAT4 after stimulation. Representative contour plots are displayed. The graph depicts mean ± SEM and represents at least two independent experiments with 2–4 mice per group. (B) The same experimental setup was used as in (A), but Vγ1.1/2- CD44hi CD27- γδ T cells were analyzed for STAT4 protein after stimulation. Representative plots for mean fluorescent intensity (MFI) are displayed. The graph depicts mean ± SEM and represents two independent experiments with 2 mice per group. (C) MLN suspensions from L. monocytogenes infected mice were either treated or untreated with IL-12/23p40 neutralizing antibody prior to stimulation with 1 MOI of WT Yptb-βla for 6 hours. Antibiotics were given 2 hours post-stimulation. Vγ1.1/2- CD44hi CD27- γδ T cells were analyzed for pSTAT4 after stimulation. Representative contour plots are displayed. The graph depicts mean ± SEM and represents at least two independent experiments with 2–4 mice per group. (D) MLN cell suspensions from L. monocytogenes infected mice were stimulated with 10 MOI of WT or YopJC172A Y. pseudotuberculosis for 24 hours. Antibiotics were given 2 hours after stimulation. Supernatants were collected 24 hours post stimulation and IL-12p70 concentration was determined via ELISA. ����p < 0.0001, ���p < 0.001, ��p < 0.01, and �p < 0.05. A repeated measures one-way ANOVA was used for (A-C). Comparisons were performed to WT Y. pseudotuberculosis in (A) and as depicted in (B-D). https://doi.org/10.1371/journal.ppat.1010103.g005 Foodborne infection with YopJC172A Y. pseudotuberculosis induces IFNγ production in adaptive Vγ4 T cells To determine whether YopJ subverts adaptive γδ T cell function in vivo, foodborne infection with WT and YopJC172A Y. pseudotuberculosis was performed on naïve Balb/c mice. As YopJC172A Y. pseudotuberculosis is attenuated in vivo [82], a one log higher (2-4x108 CFU) infection dose of YopJC172A Y. pseudotuberculosis was administered to normalize the internal bacteria burdens in the MLN between infection groups (S7A Fig). While mice infected with WT and YopJC172A Y. pseudotuberculosis lost a similar amount of weight, mice infected with YopJC172A Y. pseudotuberculosis recovered slightly faster (Fig 6A). MLN were isolated 9 days after infection to evaluate adaptive γδ T cell function. Consistent with the ex vivo stimulation of L. monocytogenes-elicited Vγ4 T cells with Y. pseudotuberculosis, Vγ1.1/2- CD44hi CD27- γδ T cells from the MLN of YopJC172A Y. pseudotuberculosis infected mice displayed enhanced IFNγ production when stimulated ex vivo compared to their WT Y. pseudotuberculosis infected counterparts (Fig 6B and 6C). When the same infectious doses were used for both WT and YopJC172A Y. pseudotuberculosis (5x107 CFU), Vγ1.1/2- CD44hi CD27- γδ T cells from the MLN of YopJC172A Y. pseudotuberculosis infected mice also displayed enhanced IFNγ PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 11 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function Fig 6. Foodborne YopJC172A Y. pseudotuberculosis infection elicits an IFNγ response from Vγ4 T cells. (A-E) Balb/c mice were foodborne infected with WT (2-4x107 CFU) or YopJC172A Y. pseudotuberculosis (2-4x108 CFU). (A) Mouse weight was assessed daily after infection. (B and C) Nine days post infection, MLN were processed into single cell suspensions and stimulated with PMA/ionomycin in the presence of brefeldin A for 4 hours. Vγ1.1/2- CD44hi CD27- γδ T cells were analyzed for IFNγ production. Representative histograms are displayed. Graphs represent mean ± SEM and are pooled from 3 experiments with a total of 5–8 mice per group. (D) Mouse survival was assessed daily after infection. anti-IL-12p40 antibody was administered at 0.2 mg/mouse on 0, 2, 4, and 6 days post infection. A Kaplan-Meier survival plot is shown. Study endpoint was 9 days post infection. The data represent 2 independent experiments with a total of 9 mice per group. (E and F) Balb/c mice were foodborne infected with 2x109 CFU L. monocytogenes to elicit a Vγ1.1/2- CD44hi CD27- γδ T cell population in vivo. 30 days post infection, immunized mice were foodborne infected with WT Yptb-βla (2-4x109 CFU) or left uninfected. (E) CFU of WT Yptb-βla were enumerated in MLN 3 days post WT Yptb-βla infection. (F) Three days post foodborne WT Yptb-βla infection, Vγ1.1/2- CD44hi CD27- γδ T cells from the MLN were analyzed for Yop translocation using the CCF4-AM assay. Representative contour plots are shown. Yop translocation (Yop+) among the indicated populations is depicted as mean ± SEM and is pooled from 2 experiments with a total of 4 mice per group. ���p < 0.001, ��p < 0.01, and �p < 0.05. A repeated measures one-way ANOVA was used for (A and F) and an unpaired t- test was used for (B, C, and E). Experimental groups were compared to WT Y. pseudotuberculosis in (A-C), uninfected controls in (E), and Vγ1.1/2- CD44hi CD27- γδ T cells in (F). A logrank test was used for survival curves in (D). https://doi.org/10.1371/journal.ppat.1010103.g006 production compared to their WT Y. pseudotuberculosis infected counterparts (S7B Fig). These experiments demonstrate that the increased IFNγ observed was not a result of a higher infectious dose nor of a higher bacteria burden at the time of analysis. As IL-12/23p40 was required for adaptive Vγ4 T cell IFNγ production in vitro (Fig 5), the impact of IL-12/23p40 was assessed in vivo. Naïve Balb/c mice were infected with WT or YopJC172A Y. pseudotubercu- losis and treated with an IL-12/23p40 neutralizing antibody or isotype control. All YopJC172A Y. pseudotuberculosis infected mice treated with anti-IL-12/23p40 succumbed by day 8 post PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 12 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function infection (Fig 6D). This data suggests that IL-12 promotes the protective capacity of catalyti- cally inactive YopJ. Similarly, IL-12 contributed to the protection of mice infected with WT Y. pseudotuberculosis. Serum was also collected on days 2, 4 and 6 after infection to assess circu- lating IL-12p70 levels. IL-12p70 was detectable 6 days after YopJC172A Y. pseudotuberculosis infection but was mostly below the limit of detection after WT Y. pseudotuberculosis infection (Fig 6E). Collectively, these data suggest that IL-12 is important in the protection of mice infected with WT or YopJC172A Y. pseudotuberculosis. Foodborne infection with WT Yptb-βla was performed to determine whether Y. pseudotu- berculosis could target adaptive Vγ4 T cells with Yop translocation in vivo. Balb/c mice were foodborne infected with L. monocytogenes to elicit a population of Vγ1.1/2- CD44hi CD27- γδ T cells as described previously [37,83]. After a return to homeostasis at 30 days post L. monocy- togenes infection, mice were foodborne infected with WT Yptb-βla. Y. pseudotuberculosis bur- den was assessed in the MLN 3 days post foodborne infection to determine whether WT Yptb- βla could establish a productive infection where Vγ4 T cells reside. Indeed, infected mice con- tained detectable Y. pseudotuberculosis in the MLN 3 days after foodborne infection (Fig 6F). Analysis of the translocation of Yop into Vγ1.1/2- CD44hi CD27- γδ T cells, myeloid cells (CD11b+), and CD4 and CD8 T cells was performed. Consistent with our in vitro observations, Vγ1.1/2- CD44hi CD27- γδ T cells and myeloid cells contained translocated Yop in vivo (Fig 6G). Y. pseudotuberculosis was relatively inefficient at Yop translocation into CD4 and CD8 T cells (Figs 6G and S7C). Collectively, these data show that foodborne YopJC172A Y. pseudotu- berculosis infection of naïve mice elicits IFNγ production in adaptive γδ T cells and that Yop can be translocated into adaptive Vγ4 T cells in vivo. Discussion In this study, we assessed the subversion of an adaptive subset of γδ T cells specialized in the promotion of pathogen resistance at the intestinal mucosa through the production of anti- infective cytokines like IFNγ and IL-17A [37]. While limited evidence suggests that Yop effec- tors directly target T cells to subvert their function, we identified that the Y. pseudotuberculosis effector molecule YopJ directly inhibits IFNγ production from adaptive CD44hi CD27- γδ T cells to subvert host immunity in mice. Additionally, we demonstrate that circulating human Vδ2+ T cells are similarly inhibited by direct translocation of YopJ, demonstrating that the direct targeting of γδ T cells by Y. pseudotuberculosis to inhibit IFNγ production is a conserved pathway of immune evasion in humans. Thus, Y. pseudotuberculosis Yop effectors translocated into murine Vγ4 T cells and human Vδ2+ T cells directly subvert their anti-microbial functions and host immunity by limiting IFNγ production. While Yersinia mediated inhibition of conventional T cells has been previously reported, studies have largely focused on the indirect subversion of T cells that is mediated by transloca- tion of Yop effectors into myeloid cells [55,82]. For example, YopJ/P appears to primarily sub- vert conventional T cell function through indirect mechanisms associated with inhibiting DC [22,26]. On the contrary, the study of γδ T cells in the context of Y. pseudotuberculosis infection has been primarily limited to the potential antigens that drive γδ T cell recognition of infection [84–89]. After phagocytosis of pathogens, activated DC migrate to lymph nodes and present antigen to T cells. APC-derived IL-12 further shapes T cell responses by providing a critical signal dur- ing T cell activation [90]. Interestingly, Y. pestis can limit both the migratory capacity of DC and the production of IL-12 [91], and Y. enterocolitica can induce programmed death of DC and inhibit antigen presentation [22]. Y. pseudotuberculosis YopJ can also indirectly limit NK cell function by interfering with DC TLR4 signaling pathways and YopP in Y. enterocolitica PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 13 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function can limit NK cell function through STAT4 inhibition [21,92]. Given that IL-12 signaling pro- motes STAT4 phosphorylation and IFNγ production in γδ T cells [29,93] and Yersinia spp. can inhibit DC, a potential extrinsic mechanism emerges for Y. pseudotuberculosis to inhibit γδ T cell responses by suppressing DC functions. In line with these observations, IL-12p40 was criti- cal for the induction of phospho-STAT4 in Vγ4 T cells after stimulation with Y. pseudotuber- culosis in in vitro cultures. In contrast, Y. pseudotuberculosis and IL-12 were unable to directly elicit IFNγ production from a highly enriched population of in vitro expanded γδ T cells sug- gesting that IL-12 is required but not sufficient for Vγ4 T cell IFNγ production. However, stimulation of MLN cell suspensions with ΔYopB or YopJC127A Y. pseudotuberculosis elicited STAT4 phosphorylation among Vγ4 T cells suggesting that translocation of Yop effectors and YopJ in particular subverts Vγ4 T cell function. Tracking Yop translocation in vitro revealed that Vγ4 T cells that contain Yop had reduced pSTAT4 levels and inhibited IFNγ production. Importantly, Vγ4 T cells that did not contain Yop effectors from the same cultures expressed higher pSTAT4 levels and comparable IFNγ production as Vγ4 T cells stimulated with ΔYopB-βla Yptb. Additionally, IL-12 levels were comparable between cultures stimulated with WT and YopJC172A Y. pseudotuberculosis. Collectively, these data suggest that functional impairment of Vγ4 T cells was mediated by direct translocation of YopJ into Vγ4 T cells and cell intrinsic mechanisms in vitro. The YopJ effector family has been increasingly described by their acetyltransferase function on serine, threonine, and lysine amino acid residues [19,94]. Serine and threonine are com- mon targets of phosphorylation to propagate signaling cascades or elicit functional conse- quences. For example, phosphorylation of STAT4 leads to dimerization and transport to the nucleus to promote transcription of STAT4 target genes. Acetylation of these target residues may inhibit phosphorylation and downstream signaling events [70]. Thus, a potential mecha- nism of YopJ subversion of Vγ4 T cells is through acetylation of STAT4 to inhibit the phos- phorylation or dimerization of STAT4. Other potential targets of YopJ acetyltransferase activity are the IL-12R and Janus kinases upstream of STAT4 activation. YopJ has also been reported to have cysteine protease, lysine acetyltransferase, ubiquitin-like protein protease, and deubiquitinase activity that may provide other potential avenues for YopJ to modulate the function of Vγ4 T cells through STAT4 [94–96]. An important aspect of Y. pseudotuberculosis pathogenesis unveiled by this work is the pref- erential targeting of a specialized subset of γδ T cells for delivery of inhibitory Yop effector molecules. Y. pseudotuberculosis injected Yop effectors into adaptive γδ T cells in a similar pro- portion as macrophages and DC and to a much greater extent than conventional CD4 or CD8 T cells. Of note, Y. pseudotuberculosis has been reported to translocate Yop effectors more effi- ciently into Treg cells than conventional CD4 T cells at high MOI to modulate their function [65]. The preferential targeting of Vγ4 T cells in this system is associated with expression of the β1-integrin by adaptive Vγ4 T cells. Additionally, the majority of adaptive Vγ4 T cells are anatomically segregated from conventional T cells in the paracortex by their localization in the interfollicular and medullary areas of the gut draining lymph nodes [38]. The distinct localiza- tion of adaptive Vγ4 T cells may facilitate interactions with Y. pseudotuberculosis in vivo. Loss of the adhesin YadA but not Inv abrogated YopJ mediated γδ T cell inhibition, suggesting that Y. pseudotuberculosis utilizes YadA to target adaptive γδ T cells for Yop translocation, consis- tent with previous studies suggesting that the adhesin Inv is largely dispensable for Y. enteroco- litica virulence [61,97]. Interestingly, this appears distinct from the targeting of conventional CD4 T cells that relies on Inv and is enhanced in the absence of YadA [11,65]. While our data demonstrates that STAT4 phosphorylation is inhibited by YopJ, our RNA- Seq analysis suggests that other targets may also be affected. One of YopJ’s known targets is the MAPK family of proteins that can have broad effects on cell proliferation, differentiation, PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 14 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function survival, and apoptosis [18]. The MAPK pathway in CD4 T cells and NK cells may also pro- mote STAT4 activity and downstream IFNγ mRNA stabilization, respectively [98,99]. Our data demonstrate a broad impact of YopJ on adaptive γδ T cell proliferation, metabolism, cell cycle, RNA/DNA processing, and ER/Golgi processing gene expression networks. These path- ways may be regulated by MAPK family member activity [100–105]. Homer motif analysis identified other potential means by which YopJ may regulate IFNγ production. For example, Ets-1 is a T-bet cofactor and necessary for Th1 IFNγ responses [106]. Increases in ETS-domain family of transcription factor motifs were associated with type 3 innate lymphoid cells (ILC3) but not Th17 cells [107], which may suggest that one of the potential mechanisms of YopJ mediated inhibition may target conserved pathways in unconventional lymphocyte popula- tions. Many NK cell receptors are also expressed on γδ T cells and may facilitate TCR indepen- dent effector functions [108–110]. In line with this, our profiling demonstrates that YopJ limits gene expression of Nkg7, which has recently been reported to promote cytotoxic granule release and inflammation during infection and cancer [72], and Prf1, which encodes the pore forming molecule perforin necessary to deliver lytic machinery into target cells [111]. Finally, interactions with Y. pseudotuberculosis YopJ led to the upregulation of Ulbp1, which encodes a stress-induced NKG2D ligand, and Idi1, which encodes an enzyme in the mevalonate pathway. As human γδ T cells respond to phospho-antigens derived from the non-mevalonate pathway in bacteria and mammalian mevalonate pathway in humans [112], this may limit the removal of translocated cells through NK or γδ T cell sensing mechanisms and suggests a broad mecha- nism to subvert human immunity. Thus, our findings suggest that adaptive Vγ4 T cells provide dynamic anti-infectious immunity that is subverted by direct translocation of YopJ. A number of studies have highlighted the importance of IFNγ production from conven- tional CD4 and CD8 T cells, NK cells, and ILC3 for Yersinia resistance [40,43,113,114], although the in vivo relevance of Yop inhibition of conventional T cells has not been addressed. Foodborne infection with YopJC172A Y. pseudotuberculosis led to an enhanced response from Vγ4 T cells, including increased IFNγ production, that was associated with a more rapid recovery of weight. As IL-12 was critical for Vγ4 T cell derived IFNγ production in vitro, the role of IL-12 after foodborne infection of Balb/c mice was assessed. Consistent with previous studies [22,40,115,116], IL-12 appeared critical for protection against foodborne WT and YopJC172A Y. pseudotuberculosis infection. As serum IL-12 was only readily detectable after YopJC172A Y. pseudotuberculosis infection, increased IL-12 may also contribute to the enhanced IFNγ response from Vγ4 T cells in vivo. Finally, assessment of cell populations tar- geted for Yop translocation in vivo was comparable to the results from the ex vivo MLN cul- tures. The highest percentage of Yop+ cells were among CD11b+ cells and Vγ4 T cells. Given the low MOI used in our ex vivo studies with the WT Yptb-βla reporter and the lack of inhibi- tion observed in Vγ4 T cells that lack Yop effectors from the same culture conditions, it is likely that intrinsic Yop effects on Vγ4 T cells is a mechanism of inhibiting Vγ4 T cell function in vivo. Thus, while Y. pseudotuberculosis can use γδ T cell intrinsic mechanisms to subvert the γδ T cell IFNγ response, multiple mechanisms may be available for Yersinia spp. to subvert γδ T cell functions to aid pathogenesis in vivo. As we previously demonstrated that foodborne but not i.v. infection led to adaptive Vγ4 T cell responses [37], our physiologic foodborne Y. pseudotuberculosis infection model revealed novel aspects of Yersinia pathogenesis and adap- tive Vγ4 T cell biology. In summary, the Y. pseudotuberculosis effector YopJ directly inhibits essential anti-effective functions of murine Vγ4 T cells and human Vδ2+ T cells. Y. pseudotuberculosis targeted Vγ4 T cells in a T3SS- and YadA-dependent process to deliver Yop effectors directly into Vγ4 T cells. Ex vivo whole tissue cultures revealed that direct inhibition of Vγ4 T cell function was the major mechanism of Vγ4 T cell subversion. YopJ translocation led to a dramatic reduction in PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 15 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function STAT4 phosphorylation levels and IFNγ production, which is important for protection from Yersinia. YopJ also inhibited a broad anti-infective gene signature. Thus, these findings add substantial insight into YopJ effector functions on murine and human γδ T cells and the patho- genesis of foodborne Y. pseudotuberculosis infection. Materials and methods Ethics statement All animal experiments were conducted in accordance with the Stony Brook University Insti- tutional Animal Care and Use Committee and National Institutes of Health guidelines. Blood collection from healthy human donors was approved by the Institutional Review Board at Stony Brook University. Mice Female 8–12 week old BALB/cJ mice were purchased from the Jackson Laboratory. Mice were euthanized by CO2 inhalation. Human studies Blood was sampled from a total of 6 adult healthy human donors of either gender between the ages of 20 and 40. Studies were designed so no randomization to experimental groups was nec- essary. Donors provided written informed consent. Bacteria Bacteria strains used in this study include: Y. pseudotuberculosis on the 32777 background WT strain, WT32777c, YopJC172A, YopHR409A, ΔYopB, YopER144A, YopTC139A, ΔYopM, ΔYpkA, ΔYopK, WT Yptb-βla, and ΔYopB Yptb-βla. Y. pseudotuberculosis on the IP2666 background WT strain, ΔYopB, ΔInv, ΔYadA, ΔInv ΔYadA, and ΔYopB ΔInv ΔYadA. See Table 1 for details. All strains were stored in 25% glycerol stocks at -80˚C. For stimulations, Y. pseudotu- berculosis strains were cultured overnight at 28˚C and 220 RPM in LB media. The following morning, Y. pseudotuberculosis was sub-cultured 1:10 in LB and 50 mM CaCl2 at 37˚C and 220 RPM for approximately 2 hours. Stimulation doses were based on the OD600. Foodborne L. monocytogenes immunization L. monocytogenes (EGDe strain) expressing a mutation in the internalin A gene (InlAM) was used for foodborne infection to facilitate epithelial cell invasion [117]. InlAM L. monocytogenes was cultured overnight at 37˚C and 220 RPM in BHI media. The following morning, L. mono- cytogenes was sub-cultured 1:10 in BHI at 37˚C and 220 RPM for approximately 2 hours. Infec- tion doses were based on the OD600. Mice were food and water deprived for 4 hours. Approximately 0.5 cm3 bread pieces were inoculated with 2x109 CFU L. monocytogenes in 50 μL. Mice were monitored to ensure the inoculated bread was consumed within 1 hour. Mice that did not fully consume bread were removed from the study. Bacterial infection doses were confirmed by plating inoculum on BHI. Foodborne Y. pseudotuberculosis infection Y. pseudotuberculosis strains (Table 1) were cultured overnight at 28˚C and 220 RPM in LB media. Infection doses were based on the OD600. Mice were food and water deprived for 4 hours. Approximately 0.5 cm3 bread pieces were inoculated with 2-4x107 CFU for WT32777 PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 16 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function Y. pseudotuberculosis, 2-4x107–2-4x108 CFU for YopJC172A Y. pseudotuberculosis, or 2x109 CFU for WT Yptb-βla infection in 50 μL. Mice were monitored to ensure the inoculated bread was consumed within 1 hour. Mice that did not fully consume bread were removed from the study. Bacterial infection doses were confirmed by plating inoculum on LB. Single cell preparations, Y. pseudotuberculosis stimulations, and flow cytometry MLN from L. monocytogenes immunized mice were harvested 9 days after immunization and mechanically dissociated using a syringe plunger through a 70 μm cell strainer into a single cell suspension. Cells were resuspended in IMDM (Gibco) supplemented with 10% fetal bovine serum, 0.01 M HEPES, 100 μM non-essential amino acids (Gibco), 2 mM L-alanyl-L- glutamine dipeptide in 0.85% NaCl or 1x Glutamax (Gibco), and 1 mM sodium pyruvate. Cells were counted using a Vi-CELL Viability Analyzer (Beckman Coulter). Cells were stimu- lated in 96 well round-bottom tissue culture treated plates with various strains of Y. pseudotu- berculosis at 1 or 10 MOI (1 MOI for WT or ΔYopB Yptb-βla and 10 MOI for all other Y. pseudotuberculosis stimulations, unless otherwise indicated) at 37˚C/5% CO2. 100 U/mL of penicillin and 100 μg/mL of streptomycin were added to cells 2 hours post-stimulation. Cells were stimulated for a total of 24 hours or as indicated. BD GolgiPlug (BD Biosciences) was added 5 hours prior to the end of stimulation. If translocation was assessed, β-lactamase Load- ing Solutions kit (Invitrogen) was used to load CCF4-AM by incubating CCF4-AM at RT with cells for 1 hour in the dark. Cells were then processed for surface staining via incubation with live/dead stain, antibody, and Fc block (BioXcell) for 20 min in the dark at 4˚C. Antibodies used included antibodies specific to CD45, CD3, TCRδ, CD8, CD4, Vγ1.1, Vγ2, CD44, CD27, F4/80, CD11b, MHCII, CD11c, and CD29 (BioLegend). Cells were fixed, permeabilized, and stained with anti-IFNγ, anti-IRF8, or anti-STAT4 using BD Cytofix/Cytoperm kit (BD Biosci- ences) for intracellular cytokine staining. Functional γδ T cell analysis was done by stimulation with BD leukocyte activation cocktail (containing PMA, ionomycin, and brefeldin A; BD Pharmingen) for 4 hours prior to staining. Flow cytometry data were acquired using a BD LSRFortessa and analyzed by FlowJo software (BD Biosciences). Cell culture supernatant was analyzed for IL-12p70 using the BioLegend ELISA MAX Deluxe Set Mouse IL-12 (p70) kit per manufacturer instructions. Human γδ T cell response Blood was drawn and collected from healthy human donors in BD Vacutainer sodium heparin tubes (BD Biosciences). Blood was diluted 1:1 with 1x PBS at room temperature. Peripheral blood mononuclear cells (PBMC) were isolated from the buffy coat of Ficoll-paque PLUS gra- dient centrifugation (GE Healthcare) for 20 min at 1,400 × g without a brake. PBMC were washed with 1x PBS at room temperature and resuspended in IMDM supplemented with 10% fetal bovine serum, 0.01 M HEPES, 100 μM Non-essential amino acids (Gibco), 2 mM L-ala- nyl-L-glutamine dipeptide in 0.85% NaCl or 1x Glutamax (Gibco), and 1 mM sodium pyru- vate. Y. pseudotuberculosis strains were cultured overnight at 28˚C and 220 RPM in LB media the night prior. The following morning, Y. pseudotuberculosis was sub-cultured 1:10 in LB and 50 mM CaCl2 at 37˚C and 220 RPM for approximately 2 hours. Stimulation doses were based on the OD600. Cells were counted using a Vi-CELL Viability Analyzer (Beckman Coulter). Cells were stimulated in 96 round-bottom tissue culture treated plates with various strains of Y. pseudotuberculosis at 1 MOI at 37˚C/5% CO2. 1x penicillin and streptomycin (100 U/mL penicillin and 100 μg/mL streptomycin) were added to cells 2 hours post-stimulation. Cells were stimulated for a total of 24 hours or as indicated. BD GolgiPlug (BD Biosciences) was PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 17 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function added 5 hours prior to the end of stimulation. If translocation was assessed, β-lactamase Load- ing Solutions kit (Invitrogen) was used to load CCF4-AM by incubating CCF4-AM at RT with cells for 1 hour in the dark. Cells were then processed for surface staining via incubation with live/dead stain, antibody, and Fc block (BioXcell) for 20 min in the dark at 4˚C. Antibodies used included antibodies specific to Vδ2, CD3, TCRδ, (BioLegend). Cells were fixed, permea- bilized, and stained with anti-IFNγ using BD Cytofix/Cytoperm kit (BD Biosciences) for intra- cellular cytokine staining. Flow cytometry data were acquired using a BD LSRFortessa and analyzed by FlowJo software (BD Biosciences). Phospho-flow cytometry After surface staining for flow cytometry, cells were washed and stained for pSTAT4 using a methanol-based approach. Cells were fixed in 4% PFA/1.5% methanol for 30 minutes in the dark at 4˚C. Cells were then washed and incubated in methanol in the dark at 20˚C for 45 min- utes. After washing, cells were stained with anti-pSTAT4 (Y693)-PE (BD Biosciences) and washed once more. Flow cytometry data were acquired using a BD LSRFortessa and analyzed by FlowJo software (BD Biosciences). Enumeration of Y. pseudotuberculosis burden MLN were crushed and diluted in media prior to plating on LB agar. Total Y. pseudotuberculo- sis burden per organ was calculated. Sequencing and analysis Samples were prepared after Y. pseudotuberculosis stimulation as described above. Cell prepa- rations were stimulated with 10 MOI of WT or YopJC172A Y. pseudotuberculosis or 1 MOI of WT Yptb-βla for 24 hours. 500 Vγ1.1/2- CD44hi CD27- γδ T cells were flow sorted directly into a tube with NEBNext Cell Lysis Buffer and Murine RNase Inhibitor and processed for RNA sequencing using NEBNext Single Cell/Low Input RNA Library Prep Kit (Illumina). Sequenc- ing was performed at the Cold Spring Harbor Laboratory sequencing core on a NextSeq500. Fastq files were produced as an output of the sequencing files. Fastq were run through FastQC to perform quality control of transcripts prior to alignment. Fastq files were pair-ended aligned to GRCm38/mm10 by way of HISAT2 and output as .BAM files [118]. Raw counts of aligned transcripts were quantified with FeatureCounts [119]. Dimensionality reduction was per- formed with PCA analysis with the axes PC1 and PC2 in R-studio [120]. To determine differ- ential expression between samples, FeatureCounts raw count matrix was analyzed through DESeq2 with a parametric fitting normalized to the geometric mean of each individual gene across samples [121]. Cutoff values for significance and quality control were a p-value of <0.05 and FDR-value of <0.10, respectively. Significantly differentially expressed genes were visual- ized on a heatmap with a dendrogram that was clustered through average linkage. The distance measurement on the dendrogram used was through the Euclidean method. Overlapping expressions between gene differential expression sets were filtered with R-studio. Upregulated and downregulated genes from the differential expression analysis were separated with R-stu- dio and these Gene IDs were used for HOMER motif analysis [74]. Parameters of analysis of each gene used were 400bp preceding the initiation site and 100bp after the initiation site. The length of the motifs analyzed was set between 8 and 10. PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 18 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function γδ T cell purification γδ T cells were expanded in vitro according to published protocols [122]. The MLN and spleen were isolated and processed into a single cell suspension 9 days post foodborne L. monocyto- genes infection of Balb/c mice. Red blood cells were lysed with red blood cell lysis buffer or ammonium chloride for 1 minute and cells from the MLN and spleen were combined. γδ T cells were enriched by negative selection using the following rat IgG primary antibodies from BioLegend: anti-CD4 (clone GK1.5), anti-CD8α (clone 53–6.7), anti-B220 (clone RA3-6B2), anti-MHC-II (clone M5/114.15.2), and anti-CD11b (clone M1/70). The MACS goat anti-rat IgG kit (Miltenyi Biotec) was used per manufacturer instructions with MACS LD columns and a QuadroMACS magnet. Enriched cells were cultured in 48-well plates coated overnight with 5 ug/ml anti-TCRδ (clone GL3). Cells were cultured in RPMI 1640 supplemented with 25 mM HEPES (Gibco), 1x glucose (Gibco), 10 g/ml folate (Sigma Aldrich), 1x sodium pyruvate (Gibco), 5x105 M 2 beta-mercaptoethanol (Sigma Aldrich), 1x Glutamax (Gibco), 1x penicil- lin-streptomycin (Gibco), and 10% FBS with 100 U/ml recombinant human IL-2. After 2 days of culture, cells were transferred into new wells with the same culture media to rest for 5 days. Cells were then stimulated with 10 MOI YopJC172A Y. pseudotuberculosis with 0.1, 1, or 10 ng/ ml of recombinant murine IL-12p70 (Peprotech). In vivo anti-IL-12p40 antibody treatment and serum collection On the day of foodborne WT or YopJC172A Y. pseudotuberculosis infection and on day 2, 4, and 6 after infection, mice were treated with 0.2 mg of anti-IL-12p40 (clone C17.8; BioLegend) by i.p. injection. Blood was collected via tail vein on day 2, 4, and 6 after infection. Blood was incubated at ambient temperature for 30 minutes before being spun down at 1500G for 10 minutes at 4˚C. Serum was isolated and analyzed for IL-12p70 with the BioLegend ELISA MAX Deluxe Set Mouse IL-12 (p70) kit. Ex vivo anti-IL-12p40 and recombinant IL-12p70 treatments At the start of Y. pseudotuberculosis stimulation of MLN cell suspensions, cultures were treated with 10 μg/ml anti-IL-12p40 (clone C17.8; BioLegend) for neutralization. In other conditions, recombinant murine IL-12p70 (Peprotech) was added at the start of Y. pseudotuberculosis stimulation of MLN cell suspensions at 2, 10, or 50 ng/ml. Statistical analysis GraphPad Prism 6 software (GraphPad Software Inc.) was used for statistical analysis. The dif- ferences between the means were compared using the statistical analysis described in the asso- ciated figure legends. All the data are presented as mean ± SEM and p < 0.05 was considered significant. �p < 0.05, ��p < 0.01, ���p < 0.001, ����p < 0.0001. Supporting information S1 Data. Excel spreadsheet containing the underlying numerical data for Figs 1A–1C, 2A and 2B, 3A, 5A–5D and 6A–6F. (XLSX) S1 Fig. Use of the Yptb-βla reporter to track Yop translocation. (A) MLN suspensions from L. monocytogenes infected mice were stimulated with WT or ΔYopB Y. pseudotuberculosis con- taining a β-lactamase translocation reporter (Yptb-βla) for 2 hours and given antibiotics. Cells were loaded with CCF4-AM dye to measure β-lactamase activity. FITC indicates CCF4-AM PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 19 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function loaded cells without translocation (Yop-) and BV421 indicates CCF4-AM loaded cells with Yop translocation (Yop+). Vγ1.1/2- CD44hi CD27- γδ T cells were analyzed for Yop transloca- tion 2 hours post stimulation at an MOI of 10. Representative contour plots are displayed. (B) MLN from L. monocytogenes infected mice were stimulated with Yptb-βla for 2 hours and given antibiotics. Yop translocation was detected as described above. The indicated cell popu- lations were analyzed for Yop translocation 2 hours after stimulation. Representative contour plots are displayed. (C) MLN from L. monocytogenes infected mice were stimulated with Yptb- βla for 2 hours and given antibiotics. Vγ1.1/2- CD44hi CD27- γδ T cells were analyzed for Yop translocation 2 hours post stimulation at the indicated MOI and quantified for Yop transloca- tion. Data consists of one experiment with 2–10 mice/group and the graphs depict the mean ± SEM in (A-C). ����p < 0.0001, ���p < 0.001, and ��p < 0.01. A t-test was used for (A), and a repeated measures one-way ANOVA was used for (C). Comparisons were performed to ΔYopB Y. pseudotuberculosis in (A) and as depicted in (C). (TIF) S2 Fig. The majority of Vγ1.1/2- CD44hi CD27- γδ T cells and γδ T cells containing Yop express β1-integrin. (A) MLN from L. monocytogenes infected mice were isolated and pro- cessed into single cell suspensions. Vγ1.1/2- CD44hi CD27- γδ T cells, CD4 T cells, and CD8 T cells were analyzed for β1-integrin expression. (B) MLN suspensions from L. monocytogenes infected mice were loaded with CCF4-AM dye and stimulated with 10 MOI WT Y. pseudotu- berculosis containing a β-lactamase translocation reporter. CCF4-AM dye reports the occur- rence of β-lactamase activity and Yop translocation. γδ T cells that contain Yop (Yop+) or do not contain Yop (Yop-) were analyzed for β1-integrin expression 2 hours after stimulation. Representative histogram plots are displayed. Data is pooled from two experiments with a total of 7 mice/group and the graphs depict the mean ± SEM in (A-C). ����p < 0.0001 and ���p < 0.001. A repeated measures one-way ANOVA was used for (A) and a t-test was used for (B), and. Comparisons were done to adaptive γδ T cells in (A) and as depicted in (B). (TIF) S3 Fig. YopJ regulates the transcriptional profile of Vγ1.1/2- CD44hi CD27- γδ T cells. (A and B) MLN from L. monocytogenes infected mice were stimulated with 10 MOI of WT Y. pseudotuberculosis (WT) or mutant YopJ Y. pseudotuberculosis (YopJC172A) for 24 hours. Anti- biotics were given 2 hours post-stimulation. Five hundred Vγ1.1/2- CD44hi CD27- γδ T cells from each stimulation were flow sorted and processed for RNA sequencing. The heat map depicts upregulated genes in Vγ1.1/2- CD44hi CD27- γδ T cells after YopJC172A Y. pseudotuber- culosis stimulation and individual genes are listed. (C-E) Genes differentially expressed (down- regulated) that overlapped between RNA sequencing analyses as displayed in the Venn diagram in (C) to select for direct effects of YopJ on Vγ1.1/2- CD44hi CD27- γδ T cells. The heat map depicts downregulated genes in Vγ1.1/2- CD44hi CD27- γδ T cells from the analysis in (C). Individual genes are listed in (E). Each experiment was performed once with biologic replicates. The cutoff for gene significance was p < 0.05 and FDR < 0.10. (TIF) S4 Fig. YopJ does not inhibit IL-17A production in Vγ1.1/2- CD44hi CD27- γδ T cells. MLN cell suspensions from L. monocytogenes infected mice were stimulated with 10 MOI of WT, YopJC172A, or ΔYopB Y. pseudotuberculosis for 24 hours. Antibiotics were given 2 hours after stimulation and brefeldin A was added for the last 5–6 hours. Vγ1.1/2- CD44hi CD27- γδ T cells were analyzed for IL-17A production after stimulation. Representative histograms are displayed. The graph depicts mean ± SEM and represents two independent experiments with PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 20 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function 4 mice per group. A repeated measures one-way ANOVA was used. � p < 0.05. (TIF) S5 Fig. YopJ impacts IFNγ related transcription factor motifs but not STAT4 protein. (A) Homer motif analysis was performed on the RNA sequencing results for the YopJC172A and WT Y. pseudotuberculosis comparison from Fig 5. The panel highlights the top transcription factor motifs of the ETS, SP/KLF, and IRF family of proteins identified in YopJC172A stimulated Vγ1.1/2- CD44hi CD27- γδ T cells. (B) MLN from L. monocytogenes infected mice were stimu- lated with 10 MOI of WT, YopJC172A, or ΔYopB Y. pseudotuberculosis for 6 hours. Antibiotics were given 2 hours post-stimulation. Vγ1.1/2- CD44hi CD27- γδ T cells were analyzed for IRF8 levels with or without anti-IL12p40 neutralizing antibody. The graph depicts the percentage of IRF8 protein expression among Vγ1.1/2- CD44hi CD27- γδ T cells after WT, YopJC172A, or ΔYopB Y. pseudotuberculosis stimulation. Data depict two pooled experiments with a total of 8 mice/group and represents the mean ± SEM. (C) The genes from the RNAseq and Homer motif analysis in Figs 4F and 4G and S3B and S5A were compared to an existing STAT4 ChIP- on-chip dataset to identify common genes. Genes from our dataset that were represented in the top 1000 genes of the Chip-on-chip dataset are displayed. (D) STAT4 KO spleens are shown in maroon and WT spleens are shown in black in representative histograms. The graph depicts the MFI of STAT4 protein expression in bulk γδ T cells. Data depicts one experiment with 4 mice/group and represents the mean ± SEM. ����p < 0.0001, ���p < 0.001, ��p < 0.01, �p < 0.05. A repeated measures one-way ANOVA was used for (B), and a t-test was used for (D). Comparisons were performed as depicted in (B) and to Naïve WT in (D). (TIF) S6 Fig. IL-12 is insufficient to induce IFNγ and does not readily overcome the inhibition of YopJ. (A) γδ T cells enriched from the MLN and spleen of L. monocytogenes infected mice were expanded with plate bound γδTCR antibody for 2 days and rested for 5 days. After expansion, ~ 50% of cells were γδ T cells, and the majority of those were Vγ4 T cells. The enrichment summary reflects the mean enrichment from 4 samples. Afterwards, γδ T cells were isolated from cultures and stimulated with YopJC172A Y. pseudotuberculosis with 0.1, 1, or 10 ng/ml IL-12p70 for 24 hours. Antibiotics were added 2 hours after stimulation and brefel- din A was added for the last 5–6 hours. Histograms display IFNγ production from Vγ1.1/2- CD44hi CD27- γδ T cells under different culture conditions. Data depicts one experiment with 4 mice pooled and split into the indicated stimulation conditions. (B) MLN cell suspensions from L. monocytogenes infected mice were stimulated with 10 MOI of WT Y. pseudotuberculo- sis in the presence of 2, 10, or 50 ng/ml IL-12p70 or 10 MOI of YopJC172A Y. pseudotuberculosis for 24 hours. Antibiotics were given 2 hours after stimulation and brefeldin A was added for the last 5–6 hours. Vγ1.1/2- CD44hi CD27- γδ T cells were analyzed for IFNγ production. Rep- resentative histograms of IFNγ production from Vγ1.1/2- CD44hi CD27- γδ T cells are dis- played. The graph depicts mean ± SEM from one experiment with 4 mice per group �p < 0.05. A repeated measures one-way ANOVA was used for comparisons to YopJC172A Y. pseudotu- berculosis in (B). (TIF) S7 Fig. The impact of foodborne infection of mice with Y. pseudotuberculosis. (A) Balb/c mice were foodborne infected with the indicated doses of WT or mutant YopJC172A Y. pseudo- tuberculosis and tissues were analyzed 9 days post-infection. Bacteria burden was quantified from the MLN. Data reflect 3–5 mice per group pooled from 3 independent experiments and the graphs depict the mean ± SEM. (B) Balb/c mice were foodborne infected with the indicated doses of WT or mutant YopJC172A Y. pseudotuberculosis. Nine days post infection, MLN PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 21 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function suspensions from WT or YopJC172A Y. pseudotuberculosis infected mice were stimulated with PMA/ionomycin and brefeldin A for 4 hours. Vγ1.1/2- CD44hi CD27- γδ T cells were analyzed for IFNγ production. Representative histograms are displayed and quantified. Data depicts one experiment with 3 mice per group. (C) Balb/c mice were foodborne infected with WT (2- 4x107 CFU) or YopJC172A Y. pseudotuberculosis (2-4x108 CFU) and treated with 0.2 mg/mouse of anti-IL12p40 on days 0, 2, 4, and 6 post infection. IL-12p70 concentrations were determined from serum at days 2, 4, and 6 post infection. Data represent 2 independent experiments with a total of 9 mice per group. Serum samples were pooled into groups of 3 per experimental con- dition. (D) Balb/c mice were foodborne infected with 2x109 CFU L. monocytogenes to elicit a Vγ1.1/2- CD44hi CD27- γδ T cell population in vivo. 30 days post infection, adaptive Vγ1.1/2- CD44hi CD27- γδ T cells from the MLN of immune mice were analyzed for Yop translocation using the CCF4-AM assay. Representative contour plots are shown. Yop translocation (Yop+) among the indicated populations represents background staining as a negative control for Fig 6F. The graph depicts the mean ± SEM and is pooled from 2 experiments with a total of 4 mice per group. ����p < 0.0001, �p < 0.05. A one-way ANOVA was used for (A), and an unpaired t-test was used for (B). Comparisons were performed to uninfected in (A) and to 5x107 WT Y. pseudotuberculosis in (B). (TIF) Author Contributions Conceptualization: Timothy H. Chu, Camille Khairallah, James B. Bliska, Brian S. Sheridan. Data curation: Timothy H. Chu, Camille Khairallah, Jason Shieh. Formal analysis: Timothy H. Chu, Camille Khairallah, Jason Shieh. Funding acquisition: Brian S. Sheridan. Investigation: Timothy H. Chu, Camille Khairallah, Rhea Cho, Zhijuan Qiu. Methodology: Timothy H. Chu, Camille Khairallah, Yue Zhang, Onur Eskiocak, Semir Beyaz, Brian S. Sheridan. Project administration: Brian S. Sheridan. Resources: David G. Thanassi, Mark H. Kaplan, Semir Beyaz, Vincent W. Yang, James B. Bliska. Supervision: Brian S. Sheridan. Validation: Timothy H. Chu. Visualization: Timothy H. Chu, Brian S. Sheridan. Writing – original draft: Timothy H. Chu. Writing – review & editing: Timothy H. Chu, Camille Khairallah, David G. Thanassi, James B. Bliska, Brian S. Sheridan. References 1. Barnes PD, Bergman MA, Mecsas J, Isberg RR. Yersinia pseudotuberculosis disseminates directly from a replicating bacterial pool in the intestine. J Exp Med. 2006; 203(6):1591–601. Epub 2006/06/07. https://doi.org/10.1084/jem.20060905 PMID: 16754724; PubMed Central PMCID: PMC2118325. 2. Autenrieth IB, Kempf V, Sprinz T, Preger S, Schnell A. Defense mechanisms in Peyer’s patches and mesenteric lymph nodes against Yersinia enterocolitica involve integrins and cytokines. Infect Immun. PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 22 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function 1996; 64(4):1357–68. Epub 1996/04/01. https://doi.org/10.1128/iai.64.4.1357-1368.1996 PMID: 8606101; PubMed Central PMCID: PMC173926. 3. Titball RW, Leary SE. Plague. Br Med Bull. 1998; 54(3):625–33. Epub 1999/05/18. https://doi.org/10. 1093/oxfordjournals.bmb.a011715 PMID: 10326289. 4. Hinnebusch BJ. Bubonic plague: a molecular genetic case history of the emergence of an infectious disease. J Mol Med (Berl). 1997; 75(9):645–52. Epub 1997/11/14. https://doi.org/10.1007/ s001090050148 PMID: 9351703. 5. 6. Titball RW, Williamson ED. Vaccination against bubonic and pneumonic plague. Vaccine. 2001; 19 (30):4175–84. Epub 2001/07/18. https://doi.org/10.1016/s0264-410x(01)00163-3 PMID: 11457543. Inglesby TV, Dennis DT, Henderson DA, Bartlett JG, Ascher MS, Eitzen E, et al. Plague as a biological weapon: medical and public health management. Working Group on Civilian Biodefense. JAMA. 2000; 283(17):2281–90. Epub 2000/05/12. https://doi.org/10.1001/jama.283.17.2281 PMID: 10807389. 7. Nakajima R, Brubaker RR. Association between virulence of Yersinia pestis and suppression of gamma interferon and tumor necrosis factor alpha. Infect Immun. 1993; 61(1):23–31. Epub 1993/01/01. https://doi.org/10.1128/iai.61.1.23-31.1993 PMID: 8418045; PubMed Central PMCID: PMC302683. 8. Cornelis GR, Boland A, Boyd AP, Geuijen C, Iriarte M, Neyt C, et al. The virulence plasmid of Yersinia, an antihost genome. Microbiol Mol Biol Rev. 1998; 62(4):1315–52. Epub 1998/12/05. https://doi.org/ 10.1128/MMBR.62.4.1315-1352.1998 PMID: 9841674; PubMed Central PMCID: PMC98948. 9. Cornelis GR, Wolf-Watz H. The Yersinia Yop virulon: a bacterial system for subverting eukaryotic cells. Mol Microbiol. 1997; 23(5):861–7. Epub 1997/03/01. https://doi.org/10.1046/j.1365-2958.1997. 2731623.x PMID: 9076724. 10. Koberle M, Klein-Gunther A, Schutz M, Fritz M, Berchtold S, Tolosa E, et al. Yersinia enterocolitica tar- gets cells of the innate and adaptive immune system by injection of Yops in a mouse infection model. PLoS Pathog. 2009; 5(8):e1000551. Epub 2009/08/15. https://doi.org/10.1371/journal.ppat.1000551 PMID: 19680448; PubMed Central PMCID: PMC2718809. 11. Maldonado-Arocho FJ, Green C, Fisher ML, Paczosa MK, Mecsas J. Adhesins and host serum factors drive Yop translocation by yersinia into professional phagocytes during animal infection. PLoS Pathog. 2013; 9(6):e1003415. Epub 2013/07/03. https://doi.org/10.1371/journal.ppat.1003415 PMID: 23818844; PubMed Central PMCID: PMC3688556. 12. Durand EA, Maldonado-Arocho FJ, Castillo C, Walsh RL, Mecsas J. The presence of professional phagocytes dictates the number of host cells targeted for Yop translocation during infection. Cell Microbiol. 2010; 12(8):1064–82. Epub 2010/02/13. https://doi.org/10.1111/j.1462-5822.2010.01451.x PMID: 20148898; PubMed Central PMCID: PMC2906667. 13. Viboud GI, Bliska JB. Yersinia outer proteins: role in modulation of host cell signaling responses and pathogenesis. Annu Rev Microbiol. 2005; 59:69–89. Epub 2005/04/26. https://doi.org/10.1146/ annurev.micro.59.030804.121320 PMID: 15847602. 14. Cornelis GR. Yersinia type III secretion: send in the effectors. J Cell Biol. 2002; 158(3):401–8. Epub 2002/ 08/07. https://doi.org/10.1083/jcb.200205077 PMID: 12163464; PubMed Central PMCID: PMC2173816. 15. Bliska JB. Yersinia inhibits host signaling by acetylating MAPK kinases. ACS Chem Biol. 2006; 1 (6):349–51. Epub 2006/12/14. https://doi.org/10.1021/cb600261k PMID: 17163770. 16. Sweet CR, Conlon J, Golenbock DT, Goguen J, Silverman N. YopJ targets TRAF proteins to inhibit TLR-mediated NF-kappaB, MAPK and IRF3 signal transduction. Cell Microbiol. 2007; 9(11):2700–15. Epub 2007/07/05. https://doi.org/10.1111/j.1462-5822.2007.00990.x PMID: 17608743. 17. Cao Y, Guan K, He X, Wei C, Zheng Z, Zhang Y, et al. Yersinia YopJ negatively regulates IRF3-medi- ated antibacterial response through disruption of STING-mediated cytosolic DNA signaling. Biochim Biophys Acta. 2016; 1863(12):3148–59. Epub 2016/11/05. https://doi.org/10.1016/j.bbamcr.2016.10. 004 PMID: 27742471. 18. Mukherjee S, Orth K. In vitro signaling by MAPK and NFkappaB pathways inhibited by Yersinia YopJ. Methods Enzymol. 2008; 438:343–53. Epub 2008/04/17. https://doi.org/10.1016/S0076-6879(07) 38024-5 PMID: 18413260; PubMed Central PMCID: PMC3432501. 19. Ma KW, Ma W. YopJ Family Effectors Promote Bacterial Infection through a Unique Acetyltransferase Activity. Microbiol Mol Biol Rev. 2016; 80(4):1011–27. Epub 2016/10/28. https://doi.org/10.1128/ MMBR.00032-16 PMID: 27784797; PubMed Central PMCID: PMC5116873. 20. Orning P, Weng D, Starheim K, Ratner D, Best Z, Lee B, et al. Pathogen blockade of TAK1 triggers caspase-8-dependent cleavage of gasdermin D and cell death. Science. 2018; 362(6418):1064–9. Epub 2018/10/27. https://doi.org/10.1126/science.aau2818 PMID: 30361383; PubMed Central PMCID: PMC6522129. 21. Rosadini CV, Zanoni I, Odendall C, Green ER, Paczosa MK, Philip NH, et al. A Single Bacterial Immune Evasion Strategy Dismantles Both MyD88 and TRIF Signaling Pathways Downstream of PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 23 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function TLR4. Cell Host Microbe. 2015; 18(6):682–93. Epub 2015/12/15. https://doi.org/10.1016/j.chom.2015. 11.006 PMID: 26651944; PubMed Central PMCID: PMC4685476. 22. Trulzsch K, Geginat G, Sporleder T, Ruckdeschel K, Hoffmann R, Heesemann J, et al. Yersinia outer protein P inhibits CD8 T cell priming in the mouse infection model. J Immunol. 2005; 174(7):4244–51. Epub 2005/03/22. https://doi.org/10.4049/jimmunol.174.7.4244 PMID: 15778387. 23. Alonso A, Bottini N, Bruckner S, Rahmouni S, Williams S, Schoenberger SP, et al. Lck dephosphoryla- tion at Tyr-394 and inhibition of T cell antigen receptor signaling by Yersinia phosphatase YopH. J Biol Chem. 2004; 279(6):4922–8. Epub 2003/11/19. https://doi.org/10.1074/jbc.M308978200 PMID: 14623872. 24. Bruckner S, Rhamouni S, Tautz L, Denault JB, Alonso A, Becattini B, et al. Yersinia phosphatase induces mitochondrially dependent apoptosis of T cells. J Biol Chem. 2005; 280(11):10388–94. Epub 2005/01/06. https://doi.org/10.1074/jbc.M408829200 PMID: 15632192. 25. Yao T, Mecsas J, Healy JI, Falkow S, Chien Y. Suppression of T and B lymphocyte activation by a Yer- sinia pseudotuberculosis virulence factor, yopH. J Exp Med. 1999; 190(9):1343–50. Epub 1999/11/02. https://doi.org/10.1084/jem.190.9.1343 PMID: 10544205; PubMed Central PMCID: PMC2195683. 26. Gerke C, Falkow S, Chien YH. The adaptor molecules LAT and SLP-76 are specifically targeted by Yersinia to inhibit T cell activation. J Exp Med. 2005; 201(3):361–71. Epub 2005/02/09. https://doi.org/ 10.1084/jem.20041120 PMID: 15699071; PubMed Central PMCID: PMC2213036. 27. Viney J, MacDonald TT, Spencer J. Gamma/delta T cells in the gut epithelium. Gut. 1990; 31(8):841– 4. Epub 1990/08/01. https://doi.org/10.1136/gut.31.8.841 PMID: 2143741; PubMed Central PMCID: PMC1378605. 28. Goodman T, Lefrancois L. Expression of the gamma-delta T-cell receptor on intestinal CD8+ intrae- pithelial lymphocytes. Nature. 1988; 333(6176):855–8. Epub 1988/06/30. https://doi.org/10.1038/ 333855a0 PMID: 2968521. 29. Ribeiro ST, Ribot JC, Silva-Santos B. Five Layers of Receptor Signaling in gammadelta T-Cell Differ- entiation and Activation. Front Immunol. 2015; 6:15. Epub 2015/02/13. https://doi.org/10.3389/fimmu. 2015.00015 PMID: 25674089; PubMed Central PMCID: PMC4306313. 30. Yin Z, Zhang DH, Welte T, Bahtiyar G, Jung S, Liu L, et al. Dominance of IL-12 over IL-4 in gamma delta T cell differentiation leads to default production of IFN-gamma: failure to down-regulate IL-12 receptor beta 2-chain expression. J Immunol. 2000; 164(6):3056–64. Epub 2000/03/08. https://doi. org/10.4049/jimmunol.164.6.3056 PMID: 10706694. 31. Li W, Kubo S, Okuda A, Yamamoto H, Ueda H, Tanaka T, et al. Effect of IL-18 on expansion of gam- madelta T cells stimulated by zoledronate and IL-2. J Immunother. 2010; 33(3):287–96. Epub 2010/ 05/07. https://doi.org/10.1097/CJI.0b013e3181c80ffa PMID: 20445349. 32. Moens E, Brouwer M, Dimova T, Goldman M, Willems F, Vermijlen D. IL-23R and TCR signaling drives the generation of neonatal Vgamma9Vdelta2 T cells expressing high levels of cytotoxic media- tors and producing IFN-gamma and IL-17. J Leukoc Biol. 2011; 89(5):743–52. Epub 2011/02/19. https://doi.org/10.1189/jlb.0910501 PMID: 21330350. 33. Sutton CE, Lalor SJ, Sweeney CM, Brereton CF, Lavelle EC, Mills KH. Interleukin-1 and IL-23 induce innate IL-17 production from gammadelta T cells, amplifying Th17 responses and autoimmunity. Immunity. 2009; 31(2):331–41. Epub 2009/08/18. https://doi.org/10.1016/j.immuni.2009.08.001 PMID: 19682929. 34. Lockhart E, Green AM, Flynn JL. IL-17 production is dominated by gammadelta T cells rather than CD4 T cells during Mycobacterium tuberculosis infection. J Immunol. 2006; 177(7):4662–9. Epub 2006/09/20. https://doi.org/10.4049/jimmunol.177.7.4662 PMID: 16982905. 35. Ribot JC, Chaves-Ferreira M, d’Orey F, Wencker M, Goncalves-Sousa N, Decalf J, et al. Cutting edge: adaptive versus innate receptor signals selectively control the pool sizes of murine IFN-gamma- or IL- 17-producing gammadelta T cells upon infection. J Immunol. 2010; 185(11):6421–5. Epub 2010/11/03. https://doi.org/10.4049/jimmunol.1002283 PMID: 21037088; PubMed Central PMCID: PMC3915338. 36. Garman RD, Doherty PJ, Raulet DH. Diversity, rearrangement, and expression of murine T cell gamma genes. Cell. 1986; 45(5):733–42. Epub 1986/06/06. https://doi.org/10.1016/0092-8674(86) 90787-7 PMID: 3486721. 37. Sheridan BS, Romagnoli PA, Pham QM, Fu HH, Alonzo F, 3rd, Schubert WD, et al. gammadelta T cells exhibit multifunctional and protective memory in intestinal tissues. Immunity. 2013; 39(1):184–95. Epub 2013/07/31. https://doi.org/10.1016/j.immuni.2013.06.015 PMID: 23890071; PubMed Central PMCID: PMC3749916. 38. Romagnoli PA, Sheridan BS, Pham QM, Lefrancois L, Khanna KM. IL-17A-producing resident mem- ory gammadelta T cells orchestrate the innate immune response to secondary oral Listeria monocyto- genes infection. Proc Natl Acad Sci U S A. 2016; 113(30):8502–7. Epub 2016/07/13. https://doi.org/ 10.1073/pnas.1600713113 PMID: 27402748; PubMed Central PMCID: PMC4968747. PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 24 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function 39. Dillen CA, Pinsker BL, Marusina AI, Merleev AA, Farber ON, Liu H, et al. Clonally expanded gamma- delta T cells protect against Staphylococcus aureus skin reinfection. J Clin Invest. 2018; 128(3):1026– 42. Epub 2018/02/06. https://doi.org/10.1172/JCI96481 PMID: 29400698; PubMed Central PMCID: PMC5824877. 40. Bohn E, Autenrieth IB. IL-12 is essential for resistance against Yersinia enterocolitica by triggering IFN-gamma production in NK cells and CD4+ T cells. J Immunol. 1996; 156(4):1458–68. Epub 1996/ 02/15. PMID: 8568248. 41. Szaba FM, Kummer LW, Duso DK, Koroleva EP, Tumanov AV, Cooper AM, et al. TNFalpha and IFN- gamma but not perforin are critical for CD8 T cell-mediated protection against pulmonary Yersinia pes- tis infection. PLoS Pathog. 2014; 10(5):e1004142. Epub 2014/05/24. https://doi.org/10.1371/journal. ppat.1004142 PMID: 24854422; PubMed Central PMCID: PMC4031182. 42. Logsdon LK, Mecsas J. The proinflammatory response induced by wild-type Yersinia pseudotubercu- losis infection inhibits survival of yop mutants in the gastrointestinal tract and Peyer’s patches. Infect Immun. 2006; 74(3):1516–27. Epub 2006/02/24. https://doi.org/10.1128/IAI.74.3.1516-1527.2006 PMID: 16495522; PubMed Central PMCID: PMC1418670. 43. Seo GY, Shui JW, Takahashi D, Song C, Wang Q, Kim K, et al. LIGHT-HVEM Signaling in Innate Lym- phoid Cell Subsets Protects Against Enteric Bacterial Infection. Cell Host Microbe. 2018; 24(2):249– 60 e4. Epub 2018/08/10. https://doi.org/10.1016/j.chom.2018.07.008 PMID: 30092201; PubMed Cen- tral PMCID: PMC6132068. 44. 45. Imperato JN, Xu D, Romagnoli PA, Qiu Z, Perez P, Khairallah C, et al. Mucosal CD8 T Cell Responses Are Shaped by Batf3-DC After Foodborne Listeria monocytogenes Infection. Front Immunol. 2020; 11:575967. Epub 2020/10/13. https://doi.org/10.3389/fimmu.2020.575967 PMID: 33042159; PubMed Central PMCID: PMC7518468. Zhang Y, Murtha J, Roberts MA, Siegel RM, Bliska JB. Type III secretion decreases bacterial and host survival following phagocytosis of Yersinia pseudotuberculosis by macrophages. Infect Immun. 2008; 76(9):4299–310. Epub 2008/07/02. https://doi.org/10.1128/IAI.00183-08 PMID: 18591234; PubMed Central PMCID: PMC2519449. 46. Palmer LE, Hobbie S, Galan JE, Bliska JB. YopJ of Yersinia pseudotuberculosis is required for the inhibition of macrophage TNF-alpha production and downregulation of the MAP kinases p38 and JNK. Mol Microbiol. 1998; 27(5):953–65. Epub 1998/04/16. https://doi.org/10.1046/j.1365-2958.1998. 00740.x PMID: 9535085. 47. Simonet M, Falkow S. Invasin expression in Yersinia pseudotuberculosis. Infect Immun. 1992; 60 (10):4414–7. Epub 1992/10/01. https://doi.org/10.1128/iai.60.10.4414-4417.1992 PMID: 1398952; PubMed Central PMCID: PMC257481. 48. Ivanov MI, Stuckey JA, Schubert HL, Saper MA, Bliska JB. Two substrate-targeting sites in the Yersinia protein tyrosine phosphatase co-operate to promote bacterial virulence. Mol Microbiol. 2005; 55(5):1346–56. Epub 2005/02/22. https://doi.org/10.1111/j.1365-2958.2005.04477.x PMID: 15720545. 49. Bliska JB, Guan KL, Dixon JE, Falkow S. Tyrosine phosphate hydrolysis of host proteins by an essen- tial Yersinia virulence determinant. Proc Natl Acad Sci U S A. 1991; 88(4):1187–91. Epub 1991/02/15. https://doi.org/10.1073/pnas.88.4.1187 PMID: 1705028; PubMed Central PMCID: PMC50982. 50. McPhee JB, Mena P, Bliska JB. Delineation of regions of the Yersinia YopM protein required for inter- action with the RSK1 and PRK2 host kinases and their requirement for interleukin-10 production and virulence. Infect Immun. 2010; 78(8):3529–39. Epub 2010/06/03. https://doi.org/10.1128/IAI.00269-10 PMID: 20515922; PubMed Central PMCID: PMC2916259. 51. Schoberle TJ, Chung LK, McPhee JB, Bogin B, Bliska JB. Uncovering an Important Role for YopJ in the Inhibition of Caspase-1 in Activated Macrophages and Promoting Yersinia pseudotuberculosis Vir- ulence. Infect Immun. 2016; 84(4):1062–72. Epub 2016/01/27. https://doi.org/10.1128/IAI.00843-15 PMID: 26810037; PubMed Central PMCID: PMC4807483. 52. Chung LK, Philip NH, Schmidt VA, Koller A, Strowig T, Flavell RA, et al. IQGAP1 is important for acti- vation of caspase-1 in macrophages and is targeted by Yersinia pestis type III effector YopM. mBio. 2014; 5(4):e01402–14. Epub 2014/07/06. https://doi.org/10.1128/mBio.01402-14 PMID: 24987096; PubMed Central PMCID: PMC4161239. 53. Harmon DE, Davis AJ, Castillo C, Mecsas J. Identification and characterization of small-molecule inhibitors of Yop translocation in Yersinia pseudotuberculosis. Antimicrob Agents Chemother. 2010; 54(8):3241–54. Epub 2010/05/26. https://doi.org/10.1128/AAC.00364-10 PMID: 20498321; PubMed Central PMCID: PMC2916352. 54. Schweer J, Kulkarni D, Kochut A, Pezoldt J, Pisano F, Pils MC, et al. The cytotoxic necrotizing factor of Yersinia pseudotuberculosis (CNFY) enhances inflammation and Yop delivery during infection by activation of Rho GTPases. PLoS Pathog. 2013; 9(11):e1003746. Epub 2013/11/19. https://doi.org/ 10.1371/journal.ppat.1003746 PMID: 24244167; PubMed Central PMCID: PMC3820761. PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 25 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function 55. 56. Zhang Y, Tam JW, Mena P, van der Velden AW, Bliska JB. CCR2+ Inflammatory Dendritic Cells and Translocation of Antigen by Type III Secretion Are Required for the Exceptionally Large CD8+ T Cell Response to the Protective YopE69-77 Epitope during Yersinia Infection. PLoS Pathog. 2015; 11(10): e1005167. Epub 2015/10/16. https://doi.org/10.1371/journal.ppat.1005167 PMID: 26468944; PubMed Central PMCID: PMC4607306. Zhang Y, Romanov G, Bliska JB. Type III secretion system-dependent translocation of ectopically expressed Yop effectors into macrophages by intracellular Yersinia pseudotuberculosis. Infect Immun. 2011; 79(11):4322–31. Epub 2011/08/17. https://doi.org/10.1128/IAI.05396-11 PMID: 21844228; PubMed Central PMCID: PMC3257923. 57. Kim J, Fukuto HS, Brown DA, Bliska JB, London E. Effects of host cell sterol composition upon inter- nalization of Yersinia pseudotuberculosis and clustered beta1 integrin. J Biol Chem. 2018; 293 (4):1466–79. Epub 2017/12/05. https://doi.org/10.1074/jbc.M117.811224 PMID: 29197826; PubMed Central PMCID: PMC5787820. 58. Bohn E, Sonnabend M, Klein K, Autenrieth IB. Bacterial adhesion and host cell factors leading to effec- tor protein injection by type III secretion system. Int J Med Microbiol. 2019; 309(5):344–50. Epub 2019/ 06/11. https://doi.org/10.1016/j.ijmm.2019.05.008 PMID: 31178419. 59. Isberg RR, Voorhis DL, Falkow S. Identification of invasin: a protein that allows enteric bacteria to pen- etrate cultured mammalian cells. Cell. 1987; 50(5):769–78. Epub 1987/08/28. https://doi.org/10.1016/ 0092-8674(87)90335-7 PMID: 3304658. 60. Schulze-Koops H, Burkhardt H, Heesemann J, von der Mark K, Emmrich F. Plasmid-encoded outer membrane protein YadA mediates specific binding of enteropathogenic yersiniae to various types of collagen. Infect Immun. 1992; 60(6):2153–9. Epub 1992/06/01. https://doi.org/10.1128/iai.60.6.2153- 2159.1992 PMID: 1587583; PubMed Central PMCID: PMC257137. 61. Deuschle E, Keller B, Siegfried A, Manncke B, Spaeth T, Koberle M, et al. Role of beta1 integrins and bacterial adhesins for Yop injection into leukocytes in Yersinia enterocolitica systemic mouse infection. Int J Med Microbiol. 2016; 306(2):77–88. Epub 2016/01/01. https://doi.org/10.1016/j.ijmm.2015.12. 001 PMID: 26718660. 62. Bliska JB, Copass MC, Falkow S. The Yersinia pseudotuberculosis adhesin YadA mediates intimate bac- terial attachment to and entry into HEp-2 cells. Infect Immun. 1993; 61(9):3914–21. Epub 1993/09/01. https://doi.org/10.1128/iai.61.9.3914-3921.1993 PMID: 7689542; PubMed Central PMCID: PMC281094. 63. Nguyen GT, McCabe AL, Fasciano AC, Mecsas J. Detection of Cells Translocated with Yersinia Yops in Infected Tissues Using beta-Lactamase Fusions. Methods Mol Biol. 2019; 2010:117–39. Epub 2019/06/10. https://doi.org/10.1007/978-1-4939-9541-7_9 PMID: 31177435; PubMed Central PMCID: PMC6733027. 64. Pasztoi M, Bonifacius A, Pezoldt J, Kulkarni D, Niemz J, Yang J, et al. Yersinia pseudotuberculosis supports Th17 differentiation and limits de novo regulatory T cell induction by directly interfering with T cell receptor signaling. Cell Mol Life Sci. 2017; 74(15):2839–50. Epub 2017/04/06. https://doi.org/10. 1007/s00018-017-2516-y PMID: 28378044; PubMed Central PMCID: PMC5491567. 65. Elfiky A, Bonifacius A, Pezoldt J, Pasztoi M, Chaoprasid P, Sadana P, et al. Yersinia Pseudotuberculo- sis Modulates Regulatory T Cell Stability via Injection of Yersinia Outer Proteins in a Type III Secretion System-Dependent Manner. Eur J Microbiol Immunol (Bp). 2018; 8(4):101–6. Epub 2019/02/06. https://doi.org/10.1556/1886.2018.00015 PMID: 30719325; PubMed Central PMCID: PMC6348704. 66. Clark MA, Hirst BH, Jepson MA. M-cell surface beta1 integrin expression and invasin-mediated target- ing of Yersinia pseudotuberculosis to mouse Peyer’s patch M cells. Infect Immun. 1998; 66(3):1237– 43. Epub 1998/03/06. https://doi.org/10.1128/IAI.66.3.1237-1243.1998 PMID: 9488419; PubMed Central PMCID: PMC108039. 67. El Tahir Y, Skurnik M. YadA, the multifaceted Yersinia adhesin. Int J Med Microbiol. 2001; 291(3):209– 18. Epub 2001/09/14. https://doi.org/10.1078/1438-4221-00119 PMID: 11554561. 68. 69. Tertti R, Skurnik M, Vartio T, Kuusela P. Adhesion protein YadA of Yersinia species mediates binding of bacteria to fibronectin. Infect Immun. 1992; 60(7):3021–4. Epub 1992/07/01. https://doi.org/10. 1128/iai.60.7.3021-3024.1992 PMID: 1612772; PubMed Central PMCID: PMC257272. Flugel A, Schulze-Koops H, Heesemann J, Kuhn K, Sorokin L, Burkhardt H, et al. Interaction of entero- pathogenic Yersinia enterocolitica with complex basement membranes and the extracellular matrix proteins collagen type IV, laminin-1 and -2, and nidogen/entactin. J Biol Chem. 1994; 269(47):29732– 8. Epub 1994/11/25. PMID: 7961965. 70. Mukherjee S, Keitany G, Li Y, Wang Y, Ball HL, Goldsmith EJ, et al. Yersinia YopJ acetylates and inhibits kinase activation by blocking phosphorylation. Science. 2006; 312(5777):1211–4. Epub 2006/ 05/27. https://doi.org/10.1126/science.1126867 PMID: 16728640. 71. Lee J, Aoki T, Thumkeo D, Siriwach R, Yao C, Narumiya S. T cell-intrinsic prostaglandin E2-EP2/EP4 signaling is critical in pathogenic TH17 cell-driven inflammation. J Allergy Clin Immunol. 2019; 143 PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 26 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function (2):631–43. Epub 2018/06/24. https://doi.org/10.1016/j.jaci.2018.05.036 PMID: 29935220; PubMed Central PMCID: PMC6354914. 72. Ng SS, De Labastida Rivera F, Yan J, Corvino D, Das I, Zhang P, et al. The NK cell granule protein NKG7 regulates cytotoxic granule exocytosis and inflammation. Nat Immunol. 2020; 21(10):1205–18. Epub 2020/08/26. https://doi.org/10.1038/s41590-020-0758-6 PMID: 32839608. 73. Malarkannan S. NKG7 makes a better killer. Nat Immunol. 2020; 21(10):1139–40. Epub 2020/08/26. https://doi.org/10.1038/s41590-020-0767-5 PMID: 32839609. 74. Heinz S, Benner C, Spann N, Bertolino E, Lin YC, Laslo P, et al. Simple combinations of lineage-deter- mining transcription factors prime cis-regulatory elements required for macrophage and B cell identi- ties. Mol Cell. 2010; 38(4):576–89. Epub 2010/06/02. https://doi.org/10.1016/j.molcel.2010.05.004 PMID: 20513432; PubMed Central PMCID: PMC2898526. 75. Pham D, Sehra S, Sun X, Kaplan MH. The transcription factor Etv5 controls TH17 cell development and allergic airway inflammation. J Allergy Clin Immunol. 2014; 134(1):204–14. Epub 2014/02/04. https://doi.org/10.1016/j.jaci.2013.12.021 PMID: 24486067; PubMed Central PMCID: PMC4209254. 76. Yin L, Li W, Xu A, Shi H, Wang K, Yang H, et al. SH3BGRL2 inhibits growth and metastasis in clear cell renal cell carcinoma via activating hippo/TEAD1-Twist1 pathway. EBioMedicine. 2020; 51:102596. Epub 2020/01/09. https://doi.org/10.1016/j.ebiom.2019.12.005 PMID: 31911271; PubMed Central PMCID: PMC7000347. 77. Pham D, Vincentz JW, Firulli AB, Kaplan MH. Twist1 regulates Ifng expression in Th1 cells by interfer- ing with Runx3 function. J Immunol. 2012; 189(2):832–40. Epub 2012/06/12. https://doi.org/10.4049/ jimmunol.1200854 PMID: 22685315; PubMed Central PMCID: PMC3392532. 78. Wang Y, Godec J, Ben-Aissa K, Cui K, Zhao K, Pucsek AB, et al. The transcription factors T-bet and Runx are required for the ontogeny of pathogenic interferon-gamma-producing T helper 17 cells. Immunity. 2014; 40(3):355–66. Epub 2014/02/18. https://doi.org/10.1016/j.immuni.2014.01.002 PMID: 24530058; PubMed Central PMCID: PMC3965587. 79. Good SR, Thieu VT, Mathur AN, Yu Q, Stritesky GL, Yeh N, et al. Temporal induction pattern of STAT4 target genes defines potential for Th1 lineage-specific programming. J Immunol. 2009; 183 (6):3839–47. Epub 2009/08/28. https://doi.org/10.4049/jimmunol.0901411 PMID: 19710469; PubMed Central PMCID: PMC2748807. 80. Xu X, Sun YL, Hoey T. Cooperative DNA binding and sequence-selective recognition conferred by the STAT amino-terminal domain. Science. 1996; 273(5276):794–7. Epub 1996/08/09. https://doi.org/10. 1126/science.273.5276.794 PMID: 8670419. 81. Yamamoto K, Miura O, Hirosawa S, Miyasaka N. Binding sequence of STAT4: STAT4 complex recog- nizes the IFN-gamma activation site (GAS)-like sequence (T/A)TTCC(C/G)GGAA(T/A). Biochem Bio- phys Res Commun. 1997; 233(1):126–32. Epub 1997/04/07. https://doi.org/10.1006/bbrc.1997.6415 PMID: 9144409. 82. Zhang Y, Ting AT, Marcu KB, Bliska JB. Inhibition of MAPK and NF-kappa B pathways is necessary for rapid apoptosis in macrophages infected with Yersinia. J Immunol. 2005; 174(12):7939–49. Epub 2005/06/10. https://doi.org/10.4049/jimmunol.174.12.7939 PMID: 15944300. 83. Khairallah C, Bettke JA, Gorbatsevych O, Qiu Z, Zhang Y, Cho K, et al. A blend of broadly-reactive and pathogen-selected Vgamma4 Vdelta1 T cell receptors confer broad bacterial reactivity of resident memory gammadelta T cells. Mucosal Immunol. 2021. Epub 2021/09/01. https://doi.org/10.1038/ s41385-021-00447-x PMID: 34462572. 84. Young JL, Goodall JC, Beacock-Sharp H, Gaston JS. Human gamma delta T-cell recognition of Yersi- nia enterocolitica. Immunology. 1997; 91(4):503–10. Epub 1997/08/01. https://doi.org/10.1046/j.1365- 2567.1997.00289.x PMID: 9378487; PubMed Central PMCID: PMC1363868. 85. Hermann E, Ackermann B, Duchmann R, Meyer zum Buschenfelde KH. Synovial fluid MHC- unrestricted gamma delta-T lymphocytes contribute to antibacterial and anti-self cytotoxicity in the spondylarthropathies. Clin Exp Rheumatol. 1995; 13(2):187–91. Epub 1995/03/01. PMID: 7656464. 86. Huang D, Chen CY, Ali Z, Shao L, Shen L, Lockman HA, et al. Antigen-specific Vgamma2Vdelta2 T effector cells confer homeostatic protection against pneumonic plaque lesions. Proc Natl Acad Sci U S A. 2009; 106(18):7553–8. Epub 2009/04/23. https://doi.org/10.1073/pnas.0811250106 PMID: 19383786; PubMed Central PMCID: PMC2678605. 87. Puan KJ, Jin C, Wang H, Sarikonda G, Raker AM, Lee HK, et al. Preferential recognition of a microbial metabolite by human Vgamma2Vdelta2 T cells. Int Immunol. 2007; 19(5):657–73. Epub 2007/04/21. https://doi.org/10.1093/intimm/dxm031 PMID: 17446209. 88. Hermann E, Lohse AW, Mayet WJ, van der Zee R, Van Eden W, Probst P, et al. Stimulation of synovial fluid mononuclear cells with the human 65-kD heat shock protein or with live enterobacteria leads to preferential expansion of TCR-gamma delta+ lymphocytes. Clin Exp Immunol. 1992; 89(3):427–33. PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 27 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function Epub 1992/09/01. https://doi.org/10.1111/j.1365-2249.1992.tb06975.x PMID: 1387595; PubMed Cen- tral PMCID: PMC1554482. 89. Deseke M, Prinz I. Ligand recognition by the gammadelta TCR and discrimination between homeosta- sis and stress conditions. Cell Mol Immunol. 2020; 17(9):914–24. Epub 2020/07/28. https://doi.org/10. 1038/s41423-020-0503-y PMID: 32709926. 90. Cui W, Joshi NS, Jiang A, Kaech SM. Effects of Signal 3 during CD8 T cell priming: Bystander produc- tion of IL-12 enhances effector T cell expansion but promotes terminal differentiation. Vaccine. 2009; 27(15):2177–87. Epub 2009/02/10. https://doi.org/10.1016/j.vaccine.2009.01.088 PMID: 19201385; PubMed Central PMCID: PMC2803112. 91. Robinson RT, Khader SA, Locksley RM, Lien E, Smiley ST, Cooper AM. Yersinia pestis evades TLR4- dependent induction of IL-12(p40)2 by dendritic cells and subsequent cell migration. J Immunol. 2008; 181(8):5560–7. Epub 2008/10/04. https://doi.org/10.4049/jimmunol.181.8.5560 PMID: 18832714; PubMed Central PMCID: PMC2640496. 92. Koch I, Dach K, Heesemann J, Hoffmann R. Yersinia enterocolitica inactivates NK cells. Int J Med Microbiol. 2013; 303(8):433–42. Epub 2013/07/03. https://doi.org/10.1016/j.ijmm.2013.05.004 PMID: 23810728. 93. Garcia VE, Jullien D, Song M, Uyemura K, Shuai K, Morita CT, et al. IL-15 enhances the response of human gamma delta T cells to nonpeptide [correction of nonpetide] microbial antigens. J Immunol. 1998; 160(9):4322–9. Epub 1998/05/09. PMID: 9574535. 94. Orth K. Function of the Yersinia effector YopJ. Curr Opin Microbiol. 2002; 5(1):38–43. Epub 2002/02/ 09. https://doi.org/10.1016/s1369-5274(02)00283-7 PMID: 11834367. 95. 96. Li C, Wang D, Lv X, Jing R, Bi B, Chen X, et al. Yersinia pestis acetyltransferase-mediated dual acety- lation at the serine and lysine residues enhances the auto-ubiquitination of ubiquitin ligase MARCH8 in human cells. Cell Cycle. 2017; 16(7):649–59. Epub 2017/01/20. https://doi.org/10.1080/15384101. 2017.1281481 PMID: 28103160; PubMed Central PMCID: PMC5397269. Zhou H, Monack DM, Kayagaki N, Wertz I, Yin J, Wolf B, et al. Yersinia virulence factor YopJ acts as a deubiquitinase to inhibit NF-kappa B activation. J Exp Med. 2005; 202(10):1327–32. Epub 2005/11/23. https://doi.org/10.1084/jem.20051194 PMID: 16301742; PubMed Central PMCID: PMC2212976. 97. Keller B, Muhlenkamp M, Deuschle E, Siegfried A, Mossner S, Schade J, et al. Yersinia enterocolitica exploits different pathways to accomplish adhesion and toxin injection into host cells. Cell Microbiol. 2015; 17(8):1179–204. Epub 2015/02/14. https://doi.org/10.1111/cmi.12429 PMID: 25678064. 98. Morinobu A, Gadina M, Strober W, Visconti R, Fornace A, Montagna C, et al. STAT4 serine phosphor- ylation is critical for IL-12-induced IFN-gamma production but not for cell proliferation. Proc Natl Acad Sci U S A. 2002; 99(19):12281–6. Epub 2002/09/06. https://doi.org/10.1073/pnas.182618999 PMID: 12213961; PubMed Central PMCID: PMC129436. 99. Mavropoulos A, Sully G, Cope AP, Clark AR. Stabilization of IFN-gamma mRNA by MAPK p38 in IL- 12- and IL-18-stimulated human NK cells. Blood. 2005; 105(1):282–8. Epub 2004/09/04. https://doi. org/10.1182/blood-2004-07-2782 PMID: 15345584. 100. Zhang W, Liu HT. MAPK signal pathways in the regulation of cell proliferation in mammalian cells. Cell Res. 2002; 12(1):9–18. Epub 2002/04/11. https://doi.org/10.1038/sj.cr.7290105 PMID: 11942415. 101. Gehart H, Kumpf S, Ittner A, Ricci R. MAPK signalling in cellular metabolism: stress or wellness? EMBO Rep. 2010; 11(11):834–40. Epub 2010/10/12. https://doi.org/10.1038/embor.2010.160 PMID: 20930846; PubMed Central PMCID: PMC2966959. 102. Wilkinson MG, Millar JB. Control of the eukaryotic cell cycle by MAP kinase signaling pathways. FASEB J. 2000; 14(14):2147–57. Epub 2000/10/29. https://doi.org/10.1096/fj.00-0102rev PMID: 11053235. 103. Sugiura R, Satoh R, Ishiwata S, Umeda N, Kita A. Role of RNA-Binding Proteins in MAPK Signal Transduction Pathway. J Signal Transduct. 2011; 2011:109746. Epub 2011/07/22. https://doi.org/10. 1155/2011/109746 PMID: 21776382; PubMed Central PMCID: PMC3135068. 104. Kopper F, Bierwirth C, Schon M, Kunze M, Elvers I, Kranz D, et al. Damage-induced DNA replication stalling relies on MAPK-activated protein kinase 2 activity. Proc Natl Acad Sci U S A. 2013; 110 (42):16856–61. Epub 2013/10/02. https://doi.org/10.1073/pnas.1304355110 PMID: 24082115; PubMed Central PMCID: PMC3801042. 105. Wei JH, Seemann J. Remodeling of the Golgi structure by ERK signaling. Commun Integr Biol. 2009; 2(1):35–6. Epub 2009/08/26. https://doi.org/10.4161/cib.2.1.7421 PMID: 19704864; PubMed Central PMCID: PMC2649298. 106. Grenningloh R, Kang BY, Ho IC. Ets-1, a functional cofactor of T-bet, is essential for Th1 inflammatory responses. J Exp Med. 2005; 201(4):615–26. Epub 2005/02/25. https://doi.org/10.1084/jem. 20041330 PMID: 15728239; PubMed Central PMCID: PMC2213045. PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 28 / 29 PLOS PATHOGENS Yersinia pseudotuberculosis subversion of γδ T cell function 107. Koues OI, Collins PL, Cella M, Robinette ML, Porter SI, Pyfrom SC, et al. Distinct Gene Regulatory Pathways for Human Innate versus Adaptive Lymphoid Cells. Cell. 2016; 165(5):1134–46. Epub 2016/ 05/10. https://doi.org/10.1016/j.cell.2016.04.014 PMID: 27156452; PubMed Central PMCID: PMC4874868. 108. Dandekar AA, O’Malley K, Perlman S. Important roles for gamma interferon and NKG2D in gamma- delta T-cell-induced demyelination in T-cell receptor beta-deficient mice infected with a coronavirus. J Virol. 2005; 79(15):9388–96. Epub 2005/07/15. https://doi.org/10.1128/JVI.79.15.9388-9396.2005 PMID: 16014902; PubMed Central PMCID: PMC1181615. 109. Nitahara A, Shimura H, Ito A, Tomiyama K, Ito M, Kawai K. NKG2D ligation without T cell receptor engagement triggers both cytotoxicity and cytokine production in dendritic epidermal T cells. J Invest Dermatol. 2006; 126(5):1052–8. Epub 2006/02/18. https://doi.org/10.1038/sj.jid.5700112 PMID: 16484989. 110. Girardi M, Oppenheim DE, Steele CR, Lewis JM, Glusac E, Filler R, et al. Regulation of cutaneous malignancy by gammadelta T cells. Science. 2001; 294(5542):605–9. Epub 2001/09/22. https://doi. org/10.1126/science.1063916 PMID: 11567106 111. Osinska I, Popko K, Demkow U. Perforin: an important player in immune response. Cent Eur J Immu- nol. 2014; 39(1):109–15. Epub 2014/01/01. https://doi.org/10.5114/ceji.2014.42135 PMID: 26155110; PubMed Central PMCID: PMC4439970. 112. Gogoi D, Chiplunkar SV. Targeting gamma delta T cells for cancer immunotherapy: bench to bedside. Indian J Med Res. 2013; 138(5):755–61. Epub 2014/01/18. PMID: 24434328; PubMed Central PMCID: PMC3928706. 113. Autenrieth IB, Beer M, Bohn E, Kaufmann SH, Heesemann J. Immune responses to Yersinia entero- colitica in susceptible BALB/c and resistant C57BL/6 mice: an essential role for gamma interferon. Infect Immun. 1994; 62(6):2590–9. Epub 1994/06/01. https://doi.org/10.1128/iai.62.6.2590-2599.1994 PMID: 8188382; PubMed Central PMCID: PMC186549. 114. Bohn E, Heesemann J, Ehlers S, Autenrieth IB. Early gamma interferon mRNA expression is associ- ated with resistance of mice against Yersinia enterocolitica. Infect Immun. 1994; 62(7):3027–32. Epub 1994/07/01. https://doi.org/10.1128/iai.62.7.3027-3032.1994 PMID: 8005693; PubMed Central PMCID: PMC302917. 115. Hein J, Sing A, Di Genaro MS, Autenrieth IB. Interleukin-12 and interleukin-18 are indispensable for protective immunity against enteropathogenic Yersinia. Microb Pathog. 2001; 31(4):195–9. Epub 2001/09/20. https://doi.org/10.1006/mpat.2001.0458 PMID: 11562172. 116. Bohn E, Schmitt E, Bielfeldt C, Noll A, Schulte R, Autenrieth IB. Ambiguous role of interleukin-12 in Yersinia enterocolitica infection in susceptible and resistant mouse strains. Infect Immun. 1998; 66 (5):2213–20. Epub 1998/05/09. https://doi.org/10.1128/IAI.66.5.2213-2220.1998 PMID: 9573110; PubMed Central PMCID: PMC108184. 117. Wollert T, Pasche B, Rochon M, Deppenmeier S, van den Heuvel J, Gruber AD, et al. Extending the host range of Listeria monocytogenes by rational protein design. Cell. 2007; 129(5):891–902. Epub 2007/06/02. https://doi.org/10.1016/j.cell.2007.03.049 PMID: 17540170. 118. Kim D, Langmead B, Salzberg SL. HISAT: a fast spliced aligner with low memory requirements. Nat Methods. 2015; 12(4):357–60. Epub 2015/03/10. https://doi.org/10.1038/nmeth.3317 PMID: 25751142; PubMed Central PMCID: PMC4655817. 119. Liao Y, Smyth GK, Shi W. featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics. 2014; 30(7):923–30. Epub 2013/11/15. https://doi.org/10. 1093/bioinformatics/btt656 PMID: 24227677. 120. Stacklies W, Redestig H, Scholz M, Walther D, Selbig J. pcaMethods—a bioconductor package pro- viding PCA methods for incomplete data. Bioinformatics. 2007; 23(9):1164–7. Epub 2007/03/09. https://doi.org/10.1093/bioinformatics/btm069 PMID: 17344241. 121. Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 2014; 15(12):550. Epub 2014/12/18. https://doi.org/10.1186/s13059-014- 0550-8 PMID: 25516281; PubMed Central PMCID: PMC4302049. 122. Collins C, Shi C, Russell JQ, Fortner KA, Budd RC. Activation of gamma delta T cells by Borrelia burg- dorferi is indirect via a TLR- and caspase-dependent pathway. J Immunol. 2008; 181(4):2392–8. Epub 2008/08/08. https://doi.org/10.4049/jimmunol.181.4.2392 PMID: 18684928; PubMed Central PMCID: PMC2832482. PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 29 / 29 PLOS PATHOGENS
null
10.1186_s12874-023-01902-y.pdf
Data Availability We provide R modules as the basis for the dashboard development on GitHub (https://github.com/CTU-Basel/viewTrial). Qualitative data that supported the development of the risk assessment and study dashboard is provided in the supplementary material.
Data Availability We provide R modules as the basis for the dashboard development on GitHub ( https://github.com/CTU-Basel/viewTrial ). Qualitative data that supported the development of the risk assessment and study dashboard is provided in the supplementary material.
Klatte et al. BMC Medical Research Methodology (2023) 23:84 https://doi.org/10.1186/s12874-023-01902-y BMC Medical Research Methodology Development of a risk-tailored approach and dashboard for efficient management and monitoring of investigator-initiated trials Katharina Klatte1*, Suvitha Subramaniam1, Pascal Benkert1, Alexandra Schulz1, Klaus Ehrlich1, Astrid Rösler1, Mieke Deschodt2,3, Thomas Fabbro1, Christiane Pauli-Magnus1† and Matthias Briel1,4† Abstract Background Most randomized controlled trials (RCTs) in the academic setting have limited resources for clinical trial management and monitoring. Inefficient conduct of trials was identified as an important source of waste even in well-designed studies. Thoroughly identifying trial-specific risks to enable focussing of monitoring and management efforts on these critical areas during trial conduct may allow for the timely initiation of corrective action and to improve the efficiency of trial conduct. We developed a risk-tailored approach with an initial risk assessment of an individual trial that informs the compilation of monitoring and management procedures in a trial dashboard. Methods We performed a literature review to identify risk indicators and trial monitoring approaches followed by a contextual analysis involving local, national and international stakeholders. Based on this work we developed a risk-tailored management approach with integrated monitoring for RCTs and including a visualizing trial dashboard. We piloted the approach and refined it in an iterative process based on feedback from stakeholders and performed formal user testing with investigators and staff of two clinical trials. Results The developed risk assessment comprises four domains (patient safety and rights, overall trial management, intervention management, trial data). An accompanying manual provides rationales and detailed instructions for the risk assessment. We programmed two trial dashboards tailored to one medical and one surgical RCT to manage identified trial risks based on daily exports of accumulating trial data. We made the code for a generic dashboard available on GitHub that can be adapted to individual trials. Conclusions The presented trial management approach with integrated monitoring enables user-friendly, continuous checking of critical elements of trial conduct to support trial teams in the academic setting. Further work is needed in order to show effectiveness of the dashboard in terms of safe trial conduct and successful completion of clinical trials. Keywords Clinical trial, Trial management, Risk-tailored monitoring, Trial dashboard †Shared senior authorship. *Correspondence: Katharina Klatte Katiklatte@icloud.com 1Department of Clinical Research, University Hospital Basel and University of Basel, Spitalstrasse 12, Basel CH- 4031, Switzerland 2Department of Public Health & Primary Care, KU Leuven, Leuven, Belgium 3Competence Centre of Nursing, University Hospitals Leuven, Leuven, Belgium 4Department of Health Research Methods, Evidence, and Impact, McMaster University, Hamilton, ON, Canada © The Author(s) 2023. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data. RESEARCHOpen Access Page 2 of 11 Introduction Randomized controlled trials (RCTs) are the gold stan- dard for assessing the effects of medical interventions. However, they are typically resource intense and pose various organisational challenges [1–3]. Inefficient man- agement and monitoring of RCTs have been identified as an important source of waste [1–5]. Monitoring efforts are traditionally quite generic and extensive, [6–8] but problems such as slow participant recruitment, con- siderable losses to follow-up, or poor data quality are often recognized too late during trial conduct delaying necessary adjustments of processes or the protocol. In addition, resources for clinical trial monitoring and man- agement are usually scarce in the academic setting and sophisticated commercial solutions can be costly [9, 10]. Organisational challenges and critical factors jeopar- dizing trial integrity and quality may vary considerably across trials; therefore, a risk assessment conducted prior to trial initiation or at certain intervals during trial con- duct may yield different risk profiles for individual trials. Trial monitoring protects the safety and rights of partici- pants, ensures data are accurate, complete and verifiable, and that the trial follows the principles of good clinical practice [11, 12]. Currently recommended risk-based trial monitoring allows for an adaptation of the monitor- ing intensity according to an initial risk assessment of a trial and has been developed to reduce resource intense onsite visits with source data verification for non-high- risk trials [1–3, 13–15, 16, 17. However, this approach typically does not consider individual risk profiles of RCTs, but rather classifies trials by generic risk catego- ries [16]. To accommodate individual trial risks, a moni- toring strategy may include several components such as centralized monitoring (evaluation of accumulated trial data performed in a timely manner at a central location), onsite monitoring (performed at investigator sites with source data verification and review of protocol-specified processes), or remote monitoring (same tasks as onsite monitoring but performed away from investigator sites) [17, 18, 19]. Trial management should provide for smooth and reli- able trial procedures including participant recruitment, randomisation, intervention application, data collection, and data cleaning [20, 21]. Data cleaning and checking of recruitment and retention rates, for instance, need to be performed in a timely fashion, so that corrective mea- sures can be taken early on and detrimental effects on the trial can be avoided [22]. Trial monitoring is most effec- tive when performed on cleaned data, because incorrect processes may be missed due to poor data quality and monitoring efforts are wasted on individual data errors. Therefore, trial management and monitoring ideally are integrated tasks that make use of accumulating data dur- ing trial conduct, i.e. continuously keeping oversight of complex study processes and performing centralized data monitoring [23–25]. The objective of this project was to develop a risk- tailored approach that integrated trial management and monitoring in investigator-initiated RCTs. We closely collaborated with relevant stakeholders (trial coordina- tors, principal investigators, data managers, trial moni- tors, statisticians) to create a user-friendly dashboard that efficiently visualizes data on critical processes of individual trials. Methods Overview of research process In the first phase of this user-centred project, [26] we developed a concept of a risk-tailored trial monitor- ing and management approach with corresponding trial dashboard (Fig. 1). We anticipated users to be primarily trial managers, principal investigators, and trial moni- tors. The development involved relevant stakeholder groups and was based on the results of systematic litera- ture reviews on existing monitoring strategies, [17] and a contextual analysis to identify current practices and needs of anticipated users. The concept and dashboard were piloted and refined in an iterative process involving different end users and other stakeholder groups. In the second phase, we performed formal user testing of the developed risk assessment and dashboard. Experiences of investigators and trial staff of one medical and one surgi- cal investigator-initiated RCT were gathered using semi- structured interviews to further refine the concept and dashboard. Setting Before the introduction of the new concept, a risk assess- ment was routinely performed by the monitoring team to assess the extent of the monitoring needed for the trial according to the ADAMON criteria. This approach allowed the rough classification of trials into the catego- ries low, medium, or high risk [27]. The new risk assess- ment incorporates many more factors related to the study specific conduct including challenges in the study management. It is not meant to categorize trials and adjust the extent of monitoring based on the category. The trial teams included in our project were not involved in other pre-trial risk assessments. Both trial teams assessing the benefits of the risk assessment and dash- board tool had started participant recruitment and data collection before the implementation of the new tool and, thus, compared it to the situation without structured risk assessment and tool support.” Systematic literature review To identify and structure components for the initial risk assessment of individual trials, we systematically Klatte et al. BMC Medical Research Methodology (2023) 23:84 Page 3 of 11 Fig. 1 Overview of the two phases of the development and user-testing of the risk-tailored approach and trial dashboard searched for published risk assessment approaches and risk indicators used to support trial oversight and to identify centres in need for support. We considered dif- ferent components and qualitative evidence from process evaluations of tested monitoring strategies summarized in a previously conducted systematic review [17]. We further considered the guideline of the European Clini- cal Research Infrastructure Network (Ecrin) [16] and the risk assessment guideline developed by the Swiss Clinical Trial Organization [28], TransCelerate metrics [29, 30], Whitham metrics [31], and the trial specific metrics used by the Medical Research Council (MRC) Clinical Trials Unit (CTU) at University College London (UCL) Trial specific metrics [32]. Results from this literature review are summarized in Supplementary Table 1. groups provided an additional opportunity for feedback and exchange of information on the risk assessment and dashboard development as well as on the application strategy. In order to get input from a national group of stakeholders in Switzerland, we contacted the national platform of the Swiss Clinical Trial Organisation for trial monitoring. Finally, we gathered experiences from inter- national methodological research groups and UK-based CTUs using risk-based approaches or study dashboards to support trial conduct. The different activities with stakeholders at all levels are summarized in Supplemen- tary Table 2. We extracted information from protocols of meetings and interviews and summarized the output in Supplementary Table 3. Contextual analysis Stakeholder involvement We set up a local, multidisciplinary working group including end users and representatives of different stakeholder groups within the Department of Clinical Research (DKF) and associated research groups at the University Hospital Basel. At this local level, we involved members from the Data Science and Data Management Teams of the DKF experienced in central monitoring, R shiny applications, dashboard development, database structures and exports; we involved trial monitors with experience in on-site and remote monitoring, knowl- edge of study site structures and processes; study coor- dinators and investigators experienced in managing RCTs. Stakeholder meetings with all members of these Gathering contextual input from various end users and the above-mentioned stakeholders guided the devel- opment of the risk-tailored approach and helped to determine relevant domains and applications to be con- sidered in the initial risk assessment. We structured the identified stakeholder needs into content related fac- tors such as the inclusion of the follow-up visits into the risk assessment, and design related factors such as the suggested separation of severity and likelihood in the assessment or the colour code for the status of queries visualized in the dashboard (Supplementary Table 3). In terms of content of the risk assessment, it became clear, for instance, that the assessment covers a wide spectrum of risks applicable to a large variety of RCTs. The design Klatte et al. BMC Medical Research Methodology (2023) 23:84 of the risk assessment guide should support the intuitive assessment by different end user groups (monitors, study managers, principal investigators). The study dashboard should reflect the outcome of the risk assessment and the design of the dashboard should enable an efficient navi- gation within the routine study procedure by end-users. The findings of the contextual analysis are summarized in Supplementary Table 3. Development and piloting of the concept and dashboard Based on the systematically reviewed literature, our contextual analysis and stakeholder input, we drafted a generic risk-assessment template. We then created trial- specific dashboards for a medical and a surgical mul- ticentre trial that differed in their risk profile, but both comprised complex study procedures and data collec- tion. The risk-tailored approach continued to evolve as we gathered contextual information, detected gaps in the assessment procedure, and identified critical components of study management. We developed R code to extract data values from exported data tables of the trial database secuTrial and summarized, compared, and calculated rel- evant information to create pathways for the identified risks. The output of these operations was then visual- ized in the trial dashboard. The piloting and refinement was an iterative process incorporating repeated feedback from the end-users and the stakeholder representatives in the project group on dashboard content, structure, user-friendly interface, and visualization of critical study data. User testing The aim of the user testing was to identify challenges in the routine use of the dashboard experienced by different user groups. Each of the six users (i.e. 2 trial managers, 2 monitors, 2 principal investigators) received a detailed manual of the features and operation mode of the study dashboard. Table 1 Domains and their attributed risk elements Domain Participant Safety and Rights Overall Study Management Device/ Medication Management Study Data Risk Elements Informed consent AE/SAE reporting and documentation Inclusion/exclusion Recruitment Retention Study procedures and endpoint assessment (e.g. bio sampling, imaging quality) Participant schedule (e.g. timeframe of visits) AE/SAE management Administration Accountability/ storage Data quality – completeness, consistency, timeliness Documentation/ storage Abbreviations: AE, adverse event; SAE, serious adverse event Page 4 of 11 We interviewed users 6–12 weeks after using the study dashboard in daily trial routine. We followed a semi- structured interview guide, which allowed for expan- sion on topics that emerged during the interview. All interviews took approximately 30  min. The interviewer (KK) transcribed the recorded interviews and extracted suggestions for improvement. We then updated the trial dashboard based on the feedback of the users and pro- vided the adapted version for further use and evaluation. Results The final concept consisted of the following three steps: trial-specific risk assessment prior to study start, selec- tion and development of data-based pathways to address identified risks, and visualization of pathways output in a trial dashboard. Trial-specific risk assessment The four trial-specific risk assessment comprised domains (participant safety and rights, overall study management, device/medication management, study data), and each domain contained several risk elements (Table  1). To better assess if these elements are critical for a specific trial and which trial components are at par- ticular risk, we determined trial assets and corresponding risk scenarios. Trial assets are conditions essential for the successful and proper conduct of a trial, e.g. visits must be scheduled and take place in the required timeframe, Serious Adverse Events (SAEs) have to be reported on time and need to be closely followed over the whole study conduct. If a trial includes many follow-up visits over a long follow-up time and assessments have to take place in a very narrow time window, this asset would be con- sidered at risk (example shown in Table 2, Part A). Other assets, for example SAE reporting and oversight, are essential for all clinical trials and, thus, are considered as a risk that applies to all trials (marked in red, Example shown in Table  2, Part B). The identified risks are then analysed in terms of severity and likelihood. For exam- ple, if many follow-up visits need to be coordinated but the time window of the endpoint assessment is wide the severity is rated as less critical. The likelihood is highly influenced by the experience of the trial team and partici- pating centres with similar trials, training and experience of all involved staff members, and the resources available for the study. The complete list of assets, as well as the corresponding risk scenarios, is provided in the full risk assessment in Supplementary Table 4. We suggest that the risk assess- ment is done by an experienced trial manager (e.g. from a trials support unit) supported by a trial monitor, a clini- cal expert, and the principal investigator. The first risk assessment should be performed before the start of the trial based on the study protocol, Case Report Forms Klatte et al. BMC Medical Research Methodology (2023) 23:84 Table 2 Example of assets and risk scenarios for risk elements in the domain Overall Study Management (Part A) and Participant Safety and Rights (Part B). Assets that apply to all trials are marked in red A) Domain Overall Study Management Risk element Participant Schedule Asset Visits/Phone calls must be within the given Timeframe Risk scenario (A) Time point of visit is critical for the endpoint assess- ment of the study (B) Large number of visits are difficult to organize and coordinate between centres and patients B) Domain Participant Safety and Rights Risk element SAE/AE Asset SAE have to be re- ported and documented correctly in the required timeframe Risk scenario Complexity of CRF or missing SOPs for SAE Reporting leads to (A) Incorrect docu- mentation and (B) Delayed report- ing of SAEs Abbreviations: CRF, case report form; SOPs, standard operating procedures; SAE, serious adverse events (CRFs), the planned and actual budget of the study, expected recruitment rates for all participating centres, information on the trial intervention, and information about planned study staff (see Appendix for detailed Manual). Pathways to manage identified risks In order to continuously manage identified risks, we cre- ated pathways that eventually allowed for tailored visu- alization of accumulating trial data and implemented action at suitable time intervals (e.g., email reminders, staff overviews) in a study dashboard. The operations applied to the exported data tables via R code are depen- dent on the specific information needed to provide a clear oversight on identified risk elements. The code is structured into modules that contain the operations of all pathways visualized in one dashboard tab (e.g. SAE management). For example, the module SAE contains operations that count the number of SAEs, determine the number of patients with SAE and calculate the ratio SAEs per patient randomized. In addition, information like severity, causality and outcome are extracted from the SAE form data table and percentages of value options (e.g. SAE outcome: Continuing, Resolved without sequel, Resolved with sequel, others) are calculated and graphi- cally displayed (Fig.  2, Panel A and B). The developed study dashboards contain tabs that visualize the output of created pathways reflecting identified study-specific risks. These tabs are based on the R modules contain- ing the pathways as well as the code required for a clear Page 5 of 11 visual presentation (value boxes, graphs, lists). When pilot testing our risk assessment guide, it became appar- ent that some risks apply to almost all trials (marked in red in the full risk assessment Supplementary Table  4). The management of these risks is, thus, based on tabs classified as “generic” in the study dashboard, while other, more seldom and study-specific risks are considered in “optional” tabs (Table 3). The content of generic tabs can also be adapted depending on, for instance, the complex- ity or time point of outcome assessment in a trial. The generic dashboard template is freely available on GitHub (https://github.com/CTU-Basel/viewTrial). Visualization of data based pathways The output of the pathways is visualized in the corre- sponding tabs in the study dashboard. The arrangement of the tabs within the study dashboard can be determined by study teams; a division into study management related tabs and oversight/study progress tabs may provide a better overview for the different user groups (principal investigator, study manager, and trial monitor). The main tabs can also contain sub-tabs. For example, the num- ber of due visits is displayed under the visits tab in the sub-category “due visits”. In this context, the definitions of due, overdue, and missed visits are dependent on the specific timeframes of the study protocol. Total num- bers are provided as well as a list of the patient ID and a direct link to the corresponding eCRF in the database (Fig. 2, Panel A). Each tab or sub-tab can represent sev- eral pathway outputs displayed in form of value boxes, graphical presentations, or lists of relevant patients. For example, the SAE management tab provides an overview on SAE prevalence in boxes, and in additional panels the user can switch between the graphical representation of SAE severity, causality, and outcome. Additionally, a list of patients with SAE is provided below, displaying infor- mation on SAE status (e.g. ongoing/closed) and a short description of the event (Fig.  2, Panel B). The informa- tion is provided for the overall study, including all ran- domized patients as numbers and percentages in boxes, while graphs differentiating between centres are provided to better assess which centres are in need for support in a certain aspect of the study conduct. In addition, the dashboard allows filtering for specific centres and time ranges of interest or choosing particular study visits from drop down menus to provide users with more detailed information (see Supplementary Fig.  1 for an example). The output of the pathways visualized in the dash- board is based on a daily export of trial data and, thus, includes up-to-date information on randomised patients and entered data. The generic and some of the optional tabs are listed in Table 3. Examples of the tabs from the two study dashboards are provided in Supplementary Figs. 2–5. The generic dashboard is accessible via GitHub Klatte et al. BMC Medical Research Methodology (2023) 23:84 Page 6 of 11 Fig. 2 Dashboard screenshots of the Visits tab, sub-tab “Due visits” (Panel A), and the Safety management tab, sub-tab “Serious adverse events” (Panel B) and generic data is provided to test the different code modules behind each tab (examples provided in Supple- mentary Figs. 6 and 7). suggestions for further elements to be included in the dashboard. A detailed summary of the results from the user testing is provided in Supplementary Table 5. User testing The user testing of our study dashboards provided posi- tive feedback in terms of improved study oversight and facilitated conduct. Trial monitors and study staff agreed that the initial risk assessment was beneficial, because it increased the awareness of critical processes in the col- lection of outcome data, enabling corrective measures at an early time point, e.g. adaptation of database struc- tures. A clear benefit perceived by all user groups was the more frequent and improved communication with trial sites; sites were better prepared for remote or on-site monitoring visits, because many issues were recognized and solved in advance. In addition, users made several Discussion Using a systematic approach involving relevant stake- holder groups, we developed a concept of risk-tailored trial monitoring and management that focuses on the identification and control of trial specific risks during trial conduct. The continuous evaluation of most impor- tant risks provides important information about the study progress, e.g. in terms of recruitment, endpoint assessment, as well as in terms of data management and data quality, e.g. CRF completion, timeliness of follow- up visits. Completeness of essential data points as the basis for analysable patient data is continuously evalu- ated and trial monitors and study managers maintain an Klatte et al. BMC Medical Research Methodology (2023) 23:84 Table 3 Structure and content of dashboard tabs Domain Participant Safety and Rights Risk Elements Informed consent Example Tabs Informed consent AE/SAE reporting and documentation AE/SAE Inclusion/exclusion Safety Overall Study Management Recruitment Recruitment Patient Characteristics Retention Retention Study procedures and endpoint assessment Bio sampling (e.g. blood samples) Imaging quality Content of Tab In case of a re-consent this tab can provide an overview of patients patients who have previously not been able to give consent themselves Provides an overview of timeliness and completeness of AE/SAE entries In case of safety-relevant inclusion or exclusion criteria, a verification of relevant information available in the database can provide ad- ditional security (e.g. blood pressure has to be within a certain range – check for the entry of blood pressure in the database) Recruitment trajectories for expected and actual recruit- ment in total and per centre (Supplementary Fig. 2) Relevant patient character- istics are summarized and presented (e.g. gender, age, background of treatment) Patients who have ended the study resulting in missing outcome data, reasons for leaving the study, kind of data collected before study end (Primary outcome data avail- able) (Supplementary Fig. 3) Overview of samples taken and availability of sample results Automated and visual verifica- tion of imaging data quality, e.g., for MRI or CT Participant schedule: Follow-up visits Overview of follow-up visits AE/SAE management Safety manage- ment (SAEs, AEs) with a particular focus on visits where primary outcome data is collected. (Fig. 2, Panel A) The Safety tab provides an overview of SAEs and AEs that have been reported in the study and information on severity and outcome of SAEs/AEs (Fig. 2, Panel B) Page 7 of 11 Generic/Optional Optional Generic Optional Functionality/Purpose To ensure patient rights and support of re-consent process through site-specific reminders, list of patients that still need a re-consent. To ensure that all AE/SAE forms are complete and that the date of first entry is within the required reporting timeframe To provide the option for addi- tional checks for inclusion/ exclu- sion criteria besides the marked list of criteria in the eCRF To monitor the progress of partici- pant recruitment enabling early action in case of slow recruitment. Generic Generic Generic Optional Optional Optional Generic To inform the study team on the accuracy of inclusion/exclusion criteria and provide an overview of the sample population in terms of relevant characteristics To monitor the progress of partici- pant retention, consider reasons for ending study in recruitment. Time point of ending the study important for amount of data analysable. To support sample management in terms of localization and status of bio sample. Important for biomarker determination. To enable early adjustments in case of low quality imaging data and ensure that the imaging data is analysable. To assist in integrating follow-up visits on time into the daily clinical routine might be difficult for trial sites. Support through remind- ers for due visits can be initiated through the dashboard. To estimate potential safety issues (e.g. SAEs occurring more often in one study arm, number of SAEs in total, number of patients with SAE) Klatte et al. BMC Medical Research Methodology (2023) 23:84 Table 3 (continued) Domain Device/ Medication Management Risk Elements Administration Accountability/ storage Example Tabs Medication Study Data Data Quality Data quality – com- pleteness, consis- tency, timeliness Documentation/ storage Content of Tab Overview of medication con- sumption based on number of patients and their current position in the medication plan per protocol and com- parison with IMP stock at sites Completeness of forms (Primary end point, secondary endpoint, SAE/AE forms) Timeliness of data entry, Number of queries, status of queries (open, resolved) (Supplementary Figs. 4,5) Page 8 of 11 Functionality/Purpose To assist in the managing of IMP stock overview and enable reminders for restocking Generic/Optional Optional Generic To increase awareness of items missing in the database Trial sites may have different chal- lenges when integrating a trial in their daily clinical routine and therefore need support in differ- ent aspects of the study conduct. Completeness and timeliness of data entry as well as query man- agement constitute indicators for need of support. Query status helps the study monitor to decide which centre needs more assistance/ on-site visit. Abbreviations: AEs, adverse events; CT, computerized tomography ;IMP, investigational medicinal product; MRI, magnetic resonance imaging; SAEs, serious adverse events overview of visit timeframes, SAE reporting, and query management. Strengths and limitations Strengths of our study are the systematic and structured process of development of the risk assessment and the trial dashboard, which included the involvement of all local stakeholder groups and the performance of a com- prehensive contextual analysis. In addition, the devel- opment was based on prior evidence gathered through systematic literature searches and exchange with interna- tional stakeholder groups. Directly involving end users in developing and evaluating the usability of our tool may facilitate the implementation process, promote wider adoption, maintain involvement, and increase user satis- faction with the concept as well as the tool [33]. Providing an R code repository for other study teams that can be adapted and applied to differently structured databases, constitutes a software-independent, affordable approach for the limited budget of investigator-initiated trials. Our study has the following limitations: First, we per- formed user testing in two ongoing RCTs only, and, thus, the spectrum of feedback may have been limited and may compromise the extrapolation of mentioned ben- efits and disadvantages to other trials. Both RCTs had already started participant recruitment when the dash- board was implemented. This allowed for a qualitative comparison of management and monitoring processes without and with the dashboard tool in place. However, it will be crucial to subsequently evaluate the impact and value of the study dashboard during the entire course of a clinical trial. Since both RCTs are still ongoing, we could not evaluate the impact of the tool on participant safety and overall trial success, including the percentage of analysable data, at the end of a trial. Lastly, we have not yet evaluated any cost-effectiveness of our developed approach, e.g. assessing whether the dashboard has the potential to reduce monitoring and management hours needed to ensure a safe and successful trial conduct. While some users felt that our dashboard would only be worthwhile for multicentre trials, others found that the costs of providing a study dashboard will always depend on the needs and preferences of the study team and the complexity of the study. Comparison with similar studies and frameworks Following the recommendations of the Clinical Trials Transformation Initiative (CTTI), effective and efficient monitoring and management needs to first determine what matters for a specific trial and focus on areas of highest risk for generating errors that matter [34, 35]. With our risk assessment guide and the study dashboard we address the need for this focus and provide a tool that supports the continuous oversight of the quality of the trial conduct. Dashboards that visualize time-dependent parameters have recently met a growing acceptance in medical and administrative health care settings [36–43]. Dashboards have been introduced to support various aspects of clinical trials, including web applications for eligibility screening and overview of the enrolment progress [41], web-based support of recruitment management and Klatte et al. BMC Medical Research Methodology (2023) 23:84 Page 9 of 11 communication; [42] graphical summaries and diagrams of the progress of patient accrual and form completion [43], feedback on data completeness by using a traffic light system [44], and automated reports of data compli- ance, protocol adherence and safety [45]. These available dashboards typically focus on specific elements of trial conduct and communication with trial sites; however, our dashboard provides a comprehensive overview of all elements of a trial identified as critical. In addition, tables and graphical representations are often limited to certain time intervals [41]. The daily export of trial data providing up-to-date trial information is part of the core idea of our approach as it enables immediate actions and improves communication with site staff. Various methods for assessing the risk of non-conform trial conduct at trial sites including central statistical monitoring have been introduced in the academic set- ting with increasing prevalence [46]. Most methods use statistical testing of all or a subset of trial data items to compare sites and identify atypical trial centres. While many methods focus on the detection of data errors and fraud, [47] triggered monitoring is frequently used to direct on-site monitoring to atypical trial sites [46]. In our approach components of central data evaluations are used to assess whether actions are required constituting some sort of triggered intervention. However, the data evaluation is not based on statistical testing, it is rather an assessment of trial progress (recruitment, retention), management challenges, and conform data collection progress. It is also not intended to categorize trials and predetermine the extent of on-site monitoring [48]. Our concept focuses on directing attention to the most criti- cal areas of a trial and should help to minimize and tailor on-site monitoring. Several commercial solutions supporting the over- all trial conduct in various aspects are readily available [9, 49–53], but for investigator-initiated trials with tight budgets such software packages typically remain unaf- fordable. We wanted to provide a comprehensive and affordable option for investigator-initiated trials that can be adapted to individual needs and preferences and fur- ther developed by the research community. Therefore, we transparently present all details of the structured risk assessment and manual as well as the generic code for our dashboard in publicly accessible repositories via GitHub. We invite users to report difficulties or sugges- tions for improvement for consideration in future modifi- cations of the generic dashboard via GitHub. Implications Besides the emphasis on the feasibility and design of clin- ical trials, measures to increase the efficiency of clinical trial conduct are needed [54]. Current challenges include premature discontinuation of a significant proportion of clinical trials, and inflated costs mainly due to delayed recruitment and organisational issues [54]. We propose a comprehensive approach integrating management and monitoring of a clinical trial into one risk management tool supporting the conduct of investigator-initiated trials. Overseeing the progress of a trial in each centre based on up-to-date information, provides the opportunity for trial monitors to prioritize centres for on-site visits or remote interactions, tailor their action to the specific issues of a centre, and guide decisions on where resources and training is needed the most. In addition, providing automated reminders for upcoming visits or sampling, overview of investigational medicinal product supply, overview of patients who need a re-consent, overview of ongoing SAEs, etc. could increase the efficiency of the trial management processes. The tool further provides the opportunity to improve the overall communication between the study team and trial sites and may increase motivation through the involvement of sites in the trial progress and the option to compliment active partici- pation in the trial. The dashboard tool is intended to address site-level monitoring, trial-wide monitoring, and finding per-patient issues. Feedback from the user testing also revealed a positive perception of study managers and investigators to improved data quality visible in the dash- board: “If incomplete is empty, I am at ease. The impact of this tool is largely dependent on the suc- cessful implementation into clinical trial practice. The perception of benefits and opportunities by stakeholders and end-users have been collected while the effectiveness of the tool in terms of analysable data collected, timeline of recruitment, conformity of SAE/Adverse Event (AE) reporting and documentation, support of the overall study management still have to be evaluated. The next step is now to implement the risk assessment as a routine step in the joint planning of clinical trials with the respective study teams. The timely generation of a dashboard on the basis of the generic template and further study-specific risks has to be organized. Strate- gies to further evaluate this implementation process as well as the effectiveness of this new approach in studies of different design and structure have to be developed. As an implementation outcome, the amount of studies tak- ing advantage of the study dashboard in relation to the studies for which a dashboard was recommended could be assessed along with the frequency of risk assessments performed per trial. The effectiveness of the concept of risk assessment and dashboard tool will be evaluated based on structured feedback from study teams on their experience and quantitative measures of the trial, e.g. proportion of analysable patients/data at the end of the trial. These evaluations will provide more information on the feasibility of study-specific dashboards supporting Klatte et al. BMC Medical Research Methodology (2023) 23:84 trial monitoring and management in the heterogeneous field of clinical trials. Conclusion In summary, the presented risk-assessment guide and dashboard tool provide a systematically developed and user-tested instrument for the risk-tailored support of trial monitoring and trial management. Feedback from the user testing of the instrument revealed many benefits for the involved stakeholder groups. However, the effec- tiveness of the dashboard in terms of a safe trial conduct and overall support for a successful completion of clinical trials needs to be further evaluated. Supplementary Information The online version contains supplementary material available at https://doi. org/10.1186/s12874-023-01902-y. Supplementary Material 1 Supplementary Material 2 Acknowledgements We would like to thank all stakeholders that provided feedback in the development process of our risk assessment and the study dashboard. Author contributions K.K., M.B. and C.P.M. wrote the main manuscript text and K.K. prepared the figures. P.B., A.S., K.E., A.R. and T.F. were involved in the development of the concept, the risk assessment and dashboard content. S.S. and K.K. developed the study-specific dashboards and the generic dashboard. K.K. conducted the interviews for the user testing. All authors reviewed the manuscript. Funding Open access funding provided by University of Basel Data Availability We provide R modules as the basis for the dashboard development on GitHub (https://github.com/CTU-Basel/viewTrial). Qualitative data that supported the development of the risk assessment and study dashboard is provided in the supplementary material. Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing Interest The authors declare no competing interests. Received: 10 September 2022 / Accepted: 23 March 2023 References 1. Yusuf S. Randomized clinical trials: slow death by a thousand unnecessary policies? CMAJ 2004;171(8):889 – 92; discussion 92 – 3. doi: https://doi. org/10.1503/cmaj.1040884 [published Online First: 2004/10/13] Page 10 of 11 3. 4. 2. Eisenstein EL, Lemons PW 2nd, Tardiff BE, et al. Reducing the costs of phase III cardiovascular clinical trials. Am Heart J. 2005;149(3):482–8. https://doi. org/10.1016/j.ahj.2004.04.049. [published Online First: 2005/05/03]. Eisenstein EL, Collins R, Cracknell BS, et al. Sensible approaches for reducing clinical trial costs. Clin Trials. 2008;5(1):75–84. https://doi. org/10.1177/1740774507087551. [published Online First: 2008/02/20]. Kasenda B, von Elm E, You J, et al. Prevalence, characteristics, and publication of discontinued randomized trials. JAMA. 2014;311(10):1045–51. https://doi. org/10.1001/jama.2014.1361. [published Online First: 2014/03/13]. Al-Shahi Salman R, Beller E, Kagan J, et al. Increasing value and reduc- ing waste in biomedical research regulation and management. Lancet. 2014;383(9912):176–85. https://doi.org/10.1016/S0140-6736(13)62297-7. [published Online First: 2014/01/15]. Tudur Smith C, Stocken DD, Dunn J, et al. The value of source data verifica- tion in a cancer clinical trial. PLoS ONE. 2012;7(12):e51623. https://doi. org/10.1371/journal.pone.0051623. [published Online First: 2012/12/20]. 7. Hearn J, Sullivan R. The impact of the ‘clinical trials’ directive on the cost 5. 6. and conduct of non-commercial cancer trials in the UK. Eur J Cancer. 2007;43(1):8–13. https://doi.org/10.1016/j.ejca.2006.09.016. [published Online First: 2006/11/23]. 8. Duley L, Antman K, Arena J, et al. Specific barriers to the con- duct of randomized trials. Clin Trials. 2008;5(1):40–8. https://doi. org/10.1177/1740774507087704. [published Online First: 2008/02/20]. CluePoints.Risk-basedmonitoring. http://cluepoints.com/risk-based-monitor- ing. 2016 [accessed 31st July 2022. 9. 10. Medidata.Strategic monitoring https://www.medidata.com/wp-content/ 11. uploads/2018/12/Strategic-Monitoring_2018Medidata_Fact-Sheet.pdf2018 [accessed 31st July 2022. International Conference on Harmonisation of technical requirements forpharmaceuticals for human use (ICH). Guideline for good clinical practice E6(R2) Available from: https://www.ema.europa.eu/en/documents/scientific- guideline/ich-e-6-r2-guideline-good-clinical-practice-step-5_en.pdf2018 [accessed 21st July 2022. 12. Medicines. Healthcare products Regulatory Agency. Good clinicalpractice guide. The Stationery Office; 2012. 13. Califf RM. Clinical trials bureaucracy: unintended consequences of well-inten- tioned policy. Clin Trials. 2006;3(6):496–502. doi: 10.1177/1740774506073173 [published Online First: 2006/12/16]. 14. Sertkaya A, Wong HH, Jessup A, et al. Key cost drivers of pharmaceutical clinical trials in the United States. Clin Trials. 2016;13(2):117–26. https://doi. org/10.1177/1740774515625964. [published Online First: 2016/02/26]. 15. Funning SGA, Eriksson K, Kettis-Lindblad. A Quality assurance within the scope of good clinical practice (GCP) - what is the cost of GCP-related activi- ties? A survey within the Swedish Association of the Pharmaceutical industry (LIF)’s members. Qual Assur J. 2009;12:3–7. 16. Ecrin. Risk-Based Monitoring Toolbox Available from: https://ecrin.org/tools/ risk-based-monitoring-toolbox2021 [ 17. Klatte K, Pauli-Magnus C, Love SB, et al. Monitoring strategies for clinical inter- vention studies. Cochrane Database Syst Rev. 2021;12:MR000051. https://doi. org/10.1002/14651858.MR000051.pub2. [published Online First: 2021/12/09]. 18. Baigent C, Harrell FE, Buyse M, et al. Ensuring trial validity by data quality assurance and diversification of monitoring methods. Clin Trials. 2008;5(1):49– 55. https://doi.org/10.1177/1740774507087554. [published Online First: 2008/02/20]. 19. Love SB, Armstrong E, Bayliss C, et al. Monitoring advances including consent: learning from COVID-19 trials and other trials running in UKCRC registered clinical trials units during the pandemic. Trials. 2021;22(1):279. https://doi. org/10.1186/s13063-021-05225-5. [published Online First: 2021/04/16]. 20. Farrell B, Kenyon S, Shakur H. Managing clinical trials. Trials. 2010;11:78. https://doi.org/10.1186/1745-6215-11-78. [published Online First: 2010/07/16]. 21. Medical Research Council. Clinical trials for tomorrow: an MRC Review of Randomised Control trials. Medical Research Council; 2003. 22. Love SB, Yorke-Edwards V, Diaz-Montana C, et al. Making a distinction between data cleaning and central monitoring in clinical trials. Clin Trials. 2021;18(3):386–88. https://doi.org/10.1177/1740774520976617. [published Online First: 2021/03/04]. 23. Swiss Clinical Trial Organisation (SCTO). F dated. Fact sheet: Central Data Monitoring in Clinical Trials – V 1.0. available from www.scto.ch/monitoring SCTO; [accessed May 3rd 2022. 24. Venet D, Doffagne E, Burzykowski T, et al. A statistical approach to central monitoring of data quality in clinical trials. Clin Trials. 2012;9(6):705–13. Klatte et al. BMC Medical Research Methodology (2023) 23:84 https://doi.org/10.1177/1740774512447898. [published Online First: 2012/06/12]. 25. Agrafiotis DK, Lobanov VS, Farnum MA, et al. Risk-based monitoring of clinical trials: an Integrative Approach. Clin Ther. 2018;40(7):1204–12. https://doi. org/10.1016/j.clinthera.2018.04.020. [published Online First: 2018/08/14]. 26. Walden A, Garvin L, Smerek M, et al. User-centered design principles in the development of clinical research tools. Clin Trials. 2020;17(6):703–11. https:// doi.org/10.1177/1740774520946314. [published Online First: 2020/08/21]. 27. von Niederhausern B, Orleth A, Schadelin S, et al. Generating evidence on a risk-based monitoring approach in the academic setting - lessons learned. BMC Med Res Methodol. 2017;17(1):26. https://doi.org/10.1186/s12874-017- 0308-6. [published Online First: 2017/02/15]. 28. Swiss Clinical Trial Organisation SCTO. Risk assessment Form for Clinical Research Projects available from: https://www.sctoplatforms.ch/en/tools/risk- assessment-form-for-clinical-research-projects-30.html2018 [accessed May 3rd 2022. 29. Wilson B, Provencher T, Gough J, et al. Defining a central monitoring capability:sharing the experience of TransCelerate BioPharma’s Approach, Part 1. Therapeutic Innov Regul Sci. 2014;48(5):529–35. https://doi. org/10.1177/2168479014546335. 30. Gough J, Wilson B, Zerola M, et al. Defining a central monitoring capability:sharing the experience of TransCelerate BioPharma’s Approach, Part 2. Therapeutic Innov Regul Sci. 2016;50(1):8–14. https://doi. org/10.1177/2168479015618696. 31. Whitham D, Turzanski J, Bradshaw L, et al. Development of a standardised set of metrics for monitoring site performance in multicentre randomised trials: a Delphi study. Trials. 2018;19(1):557. https://doi.org/10.1186/s13063-018-2940- 9. [published Online First: 2018/10/18]. 32. Diaz-Montana C, Cragg WJ, Choudhury R, et al. Implementing monitoring triggers and matching of triggered and control sites in the TEMPER study: a description and evaluation of a triggered monitoring management system. Trials. 2019;20(1):227. https://doi.org/10.1186/s13063-019-3301-z. [published Online First: 2019/04/19]. 33. Goodman MS, Sanders Thompson VL. The science of stakeholder engage- ment in research: classification, implementation, and evaluation. Transl Behav Med. 2017;7(3):486–91. https://doi.org/10.1007/s13142-017-0495-z. [published Online First: 2017/04/12]. 34. EFFECTIVE AND EFFICIENT CTTIRECOMMENDATIONS, MONITORING AS A COMPONENT OF QUALITY ASSURANCE IN, THE CONDUCT OF CLINICAL TRI- ALS Available at. : https://ctti-clinicaltrials.org/wp-content/uploads/2021/06/ CTTI_Monitoring_Recs.pdf2021 [accessed 21st July 2022. 35. CTTI. QUALITY BY DESIGN PROJECT - CRITICAL TO QUALITY (CTQ) FACTORS PRINCIPLES DOCUMENT Available at. : https://ctti-clinicaltrials.org/wp-con- tent/uploads/2021/07/CTTI_QbD_Workshop_Principles_Document.pdf 2021 [accessed 21st July 2022. 36. Simpao AF, Ahumada LM, Desai BR, et al. Optimization of drug-drug interac- tion alert rules in a pediatric hospital’s electronic health record system using a visual analytics dashboard. J Am Med Inform Assoc. 2015;22(2):361–9. https://doi.org/10.1136/amiajnl-2013-002538. [published Online First: 2014/10/17]. 37. Hartzler AL, Izard JP, Dalkin BL, et al. Design and feasibility of integrating per- sonalized PRO dashboards into prostate cancer care. J Am Med Inform Assoc. 2016;23(1):38–47. https://doi.org/10.1093/jamia/ocv101. [published Online First: 2015/08/12]. 38. Dowding D, Randell R, Gardner P, et al. Dashboards for improving patient care: review of the literature. Int J Med Inform. 2015;84(2):87–100. https://doi. org/10.1016/j.ijmedinf.2014.10.001. [published Online First: 2014/12/03]. Page 11 of 11 39. Waitman LR, Phillips IE, McCoy AB, et al. Adopting real-time surveillance dashboards as a component of an enterprisewide medication safety strategy. Jt Comm J Qual Patient Saf. 2011;37(7):326–32. https://doi.org/10.1016/ s1553-7250(11)37041-9. [published Online First: 2011/08/09]. 40. Nelson SD, Del Fiol G, Hanseler H, et al. Software Prototyping: a Case Report of Refining user requirements for a Health Information Exchange Dash- board. Appl Clin Inform. 2016;7(1):22–32. https://doi.org/10.4338/ACI-2015- 07-CR-0091. [published Online First: 2016/04/16]. 41. Arab L, Hahn H, Henry J, et al. Using the web for recruitment, screen, tracking, data management, and quality control in a dietary assessment clinical valida- tion trial. Contemp Clin Trials. 2010;31(2):138–46. [published Online First: 2009/11/21]. 42. Mattingly WA, Kelley RR, Wiemken TL, et al. Real-time enrollment dashboard for Multisite clinical trials. Contemp Clin Trials Commun. 2015;1:17–21. https://doi.org/10.1016/j.conctc.2015.09.001. [published Online First: 2016/02/16]. 43. Toddenroth D, Sivagnanasundaram J, Prokosch HU, et al. Concept and imple- mentation of a study dashboard module for a continuous monitoring of trial recruitment and documentation. J Biomed Inform. 2016;64:222–31. https:// doi.org/10.1016/j.jbi.2016.10.010. [published Online First: 2016/10/23]. Improving data entry and study compliance efficiently using immediate audit and feedback tools 5th International Clinical Trials Methodology Con- ference (ICTMC); 2019. Trials. 44. 45. Automation of clinical trial statistical monitoring. 5th International Clinical Trials Methodology Conference (ICTMC); 2019. Trials. 46. Cragg WJ, Hurley C, Yorke-Edwards V, et al. Dynamic methods for ongoing assessment of site-level risk in risk-based monitoring of clinical trials: a scoping review. Clin Trials. 2021;18(2):245–59. https://doi. org/10.1177/1740774520976561. [published Online First: 2021/02/23]. 47. Kirkwood AA, Cox T, Hackshaw A. Application of methods for central statisti- cal monitoring in clinical trials. Clin Trials. 2013;10(5):783–806. https://doi. org/10.1177/1740774513494504. [published Online First: 2013/10/17]. 48. Hurley C, Shiely F, Power J, et al. Risk based monitoring (RBM) tools for clinical trials: a systematic review. Contemp Clin Trials. 2016;51:15–27. https://doi. org/10.1016/j.cct.2016.09.003. [published Online First: 2016/10/30]. 49. PerkinElmer. Risk-based monitoring.: Available at: http://www.perkinelmer. 50. 51. com/product/risk-based-monitoring-rbm.; [accessed August 18th 2022. IQVIA. The next wave of central monitoring. Available at: https://www. iqvia.com/library/white-papers/the-next-wave-of-centralized-monitoring. [accessed August 18th 2022. Intelligence TR. Ttirials: a complete risk-based monitoring solution Available at: https://tritrials.com [accessed August 18th 2022. 52. Cyntegrity. Risk-based monitoring solutions Available at: https://cyntegrity. com [accessed August 18th 2022. 53. Plc ICON. Patient centric monitoring: preventing and learning from mistakes Available at: http://www.iconplc.com/icon-views/blog/2015/05/21/patient- centric-monitorin [accessed August 18th 2022. 54. Swiss Academy of Medical Sciences. White Paper: Clinical Research. Swiss Academies Communications 2021; 16 (4) Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Klatte et al. BMC Medical Research Methodology (2023) 23:84
null
10.1093_nar_gkad331.pdf
DA T A A V AILABILITY The online r esour ce is available without restriction at https: //www.flyrnai.org/tools/pangea/
null
Published online 1 May 2023 Nucleic Acids Research, 2023, Vol. 51, Web Server issue W419–W426 https://doi.org/10.1093/nar/gkad331 PANGEA: a new gene set enrichment tool for Drosophila and common research organisms Yanhui Hu 1 , 2 ,* , Aram Comjean 1 , 2 , Helen Attrill 3 , Giulia Antonazzo 3 , Jim Thurmond 4 , Weihang Chen 1 , 2 , Fangge Li 2 , Tiffany Chao 2 , Stephanie E. Mohr 1 , 2 , Nicholas H. Brown 3 and Norbert Perrimon 1 , 2 , 5 ,* 1 Department of Genetics, Blavatnik Institute, Harvard Medical School, Harvard University, Boston, MA 02115, USA, 2 Drosophila RNAi Screening Center, Harvard Medical School, 77 Avenue Louis Pasteur, Boston, MA 02115, USA, 3 Department of Physiology, Development and Neuroscience, University of Cambridge, Downing Street, Cambridge CB2 3DY, UK, 4 Department of Biology, Indiana University, Bloomington, IN 47405, USA and 5 Ho w ard Hughes Medical Institute, Boston, MA 02138, USA Received February 21, 2023; Revised March 28, 2023; Editorial Decision April 17, 2023; Accepted April 29, 2023 ABSTRACT GRAPHICAL ABSTRACT Gene set enrichment analysis (GSEA) plays an im- portant role in large-scale data analysis, helping sci- entists discover the underlying biological patterns o ver -represented in a gene list resulting fr om, f or e xample, an ‘omics’ stud y. Gene Ontology (GO) an- notation is the most frequently used classification mechanism for gene set definition. Here we present a new GSEA tool, PANGEA (PAthwa y, Netw ork and Gene-set Enrichment Analysis; https://www.flyrnai. org/ tools/ pangea/ ), developed to allow a more flexi- ble and configurable approach to data analysis us- ing a variety of classification sets. PANGEA allows GO analysis to be performed on different sets of GO annotations, for e xample e xcluding high-throughput studies. Beyond GO, gene sets for pathway anno- tation and protein complex data fr om v arious re- sources as well as expression and disease anno- tation from the Alliance of Genome Resources (Al- liance). In addition, visualizations of results are en- hanced by providing an option to view network of gene set to gene relationships. The tool also al- lows comparison of multiple input gene lists and ac- companying visualisation tools for quick and easy comparison. This new tool will facilitate GSEA for Drosophila and other major model organisms based on high-quality annotated inf ormation av ailable f or these species. INTRODUCTION Modern genetics and genomics owe much to work done us- ing common model organisms. These models continue to make a significant contribution to the understanding of de- velopment, metabolism, neuroscience, behaviour and dis- ease. With the onset of the ‘big data’ era has come a need for analysis platforms that deconvolute complex data from multispecies studies. Model organism databases (MODs) are knowledgebases dedicated to the cur ation, stor age and integration of species-specific data for their r esear ch com- munity. The past decade has seen a number of efforts aimed at pulling together model organism and human data to facilitate a more inter disciplinary approach; e xamples in- clude MARRVEL (Model organism Aggregated Resources for Rare Variant ExpLoration) ( 1 ), Gene2Function ( 2 ) and * To whom correspondence should be addressed. Tel: +1 6174327672; Fax: +1 6174327688; Email: perrimon@receptor.med.harvard.edu Correspondence may also be addressed to Yanhui Hu. Tel: +1 6174327672; Fax: +1 6174327688; Email: claire hu@genetics.med.harvard.edu C (cid:2) The Author(s) 2023. Published by Oxford University Press on behalf of Nucleic Acids Research. This is an Open Access article distributed under the terms of the Creati v e Commons Attribution License (http: // creati v ecommons.org / licenses / by / 4.0 / ), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. W420 Nucleic Acids Research, 2023, Vol. 51, Web Server issue the Monarch Initiati v e ( 3 ). Furthermore, the Alliance of Genome Resources (Alliance) ( 4 ), a consortium of se v en model organism databases and the Gene Ontology (GO) Consortium (GOC), formed recently with the objecti v e of building an umbrella resource from which users can navi- gate combined data within a single integrated knowledge- base. To help support such a r esour ce, the MODs ar e work- ing to reduce the di v ergence in the way the primary data is curated, stor ed and pr esented to facilita te compara ti v e and translational r esear ch. Although these integrated r esour ces allow sear ch and comparison across certain data classes, large-scale data analysis remains in the domain of stand-alone bioinfor- matic tools such as DAVID ( 5 ), TermMapper ( 6 ), GOrilla ( 7 ), PANTHER Gene List Analysis ( 8 ), WebGestalt ( 9 ) and g:Profiler ( 10 ), with a focus on processing gene list to extract a statistical measure of shared biological features, usually termed gene set enrichment analysis (GSEA). The most frequently used gene set classification in GSEA is GO annotation, which is based on the most widely-used on- tology (a hierarchical controlled vocabulary) in biological r esear ch for the wild-type molecular function(s), biologi- cal process(es) and cellular component(s) associated with a gi v en gene product ( 11 , 12 ). Many GSEA tools also incor- pora te classifica tions from other sources such as Reactome ( 13 ) and KEGG ( 14 ) pathways (e.g. DAVID, WebGestalt and g:Profiler) and, for human studies, there may be addi- tional data sources such as the Human Phenotype Ontol- ogy (HPO) ( 15 ) and Online Mendelian Inheritance in Man (OMIM) ( 16 ) (e.g. WebGestalt). The addition of gene sets beyond the GO allows users to extract more classification information, as well as seek trends and overlap in the en- riched sets. DAVID and g:Profiler are two of the few re- sources that make it possible for users to compare differ- ent sets on the same display; howe v er, the ability to interact with the results post-processing at these r esour ces is rather limited. Despite the abundance of tools, we found that they do not fully meet community needs, primarily because they were overly focused on human gene data and were not us- ing the most up to date data on other species. For exam- ple, Reactome pa thway annota tion is based on computa- tional predictions deri v ed from manually curated human pathways. Ther e ar e a few or ganism-tar geted analysis tools; the prokaryote-centred GSEA FUNAGE-Pro ( 17 ) is one example in which the underlying knowledgebase was as- sembled to cater to the needs of a specific r esear ch commu- nity. A number of useful gene classification r esour ces (e.g. pathways , complexes , gene groups) have been developed by the Drosophila RNAi Screening Center (DRSC) and Fly- Base, the Drosophila knowledgebase ( 18–22 ). Furthermore, in FlyBase and indeed across the MODs, ther e ar e se v eral types of curated data in common, including disease mod- els, phenotypes and gene expression, which could be used for GSEA. In contrast to the annotation of gene function data with GO, which is done in a consistent manner across multiple organisms, other data types ar e r epr esented within the MODs in di v erse ways, reflecting some of the technical differences in the genetics of these organisms. The Alliance was founded to integra te da ta across many MODs ( 4 ) and now provides a source of harmonised data that can also be used for GSEA. To take full advantage of the r esear ch in di v erse model organisms, we describe our creation of a new tool that we name PANGEA. Although our primary focus was on Drosophila genes, we de v eloped PANGEA to also include r at, mouse, zebr afish, nematode worm data, as well as harmonised human data to facilitate translational r esear ch. PANGEA not only incorporates additional gene set classifications from Alliance and MODs, but also have implemented the features that enhance the presentation of enrichment results by allowing the user to select sets and compare them visually to facilitate interpretation as well as making it easy to do parallel GSEA for multiple gene lists. This fle xibility enab les users to adapt the tool to their needs and allow ‘fortuitous’ discovery by widening the pool of knowledge for the purpose of analysis. MATERIALS AND METHODS Building the knowledgebase for PANGEA The gene set classification is a way to group genes based on commonality such as the same biological pathway. We have collected > 300 000 gene sets from various public r esour ces (Tables 1 , 2 ) for fruit fly D. melanogaster , the nematode worm C. elegans , the zebrafish D. rerio , the mouse M. mus- culus , the rat R. norvegicus and human H. sapiens . For an- notations based on a controlled vocabulary arranged in hi- erar chical structur e, such as gene group and phenotype an- notations from FlyBase, gene-to-gene set relationships were assembled after the hierar chical structur e was flattened. An exception was made for GO annotations, which were as- sembled in two ways, with and without being flattened, al- lowing users to choose which output is used in the anal- ysis. GO annotations include evidence codes that indicate the type of evidence supporting the annotation. For exam- ple, ‘IDA’ means that an annotation was supported by a dir ect assay, wher eas ‘ISS’ means that the annotation was inferred from sequence or structural similarity. Using such evidence codes, GO gene sets were built with additional configurations: (i) subsets based on experimental evidence codes, i.e. excluding annotations only based on phyloge- netic, sequence or structural similarity and other computa- tional analyses (IEA, IBA, IBD, IKR, IRD, ISS, ISO, ISA, ISM, IGC, RCA); (ii) subsets excluding annotations only supported by high-throughput (HTP) evidence codes (HTP, HDA, HMP, HGI and HEP); (iii) subsets of GO generic terms (GO slim) provided by the GOC ( http://geneontology. or g/docs/do wnload-ontology/#subsets ); (iv) subsets of very high-le v el GO term classifications used by FlyBase and the Alliance originally generated to support GO summary rib- bon displays. For Drosophila phenotype annotations from FlyBase, we assembled the gene-to-phenotype association using the ‘genotype phenotype data’ file available in the FlyBase Downloads page, in which phenotypes are associ- ated with individual genotypes and controlled vocabulary identifiers are indicated. This allowed us to extract only those genotypes where we could be certain that the phe- notype was associated with the perturbation of a single Drosophila melanogaster gene (i.e. single classical or in- sertional alleles). Because different resources use different gene or protein identifiers, we used an in-house mapping Nucleic Acids Research, 2023, Vol. 51, Web Server issue W421 pr ogram to synchr onise IDs to NCBI Entrez gene IDs, offi- cial gene symbols and gene identifiers of species-specific re- sources, such as MGI and FlyBase (Table 1 ). The PANGEA knowledgebase stores the information of gene set classifica- tion, gene annotation obtained from NCBI as well as the inf ormation f or ID mapping among various r esour ces. Building gene sets of pr eferr ed tissue expression To study the di v ersity and dynamics of the Drosophila transcriptome, the modENCODE consortium sequenced the transcriptome in twenty-nine dissected tissues ( 23 ) and the processed datasets are available at FlyBase ( http: //ftp.flybase.net/r eleases/curr ent/pr ecomputed files/genes/ ). A program was implemented in the Python programming language to identify genes expressed at a substantially higher le v els in one tissue versus any other tissue. This pr ogram first gr oups the RNA-seq datasets based on tissue. For example, all data related to the nervous system are grouped together. It then calculates the average reads per kilobase per million mapped r eads (RPKM) expr ession values for each gene in each tissue group. Genes were identified as pr efer entially expr essed in a gi v en tissue group if their average expression in the tissue group is 3-fold or higher than the av erage e xpression in any other tissue group. Genes with average RPKM value lower than 10 were excluded. Genes defined in this way as ‘tissue-specific’ then get annotated with the relevant tissue to generate the tissues expression classification set. Datasets used for testing Drosophila cell RNAi screen phenotype data was obtained from the DRSC ( https://www.flyrnai.org/ ) via download of a file of all available public screen ‘hits’ (results) ( https: //www.flyrnai.org/RN Ai all hits.txt ). RNAi reagents of op- timal design were selected. The criteria for optimal design were no CAN or CAR repeat, fewer than six predicted OTEs (off-target alignment sites of 19 bp) and a single gene target. CAN and CAR r epeats ar e thr ee base tandem r e- peats such as CAA CAGCA CCAT (CAN repeat, the third position can be A, G, C or T) and CAACA GCA GCAA (CAR repeat, the 3rd position can be A or G). RNAi r eagents wer e mapped to curr ent FlyBase gene identifiers using a DRSC internal mapping tool. Screens focused on major signalling pathways were selected for PANGEA anal- ysis ( 24–29 ). Proteomics data was obtained from Tang et al. ( 30 ) and high-confident prey proteins identified by mass-spec (Supplementary Table S2) were used for analy- sis. Gene set enrichment statistics used at PANGEA Hypergeometric testing is performed to calculate P val- ues for GSEA using the PypeR function in R. Bonfer- roni correction for multiple statistical tests, Benjamini- Hochberg procedure for false discovery rate adjustment, and Benjamini-Yekutieli procedure for false discovery rate adjustment were performed using the p.adjust function in R. Web tool implementation PANGEA is a SaaS (Software as a Service) w e b tool ( https: //www.flyrnai.org/tools/pangea/ ) and is built following a three-tier model, with a w e b-based user-interface at the front end, the knowledgebase at the backend, and the busi- ness logic in the middle tier communicating between the front and back ends by matching input genes with gene sets, doing statistical analysis and building visualization graphs. The front page is written in PHP using the Sym- f on y frame wor k and front-end HTML pages using the Twig template engine. The JQuery JavaScript library is used to facilitate Ajax calls to the back end, with the DataTables plugin f or displa ying tab le vie ws and Cytoscape and Veg- aLite packages for the da ta visualisa tions. The Bootstrap frame wor k and some custom CSS are used on the user in- terface. A mySQL database is used to store the knowledge- base. Both the w e bsite and databases are hosted on the O2 high-performance computing cluster, which is made avail- able by the Research Computing group at Harvard Medical School. RESULTS Pr epar ation of the classified gene sets for GSEA: the PANGEA knowledgebase GSEA relies on high-quality annotation of genes / gene products with information related to their biological func- tions. For PANGEA, we used multiple sources of annota- tion to generate > 300 000 different classes of gene func- tion for fiv e major model organisms ( D. melanogaster, C. elegans, D. rerio, M. musculus, R. norvegicus ) and human. For example, pathway annotations allow users to identify metabolic or signalling pathways that are over-represented in a gene list and help understand causal mechanisms un- derlying the observed phenotype from a scr een. P athway annotations from KEGG, PantherDB, and Reactome, as well as manually curated Drosophila gene sets, such as Fly- Base Signalling Pathways and the DRSC PathON annota- tion ( 18 , 21 ), are included in the PANGEA knowledgebase. The GO annotation set provides the comprehensi v e knowledge on gene functions and we store the gene-to-gene set relationships from GO in two ways. One is the direct gene-to-GO term associations as obtained from the gene association file while the other stores the gene-to-GO term associa tions with considera tion gi v en to child-par ent r ela- tionships. The latter is recommended for use in GSEA as it reflects the intended use of the ontology in curation prac- tice. The direct, gene-to-term set may be useful to under- stand the depth of annotation for each gene. In addition, we also generated two gene annotation subsets using evi- dence codes. The ‘experimental data only’ subset includes only those gene associations that are supported by experi- mental evidence codes. The ‘excluding high-throughput ex- periments’ subset excludes annotations only supported by HTP evidence codes. Excluding HTP data may be impor- tant to avoid bias when analysing similar studies ( 31 ). GO slim subsets are the cut-down versions of GO that give a broad ov ervie w of the ontology content without the detail of the specific, fine-grained terms. The PANGEA knowl- W422 Nucleic Acids Research, 2023, Vol. 51, Web Server issue Table 1. Species coverage by PANGEA and corresponding species-specific databases Species Abbreviation Species specific database URL Example Drosophila melanogaster Homo sapiens Mus musculus Caenorhabditis elegans Danio rerio Rattus norvegicus dm hs mm ce dr rn FlyBase https://flybase.org wg, FBgn0284084 HGNC MGI WormBase ZFIN RDG https://www.genenames.org http://www.informatics.jax.org/ https://www.wormbase.org https://zfin.org https://rgd.mcw.edu/ WNT1, HGNC:12774 Wnt1, MGI:98953 cwn-1, WBGene00000857 wnt1,ZDB-GENE-980526–526 Wnt1, RGD:1597195 Table 2. Knowledgebase of PANGEA built from various gene annotation r esour ces Type Source URL Species covered at PANGEA Gene Ontology GO http://geneontology.org/ hs,mm,rn,dr,dm,ce pathway pathway pathway pathway pathway group group group protein protein KEGG REACTOME PantherDB FlyBase pathway PathON HGNC FlyBase gene group GLAD COMPLEAT EBI protein complex phenotype AGR disease phenotype expression FlyBase phenotype AGR expression https://www.genome.jp/kegg/ https://reactome.org/ http://www.pantherdb.org/ https://flybase.org/ https: //www.flyrnai.org/tools/pathon/ https://www.genenames.org/ https://flybase.org/ https: //www.flyrnai.org/tools/glad/ https: //www.flyrnai.org/compleat/ https: //www.ebi.ac.uk/complexportal https: //www.alliancegenome.org/ https://flybase.org/ https: //www.alliancegenome.org/ hs,mm,rn,dr,dm,ce hs,mm,rn,dr,dm,ce dm dm dm hs dm dm dm hs,mm,rn,dr,dm,ce hs,mm,rn,dr,dm,ce dm mm,rn,dr,dm,ce Source update frequency irregular, usually 1–2 months unknown unknown irregular 2 months irregular unknown 2 months irregular irregular 2 months 3–4 months 2 months 3–4 months edgebase includes two sets of GO slim annotations from differ ent r esour ces. In addition to GO and pa thway annota tions, MODs provide organism-specific curation of important aspects of gene information, such as gene expression and mutant phe- notype, that are not captured in GO. The Alliance is focused on the harmonisation and centralisation of major MODs data ( 4 , 32 ). To take advantage of this effort, we integrated gene-to-tissue expression and gene-to-disease (model) asso- cia tion annota tions from the Alliance into the PANGEA kno wledgebase. As all or ganisms in the Alliance use the Disease Ontology (DO) for annotation, this set is easily comparable across species. The Alliance DO annotation set also includes disease association to model organism genes via an electronic pipeline using orthology with human dis- ease genes which expands the set provide by the MODs. Moreover, for Drosophila genes we assembled an additional gene set from phenotype annota tion a t FlyBase by extract- ing phenotype data associated with a ‘single allele’ genotype (i.e. single classical or insertional alleles), allowing users to perform meaningful enrichment analyses on this data class for the first time. Also included in PANGEA are gene group classifica- tion (eg. kinases and transcription factors) from organism- specific r esour ces (human and fly), protein complex anno- tations for multiple organisms from the EMBL-EBI Com- plex Portal ( 33 ) and COMPLEAT ( 22 ) and bespoke gene sets using Drosophila modENcode RNAseq data to iden- tify genes particularly highly expressed in one tissue (see Materials and Methods). In summary, we have assembled > 300 000 different gene sets that can be used in PANGEA to assess the enrichment of particular biological features in an input gene list. Features of the PANGEA user interface GSEA can be computationally intensi v e because of the number of gene sets being tested and potentially large num- ber of genes entered by users. Ther efor e, the step of pre- processing user’s input by mapping the input gene identi- fiers to the gene identifiers used for gene set annotation, is set up as a standalone ID mapping page (accessed by click- ing ‘Gene Id Mapping’ on the top toolbar) instead of com- bining it with the analysis step. Gene identifiers supported by PANGEA include Entrez Gene IDs, official gene sym- bols and primary gene identifiers from MODs. Users might need to analyse lists of other identifiers such as UniPro- tKB IDs and Ensembl gene IDs. Users can use ‘Gene Id Mapping’ tool and select an organism of choice to map IDs. As gene annotation is an on-going process, the gene identifiers as well as gene symbols might change over time. Even with the same type of gene identifiers such as FlyBase gene ID, the IDs used by users might be from a different FlyBase r elease. Ther efor e, ID-mapping step is an optional Nucleic Acids Research, 2023, Vol. 51, Web Server issue W423 Figure 1. Example of analysing a single gene list using PANGEA. A proteomic interaction dataset was selected from a study of the m6A methyltr ansfer ase complex MTC ( 30 ). The 75 high-confidence interactors of the four subunits of the MTC (METTL3, METTL14, Fl(2)d and Nito) identified by affinity- purified mass spectrometry from Drosophila S2R+ cells were submitted via the ‘Search Single’ option at PANGEA and enrichment analysis was performed over phenotype, GO SLIM2 BP and protein complex annotation from COMPLEAT (literature based). The result was filtered using P value 1 × 10 −5 cut-off and was illustrated as ( A ) a bar graph and ( B ) a network graph of selected gene sets from phenotype annotation and GO SLIM2 BP. Triangle nodes r epr esent gene sets and circle nodes represent genes while the edges r epr esent gene to gene set associations. ( C ) A network graph for selected gene sets from phenotype annotation and COMPLEAT protein complex annotation (literature based). but recommended first step to ensure that the entered IDs are synchronized with the IDs used by PANGEA gene set annotation. Users of FlyBase may also directly export a ‘HitList’ of genes generated in FlyBase to the tool by se- lecting the ‘PANGEA Enrichment Tool (DRSC)’ from the dropdown ‘Export’ menu (Supplementary Figure S1). An option for users to upload a background gene list for the analysis is provided; this may be useful when analysing hits from a focused screen using a kinase sub-library instead of a genome-scale library, for example. PANGEA identi- fies all relevant gene sets and provides enrichment statis- tics such as P -values, adjusted P -values, and fold enrich- ment, as well as the genes shared by the input gene list and gene set members. Users have the option to set differ- ent P -value cut-offs and visualise the results using a bar graph, the height and colour intensity of which can be customised (Figure 1 A). In addition, users can select gene sets of interest to examine the overlap of genes in differ- ent gene sets using the ‘Gene Set Node Graph’ visualisa- tion option. Nodes of different shapes in the network indi- cate genes or gene sets while edges reflect the gene-to-gene set relationship. This type of visualisation can help users identify the most relevant genes in each gene set as well as commonly shared or distinct gene members of the selected gene sets (Figure 1 B, C). An under-appreciated use of GSEA tools is that re- searchers often use them as simple gene classification tools, for example, asking ‘which genes in my list are kinases?’ to help inform further computational or experimental analy- sis. Having di v erse classification sets is important because depending on the type of data / experiment being analysed, different gene sets may be more useful than others. It is often useful to be able to compare similar gene sets from different sources to help evaluate the evidence for support. In addi- tion, PANGEA not only reports genes in an enriched gene set but also reports genes not covered by the gene set cate- gory selected, which may be interesting because of their lack of characterisation. This feature can help user answer ques- tion like ‘which genes in my list are not covered by KEGG annotation?’. W424 Nucleic Acids Research, 2023, Vol. 51, Web Server issue Figure 2. Examples of analysing multiple gene lists using PANGEA. ( A ) The prey proteins of multiple baits from AP mass-spec dataset ( 30 ) were submitted via the ‘Search Multiple’ option at PANGEA. The gene sets of protein complex annotation from COMPLEAT were selected. The comparison of enrichment over annotated protein complexes from the interacting proteins of four different baits was illustrated using a heatma p. ( B ) RN Ai screen data of signalling pathway studies were obtained from DRSC RNAi data repository ( 24 ) and the hits were submitted via the ‘Search Multiple’ option at PANGEA. The gene sets of FlyBase pathway annotation were used. The comparison of the enrichment of signalling pathway components from the screen hits of fiv e studies was illustrated using a heatmap. Often users need to anal yse m ultiple gene lists and com- par e the r esults; howe v er, majority of current w e b-based tools only allow the analysis of a single input list (plus back- ground). Thus, users have to perform comparisons manu- ally or using different tools. To address this need, PANGEA allows users to input multiple gene lists and compare results directly via a heatmap or a dot plot visualisation. For exam- ple, users might input gene hits from different phenotypic scr eens and compar e wha t pa thways, gene groups or biolog- ical processes that are common or different among results (Figure 2 ). Testing the utility of PANGEA To test the utility of PANGEA, we first analysed a pro- teomic interaction dataset from a study of the m6A methyl- tr ansfer ase complex MTC ( 30 ). In this study, using individ- ual pull-downs of the four subunits (METTL3, METTL14, Fl(2)d and Nito) of the MTC complex, high-confidence interactors were identified by mass-spectrometry from Drosophila S2R + cells. Submitting the combined list of 75 interacting proteins from the four baits via the ‘Search Single’ option at PANGEA (accessed by click- ing ‘Search Single’ on the top toolbar) for Fly and per- forming GSEA over phenotype, SLIM2 GO BP as well as protein complex annotation from COMPLEAT (liter- ature based) enrichment, we identified mRNA metabolic pr ocess (GO:0016071), pr otein folding (GO:0006457), ab- nor mal sex-deter mination (FBcv:0000436), abnor mal neu- roanatomy (FBcv:0000435), CCT complex and Spliceo- some complex among the top enriched gene sets with the −5 ) (Figure 1 A). most significant p-values (all < 1 × 10 Next, we visualised SLIM2 GO BP and phenotype annota- tions using a network gra ph. GO mRN A metabolic process hits overlap with both abnormal sex-determination and ab- normal neuroanatomy phenotypes, but GO protein folding hits onl y overla p with the abnormal neuroanatomy pheno- type (Figure 1 B). We also visualised protein complex and phenotype annotations using a different network graph, showing Spliceosome hits overlap with the phenotypes of abnor mal sex-deter mination and abnor mal neuroanatomy while the CCT complex hits only overlap with the abnor- Nucleic Acids Research, 2023, Vol. 51, Web Server issue W425 mal neuroanatomy phenotype (Figure 1 C). Enrichment of se x / reproducti v e phenotypes align with known function of MT C in r egulating the splicing of female-specific Sex lethal (Sxl) and its roles in alternati v e splicing and se xual dimor- phism, as well as the germ stem cell dif ferentia tion in the ovary ( 34 ). These GSEA results are also concordant with the fact that MTC is also known to have a significant role in neuronal mRNA regulation. The benefit of the network visualization is apparent when viewing how the gene set as- signment overlap (Figure 1 B, C), which re v eals that some of the MTC interacting proteins are associated with abnor- mal neuroanatomy phenotype and that the mechanism of the association is through the CCT complex in the pro- cess of protein folding. In contr act, the inter acting pro- teins from Spliceosome have more broad impacts related to both abnormal neuroanatomy phenotype and abnormal sex-determination phenotypes through mRNA metabolic process. We further analysed the protein complexes associated with each individual subunit using the ‘Search multiple’ option at PANGEA (accessed by clicking ‘Search Multi- ple’ on the top toolbar) and inputting the interacting pro- tein lists for each bait, then compared the enrichment re- sult using a heatmap visualisation (Figure 2 A). The re- sults indicated that some complexes, such as spliceosome subunits, are common to all MTC subunits, whereas some ar e mor e specific, such as protein complex es CCT com- plex for METTL14 and METTL13. In addition, we fur- ther analysed phenotype enrichment for proteins associated with each individual subunit using the ‘Search multiple’ op- tion, and comparison of the enrichment results shows many over lapping phenotypes, particular ly with regard to sterility (Supplementary Figure S2). At another use case, we looked at phenotypic cell screen data. Large-scale RN A interference (RN Ai) screening is a powerful method for functional studies in Drosophila . At the DRSC, datasets generated from more than one hundred scr eens ar e pub licly availab le ( 24 ). We selected fiv e screens designed to identify the genes for major signalling pathways and performed a GSEA analysis of the hits using the multi- ple gene list enrichment function of PANGEA. FlyBase sig- nalling pathway gene sets were selected and the results of the fiv e scr eens wer e compar ed side-by-side using a heatmap, w hich clearl y illustrated enrichment of the core components of the corresponding pathways, as well as potential cross- talks between pathways (Figure 2 B). These use cases of PANGEA for phenotype screening data as well as proteomics data demonstrate the value of the tool in validating screen results as well as generating new hypotheses for further study. DISCUSSION GSEA is a computational method used to identify sig- nificantly over-r epr esented gene classes within an input gene list(s) by testing against gene sets assembled based on prior knowledge. Input gene lists are typically from high- throughput screens or analyses. Here we present PANGEA, a newly developed GSEA tool with major model organ- isms as its focus, that includes gene sets that are usually not utilized by other GSEA tools, such as expression and disease annotations from the Alliance, phenotype annota- tions from FlyBase, and GO subsets with different configu- rations. PANGEA is easy to use and has new features such as allowing enrichment analyses for multiple input gene lists and gener ating gr aphical outputs that make comparisons straightf orward f or users. In addition to the use cases pre- sented here, i.e. analysing phenotypic screening and pro- teomic data, we anticipate that the tool will also facilitate analysis of gene lists from other types of data. For exam- ple, analysis of single-cell RNA-seq datasets at PANGEA might help users identify pathways and biological processes that are characteristic of various cell types. Users will also be able to answer questions on classification such as, ‘which genes in this list are kinases?’. PANGEA is designed to ac- commodate a wide range of biological data types and ques- tions, providing users with a w e b-based analysis tool that is easily accessible and user-friendly. We also note that gene classifications are not static, and the generic design of the tool means that it will be easy to update or expand PANGEA for more gene set classification and / or more species. In de v eloping PANGEA, we sought to improv e the effecti v eness of GSEA by (i) providing multiple collections of genes classified by their function in different ways (classified gene sets); (ii) ensuring the data underly- ing the classification of gene function was up to date and (iii) improving the visualization so that results from multi- ple gene sets or multiple gene lists could be compared easily. DA T A A V AILABILITY The online r esour ce is available without restriction at https: //www.flyrnai.org/tools/pangea/ . SUPPLEMENT ARY DA T A Supplementary Data are available at NAR Online. ACKNOWLEDGEMENTS We would like to thank the members of the Perrimon lab- oratory, the FlyBase consortium, the Drosophila RNAi Screening Center (DRSC), and the Transgenic RNAi Project (TRiP) for the discussion and suggestions during the design and implementation of the tool as well as the feed- back during the tool testing. Additional thanks to Gil dos Santos (Harvard, US) and Gillian Millburn (Cambridge, UK) at FlyBase for their genotype-to-phenotype work. [P41 GM132087]; FlyBase FUNDING NIH / NIGMS grant NIH / NHGRI [U41HG000739]; UK Medical Research Council [MR / W024233 / 1]; N.P. is an investigator of Howard Hughes Medical Institute. Funding for open access charge: NIH / NIGMS grant [P41 GM132087]. Conflict of interest statement. None declared. REFRENCES 1. Wang,J., Al-Ouran,R., Hu,Y., Kim,S.Y., Wan,Y.W., Wangler,M.F., Yamamoto,S., Chao,H.T., Comjean,A., Mohr,S.E. et al. (2017) W426 Nucleic Acids Research, 2023, Vol. 51, Web Server issue MARRVEL: integration of Human and Model Organism Genetic Resources to Facilitate Functional Annotation of the Human Genome. Am. J. Hum. Genet. , 100 , 843–853. 2. Hu,Y., Comjean,A., Mohr,S.E., FlyBase,C. and Perrimon,N. (2017) Gene2Function: an Integrated Online Resource for Gene Function Discovery. G3 (Bethesda) , 7 , 2855–2858. 3. Shefchek,K.A., Harris,N.L., Gargano,M., Matentzoglu,N., Unni,D., Brush,M., Keith,D., Conlin,T., Vasilevsky,N., Zhang,X.A. et al. (2020) The Monarch Initiati v e in 2019: an integrati v e data and analytic platform connecting phenotypes to genotypes across species. Nucleic Acids Res. , 48 , D704–D715. 4. Alliance of Genome Resources Consortium (2022) Harmonizing model organism data in the Alliance of Genome Resources. Genetics , 220 , https://doi.org/10.1093/genetics/iyac022 . 5. Sherman,B.T., Hao,M., Qiu,J., Jiao,X., Baseler,M.W., Lane,H.C., Imamichi,T. and Chang,W. (2022) DAVID: a w e b server for functional enrichment analysis and functional annotation of gene lists (2021 update). Nucleic Acids Res. , 50 , W216–W221. 6. Boyle,E.I., Weng,S., Gollub,J., Jin,H., Botstein,D., Cherry,J.M. and Sherlock,G. (2004) GO::TermFinder–open source software for accessing Gene Ontology information and finding significantly enriched Gene Ontology terms associated with a list of genes. Bioinformatics , 20 , 3710–3715. 7. Eden,E., Navon,R., Steinfeld,I., Lipson,D. and Yakhini,Z. (2009) GOrilla: a tool for discovery and visualization of enriched GO terms in ranked gene lists. BMC Bioinf. , 10 , 48. 8. Mi,H., Muruganujan,A., Huang,X., Ebert,D., Mills,C., Guo,X. and Thomas,P.D. (2019) Protocol update for large-scale genome and gene function analysis with the PANTHER classification system (v.14.0). Nat. Protoc. , 14 , 703–721. 9. Liao,Y., Wang,J., Jaehnig,E.J., Shi,Z. and Zhang,B. (2019) WebGestalt 2019: gene set analysis toolkit with revamped UIs and APIs. Nucleic Acids Res. , 47 , W199–W205. 10. Raudvere,U., Kolberg,L., Kuzmin,I., Arak,T., Adler,P., Peterson,H. and Vilo,J. (2019) g:profiler: a w e b server for functional enrichment analysis and conversions of gene lists (2019 update). Nucleic Acids Res. , 47 , W191–W198. 11. Gene Ontology, C. (2021) The Gene Ontology r esour ce: enriching a GOld mine. Nucleic Acids Res. , 49 , D325–D334. 12. Ashburner,M., Ball,C.A., Blake,J.A., Botstein,D., Butler,H., Cherry,J.M., Davis,A.P., Dolinski,K., Dwight,S .S ., Eppig,J.T. et al. (2000) Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat. Genet. , 25 , 25–29. 13. Gillespie,M., Jassal,B., Stephan,R., Milacic,M., Rothfels,K., Senff-Ribeiro,A., Griss,J., Sevilla,C., Matthews,L., Gong,C et al. (2022) The reactome pathway knowledgebase 2022. Nucleic Acids Res. , 50 , D687–D692. 14. Kanehisa,M., Furumichi,M., Sato,Y., Kawashima,M. and Ishiguro-Watanabe,M. (2023) KEGG for tax onom y-based analysis of pathways and genomes. Nucleic Acids Res. , 51 , D587–D592. 15. Kohler,S., Gargano,M., Matentzoglu,N., Carmody,L.C., Le wis-Smith,D., Vasile vsky,N.A., Danis,D., Balagura,G., Baynam,G., Brower,A.M. et al. (2021) The Human Phenotype Ontology in 2021. Nucleic Acids Res. , 49 , D1207–D1217. 16. Amberger,J.S. and Hamosh,A. (2017) Searching Online Mendelian Inheritance in Man (OMIM): a Knowledgebase of Human Genes and Genetic Phenotypes. Curr Protoc Bioinformatics , 58 , 1 2 1–1 2 12. 17. de Jong,A., Kuipers,O.P. and Kok,J. (2022) FUNAGE-Pro: comprehensi v e w e b server for gene set enrichment analysis of prokaryotes. Nucleic Acids Res. , 50 , W330–W336. 18. Gramates,L.S., Agapite,J., Attrill,H., Calvi,B.R., Crosby,M.A., Dos Santos,G., Goodman,J.L., Goutte-Ga tta t,D., Jenkins,V.K., Kaufman,T. et al. (2022) Fly Base: a guided tour of highlighted features. Genetics , 220 , iyac035. 19. Attrill,H., Falls,K., Goodman,J.L., Millburn,G.H., Antonazzo,G., Rey,A.J ., Marygold,S.J . and FlyBase,C. (2016) FlyBase: establishing a Gene Group r esour ce for Drosophila melanogaster. Nucleic Acids Res. , 44 , D786–D792. 20. Hu,Y., Comjean,A., Perkins,L.A., Perrimon,N. and Mohr,S.E. (2015) GLAD: an Online Database of Gene List Annotation for Drosophila. J Genomics , 3 , 75–81. 21. Ding,G., Xiang,X., Hu,Y., Xiao,G., Chen,Y., Binari,R., Comjean,A., Li,J., Rushworth,E., Fu,Z et al. (2021) Coordination of tumor growth and host wasting by tumor-deri v ed Upd3. Cell Rep. , 36 , 109553. 22. Vinayagam,A., Hu,Y., Kulkarni,M., Roesel,C., Sopko,R., Mohr,S.E. and Perrimon,N. (2013) Protein complex-based analysis framework for high-throughput data sets. Sci. Signal , 6 , rs5. 23. Brown,J.B., Boley,N., Eisman,R., May,G.E., Stoiber,M.H., Duff,M.O., Booth,B.W., Wen,J., Park,S., Suzuki,A.M. et al. (2014) Di v ersity and dynamics of the Drosophila transcriptome. Nature , 512 , 393–399. 24. Hu,Y., Comjean,A., Rodiger,J., Liu,Y., Gao,Y., Chung,V., Zirin,J., Perrimon,N. and Mohr,S.E. (2021) Fl yRN Ai.org-the database of the Drosophila RNAi screening center and transgenic RNAi project: 2021 update. Nucleic Acids Res. , 49 , D908–D915. 25. Nybakk en,K., Vok es,S.A., Lin,T.Y., McMahon,A.P. and Perrimon,N. (2005) A genome-wide RNA interference screen in Drosophila melanogaster cells for new components of the Hh signaling pathway. Nat. Genet. , 37 , 1323–1332. 26. DasGupta,R., Kaykas,A., Moon,R.T. and Perrimon,N. (2005) Functional genomic analysis of the Wnt-wingless signaling pathway. Science , 308 , 826–833. 27. Baeg,G.H., Zhou,R. and Perrimon,N. (2005) Genome-wide RNAi analysis of JAK / STAT signaling components in Drosophila. Genes Dev. , 19 , 1861–1870. 28. Kockel,L., Kerr,K.S., Melnick,M., Bruckner,K., Hebrok,M. and Perrimon,N. (2010) Dynamic switch of negati v e feedback regulation in Drosophila Akt-TOR signaling. PLos Genet. , 6 , e1000990. 29. Friedman,A.A., Tucker,G., Singh,R., Yan,D., Vinayagam,A., Hu,Y., Binari,R., Hong,P., Sun,X., Porto,M et al. (2011) Proteomic and functional genomic landscape of receptor tyrosine kinase and ras to extracellular signal-regulated kinase signaling. Sci. Signal , 4 , rs10. 30. Tang,H.W., Weng,J.H., Lee,W.X., Hu,Y., Gu,L., Cho,S., Lee,G., Binari,R., Li,C., Cheng,M.E et al. (2021) mTORC1-chaperonin CCT signaling regulates m(6)A RNA methylation to suppress autophagy. Proc. Natl. Acad. Sci. U.S.A. , 118 , e2021945118. 31. Attrill,H., Gaudet,P., Huntley,R.P., Lovering,R.C., Engel,S.R., Poux,S., Van Auken,K.M., Georghiou,G., Chibucos,M.C., Berardini,T.Z. et al. (2019) Annotation of gene product function from high-throughput studies using the Gene Ontology. Database (Oxford) , 2019 , baz007. 32. Alliance of Genome Resources, C. (2020) Alliance of Genome Resources Portal: unified model organism research platform. Nucleic Acids Res. , 48 , D650–D658. 33. Meldal,B.H.M., Perfetto,L., Combe,C., Lubiana,T., Ferreira Cavalcante,J .V., Bye,A.J .H., Waagmeester,A., Del-Toro,N., Shriv astav a,A., Barrera,E. et al. (2022) Complex Portal 2022: new curation frontiers. Nucleic Acids Res. , 50 , D578–D586. 34. Lence,T., Akhtar,J., Bayer,M., Schmid,K., Spindler,L., Ho,C.H., Kreim,N., Andrade-Navarro,M.A., Poeck,B., Helm,M. et al. (2016) m(6)A modulates neuronal functions and sex determination in Drosophila. Nature , 540 , 242–247. C (cid:2) The Author(s) 2023. Published by Oxford University Press on behalf of Nucleic Acids Research. This is an Open Access article distributed under the terms of the Creati v e Commons Attribution License (http: // creati v ecommons.org / licenses / by / 4.0 / ), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.
null
10.1021_acschembio.3c00413.pdf
null
null
pubs.acs.org/acschemicalbiology This article is licensed under CC-BY 4.0 Articles Direct Modulators of K‑Ras−Membrane Interactions Johannes Morstein,* Rebika Shrestha, Que N. Van, César A. López, Neha Arora, Marco Tonelli, Hong Liang, De Chen, Yong Zhou, John F. Hancock, Andrew G. Stephen, Thomas J. Turbyville, and Kevan M. Shokat* Cite This: ACS Chem. Biol. 2023, 18, 2082−2093 Read Online ACCESS Metrics & More Article Recommendations *sı Supporting Information ABSTRACT: Protein−membrane interactions (PMIs) are ubiq- uitous in cellular signaling. Initial steps of signal transduction cascades often rely on transient and dynamic interactions with the inner plasma membrane leaflet to populate and regulate signaling hotspots. Methods to target and modulate these interactions could yield attractive tool compounds and drug candidates. Here, we demonstrate that the conjugation of a medium-chain lipid tail to the covalent K-Ras(G12C) binder MRTX849 at a solvent-exposed site enables such direct modulation of PMIs. The conjugated lipid tail interacts with the tethered membrane and changes the relative membrane orientation and conformation of K-Ras(G12C), as shown by molecular dynamics (MD) simulation-supported NMR studies. In cells, this PMI modulation restricts the lateral mobility of K-Ras(G12C) and disrupts nanoclusters. The described strategy could be broadly applicable to selectively modulate transient PMIs. ■ INTRODUCTION Bifunctional molecules targeting biological interfaces are emerging therapeutic modalities that are undergoing a rapid expansion (e.g., PROTACS).1−4 To date, the majority of these strategies are focused on the modulation of protein−protein to target protein−membrane interactions, and methods interactions (PMIs) have remained relatively unexplored,5,6 despite their central importance in cellular signaling.7,8 Many targets in cancer signaling (e.g., Ras, PI3K, PKC, AKT) undergo transient and dynamic recruitment to the inner leaflet of the plasma membrane (PM), which could be susceptible to a relatively subtle pharmacological intervention. These targets include K-Ras4b (hereafter simply referred to as K-Ras), which is one of the most widely mutated cancer oncogenes.9−11 The lysine hypervariable region of K-Ras exhibits a patch of residues that aid in transiently associating K-Ras with the PM upon post-translational farnesylation. Inhibition of farnesyla- tion was extensively explored as a therapeutic strategy to inhibit K-Ras function but ultimately failed due to alternative rescued membrane attachment.12 More prenylation that recently, switch II pocket engagement has emerged as a direct strategy to covalently target the mutant allele K-Ras(G12C) giving rise to two clinically approved inhibitors sotorasib and adagrasib (Figure 1A).13−18 Moreover, this strategy has been translated to other mutant alleles K-Ras(G12S),19 K-Ras- (G12R),20 and K-Ras(G12D),21,22 this approach could be quite general. suggesting that by the C-terminal membrane anchor that consists of a farnesylated hexa-lysine polybasic domain. This anchor selectively associates with defined species of phosphatidylser- ine to form nanoclusters, comprising 4−6 K-Ras proteins.23−27 In addition, K-Ras diffusion is distinctive when compared to other paralogs, indicating that the lipid−protein environment that K-Ras explores is unique.11,28,29 Importantly, the specific lipid environment within K-Ras nanoclusters facilitates effector recruitment and activation.30−32 However, the precise mechanism underlying this PMI dependence in effector recruitment is currently unknown. New chemical tools that enable a precise modulation of these PMIs could therefore meet a critical need. Additionally, PMIs may present a therapeutic vulnerability that could be utilized in drug design. A number of monofunctional approaches to target PMIs have previously been reported for lipid clamp domains,8,33,34 and a screening hit for K-Ras with unique membrane-dependent behavior was found to modulate its PMIs in vitro.35 Herein, we attempted the rational design of bifunctional K- Ras(G12C) inhibitors with the capacity to directly modulate K-Ras−membrane interactions (Figure 1B). We envisioned Received: Accepted: Published: August 14, 2023 July 14, 2023 July 31, 2023 K-Ras’ association and interaction with plasma membrane lipids are essential for its function. K-Ras PMIs are mediated © 2023 The Authors. Published by American Chemical Society 2082 https://doi.org/10.1021/acschembio.3c00413 ACS Chem. Biol. 2023, 18, 2082−2093 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles Figure 1. Design and synthesis of direct modulators for K-Ras−membrane interactions. (A) Scheme of direct Ras inhibition. (B) Scheme of a direct Ras inhibitor that simultaneously modulates its membrane interaction. (C) Crystal structure of K-Ras(G12C) in complex with MRTX849 (PDB 6UT0), highlighting the solvent exposed site of MRTX849.17 (D) Chemical structures of lipidated analogues of MRTX849, C5-MRTX, C11-MRTX, C18-MRTX, and the noncovalent control compound C11′-MRTX. (E) Synthesis of lipidated MRTX849 conjugates. that the installation of a second lipid tail on the surface of K- Ras would allow for modulation of PMIs. To this end, we the solvent-exposed site of proposed the modification of known covalent binders of K-Ras(G12C) with lipophilic groups. Effects on PMIs were characterized extensively in vitro and in cellulo. ■ RESULTS AND DISCUSSION Design and Synthesis of Lipid-Conjugated K-Ras- (G12C) the crystal structure of MRTX849 bound to K-Ras(G12C)17 (PDB 6UT0) revealed partial solvent exposure of the pyrrolidine fragment of the Inhibitors. Analysis of covalently bound ligand (Figure 1C).36 We envisioned that this site could be utilized to append lipophilic groups on the surface of K-Ras with the capacity to directly interact with the membrane. To this end, a series of lipid-conjugated MRTX849 analogues with varying lipid chain lengths were designed (Figure 1D). A small-chain lipid (SCL) conjugate with a 5- carbon containing tail (C5-MRTX), a medium-chain lipid (MCL) conjugate with a 11-carbon tail (C11-MRTX), and a long-chain lipid (LCL) conjugate with an 18-carbon containing tail (C18-MRTX) were synthesized. The control compound C11′-MRTX, which replaced the cysteine-reactive acrylamide warhead with a nonreactive saturated analogue was 2083 https://doi.org/10.1021/acschembio.3c00413 ACS Chem. Biol. 2023, 18, 2082−2093 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles Figure 2. Biochemical and cell biological characterization of K-Ras(G12C) inhibitors. (A) LC/MS detection of covalent adducts of respective MRTX849-lipid conjugates to K-Ras(G12C) in vitro after 60 min. (B) Cellular target engagement using a TAMRA-Click assay.40 After 4 h incubation in H358 cells (G12C/WT), cells were harvested and incubated with TAMRA-azide to label the terminal end of lipids with a fluorophore. Pellets were blotted for RAS, and the shift of the upper band is indicative of cellular covalent engagement of K-Ras(G12C). (C) Dynamic light scattering measurement to determine critical aggregation concentration (CAC). Scattering intensity was plotted against logarithmic concentration. The origins of slope were used to identify the CAC as a starting point of aggregation. (D) Thermal stability shift assay using SYPRO Orange and covalently modified K-Ras(G12C) in vitro. (E) Cellular viability assay (CellTiter-Glo) of H358 cells with MRTX849, C5-MRTX, C11- MRTX, and C11′-MRTX (CTRL) after 72 h incubation at varying concentrations. also produced. All compounds were synthesized from a previously described MRTX849 intermediate37 and an N- functionalized prolinol derivative (Figure 1E). C11-MRTX is a Nonaggregating Potent Cellular Inhibitor of K-Ras(G12C). In vitro labeling of recombinant K-Ras(G12C) showed that C5-MRTX and C11-MRTX undergo rapid covalent modification of K-Ras(G12C), while the control compound C11′-MRTX and C18-MRTX do not label K-Ras(G12C) covalently (Figure 2A). To test if these results translate into cellular labeling of K-Ras(G12C), the alkyne moiety at the lipid terminus was utilized for copper- catalyzed azide-alkyne click chemistry38,39 leading to a shift in sodium dodecyl sulfate−polyacrylamide gel electrophoresis. Incubation of H358 (WT/G12C) cells with respective analogues of MRTX for 4 h, subsequent click labeling, SDS electrophoresis, and western blotting revealed effective cellular engagement of K-Ras(G12C) in cells by C5-MRTX and C11- MRTX as observed through a shift in SDS gel electrophoresis (Figure 2B) of the K-Ras band (note: H358 is a heterozygous cell line; partial labeling is observed due to the presence of a wildtype allele). Similar to our intact mass spectrometry experiments, covalent target engagement was not detected for C18-MRTX. We hypothesized that this could be due to an increased propensity of longer lipid tails to form aggregates, which was confirmed by a dynamic light scattering experiment (Figure 2C). Interestingly, the critical aggregation concen- tration of C5-MRTX was lower than that of MRTX849 and C11-MRTX was comparable to MRTX849. By contrast, C18- MRTX exhibited a much lower critical aggregation concen- tration (∼80 nM), which could be limiting its labeling efficiency and bioactivity. MRTX849 engages the switch II pocket of K-Ras(G12C) leading to a marked stabilization of its fold. To assess if our lipid conjugates behave similarly, we used a thermal shift assay with SYPRO Orange (Figure 2D). Notably, MRTX849-, C5- MRTX-, and C11-MRTX-labeled K-Ras(G12C) variants all showed a large thermal shift compared to nonlabeled K- Ras(G12C). At the same time, the shift between the three labeled variants exhibits no detectable differences, suggesting that the lipid tail does not strongly bind to K-Ras(G12C), which is desirable for it to potentially interact with the inner leaflet of the PM. To confirm that C11-MRTX exhibits specific cellular toxicity in K-Ras(G12C)-driven cancer cell lines, we performed a cell viability assay with C11-MRTX and the noncovalent control compound C11′-MRTX (CellTiter- Glo).13,15 C11-MRTX was found to be significantly more potent than the negative control compound (Figure 2E), which verifies that this inhibitor exhibits potent cellular activity despite the MCL conjugation. C11-MRTX Alters the Relative Conformation of K- Ras(G12C) on the PM. To study the capacity of C11-MRTX to modulate K-Ras−membrane interactions, we decided to employ coarse-grained MD simulations with Martini 3 force fields in conjunction with NMR paramagnetic relaxation enhancement (NMR-PRE) for K-Ras(G12C)·MRTX849 and K-Ras(G12C)·C11-MRTX that were chemically tethered to 2084 https://doi.org/10.1021/acschembio.3c00413 ACS Chem. Biol. 2023, 18, 2082−2093 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles Figure 3. MD simulations and NMR-PRE experiments with membrane-tethered K-Ras(G12C). (A) Model of C11-MRTX modified K-Ras(G12C) tethered to a model membrane for MD simulations. (B) MD simulations revealed transient membrane engagement through the MCL of C11- MRTX leading to a novel bianchored conformation of K-Ras(G12C) on the membrane. (C) Ranking of membrane contacts of ligands for simulation with K-Ras(G12C)·MRTX849 (top) and K-Ras(G12C)·C11-MRTX (bottom). (D) Selected peaks from K-Ras(G12C/C118S)· MRTX849 and K-Ras(G12C/C118S)·C11-MRTX on nanodisks with and without the PRE tag Tempo. (E) Structure of K-Ras(G12C) highlighting areas that are moved close to the membrane when bound to C11-MRTX relative to MRTX849 in blue and moved further away in red. (F) NMR-PRE ratios for K-Ras·MRTX849 and K-Ras·C11-MRTX tethered to nanodisks. lipid nanodisks. MD simulations of K-Ras(G12C)·C11-MRTX (Figure 3A) revealed transient binding of the C11-MRTX MCL to the membrane leading to unusual bianchored conformations of K-Ras(G12C) (Figure 3B). To visualize the membrane contacts induced by C11-MRTX relative to ligand−membrane contacts were MRTX849, counted (Figure 3C), showing frequent membrane contacts for for C11-MRTX but near zero membrane contacts the direct 2085 https://doi.org/10.1021/acschembio.3c00413 ACS Chem. Biol. 2023, 18, 2082−2093 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles Figure 4. Single-molecule tracking of K-Ras(G12C)·C11-MRTX in HeLa cells. (A) Schematic of HaloTag-tagged K-Ras used for TIRF single- particle tracking experiment. (B) TIRF image of JF549-chloroalkane labeled HaloTag K-Ras(G12C) in HeLa cells. (C) Representative trajectories of diffusion for labeled K-Ras(G12C) on the inner leaflet of the PM. Colors represent different single-molecule tracks over time. (D−F) Mean- square displacement plots calculated from the trajectories obtained for HaloTag K-Ras(G12C) labeled with 50 pM JF549 treated with no drug (black), 10 μM C11-MRTX (blue), 10 μM C11′-MRTX (green), and 10 μM MRTX849 (orange) for 30 min (D), 1 h (E), and 2 h (F) of compound incubation. In panels (E, F), the orange and green lines partially cover each other. MRTX849. To test these predictions experimentally, we tethered K-Ras(G12C) to nanodisks and conducted protein NMR studies. We observed marked chemical shift perturba- tions comparing K-Ras(G12C) bound to MRTX849 versus C11-MRTX. These shifts occurred on residues of SI, SII, and α3 regions (Figure 3D). Residue 63 from the switch II region had a particularly strong chemical shift response. This was consistent with the MD prediction of regions in K-Ras(G12C) moving into closer proximity of the membrane (marked in blue, Figure 3E). We further conducted NMR-PRE experi- ments which confirmed greater membrane proximity of residues 62 to 66 in the switch II region of K-Ras and an overall decrease in NMR-PRE ratios for β1, α3, and α4 residues (Figure 3F). K-Ras(G12C)·C11-MRTX had a longer rotational correlation time of 22.8 vs 18.4 ns for K-Ras(G12C)· MRTX849, which provided additional support for its closer membrane proximity (Figure S4). C11-MRTX Modulates the Diffusion of PM-Localized K-Ras(G12C) in Live Cells. To study if the PMI modulations observed in MD simulations and NMR experiments translate to live cells, we decided to study the lateral diffusion of labeled live cells.11 K- K-Ras(G12C) on the inner PM leaflet of internal Ras(G12C) diffusion was measured using total reflection microscopy (TIRF) employing a charge-couple 2086 https://doi.org/10.1021/acschembio.3c00413 ACS Chem. Biol. 2023, 18, 2082−2093 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles Figure 5. Nanoclustering of K-Ras(G12C)·C11-MRTX in MDCK cells. (A) TEM image of 4.5 nm gold nanoparticles immunolabeling the GFP- tagged K-Ras(G12C) at a magnification of 100,000X. (B-D) Color-coded TEM images of the gold-labeled GFP-tagged K-Ras(G12C) treated with DMSO (B), C11-MRTX (C), and MRTX849 (D). (E, F) Analysis of PM localization and nanoclustering for GFP-K-Ras(G12C) (E) and GFP-K- Ras(G12D) (F). Error bars indicate mean ± SEM of the at least 15 PM sheet images for each condition. Bootstrap tests evaluated the statistical significance of the Lmax data, while one-way ANOVA calculated the statistical significance of the gold labeling data, with * indicating p < 0.05. that device (CCD) camera for fast frame rate acquisition and a bright organic dye covalently linked to HaloTag K-Ras(G12C) overexpressed in HeLa cells (Figure 4A,B). The result demonstrates labeling of K-Ras(G12C) with C11- MRTX leads to marked changes in its dynamic diffusion along the PM. While no clear trends could be observed within 30 min (Figure 4C), C11-MRTX showed a marked reduction in diffusion rates compared to MRTX849 and C11′-MRTX after 1 h (Figure 4D) and further pronounced after 2 h (Figure 4E). We reasoned that the lateral restriction in K-Ras(G12C) mobility along the plasma membrane is a likely effect of the additional membrane contacts established by the C11-lipid tail. We further the subcellular distribution of K-Ras, for example by shifting its localization from the plasma membrane to endomembranes.41 Confocal imaging of GFP-fused K-Ras did not reveal alterations in the subcellular localization of K-Ras (Figure S5). tested if our molecules alter C11-MRTX Disrupts K-Ras(G12C) Nanoclusters. The spatial organization of K-Ras on the inner PM leaflet is critical for its physiological function. Transient nanoclusters were found to be the sites where effectors preferentially interact with K-Ras and are therefore especially critical for its physiological function.9,42 To test the lateral if C11-MRTX affects organization of K-Ras into nanoclusters, we conducted electron microscopy (EM) combined with spatial analysis43 in MDCK cells stably expressing GFP-K-Ras(G12C) or GFP- K-Ras(G12D) as control. Intact 2D PM sheets from cells treated with DMSO vehicle control, 10 μM C11-MRTX, or 10 μM MRTX849 for 2 h were fixed and labeled with 4.5 nm gold nanoparticles conjugated directly to anti-GFP antibody (Figure 5A). The gold particle spatial distributions were quantified using univariate K-functions expressed as L(r) − r. The maximum value of this function, Lmax, can be used as a summary statistic for the extent of nanoclustering. The extent of nanoclustering, L(r) − r, was plotted as a function of the length scale, r. The L(r) − r value of 1 is the 99% confidence the values above which indicate the statistically interval, meaningful clustering. Based on this K-function analysis, the EM images were color-coded to indicate the population distribution of the gold-labeled GFP-K-Ras(G12C). Larger L(r) − r values indicate more clustering (Figure 5B−D). We found that MRTX849 and C11-MRTX treatment both decreased the gold labeling density when compared with control, indicating that MRTX849 and C11-MRTX both reduced the localization of K-Ras(G12C) to the PM. C11- MRTX significantly reduced the Lmax value for GFP-K- Ras(G12C), indicating that C11-MRTX also disrupted the nanoclustering of K-Ras(G12C) (Figure 5E). Both MRTX849 and C11-MRTX had no effect on localization or nano- indicating selectivity for K- clustering of K-Ras(G12D), Ras(G12C) (Figure 5F). Combined, these data demonstrate the ability of the lipidated drug to selectively disrupt the lateral spatial organization of K-Ras(G12C) on the PM, which is function of GTP-bound K- critical Ras.30,31 the physiological for 2087 https://doi.org/10.1021/acschembio.3c00413 ACS Chem. Biol. 2023, 18, 2082−2093 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles ■ CONCLUDING REMARKS leaflet of Herein, we report a bifunctional chemical approach to directly modulate the interactions between K-Ras(G12C) and the inner the PM. This is achieved through the installation of a C11 medium-chain lipophilic group to the solvent-exposed site of the covalent K-Ras(G12C) inhibitor MRTX849. Medium-chain lipids are common in natural products, occur in drugs (e.g., fingolimod or orlistat), and may present a sweet spot for bioactive amphiphiles due to their capacity to partition, while exhibiting a lower propensity to aggregate compared to longer membrane lipids.44 In our lead molecule C11-MRTX, the conjugated lipid tail establishes new interactions with the inner leaflet of the plasma membrane, resulting in novel bi-anchored conformations of membrane- tethered K-Ras(G12C). Thereby, the nucleotide binding site and switch I/II regions are brought in closer proximity to the PM, as demonstrated through a combination of MD simulations and NMR experiments. In cells, C11-MRTX restricts the lateral mobility of K-Ras which was observed through a marked reduction in diffusion rates. Finally, C11- MRTX was found to disrupt K-Ras(G12C) nanoclusters, which are the sites of Ras effector recruitment and activation and thus essential for signal transmission of noninhibited K- Ras. Combined, these results demonstrate a targeted modulation of protein−membrane interactions. These types of interactions are ubiquitous in early steps of cellular signaling, and our strategy could be translatable to target other signaling or lipid binding factors. PMI modulators could provide useful tools to dissect the function of these interactions and hold promise for the design of novel therapeutic agents. ■ MATERIALS AND METHODS General Methods. Anhydrous solvents were purchased from Acros Organics. Unless specified below, all chemical reagents were purchased from Sigma-Aldrich, Oakwood, Ambeed, or Chemscene. Analytical thin-layer chromatography (TLC) was performed using aluminum plates precoated with silica gel (0.25 mm, 60 Å pore size, 230−400 mesh, Merck KGA) impregnated with a fluorescent indicator (254 nm). TLC plates were visualized by exposure to ultraviolet light (UV). Flash column chromatography was performed with Teledyne ISCO CombiFlash EZ Prep chromatography system, employing prepacked silica gel cartridges (Teledyne ISCO RediSep). Proton nuclear magnetic resonance (1H NMR) spectra were recorded on a Bruker Avance III HD instrument (400/100/376 MHz) at 23 °C operating with the Bruker Topspin 3.1. NMR spectra were processed using Mestrenova (version 14.1.2). Proton chemical shifts are expressed in parts per million (ppm, δ scale) and are referenced to residual protium in the NMR solvent (CHCl3: δ 7.26, MeOD: δ 3.31). Data are represented as follows: chemical shift, multiplicity (s = singlet, d = doublet, t = triplet, q = quartet, dd = doublet of doublets, dt = doublet of triplets, m = multiplet, br = broad, app = apparent), integration, and coupling constant (J) in hertz (Hz). High-resolution mass spectra were obtained using a Waters Xevo G2-XS time-of-flight mass spectrometer operating with Waters MassLynx software (version 4.2). When liquid chromatography−mass spectrometry (LC−MS) analysis of the reaction mixture is indicated in the procedure, it was performed as follows. An aliquot (1 μL) of the reaction mixture (or the organic phase of a mini-workup mixture) was diluted with 100 μL 1:1 acetonitrile/water. 1 μL of the diluted solution was injected onto a Waters Acquity UPLC BEH C18 1.7 μm column and eluted with a linear gradient of 5−95% acetonitrile/water (+0.1% formic acid) over 3.0 min. Chromatograms were recorded with a UV detector set at 254 nm and a time-of-flight mass spectrometer (Waters Xevo G2-XS). Intact Protein Mass Spectrometry. Purified K-Ras variants (4 μM final) were incubated with compounds at 50 or 100 μM (1% v/v DMSO final) in 20 mM HEPES pH 7.5, 150 mM NaCl, 1 mM MgCl2 in a total volume of 150 μL. After the noted time, the samples were analyzed by intact protein LC/MS using a Waters Xevo G2-XS system equipped with an Acquity UPLC BEH C4 1.7 μm column. The mobile phase was a linear gradient of 5−95% acetonitrile/water + 0.05% formic acid. The spectra were processed using QuantLynx, giving the ion counts observed for the most abundant species. TAMRA-Click Assay. This assay was performed as previously described.40 Briefly, cells (500,000 to 1,000,000 cells per well) were seeded into six-well ultralow attachment plates (Corning Costar #3471) and allowed to incubate at 37 °C overnight. Cells were treated with the indicated concentrations of compound combinations and then incubated at 37 °C for the indicated lengths of time. In preparation for sodium dodecyl sulfate−polyacrylamide gel electro- phoresis (SDS−PAGE) and immunoblotting, cells were pelleted at 4 °C at 500 g and washed twice with ice-cold phosphate-buffered saline (PBS). Lysis was conducted, and copper-catalyzed click chemistry was performed by addition of the following to each lysate at the following final concentrations: 1% SDS (20% SDS in water stock), 50 μM TAMRA-N3 (5 mM in DMSO stock), 1 mM TCEP (50 mM in water stock), 100 μM TBTA (2 mM in 1:4 DMSO/t-butyl alcohol stock), and 1 mM CuSO4 (50 mM in water stock). After 1 h at room temperature, the reaction was quenched with 6× Laemmli sample buffer before SDS−PAGE. Dynamic Light Scattering. Measurements were performed using a DynaPro MS/X (Wyatt Technology) with a 55 mW laser at 826.6 nm, using a detector angle of 90°. Histograms represent the average of three data sets. Differential Scanning Fluorimetry. The protein of interest was diluted with HEPES buffer [20 mM HEPES 7.5, 150 mM NaCl, 1 mM MgCl2] to 2 μM. 1 μL of SYPRO Orange (500×) was mixed with 99 μL of protein solution. This solution was dispensed into wells of a white 96-well PCR plate in triplicate (25 μL/well). Fluorescence was measured at 0.5 °C temperature intervals every 30 s from 25 to 95 °C on a Bio-Rad CFX96 qPCR system using the FRET setting. Each data set was normalized to the highest fluorescence, and the normalized fluorescence reading was plotted against temperature in GraphPad Prism 8.0. Tm values were determined as the temper- ature(s) corresponding to the maximum (ma) of the first derivative of the curve. Cell Viability Assay. Cells were seeded into 96-well white flat bottom plates (1000 cells/well) (Corning) and incubated overnight. Cells were treated with the indicated compounds in a seven-point threefold dilution series (100 μL final volume) and incubated for 72 h. Cell viability was assessed using a commercial CellTiter-Glo (CTG) luminescence-based assay (Promega). Briefly, the 96-well plates were equilibrated to room temperature before the addition of diluted CTG reagent (100 μL) (1:4 CTG reagent/PBS). Plates were placed on an orbital shaker for 30 min before recording luminescence using a Spark 20M (Tecan) plate reader. Molecular Simulations. Coordinates of K-Ras bound to MRTX849 were downloaded from the pdb database (6UTO). Missing residues in the HVR were modeled using Modeller,45 as a disordered region.46 The protein was represented using the Martini 347 coarse-grained force field in combination with the structure- based48 approach in order to maintain its secondary structure. The farnesyl group was represented using parameters published before49 and updated in order to keep consistency with Martini 3. MRTX849, C11-MRTX, and GDP molecules were modeled using the method- ology published before,50 and bead types were updated accordingly to match the Martini 3 force field interaction matrix. Harmonic bonds were used to maintain the stability of the ligands in their respective binding regions, a methodology used successfully in the past.2 A membrane lipid bilayer composed of 70:30 POPC/POPS was constructed using the “insane” tool.51 Before insertion of K-Ras, the membrane was pre-equilibrated at 310 K for 100 ns. Protein and ligands were inserted, embedding the farnesyl group into the lipid bilayer and removing overlapping Martini water beads. Systems were charge-neutralized, and ions (Na+, Cl−) were added to mimic a 150 mM ionic strength environment. Before production, system boxes were energy-minimized and trajectories were saved every 2 ns for 2088 https://doi.org/10.1021/acschembio.3c00413 ACS Chem. Biol. 2023, 18, 2082−2093 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles analysis. Each trajectory (2 total) was run for 30 μs. Simulations were carried out with GROMACS 2018.6,52 using a 20 fs time step for updating forces as recommended in the original publication. Reaction- field electrostatics was used with a Coulomb cutoff of 1.1 nm and dielectric constants of 15 or 0 within or beyond this cutoff, respectively. A cutoff of 1.1 nm was also used for calculating Lennard-Jones interactions, using a scheme that shifts the van der Waals potential to zero at this cutoff. Membranes were thermally coupled to 310 K using the velocity rescaling53 thermostat. Semi- isotropic pressure coupling was set for all systems at 1 bar using a Berendsen54 barostat with a relaxation time of 12.0 ps. DNA for Protein Production of K-Ras4b(1−185) G12C/ C118S. The gene for protein expression of Hs.K-Ras4b(1−185) initially G12C/C118S was generated from a DNA construct synthesized as a Gateway Entry clone (ATUM, Newark, CA). The construct consisted of an Escherichia coli gene-optimized fragment containing an upstream tobacco etch virus (TEV) protease site (ENLYFQ/G), followed by the coding sequence of human K- Ras4b(1−185). An entry clone was transferred to an E. coli destination vector containing an amino terminal His6-MBP (pDest- 566, Addgene #11517) tag by gateway LR recombination (Thermo Scientific, Waltham, MA). The construct generated was R949-x95- 566: His6-MBP-tev-Hs.K-Ras4b(1−185) G12C/C118S. The mem- brane scaffolding protein expression clone (pMSP delH5) was obtained from the group of Gerhard Wagner at Harvard University.55 Protein Expression and Purification. K-Ras4b(1−185) G12C/ C118S was expressed following the protocols described in Travers et al. for 15N/13C incorporation with modifications.49 Specifically, ZnCl2 was omitted and induction after IPTG addition was at 16 °C. Highly deuterated and 15N-labeled K-Ras protein was expressed using the protocols described in Chao et al.56 and purified essentially as outlined in Kopra et al.57 for K-Ras(1−169). pMSP delH5 was expressed and purified as described in Travers et al. NMR-PRE Sample Preparation and NMR Data Collection and Processing. Uniformly 15N/2H-labeled K-Ras(G12C/C118S) was first labeled with 2.5× excess of MRTX849 and C11-MRTX in 20 mM Hepes, pH 7.48, and 150 mM NaCl overnight at room temperature (∼11 h), and excess compounds were removed using a PD10 column equilibrated with 20 mM Hepes, pH 7.0, and 150 mM NaCl. Then, the MRTX849 and C11-MRTX bound K-Ras, concentration between 182 and 195 μM, were tethered to 2× excess of delH5 nanodisks composed of 63.75/30/6.25 POPC/POPS/PE MCC and 57.5/30/6.25/6.25 POPC/POPS/PE MCC/Tempo PC at room temperature overnight, followed by purification on an AKTA FPLC with a Superdex 200 Increase 10/300 column to remove nontethered K-Ras. All lipids were purchased from Avanti Polar Lipids. Empty delH5 nanodisks were made as described in Van et al. with pH 7.0 buffer.58 The final NMR buffer was 20 mM Hepes, pH 7.0, 150 mM NaCl, 0.07% NaN3, and 7.0% D2O. 280 μL of each sample was enclosed in 5 mm susceptibility-matched Shigemi tubes (Shigemi, Allison Park, PA) for NMR data collection. All NMR experiments were acquired on a Bruker AVANCE III HD spectrometer operating at 900 MHz (1H), equipped with a cryogenic triple-resonance probe. The temperature of the sample was regulated at 298 K throughout the experiments. Two-dimensional (2D) 1H,15N- TROSY-HSQC spectra were recorded with 1024 × 128 complex points for the 1H and 15N dimension, respectively, 128 scans, and a recovery delay of 1.5 s for a total collection time of 15 h. All 2D spectra were processed using NMRPipe59 and analyzed using NMRFAM-SPARKY.60 The NMR-PRE ratios were calculated from peak intensities and normalized to 1 (Figure 3F). Chemical shift perturbations (CSP) were calculated using CSP (ppm) = Sqrt((ΔN2/ 25 + ΔH2)/2) (Figure S1). The TROSY spectra for K-Ras·MRTX849 and K-Ras·C11-MRTX tethered to nanodisks without the Tempo PRE tag are shown in Figures S2 and S3, respectively. Expansion of the spectral region for residues 61 to 67 is shown in Figure 3D. To estimate the tumbling time of the K-Ras proteins in solution, 1H/15N-TRACT61 experiments were recorded as a series of one- dimensional (1D) spectra for the α and β states. For the 15N-α state, the relaxation delays were set to 0, 5, 10, 16, 22, 30, 40, 50, 64, 80, 100, 130, 170, and 240 ms. The relaxation delays for the faster- relaxing 15N-β state were set to 0, 1, 2, 4, 7, 11, 15, 20, 26, 32, 39, 47, 56, and 70 ms. Spectra for both the α and β states were recorded in a in an interleaved fashion. Each FID was single experiment accumulated for 1536 scans with a repetition delay between scans of 1.5 s for a total recording time of 18.5 h for both the α and β states. The interleaved spectra were separated in topspin using inhouse written scripts and analyzed using Mestrelab Research Mnova software. Plots showing the fits to calculate the rotational correlation time are shown in Figure S4. K-Ras·MRTX849 Backbone Chemical Shift Assignments. A sample of uniformly 13C,15N-labeled K-Ras bound to MRTX849 (6.4 mM in 20 mM Hepes, pH 7.0, with 150 mM NaCl, 1 mM MgCl2, 1 mM TCEP, 0.07% NaN3, and 7.0% D2O) was used to collect sequence-specific assignments of backbone resonances: two-dimen- sional (2D) 1H,15N-HSQC and three-dimensional (3D) HNCACB, 3D CBCA(CO)NH, 3D HNCA, 3D HN(CA)CO, 3D HNCO spectra, as well as a 3D NOESY 1H,15N-HSQC spectrum with a 100 ms mixing time. The 1H/15N assignments are shown in Figure S6. To increase the resolution of the C α cross-peaks in the 13C dimension of the 3D HNCA spectrum, band-selective shaped pulses (BADCOP) developed by optimal control theory were utilized to decoupled C α from C β nuclei.62 All NMR experiments were acquired on a Bruker AVANCE III HD spectrometer operating at 750 MHz (1H), equipped with a cryogenic triple-resonance probe. The temperature of the sample was regulated at 298 K throughout the experiments. All 3D spectra were recorded using nonuniform sampling (NUS) with sampling rates ranging between 30.5 and 33.3%. All spectra were processed using NMRPipe and analyzed in NMRFAM-SPARKY. The 3D spectra recorded with NUS were reconstructed and processed using the SMILE package available with NMRPipe. Single-Particle Tracking Experiments. HeLa cells were grown in Dulbecco’s modified Eagle medium (DMEM) (Thermo Fisher Scientific) supplemented with 1% 200 mM L-Glutamine and 10% FBS in a 6-well plate. The HaloTag fusion construct of K-Ras4b(G12C) was transiently transfected into each well using Fugene 6 transfection reagent (Promega) and 1.1 μg DNA per well. The protocol for plasmid design is described in Goswami et al.11 On the following day, cells were transferred on to plasma-cleaned coverslips (#1.5, 25 mm). On the day of imaging, the cells were first labeled with the fluorescent JF549 HaloTag ligand (Tocris) and then treated with the compounds. For labeling, the cells were first washed with 3 mL of PBS 3 times, incubated with 50 pM of JF54963 in complete media for 40 min, washed with 3 mL of PBS, and then allowed to recover in complete media for 30 min. For drug treatment, cells were first washed with 3 mL of PBS and then incubated with 10 μM of compound in complete media for the indicated time course. Single-particle tracking experiments were performed on the Nikon NStorm Ti-81 inverted microscope equipped with thermo-electric-cooled Andor iX EMCCD camera (Andor Technologies). During imaging experiments, the cells were maintained at 37 °C and 5% CO2 using a Tokai hit stage incubator (Tokai Hit Co., Ltd., Japan). The JF549 fluorescent molecules were illuminated under TIRF mode with the continuous 561 nm laser line from the Agilent laser module at 15% and imaged with an APO x100 TIRF objective with 1.49 NA (Nikon Japan). A 100 by 100-pixel region (16 × 16 μm2) of interest (ROI) was created in the cytoplasmic region of the PM in a cell and imaged at a frame rate of 10ms/frame for a total of 5000 frames. For each experiment, a minimum of 17 cells were imaged. Single-particle tracking movies were analyzed using the Localizer plugin embedded in Igor Pro software.64 Single particles in each frame were localized as spots based on the eight-way adjacency particle detection algorithm with a generalized likelihood ratio test (GLRT) sensitivity of 30 and a point spread function (PSF) of 1.3 pixels. The position of the PSF was estimated based on a symmetric 2D Gaussian fit function. If the particles persisted for more than 6 frames, they were then linked between consecutive frames into tracks. The particles were allowed a maximum jump distance of 5 pixels and blinking for one frame. For each experiment, tracks from all of the movies were combined into a single Matlab file and used to calculate mean-square displacement 2089 https://doi.org/10.1021/acschembio.3c00413 ACS Chem. Biol. 2023, 18, 2082−2093 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles plots using a home-written script in Matlab. The plots were created using GraphPad Prism software. FLIM Imaging. In this study, we conducted fluorescence lifetime imaging (FLIM) experiments on doxycycline (Dox)-inducible eGFP- tagged K-Ras4b G12C HeLa cells. To generate the Dox-inducible cell line, HeLa cells (ATCC #CCL-2) were transduced with lentivirus containing the plasmid construct R733-M42-663 (TRE3Gp > eGFP- Hs.K-Ras4b G12C) at an MOI of 1.0. The cells were cultured in DMEM media supplemented with 10× L-Glutamine, 10% fetal bovine serum (complete media), 4 μg/mL of blastocydin, and 1 μg/mL of puromycin. Prior to imaging, the cell media was replaced with complete media containing doxycycline at a concentration of 500 ng/ mL, and drug treatment was administered at 10 μM for at least 2 h. FLIM imaging was performed using an Olympus Fluoview FV1000 inverted confocal microscope equipped with the Picoquant LSM upgrade kit and Picoharp 300 TCSPC module. A picosecond pulsed diode laser for the green channel (LDH-D-C-485) was used to illuminate the samples at a repetition rate of 40 MHz, allowing us to obtain the fluorescence lifetime decay curve. PicoQuant Symphotime 64 software was utilized for fluorescence lifetime fitting and image analysis. The fluorescence decay curve was fitted to a single- component n-Exponential tailfit to calculate the fluorescence lifetime for each pixel. The color scale on the right represents the fluorescence the mean lifetime of each pixel fluorescence lifetime of eGFP-K-Ras G12C was calculated to be approximately 2.6 ns, as depicted in green within the FLIM images.65,66 in the FLIM image. Notably, EM Spatial Analysis. MDCK cells stably expressing GFP-K- Ras(G12C) or GFP-K-Ras(G12D) were maintained in Dulbecco’s modified Eagle medium (DMEM) containing 10% fetal bovine serum (FBS). Cells were treated with DMSO, C11-MRTX, or MRTX849 at a concentration of 10 μM for 2 h, followed by preparation of the cell PM for electron microscopy (EM) analysis. An EM spatial distribution method is used to quantify the extent of K-Ras protein lateral spatial segregation in the inner leaflet of the PM.26,67 Gold grids with basal PM were prepared as described previously.30,68 Briefly, MDCK cells expressing GFP-tagged K-Ras mutants were grown on pioloform and poly-L-lysine-coated gold EM grids. After treatment, intact basal PM sheets attached to the gold grids were fixed with 4% paraformaldehyde and 0.1% glutaraldehyde, labeled with 4.5 nm gold nanoparticles coupled to anti-GFP antibody, and embedded in methyl cellulose containing 0.3% uranyl acetate. Distribution of gold particles on the basal PM sheets was imaged using a JEOL JEM- 1400 transmission electron microscope at 100,000× magnification. The EM images were analyzed using ImageJ software to assign x and y coordinates to gold particles in a 1 μm2 area of interest on the PM sheets. We use Ripley’s K-function to quantify the gold particle distribution and the extent of nanoclustering eqs A and B. (A) (B) where K(r) indicates the univariate K-function for the number of gold particles (n) within a selected area (A), r is the radius or length scale, ||·|| is the Euclidean distance, the indicator function 1(·) is assigned a −1 is the value of 1 if ||xi − xj|| ≤ r and a value of 0 otherwise, and wij proportion of the circumference of a circle with center at xi and a radius ||xi − xj||. K(r) is linearly transformed to yield a parameter of L(r) − r, which is normalized on the 99% confidence interval (99% C.I.) using Monte Carlo simulations. The maximum value of the L(r) − r function Lmax provides a statistical summary for the extent of nanoclustering. For each treatment condition (DMSO, C11-MRTX, or MRTX849), at least 15 PM sheets were imaged, analyzed, and data pooled. Bootstrap tests were used to calculate the statistical significance of the nanoclustering data, while one-way ANOVA was used to estimate the statistical significance of the gold labeling density as previously described. ■ ASSOCIATED CONTENT *sı Supporting Information The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acschembio.3c00413. Chemical shift perturbation plot; plots of the alpha and beta state signal decay; experimental procedures; and compound characterization by high-resolution mass spectrometry (HRMS) and NMR (PDF) Final video (MP4) ■ AUTHOR INFORMATION Corresponding Authors Johannes Morstein − Department of Cellular and Molecular Pharmacology and Howard Hughes Medical Institute, University of California, San Francisco, California 94158, United States; Email: johannes.morstein@ucsf.edu orcid.org/0000-0002-6940-288X; Kevan M. Shokat − Department of Cellular and Molecular Pharmacology and Howard Hughes Medical Institute, University of California, San Francisco, California 94158, United States; Email: kevan.shokat@ucsf.edu orcid.org/0000-0001-8590-7741; Authors Rebika Shrestha − NCI RAS Initiative, Cancer Research Technology Program, Frederick National Laboratory for Cancer Research, Frederick, Maryland 21701, United States Que N. Van − NCI RAS Initiative, Cancer Research Technology Program, Frederick National Laboratory for Cancer Research, Frederick, Maryland 21701, United States César A. López − Theoretical Biology and Biophysics Group, Los Alamos National Laboratory, Los Alamos, New Mexico 87545, United States; orcid.org/0000-0003-4684-3364 Neha Arora − Department of Integrative Biology and Pharmacology, McGovern Medical School, University of Texas Health Science Center, Houston, Texas 77030, United States Marco Tonelli − National Magnetic Resonance Facility at Madison, Biochemistry Department, University of Wisconsin- Madison, Madison, Wisconsin 53706, United States Hong Liang − Department of Integrative Biology and Pharmacology, McGovern Medical School, University of Texas Health Science Center, Houston, Texas 77030, United States De Chen − NCI RAS Initiative, Cancer Research Technology Program, Frederick National Laboratory for Cancer Research, Frederick, Maryland 21701, United States Yong Zhou − Department of Integrative Biology and Pharmacology, McGovern Medical School, University of Texas Health Science Center, Houston, Texas 77030, United States John F. Hancock − Department of Integrative Biology and Pharmacology, McGovern Medical School, University of Texas Health Science Center, Houston, Texas 77030, United States Andrew G. Stephen − NCI RAS Initiative, Cancer Research Technology Program, Frederick National Laboratory for Cancer Research, Frederick, Maryland 21701, United States Thomas J. Turbyville − NCI RAS Initiative, Cancer Research Technology Program, Frederick National Laboratory for Cancer Research, Frederick, Maryland 21701, United States Complete contact information is available at: 2090 https://doi.org/10.1021/acschembio.3c00413 ACS Chem. Biol. 2023, 18, 2082−2093 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles https://pubs.acs.org/10.1021/acschembio.3c00413 Notes The authors declare the following competing financial interest(s): K.M.S. and J.M. are inventors on patents owned by UCSF covering K-Ras targeting small molecules. K.M.S. has consulting agreements for the following companies, which involve monetary and/or stock compensation: Revolution Medicines, Black Diamond Therapeutics, BridGene Bioscien- ces, Denali Therapeutics, Dice Molecules, eFFECTOR Therapeutics, Erasca, Genentech/Roche, Janssen Pharmaceut- icals, Kumquat Biosciences, Kura Oncology, Mitokinin, Nested, Type6 Therapeutics, Venthera, Wellspring Biosciences (Araxes Pharma), Turning Point, Ikena, Initial Therapeutics, Vevo and BioTheryX. ■ ACKNOWLEDGMENTS J.M. thanks the NCI for a K99/R00 award (K99CA277358). K.M.S. thanks NIH grant 5R01CA244550 and the Samuel Waxman Cancer Research Foundation. The authors thank J. from B. Shoichet’s lab for assistance with the O’Connell dynamic light scattering (DLS) measurement. The authors wish to acknowledge C. J. DeHart, J.-P. Denson, P. H. Frank, M. Hong, S. Messing, A. Mitchell, N. Ramakrishnan, W. for cloning, protein Burgan, K. Powell, and T. Taylor expression, protein purification, cell line production, and electrospray ionization mass spectroscopy. The authors thank J. B. Combs, P. Pfaff, and D. M. Peacock for the critical review of the manuscript. The authors also thank Q. Zheng for providing optimized conditions for the Cbz-deprotection step. This project has been funded in whole or in part with Federal funds from the National Cancer Institute, National Institutes of Health, under Contract No. 75N91019D00024. The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services nor does the mention of trade names, commercial products, or organizations imply endorsement by the U.S. Government. This study made use of the National Magnetic Resonance Facility at Madison, which is supported by NIH grants P41GM136463 and R24GM141526. ■ REFERENCES (1) Alabi, S. B.; Crews, C. M. Major Advances in Targeted Protein Degradation: PROTACs, LYTACs, and MADTACs. J. Biol. Chem. 2021, 296, No. 100647. (2) Békés, M.; Langley, D. R.; Crews, C. M. PROTAC Targeted Protein Degraders: The Past Is Prologue. Nat. Rev. Drug Discovery 2022, 21, 181−200. (3) Schreiber, S. L. The Rise of Molecular Glues. Cell 2021, 184, 3− 9. (4) Kozicka, Z.; Thomä, N. H. Haven’t Got a Glue: Protein Surface Variation for the Design of Molecular Glue Degraders. Cell Chem. Biol. 2021, 28, 1032−1047. (5) Yin, H.; Flynn, A. D. Drugging Membrane Protein Interactions. Annu. Rev. Biomed. Eng. 2016, 18, 51−76. (6) Payandeh, J.; Volgraf, M. Ligand Binding at the Protein−Lipid Interface: Strategic Considerations for Drug Design. Nat. Rev. Drug Discovery 2021, 20, 710−722. (7) Vögler, O.; Barceló, J. M.; Ribas, C.; Escribá, P. V. Membrane Interactions of G Proteins and Other Related Proteins. Biochim. Biophys. Acta, Biomembr. 2008, 1778, 1640−1652. (8) Cho, W.; Stahelin, R. V. Membrane-Protein Interactions in Cell Signaling and Membrane Trafficking. Annu. Rev. Biophys. Biomol. Struct. 2005, 34, 119−151. (9) Zhou, Y.; Hancock, J. F. Ras Nanoclusters: Versatile Lipid-Based Signaling Platforms. Biochim. Biophys. Acta, Mol. Cell Res. 2015, 1853, 841−849. (10) Cox, A. D.; Der, C. J.; Philips, M. R. Targeting RAS Membrane Association: Back to the Future for Anti-RAS Drug Discovery? Clin. Cancer Res. 2015, 21, 1819−1827. (11) Goswami, D.; Chen, D.; Yang, Y.; Gudla, P. R.; Columbus, J.; Worthy, K.; Rigby, M.; Wheeler, M.; Mukhopadhyay, S.; Powell, K.; Burgan, W.; Wall, V.; Esposito, D.; Simanshu, D. K.; Lightstone, F. C.; Nissley, D. V.; McCormick, F.; Turbyville, T. Membrane Interactions of the Globular Domain and the Hypervariable Region of KRAS4b Define Its Unique Diffusion Behavior. eLife 2020, 9, No. e47654. (12) Pass, D. V. M. W. FTase Inhibition Holds Promise for RAS Targeting and Beyond. Cancer, 2018; 90, 2. (13) Ostrem, J. M.; Peters, U.; Sos, M. L.; Wells, J. A.; Shokat, K. M. K-Ras(G12C) Inhibitors Allosterically Control GTP Affinity and Effector Interactions. Nature 2013, 503, 548−551. (14) Ostrem, J. M. L.; Shokat, K. M. Direct Small-Molecule Inhibitors of KRAS: From Structural Insights to Mechanism-Based Design. Nat. Rev. Drug Discovery 2016, 15, 771−785. (15) Janes, M. R.; Zhang, J.; Li, L.-S.; Hansen, R.; Peters, U.; Guo, X.; Chen, Y.; Babbar, A.; Firdaus, S. J.; Darjania, L.; Feng, J.; Chen, J. H.; Li, S.; Li, S.; Long, Y. O.; Thach, C.; Liu, Y.; Zarieh, A.; Ely, T.; Kucharski, J. M.; Kessler, L. V.; Wu, T.; Yu, K.; Wang, Y.; Yao, Y.; Deng, X.; Zarrinkar, P. P.; Brehmer, D.; Dhanak, D.; Lorenzi, M. V.; Hu-Lowe, D.; Patricelli, M. P.; Ren, P.; Liu, Y. Targeting KRAS Mutant Cancers with a Covalent G12C-Specific Inhibitor. Cell 2018, 172, 578−589.e17. (16) Moore, A. R.; Rosenberg, S. C.; McCormick, F.; Malek, S. RAS- Targeted Therapies: Is the Undruggable Drugged? Nat. Rev. Drug Discovery 2020, 19, 533−552. (17) Fell, J. B.; Fischer, J. P.; Baer, B. R.; Blake, J. F.; Bouhana, K.; Briere, D. M.; Brown, K. D.; Burgess, L. E.; Burns, A. C.; Burkard, M. R.; Chiang, H.; Chicarelli, M. J.; Cook, A. W.; Gaudino, J. J.; Hallin, J.; Hanson, L.; Hartley, D. P.; Hicken, E. J.; Hingorani, G. P.; Hinklin, R. J.; Mejia, M. J.; Olson, P.; Otten, J. N.; Rhodes, S. P.; Rodriguez, M. E.; Savechenkov, P.; Smith, D. J.; Sudhakar, N.; Sullivan, F. X.; Tang, T. P.; Vigers, G. P.; Wollenberg, L.; Christensen, J. G.; Marx, M. A. the Clinical Development Candidate MRTX849, a Covalent KRASG12C Inhibitor for the Treatment of Cancer. J. Med. Chem. 2020, 63, 6679−6693. (18) Lanman, B. A.; Allen, J. R.; Allen, J. G.; Amegadzie, A. K.; Ashton, K. S.; Booker, S. K.; Chen, J. J.; Chen, N.; Frohn, M. J.; Goodman, G.; Kopecky, D. J.; Liu, L.; Lopez, P.; Low, J. D.; Ma, V.; Minatti, A. E.; Nguyen, T. T.; Nishimura, N.; Pickrell, A. J.; Reed, A. B.; Shin, Y.; Siegmund, A. C.; Tamayo, N. A.; Tegley, C. M.; Walton, M. C.; Wang, H.-L.; Wurz, R. P.; Xue, M.; Yang, K. C.; Achanta, P.; Bartberger, M. D.; Canon, J.; Hollis, L. S.; McCarter, J. D.; Mohr, C.; Rex, K.; Saiki, A. Y.; Miguel, T. S.; Volak, L. P.; Wang, K. H.; Whittington, D. A.; Zech, S. G.; Lipford, J. R.; Cee, V. J. Discovery of a Covalent Inhibitor of KRASG12C (AMG 510) for the Treatment of Solid Tumors. J. Med. Chem. 2020, 63, 52−65. (19) Zhang, Z.; Guiley, K. Z.; Shokat, K. M. Chemical Acylation of an Acquired Serine Suppresses Oncogenic Signaling of K-Ras(G12S). Nat. Chem. Biol. 2022, 18, 1177−1183. (20) Zhang, Z.; Morstein, J.; Ecker, A. K.; Guiley, K. Z.; Shokat, K. M. Chemoselective Covalent Modification of K-Ras(G12R) with a Small Molecule Electrophile. J. Am. Chem. Soc. 2022, 144, 15916− 15921. (21) Wang, X.; Allen, S.; Blake, J. F.; Bowcut, V.; Briere, D. M.; Calinisan, A.; Dahlke, J. R.; Fell, J. B.; Fischer, J. P.; Gunn, R. J.; Hallin, J.; Laguer, J.; Lawson, J. D.; Medwid, J.; Newhouse, B.; J. M.; Olson, P.; Pajk, S.; Rahbaek, L.; Nguyen, P.; O’Leary, Rodriguez, M.; Smith, C. R.; Tang, T. P.; Thomas, N. C.; Vanderpool, D.; Vigers, G. P.; Christensen, J. G.; Marx, M. A. Identification of MRTX1133, a Noncovalent, Potent, and Selective KRASG12D Inhibitor. J. Med. Chem. 2022, 65, 3123−3133. (22) Vasta, J. D.; Peacock, D. M.; Zheng, Q.; Walker, J. A.; Zhang, Z.; Zimprich, C. A.; Thomas, M. R.; Beck, M. T.; Binkowski, B. F.; Identification of 2091 https://doi.org/10.1021/acschembio.3c00413 ACS Chem. Biol. 2023, 18, 2082−2093 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles Corona, C. R.; Robers, M. B.; Shokat, K. M. KRAS Is Vulnerable to Reversible Switch-II Pocket Engagement in Cells. Nat. Chem. Biol. 2022, 18, 596−604. (23) Zhou, Y.; Hancock, J. F. RAS Nanoclusters Are Cell Surface Transducers That Convert Extracellular Stimuli to Intracellular Signalling. FEBS Lett. 2023, 597, 892−908. (24) Simanshu, D. K.; Philips, M. R.; Hancock, J. F. Consensus on the RAS Dimerization Hypothesis: Strong Evidence for Lipid- Mediated Clustering but Not for G-Domain-Mediated Interactions. Mol. Cell 2023, 83, 1210−1215. (25) Hancock, J. F. Ras Proteins: Different Signals from Different Locations. Nat. Rev. Mol. Cell Biol. 2003, 4, 373−385. (26) Prior, I. A.; Muncke, C.; Parton, R. G.; Hancock, J. F. Direct Visualization of Ras Proteins in Spatially Distinct Cell Surface Microdomains. J. Cell Biol. 2003, 160, 165−170. (27) Plowman, S. J.; Muncke, C.; Parton, R. G.; Hancock, J. F. H- Ras, K-Ras, and Inner Plasma Membrane Raft Proteins Operate in Nanoclusters with Differential Dependence on the Actin Cytoskele- ton. Proc. Natl. Acad. Sci. U.S.A. 2005, 102, 15500−15505. (28) Shrestha, R.; Chen, D.; Frank, P.; Nissley, D. V.; Turbyville, T. J. Recapitulation of Cell-like KRAS4b Membrane Dynamics on Complex Biomimetic Membranes. iScience 2022, 25, No. 103608. (29) Lee, Y.; Phelps, C.; Huang, T.; Mostofian, B.; Wu, L.; Zhang, Y.; Tao, K.; Chang, Y. H.; Stork, P. J.; Gray, J. W.; Zuckerman, D. M.; Nan, X. High-Throughput, Single-Particle Tracking Reveals Nested Membrane Domains That Dictate KRasG12D Diffusion and Trafficking. eLife 2019, 8, No. e46393. (30) Zhou, Y.; Prakash, P.; Liang, H.; Cho, K.-J.; Gorfe, A. A.; Hancock, J. F. Lipid-Sorting Specificity Encoded in K-Ras Membrane Anchor Regulates Signal Output. Cell 2017, 168, 239−251.e16. (31) Zhou, Y.; Wong, C.-O.; Cho, K.; van der Hoeven, D.; Liang, H.; Thakur, D. P.; Luo, J.; Babic, M.; Zinsmaier, K. E.; Zhu, M. X.; Hu, H.; Venkatachalam, K.; Hancock, J. F. Membrane Potential Modulates Plasma Membrane Phospholipid Dynamics and K-Ras Signaling. Science 2015, 349, 873−876. (32) Zhou, Y.; Liang, H.; Rodkey, T.; Ariotti, N.; Parton, R. G.; Hancock, J. F. Signal Integration by Lipid-Mediated Spatial Cross Talk between Ras Nanoclusters. Mol. Cell. Biol. 2014, 34, 862−876. (33) Singaram, I.; Sharma, A.; Pant, S.; Lihan, M.; Park, M.-J.; Pergande, M.; Buwaneka, P.; Hu, Y.; Mahmud, N.; Kim, Y.-M.; Cologna, S.; Gevorgyan, V.; Khan, I.; Tajkhorshid, E.; Cho, W. Targeting Lipid−Protein Interaction to Treat Syk-Mediated Acute Myeloid Leukemia. Nat. Chem. Biol. 2023, 19, 239−250. (34) Katti, S. S.; Krieger, I. V.; Ann, J.; Lee, J.; Sacchettini, J. C.; Igumenova, T. I. Structural Anatomy of Protein Kinase C C1 Domain Interactions with Diacylglycerol and Other Agonists. Nat. Commun. 2022, 13, No. 2695. (35) Fang, Z.; Marshall, C. B.; Nishikawa, T.; Gossert, A. D.; Jansen, J. M.; Jahnke, W.; Ikura, M. Inhibition of K-RAS4B by a Unique Mechanism of Action: Stabilizing Membrane-Dependent Occlusion of the Effector-Binding Site. Cell Chem. Biol. 2018, 25, 1327−1336.e4. (36) Bond, M. J.; Chu, L.; Nalawansha, D. A.; Li, K.; Crews, C. M. Targeted Degradation of Oncogenic KRASG12C by VHL-Recruiting PROTACs. ACS Cent. Sci. 2020, 6, 1367−1375. (37) Chu, L.; Crews, C. M.; Dong, H.; Hornberger, K. R.; Medina, J. R.; Snyder, L.; Wang, J. Compounds and Methods for Targeted Degradation of Kras. WO Patent, WO2021207172A12021. (38) Presolski, S. I.; Hong, V. P.; Finn, M. G. Copper-Catalyzed Azide−Alkyne Click Chemistry for Bioconjugation. Curr. Protoc. Chem. Biol. 2011, 3, 153−162. (39) Thirumurugan, P.; Matosiuk, D.; Jozwiak, K. Click Chemistry for Drug Development and Diverse Chemical−Biology Applications. Chem. Rev. 2013, 113, 4905−4979. (40) Lou, K.; Steri, V.; Ge, A. Y.; Hwang, Y. C.; Yogodzinski, C. H.; Shkedi, A. R.; Choi, A. L. M.; Mitchell, D. C.; Swaney, D. L.; Hann, B.; Gordan, J. D.; Shokat, K. M.; Gilbert, L. A. KRASG12C Inhibition Produces a Driver-Limited State Revealing Collateral Dependencies. Sci. Signal. 2019, 12, No. eaaw9450. (41) Schmick, M.; Vartak, N.; Papke, B.; Kovacevic, M.; et al. KRas Localizes to the Plasma Membrane by Spatial Cycles of Solubilization, Trapping and Vesicular Transport. Cell 2014, 157, 459−471. (42) Kattan, W. E.; Hancock, J. F. RAS Function in Cancer Cells: Translating Membrane Biology and Biochemistry into New Therapeutics. Biochem. J. 2020, 477, 2893−2919. (43) Zhou, Y.; Hancock, J. F. Electron Microscopy Combined with Spatial Analysis: Quantitative Mapping of the Nano-Assemblies of Plasma Membrane-Associating Proteins and Lipids. Biophys. Rep. 2018, 4, 320−328. (44) Morstein, J.; Capecchi, A.; Hinnah, K.; Park, B.; Petit-Jacques, J.; Van Lehn, R. C.; Reymond, J.-L.; Trauner, D. Medium-Chain Lipid J. Am. Conjugation Facilitates Cell-Permeability and Bioactivity. Chem. Soc. 2022, 144, 18532−18544. (45) Eswar, N.; Webb, B.; Marti-Renom, M. A.; Madhusudhan, M. S.; Eramian, D.; Shen, M.-Y.; Pieper, U.; Sali, A. Comparative Protein Structure Modeling Using Modeller Curr. Protoc. Bioinf. 2006, 15, DOI: 10.1002/0471250953.bi0506s15. (46) Ingólfsson, H. I.; Neale, C.; Carpenter, T. S.; Shrestha, R.; López, C. A.; Tran, T. H.; Oppelstrup, T.; Bhatia, H.; Stanton, L. G.; Zhang, X.; Sundram, S.; Di Natale, F.; Agarwal, A.; Dharuman, G.; Kokkila Schumacher, S. I. L.; Turbyville, T.; Gulten, G.; Van, Q. N.; Goswami, D.; Jean-Francois, F.; Agamasu, C.; Chen, D.; Hettige, J. J.; Travers, T.; Sarkar, S.; Surh, M. P.; Yang, Y.; Moody, A.; Liu, S.; Van Essen, B. C.; Voter, A. F.; Ramanathan, A.; Hengartner, N. W.; Simanshu, D. K.; Stephen, A. G.; Bremer, P.-T.; Gnanakaran, S.; Glosli, J. N.; Lightstone, F. C.; McCormick, F.; Nissley, D. V.; Streitz, F. H. Machine Learning−Driven Multiscale Modeling Reveals Lipid- Dependent Dynamics of RAS Signaling Proteins. Proc. Natl. Acad. Sci. U.S.A. 2022, 119, No. e2113297119. (47) Souza, P. C. T.; Alessandri, R.; Barnoud, J.; Thallmair, S.; Faustino, I.; Grünewald, F.; Patmanidis, I.; Abdizadeh, H.; Bruininks, B. M. H.; Wassenaar, T. A.; Kroon, P. C.; Melcr, J.; Nieto, V.; Corradi, V.; Khan, H. M.; Domański, J.; Javanainen, M.; Martinez- Seara, H.; Reuter, N.; Best, R. B.; Vattulainen, I.; Monticelli, L.; Periole, X.; Tieleman, D. P.; de Vries, A. H.; Marrink, S. J. Martini 3: A General Purpose Force Field for Coarse-Grained Molecular Dynamics. Nat. Methods 2021, 18, 382−388. (48) Poma, A. B.; Cieplak, M.; Theodorakis, P. E. Combining the MARTINI and Structure-Based Coarse-Grained Approaches for the Molecular Dynamics Studies of Conformational Transitions in Proteins. J. Chem. Theory Comput. 2017, 13, 1366−1374. (49) Travers, T.; López, C. A.; Van, Q. N.; Neale, C.; Tonelli, M.; Stephen, A. G.; Gnanakaran, S. Molecular Recognition of RAS/RAF Complex at the Membrane: Role of RAF Cysteine-Rich Domain. Sci. Rep. 2018, 8, No. 8461. (50) Potter, T. D.; Barrett, E. L.; Miller, M. A. Automated Coarse- the Martini Force Field and Grained Mapping Algorithm for Benchmarks for Membrane-Water Partitioning. J. Chem. Theory Comput. 2021, 17, 5777−5791. (51) Wassenaar, T. A.; Ingólfsson, H. I.; Böckmann, R. A.; Tieleman, D. P.; Marrink, S. J. Computational Lipidomics with Insane: A Versatile Tool for Generating Custom Membranes for Molecular Simulations. J. Chem. Theory Comput. 2015, 11, 2144−2155. (52) Páll, S.; Abraham, M. J.; Kutzner, C.; Hess, B.; Lindahl, E. Tackling Exascale Software Challenges in Molecular Dynamics Simulations with GROMACS. In Solving Software Challenges for Exascale; Markidis, S.; Laure, E., Eds.; Springer International Publishing: Cham, 2015; pp 3−27. (53) Bussi, G.; Donadio, D.; Parrinello, M. Canonical Sampling through Velocity Rescaling. J. Chem. Phys. 2007, 126, No. 014101. (54) Berendsen, H. J. C.; Postma, J. P. M.; van Gunsteren, W. F.; DiNola, A.; Haak, J. R. Molecular Dynamics with Coupling to an External Bath. J. Chem. Phys. 1984, 81, 3684−3690. (55) Hagn, F.; Etzkorn, M.; Raschle, T.; Wagner, G. Optimized Phospholipid Bilayer Nanodiscs Facilitate High-Resolution Structure Determination of Membrane Proteins. J. Am. Chem. Soc. 2013, 135, 1919−1925. 2092 https://doi.org/10.1021/acschembio.3c00413 ACS Chem. Biol. 2023, 18, 2082−2093 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles (56) Chao, F.-A.; Dharmaiah, S.; Taylor, T.; Messing, S.; Gillette, W.; Esposito, D.; Nissley, D. V.; McCormick, F.; Byrd, R. A.; Simanshu, D. K.; Cornilescu, G. Insights into the Cross Talk between Effector and Allosteric Lobes of KRAS from Methyl Conformational Dynamics. J. Am. Chem. Soc. 2022, 144, 4196−4205. (57) Kopra, K.; Vuorinen, E.; Abreu-Blanco, M.; Wang, Q.; Eskonen, V.; Gillette, W.; Pulliainen, A. T.; Holderfield, M.; Härmä, H. Homogeneous Dual-Parametric-Coupled Assay for Simultaneous Nucleotide Exchange and KRAS/RAF-RBD Interaction Monitoring. Anal. Chem. 2020, 92, 4971−4979. (58) Van, Q. N.; López, C. A.; Tonelli, M.; Taylor, T.; Niu, B.; Stanley, C. B.; Bhowmik, D.; Tran, T. H.; Frank, P. H.; Messing, S.; Alexander, P.; Scott, D.; Ye, X.; Drew, M.; Chertov, O.; Lösche, M.; Ramanathan, A.; Gross, M. L.; Hengartner, N. W.; Westler, W. M.; Markley, J. L.; Simanshu, D. K.; Nissley, D. V.; Gillette, W. K.; Esposito, D.; McCormick, F.; Gnanakaran, S.; Heinrich, F.; Stephen, A. G. Uncovering a Membrane-Distal Conformation of KRAS Available to Recruit RAF to the Plasma Membrane. Proc. Natl. Acad. Sci. U.S.A. 2020, 117, 24258−24268. (59) Delaglio, F.; Grzesiek, S.; Vuister, G. W.; Zhu, G.; Pfeifer, J.; Bax, A. NMRPipe: A Multidimensional Spectral Processing System Based on UNIX Pipes. J. Biomol. NMR 1995, 6, 277−293. (60) Lee, W.; Tonelli, M.; Markley, J. L. NMRFAM-SPARKY: Enhanced Software for Biomolecular NMR Spectroscopy. Bioinfor- matics 2015, 31, 1325−1327. (61) Lee, D.; Hilty, C.; Wider, G.; Wüthrich, K. Effective Rotational Correlation Times of Proteins from NMR Relaxation Interference. J. Magn. Reson. 2006, 178, 72−76. (62) Coote, P. W.; Robson, S. A.; Dubey, A.; Boeszoermenyi, A.; Zhao, M.; Wagner, G.; Arthanari, H. Optimal Control Theory Enables Homonuclear Decoupling without Bloch-Siegert Shifts in NMR Spectroscopy. Nat. Commun. 2018, 9, No. 3014. (63) Grimm, J. B.; Muthusamy, A. K.; Liang, Y.; Brown, T. A.; Lemon, W. C.; Patel, R.; Lu, R.; Macklin, J. J.; Keller, P. J.; Ji, N.; Lavis, L. D. A General Method to Fine-Tune Fluorophores for Live- Cell and in Vivo Imaging. Nat. Methods 2017, 14, 987−994. (64) Dedecker, P.; Duwé, S.; Neely, R. K.; Zhang, J. Localizer: Fast, Accurate, Open-Source, and Modular Software Package for Super- resolution Microscopy. J. Biomed. Opt. 2012, 17, No. 126008. (65) Pliss, A.; Zhao, L.; Ohulchanskyy, T. Y.; Qu, J.; Prasad, P. N. Fluorescence Lifetime of Fluorescent Proteins as an Intracellular Environment Probe Sensing the Cell Cycle Progression. ACS Chem. Biol. 2012, 7, 1385−1392. J. P. Shifts in the (66) Li, W.; Houston, K. D.; Houston, Fluorescence Lifetime of EGFP during Bacterial Phagocytosis Measured by Phase-Sensitive Flow Cytometry. Sci. Rep. 2017, 7, No. 40341. (67) Diggle, P. J.; Mateu, J.; Clough, H. E. A Comparison between Parametric and Non-Parametric Approaches to the Analysis of Replicated Spatial Point Patterns. Adv. Appl. Probab. 2000, 32, 331−343. (68) Zhou, Y.; Hancock, J. F.Super-Resolution Imaging and Spatial Analysis of RAS on Intact Plasma Membrane Sheets. In Methods in Molecular Biology; Springer, 2021; Vol. 2262, pp 217−232. 2093 https://doi.org/10.1021/acschembio.3c00413 ACS Chem. Biol. 2023, 18, 2082−2093
null
10.1038/s41431-022-01214-7
null
null
www.nature.com/ejhg OPEN POLICY Scope of professional roles for genetic counsellors and clinical geneticists in the United Kingdom Position on behalf of the Association of Genetic Nurses and Counsellors and the Clinical Genetics Society Anna Middleton 1,2 ✉ 3, Catherine Houghton4, Sarah Smithson5,6, Meena Balasubramanian7,8 and Frances Elmslie9 , Nicola Taverner © The Author(s) 2022 This document is written on behalf of the two professional bodies in the United Kingdom that represent genetic counsellors (the Association of Genetic Nurses and Counsellors) and clinical geneticists (the Clinical Genetics Society) and aims to support multidisciplinary working of these professional groups highlighting within a quick-reference format, areas of shared practice and the distinctions between role profiles for a Consultant Clinical Geneticist, Principal/Consultant Genetic Counsellor and the new support role that we have termed ‘Genomic Associate’, see AGNC career structure [1]. This builds on published documents that articulate the scope of practice of the clinical genetics workforce [2] and specifically the genetic counsellor [3] and clinical geneticist [4] roles. European Journal of Human Genetics (2023) 31:9–12; https://doi.org/10.1038/s41431-022-01214-7 : , ; ) ( 0 9 8 7 6 5 4 3 2 1 In the United Kingdom clinical geneticists are medically qualified Members/Fellows of the Royal College Physicians or equivalent, where Clinical Genetics is an affiliated medical specialty. Genomic or genetic counsellors are allied health professionals with Masters level accreditation from the Genetic Counsellor Registration Board included in the Academy for Healthcare Science register and clinical scientists (genomic counselling specialty) accredited by the Health and Care Professions Council. We acknowledge there is currently variability in these roles between NHS trusts and exceptions where the scope of practice for one professional group exceeds what is provided below in Fig. 1. i.e. they acknowledge that there are some areas of practice that may have In Fig. 1 the roles are deliberately forward looking, traditionally been performed by one professional group, can now be shared with or devolved to other groups. Broadly speaking, the clinical geneticist leads on diagnostics and therapeutics and the genetic counsellor leads on psychosocial issues and care of the extended family. Both groups have skills and training in clinical genetics and there is much cross over between roles. The genomic associate leads on administrative support for the clinic, the patient and the clinical activities of the clinical geneticist and genetic counsellor. The genomic associate is part of the genetic counsellor career structure and has a clinical role that is different to a secretary; it is a position that has already been discussed in relation to the Genomics Service Specification for the National Health Service in the United Kingdom. 1Engagement and Society, Wellcome Connecting Science, Wellcome Genome Campus, Hinxton, Cambridge, UK. 2Kavli Centre for Ethics, Science, and the Public, Faculty of Education, University of Cambridge, Cambridge, UK. 3School of Medicine, Cardiff University and the All Wales Medical Genomics Service, Cardiff, UK. 4Liverpool Centre for Genomic Medicine, Liverpool Women’s NHS Foundation Trust, Liverpool, UK. 5Department of Clinical Genetics, University Hospitals Bristol and Weston NHS Foundation Trust, Bristol, UK. 6Faculty of Health Sciences, University of Bristol, Bristol, UK. 7Department of Oncology and Metabolism, University of Sheffield, Sheffield, UK. 8Sheffield Clinical Genetics Service, Sheffield Children’s NHS Foundation Trust, Sheffield, UK. 9South West Thames Centre for Genomics, St George’s University Hospitals NHS Foundation Trust, London, UK. email: Am2624@cam.ac.uk ✉ Received: 28 September 2022 Accepted: 4 October 2022 Published online: 1 November 2022 10 A. Middleton et al. Clinical Geneticist Genetic Counsellor Genomic Associate Triaging referrals Referrals are assessed and triaged Advice and guidance letters for refused referrals Access to the appointment Preparation for appointment Patient/family type seen Letters written in response to referrals that require clinical advice, but do not meet Genomic Medicine Service referral guidelines Responsibility for responding to referrals that do not require any clinical advice nor clinical contact Facilitating patient access, including establishing if patient wants to be seen, supporting minority populations to access services, supporting patients with disabilities/audio/visual impairment to access services, contacting patients to explain what clinical genetics services can offer, arranging interpreters Acting as a chaperone in clinic Arranging measurements for patients in clinic, e.g. taking patient’s weight and height Transcribing a written pedigree into electronic software Gathering relevant medical records, pathology reports, death certificates, tumour blocks Organising and obtaining familial blood or saliva samples to help confirm diagnosis in proband Obtaining record of patient choice/consent (not having the full consent conversation, but recording that it has been taken) Collating appropriate patient leaflets, consent forms for the clinic as determined by senior staff General genetics (adult or paediatric) Cancer genetics (adult or paediatric) Prenatal Physical medical examination Family history Physical examination of a patient to make a clinical diagnosis and/or to support or stratify genetic testing Specific physical examination that might be considered routine with respect to particular conditions (e.g. head measurement for a Cowden’s clinic) Taking a family history Psychosocial history Investigations Evaluating a family history to determine genetic risk Taking a detailed psychosocial history to determine effect of genetic diagnosis on individual and wider family members Medical investigations: Employ a range of tailored investigations including genetic, biochemistry, radiology, haematology etc. for clinically undiagnosed patients Routine medical investigations for specific, defined conditions, e.g. ophthalmology or audiological investigations as part of conditions involving visual and/or hearing impairment Genetic investigations: Choose appropriate genetic testing for patients with specific family history indicative of genetic risk (e.g. family history of cancer) Genetic investigations: Choose appropriate genetic testing determined by a pre - existing definitive clinical diagnosis/clinical presentation (e.g. breast cancer) Take samples (e.g. blood saliva) for genetic testing Consent Consent a patient for genetic testing Counselling and support Genomic variant interpretation Arrange and consent for cascade genetic testing amongst extended family (e.g. BRCA, Fra-X testing) Generic genetic counselling skills e.g. disclosure of diagnosis, breaking bad news etc. Supporting patients and families adjusting to a genetic diagnosis or coping without one Making appropriate onward referrals for further psychological support Identifying complex grief reactions and interpreting complex family dynamics Specific application of genetic counselling theory to person -centred care, e.g. application of reciprocal-engagement models and/or reflective practice models Interpreting gene variants to determine clinical decisions, as part of a multidisciplinary team Integrating the results of clinical presentation and investigation to determine whether a rare phenotype supports variant pathogenicity Interpreting whether an established clinical presentation supports variant pathogenicity Administration to track down relatives to provide evidence in support of variant interpretation Fig. 1 Scope of professional roles for clinical geneticist, genetic counsellor and genomic associate in the United Kingdom. The colour coding provides a guide to the professional group providing each aspect of service: green = routinely within the scope of practice, amber = within the scope of practice for some professionals, but not for the majority, red = outside of the scope of routine practice. European Journal of Human Genetics (2023) 31:9 – 12 A. Middleton et al. 11 Management and Treatment Reviewing and recommending peer-reviewed management guidelines. Writing, e.g. NICE guidance Organising appropriate disease screening and acting as patient advocate to arrange access to services Devising individual management guidelines for a rare disorder based on research evidence Prescribing pharmaceuticals or molecularly targeted therapies MDT coordination, collating agenda items, taking meeting minutes All administration required for clinic and follow up Ordering of clinic supplies, test kits, appropriate proformas, consent forms Follow Up Follow up care of the nuclear family (e.g. parents and children) Follow up care of the extended family (e.g. 2nd and 3rd degree relatives) Research Monitoring/chasing outstanding records/samples/screening and any administration work needed to support the clinical geneticists and genetic counsellors Leading or referring to research studies relating to patient’s genetic diagnosis Leading or being a site investigator for Clinical Trials of Investigational Medical Products Finding and referring to surveillance trials (e.g. for cancer screening) Referring to psychosocial research (e.g. genetic counselling or communication research) Leading genetic counselling research specifically on the evidence base behind genetic counselling practice Administration for research studies Mainstreaming Providing advice and support to other healthcare workers Education Participation in multi-disciplinary team meetings Managing and leading a specialist nurse mainstreaming team (e.g. familial hypercholesterolemia clinic, family breast screening clinics, pre - implantation genetic diagnosis within an IVF clinic) Delivering education programmes for patients, public, health professionals Developing educational material such as leaflets, interactive infographics and decision aids Liaising with patient support groups to participate in patient led events and sharing of verified information Administration for education events Management Running a genetic register Training and mentoring colleagues from genetics services Training, mentoring and supporting non-genetics healthcare colleagues Acting as Clinical Lead for a clinical genetics service Acting as Management Lead for clinical genetics service Leadership Sitting on regulatory bodies for own profession Designing professional competency-to-practice frameworks Fig. 1 Continued. REFERENCES 1. Association of Genetic Nurses and Counsellors. Career structure for genetic counsellors and support roles. 2020b. https://www.agnc.org.uk/info-education/ documents-websites/. 2. Dragojlovic N, Borle K, Kopac N, Ellis U, Birch P, Adam S, et al. The composition and capacity of the clinical genetics workforce in high-income countries: a scoping review. Genet Med. 2020;22:1437–49. 3. Middleton A. et al. The genetic counsellor role in the United Kingdom: Position on behalf of the Association of Genetic Nurses and Counsellors (AGNC), Endorsed by the Genetic Counsellor Registration Board (GCRB) and Academy for Healthcare Science (AHCS). Eur J Hum Genet. https://doi.org/10.1038/s41431-022-01212-9 (2022). 4. Clinical Genetics Society. Policies and Resources around the role of the clinical geneticist. 2022. https://www.clingensoc.org/information-and-education/policies- and-resources/. Committee on Genomics in Medicine in the UK and we thank Prof Helen Firth for starting these discussions. The two professional bodies representing genetic counsellors (Association of Genetic Nurses and Counsellors) and clinical geneticists (Clinical Genetics Society) led on the development of the conceptual framework. Consensus was reached on the scope of professional practice through discussion with members of the committee representing each professional body. The negotiations were led by AM, NT and CH on behalf of the AGNC and led by SS, FE and MB on behalf of the CGS. AUTHOR CONTRIBUTIONS AM, NT, CH, SS, and FE all contributed equally to the writing of the manuscript, MB provided feedback on the manuscript. ACKNOWLEDGEMENTS The original idea for developing a quick reference format for visualising the different roles of genetic counsellor and clinical geneticist came from discussions within the Joint FUNDING AM was funded by Wellcome grant 108413/A/15/D awarded to Wellcome Connecting Science and grant G115418 from the Kavli Foundation to the Kavli Centre for Ethics, Science and the Public, University of Cambridge. European Journal of Human Genetics (2023) 31:9 – 12 12 COMPETING INTERESTS The authors declare no competing interests. A. Middleton et al. ADDITIONAL INFORMATION Correspondence and requests for materials should be addressed to Anna Middleton. Reprints and permission information is available at http://www.nature.com/ reprints Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of license, visit http:// creativecommons.org/licenses/by/4.0/. this Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. © The Author(s) 2022 European Journal of Human Genetics (2023) 31:9 – 12
null
10.3390_molecules24122333.pdf
null
null
Article Development of a Method for the Quantification of Clotrimazole and Itraconazole and Study of Their Stability in a New Microemulsion for the Treatment of Sporotrichosis Patricia Garcia Ferreira 1, Carolina Guimarães de Souza Lima 2, Letícia Lorena Noronha 1, Marcela Cristina de Moraes 2, Fernando de Carvalho da Silva 2 Débora Omena Futuro 1 and Vitor Francisco Ferreira 1,* , Alessandra Lifsitch Viçosa 3 , 1 Departamento de Tecnologia Farmacêutica, Faculdade de Farmácia, Universidade Federal Fluminense, Niterói-RJ 24241-000, Brazil; patricia.pharma@yahoo.com.br (P.G.F.); leticianoronha95@gmail.com (L.L.N.); dfuturo@id.uff.br (D.O.F.) 2 Departamento de Química Orgânica, Instituto de Química, Universidade Federal Fluminense, Niterói-RJ 24210-141, Brazil; carolgslima@gmail.com (C.G.d.S.L.); mcmoraes@id.uff.br (M.C.d.M.); gqofernando@vm.uff.br (F.d.C.d.S.) Fundação Oswaldo Cruz (FIOCRUZ), Farmanguinhos-Manguinhos, Avenida Sinzenando Nabuco 100, Rio de Janeiro-RJ 21045-900, Brazil; alessandra.vicosa@far.fiocruz.br 3 * Correspondence: vitorferreira@id.uff.br; Tel.: +55-21-998578148 Academic Editors: Clinio Locatelli, Marcello Locatelli and Dora Melucci Received: 6 June 2019; Accepted: 20 June 2019; Published: 25 June 2019 ® Abstract: Sporotrichosis occurs worldwide and is caused by the fungus Sporothrix brasiliensis. This agent has a high zoonotic potential and is transmitted mainly by bites and scratches from infected felines. A new association between the drugs clotrimazole and itraconazole is shown to be effective against S. brasiliensis yeasts. This association was formulated as a microemulsion containing benzyl alcohol as oil, Tween 60 and propylene glycol as surfactant and cosurfactant, respectively, and water. Initially, the compatibility between clotrimazole and itraconazole was studied using differential scanning calorimetry (DSC), thermogravimetric analysis (TG), Fourier transform infrared spectroscopy (FTIR), and X-ray powder diffraction (PXRD). Additionally, a simple and efficient analytical HPLC method was developed to simultaneously determine the concentration of clotrimazole and itraconazole in the novel microemulsion. The developed method proved to be efficient, robust, and reproducible for both components of the microemulsion. We also performed an accelerated stability study of this formulation, and the developed analytical method was applied to monitor the content of active ingredients. Interestingly, these investigations led to the detection of a known clotrimazole degradation product whose structure was confirmed using NMR and HRMS, as well as a possible interaction between itraconazole and benzyl alcohol. Keywords: validation; sporotrichosis pre-development process; clotrimazole; itraconazole; stability; method 1. Introduction Sporotrichosis is a subcutaneous infectious disease with subacute to chronic evolution and with a worldwide distribution. The etiologic agent of sporotrichosis is Sporothrix schenckii, which is a thermo-dimorphic fungus that lives saprophytically in nature and is pathogenic to humans and animals [1,2]. The occurrence of sporotrichosis in animals, especially cats, as well as its transmission to humans has been reported in several countries [3]. In this context, the Brazilian state of Rio de Janeiro Molecules 2019, 24, 2333; doi:10.3390/molecules24122333 www.mdpi.com/journal/molecules molecules(cid:1)(cid:2)(cid:3)(cid:1)(cid:4)(cid:5)(cid:6)(cid:7)(cid:8)(cid:1)(cid:1)(cid:2)(cid:3)(cid:4)(cid:5)(cid:6)(cid:7) Molecules 2019, 24, 2333 2 of 15 is an epidemic area for this disease and the first one associated with zoonotic transmission related to sick felines by Sporothrix brasiliensis, the most virulent species from the S. schenckii complex [4]. The treatment of both feline and human sporotrichosis is based on the use of itraconazole 1, which contains the 1,2,4-triazole scaffold in its structure and inhibits the synthesis of sterol, a vital component of the fungus cell membrane [5,6]. Clotrimazole 2, on the other hand, is an imidazole derivative with antifungal activity that is only indicated for topical use due to its toxicity (Figure 1). Similarly to itraconazole, clotrimazole is a synthetic antifungal and its mechanism of action involves the inhibition of sterol biosynthesis [7]. In this sense, Gagini et al. [8] reported the effectiveness of the combination of itraconazole with clotrimazole against S. brasiliensis yeasts (the infective form) from feline and human sporotrichosis isolates, suggesting that clotrimazole by itself or in combination with itraconazole is potentially a new option for the treatment of sporotrichosis. Figure 1. Chemical structures of clotrimazole and itraconazole. Accordingly, the development of new pharmaceutical technologies for the use of clotrimazole and itraconazole associations is highly desirable in order to increase their efficiency in therapy, decrease adverse effects and provide, especially for felines, alternative treatments. Moreover, the use of a combination antifungal therapy is a promising approach to avoid resistance [9]. Allied to all the mentioned features, the association of known drugs is highly advantageous for the pharmaceutical industry to find innovations for the market, since they can reformulate their products in a more economically advantageous way when compared to the development of new drugs. In addition, the association of drugs already in use in the pharmaceutical market may increase their efficiency with known safety and effectiveness, reintroducing forgotten and/or discarded ones. Considering the development of new formulations, microemulsions (MEs) have attracted great interest as potential drug delivery systems, mainly due to their unique physicochemical properties such as drug solubilization and enhanced absorption properties [10,11]. MEs are a thermodynamically stable, isotropic, transparent liquid system consisting of two immiscible liquids (usually water and oil) stabilized by a film of surfactant compounds, suitably combined with a cosurfactant [12,13]. The presence of the surfactant helps to reduce the interfacial tension, making it possible to join the oil and aqueous phases [14,15]. MEs have been proposed as an innovative formulation approach to improve solubility and efficacy and reduce of the toxicity of various drugs. Therefore, when the known hydrophobicity of clotrimazole and itraconazole are taken into account, such systems could be particularly advantageous for their delivery. In light of the aforementioned concepts, this paper reports the initial research phase for the pre-development of a clotrimazole–itraconazole formulation, the first step towards a new antifungal combination. In this sense, the development and characterization of this new pharmaceutical formulation requires the evaluation of parameters such as drug release and stability. Therefore, as a further extension of our work in the field, we have developed a simple, sensitive, and specific HPLC method for the simultaneous quantification of clotrimazole and itraconazole in microemulsion. Although many researchers have investigated clotrimazole and itraconazole singly or in combination with other compounds, to the best of our knowledge, no HPLC method has been developed for the simultaneous determination of both drugs simultaneously, especially in microemulsion systems [16,17]. Finally, we performed an accelerated stability study of this formulation and the developed analytical Molecules 2019, 24, x FOR PEER REVIEW 2 of 15 to humans has been reported in several countries [3]. In this context, the Brazilian state of Rio de Janeiro is an epidemic area for this disease and the first one associated with zoonotic transmission related to sick felines by Sporothrix brasiliensis, the most virulent species from the S. schenckii complex [4]. The treatment of both feline and human sporotrichosis is based on the use of itraconazole 1, which contains the 1,2,4-triazole scaffold in its structure and inhibits the synthesis of sterol, a vital component of the fungus cell membrane [5,6]. Clotrimazole 2, on the other hand, is an imidazole derivative with antifungal activity that is only indicated for topical use due to its toxicity (Figure 1). Similarly to itraconazole, clotrimazole is a synthetic antifungal and its mechanism of action involves the inhibition of sterol biosynthesis [7]. In this sense, Gagini et al. [8] reported the effectiveness of the combination of itraconazole with clotrimazole against S. brasiliensis yeasts (the infective form) from feline and human sporotrichosis isolates, suggesting that clotrimazole by itself or in combination with itraconazole is potentially a new option for the treatment of sporotrichosis. Accordingly, the development of new pharmaceutical technologies for the use of clotrimazole and itraconazole associations is highly desirable in order to increase their efficiency in therapy, decrease adverse effects and provide, especially for felines, alternative treatments. Moreover, the use of a combination antifungal therapy is a promising approach to avoid resistance [9]. Allied to all the mentioned features, the association of known drugs is highly advantageous for the pharmaceutical industry to find innovations for the market, since they can reformulate their products in a more economically advantageous way when compared to the development of new drugs. In addition, the association of drugs already in use in the pharmaceutical market may increase their efficiency with known safety and effectiveness, reintroducing forgotten and/or discarded ones. Figure 1. Chemical structures of clotrimazole and itraconazole. Considering the development of new formulations, microemulsions (MEs) have attracted great interest as potential drug delivery systems, mainly due to their unique physicochemical properties such as drug solubilization and enhanced absorption properties [10,11]. MEs are a thermodynamically stable, isotropic, transparent liquid system consisting of two immiscible liquids (usually water and oil) stabilized by a film of surfactant compounds, suitably combined with a cosurfactant [12,13]. The presence of the surfactant helps to reduce the interfacial tension, making it possible to join the oil and aqueous phases [14,15]. MEs have been proposed as an innovative formulation approach to improve solubility and efficacy and reduce of the toxicity of various drugs. Therefore, when the known hydrophobicity of clotrimazole and itraconazole are taken into account, such systems could be particularly advantageous for their delivery. In light of the aforementioned concepts, this paper reports the initial research phase for the pre-development of a clotrimazole–itraconazole formulation, the first step towards a new antifungal combination. In this sense, the development and characterization of this new pharmaceutical formulation requires the evaluation of parameters such as drug release and stability. Therefore, as a further extension of our work in the field, we have developed a simple, sensitive, and specific HPLC method for the simultaneous quantification of clotrimazole and itraconazole in microemulsion. Although many researchers have investigated clotrimazole and itraconazole singly or in combination with other compounds, to the best of our knowledge, no HPLC method has been developed for the simultaneous determination of both drugs simultaneously, especially in microemulsion systems [16,17]. Finally, we performed an accelerated stability study of this formulation and the developed Molecules 2019, 24, 2333 3 of 15 method was applied to monitor the content of active ingredients. Interestingly, these investigations led to the detection of a known clotrimazole degradation product whose structure was confirmed using NMR and HRMS, as well as a possible interaction between itraconazole and benzyl alcohol. 2. Results and Discussion 2.1. Study of the Compatibility between Clotrimazole and Itraconazole We initiated our studies by analyzing the physicochemical properties of both active ingredients as well as their compatibility using different techniques such as thermal analyses (differential scanning calorimetry (DSC) and thermogravimetric/derivative thermogravimetry (TG)/DTG) analysis), powder X-ray diffraction (XRD) and FTIR. Initially, we proceeded to characterize the active ingredients and their combination using thermal analyses, which offer the ability to quickly screen for potential drug–drug incompatibilities. Such interactions can be of a physical or chemical nature and may affect the stability and bioavailability of the final product, compromising the therapeutic efficacy and safety [18]. The TG and DTG curves of clotrimazole (Figure 2a) showed that it is thermally stable up to 340 C, C, as showed in when its thermal decomposition starts; the highest rate of weight loss occurs at 388.6 C, where a loss of 60% of the total weight is observed. As for the DTG curve, and is finished at 421.1 C, with a maximum itraconazole, its thermal decomposition starts at 200 rate at 295.3 C and a total weight loss of 87%. The TG profile of the binary mixture of clotrimazole and itraconazole (1:1 ratio) showed two decomposition steps, indicating that the compounds undergo thermal degradation independently, although a small shift in the initial temperature of decomposition was observed, as expected. C and is finished at 348.4 ◦ ◦ ◦ ◦ ◦ ◦ Figure 2. Thermogravimetric (TG) and derivative thermogravimetry (DTG) curves for (a) clotrimazole, (b) itraconazole and (c) the binary mixture of clotrimazole and itraconazole (1:1). Next, the DSC technique was employed to further analyze the occurrence of events related to possible interactions between the drugs [19]. It is noteworthy that although such analyses are conducted upon heating the sample to high temperatures, which is not consistent with the process of drug production nor its administration to patients, they afford important information regarding the physical properties of the sample [18]. ◦ The DSC curves of the drugs showed endothermic peaks attributed to the melting of the drugs −1) between 158.5 and 175.0 for clotrimazole. On the other hand, a single endothermic event was observed in the DSC curve of the −1), which suggests a strong binary mixture, starting at 127.7 and finishing at 137.1 (∆H = −25.35 J g interaction between clotrimazole and itraconazole (Figure 3). −1) for itraconazole and 136.8 and 153.1 C (∆H = 41.6 J g C (∆H = 31.5 J g ◦ Molecules 2019, 24, x FOR PEER REVIEW 3 of 15 analytical method was applied to monitor the content of active ingredients. Interestingly, these investigations led to the detection of a known clotrimazole degradation product whose structure was confirmed using NMR and HRMS, as well as a possible interaction between itraconazole and benzyl alcohol. 2. Results and Discussion 2.1. Study of the Compatibility between Clotrimazole and Itraconazole We initiated our studies by analyzing the physicochemical properties of both active ingredients as well as their compatibility using different techniques such as thermal analyses (differential scanning calorimetry (DSC) and thermogravimetric/derivative thermogravimetry (TG)/DTG) analysis), powder X-ray diffraction (XRD) and FTIR. Initially, we proceeded to characterize the active ingredients and their combination using thermal analyses, which offer the ability to quickly screen for potential drug–drug incompatibilities. Such interactions can be of a physical or chemical nature and may affect the stability and bioavailability of the final product, compromising the therapeutic efficacy and safety [18]. The TG and DTG curves of clotrimazole (Figure 2a) showed that it is thermally stable up to 340 °C, when its thermal decomposition starts; the highest rate of weight loss occurs at 388.6 °C, as showed in the DTG curve, and is finished at 421.1 °C, where a loss of 60% of the total weight is observed. As for itraconazole, its thermal decomposition starts at 200 °C and is finished at 348.4 °C, with a maximum rate at 295.3 °C and a total weight loss of 87%. The TG profile of the binary mixture of clotrimazole and itraconazole (1:1 ratio) showed two decomposition steps, indicating that the compounds undergo thermal degradation independently, although a small shift in the initial temperature of decomposition was observed, as expected. Figure 2. Thermogravimetric (TG) and derivative thermogravimetry (DTG) curves for (a) clotrimazole, (b) itraconazole and (c) the binary mixture of clotrimazole and itraconazole (1:1). Next, the DSC technique was employed to further analyze the occurrence of events related to possible interactions between the drugs [19]. It is noteworthy that although such analyses are conducted upon heating the sample to high temperatures, which is not consistent with the process of drug production nor its administration to patients, they afford important information regarding the physical properties of the sample [18]. The DSC curves of the drugs showed endothermic peaks attributed to the melting of the drugs between 158.5 and 175.0 °C (ΔH = 31.5 J g−1) for itraconazole and 136.8 and 153.1 °C (ΔH = 41.6 J g−1) for clotrimazole. On the other hand, a single endothermic event was observed in the DSC curve of the binary mixture, starting at 127.7 and finishing at 137.1 (ΔH = −25.35 J g−1), which suggests a strong interaction between clotrimazole and itraconazole (Figure 3). Molecules 2019, 24, 2333 4 of 15 Figure 3. Differential scanning calorimetry (DSC) profile of itraconazole, clotrimazole, and the clotrimazole/itraconazole binary mixture (1:1). In order to further explore the possibility of interactions between the active ingredients, powder X-ray diffraction (PXRD) analyses were conducted. Interestingly, the diffractogram of the binary mixture (Figure 4) contained virtually all the peaks of clotrimazole and itraconazole, with no marked displacement of the peaks being observed. Furthermore, it is important to highlight that it was not possible to notice the appearance of any new peaks, which means that if there is any interaction between the drugs, it probably is not strong enough to take place in the solid state. The same observations were made in the FTIR spectra of the binary mixture, which showed the characteristic bands observed for the isolated active ingredients (For more details, see the Supplementary Materials). Figure 4. X-ray diffractograms of clotrimazole, itraconazole, and the binary mixture (1:1). With the characterization of the active ingredients and the binary mixture in hand, we proceeded to develop an HPLC method for their quantification in a newly developed microemulsion for the treatment of sporotrichosis. Molecules 2019, 24, x FOR PEER REVIEW 4 of 15 80100120140160180200220Heat flow (mW/mg)Temperature (°C) Mixture (1:1) Clotrimazole Itraconazole Figure 3. Differential scanning calorimetry (DSC) profile of itraconazole, clotrimazole, and the clotrimazole/itraconazole binary mixture (1:1). In order to further explore the possibility of interactions between the active ingredients, powder X-ray diffraction (PXRD) analyses were conducted. Interestingly, the diffractogram of the binary mixture (Figure 4) contained virtually all the peaks of clotrimazole and itraconazole, with no marked displacement of the peaks being observed. Furthermore, it is important to highlight that it was not possible to notice the appearance of any new peaks, which means that if there is any interaction between the drugs, it probably is not strong enough to take place in the solid state. The same observations were made in the FTIR spectra of the binary mixture, which showed the characteristic bands observed for the isolated active ingredients (For more details, see the Supplementary Materials). With the characterization of the active ingredients and the binary mixture in hand, we proceeded to develop an HPLC method for their quantification in a newly developed microemulsion for the treatment of sporotrichosis. 10203040ICCCCCCCCCCCCCCCCCCCCIIIIIIIICCCCCIIICCCRelative intensity (a.u.)2θ (°) Binary mixture (1:1) Itraconazole (I) Clotrimazole (C)I Figure 4. X-ray diffractograms of clotrimazole, itraconazole, and the binary mixture (1:1). Molecules 2019, 24, x FOR PEER REVIEW 4 of 15 80100120140160180200220Heat flow (mW/mg)Temperature (°C) Mixture (1:1) Clotrimazole Itraconazole Figure 3. Differential scanning calorimetry (DSC) profile of itraconazole, clotrimazole, and the clotrimazole/itraconazole binary mixture (1:1). In order to further explore the possibility of interactions between the active ingredients, powder X-ray diffraction (PXRD) analyses were conducted. Interestingly, the diffractogram of the binary mixture (Figure 4) contained virtually all the peaks of clotrimazole and itraconazole, with no marked displacement of the peaks being observed. Furthermore, it is important to highlight that it was not possible to notice the appearance of any new peaks, which means that if there is any interaction between the drugs, it probably is not strong enough to take place in the solid state. The same observations were made in the FTIR spectra of the binary mixture, which showed the characteristic bands observed for the isolated active ingredients (For more details, see the Supplementary Materials). With the characterization of the active ingredients and the binary mixture in hand, we proceeded to develop an HPLC method for their quantification in a newly developed microemulsion for the treatment of sporotrichosis. 10203040ICCCCCCCCCCCCCCCCCCCCIIIIIIIICCCCCIIICCCRelative intensity (a.u.)2θ (°) Binary mixture (1:1) Itraconazole (I) Clotrimazole (C)I Figure 4. X-ray diffractograms of clotrimazole, itraconazole, and the binary mixture (1:1). Molecules 2019, 24, 2333 5 of 15 2.2. Determination of the Concentration of Clotrimazole and Itraconazole in Microemulsions Using HPLC Analyses Considering the unique properties presented by microemulsions, in the present work, benzyl alcohol was used as an oil phase, Tween 60 as a surfactant, and propylene glycol as a cosolvent in the presence of water. These components were chosen on the basis in their previously reported applications in other pharmaceutical forms available on the international market. ® In this context, HPLC-DAD (diode array detector) was selected as an analytical tool for the simultaneous quantification of clotrimazole and itraconazole in the developed microemulsion through a rapid, simple, and isocratic method [20]. In our study, the best separation condition was achieved using a C18 analytical column with a mobile phase composed of acetonitrile and a phosphate buffered saline 0.05 M (pH 8.0 with ammonium hydroxide 1 M) in the ratio (v/v) 60:40, respectively, with a −1 flow rate and UV detection at 190 nm. A typical chromatogram is presented in Figure 5, 1 mL min with a retention time of 9.1 min being observed for clotrimazole and 10.9 min for itraconazole. Figure 5. Chromatograms of the (a) mobile phase and (b) standard solution containing a binary mixture of itraconazole and clotrimazole. Molecules 2019, 24, x FOR PEER REVIEW 5 of 15 2.2. Determination of the Concentration of Clotrimazole and Itraconazole in Microemulsions Using HPLC Analyses Considering the unique properties presented by microemulsions, in the present work, benzyl alcohol was used as an oil phase, Tween® 60 as a surfactant, and propylene glycol as a cosolvent in the presence of water. These components were chosen on the basis in their previously reported applications in other pharmaceutical forms available on the international market. In this context, HPLC-DAD (diode array detector) was selected as an analytical tool for the simultaneous quantification of clotrimazole and itraconazole in the developed microemulsion through a rapid, simple, and isocratic method [20]. In our study, the best separation condition was achieved using a C18 analytical column with a mobile phase composed of acetonitrile and a phosphate buffered saline 0.05 M (pH 8.0 with ammonium hydroxide 1 M) in the ratio (v/v) 60:40, respectively, with a 1 mL min−1 flow rate and UV detection at 190 nm. A typical chromatogram is presented in Figure 5, with a retention time of 9.1 min being observed for clotrimazole and 10.9 min for itraconazole. (a) (b) Figure 5. Chromatograms of the (a) mobile phase and (b) standard solution containing a binary mixture of itraconazole and clotrimazole. Molecules 2019, 24, 2333 6 of 15 To evaluate the linearity of the method, calibration standards of clotrimazole (5–200 µg mL −1) −1) were analyzed. A linear relationship was established for the injected and itraconazole (5–160 µg mL concentration ranges versus the peak area for both analytes, with determination coefficients greater than 0.9988 (see the calibration curves in the Supplementary Materials). The calibration curve parameters are reported in Table 1, with the linearity parameters of the method shown in Table 2. Table 1. Summary of the validation data obtained for the proposed HPLC method developed for the quantification of clotrimazole and itraconazole in microemulsions. LOD—limit of detection; LOQ—limit of quantification. Standard Solutions Parameters of the Method Validation Results Clotrimazole Itraconazole Linearity LOD LOQ Slope Interception Linearity LOD LOQ Slope Interception Calibration range (µg mL −1): 5–200 y = 233647.7939x − 312039.9299 (R2 = 0.9988) 0.84 µg mL 2.54 µg mL 233647.7939 ± 976.8015153 −312039.9299 ± 59416.57811 −1 −1 Calibration range (µg mL −1): 5–160 y = 89946.6896x − 79996.5373 (R2 = 0.9999) 0.86 µg mL 2.60 µg mL 89946.6896 ± 780.1420761 −79996.53731 ± 23351.48986 −1 −1 Table 2. Data related to the linearity of the developed HPLC method with its respective average, precision, and accuracy. Concentration (µg/mL) Clotrimazole Itraconazole Average (µg/mL) Accuracy (%) Precision (%) Average (µg/mL) Accuracy (%) Precision (%) 5 10 20 40 80 160 200 4.883 9.292 19.233 38.927 77.731 151.888 204.631 97.7 92.9 96.2 97.3 97.2 94.9 102.3 0.20 0.57 0.40 0.01 1.08 0.71 0.65 5.593 10.115 20.029 39.621 79.154 160.488 - 111.9 101.2 100.1 99.1 98.9 100.3 - 0.86 0.91 0.58 1.12 1.81 0.82 - The method’s selectivity was confirmed by the absence of interferences at the retention times of itraconazole and clotrimazole in the microemulsion prepared without the drugs (Figure 6). The purity of the compounds was checked using PDA (photodiode array) detection. The within-assay precision (repeatability) was carried out by performing six consecutive analyses of standard solution at three different concentrations for each drug on the same day. The samples were also analyzed on different days to evaluate the between-assay precision (intermediate precision). The obtained values were evaluated through the dispersion of the results by calculating the standard deviation of the measurement series. The intra- and inter-day precision relative standard deviation (RSD %) was between 1.18 and 0.8 for clotrimazole and 1.48 and 0.84 for itraconazole. The recovery of the drugs was in the range of 93.8–100.9% with RSDs below 2.35% for clotrimazole and in the range of 100.5–104.3% with RSDs below 2.40% for itraconazole. The results are given in Table 3. Molecules 2019, 24, 2333 7 of 15 Figure 6. Chromatogram obtained from the injection of the microemulsion using the developed HPLC method. Table 3. Data related to the repeatability and intermediate precision of the developed HPLC method. Samples (µg mL−1) Intra-Day Precision (Repeatability) Inter-Day Precision (Intermediate Precision) Clotrimazole Concentration −1) Found (µg mL 7 15 120 6.818 14.510 116.679 Itraconazole Concentration −1) Found (µg mL 7 70 150 7.206 70.809 152.745 Accuracy (%) 97.4 ±2.25 96.7 ±1.13 97.2 ±0.27 Accuracy (%) 102.9 ± 1.33 101.2 ± 1.15 101.8 ± 0.85 Precision (%) Concentration −1) Found (µg mL 0.47 1.18 0.28 6.865 14.075 121.108 Precision (%) Concentration −1) Found (µg mL 1.48 1.16 0.84 7.305 70.374 160.98 Accuracy (%) 98.07 ± 1.17 93.83 ± 3.17 100.92 ± 4.28 Accuracy (%) 104.35 ± 1.25 100.53 ± 2.41 100.61 ± 4.9 Precision (%) 2.35 0.95 0.28 Precision (%) 1.20 2.40 1.59 No changes were observed in the drug concentrations of the stock solutions under storage conditions. Indeed, further analyses showed that the percent recovery of clotrimazole and itraconazole were, respectively, 97.3% ± 3.15 and 91.3% ± 2.71 at room temperature (25 C) and 94.2 ± 0.34 and 88.7 ± 1.63 under refrigeration (−5 C, Table 4). Moreover, the drugs were stable for at least 30 days under storage conditions, with RSDs below 8%. ◦ ◦ Table 4. Data related to the stability of the assay of the developed HPLC method. N = 2 for each day and condition. Days Accuracy (%) Precision (%) Accuracy (%) Precision (%) Clotrimazole Itraconazole 0 7 15 30 ◦ 97.3 ± 0.94 (25 105.7 ± 0.89 (25 104.9 ± 0.07 (−5 105.3 ± 1.51 (25 105.5 ± 0.39 (−5 ◦ 97.3 ± 3.15 (25 94.2 ± 0.34 (−5 C) ◦ C) C) C) C) ◦ ◦ ◦ C) ◦ C) 1.18 0.85 0.07 1.45 0.38 0.62 0.73 101.4 ± 0.62 98.6 ± 4.48 98.4 ± 7.79 101.3 ± 0.51 100.1 ± 3.71 91.3 ± 2.71 88.7 ± 1.63 0.84 4.57 7.96 0.50 3.74 3.21 3.04 In order to evaluate the robustness of the chromatographic method, assays were carried out by changing both the column brand and ratio of the mobile phase for acetonitrile 70:30 (v/v) and a phosphate buffered saline 0.05 M (pH 8.0 with ammonium hydroxide 1 M). The alteration of the Molecules 2019, 24, x FOR PEER REVIEW 7 of 15 Figure 6. Chromatogram obtained from the injection of the microemulsion using the developed HPLC method. Table 3. Data related to the repeatability and intermediate precision of the developed HPLC method. Samples (µg mL−1) Intra-Day Precision (Repeatability) Inter-Day Precision (Intermediate Precision) Clotrimazole Concentration Found (µg mL−1) Accuracy (%) Precision (%) Concentration Found (µg mL−1) Accuracy (%) Precision (%) 7 6.818 97.4 ±2.25 0.47 6.865 98.07 ± 1.17 2.35 15 14.510 96.7 ±1.13 1.18 14.075 93.83 ± 3.17 0.95 120 116.679 97.2 ±0.27 0.28 121.108 100.92 ± 4.28 0.28 Itraconazole Concentration Found (µg mL−1) Accuracy (%) Precision (%) Concentration Found (µg mL−1) Accuracy (%) Precision (%) 7 7.206 102.9 ± 1.33 1.48 7.305 104.35 ± 1.25 1.20 70 70.809 101.2 ± 1.15 1.16 70.374 100.53 ± 2.41 2.40 150 152.745 101.8 ± 0.85 0.84 160.98 100.61 ± 4.9 1.59 No changes were observed in the drug concentrations of the stock solutions under storage conditions. Indeed, further analyses showed that the percent recovery of clotrimazole and itraconazole were, respectively, 97.3% ± 3.15 and 91.3% ± 2.71 at room temperature (25 °C) and 94.2 ± 0.34 and 88.7 ± 1.63 under refrigeration (−5 °C, Table 4). Moreover, the drugs were stable for at least 30 days under storage conditions, with RSDs below 8%. Table 4. Data related to the stability of the assay of the developed HPLC method. N = 2 for each day and condition. Days Accuracy (%) Precision (%) Accuracy (%) Precision (%) Clotrimazole Itraconazole 0 97.3 ± 0.94 (25 °C) 1.18 101.4 ± 0.62 0.84 7 105.7 ± 0.89 (25 °C) 0.85 98.6 ± 4.48 4.57 104.9 ± 0.07 (−5 °C) 0.07 98.4 ± 7.79 7.96 15 105.3 ± 1.51 (25 °C) 1.45 101.3 ± 0.51 0.50 105.5 ± 0.39 (−5 °C) 0.38 100.1 ± 3.71 3.74 30 97.3 ± 3.15 (25 °C) 0.62 91.3 ± 2.71 3.21 94.2 ± 0.34 (−5 °C) 0.73 88.7 ± 1.63 3.04 In order to evaluate the robustness of the chromatographic method, assays were carried out by changing both the column brand and ratio of the mobile phase for acetonitrile 70:30 (v/v) and a Molecules 2019, 24, 2333 8 of 15 column brand and the mobile phase did not promote any significant variations in the retention time of clotrimazole and itraconazole peaks; a good resolution was observed with retention times of 8 min for clotrimazole and 10.7 min for itraconazole (Figure 7). Figure 7. Chromatogram of clotrimazole and itraconazole obtained in the robustness studies. 2.3. Study of the Stability of a Novel Microemulsion Containing Clotrimazole and Itraconazole Subsequently, the developed method was used in the determination of clotrimazole and itraconazole in the newly developed microemulsion with the purpose of quantifying the drugs in the formulation, as well as in the accelerated stability study. Based on the assumption that possible interactions and incompatibilities may arise from the contact between the drugs over time, they were left to stand for three months, both under refrigeration and heating conditions, and further analyzed. The initial drug content of the microemulsion was taken as 100%, and the drug content over time was plotted (Figure 8), with all data being represented as mean ± SD (n = 3). For the samples stored at 5 C, no significant changes were observed for both drugs when compared to the first day. Furthermore, it is noteworthy that there was no evident interaction between clotrimazole and itraconazole at this temperature, since the peaks of both drugs were detected independently without the appearance of any additional peaks. On the other hand, when the samples that were stored at 40 C were analyzed, it was possible to notice a significant decrease in the concentration of the drugs over time, especially for clotrimazole. Additionally, a new peak could also be observed in the chromatogram of such samples (Figure 9). ◦ ◦ Molecules 2019, 24, x FOR PEER REVIEW 8 of 15 phosphate buffered saline 0.05 M (pH 8.0 with ammonium hydroxide 1 M). The alteration of the column brand and the mobile phase did not promote any significant variations in the retention time of clotrimazole and itraconazole peaks; a good resolution was observed with retention times of 8 min for clotrimazole and 10.7 min for itraconazole (Figure 7). Figure 7. Chromatogram of clotrimazole and itraconazole obtained in the robustness studies. 2.3. Study of the Stability of a Novel Microemulsion Containing Clotrimazole and Itraconazole Subsequently, the developed method was used in the determination of clotrimazole and itraconazole in the newly developed microemulsion with the purpose of quantifying the drugs in the formulation, as well as in the accelerated stability study. Based on the assumption that possible interactions and incompatibilities may arise from the contact between the drugs over time, they were left to stand for three months, both under refrigeration and heating conditions, and further analyzed. The initial drug content of the microemulsion was taken as 100%, and the drug content over time was plotted (Figure 8), with all data being represented as mean ± SD (n = 3). For the samples stored at 5 °C, no significant changes were observed for both drugs when compared to the first day. Furthermore, it is noteworthy that there was no evident interaction between clotrimazole and itraconazole at this temperature, since the peaks of both drugs were detected independently without the appearance of any additional peaks. On the other hand, when the samples that were stored at 40 °C were analyzed, it was possible to notice a significant decrease in the concentration of the drugs over time, especially for clotrimazole. Additionally, a new peak could also be observed in the chromatogram of such samples (Figure 9). 020406080020406080100Concentration (%)Time (days) Clotrimazole 40°C Clotrimazole 5°C Itraconazole 40°C Itraconazole 5°C Figure 8. Graph showing the concentration of clotrimazole and itraconazole over time in different conditions. All data is represented as mean ± SD (n = 3). Molecules 2019, 24, 2333 9 of 15 Figure 8. Graph showing the concentration of clotrimazole and itraconazole over time in different conditions. All data is represented as mean ± SD (n = 3). Figure 9. Cont. Molecules 2019, 24, x FOR PEER REVIEW 8 of 15 phosphate buffered saline 0.05 M (pH 8.0 with ammonium hydroxide 1 M). The alteration of the column brand and the mobile phase did not promote any significant variations in the retention time of clotrimazole and itraconazole peaks; a good resolution was observed with retention times of 8 min for clotrimazole and 10.7 min for itraconazole (Figure 7). Figure 7. Chromatogram of clotrimazole and itraconazole obtained in the robustness studies. 2.3. Study of the Stability of a Novel Microemulsion Containing Clotrimazole and Itraconazole Subsequently, the developed method was used in the determination of clotrimazole and itraconazole in the newly developed microemulsion with the purpose of quantifying the drugs in the formulation, as well as in the accelerated stability study. Based on the assumption that possible interactions and incompatibilities may arise from the contact between the drugs over time, they were left to stand for three months, both under refrigeration and heating conditions, and further analyzed. The initial drug content of the microemulsion was taken as 100%, and the drug content over time was plotted (Figure 8), with all data being represented as mean ± SD (n = 3). For the samples stored at 5 °C, no significant changes were observed for both drugs when compared to the first day. Furthermore, it is noteworthy that there was no evident interaction between clotrimazole and itraconazole at this temperature, since the peaks of both drugs were detected independently without the appearance of any additional peaks. On the other hand, when the samples that were stored at 40 °C were analyzed, it was possible to notice a significant decrease in the concentration of the drugs over time, especially for clotrimazole. Additionally, a new peak could also be observed in the chromatogram of such samples (Figure 9). 020406080020406080100Concentration (%)Time (days) Clotrimazole 40°C Clotrimazole 5°C Itraconazole 40°C Itraconazole 5°C Figure 8. Graph showing the concentration of clotrimazole and itraconazole over time in different conditions. All data is represented as mean ± SD (n = 3). Molecules 2019, 24, x FOR PEER REVIEW 9 of 15 020406080020406080100 Concentration (%)Time (days) Clotrimazole 40°C Clotrimazole 5°C Itraconazole 40°C Itraconazole 5°C Figure 8. Graph showing the concentration of clotrimazole and itraconazole over time in different conditions. All data is represented as mean ± SD (n = 3). (A) 30 days (40 °C) (B) 60 days (40 °C) Molecules 2019, 24, 2333 10 of 15 Figure 9. HPLC chromatograms for the samples in the stability study after (A) 30 days, (B) 60 days, and (C) 90 days. In order to investigate the formation of this compound, which might be a result of the interaction between clotrimazole and itraconazole, we conducted further studies. Initially, we sought to investigate which degradation products could be formed from the degradation of both drugs and found that the degradation of clotrimazole is well-reported under acidic conditions, giving product 3 (Figure 10). Figure 10. Reaction scheme showing the degradation of clotrimazole in acid medium. ◦ With these concepts in mind, we conducted the synthesis of compound 3 from clotrimazole by heating it at 80 C in the presence of acetonitrile and concentrated hydrochloric acid for 2 h; the product identity was confirmed using NMR and HRMS by comparing the obtained data with previous reports (for details, see the Supplementary Materials) [21]. Next, we conducted the forced degradation of a mixture of itraconazole and clotrimazole by heating both at 50 C for 24 h in a solution of acetonitrile, water, and benzyl alcohol-mimicking the microemulsion composition—and isolated the formed product using column chromatography. ◦ With both compounds in hand, we analyzed product 3 and the degradation product by HPLC using the developed method, and the comparison of the retention times of both compounds proved that, indeed, product 3 is formed from the degradation of clotrimazole under acidic conditions. Furthermore, the retention time was also a match for the product previously detected in the stability studies conducted at 40 C, which proves that under specific conditions, clotrimazole may undergo degradation in the presence of traces of acid, forming 3. However, the formation of 3 was not observed in the stability studies conducted at 5 C, which shows the viability of this novel microemulsion and encourages carrying out further studies for its development. ◦ ◦ The content of itraconazole (% w/w) in the microemulsions stored in climatic chambers also underwent a slight decrease, which was less significant when compared to clotrimazole. In order to exclude the possibility of interaction between the drugs, the decrease in itraconazole content was also investigated. However, unlike clotrimazole, degradation studies of itraconazole are not found in the literature. Thus, an aliquot was collected directly from the chromatographic system at the same retention time as the degradation product formed during the stability study; at a low intensity, it Molecules 2019, 24, x FOR PEER REVIEW 10 of 15 (C) 90 days (40 °C) Figure 9. HPLC chromatograms for the samples in the stability study after (A) 30 days, (B) 60 days, and (C) 90 days. In order to investigate the formation of this compound, which might be a result of the interaction between clotrimazole and itraconazole, we conducted further studies. Initially, we sought to investigate which degradation products could be formed from the degradation of both drugs and found that the degradation of clotrimazole is well-reported under acidic conditions, giving product 3 (Figure 10). Figure 10. Reaction scheme showing the degradation of clotrimazole in acid medium. With these concepts in mind, we conducted the synthesis of compound 3 from clotrimazole by heating it at 80 °C in the presence of acetonitrile and concentrated hydrochloric acid for 2 h; the product identity was confirmed using NMR and HRMS by comparing the obtained data with previous reports (for details, see the Supplementary Materials) [21]. Next, we conducted the forced degradation of a mixture of itraconazole and clotrimazole by heating both at 50 °C for 24 h in a solution of acetonitrile, water, and benzyl alcohol-mimicking the microemulsion composition—and isolated the formed product using column chromatography. With both compounds in hand, we analyzed product 3 and the degradation product by HPLC using the developed method, and the comparison of the retention times of both compounds proved that, indeed, product 3 is formed from the degradation of clotrimazole under acidic conditions. Furthermore, the retention time was also a match for the product previously detected in the stability studies conducted at 40 °C, which proves that under specific conditions, clotrimazole may undergo degradation in the presence of traces of acid, forming 3. However, the formation of 3 was not observed in the stability studies conducted at 5 °C, which shows the viability of this novel microemulsion and encourages carrying out further studies for its development. The content of itraconazole (% w/w) in the microemulsions stored in climatic chambers also underwent a slight decrease, which was less significant when compared to clotrimazole. In order to exclude the possibility of interaction between the drugs, the decrease in itraconazole content was also investigated. However, unlike clotrimazole, degradation studies of itraconazole are not found in the literature. Thus, an aliquot was collected directly from the chromatographic system at the same retention time as the degradation product formed during the stability study; at a low intensity, it was Molecules 2019, 24, x FOR PEER REVIEW 10 of 15 NClN2H2OH+ (cat)OHCl3+NHN4+ Figure 10. Reaction scheme showing the degradation of clotrimazole in acid medium. With these concepts in mind, we conducted the synthesis of compound 3 from clotrimazole by heating it at 80 °C in the presence of acetonitrile and concentrated hydrochloric acid for 2 h; the product identity was confirmed using NMR and HRMS by comparing the obtained data with previous reports (for details, see the Supplementary Materials) [21]. Next, we conducted the forced degradation of a mixture of itraconazole and clotrimazole by heating both at 50 °C for 24 h in a solution of acetonitrile, water, and benzyl alcohol-mimicking the microemulsion composition—and isolated the formed product using column chromatography. With both compounds in hand, we analyzed product 3 and the degradation product by HPLC using the developed method, and the comparison of the retention times of both compounds proved that, indeed, product 3 is formed from the degradation of clotrimazole under acidic conditions. Furthermore, the retention time was also a match for the product previously detected in the stability studies conducted at 40 °C, which proves that under specific conditions, clotrimazole may undergo degradation in the presence of traces of acid, forming 3. However, the formation of 3 was not observed in the stability studies conducted at 5 °C, which shows the viability of this novel microemulsion and encourages carrying out further studies for its development. The content of itraconazole (% w/w) in the microemulsions stored in climatic chambers also underwent a slight decrease, which was less significant when compared to clotrimazole. In order to exclude the possibility of interaction between the drugs, the decrease in itraconazole content was also investigated. However, unlike clotrimazole, degradation studies of itraconazole are not found in the literature. Thus, an aliquot was collected directly from the chromatographic system at the same retention time as the degradation product formed during the stability study; at a low intensity, it was possible to observe a product with a mass-to-charge ratio (m/z) of 437.1931. Considering that the itraconazole concentration change was lower than for clotrimazole, we hypothesized that an interaction of itraconazole with some other excipient of the microemulsion may be taking place. In that sense, we propose that the degradation product may be formed by the nucleophilic addition of benzyl alcohol to the methylene group linking the phenolic aromatic part with the 1,3-dioxolane ring (Figure 11); indeed, the mass-to-charge ratio was a match for the proposed product. It is worth mentioning that although we have observed a good match in a mass-to-charge ratio of 437.1931, further studies are necessary to confirm whether the proposed structure is indeed the correct one, such as the isolation and complete spectroscopic characterization of this compound, which was not possible at the scale we were working. Figure 11. Scheme showing the reaction between itraconazole and benzyl alcohol. Molecules 2019, 24, 2333 11 of 15 was possible to observe a product with a mass-to-charge ratio (m/z) of 437.1931. Considering that the itraconazole concentration change was lower than for clotrimazole, we hypothesized that an interaction of itraconazole with some other excipient of the microemulsion may be taking place. In that sense, we propose that the degradation product may be formed by the nucleophilic addition of benzyl alcohol to the methylene group linking the phenolic aromatic part with the 1,3-dioxolane ring (Figure 11); indeed, the mass-to-charge ratio was a match for the proposed product. It is worth mentioning that although we have observed a good match in a mass-to-charge ratio of 437.1931, further studies are necessary to confirm whether the proposed structure is indeed the correct one, such as the isolation and complete spectroscopic characterization of this compound, which was not possible at the scale we were working. Figure 11. Scheme showing the reaction between itraconazole and benzyl alcohol. 3. Experimental Methods 3.1. Materials for Analytical Method Development Clotrimazole and itraconazole (as a mixture of stereoisomers) standards were purchased from Merck, São Paulo, SP, Brazil. Microemulsions were prepared using Tween 60, propylene glycol, and benzyl alcohol, all purchased from Merck, São Paulo, SP, Brazil. HPLC-grade acetonitrile was acquired from J.T. Baker Inc., Phillipsburg, NJ, USA. Clotrimazole (Jintan Zhongxing Pharmaceutical Chemical Co., Ltd., Mainland, China) and itraconazole (Metrochem API, Telangana, India) were donated by Valdequimica Produtos Quimicos Ltd., São Paulo, Brazil. All solutions were prepared with ultra-pure Milli-Q water obtained from a Milli-Q Water Millipore purification system (Burlington, MA, USA). ® 3.2. Compatibility Study of Clotrimazole and Itraconazole 3.2.1. Preparation of Clotrimazole/Itraconazole Binary Mixtures The binary mixtures were prepared and homogenized by taking clotrimazole and itraconazole in a 1:1 proportion (w:w). These mixtures were further used for X-ray powder diffraction, Fourier transform infrared spectroscopy (FTIR), and thermal analyses. 3.2.2. X-ray Powder Diffraction (PXRD) PXRD patterns were collected on a Bruker D8 Venture diffractometer system (Bruker, Billerica, MA, USA) operating at 1.5406 Å, 40 kV voltage, and a current of 40 mA using a Cu Kα radiation source. The samples were contained in a flat poly(methyl methacrylate) sample holder and the data acquisition ◦ was done in a range of 5 to 70 ◦/0.1 s step size over a total period of 10 min. (2θ) at 0.019 3.2.3. Fourier Transform Infrared Spectroscopy (FTIR) The FTIR spectra of the solid samples were obtained using a Varian FT-IR 660 equipment (Varian Inc., Walnut Creek, CA, USA). A hydraulic press was used to prepare pellets for analysis. The KBr pellets contained 3 mg of a sample and 100 mg of KBr. Spectra were collected with a resolution of 4 cm −1 on the spectral domain of 3800–600 cm −1. Molecules 2019, 24, x FOR PEER REVIEW 10 of 15 NClN2H2OH+ (cat)OHCl3+NHN4+ Figure 10. Reaction scheme showing the degradation of clotrimazole in acid medium. With these concepts in mind, we conducted the synthesis of compound 3 from clotrimazole by heating it at 80 °C in the presence of acetonitrile and concentrated hydrochloric acid for 2 h; the product identity was confirmed using NMR and HRMS by comparing the obtained data with previous reports (for details, see the Supplementary Materials) [21]. Next, we conducted the forced degradation of a mixture of itraconazole and clotrimazole by heating both at 50 °C for 24 h in a solution of acetonitrile, water, and benzyl alcohol-mimicking the microemulsion composition—and isolated the formed product using column chromatography. With both compounds in hand, we analyzed product 3 and the degradation product by HPLC using the developed method, and the comparison of the retention times of both compounds proved that, indeed, product 3 is formed from the degradation of clotrimazole under acidic conditions. Furthermore, the retention time was also a match for the product previously detected in the stability studies conducted at 40 °C, which proves that under specific conditions, clotrimazole may undergo degradation in the presence of traces of acid, forming 3. However, the formation of 3 was not observed in the stability studies conducted at 5 °C, which shows the viability of this novel microemulsion and encourages carrying out further studies for its development. The content of itraconazole (% w/w) in the microemulsions stored in climatic chambers also underwent a slight decrease, which was less significant when compared to clotrimazole. In order to exclude the possibility of interaction between the drugs, the decrease in itraconazole content was also investigated. However, unlike clotrimazole, degradation studies of itraconazole are not found in the literature. Thus, an aliquot was collected directly from the chromatographic system at the same retention time as the degradation product formed during the stability study; at a low intensity, it was possible to observe a product with a mass-to-charge ratio (m/z) of 437.1931. Considering that the itraconazole concentration change was lower than for clotrimazole, we hypothesized that an interaction of itraconazole with some other excipient of the microemulsion may be taking place. In that sense, we propose that the degradation product may be formed by the nucleophilic addition of benzyl alcohol to the methylene group linking the phenolic aromatic part with the 1,3-dioxolane ring (Figure 11); indeed, the mass-to-charge ratio was a match for the proposed product. It is worth mentioning that although we have observed a good match in a mass-to-charge ratio of 437.1931, further studies are necessary to confirm whether the proposed structure is indeed the correct one, such as the isolation and complete spectroscopic characterization of this compound, which was not possible at the scale we were working. Figure 11. Scheme showing the reaction between itraconazole and benzyl alcohol. Molecules 2019, 24, 2333 12 of 15 3.2.4. Thermal Analyses DSC data were collected on a Shimadzu Differential Scanning Calorimeter DSC-60A (Shimadzu, Quioto, Japan). Approximately 4 mg samples were placed in aluminum pans, and the temperature −1 under nitrogen flow program was set to increase from 30 to 250 (50 mL min C with a heating rate of 10 C min −1). ◦ ◦ Thermogravimetric (TG) analyses were performed using a Netzsch STA 409 PC/PG (Netzsch, −1 at a heating rate of Selb, Germany) under a nitrogen atmosphere with a flow rate of 60 mL min ◦ 10 −1 over the range of 30 to 300 C and using 6 mg of sample in an aluminum cell. C min ◦ 3.3. Instruments and Chromatographic Conditions Chromatographic experiments were performed on a Shimadzu SPD-M20A system (Shimadzu, Quioto, Japan). The chromatographic separations were performed using a 150 mm × 4.6 mm i.d. (5 µm particle size) Fortis C18 column in isocratic elution mode with acetonitrile and phosphate buffered −1. saline 0.05 M pH 8.0 adjusted with ammonium hydroxide 1 M (60:40, v/v) at a flow rate of 1.0 mL min The detection wavelength was set at 190 nm, and the injection volume was 20 µL. 3.4. Standard Stock Solutions and Calibration Standards Standard stock solutions of clotrimazole and itraconazole were freshly prepared by dissolving the −1) Calibration standards in the concentration range of 5, 10, 20, 40, 80, −1 were prepared in the appropriate volumetric flasks by diluting the stock solution drugs in methanol (0.2 mg mL 160, and 200 µg mL in the mobile phase. An aliquot (20 µL) of the solution was then directly injected into the HPLC. 3.5. Sample Preparation An amount of microemulsion was accurately weighted to contain 25 mg clotrimazole and C. The sample was itraconazole in a 50 mL centrifuge tube and heated for 5 min in a water bath at 50 then removed from the bath, shaken until cooled to room temperature, and placed in an ice-methanol bath. Next, the sample was centrifuged for 5 min and extracted with chloroform (5 mL). Finally, the solvent was removed under a stream of gaseous nitrogen, and the residue was diluted in the mobile phase. ◦ 3.6. Method Validation Protocol The proposed method was validated under the optimized conditions regarding its linearity range, selectivity, sensitivity, precision, accuracy and stability of the assay according to the regulatory guidelines requirements (FDA). 3.6.1. Linearity Range The linearity range was evaluated by measuring the chromatographic peak area responses of the drugs at seven concentration levels and in triplicate. Analytical curves were constructed by plotting the peak area against the concentration of itraconazole and clotrimazole (Figures 2 and 3), which gives the regression equation. The results are presented in Table 1. 3.6.2. Selectivity To ensure the selectivity of the proposed method, drug-free microemulsions were prepared and analyzed in the described chromatographic conditions. 3.6.3. Sensitivity The sensitivity was determined by means of the limit of detection (LOD) and limit of quantification (LOQ). One of the ways to calculate the LOD (Equation (1)) and LOQ (Equation (2)) is based on the Molecules 2019, 24, 2333 13 of 15 standard deviation (σ) of the y-intercept from the regression of the calibration standard. The results are given in Table 1. LOD = 3, 3.σ s LOD (σ—standard deviation; s—slope of the calibration standard). LOQ = 10.σ s LOQ (σ—standard deviation; s—slope of the calibration standard). 3.6.4. Precision and Accuracy (1) (2) The accuracy and precision of the method were estimated by quintuplicate quality control −1 (medium QC), (QC) samples prepared using the mobile phase: 7 µg mL −1 (medium QC), −1 (high QC) for clotrimazole and 7 µg mL and 120 µg mL −1 (high QC) for itraconazole. Accuracy was established through back-calculation and 150 µg mL and expressed as the percent difference between the found and the nominal concentration for each compound, and the precision was calculated as the coefficient of variation (CV) of the replicate measurements. Calibration standards and QC samples were analyzed in three different batches in order to determine the intra and inter-batch variability. −1 (low QC), 15 µg mL −1 (low QC), 70 µg mL 3.6.5. Stability The stability of the standard solutions was investigated after storage for 7, 15, and 30 days at room temperature (25 ◦ C) and under refrigeration (−5 ◦ C) using the working solution. 3.6.6. Robustness The robustness of an analytical method is a measure of its capacity to resist changes due to small variations in parameter conditions, e.g., by using a different column. In this way, the method robustness was assessed as a function of changing the column brand for a C18 Agilent column (Agilent Technologies Inc, Santa Clara, CA, USA), (150 × 4.6 mm × 5 µm) and the ratio of the mobile phase. 3.7. Application of the Method 3.7.1. Microemulsion Preparation ® With the developed method in hand, the next step was to develop a stable microemulsion using a combination of clotrimazole and itraconazole. MEs were composed of benzyl alcohol, the non-ionic surfactant Tween 60, propylene glycol, and water. The optimum weight ratios of the components and MEs’ areas were determined using a pseudo-ternary phase diagram (data not shown in this work). The systems were prepared as previously described [22]; the surfactant (Tween 60) and cosolvent (propylene glycol) were prepared separately, and clotrimazole and itraconazol were solubilized in benzyl alcohol and added to the mixture. The pseudo-ternary phase diagrams of oil, surfactant/cosolvent, and water were set up using the water titration method. ® 3.7.2. Stability Study The stability profile of the prepared microemulsion at accelerated conditions was studied according to the ICH guidelines. The formulation was placed separately in an amber-colored screw-capped glass container and stored at 40 ± 2 C for 3 months, with sampling at 0, 30, 60, and 90 days. The samples were then evaluated for drug content using the developed HPLC method. C and 5–8 ± 3 ◦ ◦ Molecules 2019, 24, 2333 14 of 15 3.8. Characterization of the Synthetized Compounds NMR spectra were obtained using a Varian Unity Plus VXR (Varian Inc., Walnut Creek, CA, USA), 500 MHz instrument in CDCl3 solutions. The chemical shifts were reported in units of d (ppm) downfield from tetramethylsilane, which was used as an internal standard; coupling constants (J) are reported in hertz and refer to apparent peak multiplicities. High-resolution mass spectra (HRMS) were recorded on a MICROMASS Q-TOF mass spectrometer (Waters, Milford, MA, USA). 4. Conclusions The combination of clotrimazole and itraconazole in a pharmaceutical formulation is of great importance owing to the potential of generating a new option for the treatment of sporotrichosis. In this sense, the preformulation investigation using different techniques (DSC, TG, PXRD, FTIR) was essential to examine the existence of possible clotrimazole–itraconazole interactions. Furthermore, an HPLC method was developed and validated according to standard guidelines, and it is the first reported method for the simultaneous determination of clotrimazole and itraconazole in nanotechnology-based products such as microemulsions. Based on our results, it was possible to conclude that there is no other co-eluting peak along with those of interest, the method being specific for the estimation of clotrimazole and itraconazole. Interestingly, accelerated stability studies showed that a product derived from clotrimazole was formed, as well as a possible interaction between itraconazole and benzyl alcohol, when the microemulsion was conditioned at elevated temperatures (40 C). On the other hand, the studies C showed that the microemulsion is stable for at least 3 months, as no degradation conducted at 5 peaks were observed in the HPLC analysis, which allows us to infer that it is possible to guarantee the stability of the formulation under refrigeration. ◦ ◦ Supplementary Materials: Supplementary materials are available online. Figure S1: Analytical calibration curve for clotrimazole, Figure S2: Analytical calibration curve for itraconazole, Figure S3: IR spectra of clotrimazole, itraconazole and their binary mixture (1:1), Figure S4: HPLC chromatogram of compound 3, Figure S5: HPLC chromatogram of the decomposition product formed via the forced degradation of clotrimazole in the presence of itraconazole. Author Contributions: All authors have read this manuscript and concur with its submission. The contributions of each author are listed as follows: P.G.F.—Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Roles/Writing—original draft; C.G.d.S.L.—Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing—review & editing; L.L.N.—Data curation, Formal analysis, Investigation; Methodology; M.C.d.M.—Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Funding acquisition, Writing—review & editing; F.d.C.d.S.—Conceptualization, Funding acquisition, Project administration, Resources, Supervision, Validation, Visualization; A.L.V.—Conceptualization; D.O.F.—Conceptualization, Funding acquisition, Supervision, Writing—review & editing; V.F.F.—Conceptualization, Funding acquisition, Project administration, Resources, Supervision, Writing—review & editing. Funding: This study was financed in part by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior-Brasil (CAPES)-Finance Code 001, CNPq (303713/2014-3) and FAPERJ (E-26/2002.800/2017, E-26/200.930/2017). Conflicts of Interest: All authors declare that there is no conflict of interest. References 1. 2. 3. 4. Chakrabarti, A.; Bonifaz, A.; Gutierrez-Galhardo, M.C.; Mochizuki, T.; Li, S. Global epidemiology of sporotrichosis. Med. Mycol. 2015, 53, 3–14. [CrossRef] [PubMed] Gremião, I.D.; Miranda, L.H.; Reis, E.G.; Rodrigues, A.M.; Pereira, S.A. Zoonotic epidemic of sporotrichosis: Cat to human transmission. PLoS Pathog. 2017, 13, e1006077. [CrossRef] [PubMed] Rodrigues, A.M.; de Hoog, G.S.; de Camargo, Z.P. Sporothrix species causing outbreaks in animals and humans driven by animal-animal transmission. PLoS Pathog. 2016, 12, e1005638. [CrossRef] [PubMed] Barros, M.B.L.; Schubach, T.P.; Coll, J.O.; Gremião, I.D.; Wanke, B.; Schubach, A. Esporotricose: A evolução e os desafios de uma epidemia. Rev. Panam. Salud Publica 2010, 27, 455–460. [PubMed] Molecules 2019, 24, 2333 15 of 15 5. 6. 7. 8. 9. Gremião, I.D.; Menezes, R.C.; Schubach, T.M.; Figueiredo, A.B.; Cavalcanti, M.C.; Pereira, S.A. Feline sporotrichosis: Epidemiological and clinical aspects. Med. Mycol. 2015, 53, 15–21. [CrossRef] [PubMed] Bustamante, B.; Campos, P.E. Sporotrichosis: A forgotten disease in the drug research. Expert Rev. Anti-Infect. Ther. 2004, 2, 85–94. [CrossRef] [PubMed] Kadavakollu, S.; Stailey, C.; Kunapareddy, C.S.; White, S. Clotrimazole as a cancer drug: A short review. Med. Chem. 2014, 4, 722–724. [CrossRef] Gagini, T.; Borba-Santos, L.P.; Rodrigues, A.M.; Camargo, Z.P.; Rozental, S. Clotrimazole is highly effective in vitro against feline Sporothrix brasiliensis isolates. J. Med. Microbiol. 2011, 66, 1573–1580. [CrossRef] Pai, V.; Ganavalli, A.; Kikkeri, N.N. Antifungal resistance in dermatology. Indian J. Dermatol. 2018, 63, 361–368. [CrossRef] 10. Carvalho, A.L.M.; da Silva, J.A.; Lira, A.A.M.; Conceição, T.M.F.; Nunes, R.S.; Junior, R.L.C.A.; Sarmento, V.H.V.; Leal, L.B.; Santana, D.P. Evaluation of microemulsion and lamellar liquid crystalline systems for transdermal zidovudine delivery. J. Pharm. Sci. 2016, 105, 1–6. [CrossRef] 11. Padula, C.; Telò, I.; Ianni, A.D.; Pescina, S.; Nicoli, S.; Santi, P. Microemulsion containing triamcinolone acetonide for buccal administration. Eur. J. Pharm. Sci. 2018, 115, 233–239. [CrossRef] 12. Rashida, M.A.; Naza, T.; Abbasa, M.; Nazirb, S.; Younasa, N.; Majeeda, S.; Qureshic, N.; Akhtard, M.N. Chloramphenicol loaded microemulsions: Development, characterization and stability. Colloid Interface Sci. Commun. 2019, 28, 41–48. [CrossRef] Seok, S.H.; Lee, S.-A.; Park, E.-S. Formulation of a microemulsion-based hydrogel containing celecoxib. J. Drug Deliv. Sci. Technol. 2018, 43, 409–414. [CrossRef] 13. 14. Kumar, S.K.; Dhancinamoorthi, D.; Sarvanan, R.; Gopal, U.K.; Shanmugam, V. Microemulsions as a carrier for novel drug delivery: A review. Int. J. Pharm. Sci. Rev. Res. 2011, 10, 37–45. 15. Hu, X.-B.; Kang, R.-R.; Tang, T.-T.; Li, Y.-J.; Wu, J.-Y.; Wang, J.-M.; Liu, X.-Y.; Xiang, D.-X. Topical delivery of -trimethoxy-trans-stilbene-loaded microemulsion-based hydrogel for the treatment of osteoarthritis in 3,5,4 a rabbit model. Drug Deliv. Transl. Res. 2019, 9, 357–365. [CrossRef] (cid:48) 16. Hájková, R.; Sklenárová, H.; Matysová, L.; Svecová, P.; Solich, P. Development and validation of HPLC method for determination of clotrimazole and its two degradation products in spray formulation. Talanta 2007, 73, 483–489. [CrossRef] 17. Abdel-Moety, E.M.; Khattab, F.I.; Kelani, K.M.; AbouAl-Alamein, A.M. Chromatographic determination of clotrimazole, ketoconazole and fluconazole in pharmaceutical formulations. Farmaco 2002, 57, 931–938. [CrossRef] 18. Bharate, S.S.; Bharate, S.B.; Bajaj, A.N. Interactions and incompatibilities of pharmaceutical excipientes with active pharmaceutical ingredients: A comprehensive review. J. Excipients and Food Chem. 2010, 1, 3–26. [CrossRef] 19. Ceschel, G.C.; Badiello, R.; Ronchi, C.; Maffei, P. Degradation of components in drug formulations: A comparison between HPLC and DSC methods. J. Pharm. Biomed. Anal. 2003, 32, 1067–1072. [CrossRef] 20. Deshmukha, P.R.; Gaikwadb, V.L.; Tamanea, P.K.; Mahadikc, K.R.; Purohit, R.N. Development of stability-indicating HPLC method and accelerated stability studies for osmotic and pulsatile tablet formulations of Clopidogrel Bisulfate. J. Pharm. Biomed. Anal. 2019, 165, 346–356. [CrossRef] 21. Lee, T.-K.; Ryoo, S.-J.; Lee, Y.-S. A new method for the preparation of 2-chlorotrityl resin and its application to solid-phase peptide synthesis. Tetrahedron Lett. 2007, 48, 389–391. [CrossRef] 22. Nandi, I.; Bari, M.; Joshi, H. Study of isopropyl myristate micremulsion systems containing cyclodextrins to improve the solubility of two model hydrophobic drugs. AAPS PharmaSciTech. 2003, 4, 1–9. [CrossRef] Sample Availability: Samples of the compounds are available from the authors. © 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
null
10.1093_ijrl_eeac041.pdf
null
null
Utterly unbelievable: the discourse of ‘fake’ SOGI asylum claims as a form of Utterly unbelievable: the discourse of ‘fake’ SOGI asylum claims as a form of epistemic injustice epistemic injustice Nuno Ferreira Publication date Publication date 01-02-2023 Licence Licence This work is made available under the CC BY 4.0 licence and should only be used in accordance with that licence. For more information on the specific terms, consult the repository record for this item. Document Version Document Version Published version Citation for this work (American Psychological Association 7th edition) Citation for this work (American Psychological Association 7th edition) Ferreira, N. (2023). Utterly unbelievable: the discourse of ‘fake’ SOGI asylum claims as a form of epistemic injustice (Version 1). University of Sussex. https://hdl.handle.net/10779/uos.23484497.v1 Published in Published in International Journal of Refugee Law Link to external publisher version Link to external publisher version https://doi.org/10.1093/ijrl/eeac041 Copyright and reuse: Copyright and reuse: This work was downloaded from Sussex Research Open (SRO). This document is made available in line with publisher policy and may differ from the published version. Please cite the published version where possible. Copyright and all moral rights to the version of the paper presented here belong to the individual author(s) and/or other copyright owners unless otherwise stated. For more information on this work, SRO or to report an issue, you can contact the repository administrators at sro@sussex.ac.uk. Discover more of the University’s research at https://sussex.figshare.com/ International Journal of Refugee Law, 2022, Vol XX, No XX, 1–30 https://doi.org/10.1093/ijrl/eeac041 A RT I C L E S Utterly Unbelievable: The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice Nuno Ferreira* A B ST R A CT Media and political debates on refugees and migration are dominated by a discourse of ‘fake’ and ‘bogus’ asylum claims. This article explores how this discourse affects in acute ways those people claiming asylum on grounds of sexual orientation or gender identity (SOGI). In particular, the article shows how such a discourse of ‘fakeness’ goes far beyond the well-documented and often inadequate credibility assessments carried out by asylum authorities. By framing the analysis within the context of the scholarship on epistemic injustice, and by drawing on a large body of primary and secondary data, this article reveals how the discourse of ‘fake’ SOGI claims permeates the conduct not only of asylum adjudicators, but also of all other actors in the asylum system, including non-governmental organizations, support groups, legal representatives, and even asylum claimants and refugees themselves. Following from this theoretically informed exploration of primary data, the article concludes with the impossibility of determining the ‘truth’ in SOGI asylum cases, while also offering some guidance on means that can be employed to alleviate the epistemic injustice produced by the asylum system against SOGI asylum claimants and refugees. * Professor of Law, University of Sussex, United Kingdom. This contribution has been produced in the context of the ‘Sexual Orientation and Gender Identity Claims of Asylum: A  European Human Rights Challenge’ (SOGICA Project) (<https://www.sogica.org>). The Project received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (Grant Agreement No 677693). The author wishes to thank Carmelo Danisi, Moira Dustin, Nina Held, Charlotte Skeet, Bal Sokhi-Bulley, and Christina Miliou Theocharaki, as well as the anonymous journal reviewers, for their constructive feedback on earlier drafts. © The Author(s) 2023. Published by Oxford University Press. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. • 1 l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 2 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice 1. I N T RO D U C T I O N According to the United Nations High Commissioner for Refugees (UNHCR), at the end of 2021, there were 31.7 million refugees and people seeking asylum in the world.1 These individuals face numerous social and legal obstacles to obtaining international protection, including having to demonstrate the credibility of their asylum claim during the adjudica- tion process. It is the nature of refugee status determination procedures that claimants must establish their entitlement to international protection, and that authorities must scrutinize the evidence available. The credibility of asylum claims may be called into question either because different elements of the testimony are not consistent with each other (internal credibility), or the testimony is not consistent with information gathered by the asylum authorities (external credibility).2 While the need for such credibility assessment is not in itself problematic, even before the legal adjudication process starts, claimants are often al- ready labelled as ‘bogus’, their claims are presumed to be ‘fake’, and asylum authorities and the broader public alike adopt a sceptical – even a cynical – mindset.3 People are perceived as ‘potential fraudsters’ as soon as they file their asylum claims and, by assuming that their claims are ‘false’, States maintain control over their borders (for example, to reduce levels of immigration and feed into xenophobic and populist political discourses) without having to question the system of international protection or a State’s democratic credentials within the international community.4 Some researchers argue that decision makers in countries such as Spain, the United States of America (USA), and the United Kingdom (UK) seem to be trained to disbelieve5 and carry out their functions according to an ‘unwritten (meta) message of mistrust’.6 Existing scholarship has thus identified strong elements of willing- ness and consciousness in discrediting asylum claims independently of their merits.7 Discussions about ‘fake’ asylum claims are fuelled by, and contribute towards, broader anti-refugee and anti-migration rhetoric in the media and political debates.8 1 UNHCR, ‘Figures at a Glance’ <https://www.unhcr.org/en-au/figures-at-a-glance.html> ac- cessed 12 September 2022. 2 Gábor Gyulai and others, Credibility Assessment in Asylum Procedures: A Multidisciplinary Training 3 4 5 Manual, vol 1 (Hungarian Helsinki Committee 2013) 31. Jessica Anderson and others, ‘The Culture of Disbelief: An Ethnographic Approach to Understanding an Under-Theorised Concept in the UK Asylum System’ (2014) Refugee Studies Centre Working Paper Series No 102; James Souter, ‘A Culture of Disbelief or Denial? Critiquing Refugee Status Determination in the United Kingdom’ (2011) 1 Oxford Monitor of Forced Migration 48. Cécile Rousseau and Patricia Foxen, ‘Le Mythe du Réfugie Menteur: Un Mensonge Indispensable?’ [The Myth of the Lying Refugee: An Essential Lie?] (2006) 71 L’Evolution Psychiatrique 505, 506–07. Carol Bohmer and Amy Shuman, ‘Producing Epistemologies of Ignorance in the Political Asylum Application Process’ (2007) 14 Identities 603, 615. 6 Olga Jubany, ‘Constructing Truths in a Culture of Disbelief: Understanding Asylum Screening from Within’ (2011) 26 International Sociology 74, 81. Rousseau and Foxen (n 4) 510. 7 8 Gillian McFadyen, ‘The Language of Labelling and the Politics of Hospitality in the British Asylum System’ (2016) 18 British Journal of Politics and International Relations 599, 611–12; Giuseppe Salvaggiulo, ‘La Sentenza della Cassazione: “I Racconti dei Richiedenti Asilo sono Stereotipati e Troppo Simili Tra Loro”’ [The Supreme Court’s Decision: The Testimonies of Asylum Claimants Are Stereotyped and Too Similar to Each Other] La Stampa (16 January 2020) l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 3 of 30 It is also clear that the ‘genuine refugee is discursively constructed in a particular legal, political, and cultural context’.9 This affects in critical ways those asylum claims based on sexual orientation or gender identity (SOGI). SOGI claims require a discrete ana- lysis in this context on account of the particular issues they raise in relation to different aspects of asylum adjudication, especially the need for claimants to prove their SOGI identity, the role of private actors in persecution, the intense social prejudice against SOGI claimants, the role of legislation – namely criminalization – in the country of origin in sanctioning that prejudice, and the particular psychosocial challenges that these claimants face in terms of personal identity and community integration in the host State.10 SOGI claimants are often accused of ‘fabricating’ their stories,11 including in media pieces that build on the assumption that pretending a certain sexual orientation or gender identity is easy for the claimant and a sure-fire way of obtaining international protection.12 This is especially the case where there is evidence of persecution against sexual and gender minorities in particular countries of origin. However, there is no guarantee of ‘automatic protection’ under such circumstances. Claimants must still go through the refugee status determination procedure, and authorities often place particular emphasis on the credibility assessment of SOGI claims. Such an assessment may depend mostly on the claimant’s own testimony – checked against the available country of origin information (COI) – owing to the limited documentary or witness evidence generally available in such cases. Furthermore, the ‘genuineness of a LGBT refugee is prone to constant negotiation and renegotiation dependent on ongoing developments occurring within the wider cultural politics of immigration and global sexual politics’.13 As already explored by several authors, this cynical mindset in relation to SOGI claimants creates a damaging ‘culture of disbelief ’ in asylum authorities in several <https://www.lastampa.it/cronaca/2020/01/16/news/sentenza-choc-della-cassazione-i- racconti-dei-richiedenti-asilo-sono-stereotipati-e-troppo-simili-tra-loro-1.38339774/> accessed 12 September 2022; Mehta Suketu, ‘The Asylum Seeker’ (The New Yorker, 25 July 2011) <https:// www.newyorker.com/magazine/2011/08/01/the-asylum-seeker> accessed 12 September 2022. 9 Deniz Akin, ‘Discursive Construction of Genuine LGBT Refugees’ (2018) 23 Lambda Nordica 21, 23. 10 Nuno Ferreira, ‘Sexuality and Citizenship in Europe: Sociolegal and Human Rights Perspectives’ (2018) 27 Social and Legal Studies 253, 254. 11 Rousseau and Foxen (n 4). 12 Dan Bilefsky, ‘Gays Seeking Asylum in US Encounter a New Hurdle’ The New York Times (29 January 2011)  <https://www.nytimes.com/2011/01/29/nyregion/29asylum.html> accessed 12 September 2022; Francesca Ronchin, ‘Permessi di Soggiorno per i Migranti, L’Escamotage dell’Orientamento Sessuale’ [Residence Permits for Migrants, the Deception of Sexual Orientation] Corriere della Sera (23 October 2019)  <https://www.corriere.it/video- articoli/2019/10/23/permessi-soggiorno-migranti-l-escamotage-dell-orientamento-sessuale/ ece27a72-e52c-11e9-b924-6943fd13a6fb.shtml> accessed 12 September 2022. Most of the primary data and secondary sources explored in this article refer more explicitly to sexual orientation but also hold relevance in relation to gender identity, hence the scope of the article encompassing both. 13 Akin (n 9) 36. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 4 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice countries.14 In 2001, when deciding a SOGI asylum claim, a UK judge raised the possibility of ‘encouraging a flood of fraudulent Zimbabwean (and no doubt other) asylum-seekers posing as sodomites’.15 Although we have come a long way since then, an ingrained concern persists that SOGI asylum claimants may be lying about their stories. Although SOGI claims may not be statistically more prone to being used in a deceptive way,16 and while acknowledging that they may indeed be used in a deceptive way, SOGI claimants are deeply affected by the scepticism that accompanies their asylum claims. Despite this culture of disbelief being well known to scholars, policymakers, and refugees, there is limited research on what makes SOGI claims – or claimants – so un- believable as to render them ‘fake’ in the eyes of decision makers, especially in light of the thorough, objective, individualized, and sensitive process that is required to assess their claims.17 It is crucial to explore in an in-depth manner the mechanisms behind such presumptions of ‘fakeness’. This article does so through a novel, theoretically and empirically informed analysis that examines all actors in the asylum system. The analysis reveals that the discourse of ‘fake’ SOGI claims not only strongly influences asylum au- thorities (often under political pressure to refuse claims, or hardened by listening to so many terrible stories) and the wider public (influenced by populist, racist, and homo/ transphobic social trends), but also affects the most unlikely stakeholders: on the one hand, non-governmental organizations (NGOs), support groups, and legal representa- tives take it upon themselves to filter out ‘fake’ claims from the asylum system, and, on the other hand, other SOGI claimants and refugees consider it necessary to themselves identify ‘fake’ claimants in order to contribute to the groups that support them and to protect the chances of future ‘genuine’ SOGI asylum claimants obtaining international protection. This article offers a theoretically informed analysis of these dynamics by engaging with this subject matter from the perspective of the body of literature on epistemic in- justice. The analysis is also empirically informed, drawing extensively on primary data collected through fieldwork carried out in several locations in Europe between 2017 14 Carmelo Danisi and others, Queering Asylum in Europe: Legal and Social Experiences of Seeking International Protection on Grounds of Sexual Orientation and Gender Identity (Springer 2021) ch 7; Agathe Fauchier, ‘Kosovo: What Does the Future Hold for LGBT People?’ (2013) 42 Forced Migration Review 36, 38; Theo Gavrielides and others, ‘Supporting and Including LGBTI Migrants: Needs, Experiences and Good Practices (Epsilon Project)’ (IARS International Institute 2017); Jenni Millbank, ‘From Discretion to Disbelief: Recent Trends in Refugee Determinations on the Basis of Sexual Orientation in Australia and the United Kingdom’ (2009) 13 International Journal of Human Rights 391. 15 Z v Secretary of State for the Home Department [2001] UKIAT 01TH02634, para 4. 16 John Vine, ‘An Investigation into the Home Office’s Handling of Asylum Claims Made on the Grounds of Sexual Orientation: March–June 2014’ (Independent Chief Inspector of Borders and Immigration 2014) para 5.21. 17 UNHCR, ‘Guidelines on International Protection No 9: Claims to Refugee Status Based on Sexual Orientation and/or Gender Identity within the Context of Article 1A(2) of the 1951 Convention and/or Its 1967 Protocol relating to the Status of Refugees’, HCR/GIP/12/09 (23 October 2012) para 62. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 5 of 30 and 2019.18 This fieldwork – carried out in the context of the ‘Sexual Orientation and Gender Identity Claims of Asylum’ (SOGICA Project) – concentrated on Council of Europe and European Union (EU) institutions, and the countries of Germany, Italy, and the UK. It included: 143 semi-structured interviews with SOGI asylum claimants and refugees, NGOs, policymakers, decision makers, members of the judiciary, legal repre- sentatives, and other professionals; 16 focus groups with SOGI asylum claimants and refugees; 24 non-participant contextual observations of court hearings; two online sur- veys of SOGI asylum claimants and refugees and professionals working with them; and freedom of information requests relating to case studies lodged in all three countries. In order to ensure anonymity, respect participants’ agency, and distinguish between the sources, the article uses sources in the following ways: individuals are referred to either by their first name or by a pseudonym (according to their stated preference); references note the capacity in which participants were interviewed and the country in which they were based (if no capacity is specified, then the participant was an asylum claimant or a legally recognized refugee); focus groups are identified by their number and location; court hearings are identified by the level of the court, its broad geographical location, and the year in which the hearing took place; and survey respondents are referred to by a letter (S for ‘supporter’ and C for ‘claimant’) and a numerical identifier.19 The article begins with a discussion of the theoretical framework on which the sub- sequent analysis relies, with an emphasis on the relevance of the scholarship on epi- stemic injustice for asylum law and policy (part 2). In part 3, the analysis of the primary data begins by exploring how epistemic injustice operates during the asylum adjudica- tion process, and how epistemic injustice is produced by asylum decision makers. In part 4, the focus shifts to the roles of NGOs, support groups, and legal representatives, as well as asylum claimants and refugees themselves, who are often ignored in such de- bates but are undoubtedly also key actors in the discourse of ‘fake’ claims, as evidenced by the primary data. Part 5 explores key means to address the epistemic injustice pro- duced by the actors discussed in parts 3 and 4, even though achieving the ‘truth’ is ultimately impossible. Finally, part 6 reiterates the need to accept the impossibility of determining the ‘truth’ in SOGI asylum claims and to alleviate the epistemic injustice of the asylum system for SOGI claimants. ‘Fake’ and ‘truth’ are used with quotation marks throughout the article to high- light the impossibility of determining the veracity of claims. Even when a claimant may acknowledge not having a genuine SOGI claim, their sexual orientation or gender identity may, in fact, be relevant to their need for international protection, although the claimant may choose to deny this, owing to emotional, social, or cul- tural factors. 18 Ethics approval was obtained from the University of Sussex (certificate of approval for Ethical Review ER/NH285/1). Written and informed consent was obtained from all the participants. The project – including the collection of empirical data – was carried out by all the team mem- bers: Carmelo Danisi, Moira Dustin, Nuno Ferreira, and Nina Held. For full details of the methodology, see Danisi and others (n 14) ch 2; SOGICA, ‘Fieldwork’ <https://www.sogica.org/en/fieldwork/> accessed 12 September 2022. 19 l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 6 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice 2. C R E AT I N G E P I ST E M I C I N J U ST I C E I N T H E Q U E ST F O R ‘ T R U T H ’ As Foucault’s work so thoroughly explores, the quest for producing ‘truth’ has been central to the production of knowledge in the West – including in relation to sexuality – and is deeply embedded in subjective relationships of power.20 More specifically: Truth is a thing of this world: it is produced only by virtue of multiple forms of constraint. And it induces regular effects of power. Each society has its regime of truth, its ‘general politics’ of truth: that is, the types of discourse which it accepts and makes function as true; the mechanisms and instances which enable one to distinguish true and false statements, the means by which each is sanctioned; the techniques and procedures accorded value in the acquisition of truth; the status of those who are charged with saying what counts as true.21 Similarly, Bourdieu suggests that constructing a discourse as ‘true’ or ‘false’ essentially depends on the power dynamics that underpin social and institutional relationships.22 As explained by Spivak, there are a range of historical and ideological factors that pre- vent those inhabiting the ‘periphery’ – surely including asylum claimants and refugees – from being heard.23 All these scholarly contributions point to the fact that interper- sonal and institutional ‘power’ is a factor that shapes how we produce ‘truths’ and ‘lies’. Moreover, ‘truths’ and ‘lies’ are not produced according to what is ‘true’ or ‘false’ (if it were ever possible to determine this), but according to what is convenient, to order events around conformity and deviance.24 Consequently, epistemic injustice – under- stood here as injustice in the context of the production of knowledge – is rife in any system of ‘truth production’. In other words, no matter how a society produces know- ledge, there is bound to be unfairness as to who decides what is true or not, and how this is done. In the context of asylum law and policy, this includes two main forms of injustice: testimonial injustice and contributory injustice. On the one hand, testimonial injustice occurs when ‘prejudice causes a hearer to give a deflated level of credibility to a speaker’s word’,25 with such prejudice operating in relation to all different spheres of life that may affect a person’s social identity in the mind of the hearer. This entails a symbolic degradation, namely the listener undermines the other’s humanity,26 and oppresses the other by diminishing their self-confidence and thwarting their development.27 On the other hand, building on Pohlhaus’s work on 20 Michel Foucault, The History of Sexuality, vol 1 (Penguin Books 1990). 21 Michel Foucault, The Foucault Reader (Penguin Books 1991) 72–73. 22 Pierre Bourdieu, Language and Symbolic Power (Polity Press 1993). 23 Rosalind Morris (ed), Can the Subaltern Speak? Reflections on the History of an Idea (Columbia University Press 2010); Gayatri Chakravorty Spivak, ‘Can the Subaltern Speak?’ in Cary Nelson (ed), Marxism and the Interpretation of Culture (Macmillan Education 1988). 24 Michel de Certeau, Histoire et Psychanalyse Entre Science et Fiction (Gallimard 1987). 25 Miranda Fricker, Epistemic Injustice: Power and the Ethics of Knowing (Oxford University Press 2007) 1. ibid 44. ibid 58. 26 27 l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 7 of 30 ‘willful hermeneutical ignorance’,28 Dotson sees contributory injustice ‘as the circum- stance where an epistemic agent’s willful hermeneutical ignorance in maintaining and utilizing structurally prejudiced hermeneutical resources thwarts a knower’s ability to contribute to shared epistemic resources within a given epistemic community by com- promising her epistemic agency’.29 Nonetheless, epistemic injustice (also) derives from the fact that ‘institutions struc- ture interactions according to cultural norms that impede parity of participation’.30 As Doan explains, this prevents ‘people from testifying and being heard, asking relevant questions, contesting claims and standards of evidence, and otherwise participating in everyday epistemic practices as peers’31 – something that is directly relevant to the asylum system. Consequently, Doan submits that ‘epistemic injustice ought to be under- stood as rooted in the oppressive and dysfunctional epistemic norms undergirding ac- tual communities and institutions’.32 As such, struggles for epistemic recognition require changes not only at the individual level but also at the social and institutional levels. The responsibility and the initiative for undoing epistemic injustice rest not only with single individuals but with all actors in the system, without ‘occluding the agency and resistance of victims’.33 This is of direct relevance for present purposes, since all actors in the asylum system contribute to epistemic injustice which, in turn, affects SOGI asylum claimants and refugees. In fact, a transformative strategy that is able to ‘correct unjust outcomes precisely by restructuring the underlying generative framework’ may be required.34 Asylum systems are textbook examples of how the State can devise and operation- alize repressive and flawed epistemic norms. States deploy political technologies to govern the movement and conduct of refugees, namely by determining which ones are ‘bogus refugees’ and which ones are ‘persons in real need of protection’.35 Looking at asylum systems through a Foucauldian and Fanonian lens, Lorenzini and Tazzioli adopt poststructural and decolonial prisms to highlight how: the question of (the production of) truth is at the core of the mechanisms of sub- jection and subjectivation which are at stake in the processing of asylum claims. Asylum seekers are usually seen as suspect subjects who have to demonstrate that 28 Gaile Pohlhaus, ‘Relational Knowing and Epistemic Injustice: Toward a Theory of Willful Hermeneutical Ignorance’ (2012) 27 Hypatia 715. 29 Kristie Dotson, ‘A Cautionary Tale: On Limiting Epistemic Oppression’ (2012) 33 Frontiers: A Journal of Women Studies 24, 32. 30 Nancy Fraser, ‘Social Justice in the Age of Identity Politics: Redistribution, Recognition, and Participation’ in Nancy Fraser and Axel Honneth (eds), Redistribution or Recognition? A Political- Philosophical Exchange (Verso Books 2003) 29. 31 Michael Doan, ‘Resisting Structural Epistemic Injustice’ (2018) 4(4) Feminist Philosophy Quarterly 13. ibid 15. ibid 8 (emphasis in original). Fraser (n 30) 74. 32 33 34 35 Daniele Lorenzini and Martina Tazzioli, ‘Confessional Subjects and Conducts of Non-Truth: Foucault, Fanon, and the Making of the Subject’ (2018) 35 Theory, Culture & Society 71, 72. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 8 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice they really are in need of protection; yet, at the same time, they are considered as subjects incapable of telling the truth.36 In this process, more than ‘truth’, we are in the presence of the ‘production of ignor- ance’,37 showing that both ‘truth’ and ‘fakeness’ are discursively constructed. During the production of knowledge in the asylum system, there is a clear ‘struggle over truth’.38 By default, asylum systems privilege the epistemic resources of decision makers over claimants, thus legitimizing the former’s prerogative to ‘arbitrarily and am- biguously misinterpret asylum applicants’ experiences, cultures, and countries’ – the so-called ‘institutional comfort’ enjoyed by decision makers.39 In the asylum context, this institutional comfort translates into testimonial injustice in the form of denying ap- plicants’ experiences, ignoring available information, and deciding which information or criteria to use. Simultaneously, the asylum system is characterized by contributory injustice in the form of knowingly and voluntarily employing prejudiced hermeneut- ical resources to undermine the epistemic agency of the claimants.40 Testimonial and contributory injustice combined produce a powerful version of epistemic injustice in asylum systems. In the midst of such an epistemologically unfair system, asylum claimants may find themselves both dehumanized and ignored. Doubting the truth of the claimant is a vio- lence perpetrated against them, which produces and increases their (narrative) vulner- ability, and constitutes a form of epistemological and symbolic violence.41 At the same time, decision makers may see their personal experiences as universal and therefore suitable to be used as the basis for judging the veracity of claimants’ testimonies.42 As Jubany concluded from her research in Spain and the UK, based on decision makers’ ‘professional knowledge’, Chinese claimants are held to be untrustworthy, African claimants are perceived as liars, those from the Indian subcontinent are accused of being incoherent and using artificial stories, and those from Turkey are judged as cun- ning and exaggerated.43 ‘Intuition’, having a ‘feeling’, ‘just knowing’, or a certain ‘look’ are seen as legitimate means to determine the truthfulness of a claimant’s story and are used as justification for denying international protection.44 Even worse, the use of accelerated procedures (often coupled with the contested notion of ‘safe country’)45 36 ibid 72 (citations omitted). 37 Bohmer and Shuman (n 5). 38 Lorenzini and Tazzioli (n 35) 82. 39 Ezgi Sertler, ‘The Institution of Gender-Based Asylum and Epistemic Injustice: A  Structural Limit’ (2018) 4(3) Feminist Philosophy Quarterly 3. ibid 2, 16. 40 41 Massimo Prearo, ‘The Moral Politics of LGBTI Asylum: How the State Deals with the SOGI Framework’ (2020) 34 Journal of Refugee Studies 1454. 42 Rousseau and Foxen (n 4) 511. 43 Jubany (n 6) 83–84. ibid 86–87; Rousseau and Foxen (n 4) 516. 44 45 Danisi and others (n 14) ch 6.7. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 9 of 30 has rendered claimants’ speech ‘increasingly irrelevant’,46 depriving them of the oppor- tunity to fully articulate their experiences and fears of persecution.47 Reaching an ‘objective truth’ is not achievable, just as proving that a claim is ‘fake’ is not possible.48 In other words, ‘the pretense of judgment based on evidence obscures the real problem of the unavailability of necessary information’.49 Barsky notes that ‘we cannot employ the tools of discourse analysis, no matter how sophisticated, to distin- guish between truthful and untruthful statements in refugee hearings, except at a very superficial level’.50 ‘Fake’ claims are thus discursively produced: it is the discourse cre- ated by all the actors involved that labels claims as ‘fake’ and forms the subject position of the ‘fake’ claimant. This is true for SOGI claims as well: it is not possible to reach an ‘objective truth’ about them but, in the face of the ‘practical decisionism’ that asylum authorities face, ‘the various organizations and persons that claim that it is impossible to evaluate legitimately the truths of LGBT-ness are unsuccessful’.51 Historically, mem- bers of SOGI minorities had to hide their true identity and desires – and so society was full of ‘fake heterosexuals’ – but now, in a sort of inversion of the ‘politics of truth’, the fear is one of ‘fake homosexuals’.52 In this tangled web of the ‘politics of truth’, decision makers and other actors may overlook the fact that both sexual orientation and gender identity are socially constructed, culturally heterogeneous, fluid, complex, performed, and negotiated categories.53 A greater awareness of the nature of sexual orientation and gender identity would facilitate asylum decisions that more sensitively and accurately engage with SOGI claims, in ways that are also more socially and culturally appropriate. In a Foucauldian sense, the ‘fake’ SOGI claim and ‘fake’ SOGI claimant’s subject position are (also) discursively produced, thus constituting a sub-category of ‘fake’ claims. As a consequence, ‘only those whose sexual and gender practices are intelligible according to hegemonic gender and sexuality norms can become eligible for permitted border-crossing’, thus further entrenching the fixed, homonormative sexual ontologies 46 Lorenzini and Tazzioli (n 35) 82. 47 An infamous version of this phenomenon can be seen in the UK’s Detained Fast Track system for detained individuals, whereby people were deported without being given the opportunity to appeal against negative Home Office decisions. The system was declared unlawful by the High Court in Detention Action v First-tier Tribunal (Immigration and Asylum Chamber) [2015] EWHC 1689 (Admin). The negative practical consequences of such systems are illustrated in the case of PN, a Ugandan lesbian claimant: see PN (Uganda) v Secretary of State for the Home Department [2020] EWCA Civ 1213. 48 Rousseau and Foxen (n 4) 518. 49 Bohmer and Shuman (n 5) 622. 50 Robert F Barsky, Arguing and Justifying: Assessing the Convention Refugees’ Choice of Moment, Motive and Host Country (Ashgate Publishing 2000) 14. 51 Maja Hertoghs and Willem Schinkel, ‘The State’s Sexual Desires: The Performance of Sexuality in the Dutch Asylum Procedure’ (2018) 47 Theory and Society 691, 697. 52 Eric Fassin and Manuela Cordero Salcedo, ‘Becoming Gay? Immigration Policies and the Truth of Sexual Identity’ (2015) 44 Archives of Sexual Behavior 1117, 1121. ibid 1121–24. 53 l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 10 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice that underlie the asylum system.54 The asylum system adopts a ‘privileged configur- ation of sexual orientation [that] reflects a particular historical configuration of gen- dered, raced and classed interests and experiences’.55 Moreover, while not every denial of international protection to a SOGI claimant is an instance of epistemic injustice, the asylum adjudication process becomes a ‘test of sexual veracity by means of a truthful performance’, on the basis of the ‘facticity of sexuality’, thereby legitimizing and sanc- tioning certain gender and sexuality performances but not others.56 The following parts of the article explore how all actors in the asylum system play a role in the ‘politics of truth’ of SOGI claims. 3. T H E ‘ U N T R U T H ’ O F T H E A S Y LU M A D J U D I C AT I O N P RO C E D U R E The evidence examined for this article revealed that at both an administrative and ju- dicial level, there is significant institutional comfort relating to SOGI-based asylum claims (see part 2). Decision makers may not only be sceptical about such claims, but may deny that there is any ‘truth’ to them. Through their disbelief, decision makers exer- cise their power to produce testimonial injustice and reduce the humanity of claimants. As Victor – a SOGI asylum claimant participant in the UK – put it, decision makers: wouldn’t want to listen to you. … If you try to explain something [to] the person, it is like you are offending them for you being there to, you know, to understand for them, you are already offending them [and] everything you are saying is not true.57 Decision makers’ role in the production of epistemic injustice is also apparent in their inclination to believe that a SOGI claim is ‘fake’ when there is simply an increase in the number of such claims.58 For example, Titti, a decision maker in Italy, spoke of ‘huge peaks’ in SOGI claims, of having heard about 15 such claims in one month in an Italian region, which prompted her to examine them more carefully. Bilal, a UK Home Office presenting officer, also expressed scepticism after an increase in SOGI claims: ‘I think I have had some cynicism … the gay Pakistani cases, because there seemed suddenly to be a huge raft and they all had very similar narratives’. Similarly, in Germany, an NGO participant reported that even gay decision makers were ‘extremely suspicious’ about a rise in SOGI asylum claims, thus leading to an increase in the number of rejections59 and demonstrating decision makers’ power to deny the ‘truth’ of claimants’ testimonies. 54 Mariska Jung, ‘Logics of Citizenship and Violence of Rights: The Queer Migrant Body and the Asylum System’ (2015) 3 Birkbeck Law Review 311, 324. 55 David AB Murray, ‘Real Queer: “Authentic” LGBT Refugee Claimants and Homonationalism in the Canadian Refugee System’ (2014) 56 Anthropologica 21, 22. 56 Hertoghs and Schinkel (n 51) 691, 693. 57 Focus Group No 2, Glasgow, UK. 58 Celeste, social worker, Italy. Several participants described them as ‘fashionable stories’. 59 Thomas, NGO volunteer, Germany. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 11 of 30 Another recurrent theme in the discourse of ‘fake’ claims is the degree of similarity between different claimants’ testimonies. Maria Grazia, a decision maker in Italy, be- came aware of this soon after assuming her role: I realised how much the SOGI element is exploited. It is not a perception induced by a particularly backward policy from a certain political field. This element is really used to get protection … Yes, when I realised that the stories are all similar. … [T]he first time I made an appeal before a Court, I had an asylum claimant who had brought me a page from a newspaper in Nigeria where there was a photo of a man on the ground full of blood and a photo of the applicant, wanted for homosexuality. And I thought ‘Damn, how will the judge not believe this story? It is also in the newspaper’. And in the commission they told me ‘Look, these are photomontages and in reality the story they bring is always this: relationship with the partner, partner killed because of being homosexual, escape …’ And the grim, particularly violent element is always added in. Similarly, a German judge, Oscar, said that: the more you have listened to asylum claimants from a country, the sooner you will notice whether this really happens [claimants using fake stories] or if that is more likely. These are stories that are passed on from asylum claimant to asylum claimant and which they always try to use here [in court]. So, typical stories. A similarity between stories can, however, also be due to legal representatives some- times promoting ‘pre-prepared’ stories to their clients,60 which can lead to more rejec- tions by the authorities. In any case, it is clear that such similarities prompt decision makers to use their power to undervalue testimonies and interpret evidence in a way that undermines it, thus producing testimonial and contributory injustice. Interestingly, unique stories are also often seen as questionable, as they do not fit the scenarios familiar to decision makers.61 For example, Sofia and Emma, NGO workers in Germany, explained that asylum authorities may reject the ‘truth’ of a claimant’s tes- timony simply because it is different from other asylum claims: one [woman] who has experienced forced prostitution in China, so from Uganda to China, then she had different [experiences], then fled to other African coun- tries, where she was raped, and then [fled] again to Germany, where she has been almost forcibly prostituted. And … she is also lesbian, and with her partner, so to speak, and different things … escaped, and so, for the Federal Office, this is so blatant that it cannot be credible. The perfect fit of testimonies with publicly known events or common perceptions of SOGI minorities is also a reason for decision makers to label a story as ‘fake’ and deny 60 This has been observed in the Canadian context, for instance. See Rousseau and Foxen (n 4) 513. 61 ibid. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 12 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice the claimant’s epistemic agency. Italian decision maker Roberto explained how, at a training session, [w]e projected images of public facts, of things of this type [described in a case study] that happened in Nigeria, to show these things may happen … the classic fake claim is produced like this, put together subsequently. That is, people know that these things happen in their country, they tell you with extreme precision what they read in a newspaper or they heard in their communities, but there is no … ‘And have you had problems with your family? How did you live it?’, ‘Well’, ‘Are you in touch with your father or your mother?’, ‘Yes’, ‘And what do they tell you?’, ‘Nothing’ … Everything is missing. The experience is an individual experi- ence, unique, not repeatable but cannot be devoid of any form of perception. In the UK, there are also concerns that SOGI claims may be ‘fake’ when claimants cor- respond ‘too neatly’ to SOGI stereotypes: I think it is possibly the case that the people who see an advantage in making a claim based on [sexual] orientation will not really understand what [sexual] orientation is about, and will … go in for a stereotypical presentation. Doesn’t mean to say that what could be perceived as stereotypical may not actually be someone’s choice, they may wish to advertise themselves in some way, but that is one type of thing, I think, which would tend to indicate … a claim that didn’t have any sort of substance to it.62 It is a clear illustration of the discursive production of sexual orientation and gender identity that claimants are expected to fit Western stereotypes of what being an ‘out and proud’ LGBTIQ+ person means.63 At the same time, however, they must not fit those stereotypes too neatly or they will be accused of ‘faking’ their stories.64 Claimants from certain countries of origin seem to be regarded with particular scep- ticism by decision makers, who may use their institutional comfort to deny the ‘truth’ of those claimants’ testimonies. For example, Barbara, a lawyer in Germany, asserted that decision makers have basic prejudices against some countries of origin and as- sume that claims from those countries are always fabricated. These countries include Cameroon, Eritrea, Ethiopia, Nigeria, and The Gambia.65 As Daniele, a decision maker in Italy, explained: I believe so, that there is an X number of [fake] claims, more or less significant depending on the country [of origin], because there are countries – and this is known informally – or nationalities in relation to which the simple fact of 62 Adrian, judge, UK. 63 The acronym LGBTIQ+ stands for lesbian, gay, bisexual, trans, intersex, queer, and others. 64 Danisi and others (n 14) ch 7.5. 65 Barbara, lawyer, Germany; Chiara, NGO worker, Italy; Celeste and Susanna, social workers, Italy; Damiano, lawyer, Italy; Diego, Giulia, Giulio, Jonathan, and Riccardo, LGBTIQ+ group volunteers, Italy; Emilia, judge, Germany; Nelo, Italy; Roberto, decision maker, Italy. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 13 of 30 presenting this [SOGI] claim could be a disgrace, so it is very difficult for one to do it falsely. That is, it is very difficult for a Malian to present a claim based on sexual orientation falsely, if he is not homosexual. Because in this environment, from a cultural point of view, the origin, etc, it’s really a heavy thing. Instead, there are [countries of] origin for which the problem is minor. For Nigerians, for ex- ample, this type of claim is made with greater ease, even motivating one’s sexual orientation in a somewhat extravagant way … in sum, I must tell you the truth. l D o w n o a d e d f r o m h t t Daniele acknowledged that asylum claimants from some African countries are unlikely to ‘fake’ a SOGI claim, as there is enormous stigma associated with being a member of a SOGI minority, potentially even leading to exclusion from the diaspora community. Yet, the discourse of ‘fake’ claims persists in relation to some countries. Italian decision maker Roberto shared his scepticism about Nigerian claimants claiming to be gay: since I’m here, I have only heard a Turkish national claiming asylum for being transgender … a Somali national for being homosexual, no one from Eritrea. It’s clear that the great weight [in SOGI asylum] of some nationalities [like Nigerians] makes you be more doubtful. Similarly, Filippo, a senior judge in Italy, commented that some colleagues do not wish to listen to asylum claimants because they sell each other ‘absurd stories’, especially when they arrive from particular countries, such as Nigeria. This inclination to sus- pect the ‘fakeness’ of SOGI claims relating to certain countries of origin can worsen when decision makers are mainly, or only, allocated claims from certain geographical areas,66 and has a clear gendered dimension, as illustrated by this example relating to Nigerian women: If you come from Nigeria or come from Benin City, you are 100 per cent a victim of trafficking. So whatever you say about why you ran away, the commission will use the lens of trafficking. And therefore it [the claim] is considered untrue, be- cause you are a victim of trafficking.67 Julian, a SOGI asylum claimant in Germany, also spoke about the bias German de- cision makers frequently show towards female claimants from Uganda: ‘My interviewer was really biased. I entered and he said “Oh, you’re from Uganda, I guess you’re now going to tell me that lesbian story”. Before I could even start’. Such outright denial of claimants’ truthfulness on the basis of their country of origin evidences both testimo- nial and contributory injustice. Epistemic injustice is increased by the fact that, in practice, the discourse of ‘fake’ claims also seems to raise the standard of proof, as decision makers appear to require further evidence to ensure the claimant is not fabricating their story.68 Bilal, a UK Home Office presenting officer, expressed such concern: 66 Rousseau and Foxen (n 4) 517. 67 Celeste, social worker, Italy. 68 Silver, Italy. p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 14 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice some people are exploiting the lack of evidence because you don’t need to produce any, so you can pretend to be, say, a gay man or woman, and be suc- cessful because you don’t need to produce any evidence. So there is, there is an avenue for … you know, because of that hole in the system being exploited. Although asylum claimants (SOGI or otherwise) do need to produce evidence to sup- port their claims,69 the perception that it is easy to succeed in (unsubstantiated) SOGI claims seems to be in the mind of this official. In a more extreme example of testimonial injustice and abuse of institutional com- fort, a judge during a 2018 court observation in Hesse, Germany, asserted at the begin- ning of the hearing that he did not believe the claimant, and intimidated a supporting witness by telling him that he could receive a 12-month prison sentence if he provided false information. For the judge, the claimant’s story was not credible: ‘This story is so deceitful, it’s unbelievable! He has five children and tells me that he is gay all the way! That is unbelievable!’ The assumption that a gay man could not biologically father chil- dren dominated the judge’s thinking, reflecting a stereotypical view that pervaded the appeal hearing with a presumption of ‘fakeness’. The concern that witnesses may contribute to ‘fake’ claims was also highlighted by judges during the fieldwork, rendering witnesses victims of testimonial injustice as well. For example, a judge in the UK stated: One issue we have had is witnesses who’ve given evidence in other cases … this can mean they are active in their own community but can lead to witnesses for hire. We had a situation [a couple of years ago] of claimants from Pakistan and [the] same witnesses came along … Then another issue is social media conver- sations … usually the other person isn’t called as witness, usually they say they don’t know where the person is, but this is evidence that I had a relationship with X. The problem is that falls foul of [the] view that we decide on the basis of oral evidence and if you can’t cross-examine, how much weight can you put on it?70 The emphasis on oral evidence, despite the availability of other (written) evidence, is detrimental to SOGI claimants, as many potential supporting witnesses may not wish to offer oral evidence for fear of ‘coming out’ and being exposed to harm, stigma, or discrimination. It is a form of contributory injustice that becomes even more worrying when the skin colour of witnesses influences judges’ assessments of the genuineness of the claims. As an NGO volunteer in the UK observed: ‘If you take lots of witnesses to court, if they are white and middle class, they are believed’. Conversely, in a case relating to two Pakistani claimants, the judge said that a Pakistani couple were not ‘worth much’ as witnesses.71 69 In the UK, for example, claimants are expected to submit evidence to support a sexual orienta- tion claim, even if just in the form of an oral testimony: UK Home Office, ‘Asylum Interviews’ (Version 7.0, 2019) 31–32. 70 Ernest, judge, UK. 71 Joseph, NGO volunteer, UK. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 15 of 30 Overall, there are clear signs that judges often believe that SOGI claims are fabri- cated, rendering the judges key actors in the epistemic injustice that entraps SOGI asylum claimants: ‘evidence doesn’t seem to persuade some judges at all’.72 Yet, there are also positive examples of judges who refuse to reproduce prejudices against such claim- ants or to contribute to the discourse of ‘fake’ claims. For instance, during an appeal hearing observed in the UK, a judge reassured the appellant that ‘the fact that you’ve had a son doesn’t mean you’re not a lesbian’.73 Silvana, a judge in Italy, suggested that the polemics of ‘fake’ claims are exaggerated and stereotypical, fuelled by the media. As she put it, we should be more concerned about the persecution and discrimination experi- enced by SOGI minorities around the world: It is absolutely normal that you go to a country where homosexuality is not a crime from a country in which it is a crime. Instead, the question that should be asked is how come so many countries still criminalise homosexuality. If there were not so many countries criminalising homosexuality, there would be far fewer requests for protection, I believe. The experiences shared by participants reflect serious degrees of testimonial and con- tributory injustice in the refugee status determination process. However, as the next part of the article shows, decision makers are not the only actors in the asylum system who determine which SOGI claims are seen as ‘true’ and which are seen as ‘fake’. 4. ‘FA K E’ C L A I M S D I S CO U R S E A M O N G ST C I V I L S O C I ET Y   A CTO R S Civil society actors – understood here as the range of non-governmental actors active in the field of asylum,74 including NGOs, support groups, and legal representatives, as well as claimants and refugees themselves – also play a role in the power dynamics that shape the discursive construction of what is ‘true’ or ‘fake’ in SOGI asylum claims. While activists ‘contest the sexual and territorial borders’, they also ‘unwillingly con- tribute to their re-inscription’, thus becoming ‘border performers’ and reinforcing State formations.75 McGuirk similarly asserts that NGOs, while ‘ostensibly resisting these constructions, paradoxically create new ones, embedded in wider homonationalist dis- courses that promote a clear victim/savior binary’, mainly owing to the need to attract donations and media attention.76 NGOs working in this field thus dedicate much time and energy to grappling with ‘popular imaginaries’ concerning ‘people pretending to 74 72 Bilal, UK Home Office presenting officer. 73 First-tier Tribunal, London, 2018. Simone Chambers and Jeffrey Kopstein, ‘Civil Society and the State’ in John S Dryzek, Bonnie Honig, and Anne Phillips (eds), The Oxford Handbook of Political Theory (Oxford University Press 2006) 363. Jung (n 54) 333–34. Siobhán McGuirk, ‘Neoliberalism and LGBT Asylum: A Play in Five Acts’ in Siobhán McGuirk and Adrienne Pine (eds), Asylum for Sale: Profit and Protest in the Migration Industry (PM Press 2020) 269. On homonationalism, see Jasbir K Puar, Terrorist Assemblages: Homonationalism in Queer Times (Duke University Press 2007). 75 76 l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 16 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice be gay to get asylum’.77 Martorano notes that NGOs in the field of migration face ten- sions between their humanitarian and ethical values, on the one hand, and the bureau- cratic demands of institutions, on the other, eventually replicating the asylum system’s selective policies of assistance for material and moral reasons.78 As this part explores, non-governmental actors often find themselves trapped in the ‘politics of truth’ of the asylum system and are pushed to contribute to the harmful discourse of ‘fake’ claims, even if unwittingly or reluctantly. Some are tolerant of this role; others resist it, refusing to judge someone else’s ‘truth’. Some NGOs and support groups tend to adopt a relatively ‘hands-off ’ approach in relation to determining the veracity of SOGI claims, showing understanding for pos- sible contradictions and changes of narrative: sometimes, even knowing that the story was false, we know of people who have had it [international protection], sorry if … but on the other hand, people about whom we had no doubts and instead have not [been granted international pro- tection] … because they contradicted themselves, because when they arrived in Italy they said something else … because they are stunned by the journey, be- cause they are afraid, they don’t know what to expect, they don’t know that it [sexual orientation and gender identity] is a [ground for asylum request].79 Others are more ‘hands on’, identifying claims they perceive to be ‘fake’ and thus using their relative power to become actors in the discursive production of SOGI and epi- stemic injustice. In line with scholarly work that has identified this phenomenon in the Italian context,80 the fieldwork conducted for the present project found this dynamic operating in support groups: Let’s say that if they come into contact with us, we filter them out first, so we try not to pursue cases in which we don’t believe, but I would say that if I esti- mate the requests for assistance and those we decided to pursue, it’s more or less fifty-fifty.81 Social workers employed in NGO contexts also shared these concerns: I think in relation to The Gambia maybe [we have fake claims]. Because there was an absurd boom in 2014 in requests for reasons of sexual orientation, in the sense … obviously also connected with the question that there is more infor- mation. I  believe that many [claimants] before didn’t know that they had this 77 McGuirk (n 76) 271. 78 Noemi Martorano, ‘I Gruppi di Supporto Alle e Ai Richiedenti Asilo LGBTI in Italia: Modelli Organizzativi e Tensioni Associative’ [Support Groups for LGBTI Asylum Claimants in Italy: Organizational Models and Associative Tensions] in Massimo Prearo and Noemi Martorano (eds), Migranti LGBT: Pratiche, Politiche Contesti di Accoglienza (Edizioni ETS 2020) 149–51. 79 Anna, LGBTIQ+ group volunteer, Italy. 80 Martorano (n 78) 149–80. 81 Giulia, LGBTIQ+ group volunteer, Italy. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 17 of 30 possibility, but also because they saw that other compatriots have had [inter- national protection].82 Without in any way belittling the essential work carried out by so many NGOs working with migrants and refugees, and while fully understanding NGOs’ need to prioritize limited resources, this approach translates – even if unconsciously – into a methodo- logical homonormativity, in line with a tendency by solidarity movements to construct the ‘ideal subject of solidarity’.83 By doing this, ‘activists respond to and reconstruct the dominant rhetoric, a rhetoric on the basis of which queer and migrant people are ex- cluded and their presence [is] made illegitimate’.84 Even though they may wish to resist the logics of normativity and unleash the power of queer politics, some NGO staff and volunteers – by acting as ‘preliminary judges’ and refusing assistance to those claimants whose testimonies are not believed to be ‘true’85 – mimic the culture of disbelief of decision makers and thus reinforce State-sponsored policies of subjection and assimila- tion.86 In the process, they deprive claimants of their epistemic agency. Amongst these civil society actors are legal practitioners, who play a key role in guiding (or, sometimes, misguiding) claimants through their asylum journey, thereby co-producing the epistemic injustice that entraps them. Legal practitioners are often the first to be wary of ‘standard’ and ‘cyclical’ stories when approached by new clients.87 In Germany, for instance, one lawyer stated that: It’s true that there are … refugees faking [sexual orientation or gender identity]. Probably more women than men, because for men, male homophobia is much bigger, so, I mean, that is certainly a bigger challenge for men … it happened to me that I was sent a woman by the lesbian counselling centre and then she came again a half year later and was pregnant and then told me ‘well, what should I have said, then?’ … That is surely very aggravating. But it happens – I think the figures are not that big.88 Similarly, in Italy, Mara, a lawyer working for an LGBTIQ+ NGO, said that: [W]e do make them follow a process and it is a psychological process, a journey with the mediator, with the operator, we try to make them participate in some ac- tivities that can also be language courses, to try to understand if there is a genuine interest … or whether it is only functional to obtaining the [NGO membership] 82 Susanna, social worker, Italy. 83 Anna Carastathis and Myrto Tsilimpounidi, ‘Methodological Heteronormativity and the 84 “Refugee Crisis”’ (2018) 18 Feminist Media Studies 1120, 1121. Jung (n 54) 315. 85 Martorano (n 78) 168. 86 Jung (n 54)  316–17. On ‘queer politics’ more generally, see eg Michael Warner (ed), Fear of a Queer Planet: Queer Politics and Social Theory (University of Minnesota Press 1993); James Penney, After Queer Theory: The Limits of Sexual Politics (Pluto Press 2015). 87 Bohmer and Shuman (n 5) 614. 88 Janina, lawyer, Germany. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 18 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice card. … Yes, yes [we do a screening]. Some [claimants] already arrive after the [interview with the] commission, with the rejection and when they have to do the appeal, then we become even more suspicious … It is obvious that we can never be completely sure, but, in short, we try to work on it. In the UK, a volunteer with an LGBTIQ+ support group said that ‘[s]ome solicitors just don’t believe their LGBTQ clients, some feel very uncomfortable around the issue of sexuality as reason for protection’.89 It is unclear whether this was on account of homophobia or for another reason, but such accounts reflect the role legal represen- tatives play in the discursive production of knowledge about asylum claimants’ sexual orientation or gender identity and the epistemic injustice that results. There is a sense that it is possible to ‘know the fake ones from the real ones’,90 des- pite the fact that determining the objective ‘truth’ about someone’s sexual orientation or gender identity is impossible, given the socially and culturally constructed nature of these notions. Both the scholarly literature and asylum policy largely ignore that claim- ants and refugees are themselves key actors in this ‘politics of truth’. As such, they are co-opted by the asylum system to perpetuate the epistemic injustice that underpins the system, and on which the system depends in order to achieve its aims. Some claimants who volunteer with NGOs and support groups are indeed keen on ‘sifting out’ those who do not seem to have ‘genuine’ claims: So when somebody say, is he gay? First of all making intention clear, we send our missionaries on ground, we monitor the person, we know if he’s really a gay. And when we are satisfied … then we give him our membership card.91 they [claimants] are the first ones not to want within the group people who are not really homosexuals, they do not want us to use up our reputation as an asso- ciation for people who are not homosexuals, because they say ‘then, if we help everyone, the commission does not believe us anymore and therefore we cannot help more people’.92 The need to preserve the reputation of NGOs and support groups in order to retain their capacity to support SOGI claimants thus leads to assessments of the genuineness of new claimants, sometimes rendering claimants themselves part of the epistemic in- justice inflicted on one another. An NGO’s reputation cannot be sacrificed by ‘fake’ claims – something observed by Giametta in the French context and Martorano in the Italian context.93 In particular, fellow nationals of potential SOGI claimants function as subjective and powerful ‘filters’, acting as unofficial assessors of the ‘truth’ of their claims: 89 Survey respondent S110. 90 Alain, Italy. 91 Kennedy, Italy. 92 Giulia, LGBTIQ+ group volunteer, Italy. 93 Calogero Giametta, ‘New Asylum Protection Categories and Elusive Filtering Devices: The Case of “Queer Asylum” in France and the UK’ (2020) 46 Journal of Ethnic and Migration Studies 142, 148; Martorano (n 78) 152–53. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 19 of 30 I can’t tell you that we realise it immediately but only after a few questions, also be- cause they [our group members] come from those same countries, etc, when a new one arrives and says ‘I’m from The Gambia, I’m gay’, we have ten from The Gambia who listen to him and, when they tell their story, they are able to contradict him or to notice inconsistencies and then we decide fairly quickly which cases to pursue.94 now we have started to support the new guys with guys of the same nationality who were already with [our group] for many months, so that they are aware of the social and cultural dynamics of the country in question … and that they know how the society of the country in question reacts to homosexuals … who then speaks to us privately and tells us ‘Look, things in Nigeria don’t work that way. Society would never have reacted this way, so he’s lying’.95 ‘Genuine’ SOGI claimants not only take part in this discursive construction of ‘fake’ claims, but also express great frustration about such claimants: I see a lot of straight men come here and say that they’re gay and they’re not gay and they got acceptance. And it kind of makes you feel, you know? Some type of way. Because you’re from Jamaica and you know these men are not gay.96 It is not always clear how some claimants ‘know’ that other claimants are not members of a SOGI minority. Some members of focus groups in Germany felt particularly upset by the injustice of ‘fake’ claims and where this left ‘genuine’ claimants: I had trouble with this Jamaican from the camp, and we know he is not gay be- cause he told us, and even one time he caught an infliction because I was like saying to him ‘You say you’re not gay, then why do you come to Germany to seek refuge as gay? You are just mashing up Germany for people like us who really want to seek refugee status. You’re not gay, so why are you here?’ … Even a guy at our camp is not a gay and he got through. And his friend that is truly gay didn’t get through. He got turned down like us.97 But the straight guys who come here and seek asylum, they just come to make money and they know after two, three years they can go back home because they have saved enough money. And the thing that they don’t understand, they come here and they spoil the opportunity that we as gay people get to come here and seek asylum.98 SOGI asylum claimants thus fear that ‘fake’ claims (or perceptions thereof) will hurt their chances of obtaining a positive decision, especially as decision makers may be- come suspicious about the increasing numbers of SOGI claims. This has also been 94 Giulia, LGBTIQ+ group volunteer, Italy. 95 Nicola, LGBTIQ+ group volunteer, Italy. 96 Angel, Germany. 97 Trudy Ann, Germany. 98 Emroy, Focus Group No 1, Hesse, Germany. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 20 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice observed in the context of resettlement work carried out by UNHCR in Turkey, where SOGI claimants become self-appointed screening officers to determine the ‘inauthen- ticity’ and ‘un-deservingness’ of fellow claimants.99 While these concerns are under- standable, it is important to acknowledge that, by adopting such a ‘filtering logic’, civil society actors find themselves unexpectedly co-opted into carrying out ‘perverse prac- tices of policing’, border control, and surveillance,100 simultaneously becoming actors in the epistemic injustice underlying the asylum system. The knowledge contributed and produced by NGOs, legal representatives, asylum claimants, and refugees may play an important and legitimate role in building reliable and up-to-date COI. Yet, such knowledge is not devoid of stereotypes and generaliza- tions, and it can be used to the detriment of SOGI claimants with genuine claims. The irony of supporters and refugees undermining the ‘truth’ of other asylum claim- ants did not escape some of the participants interviewed in this project, whose role in the system can be described as one of ‘counter-conduct’ and resistance against the epi- stemic injustice and dehumanization experienced by SOGI claimants. Seth, an NGO worker in the UK, articulated his frustration at these dynamics in striking terms: ‘“[A]s chief puff I decree that, you know, he is a member of my tribe, so therefore, you know … you know, grant him asylum”. You know, it is ridiculous. … And who am I to sit in judgement’. Responses to potentially ‘fake’ claims in host countries should thus be more sophisticated and socially aware: there is an exaggerated alarmism in relation to this specific subject, because it is true that we know … [of] an increase in the number of [SOGI] claims, which is understandably coherent with an increase in flows and consequently consistent with the greater awareness that now exists, and of the training that once did not exist. … The answers that were given to interpret or manage [an increase in ‘fake’ claims] were not of a social nature. They [the answers] have been from a perspec- tive of demonisation, derision, denunciation, criticism.101 Some NGOs adopt a more constructive approach that avoids the traps of epistemic injustice, for example by offering support to any claimant who requires assistance, even if their claims may seem dubious.102 The use of the limited resources can still be ra- tionalized by imposing some requirements. For example, Joseph, a volunteer with an LGBTIQ+ group in the UK, referred to requiring a minimum period of interaction with the NGO before a supporting statement is produced: ‘[Avoiding “fake” claims] is one of the reasons we said that we wouldn’t, we would not write support letters until people had been coming [to the group] for six months’. Other participants pointed out the risk of generalizing from individual experiences, and the difficulty of ‘faking’ claims: 99 Mert Koçak, ‘Who Is “Queerer” and Deserves Resettlement? Queer Asylum Seekers and Their Deservingness of Refugee Status in Turkey’ (2020) 29 Middle East Critique 29. 100 Giametta (n 93) 147; Martorano (n 78) 172. 101 Vincenzo, LGBTIQ+ group volunteer, Italy. 102 Sara Cesaro, ‘The (Micro-)Politics of Support for LGBT Asylum Seekers in France’ in Richard CM Mole (ed), Queer Migration and Asylum in Europe (University College London Press 2021) 228–29. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 21 of 30 I haven’t met anybody here that I don’t believe is gay. Because I also believe that it is an extreme hurdle for people from this cultural centre to apply for asylum on this [SOGI] basis anyway, if he is not really gay (if his family is here, that’s even out of the question). … Well, I think that’s difficult to do, culturally, since people would have to be good actors.103 I cannot rule that out [‘fake’ claims], but for most people who reveal their sexu- ality or their sexual identity, I think that, … they do that very authentically … there are also very many risks that come with it and therefore it is also a particu- larly vulnerable status that one then has [as a SOGI asylum claimant]. And vol- untarily exposing oneself to that, I do not know, I find that rather unrealistic.104 It was also clear to some participants that attempts to assert the genuineness of SOGI claims replicated the injustice of some asylum authorities’ practices and prerogatives, which NGOs should not emulate: ‘How do I know if the person is really lesbian or gay?’ And that totally upsets me, because I think, when you grow up as a queer or lesbian person and face so many prejudices and somehow so much discrimination … Who would volun- tarily choose this kind of ‘identity’ as an identity? … And I think, these are really rare cases where people would lie about this. … these are mostly people from the [decision-making] institutions that ask such questions and possibly … ‘so, they are not gay, lesbian, trans’ and that … they do not know … do not understand the complexity of living a queer lifestyle. And yes, the stigma associated with it in society, in the family, in the psyche of that person … And whether that is a lie or truth, that’s very … I do not know … absurd.105 While not being able to completely rule out the possibility of claimants fabricating SOGI claims, these participants found it highly unlikely, considering the socio-cultural environment that asylum claimants have to navigate. Ashley, a psychotherapist in the UK, noted that ‘if you haven’t lived with the experience of clandestine sexuality, you won’t be able to fake or feign the language and methods and devices that you use to get through it’. Damiano, a lawyer in Italy, also emphasized how much more difficult life in reception centres could be once it was known that a claimant had a SOGI-based claim. Moreover, it is important to recognize the desperate circumstances that may lead someone not to be entirely honest about their claim, as well as to understand the subjectively, socially, culturally constructed, and fluid nature of sexual orientation and gender identity: But of course, there are cases of people – and one cannot blame them individu- ally – who have experienced or have heard that being gay is a good reason to be recognised [as a refugee] and then try this. It is a way out of the delays in their individual situation.106 103 Thomas, NGO volunteer, Germany. 104 Louis, LGBTIQ+ group volunteer, Germany. 105 Mariya, NGO worker, Germany (emphasis in original). 106 Thomas, NGO volunteer, Germany. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 22 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice for somebody to come repeatedly month in, month out, to a LGBT support group and stuff, then if they are not LGBT, then maybe there is questions at the back of their mind [about their sexual orientation or gender identity] or maybe there is some, you know … and even if they are not [LGBTIQ+], it is ok.107 Above all, some NGO workers make a conscious choice not to assume the role of a decision maker or to follow the way the authorities exercise power: ‘We do not want to play BAMF 2 here’.108 Crucially, and as the scholarship on epistemic injustice highlights, they demonstrate awareness of the fact that there is no verifiable ‘truth’ in respect of a person’s sexual orientation or gender identity: you really can never know that. I’m not [able to], anyway, I could never tell if anyone is gay, lesbian, trans, bi, intersex, such a declaration can only be made about oneself, and even that is flexible, yes … that’s why I always take it as it comes.109 my job is not to make that decision [whether someone is telling the ‘truth’ or not] and I  find that if you let your mind go into that, you make that decision about whether or not somebody is telling the truth, I think that makes you a bad lawyer, because who am I to make that decision? … I don’t go there. … That is not my job.110 Although such an approach may impose a higher workload on these NGOs, it seems to be accepted as a way for relevant NGO staff or volunteers to avoid having to make judgements. Given the impossibility of determining what is actually ‘true’, it is impera- tive to identify the key means to address the toxic effects of the ‘politics of truth’ and the vigilante approach that various actors in the asylum system – whether public author- ities or civil society – may have towards SOGI claimants. 5. U S I N G R E F U G E E L AW A N D P O L I C Y TO V I N D I C AT E S O G I R E F U G E E S ’ O W N ‘ T R U T H S ’ The analysis so far has made it clear that: (1) it is not possible to determine the ‘truth’ about someone’s sexual orientation or gender identity, and (2) SOGI claimants see their epistemic agency seriously and continuously damaged by the asylum system (even if they reclaim agency in a variety of other ways).111 While bearing in mind that ‘truth’ is not achievable, we also need to accept that – at least for the foreseeable fu- ture – asylum systems will continue to pursue some sort of objectivity. That being the case, this part attempts to discuss some policy-oriented means to alleviate the epistemic injustice experienced by SOGI asylum claimants. The proposals below fall 107 Seth, NGO worker, UK. 108 Thomas, NGO volunteer, Germany. ‘BAMF’ stands for ‘Bundesamt für Migration und Flüchtlinge’, the German Federal Office for Migration and Refugees. 109 Matthias, social worker, Germany. 110 Deirdre, lawyer, UK. 111 In the words of Bohmer and Shuman, the ‘process deprives the asylum applicants of the right to determine what counts in their own stories’. Bohmer and Shuman (n 5) 624. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 23 of 30 short of a transformative strategy, as suggested by Fraser (see part 2 above), and do not include all the measures that would be necessary at an individual, social, and institu- tional level in order to eliminate epistemic injustice completely in the asylum system. Nevertheless, they offer a pragmatic and realistic approach to mitigating the problems identified above, even within a generally hostile, populistic, and xenophobic political environment.112 The ‘fake’ claims debate should include an honest acknowledgment of the possibility that some claims may not be entirely genuine, but the discussion cannot stop there: I think there is an element of truth [in the ‘fake’ claims debate]. I mean, I think any system in the world, regardless of what, will be abused by some people, for some purposes. I think that is not something we can deny, I don’t think it is so much of a problem necessarily as it is often made out to be. I think there is a lot of fear around that. I also don’t think that the fact that there are some bad apples should prevent genuine cases from receiving the consideration that they actually deserve.113 Many participants acknowledged the desperation felt by asylum claimants to escape persecution and obtain international protection. Desperation can ‘legitimately’ make claimants lose ‘perspective’ and present stories that are not their own in the hope of increasing their chances of being granted international protection (for instance, if they know someone else was successful with that story).114 The lack of available information about SOGI asylum (that sexual orientation or gender identity can be the basis of a claim) when claimants arrive in Europe and/or lodge a claim – combined with claim- ants’ frequent lack of knowledge about the way SOGI minorities are treated in host countries, fear of discrimination by the host community and by their own diaspora, and internalized homo/transphobia – can also understandably lead claimants to em- bellish their fear of persecution.115 Additionally, Chiara, an NGO worker in Italy, made the point that, even if a claimant is not entirely honest in their testimony, this is not necessarily an ‘abuse of the system’, in the sense that the claimant may nevertheless be deserving of international protection. The focus should instead be on those who profit economically from ‘selling stories’ (such as smugglers),116 and from training claimants in how to use those stories: We have also had reports that there are organisations that even train people on how to present themselves as being gay in asylum procedures, because even if the person is not necessarily gay themselves, because it will help your process. And that there are again, apparently, some organisations that charge for such services. … I  think that is a more pressing issue. I  feel this is maybe a bit contentious, 112 Danisi and others (n 14) ch 4.1. 113 Jules, staff member, ILGA-Europe. 114 Giuseppe, lawyer, Italy; Sofia and Emma, NGO workers, Germany; Terry, member of the European Parliament. 115 Damiano, lawyer, and Valentina, social worker, Italy. 116 Helena, staff member, European Asylum Support Office (EASO), now European Union Agency for Asylum (EUAA). l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 24 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice I feel that anyone who is seeking asylum and who goes through all the effort and hassle and trouble of coming here and seeking asylum whether or not they are gay, whatever their sexual orientation or their gender identity, clearly there was a reason strong enough to motivate them to come, and they should be given a fair chance. So, I don’t necessarily have very strong qualms about people trying to maximise their chances … so long of course that it doesn’t count to a scale that it actually affects those who generally need this particular means. My problem would then become more with those who start to profit off it.117 Media reports from the Netherlands and the USA, for example, affirm concerns that there are people who exploit asylum claimants by selling them stories of successful SOGI claims.118 The focus should thus be on those exploiting SOGI claimants rather than on the risk that some SOGI claimants may not be entirely ‘truthful’. In this context, the ‘filtering’ role played by civil society actors is unwelcome, and these actors’ doubts about whether a claim is genuine are often perceived as judgmental. Consequently, claimants affected by this exercise of power by civil society actors have expressed sad- ness and frustration at being dehumanized and deprived of their ‘truth’ by those from whom they seek support: Because when we come to the groups, we need comfort. We need comfort. We need counselling, we need help. Not to be judged, not to be judged. There is no point why you judge me, when I come to the group, you wait to the Home Office to decide for me, why do you judge me? … You wait for the Home Office and you decide, yes. You don’t need to upset people.119 Some participants also referred to the excessive ‘craving for truth’,120 and, more gener- ally, how this pressure reflects the prejudice and arrogance of civil society actors: I don’t like it either that an association says ‘Ah, but for me he is not gay’. But how can you say that? Again, the LGBTI community also has prejudices … And then, how can you pretend to have the right to judge that a person who comes from a country totally different from yours, does not speak your language, has a totally different mind-set, you say ‘for me he is not gay’. But on what basis do you say that? Even in that case, you have to … put aside certain prejudices that some LGBTI volunteers have, and think that in any case those who have to take a decision are the commission … and that the decision should not be made in the sense that the person must prove irrefutably that they are LGBTI, but that they can offer a story that is more or less coherent.121 117 Jules, staff member, ILGA-Europe. 118 See examples in Bilefsky (n 12); Marion MacGregor, ‘Dutch Government Cracks Down on Ugandan Asylum Seekers after “Fake” LGBT Claims’ (InfoMigrants, 10 November 2020)  <https://www.infomigrants.net/en/post/28401/dutch-government-cracks-down-on- ugandan-asylum-seekers-after-fake-lgbt-claims> accessed 12 September 2022. 119 Miria, UK. 120 Giulia, LGBTIQ+ group volunteer, Italy. 121 Cristina, UNHCR official, Italy. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 25 of 30 The ‘truth’ about SOGI asylum claims is unachievable, since both ‘truth’ and ‘fakeness’ about a person’s sexual orientation or gender identity are discursively produced by all actors in the asylum system. Nonetheless, from a pragmatic and policy perspective, it is important to use all the tools available to make the asylum system fairer for SOGI claimants and to enhance its epistemic justice. Five are identified below. First, claimants should be provided with comprehensive information about key aspects of the asylum system when they first lodge their claim, including that sexual orientation and gender identity can be the basis for an asylum claim.122 The fact that this does not happen currently renders it more difficult for States to fulfil their obliga- tion to identify claimants’ special procedural needs.123 Secondly, the right to free legal assistance and representation should be expanded beyond appeal procedures,124 as well as funded more substantially by domestic authorities, to ensure quality representation at all stages of the asylum procedure. This would allow SOGI claimants to lodge better developed initial claims, supported by evidence and informed by sound legal advice, which is not currently the situation in Europe.125 Thirdly, asylum procedures need to be informed by greater respect for claimants’ rights and dignity, as well as a stronger spirit of empathy. This study’s fieldwork showed that this does not happen at present.126 It is essential to ensure that SOGI claimants have enough time to prepare adequately and present their cases effectively. More care needs to be invested in the choice of locations for asylum interviews, training in inter- view techniques, and the quality of interpreting services, as well as ensuring that claim- ants have an opportunity to clarify any apparent contradictions. Overall, a relationship of trust between the claimant and the decision maker needs to be fostered.127 Fourthly, should a decision maker retain doubts after the interview, it is important to apply the principle of the benefit of the doubt whenever possible. It is clear that this principle is not currently applied with the consistency and scope it warrants, either at a domestic or an international level.128 This is compounded by the fact that the claimant’s self-identification in terms of sexual orientation or gender identity is not afforded suf- ficient value: it may not be the end of the matter,129 but it is undoubtedly the starting 122 Nuno Ferreira, ‘Reforming the Common European Asylum System: Enough Rainbow for Queer Asylum Seekers?’ [2018] Rivista di Studi Giuridici sull’Orientamento Sessuale e l’Identità di Genere 25, 33. 123 European Council on Refugees and Exiles, ‘The Concept of Vulnerability in European Asylum Procedures’ (2017) 21. 124 Currently, and within the Common European Asylum System, this right refers only to appeal procedures: see Directive 2013/32/EU of 26 June 2013 on common procedures for granting and withdrawing international protection (recast) [2013] OJ L180/60, art 20. 125 Danisi and others (n 14) ch 6.2.2. 126 127 128 129 ibid ch 6. ibid ch 11.3.2. ibid ch 7.4.1; Nuno Ferreira, ‘An Exercise in Detachment: The Strasbourg Court and Sexual Minority Refugees’ in Mole (ed), Queer Migration and Asylum in Europe (n 102). Joined Cases C–148/13, C–149/13, and C–150/13, A, B and C v Staatssecretaris van Veiligheid en Justitie [2014] ECLI:EU:C:2014:2406, para 49. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 26 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice point, and decision makers need to take that seriously.130 Rather, some decision makers go as far as reversing the presumption of truth reflected in the principle of the benefit of the doubt and believing that, in case of doubt, a story should be assumed to be false.131 Yet, not only is respect for the principle of the benefit of the doubt a legal requirement, according to UNHCR,132 it is also advisable from a policy perspective: Because, you can even make an argument, I think, that if somebody is so des- perate to stay, that they are actually willing to lie about their sexuality and tell you that they’re, they are gay or whatever, you know, where they know that within their own society this is something which is not seen as acceptable, which does put them at risk … You have got to be pretty desperate to lie about it, so you know … I belong to the group that tends to do benefit of the doubt.133 Some decision makers do seem to be conscious of the need to adopt a lower standard of proof and apply the benefit of the doubt whenever possible: it was bollocks [a ‘fake’ claim], really. And you do get cases like that, yes, of course, you do, yes, and it makes judges battle weary and cynical, of course. And you have got to put that on one side all the time. … But, you know, you remind yourself all the time, it is a lower standard, lower standard [of proof]. It is not a balance of probabilities, it is the lower standard, and if in doubt you must give, you must give the benefit of the doubt.134 In conjunction with a lower standard of proof and the benefit of the doubt, emphasis should shift from the claimant’s ‘true’ sexual orientation or gender identity to the risk of persecution, conditions in the country of origin, and the quality of COI.135 This would better balance decision makers’ determination of the ‘truth’ of SOGI claimants’ mem- bership of a particular social group with an analysis of the risks facing claimants if they are returned to their countries of origin. Fifthly, decision makers would benefit from better training and working conditions, to avoid lack of preparation, burnout, and desensitization. This fatigue and loss of em- pathy over the years have been documented, for example, in Canada.136 Mandatory and regular training on general SOGI matters and SOGI-related COI – including the so- cial and cultural nature and variations of SOGI – would equip decision makers with more appropriate knowledge and skills to deal with SOGI claims in a non-stereotypical 130 Moira Dustin and Nuno Ferreira, ‘Improving SOGI Asylum Adjudication: Putting Persecution Ahead of Identity’ (2021) 40(3) Refugee Survey Quarterly 315. Jubany (n 6) 87. 131 132 UNHCR, Handbook on Procedures and Criteria for Determining Refugee Status and Guidelines on International Protection under the 1951 Convention and the 1967 Protocol relating to the Status of Refugees, HCR/1P/ENG/REV.4 (1979, reissued 2019) paras 203–04. Jean, member of the European Parliament. 133 134 Harry, senior judge, UK. 135 Dustin and Ferreira (n 130). 136 Rousseau and Foxen (n 4) 517. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 27 of 30 or uncynical manner. Moreover, as Helena, a staff member in the European Asylum Support Office (EASO, now European Union Agency for Asylum, EUAA), argued, de- cision makers are invariably affected by the stories of war, rape, and torture to which they listen on a daily basis. According to Jubany, the fact that (in Spain and the UK) there are fewer female than male decision makers also means that the female decision makers frequently listen to stories of rape and sexual violence, thus contributing to greater scepticism and desensitization.137 Finding that these stories must be to some extent ‘fake’ becomes a natural protective mechanism.138 States thus need to improve the training and working conditions of decision makers by providing mandatory and regular training, flexible working conditions, career breaks, and appropriate forms of staff support, including counselling and training in vicarious trauma and self-care,139 as well as abstaining from putting decision makers under any form of pressure to reject asylum claims. None of these suggestions will help determine the ‘truth’ in SOGI claims; such an endeavour is doomed to fail. Nevertheless, the five broad recommendations delineated here can assist in increasing the epistemic justice of the asylum system for SOGI claim- ants – as well as potentially for all asylum claimants – as they have the potential to help claimants have a greater say (both quantitatively and qualitatively) in the discursive construction of the ‘truth’ of their claims. By pursuing greater respect for the right to information, investing in legal aid, improving asylum procedures, applying the prin- ciple of the benefit of the doubt, and improving the training and working conditions of decision makers, we could further reduce the already negligible risk of ‘fake’ SOGI claims. By setting the example and operating an asylum adjudication system that re- spects claimants’ ‘truths’ and does not indiscriminately label their stories as ‘fake’, civil society actors would, in turn, gradually discard their roles as ‘filters’ of ‘fakeness’. NGOs’ institutional reputations would not be damaged if they occasionally offered support to a claimant not being, or not having undergone, exactly what their testimony states, since what matters is to respect claimants’ rights and to treat them with impartiality and humanity. The principle of the benefit of the doubt, in particular, would support all actors in the asylum system in a journey towards greater empathy, belief, and re- spect, better fulfilling the aims of the international protection system. Crucially, this would support refugees’ struggles for epistemic recognition and, at the same time, give them more power to define their own identities and prevent asylum authorities from dictating the terms. 6. CO N C LU S I O N SOGI asylum claimants face the impossible task of proving they are queer enough but not too queer, proving they come from a country where SOGI minorities face enough risk of persecution but where there is not a generalized risk of violence, and, above all, proving the ‘truth’ of their claim where decision makers commonly have a mindset im- bued with scepticism, cynicism, and prejudice. It is all too easy to consider a claim to 137 Jubany (n 6) 84. 138 Deirdre, lawyer, UK. 139 Danisi and others (n 14) ch 11.3.1. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 28 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice be ‘fake’, which renders the asylum system deeply unjust from an epistemic perspective. By adopting the Foucauldian-inspired body of literature on epistemic injustice as a the- oretical framework, this article has identified the crucial ways in which SOGI claimants are deprived of epistemic agency, not only by asylum authorities, but also by NGOs, support groups, legal representatives, and other SOGI claimants and refugees. While relying mostly on empirical data collected in Germany, Italy, and the UK, and notwith- standing any country-level disparities, both the study participants and the documen- tary sources confirmed that all actors in the asylum system to some extent contribute to discourses on ‘fake’ claims. This justifies the concern expressed in this article that asylum systems across Europe and elsewhere are designed in a way that seeks to estab- lish a ‘truth’ that cannot be established, and to deny SOGI claimants their ‘truth’. The topic of ‘fake’ claims is most often used by anti-migration and anti-refugee politicians as part of a xenophobic and racist rhetoric. This applies to asylum claims in general, and SOGI ones in particular, thus often also reflecting homophobia and transphobia. That may explain why discussing ‘fake’ claims seems taboo in academic circles and grey literature. Instead, this article has faced this issue without subterfuge: there may be SOGI claims that lack complete veracity, but then again, ‘truth’ in relation to a person’s sexual orientation or gender identity is illusory, since it is largely subject- ively, socially, and culturally constructed. The theoretically informed and empirically grounded approach employed here may usefully be replicated in relation to other categories of asylum claims, such as those based on religious grounds or gender-based violence, which are also severely affected by discourses of ‘fakeness’ and difficulties with standards and burdens of proof.140 If ‘fake’ claims exist, they are undoubtedly few – ‘exceedingly rare’, in the words of Neilson and Adams.141 More importantly, nobody can claim the role of final arbiter of the ‘truth’, as any system of production of ‘knowledge’ and ‘truth’ is discursively con- structed, shaped by power relations, and characterized by epistemic injustice. The ‘fake’ claimant – especially if thought of as a pervasive and dangerous phenomenon – is thus a myth: a convenient myth to help society make sense of a challenging situation, and design a solution for it.142 As Jean, a member of the European Parliament, said: I think it [fake claims] is another part of the mythology. I would be very inter- ested to see what the figures are on that, because I am willing to bet that most Member States don’t have them. … [I]t is one of those claims that … I think is invented for a purpose. … lot of countries work with the culture of disbelief, the idea that somehow, you know, this [sexual orientation or gender identity] almost 140 See eg Uwe Berlit, Harald Doerig, and Hugo Storey, ‘Credibility Assessment in Claims based on Persecution for Reasons of Religious Conversion and Homosexuality: A Practitioners Approach’ (2015) 27 International Journal of Refugee Law 649; Isabella Mighetto, ‘The Contingency of Credibility: Gender-Related Persecution, Traumatic Memory and Home Office Interviews’ (2016) 3 SOAS Law Journal 1. 141 Victoria Neilson and Lori Adams, ‘Gay Asylum Seekers’ The New York Times (7 February accessed <https://www.nytimes.com/2011/02/07/opinion/lweb07gay.html> 2011) 12 September 2022. 142 Rousseau and Foxen (n 4) 507. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 29 of 30 is a sort of privileged grounds for claim. … I cannot think of anything that I have seen in terms of evidence, that would back that statement, at all. Another member of the European Parliament, Terry, had a similar view: the step to say ‘I am a gay man’, or the step to say ‘I am a trans woman’, without being it, just, you know, to get asylum, and to have it easier … is so high [large] that the number of people who would actually do that and then can tell a cred- ible story about how they were suffering from this, and how it made their life different, very difficult … that the attention that is given to this in the media is completely over the top. In other words, if there is an ‘abuse’, it is an ‘abuse’ committed by States that construct ‘bogus asylum seekers’ and ‘irregular migrants’.143 Our response should thus be at a policy and social level, to facilitate legal and documented migration paths. This would help prevent people providing embellished accounts instead of their own stories be- cause they are desperate. There may only be discursively constructed ‘truth’ and ‘fakeness’ rather than ob- jective ones. But to the extent that one is obliged to try to ‘prove’ something – as asylum claimants are – then systems and processes should facilitate epistemic justice as much as possible. Telling one’s story – even when including experiences of violence – can be empowering,144 but that is frustrated if the listener denies the experiences being recounted and thus dehumanizes the speaker. In fact, denying the claimant’s testi- mony can be even more traumatizing and distressing for the claimant than the original trauma.145 Yet, the need to safeguard the ‘integrity of the system’ is used as an excuse to search for models of decision making that can expunge ‘false’ SOGI claims.146 SOGI claims are thus a powerful example of the disturbing epistemic injustice that asylum systems produce. Decision makers involved with SOGI claims enjoy a clear ‘institutional comfort’ that is used to facilitate testimonial and contributory injustice.147 This not only results in excessive and inappropriate use of discretion by decision makers,148 but also feeds into a toxic discourse of ‘fakeness’. While it may not be possible to completely domesticate such discretion and eradicate the discourse of ‘fake’ claims, it is realistic to combat and reduce the current testimonial and contributory injustices in SOGI claims. As explored above, the focus should be on ensuring respect for the right to information, investing 143 Valentina, social worker, Italy. 144 Amanda Burgess-Proctor, ‘Methodological and Ethical Issues in Feminist Research with Abused Women: Reflections on Participants’ Vulnerability and Empowerment’ (2015) 48 Women’s Studies International Forum 124. 145 Rousseau and Foxen (n 4) 519. 146 M Yanick Saila-Ngita, ‘Sex, Lies, and Videotape: Considering the ABC Case and Adopting the DSSH Method for the Protection of the Rights of LGBTI Asylum Seekers’ (2018) 24 Southwestern Journal of International Law 275, 298. 147 Sertler (n 39). 148 Danisi and others (n 14) ch 7. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 30 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice in legal aid, improving asylum procedures, applying the principle of the benefit of the doubt, and improving decision makers’ training and working conditions. A more trans- formative strategy – one that completely eliminates epistemic injustice in asylum sys- tems – should be the long-term aim. Indeed, it is a moral obligation, and ‘to be human is to be moral, and you cannot have a day off when it suits you’.149 l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 149 Lloyd Jones, Mister Pip ( John Murray 2008) 180.
null
10.1371_journal.pbio.3002453.pdf
Data Availability Statement: All data supporting the findings of this manuscript are available on the Open Science Framework at osf.io/3kyvw.
All data supporting the findings of this manuscript are available on the Open Science Framework at osf.io/3kyvw .
RESEARCH ARTICLE Cell size homeostasis is tightly controlled throughout the cell cycle Xili Liu1, Jiawei Yan2, Marc W. KirschnerID 1* 1 Department of Systems Biology, Harvard Medical School, Boston, Massachusetts, United States of America, 2 Department of Chemistry, Stanford University, Stanford, California, United States of America * marc@hms.harvard.edu Abstract AU : Pleaseconfirmthatallheadinglevelsarerepresentedcorrectly: To achieve a stable size distribution over multiple generations, proliferating cells require a means of counteracting stochastic noise in the rate of growth, the time spent in various phases of the cell cycle, and the imprecision in the placement of the plane of cell division. In the most widely accepted model, cell size is thought to be regulated at the G1/S transition, such that cells smaller than a critical size pause at the end of G1 phase until they have accu- mulated mass to a predetermined size threshold, at which point the cells proceed through the rest of the cell cycle. However, a model, based solely on a specific size checkpoint at G1/S, cannot readily explain why cells with deficient G1/S control mechanisms are still able to maintain a very stable cell size distribution. Furthermore, such a model would not easily account for stochastic variation in cell size during the subsequent phases of the cell cycle, which cannot be anticipated at G1/S. To address such questions, we applied computation- ally enhanced quantitative phase microscopy (ceQPM) to populations of cultured human cell lines, which enables highly accurate measurement of cell dry mass of individual cells throughout the cell cycle. From these measurements, we have evaluated the factors that contribute to maintaining cell mass homeostasis at any point in the cell cycle. Our findings reveal that cell mass homeostasis is accurately maintained, despite disruptions to the nor- mal G1/S machinery or perturbations in the rate of cell growth. Control of cell mass is gener- ally not confined to regulation of the G1 length. Instead mass homeostasis is imposed throughout the cell cycle. In the cell lines examined, we find that the coefficient of variation (CV) in dry mass of cells in the population begins to decline well before the G1/S transition and continues to decline throughout S and G2 phases. Among the different cell types tested, the detailed response of cell growth rate to cell mass differs. However, in general, when it falls below that for exponential growth, the natural increase in the CV of cell mass is effec- tively constrained. We find that both mass-dependent cell cycle regulation and mass-depen- dent growth rate modulation contribute to reducing cell mass variation within the population. Through the interplay and coordination of these 2 processes, accurate cell mass homeosta- sis emerges. Such findings reveal previously unappreciated and very general principles of cell size control in proliferating cells. These same regulatory processes might also be opera- tive in terminally differentiated cells. Further quantitative dynamical studies should lead to a better understanding of the underlying molecular mechanisms of cell size control. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Liu X, Yan J, Kirschner MW (2024) Cell size homeostasis is tightly controlled throughout the cell cycle. PLoS Biol 22(1): e3002453. https:// doi.org/10.1371/journal.pbio.3002453 Academic Editor: Jonathon Pines, The Institute of Cancer Research, UNITED KINGDOM Received: September 15, 2023 Accepted: November 28, 2023 Published: January 5, 2024 Copyright: © 2024 Liu et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All data supporting the findings of this manuscript are available on the Open Science Framework at osf.io/3kyvw. Funding: This work was funded by the National Institute of General Medical Sciences (5RO1GM26875-42 to MWK, 5R35GM145248 to MWK) and National Institute on Aging (1R56AG073341 to MWK, 5R01AG073341 to MWK). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 1 / 34 Cell size homeostasis is tightly controlled throughout the cell cycle Introduction AIC, Akaike information criterion; Abbreviations: AU : Anabbreviationlisthasbeencompiledforthoseusedinthetext:Pleaseverifythatallentriesarecorrect: BI, bilinear; ceQPM, computationally enhanced quantitative phase microscopy; CV, coefficient of variation; DA std, standard deviation of Division Asymmetry; DMEM, Dulbecco’s Modified Eagle Medium; ERA, ergodic rate analysis; FBS, fetal bovine serum; PFS, Perfect Focus System; Rb, retinoblastoma; SE, sub-exponential; SLBP, stem-loop binding protein. The size distribution of a population of proliferating cells is accurately maintained over many generations, despite variability in the growth rate and the duration of the cell cycle in individ- ual cells, as well as the imprecision in the equipartition of daughter cells at mitosis. Each of these factors is known to contribute to a dispersion in cell size within a population [1]. It has long been evident that there must be some “correction” mechanism that would act within indi- vidual cells to counteract the combined effects of all the sources of random variation and thereby ensure a stable size distribution in the population over many generations [2]. Studies on mammalian and yeast cell size up to now have focused on 1 attractive and plausible mecha- nism for size homeostasis: a dependence of the G1 length inversely with cell size. Theoretically, such a mechanism should allow small cells to “catch up” with larger cells by spending a longer time growing in the G1 phase. Such a process would be expected to reduce cell size variation by normalizing size at the point of S phase entry [2–9]. Several molecular players in this pro- cess have been suggested, such as the dilution of retinoblastoma (Rb) protein [6,9,10] and the activation of p38 MAPK kinase [11,12]. However, such a mechanism, while attractive for its simplicity, cannot in principle fully explain the constancy in the cell size distribution over many generations. Specifically, if G1 length regulation were the only operative mechanism, cells would have no way to anticipate the random variation introduced during the subsequent nonG1 cell cycle phases, a period longer than G1 in most proliferating cell types. Nevertheless, most proliferating cell populations, regardless of their surrounding environment and genetic background, manage to achieve highly accurate size homeostasis [13]. In 1985, Zetterberg and colleagues reported that the variation of G1 length in mouse fibro- blast cells accounted for most of the variation in cell cycle length when cells switched from qui- escence to proliferation [14]. However, a later study in several cell lines found the G1, S, and G2 phase lengths had comparable variability and were all positively correlated with the cell cycle length in normal cycling populations [15], implying a dependency of cell cycle phase lengths on cell size outside of G1. Furthermore, regulation of the S and G2 lengths is known to make a con- tribution to size homeostasis in lower eukaryotic organisms, such as budding and fission yeasts [16–18]. However, evidence of size-dependent regulation outside of G1 has seldom been reported in mammalian cells [4,7]. Little is known about whether the nonG1 phases play an appreciable role in maintaining mammalian cell size homeostasis or whether variation in cell size introduced in the nonG1 phases is somehow fully compensated at the next G1/S transition. An alternative approach for regulating cell size, other than regulating it at S phase entry or in the length of other cell cycle phases, would be to regulate cell growthAU : PerPLOSstyle; italicsshouldnotbeusedforemphasis:Hence; allitalicizedwordshavebeenchangedtoregulartextthroughoutthearticle: [1,19]. A few studies have suggested various types of size-dependent growth rate modulation in cultured cells. For example, Cadart and colleagues found that the slope of volume growth rate versus cell volume decreases for large cells at birth [7]; Neurohr and colleagues found that volume growth rate slows down in excessively large senescent cells [20]; and Ginzberg and colleagues found that nuclear area, an approximate proxy for cell size, is negatively correlated with growth rate at 2 points during the cell cycle [8]. Though such observations have been noted, there has been lit- tle said about their quantitative importance. Furthermore, it is hard to evaluate the various types of growth modulation, as they were discovered in different systems using different physi- cal proxies for cell size, such as cell volume and nuclear area. Hence, little can be concluded about whether these processes coexist in the same cell, are specific to certain cell types, or are only reflected in certain types of cell measurement. Compared to studies on cell cycle control, cell growth control has received little attention. In keeping with a previous study in bacteria [21], we wish to distinguish between “size con- trol” and “size homeostasis.” We will use the term “size control” to refer to the regulation of PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 2 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle the mean size, such as when the mean size in a population of cells responds to a change of envi- ronment or when cells differentiate into a different cell type; whereas, we reserve the term “size homeostasis” for the control of the variance around the mean size of a population in a defined steady-state condition. Though these 2 processes may turn out to be mechanistically related, we cannot assume that they share the same mechanism. In this study, our focus is on the less well studied but perhaps more common process of size homeostasis. We used cultured cell lines because primary cells can take a very long time to reach a stable cell size in culture, whereas cell lines are much more stable and reproducible. Furthermore, cell lines have been well characterized; hence, observations from different laboratories can be readily compared and experiments can be easily replicated. Finally, we expect that size regulation would occur in all cell types, normal and transformed, embryonic and differentiated. Like other general cellu- lar mechanisms, such as mitosis, DNA replication, and protein secretion, it is highly likely that underlying general mechanisms are conserved. To test this generality, we have studied size reg- ulation during the cell cycle in several human cell lines of diverse origins, cultured under dif- ferent conditions. Cell size can be expressed either in terms of mass or volume. Cell volume tends to be a more passive response than mass to the mechanical and osmotic conditions occurring during the cell cycle and differentiation [22–25]. Hence, we have chosen to focus on cell mass homeo- stasis. There are excellent experimental means to measure cell mass in suspension culture [26], but it is much harder to measure cell mass accurately when cells are attached to a substratum, which is closer to the physiological context for most mammalian cell types. This single experi- mental limitation has thwarted the study of cell mass homeostasis and growth rate control in the most well-studied systems. Measuring the mass of a single cell on a culture dish accurately is surprisingly difficult. Furthermore, determining the growth rate from the time derivative of the mass is even more challenging [27,28]. The study of cell mass growth rate regulation in attached cells with sufficient precision to distinguish between different models of growth con- trol required the development of new methods. To this end, we recently developed computa- tionally enhanced quantitative phase microscopy (ceQPM), which measures cell dry mass (the cell’s mass excluding water) by the refractive index difference between cell and medium to a precision of better than 2% [29]. To describe statistically significant features of cell mass and growth rate regulation, we tracked single-cell growth and the timing of cell cycle events at a scale of thousands of cells per experiment. Using this improved technology, we could investi- gate the process of cell mass accumulation relative to cell cycle progression throughout the cell cycle. From these improved measurements, we could derive new understandings of cell mass homeostasis during the cell cycle in several cultured cell lines. The results challenge existing theories of cell mass (or, more colloquially, cell size) homeostasis and suggest further mecha- nistic experiments. Results Cell mass variation is tightly controlled and largely independent of the state of the G1/S circuitry “Cell mass homeostasis” can be strictly defined as the maintenance of a stable distribution of cell mass over generations in a population of proliferating cells. Expressed mathematically, at homeostasis, the coefficient of variation (CV) of cell mass at division, CVd, should be lower than the CV of cell mass at birth, CVb. And, the two should fulfill the equation adapted from Huh and colleagues [30]: PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 CV2 b ¼ CV2 d þ Q2; ðEq1Þ 3 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle where Q denotes the partition error, with Q2 ¼ <ðm1(cid:0) m2Þ2> <m1þm2>2 ; m1 and m2 are the birth masses of the 2 daughter cells of the same mother cell, respectively. By monitoring the proliferation and growth of HeLa cells by ceQPM, we found that the cells were indeed at such a homeostatic state, as the difference between the left- and right-hand sides of Eq 1 was negligible (S1 Fig). To explore this homeostasis further, we considered an abstract model of how the cell mass variation of a cell population evolves with cell cycle progression (Section 1 in S1 Text). If there were no operative controls and cell mass grew exponentially (dm ¼ am) (Fig 1A), the cell mass dt CV would be expected to increase super-exponentially as the cells traverse the cell cycle due to the variation of the growth exponent, α, among cells (Fig 1B). Furthermore, the variation in cell cycle length and the partition error would further contribute to the cell mass variation (quantified by the birth mass CV) at each generation (Fig 1C). To maintain cell mass Fig 1. Cell mass variation is tightly controlled in mammalian cell lines and is robust to perturbations in G1/S (A–C) An abstract model of cell mass homeostasis at different G1 regulation strengths, regulation or growth rate. AU : AbbreviationlistshavebeencompiledforthoseusedinFigs1to5:Pleaseverifythatallentriesarecorrect: represented by the slope of G1 length vs. birth mass correlation. The corresponding model and simulation parameters are in the Section 1 in S1 Text. In the model, we assume cells grow exponentially, and the G1 length control is the only mechanism to reduce cell mass variation. (A) Correlations between G1 length and birth mass. Blue: no G1 length control; red: with strong G1 length control; yellow: with weak G1 length control. (B) Cell mass CV changes with cell cycle progression during 1 cell cycle with the corresponding G1 length regulation in (A). (C) Birth mass CV changes across generations with the corresponding G1 length regulation in (A). (D–G) The mean birth mass (D), birth mass CV (E), division mass CV (F), and DA std. (G) for different cell lines. (H–K) The mean birth mass (H), birth mass CV (I), division mass CV (J), and DA std. (K) for RPE-1 and U2OS cells in normal culture medium, medium with 50 nM palbociclib, and medium with 100 nM rapamycin at cell mass homeostasis. Error bars in (D–K) indicate the standard deviation of 3 or more measurements. The data underlying this figure and the scripts used to generate the plots are available on the Open Science Framework at osf.io/3kyvw. CV, coefficient of variation; DA std., standard deviation of Division Asymmetry. https://doi.org/10.1371/journal.pbio.3002453.g001 PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 4 / 34 BirthG1/SDivisionCell cycle progression0.10.150.20.25Cell mass CV200400600800Birth mass0102030G1 length5101520GenerationBirth mass CV0.10.20.30.60.70.8HT1080HeLaRPE-1U2OSSaos-20200400Mean birth mass (pg)HT1080HeLaRPE-1U2OSSaos-200.10.20.3Birth mass CVHT1080HeLaRPE-1U2OSSaos-200.10.20.3Division mass CVHT1080HeLaRPE-1U2OSSaos-200.020.040.06DA Std.DEFGABCControl50nM Palb100nM Rapa0200400600Mean birth mass (pg)Control50nM Palb100nM Rapa00.20.4Birth mass CVControl50nM Palb100nM Rapa00.20.4Division mass CVControl50nM Palb100nM Rapa00.020.040.06DA Std.RPE-1U2OSHIJKPLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle homeostasis, these accumulated discrepancies must be offset by a reduction of variability by some processes during the cell cycle. If, as suggested in both in vitro and in vivo systems [4,6], the G1/S checkpoint were the principal “size control checkpoint” (Fig 1A), we would expect the reduction in cell mass variation to occur before or at the G1/S transition. The cell mass CV would then be expected to increase super-exponentially after G1/S due to the lack of any oper- able size control processes in the nonG1 phases. Therefore, the CV reduction before G1/S would have to greatly undershoot the birth mass CV to anticipate and compensate for the cell mass variability that would accumulate during the nonG1 phases (Fig 1B). If the G1/S control were weakened by genetic mutation or pharmacological perturbation (Fig 1A), the reduction in cell mass CV before G1/S would be expected to decrease, and the uncorrected error would cause an increase in the division mass CV (Fig 1B). Such a population would eventually reach a new homeostatic state with higher birth and division mass CVs in order for Eq 1 to be ful- filled (Fig 1C). Therefore, the birth mass CV at homeostasis can be used as an indicator of the stringency of the control on cell mass homeostasis. To investigate how different forms of G1/S control might affect cell mass homeostasis, we compared various human cancer cell lines, each with different G1/S deficiencies, and RPE-1, a cell line with a wild-type G1/S transition [7,12,31] (S1 Table). To evaluate the stringency of the control mechanisms regulating cell mass homeostasis, we measured the birth and division mass CVs of live cell populations from short-term videos using ceQPM. We define the Divi- sion Asymmetry, DA ¼ m1;2 , where m1 and m2 represent the birth masses of the 2 daughter m1þm2 cells, and m1,2 denotes the mass of either of the daughter cells. For a population that divides with perfect symmetry, the distribution of DA should be precisely at 0.5 without any disper- sion. But if either daughter cell were larger or smaller than half the mother cell mass, its DA would deviate from 0.5. The standard deviation of DA (DA std.) quantitatively represents the fidelity of cytokinesis, and it is more commonly used than the partition error Q in Eq 1 [17,32]. Despite the considerable variation in cell mass across the different cell lines (the mean birth mass of the largest cell line, HT1080, is 1.85-fold greater than the smallest cell line, Saos- 2) (Fig 1D), the difference in birth mass CV is less than 15% for each cell line (Fig 1E); the divi- sion mass CV and DA std. for these cell lines were also comparable (Fig 1F and 1G). Note that the measurement error of ceQPM is negligible (less than 2%) compared to the birth and divi- sion mass CVs. To assess the robustness of the birth mass CV to perturbations in the G1/S transition, we perturbed G1/S regulation in both RPE-1 and U2OS cells using a well-characterized CDK4/6 inhibitor, palbociclib [33]. Although U2OS cells have intact Rb proteins, which have been reported to govern the G1/S transition [4,6,34], they carry deficiencies in other G1/S regulators (S1 Table) and are much less sensitive to palbociclib than RPE-1, which has intact G1/S cir- cuitry (S2A and S2B Fig). Both cell lines were examined at a low dose of palbociclib, where there was a delay in G1/S but no arrest of the cell cycle (11). We measured the dry mass of RPE-1 and U2OS cells after being cultured for more than 1 week in palbociclib, at which point the mass distribution of each cell line had reached a new steady state. It had been shown previ- ously that a low dose of palbociclib weakens the negative correlation between birth size and G1 length (like the yellow curve in Fig 1A) [11]. Thus, if G1 regulation were essential for cell mass homeostasis, we would expect the birth mass CV to increase with palbociclib treatment (like the yellow curve in Fig 1C). Surprisingly, although the mean mass at birth had increased by 1.68-fold and 1.13-fold, respectively, in RPE-1 and U2OS cells (Fig 1H), the birth mass CV for either cell line hardly changed and in fact slightly decreased (a 4% and 3% reduction for RPE-1 for U2OS cells, respectively) (Fig 1I). Similarly, the division mass CV and the standard devia- tion of Division Asymmetry, DA std., also hardly changed after exposure of both cell lines to PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 5 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle palbociclib (Fig 1J and 1K). These very small changes in mass CVs indicate that the control of mass homeostasis still operates accurately, despite strong perturbation of the G1/S transition. Since disruption and delay of the cell cycle at G1/S did not appear to affect mass homeosta- sis, we examined the inhibition of cell growth for effects on cell mass variability. We used rapa- mycin, a specific inhibitor of mTOR [35], which has pervasive knock-on effects on protein synthesis and degradation [36]. When RPE-1 and U2OS cultures were exposed to rapamycin, the steady-state birth mass decreased by 27% and 20%, respectively (Fig 1H). However, there were no significant changes in the birth mass CV, division mass CV, or DA std. (changes less than 8% were observed) (Fig 1I–1K). Therefore, it appears that mass homeostasis is strongly buffered, even when mass is greatly perturbed. Cell mass variation is regulated throughout the cell cycle Using ceQPM, we can now ask at what points during the cell cycle variation in cell mass occurs and at what points it is suppressed. We used the CV as a metric of cell mass variation and mea- sured it throughout the cell cycle in live RPE-1 and HeLa cells. To correlate the CV with the state of the cell cycle, we utilized fluorescently tagged geminin degron as the cell cycle marker. Geminin is a protein that regulates DNA replication. Possessing a destruction sequence like cyclin B, geminin is degraded precisely at mitosis and begins to accumulate precisely at the G1/S transition (S3A Fig) [37]. We aligned individual cell mass trajectories (S3B Fig) by nor- malizing the length of the G1 segment to 0–0.5 and that of the nonG1 segment to 0.5–1 and then calculated the CV of these normalized cell mass trajectories with cell cycle progression. In RPE-1 cells, the cell mass CV was found to be reduced throughout the cell cycle (Fig 2A), whereas in HeLa cells, the cell mass CV increased in the G1 phase before declining in the nonG1 phases (Fig 2B). Neither cell line exhibited a minimum cell mass CV at the G1/S transi- tion, as would be predicted by conventional G1 length control models (Fig 1B). To examine the regulation of cell mass variation further in various cell lines and under dif- ferent conditions, we calculated the cell mass CV profile as a function of cell cycle progression from fixed cells, which provided much higher throughput than our live cell measurements. Using ergodic rate analysis (ERA) (38), we defined a cell cycle mean path and divided it into 13 to 14 segments evenly spaced in time, based on measurements of DNA content and fluores- cently tagged geminin degron. We applied this analysis to hundreds of thousands of fixed cells (S4A Fig). By definition DNA replication occurs exclusively in the S phase, whereas geminin accumulation starts at the G1/S transition (S4B–S4E, S4H, and S4J Fig) [38]. Though these 2 markers provide good resolution in late G1 and S phases, they have poor temporal resolution in the early G1 and G2-M phases due to inaccuracy in cell cycle stage identification (S4F Fig). Therefore, we focused our analyses exclusively on the cell mass CV in the late G1 and S phases, employing large numbers of fixed cells. We applied this approach to 4 cell lines: RPE-1, HeLa, U2OS, and HT1080. The cell mass CV profiles in fixed RPE-1 and HeLa cells (Fig 2C and 2D) were similar to what we had previ- ously found in the live cell trajectories (Fig 2A and 2B), further validating the use of fixed cells to extract cell mass CV profiles. We found that in RPE-1 and U2OS cells, the cell mass CV declined in late G1 (Fig 2C and 2E), as would be expected from conventional models where regulation of the G1 length is thought to be the sole means for normalizing cell size. However, we were surprised to find that the CV of cell mass then continued to decrease progressively through S phase. Most strikingly, in HeLa and HT1080 cells, there was virtually no reduction in cell mass CV in late G1; the major decrease only took place in S phase (Fig 2D and 2F). These quantitative differences in cell mass CV profiles may depend on the status of the G1/S circuitry in these cell lines (S1 Table). These observations are completely at odds with the G1/S PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 6 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle Fig 2. Cell mass variation is regulated throughout the cell cycle. (A, B) Cell mass CV change with cell cycle progression measured in live RPE-1 (n = 89) (A) and HeLa cells (n = 223) (B). The red solid lines denote the cell mass CV of the population; the pink shadows show the 95% confidence interval; the dashed line indicates the G1/S transition. (C–H) The profiles of how cell mass CV changes with cell cycle progression at cell mass homeostasis measured in fixed RPE-1 (C), HeLa (D), U2OS (E), and HT1080 (F) cells, as well as RPE-1 cells that had reached the new cell mass homeostasis with 50 nM palbociclib (G) or 100 nM rapamycin (H). The cell cycle stages were identified by DNA content and log(mAG-hGeminin) as illustrated in S4B–S4F, S4H, and S4J Fig; the late G1 and S phases are indicated by areas shaded in purple and orange, respectively; error bars are the standard error of CV, (CV= where n is the cell number at the corresponding cell cycle stage (n > 135 for all conditions). The data underlying this figure and the scripts used to generate the plots are available on the Open Science Framework at osf.io/3kyvw. CV, coefficient of variation. ffiffiffiffiffi 2n p ), https://doi.org/10.1371/journal.pbio.3002453.g002 transition playing the dominant role in cell size control, although it may remain a critical point for cell cycle regulation [1,19,34]. Note that the decrease in cell mass CV cannot be explained by a reduction in noise because even if noise went to zero at some point, the CV would remain at its previous value. We believe that a very strong conclusion can be drawn from these phenomenological measurements: there must be feedback between cell mass and cell growth rate or between cell mass and cell cycle outside of the G1 phase. The effect of this feedback would be to effectively reduce existing variation in the population in nonG1 phases of the cell cycle. Since palbociclib and rapamycin had little or no effect on the birth and division mass CVs (Fig 1I and 1J), we wondered whether they affected the timing of mass CV regulation during the cell cycle. Consequently, we carefully measured the cell mass CV profiles in fixed RPE-1 cells that had reached new cell mass homeostasis with either drug. Both drugs altered the PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 7 / 34 0.20.40.60.8Cell cycle progression0.10.120.140.16Cell mass CVHeLa0.20.40.60.8Cell cycle progression0.150.20.250.3Cell mass CVRPE-1G1/SCell cycle progression0.10.20.30.4Cell mass CVRPE-1G1/SCell cycle progression0.150.20.25Cell mass CVHeLaG1/SCell cycle progression0.10.20.30.4Cell mass CVU2OSG1/SCell cycle progression0.10.20.30.4Cell mass CVHT1080Late G1SABCDEFGHG1/SCell cycle progression0.20.30.4Cell mass CVRPE-1 50 nM PalbG1/SCell cycle progression0.10.20.30.4Cell mass CVRPE-1 100 nM RapaPLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle duration of the cell cycle phases and particularly extended the G1 phase (S2 Table). As we had done above with untreated cells, we computed the cell cycle mean path of treated cells and examined their cell mass CV as a function of cell cycle progression (S4G–S4J Fig). Strikingly, we found that disrupting the G1/S transition with palbociclib led to a slight increase in cell mass CV in late G1, followed by a much greater reduction in cell mass CV during the S phase (Fig 2G). Conversely, inhibiting cell growth with rapamycin caused a greater reduction of cell mass CV in late G1, and the reduction in S phase became smaller (Fig 2H). These results sug- gest that the regulation of mass CV during S phase can compensate for the mass CV reduction in late G1. Thus, when there is an insufficient or excessive reduction in mass CV in late G1 due to the inhibition of the G1/S transition or growth, respectively, there is a corresponding change in the mass CV in S phase, which acts to maintain the mass CV reduction at division at the same level. Feedback by cell mass not only acts on the duration of G1, but also on the durations of S and G2 phases To investigate further cell mass regulation outside of the G1 phase, we needed to better opti- mize the resolution of the cell cycle markers we had employed. We therefore adopted 2 cell cycle markers for live cells that bracketed S phase: mAG-hGeminin [37] and mTurquoi- se2-SLBP [39]. The APCCdh1 substrate, geminin, starts to accumulate in the nucleus at S phase entry [40], whereas the histone mRNA stem-loop binding protein, SLBP, is rapidly degraded at the end of the S phase [41] (S5A Fig). Unlike the conventional PCNA or DNA ligase I mark- ers, which label replication foci during the S phase [42,43], geminin and SLBP are diffusive in the nucleus and more suitable for the relatively low spatial resolution of the QPM camera. With these 2 markers, we could accurately quantify the durations of G1, S, and G2-M phases. Since the duration of M phase is remarkably constant [15], we attributed most of the variation in G2-M duration to the G2 phase itself. We verified that the timing of S phase, as identified by geminin and SLBP, was consistent with the timing of S phase identified by the DNA ligase I foci (S5B and S5C Fig). None of the markers affected the length of any of the cell cycle phases nor did they affect the mass-dependent regulation of the duration of the cell cycle phases (S3 Table). Moreover, the identification of the cell cycle phases (G1, S, and G2-M) using geminin and SLBP exhibited a similar variability in their lengths as those shown using PCNA as a marker of S-phase by Araujo and colleagues [15] (S2 Table). Therefore, we could be confident that the geminin and SLBP markers faithfully reported the cell cycle phase durations and did not change the physiology of these processes. Using this approach, we confirmed the well-established existence of cell size-dependent reg- ulation of G1 length with ceQPM. Consistent with previous findings [3,4,6–8,12], we found that the G1 length was negatively correlated with birth mass in both non-transformed and transformed cell lines, RPE-1 (Fig 3A) and HeLa (Fig 3E), respectively. The correlation was stronger in RPE-1 than in HeLa cells (Fig 3A and 3E and S4 Table). We also investigated the mass-dependent regulation of the durations of both S and G2 phases. S and G2-M phase lengths negatively correlated with the initial mass of the corresponding periods in both RPE-1 and HeLa cells (Fig 3B, 3C, 3F, and 3G). For RPE-1 cells, the correlations of cell cycle phase length with initial mass in S and G2 were weaker than that in G1, yet they were significant (Fig 3A–3C and S4 Table), demonstrating that regulation of cell mass variation can occur through regulating the durations of S and G2 phases in non-transformed cells with an intact cell cycle network, including an intact G1/S transition. This contrasts to the conventional models that would have predicted G1 length to vary inversely with mass while leaving other phases unaf- fected. We also found in HeLa cells that the negative correlation between cell cycle phase PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 8 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle Fig 3. The negative regulation of the durations of the G1, S, and G2 phases by cell mass. (A–D) The correlations between the lengths of the G1 (A), S (B), G2-M phases (C), and the full cell cycle (D) and the initial mass of the corresponding period in RPE-1 cells. The bottom panels indicate the correlation; the top panels are the distributions of the initial mass. Each gray dot in the bottom panels is an observation; R is the correlation coefficient of the gray dots; black squares indicate the average of each cell mass bin; error bars are the SEM; solid black line is the best fit of the black squares (S4 Table). The red shaded area in the top panel indicates the cell mass range that is affected by the minimal cell cycle phase length limit, with the text indicating the percentage of affected cells in the distribution. (E–H) The correlations between the length of the G1 (E), S (F), G2-M phases (G), and the full cell cycle (H) and the initial mass of the corresponding period in HeLa cells. (I, J) The correlations between birth and division masses in RPE-1 (I) and HeLa (J) cells. Each gray dot is an observation; black squares are the average of each cell mass bin; error bars are SEM. The solid black line is the best linear fit of the gray dots; the text indicates the function of the best fit; the red line is the prediction of the best fit in (D) or (H), respectively, assuming that cells grow exponentially (Materials and methods). The data underlying this figure and the scripts used to generate the plots are available on the Open Science Framework at osf.io/3kyvw. SEM, standard error of the mean. https://doi.org/10.1371/journal.pbio.3002453.g003 length and mass was much stronger in the S phase, with a correlation coefficient of −0.29, compared to that in the G1 phase, which had a correlation coefficient of −0.20 (Fig 3E and 3F and S4 Table). It is worth noting that although RPE-1 has more stringent G1/S control than HeLa, the overall dependency of cell cycle length on cell mass was not stronger (Fig 3D and 3H and S4 Table). These studies challenge the G1/S checkpoint model, as mass-dependent cell cycle regulation is not restricted to the change in the length of G1 phase as predicted [2,44,45], but rather it is accompanied by changes in the lengths of the other phases of the cell cycle. Upon closer examination of the binned correlations, we observed a fixed minimum limit for the length of nearly every phase of the cell cycle, as well as the length of the entire cell cycle in RPE-1 and HeLa cells (Fig 3A, 3B, and 3D–3H). These limits are not further reduced in large cells. To summarize these findings, we employ 2 graphical representations for these cor- relations: a linear model and a bilinear model, comprised of 2 line segments. With these, we fit the binned correlations of mass and cell cycle phase lengths. We found that a bilinear model PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 9 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle provided a better fit for all phases of RPE-1 and HeLa cells, with the exception of the G2-M phase in RPE-1 cells (Fig 3A–3H and S4 Table). This graphical relationship implies that regula- tion of the durations of cell cycle phases cannot effectively control the mass of large cells. To illustrate the impact of the minimal cell cycle length on cell mass variation, we conducted sim- ulations to observe the mean and CV of cell mass within a cell population across generations, while varying the fraction of cells affected by the minimal length limit (Section 2 in S1 Text). The simulations show that as the minimal cell cycle length applies to more and more cells, the homeostatic birth mass CV increases. The system eventually loses homeostasis when the mini- mal cell cycle length is imposed on more than 40% of the cell population (S6 Fig). In these experiments, we found that the slope of a graph of birth masses versus division masses was close to 1 in both RPE-1 (Fig 3I) and HeLa cells (Fig 3J), consistent with the adder- like behavior seen previously [7]. The adder model is interpreted as a behavior where cells add a constant amount of mass during the cell cycle regardless of their birth mass. Furthermore, we found in our measurements that each cell cycle phase exhibited an adder-like behavior (S7 Fig), making the full cell cycle a sequential adder. Such behaviors challenge the interpretation that, in mammalian cells, mass regulation arises from a combination of a G1 sizer and a nonG1 timer [19]. Rather, the present findings strongly suggest that each cell cycle phase, except for M phase, contributes to cell mass homeostasis. Moreover, the fitted function of birth mass and cell cycle length correlation cannot fully explain the adder behavior. This is par- ticularly the case for large cells, under the assumption of exponential growth (Fig 3I and 3J). This discrepancy is at least partially due to the existence of a minimal cell cycle phase length. These new results underscore the need for a process of non-exponential growth (or what we term “growth rate modulation”) to maintain cell mass homeostasis in the mammalian cells we have studied, rather than relying solely on processes of cell cycle regulation. Mass-dependent growth rate modulation reduces the CV of cell mass during cell cycle progression The simplest mathematical model for cell growth kinetics, which requires no size sensing or feedback mechanisms, is an exponential model where the growth rate is proportional to size. This has been particularly successful in describing growth in bacteria and can be rationalized by a process of ribosome-dependent ribosome biosynthesis [26,46]. This simple exponential model, however, causes variation in cell size to amplify as cells progress through the cell cycle (Fig 1B and Section 3.1 in S1 Text). Contradicting this model, several studies have found that although large cells generally grow faster than small cells, growth is not precisely exponential in mammalian cells [7,26,29]. Such a lack of exponential growth might in itself lead to a reduc- tion in cell size variation. Various previous studies suggested the dependency of growth rate on cell size changes with cell size and cell cycle stage [7,8,20,47–50]. Recent studies by us and others have found growth rate oscillations [29,51], where a cell alternates between increases and decreases in growth rate. To explore the dependence of growth rate on cell mass in proliferating cells, we measured the growth rate in a 3-h time window and computed its correlation with cell mass at time zero. We examined how growth rate correlated with cell mass in 18,000 HeLa cells and found that the relation of mass to growth was close to exponential, except for a slight depression for large cells (S8A Fig). Nevertheless, when we segregated the cells into 4 cell cycle phases, we uncov- ered distinct cell cycle dependencies in such correlations, which were originally masked by pooling all cells for analysis (S8B Fig). An even closer look at the data, with cells categorized into 14 equally divided cell cycle stages, revealed positive-to-negative correlation transitions at various points in the cell cycle (S8C Fig). The slope of the linear relation between cell mass and PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 10 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle growth rate for cells in different stages of the cell cycle indicated stronger modulation (greater deviation from the expected slope of exponential growth) in the late G1 and G2-M phases (S8D Fig), consistent with S8B Fig and previous studies [8,38]. However, the proportionality is sub-exponential in most of the cell cycle stages (S8D Fig), suggesting a global process that inherently limits the growth of large cells. When we investigated the mass versus growth correlations in finely divided cell cycle stages, we found subtle features. Yet, such studies require very large numbers of cells and very accurate growth rate measurements. Coarser cell cycle discrimination leads to a loss of this kind of infor- mation on subtle changes in the growth rate, it nevertheless adds greater statistical power to conclusions about overarching aspects of mass-dependent growth regulation. Therefore, there is a practical tradeoff between high cell cycle resolution of the growth analyses and the statistical reliability of the findings. In the following analyses, we aimed for stronger statistical significance and therefore partitioned cells more crudely into the G1 and nonG1 phases, focusing on the most salient features of growth rate modulation. This level of resolution was sufficient to reveal previously undiscovered features, which serve to correct our current understandings. Measuring the correlation between cell mass and growth rate in 5 different cell lines, we found that each cell line behaved somewhat differently. In RPE-1 cells, growth was propor- tional to cell mass, but the proportionality was much less than exponential, with a significant nonzero intercept (Fig 4A). In HeLa cells, the proportionality between growth and cell mass is much closer to, but slightly less than exponential in both G1 and nonG1 phases (Fig 4B). The observed mass versus growth correlations in short-term measurements in RPE-1 and HeLa cells were consistent with their long-term growth trajectories (S9 Fig), showing nearly linear growth in RPE-1 and a slight deviation from exponential growth in HeLa cells. Therefore, we could confirm that the observed deviation from exponential growth is not due to inspection or sampling bias caused by the short-term measurement [52], but truly signifies the inherent growth law of the cells. In U2OS cells, the correlation was close to exponential for all cells in nonG1 and most cells in G1 phase, but it was abruptly negative for the 15% largest cells in G1 phase (Fig 4C). In HT1080 cells, growth was close to exponential for small cells but transi- tioned to nearly linear growth in large cells during both G1 and nonG1 phases (Fig 4D). A bilinear model provided a significantly better fit than a simple linear model for cells in the nonG1 phase, indicating the significance of this transition in mass versus growth correlation as cells became larger (S5 Table). In Saos-2 cells, growth was exponential except for a slight deviation for large cells in nonG1 phase (Fig 4E). Taken together, these results indicate that the mathematical description of growth rate is not simply exponential in the cell lines we have investigated, and that different cell lines display different characteristics of mass dependency at different phases of the cell cycle. To better compare the behaviors of different cell lines, we normalized the mass versus growth correlations, using the means of birth mass and cell cycle length (S6 Table). Since DNA copy number affects the correlation intercepts (Fig 4A–4D), we focused solely on the slope of the correlations. We could distinguish 2 general types of growth rate modulation (Fig 4F and S6 Table). In the first type, growth is linearly related to cell mass, but with a slope lower than exponential growth (RPE-1 and HeLa). We refer to this as sub-exponential (SE) modulation. In the second type, the slope of the mass versus growth correlation is close to exponential for small cells but becomes less positive or even negative for large cells (U2OS G1, HT1080, and SaoS-2 nonG1). We refer to this as bilinear (BI) modulation. For U2OS cells in the nonG1 phase and SaoS-2 cells in the G1 phase, the correlation slope is not significantly different from exponential growth, suggesting minimal regulation. Other studies had proposed that growth rate modulation contributes to cell size homeosta- sis [1,7,8,19,38]. However, most of these claims were speculative and lacked sufficient PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 11 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle Fig 4. Growth rate dependence on mass differs in different cell lines, and growth rate modulation can effectively reduce cell mass CV during the cell cycle. (A–E) Correlations between cell mass and growth rate in the G1 (blue) and nonG1 (red) phases for RPE-1 (A), HeLa (B), U2OS (C), HT1080 (D), and Saos-2 (E) cells. Filled squares represent the median growth rate of each bin; error bars show SEM. The black dashed lines indicate the expected behavior for exponential growth. The solid blue and red lines are the best fit of the filled squares (S5 Table). (F) The observed conditions were categorized into 3 types: sub-exponential, bilinear, and no modulation. (G) Contour plot illustrating the change in cell mass CV during the entire cell cycle for SE growth rate modulation, as a function of the mean and CV of α0 and β0, obtained from numerical simulations (Section 3.1 in S1 Text). (H, I) Contour plots illustrating the change in cell mass CV during the G1 (H) and nonG1 phases (I) for BI growth rate modulation, as a function of the means of γ0 and m0 t, obtained from numerical simulations (Section 3.2 in S1 Text). These simulations assumed a 20% CV in α0. Solid circles in (G–I) indicate the corresponding positions in the contour plots when adopting parameter values from the experimental data. The data underlying this figure and the scripts used to generate the plots are available on the Open Science Framework at osf.io/3kyvw. BI, bilinear; CV, coefficient of variation; SE, sub- exponential; SEM, standard error of the mean. https://doi.org/10.1371/journal.pbio.3002453.g004 quantitative support. The work by Cadart and colleagues in 2018 stands out as an exception, as it quantitated the correlation between birth size and growth rate [7]. Accurate and quantitative correlations between growth rate and cell size are essential for a thorough assessment of the impact of growth rate regulation. Nevertheless, due to the scarcity of high-quality experimental data, most theoretical investigations into cell size homeostasis have disregarded growth rate regulation completely and focused solely on the regulation of cell cycle length, often assuming exponential growth [53–56]. In this study, we addressed this gap in previous studies by investi- gating theoretically whether the types of growth rate modulation we observed could effectively reduce cell mass variation. Using stochastic models and simulations, we focused on the influ- ence of growth rate modulation and growth rate noise on the cell mass CV over 1 cell cycle. PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 12 / 34 200400600Cell mass (pg)05101520Growth rate (pg/hr)Saos-2200400600Cell mass (pg)0102030Growth rate (pg/hr)RPE400600800Cell mass (pg)10152025Growth rate (pg/hr)HeLa 200400600800Cell mass (pg)05101520Growth rate (pg/hr)U2OS4006008001000Cell mass (pg)10203040Growth rate (pg/hr)HT1080ABCDEFIGHG1nonG1ExponentialCV(0.5)2-CV(0)2-0.03-0.025-0.02-0.015-0.01-0.00500HT1080U2OS11.52m'-3-2-10'(cid:2)Type 1Type 1 Condi(cid:2)on Modula(cid:2)on type RPE-1 HeLa Sub-exponen(cid:2)alType 1 U2OS G1 Bilinear U2OS nonG1 None HT1080 G1 Bilinear HT1080 nonG1 Bilinear SaoS2 G1 None SaoS2 nonG1 Bilinear Sub-exponen(cid:2)alCV(1)2-CV(0)2-0.03-0.02-0.01000.010.020.030.04HeLaRPE-100.20.40.6'00.10.20.30.4CV'=CV'CV(1)2-CV(0.5)2-0.006-0.004-0.002000.0020.0040.0060.008HT1080Soas-211.52m'00.10.20.30.4'(cid:2)PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle Initially for convenience, we assumed that all cells divided at the same cell cycle length. Subse- quently in more comprehensive models, we incorporated cell cycle regulation and noise, as discussed in a later section. In the absence of any growth rate modulation, we might imagine that cell mass should accu- mulate exponentially, as has been found in bacteria [26]. This would cause the cell mass CV to increase super-exponentially due to stochastic variation in growth rate (Section 3.1 in S1 Text). When growth rate modulation is in the SE form (Fig 4F and S6 Table), the slope of the correlation between cell mass and growth rate is lower than that of exponential growth. This can be described by the equation: dm0 terms: α0m0 represents the part of growth rate proportional to cell mass, whereas β0 represents the part independent of cell mass. Here, m0 and t0 are the cell mass and cell cycle progression time normalized by the means of birth mass and cell cycle length, respectively (Section 3.1 in S1 Text). According to the definition of sub-exponential growth, the mean of α0 is smaller than ln2 and greater than 0, and the mean of β0 is determined by α0 when assuming that the mean division mass is twice the mean birth mass, a requirement for maintaining mass homeostasis. For simplicity, we first assumed that α0 and β0 have the same CV, but we also examined how the CV of either parameter affected the results in the Supporting information (S10 Fig). dt0 ¼ a0m0 þ b0, where the growth rate is composed of 2 During the initial stages of the cell cycle, the cell mass CV consistently decreases, with the rate of decrease negatively correlated with the mean of α0 and independent of the CVs of α0 and β0 (S10A, S10C, and S10F Fig and Section 3.1 in S1 Text). As the cell cycle progresses, the rate of mass CV reduction slows down, and the mass CV may even increase during the later period of the cell cycle (S10B, S10D, and S10G and Section 3.1 in S1 Text). The overall change in the cell mass CV throughout the cell cycle depends on both the mean of α0 and the CVs of α0 and β0. The smaller mean of α0 and lower CV of α0 and β0 result in a more significant reduc- tion in the cell mass CV (Figs 4G, S10E, and S10H). In summary, growth rate variability (char- acterized by the CVs of α0 and β0) amplifies cell mass variation, while strong growth rate modulation (small α0) can reduce cell mass variation throughout the cell cycle. To assess whether growth rate modulation in RPE-1 and HeLa cells can cause cell mass CV reduction throughout the cell cycle, we derived the parameters from the experimental data. The mean of α0 was determined based on the mean correlations in Fig 4A and 4B (S6 Table). To estimate the variation in α0, we used long-term live-cell growth trajectories. The CV of α0 was found to be independent of cell mass (S11A and S11B Fig). The variability in α0 arises from 2 sources: stochastic partitioning of cellular contents during cell division (intercellular variability) and intrinsic fluctuations in biochemical reactions (intracellular variability) [57]. The former, determined at birth, is a major contributor to cell mass variation, while the effect of the latter gradually cancels out over time, exerting minimal impact on cell mass variation. Therefore, we focused on the intercellular variability and estimated it by calculating the varia- tion among the means of individual growth trajectories (S11C Fig). The CV of α0 was esti- mated to be 0.33 for RPE-1 and 0.23 for HeLa cells, respectively. It is challenging to isolate the variation in β0 from measurement error, thus we conducted simulations with β0 having the same CV as α0 or with the CV of β0 being equal to zero. Using these parameters, we found that both RPE-1 and HeLa cells could reduce the cell mass CV after 1 cell cycle (Figs 4G, S10E, and S10H). Since the minimal requirement for cell mass homeostasis is to have a lower cell mass CV at division than that at birth, we concluded that growth rate modulation alone is sufficient to maintain cell mass homeostasis in RPE-1 and HeLa cells. When a plot of growth rate versus mass is in the BI form (Fig 4F and S6 Table), the slope of the mass versus growth correlation is close to exponential for small cells and becomes less posi- tive or even negative in large cells. This can be described by the equation: PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 13 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle t t (cid:0) (cid:0) � � � t (cid:0) g0m0 m0 � m0 t þ g0m0 þ a0m0 (cid:0) dm0 , where the mean of α0 is close to ln2 dt0 ¼ a0m0 m0 < m0 and the mean of γ0 is smaller than ln2 (Section 3.2 in S1 Text). The first term on the right side of the equation represents the exponential portion of the mass versus growth rate correlation, while the second term describes the part where growth rate modulation takes effect. Here, γ0 0 signifies the cell mass at which this modulation indicates the strength of modulation and mτ begins to take effect. Both γ0 and mτ mass, respectively. Our findings indicate that the increase in cell mass CV is primarily driven by the CV of α0 (S12A–S12D Fig). Additionally, we investigated the impact of the means of γ0, 0 on the change in cell mass CV throughout the cell cycle. We found that the smaller the and mτ means of γ0 and mτ cells it affects, the greater the cell mass CV reduction (Figs 4H and 4I and S12). 0, which means the stronger the modulation on growth rate and the more 0 are normalized by the means of cell cycle length and birth To assess whether the growth rate modulation on its own in U2OS, HT1080, and SaoS-2 cells can also lead to a reduction in cell mass CV, we simulated the changes in cell mass CV during the G1 or nonG1 phase using values of γ0 and mτ 0 obtained from the experimental data. When assuming a 20% CV for α0, growth rate modulation was found to decrease the cell mass CV in the G1 phase for U2OS and HT1080 cells (Fig 4H), as well as in the nonG1 phase for HT1080 cells (Fig 4I). However, it was not sufficient to reduce the cell mass CV in the nonG1 phase for SaoS-2 cells (Fig 4I). As the CV of α0 increases, the reduction in cell mass CV becomes less pronounced (S12E–S12H Fig). Eventually, all 3 cell lines fail to reduce cell mass CV at a 40% CV for α0 (S12G and S12H Fig). Notably, despite U2OS G1 cells exhibiting a greatly negative γ0 value, which indicated an exceptionally strong growth rate modulation, its effectiveness in reducing cell mass CV was lower than that of HT1080 G1 cells due to a smaller 0. proportion of affected cells in U2OS, represented by a larger mτ In summary, we found diverse patterns of correlation between cell mass and growth rate in different cell lines, and within the same cell line measured at different cell cycle stages. We developed stochastic models to explore the impact of different mass versus growth correlations on the change in cell mass CV throughout the cell cycle. These models are representations of the data itself and not contrived schemes. They suggest strongly that in many cases sub-expo- nential growth, either for all cells or even for a subset of cells, can be an effective means of reducing cell mass CV and can ensure cell mass homeostasis. Regulation of the cell cycle and regulation of growth rate compensate for each other to maintain cell mass homeostasis Both mass-dependent regulation of the progression through the cell cycle and mass-dependent regulation of growth rate are used by cells to reduce cell mass variation. To evaluate the relative importance of these processes in maintaining cell mass homeostasis, we have tried to perturb each mechanism individually in RPE-1 cells. To disrupt mass-dependent regulation of G1 length, we slowed entry into S phase using pal- bociclib, an inhibitor that specifically blocks the activation of Cdk4,6, which is required for entry into S phase [33]. As discussed, low concentrations of palbociclib increased the mean cell mass and, as expected, prolonged the cell cycle length by elongating the G1 phase (Fig 1H and S2 Table). However, once treated cells reached a new homeostatic state, the CV of birth mass remained unchanged compared to untreated cells (Fig 1I). When we analyzed the duration of each cell cycle phase as a function of cell mass, we found a reduced impact of cell mass on G1 phase length coupled with an enhanced impact on S phase length, characterized by the slopes and the correlation coefficients of the correlations between cell mass and the durations of these phases (Fig 5A, 5B, and 5K). Additionally, the mass-dependent regulation of G2 phase was diminished, yet still statistically significant (p = 0.0057) (Fig 5C and 5K). These opposite PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 14 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle changes in G1 and S phase regulation suggest that the mass-dependent regulation of S phase had effectively compensated for a weakened impact of cell mass on G1 length regulation. Hence, in specific circumstances such as palbociclib treatment, S phase can become the pri- mary period responsible for reducing cell mass variation (Fig 2G). Nevertheless, such compen- sation ultimately proves insufficient, resulting in a diminished cell mass-dependent regulation of the entire cell cycle length (Fig 5D and 5K). To maintain the birth mass CV at the same level as untreated cells, additional regulation of growth rate is required to further reduce cell mass variation during the cell cycle. Indeed, we found that the correlations between cell mass and growth rate in palbociclib-treated cells were even closer to linear growth compared to untreated cells (Fig 5E), implying a stronger growth rate modulation and a greater reduction in cell mass variation through growth rate regulation. The unchanged CV of birth mass when cells are treated with the G1/S inhibitor, palbociclib (Fig 1I), is a collective result of the inter- play between mass-dependent cell cycle regulation and mass-dependent growth rate regula- tion. Thus, the cell mass CV is maintained despite a significant increase in the mean birth mass (Fig 1H). In a converse experiment, we specifically perturbed cell growth rate. We treated cells with rapamycin to inhibit mTOR activity. Treatment with rapamycin resulted in an elongation of the cell cycle (S2 Table) and a decrease in mean cell mass (Fig 1H). Similar to the results with palbociclib treatment, rapamycin treatment left the birth mass CV unchanged (Fig 1I). Cell mass-dependent feedback on G1 length was enhanced in the presence of rapamycin (Fig 5F and 5K), while feedback on the S and G2-M phases were weakened (Fig 5G, 5H, and 5K). Additionally, the minimal lengths of all cell cycle phases were slightly increased compared to untreated cells (Fig 5F–5H). In the presence of rapamycin, the cell mass fed back more strongly on the entire cell cycle length, as indicated by the more negative slope and correlation coeffi- cient of the mass versus cell cycle length correlation (Fig 5I and 5K). Furthermore, the relative strengths of correlations between cell mass and cell cycle phase lengths aligned with the reduced cell mass CV in the corresponding phases: for example, cell mass CV was primarily reduced in the G1 phase with rapamycin treatment (Fig 2H), consistent with the strengthened cell cycle regulation in the G1 phase. On the other hand, we found that the slopes of the mass versus growth correlations in both the G1 and nonG1 phases closely resembled that of expo- nential growth (Fig 5J), suggesting a weaker role of growth rate regulation in maintaining cell mass homeostasis when growth rate is inhibited by rapamycin. From the experiments described above, mass-dependent cell cycle regulation and mass- dependent growth rate modulation must interact with each other to maintain the birth mass CV at a consistent level even when the G1/S transition or cell growth rate is perturbed, result- ing in significant changes in the mean birth mass. After studying the feedback of cell mass on cell cycle length and growth rate under many different circumstances, we felt a need for a new way to compare the response of each under different conditions. We have found it convenient to define a new parameter to represent the strength of this linkage. We utilized the normalized slope of birth mass versus cell cycle length correlation as the parameter λ0, which quantifies the strength of mass-dependent cell cycle regulation. The value of λ0 is always negative. A more negative value of λ0 indicates stronger regulation. Additionally, since the slopes of the cell mass versus growth rate correlations in the G1 and nonG1 phases were similar in RPE-1 and HeLa cells, we found it useful to calculate the average slope of these phases and normalized it by the mean doubling time to represent the strength of mass-dependent growth rate regulation, which we denoted as α0. The value of α0 is smaller than or equal to ln2, which represents expo- nential growth. A smaller value of α0 indicates a greater deviation from exponential growth and thus a stronger modulation of growth rate. We found an inverse correlation between λ0 and α0 across all the conditions we have investigated (Fig 5L and S7 Table), suggesting a PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 15 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle Fig 5. The compensatory roles of mass-dependent cell cycle regulation and mass-dependent growth rate regulation in maintaining cell mass homeostasis. (A–D) The correlations between the lengths of the G1 (A), S (B), G2-M phases(C), and the full cell cycle (D) and the initial mass of the corresponding period in RPE-1 cells treated with 50 nM palbociclib. Each gray dot is an observation; black squares indicate the average of each cell mass bin; error bars are SEM; solid black line is the best fit of the black squares; solid red lines are the corresponding correlations in untreated RPE-1 cells. (E) Correlations between cell mass and growth rate in the G1 (blue) and nonG1 (red) phases for RPE-1 cells treated with 50 nM palbociclib. Filled squares represent the median growth rate of each bin; error bars show SEM. The black dashed lines indicate the expected behavior for exponential growth. The solid blue and red lines are the best fit of the filled squares. (F–I) The correlation between the lengths of the G1 (F), S (G), G2-M phases (H), and the full cell cycle (I) and the initial mass of the corresponding period in RPE-1 cells treated with 100 nM rapamycin. (J) Correlations between cell mass and growth rate in the G1 (blue) and nonG1 (red) phases for RPE-1 cells treated with 100 nM rapamycin. (K) Kendall rank correlations between the duration of indicated cell cycle phase and cell mass at the initiation of the respective phase, in untreated RPE-1 cells, RPE-1 treated with 50 nM palbociclib, and RPE-1 treated with 100 nM rapamycin. (L) The correlation between the normalized slope of birth mass vs. cell cycle length correlation, λ0, and the normalized slope of cell mass vs. growth rate correlation, α0, depicted for untreated HeLa and RPE-1 cells, as well as RPE-1 cells treated with palbociclib or rapamycin. The values of λ0 and α0 used in this plot are listed in S7 Table. (M) The contribution of each control mechanism shown as the reduction in the simulated division mass CV when the respective control mechanism is included compared to that without any control mechanisms. Simulation parameters were obtained from experimental data measured in untreated HeLa and RPE-1 cells, as well as RPE-1 cells treated with palbociclib or rapamycin. The data underlying this figure and the scripts used to generate the plots are available on the Open Science Framework at osf.io/3kyvw. CV, coefficient of variation; SEM, standard error of the mean. https://doi.org/10.1371/journal.pbio.3002453.g005 compensatory effect between the regulation of cell cycle and growth rate (i.e., the strengths of these regulatory processes tend to change in opposite directions). For example, when cell cycle regulation was inhibited (e.g., by palbociclib), the modulation of growth rate became stronger, and conversely, when growth rate regulation was inhibited (e.g., by rapamycin), the modula- tion of cell cycle length became stronger. These findings highlight the compensatory roles played by these 2 processes in maintaining cell mass homeostasis. To illustrate further the compensatory roles of regulation on cell cycle and growth rate, we developed a stochastic model to simulate changes in cell mass variation throughout the cell cycle (Section 4 in S1 Text). In this model, we considered 3 factors that could contribute to the increase of cell mass variation: variability in cell cycle length, variability in growth rate, and noise in cell mass partition during mitotic division. For simplicity, we only considered PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 16 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle intercellular noise as the source of growth rate variability, which is due to stochasticity in the partitioning of cellular contents during cell division, as previously discussed (S11C Fig and Section 3.1 in S1 Text). As control mechanisms, we considered mass-dependent regulation of the duration of G1 and nonG1 phases separately, and we also considered mass-dependent growth modulation throughout the entire cell cycle. We chose all the parameters in this model from our actual experimental data and evaluated the impact of each control mechanism by comparing the cell mass CV at division with and without these control mechanisms. Notably, we observed some discrepancies between the simulated division mass CV, incorporating all 3 control mechanisms, and the values measured in experiments (S8 Table). These may arise from the simplification of variability in growth rate (S11C Fig and Section 4 in S1 Text), which effectively influences cell mass variation (S13 Fig) but is quite challenging to estimate accu- rately from experimental data. Nevertheless, these simulations largely reflect the relative signif- icance of each control mechanism in maintaining cell mass homeostasis. The model results indicate that in RPE-1 cells, the regulation of G1 length plays a slightly greater role compared to nonG1 length regulation, but both are overshadowed by the modula- tion of growth rate (Fig 5M). When the G1/S control is inhibited by palbociclib, the contribu- tion of G1 length regulation slightly decreases, the contribution of nonG1 regulation slightly increases, and the role of growth rate modulation becomes even more dominant (Fig 5M). On the other hand, inhibiting growth with rapamycin leads to an increase in the dominance of G1 length regulation, with its contribution now comparable to that of growth rate modulation, while the impact of nonG1 regulation becomes smaller (Fig 5M). In HeLa cells, the cell mass variation is considerably smaller than that in RPE-1 cells (S8 Table) when not including any control mechanisms, due to the lower variation in growth rate in HeLa cells. It is worth noting that HeLa cells possess a mutated G1/S network. Its ranking of contributions from the 3 mech- anisms is similar to the scenario observed in RPE-1 cells treated with palbociclib, which dis- rupts the G1/S transition. Specifically, in HeLa cells, the contribution of growth rate modulation outweighs that of nonG1 length regulation, which, in turn, outweighs that of G1 length regulation (Fig 5M). These findings collectively reveal compensatory roles of cell cycle and growth rate regula- tion in reducing cell mass variation, particularly distinguishing the regulation of G1 length and the regulation of growth rate. Generally, growth rate modulation, rather than cell cycle regulation, is the more dominant mechanism. When one feedback process is hindered, other mechanisms become relatively stronger to maintain cell mass variation at a similar level. Growth rate modulation, rather than cell cycle regulation, consistently plays the predominant role in reducing cell mass CV, regardless of whether or not the cells possess an intact G1/S cir- cuit. In the most extreme case, we studied when the growth rate is inhibited by rapamycin, the contribution of growth rate modulation is on par with that of G1 length regulation. These observations contradict the conventional size control models [1,14,19,44,58–62], which predict that G1/S control is the primary contributor to size homeostasis in mammalian cells. Other explanations for how a population of cells might reduce its cell mass variation We evaluated additional processes that could potentially contribute to the reduction in cell mass CV but were not accounted for in our stochastic model. In principle, any process that affects the likelihood of cell division or cell viability differentially in large and small cells could influence the distribution of cell mass within a population. To estimate the importance of such effects, we examined the rate of cell death and cell cycle arrest through long-term measure- ments of cell growth and proliferation. During the 48 to 72-h duration of our cell PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 17 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle measurements, we defined cell cycle arrest events as instances where a cell remained in the same cell cycle phase while its mass continued to increase throughout the experiment. Further- more, cell death was identified by a sudden and drastic decrease in cell dry mass, suggesting cell membrane permeabilization. We found events of cell cycle arrest or cell death in the culture affected no more than 2% of cells in all the conditions that were studied (S9 Table). In particular, neither cell cycle arrest nor cell death occurred frequently enough to contribute significantly to cell mass homeostasis in any of the experiments that we have described. It is worth noting that the remarkably low frequency of cell cycle arrest in cells treated with rapamycin and palbociclib at the drug con- centrations used in this study suggests that these drugs at low concentrations do not induce quiescence or senescence at the population level (S9 Table). Furthermore, the concentrations of these drugs did not appear to be toxic enough to cause significant cell death (S9 Table). One intriguing observation was that some large RPE-1 cells treated with palbociclib experienced a partial loss of cytoplasm during mitosis (S9 Table and S1 Movie). This cytoplasmic loss could be attributed to incomplete cortical contraction during mitotic rounding [63]. The amount of mass loss appeared to be random. Notably, these rare events, accounting for approximately 0.5% of cells, did not have a significant impact at the population level on cell mass homeostasis in the presence of palbociclib. It is worth noting that although these mechanisms were of negligible importance in the spe- cific experimental setting of our study, they might still play a significant role in a tissue setting, for example, during wound healing, regeneration, aging, and/or disease. A picture of cell mass homeostasis in proliferating cells Homeostasis refers to the maintenance of a balance between inherent noise in cellular pro- cesses and feedback control mechanisms that correct for them. In proliferating cells, this noise arises from stochastic variation in growth rate, cell cycle length, and cell mass partitioning dur- ing mitosis. To reduce cell mass variation, mass-dependent regulation can occur through the control of cell cycle progression, growth rate, or both. To illustrate mass regulation graphically as a balance between noise and control mecha- nisms, we have depicted the concept of cell mass homeostasis as a “teeter-totter” (Fig 6). Sto- chastic noise and feedback control mechanisms are represented as opposing forces on either side of the lever’s fulcrum; the sizes of the icons represent the importance of the processes, as determined from the stochastic models (S8 Table). When these effects are balanced, the system reaches a steady state. In cell lines like RPE-1, where the G1/S circuit is intact, the relative importance of the control mechanisms can be ranked from greatest (heaviest on the teeter-tot- ter) to smallest (lightest on the teeter-totter) as follows: growth rate modulation, G1 length reg- ulation, and nonG1 length regulation. A perturbation of the system leads to changes in the stochastic nature of the processes and affects the operation of specific control mechanisms. When this happens, other control mechanisms compensate for these changes and restore the balance. For example, when G1/S control is inhibited, either through pharmacological inhibi- tors, such as palbociclib, or genetic mutations in the G1/S circuitry, as seen in HeLa cells, the contribution of G1 length regulation is reduced. In response, nonG1 length regulation and growth rate modulation become more significant. Conversely, when growth rate modulation is inhibited, such as by rapamycin, G1 length regulation becomes more important, and growth rate modulation contributes less. Overall, the teeter-totter of cell mass homeostasis is robustly balanced through the compen- satory interactions of these different control processes within the cell. It is likely that the coor- dination and adjustment of these compensatory mechanisms at the molecular level are crucial PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 18 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle Fig 6. The teeter-totter model of cell mass homeostasis. Cell mass homeostasis requires a balance between stochastic noise and control mechanisms. In unperturbed cells with an intact G1/S circuitry, the weights of control mechanisms from the heaviest to the lightest are the growth rate modulation, G1 length regulation, and nonG1 length regulation. When G1/S control is perturbed, the impact of the G1 length regulation becomes smaller, and the nonG1 length regulation and growth rate modulation become larger to compensate. When the growth rate modulation is suppressed, the G1 length regulation plays a more prominent role in compensating for the reduced impact of growth rate modulation. https://doi.org/10.1371/journal.pbio.3002453.g006 for cellular survival under changing conditions. While our understanding of how these mecha- nisms achieve balance has advanced, further study is needed to elucidate how they coordinate and adapt their compensation at the molecular level to maintain balance under changed condi- tions and how this plays out in health, disease, aging, etc. Discussion To summarize: in examining cell mass homeostasis, we found that stochastic variation in cell mass in proliferating cells is tightly controlled throughout the cell cycle (Fig 2) via mass-depen- dent regulation of cell growth rate (Fig 4) and mass-dependent regulation of cell cycle progres- sion (Fig 3). Generally speaking, among the cell lines and cell cycle and cell growth inhibitors that we have employed (including those previously studied and analyzed), we conclude that the G1/S transition does not appear to be a privileged place where cell mass regulation is imposed. Rather mass regulation occurs throughout the cell cycle phases. The compensation that keeps stochastic variation of mass in check emerges from an interplay of these mecha- nisms and results in effective cell mass regulation. Not only is homeostasis maintained, but it is also maintained at high stringency, as indicated by the narrow distribution of cell mass at birth (Fig 1). Furthermore, cell mass homeostasis is robust to changes in genetic background and is resistant to manipulations of the G1/S transition or perturbation of mTOR activity (Fig 1). The birth size CVs measured in many proliferating bacterial, yeast, mammalian, and plant cells fall in a relatively small range (from 11% to 25%) (S10 Table), which is comparable to the birth weight CV of a human fetus [64]. Although it is not clear whether such strict control is explicitly selected for during evolution or merely a by-product of some other selection [65,66], cell size homeostasis appears to be highly regulated and presumably important. Though we PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 19 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle focused on cultured human cell lines in this study, the mechanisms underlying cell size homeostasis, just as the mechanisms underlying the cell cycle itself, are likely to be conserved. In this study, we utilized ceQPM [29] as a means of measuring cell dry mass, providing a complementary approach to previous studies that focused on cell volume as an indicator of cell size [3,7]. We found that many aspects of the behavior of cell mass, as directly measured by ceQPM, were consistent with studies of cell volume, particularly those reported by Cadart and colleagues, who obtained high-quality cell volume data [7]. For example, in line with their observations, we also identified inverse correlations between initial mass and cell cycle phase duration in both the G1 and nonG1 phases in HeLa cells (Fig 3), the existence of a minimal duration of the G1 phase (Fig 3), the “adder”-like correlation between the birth and division masses (Fig 3), and the coordination between mass-dependent cell cycle regulation and growth rate modulation in maintaining cell mass homeostasis (Fig 5). This consistency is further sup- ported by our recent findings that cell volume usually changes proportionally with cell mass in cultured proliferating cells, except during mitosis, resulting in a narrow distribution of cell mass density [67]. However, we were able to observe more detailed discrepancies in the regula- tion of mass and volume growth. For example, while Cadart and colleagues reported that vol- ume growth rate is dependent on cell volume at birth [7], we found that mass growth rate is related to cell mass at any point of the cell cycle, and this relationship varies across different cell cycle stages (Figs 4 and S8). Moreover, the noise in mass growth rate appears to affect the slope of the correlation (S11 Fig), in contrast to Cadart and colleagues’ findings of noise pri- marily impacting the intercept of volume growth rate [68]. These discrepancies may be attrib- uted to inherent differences in the factors affecting mass or volume and the speed and mechanisms by which cells respond to perturbations or fluctuations in mass or volume [69]. Aside from confirming previous discoveries, our findings took a significant step forward in exploring mechanisms underlying cell mass homeostasis. Extensive data collection on large populations of cells was possible thanks to the high-throughput of ceQPM [29]. From these extensive measurements, we derived reliable correlations between cell mass, the durations of cell cycle phases, and the growth rate. We studied these across multiple cell lines and under various pharmacologic perturbations. We were able to fit such data to simple functions (Figs 3, 4, and 5), which facilitated our ability to derive quantitative models. These models, in turn, facilitated our interpretation of the underlying cellular responses. For example, we showed how G1, S, and G2 phases are each under negative regulation by cell mass in both transformed and non-transformed cells (Fig 3). A particularly noteworthy discovery was the identification of a minimum length for each phase of the cell cycle in large cells, which explains the limited impact of cell cycle regulation on very large cells, leaving the underlying process to growth rate modulation (Fig 3). We further demonstrated that growth rate is modulated differently in dif- ferent cell types or cell lines (Fig 4). Such comprehensive characterization of growth regulation was not previously possible without the extensive and precise measurements of cell mass and growth rate by ceQPM [29]. When we perturbed cells by inhibiting the G1/S transition or sup- pressing the growth rate (Fig 5), ceQPM enabled us to go beyond the qualitative phenomena observed in previous studies [8,11,12]. It allowed us not only to determine the average changes in cell mass, cell cycle phase duration, and growth rate but also to measure these qualities at the single-cell level, tracking the individual cells over time. This enabled us to derive important quantitative correlation functions. These functions in turn allowed us to write deterministic equations, incorporate stochastic noise, and ultimately develop a stochastic model. With this model, we could estimate the relative weight of each of the regulatory mechanisms employed in maintaining cell mass homeostasis and finally deduce how the weights of these separate mechanisms depend on each other (Fig 5). PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 20 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle One simple finding stands out. It has been generally assumed, and widely cited in review articles and textbooks of biology, that G1 length regulation is the predominant or even the sole mechanism controlling cell size during the cell cycle [1,14,19,44,58–62]. There was always an appeal of this simple mechanism, as it made perturbation of the cell cycle at G1/S the whole process for cell size control. We now can say that this is clearly not the case. Our current highly quantitative studies involving at least hundreds of cells per condition demonstrated that, at least for the cell lines we employed, the impact of G1 length regulation on constraining cell mass CV within a proliferating cell population is much less significant than the modulation of cells’ mass accumulation (growth) rate (Fig 5). This holds true for non-transformed cells with intact G1/S control. Furthermore, even in the presence of growth inhibition induced by rapa- mycin, the contribution of growth rate modulation to cell mass CV reduction is no less than that of G1 length regulation. Why would there be size-dependent growth rate regulation if regulation of cell cycle pro- gression were sufficient to control cell size? With so many essential genes in the genome, it seems like a weak argument to claim that having 2 separate mechanisms provides increased security for survival. We propose instead that they serve 2 separate functions. Control of the G1 length might be used primarily to set the cell size for a given cell type. In this view, the G1/ S transition is hard-wired into developmental pathways like the MAP kinase pathway or the BMP pathway through proteins like TGFβ. By contrast, control of cell growth might be pri- marily used for a different purpose: maintaining cell size homeostasis of any given cell type against environmental or stochastic variation. It makes more sense that the targeted mean size of a given cell type is controlled by a few key molecular players downstream of specific hor- monal or nutrient signals or cellular differentiation. Those molecular players (such as CDK4/6 or other CDK inhibitors) were described as a cell size “dial” in a previous model by Tan and colleagues [11]. However, once cells are programmed to adopt a defined size in their new state, they would still require a mechanism to maintain size homeostasis around that new mean by buffering against environmental or internal stochastic fluctuation. Consistent with the work presented here (Fig 1) and studies in budding and fission yeasts [13,17], deletion or overex- pression of the G1/S inhibitors change the mean size dramatically but have only limited effects on the variation of cell size. Furthermore, systems that only act at a single gate for size variation would fail to provide continuous feedback on size variation and would have difficulty correct- ing noise introduced after that gate operates, which in this case is early in the cell cycle [70]. By contrast, growth rate regulation, particularly sub-exponential growth, where growth rate is proportional to cell mass but exhibits a slope smaller than that of exponential growth, proves to be a highly effective means for reducing cell mass variation throughout the cell cycle (Fig 4 and Section 3 in S1 Text). The effectiveness of this mechanism is bolstered by its operation throughout the entire cell cycle and in the whole cell size range. This form of regulation would be more effective than growth rate modulation restricted to short periods of the cell cycle and only in large cells, as suggested by previous studies [1,8,20,38,49]. Unraveling the determinant factors that underlie the sub-exponential scaling between growth rate and cell mass will likely shed light on the coordination between size-dependent biomass synthesis, nutrient transporta- tion, and macromolecule destruction [71]. We can imagine that pathological conditions, such as aging related diseases, may target growth rate regulation and therefore affect cells at differ- ent stages of cell cycle or even non-growing cells. Aside from the mass-dependent regulation on the G1 length and cell growth rate, the regu- lation of nonG1 phase lengths also contributes significantly to the reduction of cell mass varia- tion (Fig 5). This is presumably due to the fact that cell cycle phases outside of G1 have non- negligible negative correlations with cell mass (Fig 3) and often occupy a larger portion of the cell cycle than the G1 phase (S2 Table). The mechanisms regulating G2 length have been PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 21 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle mainly studied in fission yeast, where the G2/M transition acts as the major size control check- point [17,72–74]. Mammalian cells share homologous components of this G2/M regulation with fission yeast [75,76], suggesting that similar mechanisms might function during this stage in mammalian cells. However, further investigation beyond citing simple homology will be needed to confirm this possibility. The regulation of S phase length as a means for controlling cell size in mammalian cells has been rarely explored. One potential mechanism of size-depen- dent S phase length regulation could involve the control of the number of replication com- plexes. If the number of forks were proportional to the total cell size, so that small cells made fewer forks, this could serve to lengthen S phase [77]. If cell size were to affect the number of active origins or DNA replication speed, it might also affect the level of DNA damage due to the under-replicated regions [77–79]. Replication stress is not uncommon in normal cycling populations, as evidenced by the presence of DNA lesions in more than 20% of G1 cells in non-transformed cell lines [80]. If the occurrence of replication stress were influenced by cell size and if it led to forms of DNA damage that could be resolved, it could potentially drive tumorigenesis or senescence in a cell size-dependent manner, resulting in heterogeneous behavior in a genetically uniform population. This scenario might hold clinical significance and thus deserves further investigation. Additional research is needed to establish the relation- ship between the probability of replication stress and cell size during S phase. Furthermore, the actual mechanism of S phase length regulation could be more complicated than the size- dependent replication fork number. The negative correlation between cell mass and S phase length is strengthened in palbociclib-treated RPE-1 cells compared to untreated cells (Fig 5), suggesting more complex crosstalk between the G1 and S phase regulation that cannot be fully explained by the size-dependent replication fork number. In line with previous research [7], we found that both RPE-1 and HeLa cells exhibit adder- like behaviors (Fig 3I–3J). More specifically, they demonstrate sequential adder behaviors, wherein each phase of their cell cycles can be mathematically expressed as an adder, with the correlation between the masses at the beginning and the end of each phase having a slope close to one (S7 Fig). Our focus in this study is not on their adherence to an adder model. Rather, we emphasize the existence of size control mechanisms across all cell cycle phases. Such regula- tion could manifest at multiple cell cycle checkpoints by controlling the duration of individual cell cycle phases, operate throughout the cell cycle through continuous monitoring and adjust- ing the rate of mass accumulation, or more likely, be a combination of both. If the time resolu- tion of the measurements were sufficiently high, we might be able to observe that each fine segment of the cell cycle follows an adder behavior. Such a mechanism would require that a cell continually “knows” how large it is and how large it should be at any point of the cell cycle. How might cells sense their size relative to a changing standard that changes with cell cycle progression? How would such a mechanism respond differently in different cell types, differ- ent nutrient conditions, and to pharmacological perturbations? A proposed mechanism of cell size sensing relies on some form of disproportionality of molecular components or signals to cell size. For example, cells might sense size through the sub-scaling of inhibitors or super-scal- ing of activators to regulate their cell cycle length [6,10,44,81]. Cell mass accumulation requires nutrient provision, transcription, translation, and degradation; any rate-limiting step might serve as a size sensor. It has also been proposed that cells may sense size and modulate growth rate by DNA limitation, cytoplasmic dilution, surface-to-volume ratio, sublinear proportional- ity between metabolic rate and cell size, transport efficiency, and other such mechanisms [20,70,82–84]. We have found that different cell lines modulate their growth rates differently. It is of course plausible that each cell line we investigated employs a distinct size-sensing mech- anism and a distinct mode of response of mass accumulation. However, it seems more likely that all cell lines share a universal mechanism that allows various forms of growth rate PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 22 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle modulation under particular conditions. One potential candidate for this universal mechanism would be the mTOR pathway, which governs biomass synthesis and responds to various upstream signals [36,85]. Therefore, we suggest that an investigation of how the mTOR path- way responds to cell size could be informative. Additionally, growth rate regulation exhibits cell cycle-specific patterns and even intrinsic oscillations [29,48,50,51]. The likely coexistence of multiple forms of regulation could complicate any investigation. Future studies might bene- fit from isolating each mechanism, perhaps by identifying conditions where only one of the processes is dominant. Situations such as cell cycle arrest and size enlargement (so called cellu- lar senescence) triggered by DNA damage or other stresses are of particular interest in this regard [44,86]. Such phenomena may help us disentangle size-dependent growth regulation from other forms of cell cycle-dependent growth regulation, thus allowing us to focus on the effects of cell size on growth rate using the methods we employed in this study. In summary, the use of ceQPM to quantify single-cell dry mass, mass growth rate, and cell cycle progression has provided the currently most accurate, complete, and quantitative description of cell mass homeostasis in mammalian cells. In this paper, we have also showcased the often-underappreciated power of phenomenological descriptions. Such descriptions have been proven to be inherently powerful in physics and chemistry. The observed reduction in the coefficient variation of cell mass within a proliferating population throughout the cell cycle unequivocally rules out the possibility that cells control mass solely or principally by control- ling the length of the G1 phase at the G1/S transition. While this result is far from a complete answer to the problem of cell size homeostasis and does not yet provide specific molecular mechanisms, it nevertheless can serve as a guide for future investigation. It redirects our focus away from the G1/S transition or any specific cell cycle transitions in cell size homeostasis. We propose instead focusing on the molecular-level mechanisms governing size-dependent regu- lation of growth rate, as this appears to be the predominant player and holds greater promise in elucidating how cells maintain a stable size distribution. Our findings, which reveal com- pensatory responses to perturbing size, suggest the existence of previously underappreciated regulatory pathways in cell size regulation. Specifically, we suggest that there is a need to exam- ine how cell size feeds back on the anabolic or proteostatic machinery. As of now, we are still in the early stage of describing the phenomenon of size homeostasis in quantitative terms. These efforts prove that we have much to learn about the regulatory cir- cuits that tell a cell how large it is and how large it should be at any given time or in any given circumstance. Studying cell size homeostasis in cultured cells can lay the groundwork for future investigations into size control in vivo and its implications for disease, thereby expand- ing our understanding of cell physiology. Materials and methods Cell culture and chemical treatment HeLa mAG-hGem, RPE-1 mAG-hGem, HT1080 mAG-hGem mKO2-hCdt1, and HeLa mAG-hGEM DNA-ligase-dsRed cells were made in previous studies by our laboratory [38,87]. U2OS mAG-hGem and Saos-2 mAG-hGEM cells were generated by lentivirus infection in this study. Lentivirus carrying mTurquoise2-SLBP was purchased from Addgene (83842-LV) to make HeLa mAG-hGem mTurquoise2-SLBP, RPE-1 mAG-hGem mTurquoise2-SLBP, and HeLa mAG-hGEM DNA-ligase-dsRed mTurquoise2-SLBP. Single clones of stable expression were selected for each cell line. Cells were incubated at 37˚C with 5% CO2 in Dulbecco’s Modi- fied Eagle Medium (DMEM) (11965; Thermo Fisher Scientific) with 25 mM HEPES (15630080; Thermo Fisher Scientific) and 10 mM sodium pyruvate (11360070; Thermo Fisher Scientific), or McCoy’s 5A Medium (16600082; Thermo Fisher Scientific). Both media were PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 23 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle supplemented with 10% fetal bovine serum (FBS) (16000044; Thermo Fisher Scientific) and 1% penicillin/streptomycin (15140122; Thermo Fisher Scientific). Palbociclib was purchased from Selleckchem (PD-0332991) and rapamycin was purchased from LC Laboratories (R- 5000). Live cell imaging Cells were imaged at 10× magnification by an Eclipse Ti microscope with the Perfect Focus System (PFS) (Nikon, Japan) and an SID4BIO camera (Phasics, France). Nikon NIS-Elements AR ver. 4.13.0.1 software with the WellPlate plugin was used to acquire images. A home-made incubation chamber was used to maintain a constant environment of 36˚C and 5% CO2 dur- ing imaging. Cells were seeded on 6-well glass bottom plates (P06G-1.5-14-F; MatTek) at a density of 1,500 cells/cm2 3 h before long-term imaging or 3,500 cells/cm2 16 h before short- term imaging. Before time-lapse imaging was started, mineral oil (M8410; Millipore Sigma) was added into each well to prevent media evaporation. In the long-term experiments studying the cell cycle regulation, cells were monitored for 48 or 72 h. In the short-term experiments studying growth rate modulation, cells were monitored for 3 h. For all experiments, the phase images were acquired every 30 min, and the fluorescence images were acquired every 1 h. Cell fixation and cell cycle identification After the short-term time-lapse imaging, the mineral oil was gently removed by aspiration. Cells were fixed with 4% paraformaldehyde (RT 157–8; Electron Microscopy Sciences) and stained with Hoechst 33342 (62249; Thermo Fisher Scientific) at a final concentration of 1 μm. The cells were then imaged by QPM again to identify their cell cycle stages. QPM image processing and data analysis The QPM images were processed by the ceQPM method developed previously [29] and con- ducted on the O2 high-performance computing cluster at Harvard Medical School. To test the significance of the minimal cell cycle phase length, we fitted the binned correla- tions between the initial mass and cell cycle phase duration in Figs 3 and 5 with 2 alternative models. A linear model y = a1x+b1, and a bilinear model y ¼ a2x þ b2ðx � x0Þ; y ¼ a2x0 þ b2ðx > x0Þ, where y is the cell cycle phase length, x is the ini- tial mass, a1, b1, a2, b2, and x0 are the fitting parameters. We used the Akaike information crite- rion (AIC) to compare the goodness of fits. A smaller AIC indicates a better fit, and the relative likelihood p_linear or p_bilinear predicts the probability that the alternative model is a better fit when the linear or bilinear model has the smaller AIC [88]. Since the correlations between the initial mass and cell cycle phase duration were not linear, we utilized the Kendall’s rank correlation coefficient to represent the correlation strength. This coefficient is more suit- able for our data as it does not assume a linear relationship, unlike the widely used Pearson correlation coefficient [89]. To evaluate whether the cell cycle control could explain the adder behavior in Fig 3I and 3J, we assumed cells grow exponentially at the rate of α = ln(2)/DT, where DT is the averaged cell cycle length. The division mass could be predicted by md = mbeαT, T = f(mb), where f is the best-fitted function in the alternative models of cell cycle length versus birth mass. To fit the binned correlation between growth rate and cell mass in Figs 4 and 5, we employed 2 alternative models: a linear model dm dt dm dt Þ þ gm þ amt þ b (cid:0) gmt ¼ am þ b ð Þ m < mt ð ð ¼ am þ b and a bilinear model ð Þ m � mt Þ, where y is the growth rate, x is PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 24 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle the cell mass, α, β, a, b, γ, and mτ are the fitting parameters. We used the AIC to estimate the goodnesses of fits. Supporting information S1 Text. Models used in this study. (DOCX) d þ Q2; CV2 S1 Fig. The left- and right-hand sides of Eq 1 and their difference quantified in HeLa cells. d is indicated in black, Q2 is indicated in white; error bars are the In the term, CV2 standard deviation of 8 experiments. The data underlying this figure and the scripts used to generate the plots are available on the Open Science Framework at osf.io/3kyvw. (EPS) S2 Fig. RPE-1 and U2OS sensitivity to palbociclib. The mean cell mass of the population (A) and the percentage of G1 cells quantified by low Geminin expression (B) after being treated in palbociclib at the indicated concentrations for 2 days. Dashed black lines show the concentra- tion (50 nM) chosen for the analyses in this study. The data underlying this figure and the scripts used to generate the plots are available on the Open Science Framework at osf.io/ 3kyvw. (EPS) S3 Fig. mAG-hGeminin (A) and cell mass (B) trajectories of a representative HeLa cell. Dashed lines denote the timing of the G1/S transition identified by the initiation of geminin accumulation. The data underlying this figure and the scripts used to generate the plots are available on the Open Science Framework at osf.io/3kyvw. (EPS) S4 Fig. Segregation of cells into stages along the cell cycle mean path. (A) The 2D plane of the logarithmic scale of mAG-hGeminin intensity, log(Geminin), and the intensity of Hoechst fluorescence, DNA, in asynchronous RPE-1 cells. Black contours indicate cell number density; the solid red line is the cell cycle mean path; filled red circles show the centroids of the chosen stages along the mean path; the stages are evenly separated in the time axis computed by the ERA method [38]. (B–E) The averages of log(Geminin) (blue) and DNA content (red) change with cell cycle progression in different cell lines. X-axes are calculated by the ERA method [38]. The cell cycle is segregated into 4 phases indicated by color-shaded areas: the early G1 phase from birth to the onset of geminin accumulation, the late G1 phase from the initiation of geminin accumulation to the onset of DNA replication, the S phase covering DNA replication, and the G2-M phase where geminin and DNA accumulation plateau. (F) Error in computed cell mass CV caused by inaccurate cell cycle stage identification. The cell dry mass and cell cycle markers data were from Fig 2D. We added 10% random Gaussian noise to each cell’s position in the log(Geminin)-DNA plane. The cells were reassigned to cell cycle stages accord- ing to their new positions, and the cell mass CV of each stage was computed. The solid black line and error bars indicate the mean and standard deviation of computed cell mass CVs of 100 simulations; the first and last stages were truncated due to having much higher cell num- bers and variations than other stages. (G, I) The 2D planes of log(Geminin) and DNA content in RPE-1 cells in 50 nM palbociclib (G) or 100 nM rapamycin (I). The red line and filled circles are the cell cycle mean path and centroids of stages calculated from the treated cells. (H, J) The averages of log(Geminin) (blue) and DNA content (red) change with cell cycle progression in RPE-1 cells in 50 nM palbociclib (H) or 100 nM rapamycin (J). The data underlying this figure and the scripts used to generate the plots are available on the Open Science Framework at osf. PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 25 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle io/3kyvw. (EPS) S5 Fig. The geminin and SLBP markers faithfully report the timing and duration of S phase. (A) The trajectories of dsRed-DNA-ligase I foci, mAG-hGeminin, and mTurquoi- se2-SLBP in a representative HeLa cell. Open circles are the raw data; solid colored lines are the spline interpolations; dashed yellow and pink lines mark the S phase start and end, respec- tively. (B–D) Correlations between the S phase start (B), end (C), and duration (D) identified by the dsRed-DNA-ligase foci or mAG-hGeminin and mTurquoise2-SLBP combined. Each black dot is one observation; Solid red lines are the best linear fit. Texts indicate the functions of the solid red lines and the Pearson correlations of the black dots. The data underlying this figure and the scripts used to generate the plots are available on the Open Science Framework at osf.io/3kyvw. (EPS) S6 Fig. The impact of minimal cell cycle length on cell mass homeostasis, indicated by the birth mass CV (A) and mean birth mass (B) changing with simulated generations. Differ- ent colors show the percentage of cells affected by the minimal cell cycle length in the popula- tion of the first generation of simulations. The data underlying this figure and the scripts used to generate the plots are available on the Open Science Framework at osf.io/3kyvw. (EPS) S7 Fig. The sequential adder behavior in RPE-1 and HeLa cells. (A, D) The correlations between birth mass and mass at G1/S in RPE-1 (A) and HeLa (D) cells. (B, E) The correlations between mass at G1/S and mass at S/G2 in RPE-1 (B) and HeLa (E) cells. (C, F) The correla- tions between mass at S/G2 and division mass in RPE-1 (C) and HeLa (F) cells. Each gray dot is an observation; black squares are the average of each cell mass bin; error bars are the stan- dard error of means (SEMs). Solid black lines are the best linear fits of the gray dots; texts indi- cate the functions of the solid black lines. The data underlying this figure and the scripts used to generate the plots are available on the Open Science Framework at osf.io/3kyvw. (EPS) S8 Fig. Growth rate modulation in HeLa cells. (A) The correlation between cell mass and growth rate in HeLa cells when pooling all cells together. Each gray dot is an observation in the 3-h measurements, n = 18,334. Black squares are the median growth rate of each mass bin; error bars are SEMs. The solid black line is the best fit of the black squares (S5 Table). The dashed black line indicates exponential growth. (B, C) The correlations between cell mass and growth rate in HeLa cells in 4 cell cycle phases (B) and one fine stage of the cell cycle (C). The stages were determined by log(Geminin) and DNA using the ERA method [38], as indicated in S4C Fig. Filled squares are the median growth rate of each mass bin; error bars are SEMs. The solid lines are the best fit of the filled squares (S5 Table). The dashed black line in (B) indi- cates exponential growth. (D) The slope of the linear relationship between cell mass and growth rate plotted against cell cycle progression. The short-dashed line indicates the expected slope for exponential growth. The data underlying this figure and the scripts used to generate the plots are available on the Open Science Framework at osf.io/3kyvw. (EPS) S9 Fig. Specific growth rate changes with cell cycle progression in RPE-1 (A) and HeLa cells (B) in G1 (blue) and nonG1 (red) phases. Since the binned correlation could be affected by inspection bias [52], we investigated how the specific growth rate (growth rate divided by mass) changed with cell cycle progression from the long-term trajectories as recommended by PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 26 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle Kar and colleagues [52]. We arbitrarily assumed the G1 or nonG1 phase each occupies half of the cell cycle when normalizing the length of the growth trajectories. Solid blue and red lines are the means of the normalized growth trajectories of the G1 and nonG1 segments; the shaded areas indicate SEM. Dashed lines are the expected curves of exponential growth; short- dashed lines are the expected curves of linear growth, assuming the cells behave like an adder. The data underlying this figure and the scripts used to generate the plots are available on the Open Science Framework at osf.io/3kyvw. (EPS) S10 Fig. Simulation results for the sub-exponential growth rate modulation. (A, C, F) Con- tour plots illustrating the rate of change in cell mass CV at the beginning of the cell cycle (t0 = 0) when assuming CVα0 = CVβ0 (A), CVβ0 = 0 (C), or CVα0 = 0 (F), respectively. Here, μα0 repre- sents the mean of α0. (B, D, G) Contour plots illustrating the rate of change in cell mass CV at the end of the cell cycle (t0 = 1) when assuming CVα0 = CVβ0 (B), CVβ0 = 0 (D), or CVα0 = 0 (G), respectively. (E, H) Contour plots illustrating the overall change in cell mass CV throughout the cell cycle when assuming CVβ0 = 0 (E), or CVα0 = 0 (H), respectively. Solid circles indicate the corresponding positions in the contour plots when adopting parameter values from the experimental observations of RPE-1 and HeLa cells. The data underlying this figure and the scripts used to generate the plots are available on the Open Science Framework at osf.io/ 3kyvw. (EPS) S11 Fig. Estimating the variability in α0 for RPE-1 and HeLa cells. (A, B) The variability of the specific growth rate, defined as the growth rate divided by cell mass, does not change with cell mass for RPE-1 (A) and HeLa (B) cells. Blue squares and lines indicate the means and stan- dard deviations of live cell growth trajectories, which are binned by cell mass. The black lines show ð1 � �CV Þ �gr j, where �CV is the average CV in specific growth rate for all cell mass bins, and (cid:0) gr j is the average specific growth rate for each bin. (C) Schematic illustrating the defini- tions of intercellular and intracellular variability in α0. Solid lines are representative live cell growth trajectories. Dashes lines represent the means of each trajectory. Intercellular variabilty is defined as the variaiton among the means of each trajectories, while intracellular variability is defined as the fluctruation within individual trajectories. The data underlying this figure and the scripts used to generate the plots are available on the Open Science Framework at osf.io/ 3kyvw. (EPS) t , when applying variability t (C), or equal variability to all 3 parameters (D). S12 Fig. Simulation results for the Bilinear growth rate modulation. (A–D) Three-dimen- sional (3D) volumetric plots showing how the change in cell mass CV throughout the cell cycle responds to the means of γ0 and m0 t, represented by μα0 and mm0 (CV) to only 1 parameter of α0 (A), γ0 (B), m0 The slice planes are orthogonal to the CV axis at CV = 0.2. (E, F) Contour plots illustrating the change in cell mass CV during the G1 (E) and nonG1 phases (F) with the means of γ0 and m0 t when assuming a 30% CV in α0. (G, H) Contour plots illustrating the change in cell mass CV during the G1 (G) and nonG1 phases (H) with means of γ0 and m0 t when assuming a 40% CV in α0. Solid circles in (E–H) indicate the corresponding positions in the contour plots when adopting parameter values from the experimental data. The data underlying this figure and the scripts used to generate the plots are available on the Open Science Framework at osf.io/ 3kyvw. (EPS) PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 27 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle S13 Fig. Impact of growth rate variability on division mass CV, CV(mi(Ti)), in the stochas- tic model. The stochastic model is described in Section 4, Scenario IX in S1 Text. All parame- ter values used in this simulation are listed in the table at the end of Section 4, with the exception of CVgr, which is varied in this simulation. Solid blue lines indicate the simulation results. Filled blue circles are the division mass CV when simulated with the CVgr estimated from experimental data. Dashed black lines represent the division mass CV measured in experiments. The data underlying this figure and the scripts used to generate the plots are available on the Open Science Framework at osf.io/3kyvw. (EPS) S1 Table. Characteristics of the human cell lines used in this study. (DOCX) S2 Table. The durations of cell cycle phases for HeLa, RPE-1, RPE-1 in 100 nM rapamycin or 50 nM palbociclib at cell mass homeostasis. MAD is the median absolute deviation, and nMAD is MAD normalized by the median in robust statistics. (DOCX) S3 Table. Comparing cell cycle phase durations and mass versus phase length correlations with and without the mTurquoise2-SLBP marker in HeLa cells. (DOCX) S4 Table. Comparison of the linear and bilinear fits for the cell mass vs. cell cycle phase length correlations. The significantly better fits (p_bilinear or p_linear < 0.05) and the signifi- cant negative correlations (p < 0.05) are highlighted. (DOCX) S5 Table. Comparison of the linear and bilinear fits for the cell mass vs. growth rate corre- lations. The significantly better fits (p_bilinear or p_linear < 0.05) are highlighted. (DOCX) S6 Table. The normalized fitting parameters for the cell mass vs. growth rate correlations for different cell lines. For correlations fitted better by the linear model, dm dt normalized parameters α0 and β0 are listed in the table, with α0 = α<T>, b0 ¼ b <T> <T> and <mb> are the means of cell cycle length and cell birth mass, respectively. For expo- nential growth, α0 = ln2 ~= 0.693. For correlations fitted better by the bilinear model, ¼ am þ b ð dm dt b0, γ0, and m0 The correlation slopes, α0, a0, and γ0, lower than 0.75 or higher than 1.25-fold (arbitrarily cho- sen thresholds) of ln2 were highlighted. SE and BI denote the type of growth rate modulation, where SE stands for sub-exponential and BI stands for bilinear. (DOCX) Þ þ gm þ amt þ b (cid:0) gmt t are listed in the table, with a0 = a<T>, b0 ¼ b <T> Þ, the normalized parameters a0, <mb> ; g0 ¼ g < T >; m0 <mb>. ¼ am þ b, the <mb>, where ð Þ m < mt ð Þ m � mt t ¼ mt ð S7 Table. The values of λ0 and α0 used in Fig 5L, for untreated HeLa and RPE-1 cells, as well as RPE-1 cells treated with 50 nM palbociclib or 100 nM rapamycin. (DOCX) S8 Table. Contribution of each factor to cell mass variation, as indicated by the division mass CV simulated using the stochastic model in Section 4 in S1 Text. The values reported in this table are the average of division mass CVs obtained from 50 simulations. (DOCX) PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 28 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle S9 Table. The frequencies of cell death, cell cycle arrest, and cytoplasmic loss observed in the long-term measurements in HeLa, RPE-1, RPE-1 in 100 nM rapamycin, and RPE-1 in 50 nM palbociclib when cells have reached cell mass homeostasis. (DOCX) S10 Table. Birth size CVs, division size CVs, and DA stds. reported in the literature. (DOCX) S1 Movie. Time-lapse quantitative phase images of RPE-1 cells in 50 nM palbociclib; the time interval is 30 min; the yellow arrow indicates the lost cytoplasmic mass of a mitotic cell (red arrow); the scale bar indicates 100 μm. (GIF) Acknowledgments We thank the Nikon Imaging Center at Harvard Medical School for sharing its resources. We thank the Research Computing Group at Harvard Medical School for providing support for the O2 Computing Cluster for imaging processing and data storage. We thank Johan Paulsson, Ariel Amir, Prathitha Kar, Ethan Levien, Ahmed Rattani, Wenzhe Ma, Gabriel Neurohr, Simon Gemble, and Renata Basto for insightful suggestions. We thank William Ratzan for proofreading the manuscript. Author Contributions Conceptualization: Xili Liu, Marc W. Kirschner. Data curation: Xili Liu. Formal analysis: Xili Liu, Jiawei Yan. Funding acquisition: Marc W. Kirschner. Investigation: Xili Liu. Methodology: Xili Liu. Writing – original draft: Xili Liu, Jiawei Yan, Marc W. Kirschner. Writing – review & editing: Xili Liu, Marc W. Kirschner. References 1. Ginzberg MB, Kafri R, Kirschner M. On being the right (cell) size. Science [Internet]. 2015; 348 (6236):1245075–1245075. Available from: http://www.sciencemag.org/cgi/doi/10.1126/science. 1245075. 2. Killander D, Zetterberg a. A quantitative cytochemical investigation of the relationship between cell mass and initiation of DNA synthesis in mouse fibroblasts in vitro. Exp Cell Res [Internet]. 1965 Oct; 40 (1):12–20. Available from: http://www.ncbi.nlm.nih.gov/pubmed/5838935. https://doi.org/10.1016/ 0014-4827(65)90285-5 PMID: 5838935 3. Varsano G, Wang Y, Wu M, Varsano G, Wang Y, Wu M. Probing Mammalian Cell Size Homeostasis by Channel-Assisted Cell Reshaping. Cell Rep [Internet]. 2017; 20(2):397–410. https://doi.org/10.1016/j. celrep.2017.06.057 PMID: 28700941 4. Xie S, Skotheim JM. A G1 Sizer Coordinates Growth and Division in the Mouse Epidermis. Curr Biol [Internet]. 2020; 30(5):916–924.e2. https://doi.org/10.1016/j.cub.2019.12.062 PMID: 32109398 5. Dolznig H, Grebien F, Sauer T, Beug H, Mu¨ llner EW. Evidence for a size-sensing mechanism in animal cells. Nat Cell Biol [Internet]. 2004 Sep [cited 2012 Dec 5]; 6(9):899–905. Available from: http://www. ncbi.nlm.nih.gov/pubmed/15322555. https://doi.org/10.1038/ncb1166 PMID: 15322555 PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 29 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle 6. Zatulovskiy E, Zhang S, Berenson DF, Topacio BR, Skotheim JM. Cell growth dilutes the cell cycle inhibitor Rb to trigger cell division. Science (80-) [Internet]. 2020 Jul 24; 369(6502):466–71. Available from: http://science.sciencemag.org/. https://doi.org/10.1126/science.aaz6213 PMID: 32703881 7. Cadart C, Monnier S, Grilli J, Sa´ez PJ, Srivastava N, Attia R, et al. Size control in mammalian cells involves modulation of both growth rate and cell cycle duration. Nat Commun [Internet]. 2018; 9(1). https://doi.org/10.1038/s41467-018-05393-0 PMID: 30115907 8. Ginzberg MB, Chang N, Kafri R, Kirschner MW. Cell size sensing in animal cells coordinates anabolic growth rates and cell cycle progression to maintain cell size uniformity. Elife [Internet]. 2018;(1):123851 +. https://doi.org/10.7554/eLife.26957 PMID: 29889021 9. 10. 11. 12. Zhang S, Zatulovskiy E, Arand J, Sage J, Skotheim JM. The cell cycle inhibitor RB is diluted in G1 and contributes to controlling cell size in the mouse liver. Front Cell Dev Biol [Internet]. 2022 Aug 25; 10. Available from: https://www.frontiersin.org/articles/10.3389/fcell.2022.965595/full. https://doi.org/10. 3389/fcell.2022.965595 PMID: 36092730 Lanz MC, Zatulovskiy E, Swaffer MP, Zhang L, Ilerten I, Zhang S, et al. Increasing cell size remodels the proteome and promotes senescence. Mol Cell [Internet]. 2022 Sep; 82(17):3255–3269.e8. Avail- able from: https://linkinghub.elsevier.com/retrieve/pii/S1097276522007134. https://doi.org/10.1016/j. molcel.2022.07.017 PMID: 35987199 Tan C, Ginzberg MB, Webster R, Iyengar S, Liu S, Papadopoli D, et al. Cell size homeostasis is main- tained by CDK4-dependent activation of p38 MAPK. Dev Cell [Internet]. 2021 May;1–14. https://doi.org/ 10.1016/j.devcel.2021.04.030 PMID: 34022133 Liu S, Ginzberg MB, Patel N, Hild M, Leung B, Li Z, et al. Size uniformity of animal cells is actively main- tained by a p38 MAPK-dependent regulation of G1-length. Elife [Internet]. 2018 Mar 29; 7(April):1–27. Available from: https://elifesciences.org/articles/26947. https://doi.org/10.7554/eLife.26947 PMID: 29595474 13. Chen Y, Zhao G, Zahumensky J, Honey S, Futcher B. Differential Scaling of Gene Expression with Cell Size May Explain Size Control in Budding Yeast. Mol Cell [Internet]. 2020; 78(2):359–370.e6. https:// doi.org/10.1016/j.molcel.2020.03.012 PMID: 32246903 14. Zetterberg A, Larsson O. Kinetic analysis of regulatory events in G1 leading to proliferation or quies- cence of Swiss 3T3 cells. Proc Natl Acad Sci U S A [Internet]. 1985 Aug 1; 82(16):5365–9. Available from: http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=390569&tool= pmcentrez&rendertype=abstract. https://doi.org/10.1073/pnas.82.16.5365 PMID: 3860868 15. Araujo AR, Gelens L, Sheriff RSM, Santos SDM. Positive Feedback Keeps Duration of Mitosis Tempo- rally Insulated from Upstream Cell-Cycle Events. Mol Cell [Internet]. 2016; 64(2):362–375. https://doi. org/10.1016/j.molcel.2016.09.018 PMID: 27768873 16. Garmendia-Torres C, Tassy O, Matifas A, Molina N, Charvin G. Multiple inputs ensure yeast cell size homeostasis during cell cycle progression. Elife [Internet]. 2018 Jul 4; 7:1–27. Available from: https:// elifesciences.org/articles/34025. https://doi.org/10.7554/eLife.34025 PMID: 29972352 17. Sveiczer A, Novak B, Mitchison JM. The size control of fission yeast revisited. J Cell Sci [Internet]. 1996 Dec; 109(Pt 1):2947–57. Available from: http://www.ncbi.nlm.nih.gov/pubmed/9013342. https://doi.org/ 10.1242/jcs.109.12.2947 PMID: 9013342 18. 19. Turner JJ, Ewald JC, Skotheim JM. Cell Size Control in Yeast. Curr Biol [Internet]. 2012 May [cited 2012 May 7]; 22(9):R350–9. Available from: http://linkinghub.elsevier.com/retrieve/pii/ S0960982212001923. https://doi.org/10.1016/j.cub.2012.02.041 PMID: 22575477 Zatulovskiy E, Skotheim JM. On the Molecular Mechanisms Regulating Animal Cell Size Homeostasis. Trends Genet [Internet]. 2020; 36(5):360–72. https://doi.org/10.1016/j.tig.2020.01.011 PMID: 32294416 20. Neurohr GE, Terry RL, Lengefeld J, Bonney M, Brittingham GP, Moretto F, et al. Excessive Cell Growth Causes Cytoplasm Dilution And Contributes to Senescence. Cell [Internet]. 2019 Feb; 176(5):1083– 1097.e18. Available from: http://www.ncbi.nlm.nih.gov/pubmed/30739799. https://doi.org/10.1016/j. cell.2019.01.018 PMID: 30739799 21. Si F, Le Treut G, Sauls JT, Vadia S, Levin PA, Jun S. Mechanistic Origin of Cell-Size Control and Homeostasis in Bacteria. Curr Biol [Internet]. 2019; 29(11):1760–1770.e7. https://doi.org/10.1016/j. cub.2019.04.062 PMID: 31104932 22. Zlotek-Zlotkiewicz E, Monnier S, Cappello G, Le Berre M, Piel M. Optical volume and mass measure- ments show that mammalian cells swell during mitosis. J Cell Biol [Internet]. 2015; 211(4):765–774. Available from: http://jcb.rupress.org/content/211/4/765.abstract. https://doi.org/10.1083/jcb. 201505056 PMID: 26598614 23. Cooper KL, Oh S, Sung Y, Dasari RR, Kirschner MW, Tabin CJ. Multiple phases of chondrocyte enlargement underlie differences in skeletal proportions. Nature [Internet]. 2013 Mar 13 [cited 2013 Mar PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 30 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle 14]; 495(7441):375–8. Available from: http://www.nature.com/doifinder/10.1038/nature11940. PMID: 23485973 24. Son S, Kang JH, Oh S, Kirschner MW, Mitchison TJ, Manalis S. Resonant microchannel volume and mass measurements show that suspended cells swell during mitosis. J Cell Biol [Internet]. 2015; 211 (4):757–763. Available from: http://www.jcb.org/cgi/doi/10.1083/jcb.201505058. PMID: 26598613 25. Venkova L, Vishen AS, Lembo S, Srivastava N, Duchamp B, Ruppel A, et al. A mechano-osmotic feed- back couples cell volume to the rate of cell deformation. Elife [Internet]. 2022; 11:2021.06.08.447538. Available from: https://doi.org/10.1101/2021.06.08.447538%0Ahttps://www.biorxiv.org/content/10. 1101/2021.06.08.447538v2%0Ahttps://www.biorxiv.org/content/10.1101/2021.06.08.447538v2. abstract. 26. Godin M, Delgado FF, Son S, Grover WH, Bryan AK, Tzur A, et al. Using buoyant mass to measure the growth of single cells. Nat Methods [Internet]. 2010 [cited 2013 Jan 10]; 7(5):387–90. Available from: http://www.nature.com/nmeth/journal/vaop/ncurrent/full/nmeth.1452.html. https://doi.org/10.1038/ nmeth.1452 PMID: 20383132 27. Zangle TA, Teitell MA. Live-cell mass profiling: an emerging approach in quantitative biophysics. Nat Methods [Internet]. 2014; 11(12):1221–1228. Available from: http://www.nature.com/doifinder/10.1038/ nmeth.3175. PMID: 25423019 28. Popescu G, Park K, Mir M, Bashir R. New technologies for measuring single cell mass. Lab Chip [Inter- net]. 2014; 14(4):646–652. Available from: http://xlink.rsc.org/?DOI=C3LC51033F. https://doi.org/10. 1039/c3lc51033f PMID: 24322181 29. Liu X, Oh S, Peshkin L, Kirschner MW. Computationally enhanced quantitative phase microscopy reveals autonomous oscillations in mammalian cell growth. Proc Natl Acad Sci U S A [Internet]. 2020 Nov 3; 117(44):27388–99. Available from: http://www.pnas.org/lookup/doi/10.1073/pnas.2002152117. PMID: 33087574 30. Huh D, Paulsson J. Random partitioning of molecules at cell division. Proc Natl Acad Sci U S A [Inter- net]. 2011 Sep 6; 108(36):15004–9. Available from: https://pnas.org/doi/full/10.1073/pnas.1013171108. PMID: 21873252 31. Scott SJ, Suvarna KS, D’Avino PP. Synchronization of human retinal pigment ephitilial-1 (RPE-1) cells in mitosis. J Cell Sci [Internet]. 2020 Jan 1; Available from: https://journals.biologists.com/jcs/article/doi/ 10.1242/jcs.247940/266615/Synchronization-of-human-retinal-pigment. 32. Sung Y, Tzur A, Oh S, Choi W, Li V, Dasari RR, et al. Size homeostasis in adherent cells studied by syn- thetic phase microscopy. Proc Natl Acad Sci U S A [Internet]. 2013 Oct 8 [cited 2014 Mar 21]; 110 (41):16687–92. Available from: http://www.ncbi.nlm.nih.gov/pubmed/24065823. https://doi.org/10. 1073/pnas.1315290110 PMID: 24065823 33. Fry DW, Harvey PJ, Keller PR, Elliott WL, Meade M, Trachet E, et al. Specific inhibition of cyclin-depen- dent kinase 4/6 by PD 0332991 and associated antitumor activity in human tumor xenografts. Mol Can- cer Ther [Internet]. 2004 Nov; 3(11):1427–38. Available from: http://www.ncbi.nlm.nih.gov/pubmed/ 15542782. PMID: 15542782 34. Morgan D. The cell cycle: principles of control. Oxford Univesity Press; 2007. 35. Li J, Kim SG, Blenis J. Rapamycin: One drug, many effects. Cell Metab [Internet]. 2014; 19(3):373–9. https://doi.org/10.1016/j.cmet.2014.01.001 PMID: 24508508 36. Saxton RA, Sabatini DM. mTOR Signaling in Growth, Metabolism, and Disease. Cell Cell Press. 2017; 168:960–976. 37. Sakaue-Sawano A, Kurokawa H, Morimura T, Hanyu A, Hama H, Osawa H, et al. Visualizing spatiotem- poral dynamics of multicellular cell-cycle progression. Cell [Internet]. 2008 Feb 8 [cited 2013 Feb 10]; 132(3):487–98. Available from: http://www.ncbi.nlm.nih.gov/pubmed/18267078. https://doi.org/10. 1016/j.cell.2007.12.033 PMID: 18267078 38. Kafri R, Levy J, Ginzberg MB, Oh S, Lahav G, Kirschner MW. Dynamics extracted from fixed cells reveal feedback linking cell growth to cell cycle. Nature [Internet]. 2013 Feb 27 [cited 2013 Feb 27]; 494 (7438):480–3. Available from: http://www.nature.com/doifinder/10.1038/nature11897. PMID: 23446419 39. Bajar BT, Lam AJ, Badiee RK, Oh Y-H, Chu J, Zhou XX, et al. Fluorescent indicators for simultaneous reporting of all four cell cycle phases. Nat Methods [Internet]. 2016 Dec 31; 13(12):993–6. Available from: https://www.nature.com/articles/nmeth.4045. https://doi.org/10.1038/nmeth.4045 PMID: 27798610 40. McGarry TJ, Kirschner MW. Geminin, an Inhibitor of DNA Replication, Is Degraded during Mitosis. Cell [Internet]. 1998 Jun; 93(6):1043–53. Available from: https://linkinghub.elsevier.com/retrieve/pii/ S009286740081209X. https://doi.org/10.1016/s0092-8674(00)81209-x PMID: 9635433 41. Whitfield M, Zheng L, Baldwin A, Ohta T, Hurt M, Marzluff W. Stem-loop binding protein, the protein that binds the 30 end of histone mRNA, is cell cycle regulated by both translational and posttranslational PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 31 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle mechanisms. Cell Biol [Internet]. 2000 [cited 2014 Jan 15]; 20:4188–98. Available from: http://mcb.asm. org/content/20/12/4188.short. https://doi.org/10.1128/MCB.20.12.4188-4198.2000 PMID: 10825184 42. Leonhardt H, Rahn H-P, Weinzierl P, Sporbert A, Cremer T, Zink D, et al. Dynamics of DNA Replication Factories in Living Cells. J Cell Biol [Internet]. 2000 Apr 17; 149(2):271–80. Available from: https:// rupress.org/jcb/article/149/2/271/32118/Dynamics-of-DNA-Replication-Factories-in-Living. https://doi. org/10.1083/jcb.149.2.271 PMID: 10769021 43. Cardoso MC, Joseph C, Rahn H-P, Reusch R, Nadal-Ginard B, Leonhardt H. Mapping and Use of a Sequence that Targets DNA Ligase I to Sites of DNA Replication In Vivo. J Cell Biol [Internet]. 1997 Nov 3; 139(3):579–87. Available from: https://rupress.org/jcb/article/139/3/579/737/Mapping-and-Use-of-a- Sequence-that-Targets-DNA. https://doi.org/10.1083/jcb.139.3.579 PMID: 9348276 44. Xie S, Swaffer M, Skotheim JM. Eukaryotic Cell Size Control and Its Relation to Biosynthesis and Senescence. Annu Rev Cell Dev Biol [Internet]. 2022 Oct 6; 38(1):291–319. Available from: https:// www.annualreviews.org/doi/10.1146/annurev-cellbio-120219-040142. PMID: 35562854 45. Cooper S. Control and maintenance of mammalian cell size. BMC Cell Biol. 2004; 5:1–21. 46. Scott M, Hwa T. Bacterial growth laws and their applications. Curr Opin Biotechnol [Internet]. 2011 May 16 [cited 2011 Jun 27]; 22(4):559–65. Available from: http://www.ncbi.nlm.nih.gov/pubmed/21592775. https://doi.org/10.1016/j.copbio.2011.04.014 PMID: 21592775 47. Son S, Tzur A, Weng Y, Jorgensen P, Kim J, Kirschner MW, et al. Direct observation of mammalian cell growth and size regulation. Nat Methods [Internet]. 2012 Sep [cited 2012 Nov 5]; 9(9):910–2. Available from: http://www.ncbi.nlm.nih.gov/pubmed/22863882. https://doi.org/10.1038/nmeth.2133 PMID: 22863882 48. Mu L, Kang JH, Olcum S, Payer KR, Calistri NL, Kimmerling RJ, et al. Mass measurements during lym- phocytic leukemia cell polyploidization decouple cell cycle- And cell size-dependent growth. Proc Natl Acad Sci U S A [Internet]. 2020 Jul 7; 117(27):15659–65. Available from: http://biorxiv.org/cgi/content/ short/2019.12.17.879080v1?rss=1&utm_source=researcher_app&utm_medium=referral&utm_ campaign=RESR_MRKT_Researcher_inbound. https://doi.org/10.1073/pnas.1922197117 PMID: 32581119 49. Liu S, Tan C, Melo-gavin C, Mark KG, Ginzberg MB, Blutrich R, et al. Large cells activate global protein degradation to maintain cell size homeostasis. bioRxiv. 2021:1–31. 50. Miettinen TP, Kang JH, Yang LF, Manalis SR. Mammalian cell growth dynamics in mitosis. Elife [Inter- net]. 2019 May 7; 8:1–29. Available from: https://elifesciences.org/articles/44700. https://doi.org/10. 7554/eLife.44700 PMID: 31063131 51. Ghenim L, Allier C, Obeid P, Herve´ L, Fortin J-Y, Balakirev M, et al. A new ultradian rhythm in mamma- lian cell dry mass observed by holography. Sci Rep [Internet]. 2021; 11(1):1–12. https://doi.org/10. 1038/s41598-020-79661-9 PMID: 33446678 52. Kar P, Tiruvadi-Krishnan S, Ma¨nnik J, Ma¨ nnik J, Amir A. Distinguishing different modes of growth using single-cell data. Elife [Internet]. 2021 Dec 2; 10. Available from: https://elifesciences.org/articles/72565. https://doi.org/10.7554/eLife.72565 PMID: 34854811 53. Amir A. Cell Size Regulation in Bacteria. Phys Rev Lett [Internet]. 2014 May 23; 112(20):208102. Avail- able from: https://link.aps.org/doi/10.1103/PhysRevLett.112.208102. 54. Thomas P. Analysis of Cell Size Homeostasis at the Single-Cell and Population Level. Front Phys [Inter- net]. 2018 Jun 26; 6(June). Available from: https://www.frontiersin.org/article/10.3389/fphy.2018. 00064/full. 55. Vargas-Garcia CA, Bjo¨rklund M, Singh A. Modeling homeostasis mechanisms that set the target cell size. Sci Rep [Internet]. 2020 Aug 18; 10(1):13963. Available from: https://www.nature.com/articles/ s41598-020-70923-0. https://doi.org/10.1038/s41598-020-70923-0 PMID: 32811891 56. Ho P-Y, Lin J, Amir A. Modeling Cell Size Regulation: From Single-Cell-Level Statistics to Molecular Mechanisms and Population-Level Effects. Annu Rev Biophys [Internet]. 2018 May 20; 47(1):251–71. Available from: https://www.annualreviews.org/doi/10.1146/annurev-biophys-070317-032955. PMID: 29517919 57. Thomas P, Terradot G, Danos V, Weiße AY. Sources, propagation and consequences of stochasticity in cellular growth. Nat Commun [Internet]. 2018; 9(1):1–11. https://doi.org/10.1038/s41467-018-06912- 9 PMID: 30375377 58. Cook M, Tyers M. Size control goes global. Curr Opin Biotechnol [Internet]. 2007 Aug [cited 2012 Nov 5]; 18(4):341–50. Available from: http://www.ncbi.nlm.nih.gov/pubmed/17768045. https://doi.org/10. 1016/j.copbio.2007.07.006 PMID: 17768045 59. Echave P, Conlon IJ, Lloyd AC. Cell Size Regulation in Mammalian Cells. Cell Cycle [Internet]. 2007 Oct 28 [cited 2015 Feb 16]; 6(2):218–24. Available from: http://www.tandfonline.com/doi/abs/10.4161/ cc.6.2.3744. PMID: 17245129 PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 32 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle 60. Jorgensen P, Tyers M. How cells coordinate growth and division. Curr Biol [Internet]. 2004 Dec 14 [cited 2012 Mar 6]; 14(23):R1014–27. Available from: http://www.ncbi.nlm.nih.gov/pubmed/15589139. https://doi.org/10.1016/j.cub.2004.11.027 PMID: 15589139 61. Umen JG. The elusive sizer. Curr Opin Cell Biol [Internet]. 2005 Aug [cited 2012 May 21]; 17(4):435– 41. Available from: http://www.ncbi.nlm.nih.gov/pubmed/15978795. https://doi.org/10.1016/j.ceb.2005. 06.001 PMID: 15978795 62. 63. Liu S, Tan C, Tyers M, Zetterberg A, Kafri R. What programs the size of animal cells? Front Cell Dev Biol [Internet]. 2022 Nov 1; 10(November):1–19. Available from: https://www.frontiersin.org/articles/10. 3389/fcell.2022.949382/full. https://doi.org/10.3389/fcell.2022.949382 PMID: 36393871 Taubenberger A V., Baum B, Matthews HK. The Mechanics of Mitotic Cell Rounding. Front Cell Dev Biol [Internet]. 2020 Aug 6; 8(August):1–16. Available from: https://www.frontiersin.org/article/10.3389/ fcell.2020.00687/full. https://doi.org/10.3389/fcell.2020.00687 PMID: 32850812 64. Voldner N, Frey Frøslie K, Godang K, Bollerslev J, Henriksen T. Determinants of birth weight in boys and girls. human_ontogenetics [Internet]. 2009 Mar 18; 3(1):7–12. Available from: https://onlinelibrary. wiley.com/doi/10.1002/huon.200900001. 65. Amir A. Is cell size a spandrel? Elife [Internet]. 2017 Jan 19; 6:1–8. Available from: https://elifesciences. org/articles/22186. https://doi.org/10.7554/eLife.22186 PMID: 28102818 66. ElGamel M, Mugler A. Effects of molecular noise on cell size control. 2023;(1). Available from: http:// arxiv.org/abs/2303.15232. 67. Liu X, Oh S, Kirschner MW. The uniformity and stability of cellular mass density in mammalian cell cul- ture. Front Cell Dev Biol [Internet]. 2022 [cited 2022 Oct 14]; 10. Available from: https://internal-journal. frontiersin.org/articles/10.3389/fcell.2022.1017499/full. https://doi.org/10.3389/fcell.2022.1017499 PMID: 36313562 68. Cadart C, Venkova L, Piel M, Cosentino Lagomarsino M. Volume growth in animal cells is cell cycle dependent and shows additive fluctuations. Elife [Internet]. 2022 Jan 28; 11. Available from: https:// elifesciences.org/articles/70816. https://doi.org/10.7554/eLife.70816 PMID: 35088713 69. Cadart C, Venkova L, Recho P, Lagomarsino MC, Piel M. The physics of cell-size regulation across timescales. Nat Phys [Internet]. 2019; 15(10):993–1004. https://doi.org/10.1038/s41567-019-0629-y 70. Bjo¨rklund M. Cell size homeostasis: Metabolic control of growth and cell division. Biochim Biophys Acta —Mol Cell Res [Internet]. 2019 Mar; 1866(3):409–17. https://doi.org/10.1016/j.bbamcr.2018.10.002 PMID: 30315834 71. Cadart C, Heald R. Scaling of biosynthesis and metabolism with cell size. Schroer T, editor. Mol Biol Cell [Internet]. 2022 Aug 1; 33(9):1–6. Available from: https://www.molbiolcell.org/doi/10.1091/mbc. E21-12-0627. PMID: 35862496 72. Turner JJ, Ewald JC, Skotheim JM. cell size control in yeast. Curr Biol. 2012; 29(6):997–1003. https:// doi.org/10.1016/j.cub.2012.02.041 PMID: 22575477 73. Navarro FJ, Nurse P. A systematic screen reveals new elements acting at the G2/M cell cycle control. Genome Biol [Internet]. 2012; 13(5):R36. Available from: http://genomebiology.biomedcentral.com/ articles/10.1186/gb-2012-13-5-r36. https://doi.org/10.1186/gb-2012-13-5-r36 PMID: 22624651 74. Keifenheim D, Sun XM, D’Souza E, Ohira MJ, Magner M, Mayhew MB, et al. Size-Dependent Expres- sion of the Mitotic Activator Cdc25 Suggests a Mechanism of Size Control in Fission Yeast. Curr Biol [Internet]. 2017; 27(10):1491–1497.e4. https://doi.org/10.1016/j.cub.2017.04.016 PMID: 28479325 75. Donzelli M, Draetta GF. Regulating mammalian checkpoints through Cdc25 inactivation. EMBO Rep [Internet]. 2003 Jul; 4(7):671–7. Available from: https://www.embopress.org/doi/10.1038/sj.embor. embor887. PMID: 12835754 76. Zhao RY, Elder RT. Viral infections and cell cycle G2/M regulation. Cell Res [Internet]. 2005 Mar; 15 (3):143–9. Available from: https://www.nature.com/articles/7290279. https://doi.org/10.1038/sj.cr. 7290279 PMID: 15780175 77. Gemble S, Bernhard SV, Srivastava N, Wardenaar R, Nano M, Mace´ A-S, et al. Mechanisms of genetic instability in a single S-phase following whole genome doubling. bioRxiv [Internet]. 2021;2021.07.16.452672. Available from: https://www.biorxiv.org/content/10.1101/2021.07.16. 452672v1%0Ahttps://www.biorxiv.org/content/10.1101/2021.07.16.452672v1.abstract. 78. Mei L, Cook JG. Efficiency and equity in origin licensing to ensure complete DNA replication. Biochem Soc Trans [Internet]. 2021 Nov 1; 49(5):2133–41. Available from: https://portlandpress.com/ biochemsoctrans/article/49/5/2133/229829/Efficiency-and-equity-in-origin-licensing-to. https://doi.org/ 10.1042/BST20210161 PMID: 34545932 79. Maya-Mendoza A, Moudry P, Merchut-Maya JM, Lee M, Strauss R, Bartek J. High speed of fork pro- gression induces DNA replication stress and genomic instability. Nature [Internet]. 2018; 559 (7713):279–284. https://doi.org/10.1038/s41586-018-0261-5 PMID: 29950726 PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 33 / 34 PLOS BIOLOGY Cell size homeostasis is tightly controlled throughout the cell cycle 80. Arora M, Moser J, Phadke H, Basha AA, Spencer SL. Endogenous Replication Stress in Mother Cells Leads to Quiescence of Daughter Cells. Cell Rep [Internet]. 2017; 19(7):1351–1364. https://doi.org/10. 1016/j.celrep.2017.04.055 PMID: 28514656 81. Chen Y, Futcher B. Scaling gene expression for cell size control and senescence in Saccharomyces cerevisiae. Curr Genet [Internet]. 2021 Feb 5; 67(1):41–7. Available from: http://link.springer.com/10. 1007/s00294-020-01098-4. https://doi.org/10.1007/s00294-020-01098-4 PMID: 33151380 82. Lin J, Amir A. Homeostasis of protein and mRNA concentrations in growing cells. Nat Commun [Inter- net]. 2018 Dec 29; 9(1):4496. Available from: https://www.biorxiv.org/content/early/2018/01/29/255950. https://doi.org/10.1038/s41467-018-06714-z PMID: 30374016 83. Harris LK, Theriot JA. Relative Rates of Surface and Volume Synthesis Set Bacterial Cell Size. Cell [Internet]. 2016 Jun; 165(6):1479–92. Available from: https://linkinghub.elsevier.com/retrieve/pii/ S0092867416306481. https://doi.org/10.1016/j.cell.2016.05.045 PMID: 27259152 84. Rishal I, Kam N, Perry RBT, Shinder V, Fisher EMCC, Schiavo G, et al. A Motor-Driven Mechanism for Cell-Length Sensing. Cell Rep [Internet]. 2012; 1(6):608–616. https://doi.org/10.1016/j.celrep.2012.05. 013 PMID: 22773964 85. Laplante M, Sabatini DM. mTOR signaling in growth control and disease. Cell [Internet]. 2012 Apr 13 [cited 2014 Jul 9]; 149(2):274–93. Available from: http://www.pubmedcentral.nih.gov/articlerender.fcgi? artid=3331679&tool=pmcentrez&rendertype=abstract. https://doi.org/10.1016/j.cell.2012.03.017 PMID: 22500797 86. Kumari R, Jat P. Mechanisms of Cellular Senescence: Cell Cycle Arrest and Senescence Associated Secretory Phenotype. Front Cell Dev Biol [Internet]. 2021 Mar 29 [cited 2022 Nov 14]; 9:485. Available from: https://www.frontiersin.org/articles/10.3389/fcell.2021.645593/full. https://doi.org/10.3389/fcell. 2021.645593 PMID: 33855023 87. Ginzberg MB. Size control and uniformity in animal cells. Harvard University; 2015. 88. Burnham KP, Anderson DR. Model Selection and Multimodel Inference [Internet]. Burnham KP, Ander- son DR, editors. New York, NY: Springer New York; 2004. Available from: http://link.springer.com/10. 1007/b97636. 89. Abdi H. Kendall Rank Correlation Coefficient. In: The Concise Encyclopedia of Statistics [Internet]. New York, NY: Springer New York; 2008. p. 278–81. Available from: http://link.springer.com/10.1007/ 978-0-387-32833-1_211. PLOS Biology | https://doi.org/10.1371/journal.pbio.3002453 January 5, 2024 34 / 34 PLOS BIOLOGY
null
10.1186_s12874-019-0884-8.pdf
Availability of data and materials Restrictions by the data custodians mean that the datasets are not publicly available or able to be provided by the authors. Researchers wanting to access the datasets used in this study should refer to the Centre for Health Record Linkage application process (www.cherel.org.au/apply-for-linked-data).
null
Tervonen et al. BMC Medical Research Methodology (2019) 19:245 https://doi.org/10.1186/s12874-019-0884-8 R E S E A R C H A R T I C L E Open Access Using data linkage to enhance the reporting of cancer outcomes of Aboriginal and Torres Strait Islander people in NSW, Australia Hanna E. Tervonen, Stuart Purdie and Nicola Creighton* Abstract Background: Aboriginal people are known to be under-recorded in routinely collected datasets in Australia. This study examined methods for enhancing the reporting of cancer incidence among Aboriginal people using linked data methodologies. Methods: Invasive cancers diagnosed in New South Wales (NSW), Australia, in 2010–2014 were identified from the NSW Cancer Registry (NSWCR). The NSWCR data were linked to the NSW Admitted Patient Data Collection, the NSW Emergency Department Data Collection and the Australian Coordinating Register Cause of Death Unit Record File. The following methods for enhancing the identification of Aboriginal people were used: ‘ever-reported’, ‘reported on most recent record’, ‘weight of evidence’ and ‘multi-stage median’. The impact of these methods on the number of cancer cases and age-standardised cancer incidence rates (ASR) among Aboriginal people was explored. Results: Of the 204,948 cases of invasive cancer, 2703 (1.3%) were recorded as Aboriginal on the NSWCR. This increased with enhancement methods to 4184 (2.0%, ‘ever’), 3257 (1.6%, ‘most recent’), 3580 (1.7%, ‘weight of evidence’) and 3583 (1.7%, ‘multi-stage median’). Enhancement was generally greater in relative terms for males, people aged 25–34 years, people with cancers of localised or unknown degree of spread, people living in urban areas and areas with less socio-economic disadvantage. All enhancement methods increased ASRs for Aboriginal people. The weight of evidence method increased the overall ASR by 42% for males (894.1 per 100,000, 95% CI 844.5–945.4) and 27% for females (642.7 per 100,000, 95% CI 607.9–678.7). Greatest relative increases were observed for melanoma and prostate cancer incidence (126 and 63%, respectively). ASRs for prostate and breast cancer increased from below to above the ASRs of non-Aboriginal people with enhancement of Aboriginal status. Conclusions: All data linkage methods increased the number of cancer cases and ASRs for Aboriginal people. Enhancement varied by demographic and cancer characteristics. We considered the weight of evidence method to be most suitable for population-level reporting of cancer incidence among Aboriginal people. The impact of enhancement on disparities in cancer outcomes between Aboriginal and non-Aboriginal people should be further examined. Keywords: Neoplasms, Indigenous, Australia, Data linkage * Correspondence: nicola.creighton@health.nsw.gov.au Cancer Institute NSW, PO Box 41, Alexandria, Sydney, NSW 1435, Australia © The Author(s). 2019 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated. Tervonen et al. BMC Medical Research Methodology (2019) 19:245 Page 2 of 9 Background Aboriginal people are known to be under-recorded in routinely collected datasets [1–3]. Reasons for under- recording are complex and include a lack of awareness and training to ask about Aboriginal status among health staff, and among Aboriginal people concerns about how the question was asked, racism and discrimination, priv- acy, a lack of cultural safety and difficulties in tracing identity [4]. Under-recording of Aboriginal status gener- ally results in under-estimation of absolute measures of health indicators [5, 6]. It is possible to enhance reporting of health outcomes of Aboriginal people by linking data from several sources [7]. For example, Randall and colleagues showed that different enhancement methods using linked data in- creased the number of hospital admissions for Aborigi- nal people with varying impacts on admission and mortality ratios [6]. Several different methods for enhan- cing identification of Aboriginal people have been used, with no consensus on the optimal method. Australian guidelines on data linkage related to Aboriginal people recommend comparing the impact of several methods and choosing the optimal method based on the purpose of the analysis and characteristics of the datasets [7]. Aboriginal people are under-recorded in the New South Wales Cancer Registry (NSWCR) despite increased record- ing of Aboriginal status over time [3]. In the early 1980s, more than 80% of people on the NSWCR had unknown Aboriginal status, which had dropped to approximately 13% by 1999. A previous study examining the feasibility of en- hancement of reporting of Aboriginal people using linked data from several data sources, including NSWCR, found that the number of cancer cases, and hence cancer incidence, for Aboriginal people increased following enhancement [2]. Estimates of health outcomes among Aboriginal people and the size of disparities compared with non-Aboriginal people can change depending on how Aboriginal status is reported and which enhancement method is used [5, 6]. Ac- curate and complete recording of Indigenous status is needed to reliably measure cancer outcomes, identify dispar- ities and produce information about cancer among Indigenous people globally. Cancer registries are a key source of information for reporting cancer outcomes yet there are very few studies examining the impact of under- recording of Indigenous status on cancer incidence [8]. This study examined the impact of linked data enhancement methods on the number of cancer cases and cancer inci- dence rates among Aboriginal people in NSW, Australia, using common algorithms and population-based datasets. Methods Study design and data sources This was a retrospective cohort study using linked invasive routinely-collected health data. All cases of cancer diagnosed and recorded in the NSWCR between 2010 and 2014 were included in the analyses. The NSWCR is a statutory population-based cancer registry which collects information about all invasive cancers di- agnosed in NSW, Australia. Information about Aborigi- nal and Torres Strait Islander status in the NSWCR comes from multiple sources, such as hospital treatment episodes and death registration [3]. Pathology reports do not include information about Aboriginal and Torres Strait Islander status and, therefore, this information is missing if the NSWCR only receives a pathology notifi- cation. The NSWCR uses a progressive positive identifi- cation algorithm with a single notice from any source indicating a person to be Aboriginal or Torres Strait Is- lander taking precedence over any other information. Aboriginal and Torres Strait Islander status is assigned at a person level, rather than individual cancer case level. Torres Strait Islander people are included with Aborigi- nal people throughout this study due to the small num- ber of people from the Torres Strait Islands residing in NSW and in recognition that Aboriginal people are the original inhabitants of NSW [4]. The NSWCR data were linked to the NSW Admitted Pa- tient Data Collection (APDC), the NSW Emergency De- partment Data Collection (EDDC) and the Australian Coordinating Registry Cause of Death Unit Record File (COD URF). The APDC includes records of all hospital ad- missions in NSW public and private hospitals and day pro- information on the EDDC includes cedure centres, presentations to emergency departments of public hospitals in NSW, and the COD URF includes information about deaths occurring in NSW. Data linkage was performed by the Centre for Health Record Linkage (CHeReL). The CHeReL uses Choicemaker software to perform probabilis- tic linkage of personal identifiers using a privacy-preserving protocol (http://www.cherel.org.au). The datasets used in this study are in the CHeReL’s Master Linkage Key. The CHeReL implements quality assurance procedures and per- forms clerical review of a sample of records to keep the es- timated false positive and false negative linkage rate to less than 5 per 1000. The CHeReL provided a unique and arbi- trary “Project Person Number” which enabled the records in each study dataset to be joined for an individual without the researchers accessing personal identifiers. The APDC data covered a period between July 2001 and December 2017, the EDDC between January 2005 and De- cember 2017, and the COD URF between January 1985 and December 2015. Aboriginal status is self-reported in the APDC and EDDC and is provided by the next-of-kin in the COD URF. Population data were based on data from the Australian Bureau of Statistics and obtained through the Se- cure Analytics for Population Health Research and Intelligence (SAPHaRI) data warehouse (Centre for Epi- demiology and Evidence, NSW Ministry of Health). Tervonen et al. BMC Medical Research Methodology (2019) 19:245 Page 3 of 9 This project was approved by the NSW Population and Health Services Research Ethics Committee (HREC/ 15/CIPHS/15) and the Aboriginal Health and Medical Research Council Ethics Committee (HREC Ref. No. 1201/16). Subject matter advice and Aboriginal commu- nity input was sought from the Cancer Institute NSW Aboriginal Advisory Group. Enhancement methods The following methods for enhancing the reporting of ‘ever re- cancer among Aboriginal people were used: ported as Aboriginal’ [7], ‘Aboriginal on most recent rec- ord’ [7], ‘weight of evidence’ [2] and ‘multi-stage median’ [9] (Table 1). These methods were selected because they are among the most commonly used methods, represent a combination of simple and complex enhancement methods and are likely to provide a range of estimates. If a person was recorded as Aboriginal on the NSWCR or on the COD URF, a person was considered to be Abori- ginal in the analyses. Our aim was to correct for under- recording of Aboriginal people in the NSWCR, so we only considered changing the status of those recorded as non-Aboriginal or with unknown status in the NSWCR. We considered the risk of a person being wrongly in the COD URF to be low identified as Aboriginal since the information is provided by the next-of-kin. Otherwise the four enhancement methods were ap- plied to the data according to the descriptions pro- vided in Table 1. Statistical analysis The number, proportion and characteristics of cases re- ported as Aboriginal using the NSWCR information and the four enhancement methods were compared. Character- istics considered in this study were: sex, age at diagnosis, year of diagnosis, cancer site, degree of spread (localised, re- remoteness gional, distant, unknown) (major cities, inner regional, outer regional, remote/very re- mote) [11], and area-based socio-economic disadvantage residential [10], Table 1 The enhancement methods used in the analyses Method Description Ever reported [7] Recorded as being Aboriginal at least once in any of the data sources. Most recent record [7] Weight of evidence [2] Multi-stage median [9] Recorded as being Aboriginal in the most recent record in any of the data sources. Recorded as Aboriginal if 1) there are three or more units of information and at least two indicate that the person is Aboriginal; 2) if there are one or 2 units of information and at least one identifies the person as Aboriginal. The weight of evidence method is applied in a two- step process: firstly to each dataset individually; and then treating the results for each dataset as units of information. (Index of Relative Socio-economic Disadvantage quintiles) [12]. For descriptive analyses, cancer sites were classified using clinical cancer grouping [13]. considered Age-standardised cancer incidence rates (ASR) were calculated for non-Aboriginal and Aboriginal people using the NSWCR Aboriginal status variable before enhancement. Cases with unknown Aboriginal status For Aboriginal non-Aboriginal. were people, cancer incidence was also calculated using the variables created by the four enhancement methods. Direct age-standardisation was calculated using the 2001 Australian standard population and NSW popu- lation data based on data from the Australian Bureau of Statistics [14]. Results were reported as rates per 100,000 with 95% confidence intervals (CIs) for all cancers and for the following sites: (female) breast (International Statistical Classification of Diseases and Related Health Problems, 10th Revision, Australian Modification code C50), colorectal (C18-C20), pros- tate (C61), lung (C34), melanoma (C43), and cervical cancer (C53). The impact of different enhancement methods on the number of cases and on ASRs was examined in relative terms (% increase compared with the NSWCR variable). Analyses were performed using SAS Version 9.4 (SAS Institute, Cary, NC). Results invasive cancer were diag- Overall 204,948 cases of nosed in NSW in 2010–2014. Of these, 2703 (1.3%) were diagnosed among Aboriginal people based on the NSWCR Aboriginal status variable. There were 28,572 cases of cancer with unknown Aboriginal sta- tus (13.9%). After enhancement, the number of cases among Aboriginal people increased to 4184 (2.0%, ‘ever’), 3257 (1.6%, ‘weight ‘multi-stage median’). of evidence’) and 3583 (1.7%, The majority of cancer cases with a status change after enhancement were originally recorded as non- Aboriginal, rather than unknown Aboriginal status. For example, of the 877 cases of cancer with a status enhanced to Aboriginal using the weight of evidence method, 74% (n = 651) were recorded as non- Aboriginal and 26% (n = 226) had unknown Aborigi- nal status on the NSWCR. ‘most recent’), 3580 (1.7%, Relative enhancement (per cent increase) was generally greater for males, people aged 25–34 years, people with cancers of unknown or localised degree of spread, people living in urban areas and areas with less socio- economic disadvantage (Table 2). Overall the ASR among Aboriginal people was 559.9 per 100,000 (95% CI 535.3–585.3) before enhancement. All enhancement methods increased ASRs overall and for both males and females (Table 3, Fig. 1). The greatest Tervonen et al. BMC Medical Research Methodology (2019) 19:245 Page 4 of 9 Table 2 Impact of enhancement on the number of cancer cases and relative increase (%) among Aboriginal people by demographic and cancer characteristics, 2010–2014 NSWCR a Ever reported Most recent record Weight of evidence Multi-stage median n % n % Increase (%) b n % Increase (%) b n % Increase (%) b n % Increase (%) b Sex Female Male Age at diagnosis 0–14 15–24 25–34 35–44 45–54 55–64 65–74 75–84 85+ Year of diagnosis 2010 2011 2012 2013 2014 Clinical cancer group Skin Head and neck Upper gastrointestinal Colorectal Respiratory Bone and connective tissue Breast Gynaecological Urogenital Eye and neurological Thyroid and other endocrine Lymphohaematopoietic Ill-defined and unknown primary sites Degree of spread Localised Regional Distant Unknown Remoteness Major Cities Inner Regional 1329 1.5 1920 2.1 44.5% 1575 1.7 18.5% 1689 1.9 27.1% 1701 1.9 28.0% 1374 1.2 2264 2.0 64.8% 1682 1.5 22.4% 1891 1.7 37.6% 1882 1.6 37.0% 49 47 86 224 521 714 660 332 70 497 532 535 569 570 91 145 340 299 430 25 314 179 416 45 64 258 97 819 668 661 555 4.4 3.0 2.0 2.2 2.2 1.6 1.2 0.8 0.4 1.3 1.3 1.3 1.4 1.4 0.4 2.6 2.1 1.2 2.2 1.7 1.2 2.1 0.9 1.4 1.2 1.2 1.9 1.0 1.5 2.2 1.1 61 68 143 322 727 5.5 4.4 3.4 3.2 3.0 24.5% 44.7% 66.3% 43.8% 39.5% 1076 2.4 50.7% 1058 1.9 60.3% 570 159 766 830 836 851 901 284 205 442 442 553 34 472 249 803 68 109 401 122 1.3 0.8 1.9 2.1 2.0 2.0 2.1 1.3 3.6 2.7 1.8 2.8 2.3 1.9 2.9 1.8 2.2 2.1 1.9 2.3 71.7% 127.1% 54.1% 56.0% 56.3% 49.6% 58.1% 212.1% 41.4% 30.0% 47.8% 28.6% 36.0% 50.3% 39.1% 93.0% 51.1% 70.3% 55.4% 25.8% 54 61 120 276 627 867 785 388 79 594 639 646 683 695 177 171 367 356 464 29 391 207 547 51 81 313 103 4.9 3.9 2.8 2.8 2.6 1.9 1.4 0.9 0.4 1.5 1.6 1.6 1.6 1.7 0.8 3.0 2.3 1.4 2.4 2.0 1.5 2.4 1.2 1.6 1.5 1.5 2.0 10.2% 29.8% 39.5% 23.2% 20.3% 21.4% 18.9% 16.9% 12.9% 19.5% 20.1% 20.7% 20.0% 21.9% 94.5% 17.9% 7.9% 19.1% 7.9% 16.0% 24.5% 15.6% 31.5% 13.3% 26.6% 21.3% 6.2% 57 62 128 289 661 949 878 451 105 664 704 710 745 757 212 186 394 385 508 32 413 222 627 56 93 344 108 5.1 4.0 3.0 2.9 2.8 2.1 1.6 1.0 0.6 1.7 1.7 1.7 1.8 1.8 1.0 3.3 2.4 1.5 2.6 2.2 1.6 2.6 1.4 1.8 1.8 1.6 2.1 16.3% 31.9% 48.8% 29.0% 26.9% 32.9% 33.0% 35.8% 50.0% 33.6% 32.3% 32.7% 30.9% 32.8% 133.0% 28.3% 15.9% 28.8% 18.1% 28.0% 31.5% 24.0% 50.7% 24.4% 45.3% 33.3% 11.3% 56 64 130 293 675 945 876 447 97 661 705 712 747 758 216 184 392 379 502 33 420 223 633 57 94 343 107 5.1 4.1 3.0 2.9 2.8 2.1 1.6 1.0 0.5 1.7 1.7 1.7 1.8 1.8 1.0 3.2 2.4 1.5 2.6 2.2 1.7 2.6 1.4 1.8 1.8 1.6 2.1 14.3% 36.2% 51.2% 30.8% 29.6% 32.4% 32.7% 34.6% 38.6% 33.0% 32.5% 33.1% 31.3% 33.0% 137.4% 26.9% 15.3% 26.8% 16.7% 32.0% 33.8% 24.6% 52.2% 26.7% 46.9% 32.9% 10.3% 1467 1.8 79.1% 1086 1.3 32.6% 1209 1.5 47.6% 1212 1.5 48.0% 964 783 970 2.2 2.6 2.0 44.3% 18.5% 74.8% 767 686 718 1.8 2.3 1.5 14.8% 3.8% 29.4% 857 713 801 2.0 2.3 1.6 28.3% 7.9% 44.3% 850 713 808 2.0 2.3 1.7 27.2% 7.9% 45.6% 1206 0.9 1998 1.4 65.7% 1472 1.1 22.1% 1639 1.2 35.9% 1633 1.2 35.4% 831 1.7 1277 2.6 53.7% 1001 2.0 20.5% 1098 2.2 32.1% 1109 2.2 33.5% Tervonen et al. BMC Medical Research Methodology (2019) 19:245 Page 5 of 9 Table 2 Impact of enhancement on the number of cancer cases and relative increase (%) among Aboriginal people by demographic and cancer characteristics, 2010–2014 (Continued) NSWCR a Ever reported Most recent record Weight of evidence Multi-stage median n % n % Increase (%) b n % Increase (%) b n % Increase (%) b n % Increase (%) b Outer Regional 530 3.4 738 4.7 39.2% Remote/ very remote 136 Socio-economic disadvantage quintilec 12.7 171 15.9 25.7% 620 164 4.0 17.0% 15.3 20.6% 678 165 4.3 27.9% 15.4 21.3% 675 166 4.3 27.4% 15.5 22.1% Q1: Least disadvantaged Q2 Q3 Q4 Q5: Most disadvantaged 131 352 474 843 903 0.3 0.9 1.1 1.8 2.4 280 596 785 0.7 1.6 1.9 113.7% 69.3% 65.6% 157 441 580 0.4 1.1 1.4 19.8% 25.3% 22.4% 194 487 650 0.5 1.3 1.6 48.1% 38.4% 37.1% 195 485 654 0.5 1.3 1.6 48.9% 37.8% 38.0% 1219 2.6 44.6% 1004 2.2 19.1% 1082 2.3 28.4% 1087 2.3 28.9% 1304 3.4 44.4% 1075 2.8 19.0% 1167 3.1 29.2% 1162 3.1 28.7% aNSWCR: Aboriginal status variable in the NSW Cancer Registry bRelative increase compared with the number of cases based on the NSW Cancer Registry Aboriginal status variable cIndex of Relative Socio-economic Disadvantage increases were detected when using the ‘ever reported’ and the smallest increases when using the ‘most recent’ method. Enhancement increased incidence rates more for males than females. For example, the ‘weight of evi- dence’ method increased the ASR by 42% for males (894.1 per 100,000, 95% CI 844.5–945.4) and 27% for fe- males (642.7 per 100,000, 95% CI 607.9–678.7). In site-specific analyses, all enhancement methods in- creased ASRs for all sites compared with rates estimated using the NSWCR Aboriginal status variable (Table 3, Fig. 2). Again, the ‘ever reported’ method demonstrated the greatest increases while the ‘most recent’ method re- sulted in the smallest increases. Greatest relative in- creases were observed for melanoma and prostate cancer incidence, with increases of 126 and 63% respect- ively, using the ‘weight of evidence’ method. Discussion All enhancement methods increased both the number of cancer cases and age-standardised cancer incidence rates among Aboriginal people. The ‘ever reported’ method dem- onstrated the greatest increases and ‘most recent’ method the smallest increases, while the other two methods were very similar to each other and between these two extrem- ities. When using the ‘weight of evidence’ method, the ma- jority (74%) of cases with enhanced Aboriginal status were previously recorded as non-Aboriginal on the NSWCR. This indicates misclassification in the NSWCR Aboriginal status variable and highlights the need to correct this mis- classification and not solely focus on decreasing the num- ber of people with unknown Aboriginal status in the NSWCR and in the information received by the NSWCR from notifiers. Aboriginal and Torres Strait Islander status is self-reported at NSW health facilities and people to identify [4]. There have been may choose not culturally to provide strengthened procedures at a state level to improve the collection of Aboriginal and Torres Strait Islander status in NSW health facilities [15] as well as local initiatives safe health care throughout the study period. These factors are likely to have increased the willingness of people to self- identify as Aboriginal or Torres Strait Islander and improved identification at the point of care in more recent years. Linked data enhances the reporting of Aboriginal status because it brings together informa- tion on Aboriginal status that is not available to the NSWCR through people choosing to identify as Abo- riginal after diagnosis or at facilities that have not provided cancer care. Enhancement was generally greater in relative terms for males, people aged 25–34 years at diagnosis, people living in urban and less disadvantaged areas and for people with a cancer of localised or unknown degree of spread. Several factors are likely to explain these pat- terns, such as sources of cancer notifications and treat- ment patterns (e.g. the likelihood of admission for surgery). People diagnosed with cancers with good prog- nosis are less likely to be hospitalised or die which de- creases the likelihood of recording the Aboriginal status on the NSWCR. If the NSWCR only receives pathology notification, Aboriginal status information will be miss- ing. This is more likely to apply to cancers such as mela- nomas and prostate cancers, both of which showed greater levels of enhancement. A previous NSW study reported that enhancing Abori- ginal status for reporting deaths resulted in greater en- hancements for older people, for people living in urban areas and for those with chronic health conditions [16]. Another NSW study examining the im- pact on enhancement on hospital admissions reported for females, Tervonen et al. BMC Medical Research Methodology (2019) 19:245 Page 6 of 9 ) % ( e s a e r c n I b . 6 3 3 . 9 7 2 . 1 9 3 . 6 1 3 . 4 6 2 . 7 7 3 . 9 7 1 . 4 5 1 . 6 0 2 . 4 0 2 1 . 5 6 3 1 . 7 4 0 1 . 1 5 3 . 0 5 1 . 5 1 6 . – 4 9 1 7 ( . – 4 0 1 6 ( . – 3 7 2 8 ( . 0 8 4 7 . ) 4 7 7 7 . 2 5 4 6 . ) 2 1 8 6 . 6 5 7 8 . ) 7 5 2 9 ) I C % 5 9 ( R S A ) % ( . ) 9 4 9 – 1 4 7 ( . . ) 6 0 9 – 2 4 6 ( . . ) 6 0 1 1 – 6 6 7 ( . . 1 4 8 . 7 6 7 . 7 2 9 . – 4 1 0 1 ( . 5 2 1 1 . ) 4 4 2 1 . ) 6 1 1 1 – 3 3 8 ( . . 8 6 9 . – 5 4 1 1 ( . 3 3 3 1 . ) 0 4 5 1 . ) 7 3 4 – 7 0 3 ( . . ) 4 5 3 – 3 0 2 ( . . ) 6 9 5 – 5 7 3 ( . . 8 6 3 . 2 7 2 . 7 7 4 . – 2 7 2 1 ( . 1 2 4 1 . ) 1 8 5 1 . ) 3 2 2 – 5 2 1 ( . . 9 6 1 . – 6 0 8 1 ( . 6 3 0 2 . ) 3 8 2 2 e s a e r c n I b . 6 4 3 . 4 7 2 . 1 2 4 . 1 5 3 . 5 5 2 . 1 8 4 . 1 0 2 . 2 6 1 . 3 4 2 . 3 6 2 1 . 4 7 1 1 . 6 9 2 1 . 1 3 3 . 6 3 1 . 7 2 6 ) I C % 5 9 ( R S A . – 8 4 2 7 ( . – 9 7 0 6 ( . – 5 4 4 8 ( . 7 3 5 7 . ) 4 3 8 7 . 7 2 4 6 . ) 7 8 7 6 . 1 4 9 8 . ) 4 5 4 9 . ) 4 7 9 – 1 6 7 ( . . ) 0 0 9 – 9 3 6 ( . . ) 0 9 1 1 – 2 2 8 ( . . 3 6 8 . 2 6 7 . 7 9 9 . – 3 3 0 1 ( . 6 4 1 1 . ) 7 6 2 1 . ) 4 2 1 1 – 0 4 8 ( . . 5 7 9 . – 1 8 1 1 ( . 4 7 3 1 . ) 7 8 5 1 . ) 9 4 4 – 4 1 3 ( . . ) 9 2 3 – 4 8 1 ( . . ) 1 7 6 – 7 1 4 ( . . 8 7 3 . 0 5 2 . 5 3 5 . – 2 5 2 1 ( . 0 0 4 1 . ) 0 6 5 1 . ) 1 2 2 – 3 2 1 ( . . 7 6 1 . – 0 2 8 1 ( . 2 5 0 2 . ) 1 0 3 2 ) % ( e s a e r c n I b . 3 9 1 . 6 6 1 . 8 1 2 . 6 1 2 . 1 6 1 . 2 8 2 2 9 . 0 7 . . 6 1 1 . 1 0 7 . 2 2 9 . 9 4 5 . 4 4 2 . 9 0 1 . 0 7 3 ) I C % 5 9 ( R S A . – 1 1 4 6 ( . – 7 5 5 5 ( . – 0 2 2 7 ( . 9 7 6 6 . ) 3 5 9 6 . 5 8 8 5 . ) 6 2 2 6 . 7 6 6 7 . ) 0 3 1 8 . ) 0 8 8 – 1 8 6 ( . . ) 7 3 8 – 7 8 5 ( . . ) 5 3 0 1 – 8 0 7 ( . . 7 7 7 . 5 0 7 . 3 6 8 . ) 7 5 1 1 – 4 3 9 ( . . 2 4 0 1 . ) 1 4 0 1 – 8 6 7 ( . . 8 9 8 . – 1 5 0 1 ( . 3 3 2 1 . ) 4 3 4 1 . ) 2 4 3 – 2 3 2 ( . . ) 0 9 2 – 3 6 1 ( . . ) 7 6 4 – 2 7 2 ( . . 4 8 2 . 1 2 2 . 1 6 3 . – 8 6 1 1 ( . 9 0 3 1 . ) 1 6 4 1 . ) 6 1 2 – 0 2 1 ( . . 3 6 1 . – 8 1 5 1 ( . 7 2 7 1 . ) 3 5 9 1 ) % ( e s a e r c n I b . 0 4 6 . 7 8 4 . 4 0 8 . 5 3 6 . 8 3 4 . 7 9 8 . 0 2 3 . 7 3 2 . 0 1 4 . 1 3 4 2 . 0 3 1 2 . 8 4 6 2 . 1 3 5 . 0 2 3 . 5 3 2 1 ) I C % 5 9 ( R S A ) I C % 5 9 ( R S A ) I C % 5 9 ( R S A . ) 6 1 5 9 – 2 5 8 8 ( . . ) 1 0 9 7 – 2 2 1 7 ( . . 0 8 1 9 . – 3 5 3 5 ( . 5 0 5 7 . – 0 4 7 4 ( . - 0 7 7 0 1 ( . 3 5 3 1 1 . ) 5 5 9 1 1 . – 7 8 8 5 ( . 9 9 5 5 . ) 3 5 8 5 . 6 4 0 5 . ) 5 6 3 5 . 4 9 2 6 . ) 7 1 7 6 . ) 0 3 9 4 – 7 8 8 4 ( . . ) 5 0 2 4 – 0 5 1 4 ( . . ) 0 1 8 5 – 2 4 7 5 ( . . 8 0 9 4 s n o s r e P s r e c n a c l l A . 7 7 1 4 l s e a m e F . 6 7 7 5 l s e a M . ) 0 7 1 1 – 9 2 9 ( . . 5 4 0 1 . ) 2 3 7 – 3 5 5 ( . . ) 3 2 0 1 – 7 3 7 ( . . 3 7 8 . ) 0 3 7 – 8 9 4 ( . . ) 4 0 5 1 – 1 7 0 1 ( . . ) 6 8 3 1 – 9 3 1 1 ( . . ) 2 9 1 1 – 8 9 8 ( . . ) 7 8 7 1 – 7 4 3 1 ( . . 7 7 2 1 . 9 5 2 1 . 8 3 0 1 . 8 5 5 1 . ) 5 6 6 – 9 8 4 ( . . ) 6 5 4 – 7 7 2 ( . . ) 5 3 0 1 – 7 8 6 ( . . 3 7 5 . 0 6 3 . 0 5 8 . ) 4 2 8 – 9 3 5 ( . . ) 3 6 0 1 – 2 5 8 ( . . ) 8 7 9 – 5 1 7 ( . . 9 3 6 . 7 0 6 . 3 7 6 . 4 5 9 . 9 3 8 . ) 6 8 5 – 1 7 5 ( . . ) 8 9 4 – 9 7 4 ( . . ) 2 9 6 – 9 6 6 ( . . ) 1 3 4 – 8 1 4 ( . . ) 1 4 3 – 5 2 3 ( . . ) 2 9 2 1 – 6 3 9 ( . . 5 0 1 1 . ) 7 4 5 – 6 2 5 ( . . ) 6 1 2 – 6 2 1 ( . . ) 2 7 1 – 1 7 ( . . ) 6 2 3 – 8 5 1 ( . . 7 6 1 . 5 1 1 . 3 3 2 . ) 5 1 5 – 1 0 5 ( . . ) 2 1 4 – 4 9 3 ( . . ) 6 4 6 – 3 2 6 ( . . 3 3 3 . 6 3 5 . 8 0 5 . 3 0 4 . 4 3 6 . ) 2 8 7 1 – 1 5 4 1 ( . . 1 1 6 1 . ) 9 8 1 1 – 6 2 9 ( . . 2 5 0 1 . ) 3 5 2 – 5 4 1 ( . . 4 9 1 . ) 9 9 1 – 5 0 1 ( . . 7 4 1 . ) 3 2 2 1 – 2 9 1 1 ( . . 8 0 2 1 . ) 2 7 – 4 6 ( . 8 6 . . ) 8 1 1 3 – 5 3 5 2 ( . . 8 1 8 2 . – 9 7 0 1 ( . 1 6 2 1 . ) 1 6 4 1 . ) 5 9 6 1 – 9 5 6 1 ( . . 7 7 6 1 . 8 7 5 . 9 8 4 . 1 8 6 s n o s r e P l s e a m e F l s e a M l a t c e r o o C l g n u L . 5 2 4 s n o s r e P l s e a m e F l s e a M s n o s r e P l s e a m e F l s e a M a m o n a e M l l ) s e a m e F ( t s a e r B e t a t s o r P i x v r e C i n a d e m e g a t s - i t l u M e c n e d v e i f o t h g e W i d r o c e r t n e c e r t s o M d e t r o p e r r e v E a R C W S N 4 1 0 2 – 0 1 0 2 , l e p o e p l i a n g i r o b A d n a l i a n g i r o b A - n o n g n o m a s e t a r e c n e d c n i i l e p o e p l i a n g i r o b A l i a n g i r o b A - n o N a R C W S N l e p o e p r e c n a c d e s i d r a d n a t s - e g A 3 e l b a T l n o i t a u p o p d r a d n a t s n a i l a r t s u A 1 0 0 2 e h t o t d e s i d r a d n a t s y l t c e r i d ; s l a v r e t n i e c n e d i f n o c % 5 9 h t i w 0 0 0 0 0 1 , r e p e t a r e c n e d i c n i r e c n a c d e s i d r a d n a t s - e g A : ) I C % 5 9 ( R S A l e b a i r a v s u t a t s l i a n g i r o b A y r t s i g e R r e c n a C W S N e h t n o d e s a b e t a r e c n e d i c n i h t i w d e r a p m o c e s a e r c n i y r t s i g e R r e c n a C W S N e h t n i l e b a l i a v a l e b a i r a v s u t a t s l i a n g i r o b A : R C W S N a e v i t a e R b l Tervonen et al. BMC Medical Research Methodology (2019) 19:245 Page 7 of 9 Fig. 1 Age-standardised cancer incidence rates among Aboriginal people using the NSW Cancer Registry (NSWCR) Aboriginal status variable and four enhancement methods, 2010–2014. (see Table 3 for underlying data and 95% confidence intervals) greater enhancement for earlier years of admission, major cities, private hospitals and varying impact by age depending on the enhancement method used [6]. Differ- ent factors impact on enhancement depending on the health outcome of interest and the datasets used in analyses. Lung and cervical cancers saw the smallest increases in incidence rates. Both these cancers have a greater burden in Aboriginal compared with non-Aboriginal people [17]. Due to the poor prognosis, death certificate information is available for most people diagnosed with lung cancer, in- creasing the likelihood of Aboriginal status recording. It is likely that enhancement had a smaller impact on lung cancer incidence rates because the existing NSWCR Abo- riginal status already had relatively good capture. The rela- tively smaller increase in the incidence of cervical cancer may due to relatively good capture on the NSWCR, but may also be due to other factors such the patterns of hos- pitalisation and capture of Aboriginal status at the point of care for what is generally a younger cohort of women. Enhancing the reporting of cancer outcomes of Aborigi- nal people might have a major impact on observed dispar- ities between Aboriginal and non-Aboriginal people. For example, according to national statistics [17] and our Fig. 2 Age-standardised cancer incidence rates by site among Aboriginal people using the NSW Cancer Registry (NSWCR) Aboriginal status variable and four enhancement methods, 2010–2014. (see Table 3 for underlying data and 95% confidence intervals) Tervonen et al. BMC Medical Research Methodology (2019) 19:245 Page 8 of 9 Increased breast cancer analyses using the NSWCR Aboriginal status variable, Abo- riginal people have lower breast and prostate cancer inci- dence rates compared with non-Aboriginal people. This pattern has also been reported among Indigenous peoples in many international jurisdictions and has been proposed as being related to the prevalence of risk factors for these cancers and competing causes of death [18]. After enhance- ment our results indicated higher breast and prostate cancer incidence among Aboriginal people than non- Aboriginal people in NSW. This finding has implications on widely held views on risk of these cancers among Indi- genous peoples. Higher breast cancer rates have been re- ported among Indigenous people (Māori) in New Zealand using the national population-based cancer registry which includes links to a national health database to improve identification [18]. incidence among Indigenous people have been reported in two United States (US) states using data linkage between cancer registries and health service data [19, 20]. Our results also highlight the burden of melanoma among Aboriginal people which warrants further discussion on prevention strategies and actions. After enhancement our results indi- cated substantially higher incidence than when using the NSWCR Aboriginal status variable, but still lower rates compared with non-Aboriginal people (except when using the ‘ever reported’ method). The effect of under-recording of Indigenous status should be investigated in more juris- dictions. Cancer is the second leading cause of death and among the leading causes of burden of disease among Abo- riginal people in Australia [21]. The findings of our study highlight the impact of cancer on Aboriginal people and the need for cancer control to improve health outcomes. Cancer control programs should have a special focus on Aboriginal people considering that their cancer burden may be higher than expected. Australian cancer screening programs are already targeting Aboriginal people due to lower participation rates [17]. Future research should also examine the impact of en- hancement on other cancer outcomes, such as mortality, survival and the likelihood of being diagnosed with ad- vanced stage disease. Studies have shown that Aboriginal people are more likely to be diagnosed with advanced stage cancer than non-Aboriginal people [22, 23]. We found greatest enhancement for people diagnosed with localised or unknown degree of spread, which may impact on the likelihood of Aboriginal people being diagnosed with advanced cancer in comparison with non-Aboriginal people and affect estimates of disparities in survival out- comes since localised cancers have much better prognosis. Based on these results and consultation with the Can- cer Institute NSW Aboriginal Advisory Group, the ‘weight of evidence’ method was considered to be the most suitable for further reporting of cancer outcomes for Aboriginal people. The ‘weight of evidence’ method utilises information from several sources but is still rela- tively straightforward to use and report. It provides a balance between enhancing the identification of Aborigi- nal people and reducing misclassification of non- Aboriginal people as Aboriginal. This method was devel- oped and is also used by the NSW Ministry of Health [6]. Studies have pointed out that ‘ever reported’ may re- sult in misclassification and over-reporting [1, 6]. It should be noted that an enhanced Aboriginal identifier is a statistical construct that enables improved reporting of cancer outcomes using historical data but potentially includes some inaccuracies due to errors in the source datasets and incorrect linkages [2]. Collection of accur- ate information at the point of care remains vital. Limitations include that if a person was recorded as Abo- riginal on the NSWCR or death certificate, this information was accepted. Although there is a possibility for positive misclassification this is likely to be low since the information is provided by the next-of-kin. Numerator-denominator bias is a known issue affecting observed cancer burden in Indi- genous populations internationally because incidence and population data are derived using different data collection methodologies [8]. Population denominators can be unreli- able due to under-participation of Aboriginal people and varying propensity to identify as Aboriginal in censuses. The Australian Bureau of Statistics (ABS) estimates Aboriginal and Torres Strait Islander populations using self-reported information in the Australian Census data with adjustment for undercount using a household survey following the cen- sus [14]. An increase in the number of people self- identifying as Aboriginal or Torres Strait Islander has been observed, with people who did not self-identify in the 2011 Australian Census choosing to identify in the subsequent 2016 Census [24]. In our study, enhancement of the numer- ator is likely to reduce the under-estimation of cancer inci- dence that is common in cancer incidence estimates for Indigenous people [8]. However, without enhancement of the denominator using the same methodologies it may lead to over-estimation of incidence rates. Linkage of the cancer registry, census, hospital and mortality data would enable cancer outcomes for Aboriginal people to be estimated with reduced numerator-denominator bias. Conclusions All data linkage enhancement methods increased the number of cancer cases and cancer incidence rates for Aboriginal people. Enhancement varied by demographic and cancer characteristics. We considered the ‘weight of evidence’ method to be most suitable for future analyses of cancer outcomes of Aboriginal people. Enhancing the reporting of cancer outcomes of Aboriginal people can have major impacts on cancer disparities between Abori- ginal and non-Aboriginal people and this should be fur- ther examined. Tervonen et al. BMC Medical Research Methodology (2019) 19:245 Page 9 of 9 Abbreviations ABS: Australian Bureau of Statistics; APDC: Admitted Patient Data Collection; ASR: Age-standardised cancer incidence rate; CI: Confidence Intervals; COD URF: Cause of Death Unit Record File; EDDC: Emergency Department Data Collection; NSW: New South Wales; NSWCR: New South Wales Cancer Registry; US: United States Acknowledgements The authors would like to thank the Aboriginal Advisory Group of the Cancer Institute NSW for their valuable advice and comments. The Cause of Death Unit Record File (COD URF) is provided by the Australian Coordinating Registry for COD URF on behalf of Australian Registries of Births, Deaths and Marriages, Australian Coroners and the National Coronial Information System. We would also like to thank the Centre for Epidemiology and Evidence, NSW Ministry of Health for providing access to the population data and the Centre for Health Record Linkage for their assistance with this project. Authors’ contributions NC had the original idea for the study. SP and HET analysed the data. HET and NC conducted the literature searches. HET wrote the first draft of the manuscript. All authors contributed to the interpretation of the results, read and approved the final manuscript. Funding Not applicable. Availability of data and materials Restrictions by the data custodians mean that the datasets are not publicly available or able to be provided by the authors. Researchers wanting to access the datasets used in this study should refer to the Centre for Health Record Linkage application process (www.cherel.org.au/apply-for-linked-data). Ethics approval and consent to participate This project was approved by the NSW Population and Health Services Research Ethics Committee (HREC/15/CIPHS/15) and the Aboriginal Health and Medical Research Council Ethics Committee (HREC Ref. No. 1201/16). The data sources were collected under legislation and individual consent was not required for the use of the de-identified data in this project. Subject matter advice and Aboriginal community input was sought from the Cancer Institute NSW Aboriginal Advisory Group. Consent for publication Not applicable. Competing interests The authors declare that they have no competing interests. Received: 5 March 2019 Accepted: 5 December 2019 References 1. Kennedy B, Howell S, Breckell C. Indigenous identification in administrative data collections and the implications for reporting Indigenous health status. Technical Report no. 3. Brisbane: Health Statistics Centre, Queensland Health; 2009. Population and Public Health Division. Improved Reporting of Aboriginal and Torres Strait Islander Peoples on Population Datasets in New South Wales using Record Linkage – a Feasibility Study. Sydney: NSW Ministry of Health; 2012. Cancer Institute NSW. Cancer in NSW Aboriginal peoples: completeness and quality of Aboriginal status data on the NSW Central Cancer registry. Sydney: Cancer Institute NSW; 2012. NSW Aboriginal Affairs. Aboriginal identification: the way forward. An Aboriginal peoples’ perspective. Sydney: NSW Government; 2015. Thompson SC, Woods JA, Katzenellenbogen JM. The quality of Indigenous identification in administrative health data in Australia: insights from studies using data linkage. BMC Med Inform Decis. 2012;12:133. Randall DA, Lujic S, Leyland AH, Jorm LR. Statistical methods to enhance reporting of Aboriginal Australians in routine hospital records using data linkage affect estimates of health disparities. Aust NZ J Publ Health. 2013; 37(5):442–9. 2. 3. 4. 5. 6. 7. 8. 9. 10. Australian Institute of Health and Welfare, Australian Bureau of Statistics. National best practice guidelines for data linkage activities relating to Aboriginal and Torres Strait Islander people. AIHW Cat. No. IHW 74. Canberra: AIHW; 2012. Sarfati D, Robson B. Equitable cancer control: better data needed for indigenous people. Lancet Oncol. 2015;16(15):1442–4. Christensen D, Davis G, Draper G, Mitrou F, McKeown S, Lawrence D, et al. Evidence for the use of an algorithm in resolving inconsistent and missing Indigenous status in administrative data collections. Aust J Soc Issues. 2014; 49:423–49. Esteban D, Whelan S, Laudico A, Parkin DM. Manual for cancer registry personnel. IARC Technical Report No 10. Lyon: International Agency for Research on Cancer; 1995. 11. Australian Bureau of Statistics. 1216.0.15.003 - Australian Standard Geographical Classification (ASGC) Remoteness Area Correspondences. Canberra: ABS; 2011. 12. Australian Bureau of Statistics. 2039.0 - Information Paper: An Introduction to Socio-Economic Indexes for Areas (SEIFA), 2006. Canberra: ABS; 2008. 13. Cancer Institute NSW. Glossary. https://www.cancerinstitute.org.au/ glossary#term-Clinical-cancer-group. Accessed 14 Jun 2019. 14. Australian Bureau of Statistics. 3238.0 - Estimates and Projections, Aboriginal and Torres Strait Islander Australians, 2001 to 2026. Canberra: ABS; 2014. 15. Centre for Aboriginal Health. Aboriginal and Torres Strait Islander Origin - 16. Recording of Information of Patients and Clients. Sydney: NSW Health; 2012. https://www1.health.nsw.gov.au/PDS/pages/doc.aspx?dn=PD2012_042 . Accessed 14 June 2019. Taylor LK, Bentley J, Hunt J, Madden R, McKeown S, Brandt P, et al. Enhanced reporting of deaths among Aboriginal and Torres Strait Islander peoples using linked administrative health datasets. BMC Med Res Methodol. 2012;12(1):91. 17. Australian Institute of Health and Welfare. Cancer in Aboriginal & Torres Strait Islander people of Australia. Web report Cat no. CAN 109: AIHW; 2018. https://www.aihwgovau/reports/cancer/cancer-in-indigenous-australians/ contents/table-of-contents. Accessed 14 Jun 2019 18. Moore SP, Antoni S, Colquhoun A, Healy B, Ellison-Loschmann L, Potter JD, et al. Cancer incidence in indigenous people in Australia, New Zealand, Canada, and the USA: a comparative population-based study. Lancet Oncol. 2015;16(15):1483–92. 19. Partin MR, Rith-Najarian SJ, Slater JS, Korn JE, Cobb N, Soler JT. Improving 20. cancer incidence estimates for American Indians in Minnesota. Am J Public Health. 1999;89(11):1673–7. Foote M, Matloub J, Strickland R, Stephenson L, Vaughan-Batten H. Improving cancer incidence estimates for American Indians in Wisconsin. WMJ. 2007;106(4):196–204. 21. Australian Institute of Health and Welfare. Australian Burden of Disease Study: Impact and causes of illness and death in Aboriginal and Torres Strait Islander people 2011. Australian Burden of Disease Study series no. 6 Cat no. BOD 7. Canberra: AIHW; 2016. 22. Gibberd A, Supramaniam R, Dillon A, Armstrong BK, O'Connell DL. Are 23. Aboriginal people more likely to be diagnosed with more advanced cancer? Med J Australia. 2015;202(4):195–9. Tervonen HE, Walton R, You H, Baker D, Roder D, Currow D, et al. After accounting for competing causes of death and more advanced stage, do Aboriginal and Torres Strait Islander peoples with cancer still have worse survival? A population-based cohort study in New South Wales. BMC Cancer. 2017;17(1):398. 24. Markham D, Biddle N. Indigenous population change in the 2016 census. CAEPR census paper no. 1. Canberra: Centre for Aboriginal Economic Policy Research, Australian National University; 2016. Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
null
10.1371_journal.pone.0229895.pdf
Data Availability Statement: All relevant data are within the manuscript.
All relevant data are within the manuscript.
RESEARCH ARTICLE A simulation based difficult conversations intervention for neonatal intensive care unit nurse practitioners: A randomized controlled trial Roberta Bowen1, Kate M. Lally2,3, Francine R. Pingitore3,4, Richard Tucker1, Elisabeth C. McGowan1,3, Beatrice E. LechnerID 1,3* 1 Department of Neonatology, Women & Infants Hospital, Providence, RI, United States of America, 2 Program in Palliative Care, Care New England Health System, Providence, RI, United States of America, 3 Warren Alpert Medical School of Brown University, Providence, RI, United States of America, 4 Department of Pediatrics, Hasbro Children’s Hospital, Providence, RI, United States of America * blechner@wihri.org Abstract Background Neonatal nurse practitioners are often the front line providers in discussing unexpected news with parents. This study seeks to evaluate whether a simulation based Difficult Con- versations Workshop for neonatal nurse practitioners leads to improved skills in conducting difficult conversations. Methods We performed a randomized controlled study of a simulation based Difficult Conversa- tions Workshop for neonatal nurse practitioners (n = 13) in a regional level IV neonatal intensive care unit to test the hypothesis that this intervention would improve communica- tion skills. A simulated test conversation was performed after the workshop by the inter- vention group and before the workshop by the control group. Two independent blinded content experts scored each conversation using a quantitative communication skills per- formance checklist and by assigning an empathy score. Standard statistical analysis was performed. Results Randomization occurred as follows: n = 5 to the intervention group, n = 7 to the control group. All participants were analyzed in each group. Participation in the simulation based Difficult Conversations Workshop increases participants’ empathy score (p = 0.015) and the use of communication skills (p = 0.013) in a simulated clinical encounter. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Bowen R, Lally KM, Pingitore FR, Tucker R, McGowan EC, Lechner BE (2020) A simulation based difficult conversations intervention for neonatal intensive care unit nurse practitioners: A randomized controlled trial. PLoS ONE 15(3): e0229895. https://doi.org/10.1371/journal. pone.0229895 Editor: Karen-Leigh Edward, Swinburne University of Technology, AUSTRALIA Received: December 6, 2019 Accepted: February 16, 2020 Published: March 9, 2020 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0229895 Copyright: © 2020 Bowen et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the manuscript. PLOS ONE | https://doi.org/10.1371/journal.pone.0229895 March 9, 2020 1 / 12 PLOS ONE Funding: The authors received no specific funding for this work. Competing interests: The authors have declared that no competing interests exist. Conclusions Our study demonstrates that a lecture and simulation based Difficult Conversations Work- shop for neonatal nurse practitioners improves objective communication skills and empathy in conducting difficult conversations. Difficult conversations simulation in the NICU Introduction The ability to communicate effectively with patients’ families is an essential skill for those caring for infants in the neonatal intensive care unit (NICU). Delivering bad news is a skill set not typi- cally taught in the formal education of advanced practice registered nurses. In the United States, these advanced practice registered nurses, or nurse practitioners (NPs), provide care alongside physicians, often in a role similar to the physician’s role and sometimes in lieu of the physician. In the NICU, neonatal NPs diagnose and treat infants, perform procedures and interact with and provide support to parents. Thus, acquisition of skills for leading difficult conversations is essential for nurse practitioners to be successful in their full scope of practice. Conducting research on communication skills training in the clinical setting is challenging and the current status of the field does not allow for the identification of one gold standard [1, 2]. Even fewer studies exist in the context of neonatology. Neonatal NPs feel that their education is lacking in this key component of practice [3], and studies of NICU communication skills did not include NPs in the assessment [4] or only measured NPs’ self-reported and thus subjective outcomes [5]. The complicated communication task of delivering bad news to the parents of infants is fraught with discomfort and uncertainty for the practitioner delivering the news [6], especially given that bad news around the birth of an infant is not in line with parental expectations. Most clinicians rely on skills demonstrated by mentors or those learned by trial and error, despite the fact that taking part in a formal program to enhance communication skills leads to an improvement in communication skills [7, 8], while studies have demonstrated that patients desire good communication [9] and that communication skills can be taught and retained [10]. Parents of infants in the NICU are at very high risk for adverse mental health outcomes [11]. Thus, communication approaches used by the medical team, including NPs, gain utmost importance. When working in level 1 and 2 community hospital nurseries, neonatal nurse practitioners are often the front line providers in discussing unexpected news with parents. Thus, we sought to evaluate the hypothesis that a lecture and simulation based Difficult Con- versations Workshop for the neonatal nurse practitioners will increase skill in conducting dif- ficult conversations with patients’ families. Materials and methods We performed a randomized controlled prospective study of a simulation based Difficult Con- versations Workshop for NICU nurse practitioner staff at a large regional level IV NICU in the Northeast of the United States. The research related to human use has been approved by the Women & Infants Hospital Institutional Review Board. Written informed consent was obtained. In this 80 bed level 3 NICU, a simulation based Difficult Conversations Workshop is part of the training program for the neonatal-perinatal medicine fellows. Participants The clinical NICU nurse practitioner group consists of 31 nurse practitioners, who work in a level IV regional NICU as well as multiple level II community hospital NICUs. All NPs were PLOS ONE | https://doi.org/10.1371/journal.pone.0229895 March 9, 2020 2 / 12 PLOS ONE Difficult conversations simulation in the NICU invited to participate in the study. The NPs were recruited to participate in the study using email as well as a presentation of the study by one of the study authors at a monthly NP staff meeting. Recruitment and workshop were performed from May 2016 to July 2016. Each three hour session of the simulation based Difficult Conversations Workshop consisted of 4–6 par- ticipants. Participants in each session were randomized using the web-based randomization tool Randomizer.org to either the intervention or control group. Simple randomization was performed with a randomization allocation of 1:1. Randomization was performed at the begin- ning of the workshop. Both groups participated in a three hour workshop. Study structure (Fig 1) Prior to randomization, all study participants (intervention group and control group) filled out an anonymous pre-workshop survey. Then, after randomization, the control group per- formed the Test Scenario, which was a standardized clinically relevant simulation scenario using trained improvisational actors as parents. They then took part in the simulation based Difficult Conversations Workshop so as to allow them the opportunity to benefit from the learning opportunity. The intervention group, on the other hand, took part in the simulation based Difficult Conversations Workshop prior to performing the Test Scenario. At the end of the Workshop and Test Scenarios, all participants filled out a post-workshop survey. Data col- lection on the pre- and post-workshop surveys ascertained demographics, past experiences with communication skills training, past experiences leading difficult conversations in the clin- ical setting, as well as feedback on the workshop. The workshop took place in the Care New England Simulation Center at Women & Infants Hospital. Simulation based difficult conversations Workshop The simulation based Difficult Conversations Workshop was a 4.5 hour workshop that con- sisted of three components (Fig 1). First, the participants were presented with a lecture on dif- ficult conversation communication skills. This lecture was 30 minutes long and highlighted the basic tenets of communication skills in healthcare. Next, each participant took part in a simulation Teaching Scenario, a clinically relevant practice difficult conversation with a trained improvisational actor that was about ten minutes long, while remaining participants observed the scenario via live video. Finally, at the end of the Teaching Scenarios, a facilitated debriefing session was held for all participants. This debriefing session was usually an hour to two hours in length. The workshop was led by a neonatologist who is the director of and trainer in the Difficult Conversations for Neonatal Fellows Training program. Each simulated Teaching Scenario reflected a situation typical of the NICU NP’s work environment. The trained actors functioned in the role of a parent during the simulated difficult conversations. Performance assessment The Test Scenario was a 10 minute conversation with a trained improvisational actor in a sim- ulated standardized clinical scenario. The encounter took place in the Women & Infants Hos- pital Simulation Center and was videotaped, but not shown via live video to any intervention NPs, control NPs or trainers (in contrast to the Teaching Scenarios). This was done to main- tain the integrity of the standardized Test Scenario for all intervention and control NPs. The Test Scenario simulation was scored at a later date independently by two blinded content expert observers. One observer was a board certified palliative care physician; the other observer was a doctorally prepared pediatric psychiatric clinical nurse specialist with expertise in interpersonal communication and relationships. These observers did not work with or know any of the participants and were blinded to participant group. In order to assess the PLOS ONE | https://doi.org/10.1371/journal.pone.0229895 March 9, 2020 3 / 12 PLOS ONE Difficult conversations simulation in the NICU Fig 1. Study flow diagram. NNP = neonatal nurse practitioner. https://doi.org/10.1371/journal.pone.0229895.g001 performance of each participant, the observers completed a quantitative communication skills performance checklist as well as assigning an empathy score to rate the participant’s level of empathy on a scale of 1 (no empathy) to 10 (extremely empathetic) (Fig 2). The quantitative communication skills performance checklist was developed using a two-step approach. A review of the literature was performed for communication skill checklists, then the final check- list was curated by the authors via expert consensus. The empathy score was developed via expert consensus. PLOS ONE | https://doi.org/10.1371/journal.pone.0229895 March 9, 2020 4 / 12 PLOS ONE Difficult conversations simulation in the NICU Fig 2. Evaluation tool utilized by blinded independent content experts to evaluate recorded Difficult Conversations Test Scenarios performed by participants. NNP = neonatal nurse practitioner. https://doi.org/10.1371/journal.pone.0229895.g002 PLOS ONE | https://doi.org/10.1371/journal.pone.0229895 March 9, 2020 5 / 12 PLOS ONE Difficult conversations simulation in the NICU Teaching & Test Scenarios In Teaching Scenario #1, a mother was informed that child protective services had informed the hospital that they would investigate the mother after her twins were born. In Teaching Scenario #2, a mother was informed that her infant had failed a congenital heart disease screening and needed to be transferred to a regional NICU to rule out congenital heart disease. In the Test Sce- nario, a mother was told that there was clinical suspicion of Down Syndrome in her newborn. Data analysis Statistical analysis was performed as follows. Differences in exhibition of communication skills between the groups was tested using Fisher’s exact test, and numbers of skills demonstrated and empathy scores were compared via the Student’s t-test. Inter-rater reliability on the scoring of the Test Scenario was measured for the communica- tion skills items using a pooled kappa statistic. Rater agreement on empathy scores was calcu- lated using the two one-sided t-tests (TOST) method, with agreement limits of ±3 points. Results 13 out of 31 participated; n = 5 in the intervention group, n = 7 in the control group. One video could not be assessed due to technical difficulties with sound recording. Demographics of the group and experience with difficult conversations as a trainee and in the clinical setting are presented in Table 1. Table 1. Participant demographics and experience with difficult conversations. Survey questions Number of years as an NP taking care of infants 0–1 2–5 6–10 > 10 Average number of weekly hours worked 12–24 25–32 33–40 41–55 > 55 NICU level most often worked in 3 or 4 2 1 Any work in level 1/2 community hospital nursery yes Take transport call yes Received education during training/career on communicating bad news to the family of an infant yes Number of times present in the past year when bad news was given to the family of an infant 0 n = 13 (%) 1 (8) 5 (38) 2 (15) 4 (31) 0 (0) 2 (17) 3 (25) 3 (25) 4 (33) 12 (92) 1 (8) 0 (0) 9 (69) 7 (54) 2 (15) 0 (0) (Continued ) PLOS ONE | https://doi.org/10.1371/journal.pone.0229895 March 9, 2020 6 / 12 PLOS ONE Difficult conversations simulation in the NICU Table 1. (Continued) Survey questions 1–2 3 or more Number of times in the past year you gave bad news to the family of an infant 0 1–2 3 or more Extent to which you feel competent to deliver bad news to a family of an infant in your care not competent somewhat competent moderately competent competent don’t know https://doi.org/10.1371/journal.pone.0229895.t001 n = 13 (%) 3 (23) 10 (77) 3 (23) 5 (38) 5 (38) 2 (15) 7 (54) 3 (23) 1 (8) 0 (0) Participation in the simulation based Difficult Conversations Workshop increases the use of communication skills in a simulated clinical encounter and increases participants’ empathy score. In the intervention group, the mean number of predefined communication skill behaviors exhibited by each participant was higher than in the control group (12 skills compared to 8 skills per scenario; p = 0.013). Among the individual communication skill behaviors compared individually between the groups, only “asks parents open-ended questions” was significantly higher in the post intervention group (p = 0.047). In the intervention group, the mean empa- thy score was higher compared to the control group (8.4 compared to 6.2; p = 0.015). Independent of participation in the simulation based Difficult Conversations Workshop, some communication skills are used more often than others. The frequency with which individual communication skills were applied in simulated clini- cal encounters was similar among the two groups. Some skills, such as “Introduces/Re-intro- duces self”, were almost always displayed, while others, such as “Asks parents to repeat back” were never displayed (Table 2). Interobserver agreement between the two independent blinded reviewers in communica- tion skill scores was 74% agreement overall (range for individual participants between 59 and 94% and for individual skills between 42% and 100%) with an interrater reliability pooled kappa of 0.77. In the empathy score, the top three scores and the bottom two were identical between the two reviewers, while there were some differences in the middle of the field. The two one sided t-tests demonstrated equivalence of empathy score differences, with differences within three points considered equivalent (p = 0.0030). On the post-intervention survey, participants rated the workshop between 5.8 and 6.0 on a variety of measures on a 6.0 scale (Table 3). Discussion Our study demonstrates that an intervention consisting of a structured lecture and simulation based communication skills workshop for neonatal NPs leads to an increase in the use of spe- cific communication skills as well as improvement in a perceived empathy score in a simulated difficult conversation setting. This is the first study assessing these objective outcomes in neo- natal nurse practitioners, while a previous study demonstrated improved self-reported confi- dence in difficult conversations in neonatal fellows and nurse practitioners [5]. As nurse PLOS ONE | https://doi.org/10.1371/journal.pone.0229895 March 9, 2020 7 / 12 PLOS ONE Difficult conversations simulation in the NICU Table 2. Utilized communications skills. Communications skill Introduces/Re-introduces self Body Position (Seated/Positioned at eye level to parent; not hovering over parent; lean forward toward parent) Makes statements that furnish hope (“I hope I am wrong about this”) Summarizes and makes a follow up plan. Assures parents they will be available Avoids medical jargon (“atypical features” instead of dysmorphic features/Down Syndrome/Trisomy 21) Uses expressions that communicate empathy (“I wish I had better news”) Uses the baby’s name during the conversation Suggests additional supportive resources for the parents (chaplain, social worker, etc) Asks what the parent(s) know/suspect Speaks slowly in short simple sentences Acknowledges the parents’ emotions (“I can see how worried you are,” “I know this must be shocking,” “It’s OK to cry,” “I can see that you don’t know what to say”) Asks parents open-ended questions Asks parent(s) if there is anyone else they would like to be present for the meeting Foreshadows the bad news (“I’m sorry but I have bad news”) Pauses consciously and allows for silence after delivering bad news If visitors present, gives family a choice on who should be present for the meeting Asks parents to repeat back what they have been told https://doi.org/10.1371/journal.pone.0229895.t002 Intervention group (n = 5) (%) Control group (n = 7) (%) 100 100 100 100 100 80 80 90 50 80 40 80 70 70 60 00 00 86 86 86 79 64 64 50 50 43 43 43 36 29 29 21 00 00 practitioners are important members of the multidisciplinary teams providing care for neo- nates in many NICUs across the United States, it is important to train neonatal NPs in difficult conversations and breaking bad news, particularly in the current changing climate in health- care, where NPs are providing more and more care in academic medical centers as well as community hospitals. Objectively assessable specific communication skills are important components of difficult conversations in the clinical setting. Our results are supported by other studies, which have shown that simulation is an effective tool for realistic training in difficult conversations in the context of neonatal care, for example in decision-making at the limits of viability [12], as well as in pediatrics [8], and leads to improved actual communication skills [7]. Studies also Table 3. Participant workshop evaluations. Mean score (1–6 [extremely ineffective/unsatisfactory—extremely effective/outstanding]) The lecture on communication skills was helpful/informative The simulation was helpful/informative The facilitated debriefing was helpful/informative The environment felt safe to ask questions/share thoughts After attending the workshop, I feel more competent to lead a difficult conversation Overall satisfaction with the session The workshop should be part of neonatal NP orientation/training https://doi.org/10.1371/journal.pone.0229895.t003 n = 13 5.9 5.9 6.0 6.0 5.8 6.0 6.0 PLOS ONE | https://doi.org/10.1371/journal.pone.0229895 March 9, 2020 8 / 12 PLOS ONE Difficult conversations simulation in the NICU demonstrate that taking part in a formal program aimed at improving communication skills in difficult conversations influences pediatric provider confidence in managing difficult clinical scenarios [13], as well as leading to better humanistic skills and better delivery of bad news [14] and to improved knowledge and comfort levels in communication [15]. Communication skills training has also been used successfully to improve residents’ skills in code status discussions [16], for genetic counseling training [17], communication for anes- thesia residents[10] and the disclosure of medical errors [18]. Furthermore, simulation has been shown to improve long term retention of skills and self-reported changes in behavior [19] [15] [20] as well as the long term retention of confidence in one’s communication skills in breaking bad news [21]. While it is important to assess objective communication skills in difficult conversations, not every aspect of the level of skill that care providers demonstrate can be assessed using a specific skill checklist. In addition to the objective communication skills that are necessary for breaking bad news, empathy plays a significant role in patient-provider communication. Parents prioritize communication[9] and want caring providers, for example when receiving prenatal consults by neonatologists for congenital anomalies [22]. Additionally, studies have shown that physician empathy is associated with increased adherence to therapy and improved clinical outcomes [23–25]. While possible differences in communication style between physi- cians and nurse practitioners have not been studied, nurses’ and physicians’ patterns of com- munication differ in enacted NICU conversations; physicians provide more biomedical information while nurses provide more psychosocial information [4]. Furthermore, empathy is an important component of the patient-practitioner interaction from the practitioner perspective as well. For example, empathy in medical students is associ- ated with a decrease in burnout [26, 27]. Thus, the empathy score was utilized as an additional marker of the interaction between the provider and patient. Our study demonstrates that lecture plus simulation based training improves empathy per- ceived by an expert observer in addition to improving objective communication skills. Empa- thy does not lend itself to one simple definition. One approach to categorizing empathy is into cognitive empathy vs. affective empathy, reviewed in [28], in which cognitive empathy is asso- ciated with external traits that can be learned, while affective empathy is not. Thus, for the pur- poses of this study, we defined empathy as a cognitive and thus behavioral trait, which consequently is a characteristic that lends itself to modification by training. Despite the abun- dance of alternative definitions for empathy [28], the two independent content expert observ- ers were able to assess empathy in the Test Scenarios with high interrater reliability. They ranked the level of empathy that participants displayed in scenarios similarly: both their high- est and lowest ranking participants were identical, irrespective of the actual number of the empathy score on the scale. Such concordance was achieved despite the fact that the content expert observers were not trained to look for specific signs, but received the sole instructions to score scenarios on a scale of 1 (no empathy) to 10 (extremely empathic). One limitation of this study is that individual participants were not tested using both a pre- and a post-intervention scenario, given that the increased time commitment necessary for that experimental model was not possible due to participants’ clinical staffing requirements. The disadvantage of not having the same participants in both the pre- and post-intervention group is a decrease in the signal to noise ratio. However, given that we nonetheless saw significant improvement in both specific skills and overall empathy scores, we hypothesize that the results would have been even stronger if interpersonal differences had been accounted for using the same participants for both arms of the study. Another limitation of the study is that the communication skill result and the empathy score result may not be independent variables, as it is possible that the observing content PLOS ONE | https://doi.org/10.1371/journal.pone.0229895 March 9, 2020 9 / 12 PLOS ONE Difficult conversations simulation in the NICU expert evaluators were subconsciously influenced in their assessment of the empathy score based on the number of communication skills demonstrated. If this were the case, this would not detract from the validity of the results. In fact, this mechanism, if it were at play in these assessments, would support the hypothesis that empathy can be learned as a specific skill set, aligning with the cognitive/behavioral definition of empathy, thus suggesting that specific learned communicative behavioral skills may impact the patient’s perception of empathy. Furthermore, we recognize that only 13/31 NPs took part in this workshop. Upon further investigation, the most common reason for non-participation included the complex schedul- ing of clinical load. To see if these results are generalizable, a larger cohort may be needed. Nonetheless, this small study demonstrated a difference between the intervention and control groups. Since others have shown that trainees and their program directors are more lenient in their assessment of communication simulation performance compared to patients and communica- tion experts [29], an advantage of this study is that we utilized independent blinded content experts to perform the video assessments for both the skills assessment and the empathy score. An additional approach that may improve difficult conversation skill and empathy scores may be to incorporate erroneous examples into the lecture component of the workshop, as these have been shown to improve breaking bad news simulation performance in nursing stu- dents [30]. In summary, our study demonstrates that a lecture and simulation based Difficult Conver- sations Simulation workshop improves objective communication skills and empathy in neona- tal nurse practitioners in conducting difficult conversations with patients’ families as perceived by an expert observer. Future studies will need to address the long term retention of learned communication skills as well as the transfer of communication skills from simulation settings to actual clinical practice. Author Contributions Conceptualization: Roberta Bowen, Elisabeth C. McGowan, Beatrice E. Lechner. Data curation: Roberta Bowen, Kate M. Lally, Francine R. Pingitore, Elisabeth C. McGowan, Beatrice E. Lechner. Formal analysis: Richard Tucker, Elisabeth C. McGowan, Beatrice E. Lechner. Investigation: Roberta Bowen, Kate M. Lally, Francine R. Pingitore, Elisabeth C. McGowan, Beatrice E. Lechner. Methodology: Kate M. Lally, Richard Tucker, Elisabeth C. McGowan, Beatrice E. Lechner. Project administration: Roberta Bowen, Elisabeth C. McGowan, Beatrice E. Lechner. Resources: Beatrice E. Lechner. Supervision: Beatrice E. Lechner. Writing – original draft: Roberta Bowen, Beatrice E. Lechner. Writing – review & editing: Roberta Bowen, Kate M. Lally, Francine R. Pingitore, Richard Tucker, Elisabeth C. McGowan, Beatrice E. Lechner. References 1. Ditton-Phare P, Loughland C, Duvivier R, Kelly B. Communication skills in the training of psychiatrists: A systematic review of current approaches. Aust N Z J Psychiatry. 2017; 51(7):675–92. https://doi.org/ 10.1177/0004867417707820 PMID: 28462636 PLOS ONE | https://doi.org/10.1371/journal.pone.0229895 March 9, 2020 10 / 12 PLOS ONE Difficult conversations simulation in the NICU 2. Deveugele M. Communication training: Skills and beyond. Patient Educ Couns. 2015; 98(10):1287–91. https://doi.org/10.1016/j.pec.2015.08.011 PMID: 26298220 3. Botwinski C. NNP education in neonatal end-of-life care: a needs assessment. MCN Am J Matern Child Nurs. 2010; 35(5):286–92. https://doi.org/10.1097/NMC.0b013e3181e62440 PMID: 20706099 4. Lamiani G, Meyer EC, Browning DM, Brodsky D, Todres ID. Analysis of enacted difficult conversations in neonatal intensive care. J Perinatol. 2009; 29(4):310–6. https://doi.org/10.1038/jp.2008.228 PMID: 19148109 5. Boss RD, Urban A, Barnett MD, Arnold RM. Neonatal Critical Care Communication (NC3): training NICU physicians and nurse practitioners. J Perinatol. 2013; 33(8):642–6. https://doi.org/10.1038/jp. 2013.22 PMID: 23448940 6. Dosanjh S, Barnes J, Bhandari M. Barriers to breaking bad news among medical and surgical residents. Med Educ. 2001; 35(3):197–205. https://doi.org/10.1046/j.1365-2923.2001.00766.x PMID: 11260440 7. Meyer EC, Brodsky D, Hansen AR, Lamiani G, Sellers DE, Browning DM. An interdisciplinary, family- focused approach to relational learning in neonatal intensive care. J Perinatol. 2011; 31(3):212–9. https://doi.org/10.1038/jp.2010.109 PMID: 20706191 8. Tobler K, Grant E, Marczinski C. Evaluation of the impact of a simulation-enhanced breaking bad news workshop in pediatrics. Simul Healthc. 2014; 9(4):213–9. https://doi.org/10.1097/SIH. 0000000000000031 PMID: 24787559 9. Meyer EC, Ritholz MD, Burns JP, Truog RD. Improving the quality of end-of-life care in the pediatric intensive care unit: parents’ priorities and recommendations. Pediatrics. 2006; 117(3):649–57. https:// doi.org/10.1542/peds.2005-0144 PMID: 16510643 10. Karam VY, Barakat H, Aouad M, Harris I, Park YS, Youssef N, et al. Effect of a simulation-based work- shop on breaking bad news for anesthesiology residents: an intervention study. BMC Anesthesiol. 2017; 17(1):77. https://doi.org/10.1186/s12871-017-0374-7 PMID: 28615002 11. Roque ATF, Lasiuk GC, Radunz V, Hegadoren K. Scoping Review of the Mental Health of Parents of Infants in the NICU. J Obstet Gynecol Neonatal Nurs. 2017; 46(4):576–87. https://doi.org/10.1016/j. jogn.2017.02.005 PMID: 28506679 12. Boss RD, Donohue PK, Roter DL, Larson SM, Arnold RM. "This is a decision you have to make": using simulation to study prenatal counseling. Simul Healthc. 2012; 7(4):207–12. https://doi.org/10.1097/SIH. 0b013e318256666a PMID: 22569285 13. Dickens DS. Building competence in pediatric end-of-life care. J Palliat Med. 2009; 12(7):617–22. https://doi.org/10.1089/jpm.2009.0032 PMID: 19594346 14. Vetto JT, Elder NC, Toffler WL, Fields SA. Teaching medical students to give bad news: does formal instruction help? J Cancer Educ. 1999; 14(1):13–7. https://doi.org/10.1080/08858199909528567 PMID: 10328318 15. Vadnais MA, Dodge LE, Awtrey CS, Ricciotti HA, Golen TH, Hacker MR. Assessment of long-term knowledge retention following single-day simulation training for uncommon but critical obstetrical events. J Matern Fetal Neonatal Med. 2012; 25(9):1640–5. https://doi.org/10.3109/14767058.2011. 648971 PMID: 22191668 16. Szmuilowicz E, Neely KJ, Sharma RK, Cohen ER, McGaghie WC, Wayne DB. Improving residents’ code status discussion skills: a randomized trial. J Palliat Med. 2012; 15(7):768–74. https://doi.org/10. 1089/jpm.2011.0446 PMID: 22690890 17. Holt RL, Tofil NM, Hurst C, Youngblood AQ, Peterson DT, Zinkan JL, et al. Utilizing high-fidelity crucial conversation simulation in genetic counseling training. Am J Med Genet A. 2013; 161A(6):1273–7. https://doi.org/10.1002/ajmg.a.35952 PMID: 23633180 18. Matos FM, Raemer DB. Mixed-realism simulation of adverse event disclosure: an educational method- ology and assessment instrument. Simul Healthc. 2013; 8(2):84–90. https://doi.org/10.1097/SIH. 0b013e31827cbb27 PMID: 23334365 19. Hughes S, Cusack J, Fawke J. PC.68 Learning point recall and self-perceived behavioural change fol- lowing multi-disciplinary high fidelity point of care simulation training. Arch Dis Child Fetal Neonatal Ed. 2014; 99 Suppl 1:A59. 20. Bender J, Kennally K, Shields R, Overly F. Does simulation booster impact retention of resuscitation procedural skills and teamwork? J Perinatol. 2014; 34(9):664–8. https://doi.org/10.1038/jp.2014.72 PMID: 24762413 21. Karkowsky CE, Landsberger EJ, Bernstein PS, Dayal A, Goffman D, Madden RC, et al. Breaking Bad News in obstetrics: a randomized trial of simulation followed by debriefing or lecture. J Matern Fetal Neonatal Med. 2016; 29(22):3717–23. https://doi.org/10.3109/14767058.2016.1141888 PMID: 26786087 PLOS ONE | https://doi.org/10.1371/journal.pone.0229895 March 9, 2020 11 / 12 PLOS ONE Difficult conversations simulation in the NICU 22. Miquel-Verges F, Woods SL, Aucott SW, Boss RD, Sulpar LJ, Donohue PK. Prenatal consultation with a neonatologist for congenital anomalies: parental perceptions. Pediatrics. 2009; 124(4):e573–9. https://doi.org/10.1542/peds.2008-2865 PMID: 19736266 23. Hojat M, Louis DZ, Markham FW, Wender R, Rabinowitz C, Gonnella JS. Physicians’ empathy and clin- ical outcomes for diabetic patients. Acad Med. 2011; 86(3):359–64. https://doi.org/10.1097/ACM. 0b013e3182086fe1 PMID: 21248604 24. Rakel D, Barrett B, Zhang Z, Hoeft T, Chewning B, Marchand L, et al. Perception of empathy in the ther- apeutic encounter: effects on the common cold. Patient Educ Couns. 2011; 85(3):390–7. https://doi. org/10.1016/j.pec.2011.01.009 PMID: 21300514 25. Del Canale S, Louis DZ, Maio V, Wang X, Rossi G, Hojat M, et al. The relationship between physician empathy and disease complications: an empirical study of primary care physicians and their diabetic patients in Parma, Italy. Acad Med. 2012; 87(9):1243–9. https://doi.org/10.1097/ACM. 0b013e3182628fbf PMID: 22836852 26. Brazeau CM, Schroeder R, Rovi S, Boyd L. Relationships between medical student burnout, empathy, and professionalism climate. Acad Med. 2010; 85(10 Suppl):S33–6. https://doi.org/10.1097/ACM. 0b013e3181ed4c47 PMID: 20881699 27. Gleichgerrcht E, Decety J. Empathy in clinical practice: how individual dispositions, gender, and experi- ence moderate empathic concern, burnout, and emotional distress in physicians. PLoS One. 2013; 8 (4):e61526. https://doi.org/10.1371/journal.pone.0061526 PMID: 23620760 28. Preusche I, Lamm C. Reflections on empathy in medical education: What can we learn from social neu- rosciences? Adv Health Sci Educ Theory Pract. 2016; 21(1):235–49. https://doi.org/10.1007/s10459- 015-9581-5 PMID: 25597025 29. Wayne DB, Cohen E, Makoul G, McGaghie WC. The impact of judge selection on standard setting for a patient survey of physician communication skills. Acad Med. 2008; 83(10 Suppl):S17–20. https://doi. org/10.1097/ACM.0b013e318183e7bd PMID: 18820492 30. Schmitz FM, Schnabel KP, Stricker D, Fischer MR, Guttormsen S. Learning communication from erro- neous video-based examples: A double-blind randomised controlled trial. Patient Educ Couns. 2017; 100(6):1203–12. https://doi.org/10.1016/j.pec.2017.01.016 PMID: 28179074 PLOS ONE | https://doi.org/10.1371/journal.pone.0229895 March 9, 2020 12 / 12 PLOS ONE
null
10.1088_1361-6579_ad0426.pdf
Data availability statement The data cannot be made publicly available upon publication because they are not available in a format that is sufficiently accessible or reusable by other researchers. The data that support the findings of this study are available upon reasonable request from the authors.
Data availability statement The data cannot be made publicly available upon publication because they are not available in a format that is sufficiently accessible or reusable by other researchers. The data that support the findings of this study are available upon reasonable request from the authors.
Physiol. Meas. 44 (2023) 125004 https://doi.org/10.1088/1361-6579/ad0426 RECEIVED 20 June 2023 REVISED 11 September 2023 ACCEPTED FOR PUBLICATION 17 October 2023 PUBLISHED 15 December 2023 PAPER Photoplethysmography-based cuffless blood pressure estimation: an image encoding and fusion approach Yinsong Liu1 1 Department of School of Electronic Engineering, Beijing University of Posts and Telecommunication, Beijing 100876, Peopleʼs Republic , Junsheng Yu1,2,3,* and Hanlin Mou4 of China 2 School of Physics and Electronic Information, Anhui Normal University, Wuhu 241003, Peopleʼs Republic of China 3 School of Intelligence and Digital Engineering, Luoyang Vocational College of Science and Technology, Luoyang 471000, Peopleʼs Republic of China 4 Chinese Academy of Sciences Aerospace Information Research Institute, Beijing 100094, Peopleʼs Republic of China * Author to whom any correspondence should be addressed. E-mail: ysliu@bupt.edu.cn, jsyu@bupt.edu.cn and mouhl@aircas.ac.cn Keywords: blood pressure, time series to image conversion, deep learning, photoplethysmography, physiological signals Abstract Objective. Photoplethysmography (PPG) is a promising wearable technology that detects volumetric changes in microcirculation using a light source and a sensor on the skin’s surface. PPG has been shown to be useful for non-invasive blood pressure (BP) measurement. Deep learning-based BP measurements are now gaining popularity. However, almost all methods focus on 1D PPG. We aimed to design an end-to-end approach for estimating BP using image encodings from a 2D perspective. Approach. In this paper, we present a BP estimation approach based on an image encoding and fusion (BP-IEF) technique. We convert the PPG into five image encodings and use them as input. The proposed BP-IEF consists of two parts: an encoder and a decoder. In addition, three kinds of well- known neural networks are taken as the fundamental architecture of the encoder. The decoder is a hybrid architecture that consists of convolutional and fully connected layers, which are used to fuse features from the encoder. Main results. The performance of the proposed BP-IEF is evaluated on the UCI database in both non-mixed and mixed manners. On the non-mixed dataset, the root mean square error and mean absolute error for systolic BP (SBP) are 13.031 mmHg and 9.187 mmHg respectively, while for diastolic BP (DBP) they are 5.049 mmHg and 3.810 mmHg. On the mixed dataset, the corresponding values for SBP are 4.623 mmHg and 3.058 mmHg, while for DBP the values are 2.350 mmHg and 1.608 mmHg. In addition, both SBP and DBP estimation on the mixed dataset achieved grade A compared to the British Hypertension Society standard. The DBP estimation on the non-mixed dataset also achieved grade A. Significance. The results indicate that the proposed approach has the potential to improve on the current mobile healthcare for cuffless BP measurement. 1. Introduction The accelerated pace of modern society, along with people’s more excessive lifestyles, has resulted in an increasing number of individuals suffering from cardiovascular disease (CVD) such as hypertension and hyperlipidemia, which is significantly affecting people’s overall health. As the number of CVD cases continues to increase, this disease poses a significant health threat, and its high mortality rate has generated significant concern within the medical community. Early prevention of CVD has emerged as a crucial challenge in the field of modern medicine. Blood pressure (BP) is the lateral pressure of blood against the blood vessel wall, which is a significant indication of the human circulatory system (Rundo et al 2018). It is widely recognized as an indicator of health, and forms the fundamental basis for diagnosis and treatment in clinical practice. As a result, accurate BP monitoring has become critical in the prevention and management of CVD. The invasive method and the cuff-based sphygmomanometer method are the two primary approaches for BP monitoring. The invasive © 2023 Institute of Physics and Engineering in Medicine Physiol. Meas. 44 (2023) 125004 Y Liu et al method is used in intensive care units to directly measure BP in the most precise way possible using a catheter put into the artery. Due to the invasiveness of the catheter, this procedure is typically performed by doctors and specialist nurses (Griffin et al 2014). As a result, invasive methods are unsuitable for routine BP monitoring. However, cuff-based sphygmomanometers restrict body movement during measurement and the method, being cumbersome, is unsuitable for prolonged BP monitoring (Kaniusas et al 2006). Therefore, cuffless BP measuring technologies have attracted a lot of attention. The photoplethysmogram (PPG) signal, a pulsatile waveform reflecting blood volume changes in peripheral tissue, has been explored as a potentially valuable signal in non-invasive BP monitoring. Many studies (Hosanee et al 2020, Elgendi et al 2019) have clearly demonstrated the association between PPG and BP. Pulse transfer, pulse arrival time, and pulse wave velocity have been frequently utilized for early cuffless BP estimation (Chua and Heneghan 2006, Zhang et al 2011, Ma et al 2018). However, these methods require calibration for specific model parameters while potentially ignoring detailed PPG characteristics related to BP. The development of artificial intelligence technologies has led to the widespread utilization of deep learning algorithms in BP estimation. Most research employs 1D PPG signals as input, and two main approaches exist in this domain. One approach involves manual feature extraction, where significant features are manually extracted from the PPG signal and subsequently fed into a neural network for BP calculation. The other approach is an end-to-end method that automatically extracts features from the PPG signal and estimates BP values using a neural network. The former needs manual feature extraction, which is time-consuming and not suitable. However, the latter approach has limitations and cannot guarantee the accuracy of the features. The first feature extraction method utilizes the features extracted from the PPG signal as the input for the neural networks. Kurylyak et al (2013) demonstrated that artificial neural networks outperform traditional linear regression algorithms by extracting a set of properties from PPG signals. Duan et al (2016) used three distinct feature sets, each consisting of 11 features, to predict systolic BP (SBP), diastolic BP (DBP), and mean BP (MBP). The mean absolute error (MAE) for SBP, DBP, and MBP in this research is 4.77 ± 7.68, 3.67 ± 5.69, and 3.85 ± 5.87mmHg, respectively. Yang et al (2020) extracted 90 complex features from PPG and electrocardiogram (ECG) signals and utilized three regression models (ANN, SVM, and LASSO) to predict SBP and DBP. Wu et al (2018) utilized handcrafted features and personal features as input to design a deep neural network for BP estimation. Wang et al (2023) used ECG and PPG signals to estimate BP. They employed nine different feature parameters and a classification algorithm to develop multiple linear regression models for BP estimation. In their study, the MAE values were 4.46 mmHg for SBP and 4.20 mmHg for DBP. A typical problem with these methods is that the accuracy of the final BP estimation is strongly affected by the quality of features extracted from PPG, thus feature extraction should be performed carefully. In the second end-to-end approach, the PPG signal undergoes several pre-processing steps before being input into the neural networks. Mou and Yu (2021) employed convolution neural network (CNN) in combination with long short-term memory network (LSTM) to enhance continuous BP estimation. Baek et al (2019) employed PPG and ECG as input to train a deep convolutional neural network. Paviglianiti et al (2021) utilized PPG and ECG as inputs, and they used multiple neural networks to predict BP. Eom et al (2020) proposed an end-to-end BP estimate approach that is based on CNN and recurrent neural network (RNN), utilizing ECG and PPG as input. In their study, the MAE for SBP and DBP was achieved at 1.10mmHg and 0.58 mmHg, respectively. Zhong et al (2023) proposed a model for continuous cuffless BP estimation using a mixed attention gating U-Net. In their study, they achieved an MAE of 3.49mmHg for SBP and 2.11 mmHg for DBP. Hu et al (2022) proposed an end-to-end continuous BP estimation method, named MSF-MTLNet. They utilized multi-task learning techniques and attention mechanisms for searching and fusing multi-scale features. The MAE for SBP and DBP reached 0.97 ± 8.87 mmHg and 0.55 ± 4.23 mmHg, respectively. Qin et al (2023) proposed a deep learning network that is based on a modified ResNet34 and channel attention mechanism for continuous BP prediction using only PPG signals. The model was evaluated using the UCI dataset, and obtained MAE values of 5.98 mmHg for SBP and 3.24 for DBP. These methods often rely on 1D CNNs to extract features automatically. However, CNNs are widely utilized in the field of computer vision, and it remains to be verified whether 1D CNNs are effective for feature extraction. All the aforementioned methods typically utilize 1D PPG signals. Several prior studies have employed image tools to analyze and process 1D biosignals. Chen et al (2018) used the visibility graph to analyze intraspinal pressure fluctuations after spinal cord injury, and the data showed that clinically important information can be captured using the visibility graph. Shao (2010) built a complex network for the sequence of heartbeat intervals and confirmed that the assortative coefficient of associated networks could differentiate between healthy subjects and patients with congestive heart failure. Ji et al (2016) used the visibility graph to analyze EEG signals to classify normal people and people with workplace stress. To the best of our knowledge, Wang et al (2021) were the first to use the visibility graph for the challenge of BP estimation and they employed a natural visibility graph (NVG) constructed from the PPG to train convolutional neural networks. However, their experiments were based on mixed datasets, resulting in typically low estimating errors. Additionally, the creation of a visibility 2 Physiol. Meas. 44 (2023) 125004 Y Liu et al graph may result in the loss of many features relating to BP, as it does not consider the morphological properties of the PPG signal. In terms of the problem identified in the related literature, our objective is to develop a new approach to estimate BP from an image-based perspective in order to achieve an improved performance. In this paper, we propose a novel method for BP estimation based on an end-to-end image encoding and fusion method (BP- IEF). The proposed method consists of two main components: the encoder and the decoder. The encoder is responsible for extracting the high-dimensional features from the image encodings, while the decoder is used to fuse these features and calculate the BP values. Five different image encodings, namely the PPG figure, NVG, horizontal visibility graph (HVG), Gramian angular summary field (GASF), and Gramian angular difference field (GADF), are employed as inputs in our method. Due to the variability of the cardiovascular system among individuals, the subject-independent nature of the training set and the test set results in generally low BP estimation errors. To evaluate the effectiveness of our method, we designed two dataset segmentation strategies based on the UCI dataset: mixed and non-mixed. The experimental results demonstrate the competitive performance of our proposed method. The objective of this paper is to provide a reliable and accurate estimation of SBP and DBP utilizing PPG signals in the form of images. Our contributions are as follows: (1) Based on an image perspective, we propose a novel method named BP-IEF for estimating BP from PPG signals using the feature fusion technique. BP-IEF consists of encoder and decoder components. The encoder extracts high-dimensional features from image encoding, while the decoder fuses high- dimensional features from multiple image encodings. In comparison to the state-of-the-art approaches, our method shows promising results. (2) We selected five alternative image encodings as input for BP-IEF, with each encoding reflecting a distinct characteristic of the PPG waveform. ResNet18, MobileNetv3, and Vit were employed by the encoder for feature extraction. A feature fusion network was designed for the decoder component to fully fuse the features of the distinct image encodings. Consequently, BP-IEF combines the advantages of diverse image encodings to achieve more accurate BP measurements. (3) We performed validation of BP-IEF on a publicly available dataset, utilizing both non-mixed and mixed dataset settings. The experimental results have demonstrated the effectiveness of our proposed BP-IEF. It can estimate BP accurately. The remainder of this paper is organized as follows: In section 2, we provide a description of the dataset, image encodings, and our proposed BP-IEF approach. In section 3, we present the numerical results and analyze the performance of our approach. Section 4 concludes with a brief summary. 2. Material and methods This section begins with a brief description of the UCI dataset. Subsequently, we introduce our data processing steps. This is followed by an explanation of the image encodings we utilized. Lastly, we provide a detailed demonstration of our proposed BP-IEF. 2.1. Dataset The experiments in this manuscript used the cuffless BP estimation dataset sourced from the University of California, Irvine (UCI) machine learning library (Kachuee et al 2015). This dataset is a subset of the Multiparametric Intelligent Monitoring in Intensive Care (MIMIC) II waveform database (Goldberger et al 2000). Because the MIMIC database contains a large quantity of damaged and distorted signals, the UCI dataset underwent several signal pre-processing procedures, including smoothing, filtering, and outlier reduction. As a result, the UCI dataset contains clean and valid signal recordings. Each of the 12,000 recordings in the UCI dataset contains fingertip PPG, ECG, and arterial BP (ABP) signals, captured at a frequency of 125 Hz. In this paper, the PPG and ABP signals of 450 records from the UCI dataset were used. The PPG signal recordings were further divided into non-overlapping windows of length 2 s for later BP estimation. To ensure fairness, in this manuscript, each patient contributes 150 PPG windows for the experiments conducted in this manuscript. The maximum and minimum ABP values within each ABP window were utilized as the reference SBP and DBP values for the corresponding PPG window. To accurately evaluate the performance of our method, we used both non-mixed and mixed dataset schemes. In the non-mixed dataset, we randomly selected 360 patients out of the total 450 patients as the training set, leaving the remaining 90 patients for the test set. In 3 Physiol. Meas. 44 (2023) 125004 Y Liu et al Figure 1. PSD analysis of the PPG signal. Figure 2. Example of a two-second windowing process. the mixed dataset, we randomly selected 120 windows for each patient in the training set and 30 windows for the test set. 2.2. Data pre-processing The authors of the UCI dataset performed several pre-processing procedures, as illustrated below. Firstly, all signals were smoothed using a simple averaging filter to eliminate signal blocks with severe discontinuities. Secondly, signal blocks with irregular and unacceptable human BP values were eliminated. Additionally, signal blocks with unacceptable heart rates were removed. Finally, the autocorrelation of the PPG signal, which indicates the similarity between successive pulses, was calculated, and blocks with high alteration were removed. By applying these steps to all samples in the database, we assert that the processed database is adequately filtered for usage. We randomly selected a patient’s PPG signal and analyzed its power spectral density (PSD), as shown in figure 1. The analysis reveals that the predominant frequency components are distributed within the interval of 0–5 Hz. Low-frequency components contain valuable information regarding neural activities and the regulation of the cardiovascular system. Moreover, it indicates effective filtering of high-frequency noise. Therefore, we solely conducted a window screening process without employing any additional filtering operations. Given that neural networks generally require a certain input size, this paper employs a two-second PPG window consisting of 250 samples. As shown in figure 2, we first use the peak detection algorithm 4 Physiol. Meas. 44 (2023) 125004 Y Liu et al Figure 3. SBP and DBP distribution in the non-mixed dataset. Figure 4. SBP and DBP distribution in the mixed dataset. (Elgendi et al 2013) to identify the location of the systolic peak. Subsequently, 124 samples to the left and 125 samples to the right of the systolic peak are selected to construct the PPG window. This process is repeated until all PPG windows are formed. To prevent overlap between PPG windows, any systolic peaks that fall within the created windows are removed. Additionally, a screening process is implemented to ensure the integrity of the PPG windows. The peaks in the autocorrelation signal are utilized to assess the quality of each PPG window. A predetermined threshold of 0.65 is empirically established for the maximum autocorrelation. Following this, the PPG windows are normalized to a range of [0, 1]. Each PPG window is associated with an ABP window positioned at the same time interval. The reference SBP and DBP values for the corresponding PPG windows were calculated using the maximum and minimum ABP values within the window. Physiologically appropriate systolic and diastolic BP ranges were 75 to 165 mmHg and 40 to 80 mmHg, respectively (Schrumpf et al 2021). 5 Physiol. Meas. 44 (2023) 125004 Y Liu et al Figure 5. Five types of image encodings for PPG. Consequently, any windows with BP values outside of these ranges were removed. Each PPG window can generate a pair of SBP/DBP estimations. After analyzing the distributions of SBP and DBP in the dataset (see figures 3 and 4), it was observed that the mean and variance of each SBP and DBP were significantly different. These differences posed challenges for solving the multivariate regression models. To address this issue, a BP normalization procedure was conducted with the aim of reducing the disparity between SBP and DBP. Without normalization, the models would disproportionately emphasize the variable with the larger range due to the varying ranges of the two variables. The normalization process utilized the SBP and DBP ranges mentioned earlier. The formula is as follows, and the maximum and minimum values can be derived from the range of variables: Z = - min X - max min , X = SBP or DBP Inverse normalization is required when the models make predictions. X = Z max min · ( - ) + min 1( ) ( ) 2 Since our modified models require the input to have dimensions of 1 × 224 × 224, we perform a resize operation on the image encodings. Figure 5 displays five different types of image encodings for a two-second PPG window to help explain this process. 2.3. Image encodings This subsection outlines the image encodings used in the conversion of 1D PPG signals into 2D images. 2.3.1. PPG figure Plotting is a critical signal analysis approach in the field of signal processing because it visually displays the period, amplitude, phase, and other characteristics of a signal, aiding human comprehension and study. Wang et al (2021) proposed using an image perspective to analyze PPG signals and successfully applied this approach to estimate BP. However, the visibility graph only illustrates the relative magnitude between sample points. The entire plotting of the PPG signal can keep all of the details of the original signal while also improving BP estimate accuracy due to CNN’s superior feature learning capacity. The PPG signal plotting is done using a figure size of 2.5× 2.5 inches and a dpi setting of 100, as shown in figure 5. The objective is to align each pixel point with a sample point in the PPG signal, which is more suitable for training CNNs. 6 Physiol. Meas. 44 (2023) 125004 Y Liu et al Figure 6. Examples of NVG (left) and HVG (right). 2.3.2. Visibility graph Lacasa et al (2008) proposed the visibility graph as a geometric approach for creating complex networks from time series. Visibility graphs can be used to model time series into complex networks, and then use graph theory to investigate time series characteristics. The main principle underlying the visibility graph technique is the one- to-one correlation between time series samples and nodes in the visibility graph. Two nodes are considered related if they can be connected by a visibility line. The created visibility graph is unaffected by changes in the vertical and horizontal axes (translations and rescaling). The visibility graph offers an effective way to store both time series characteristics and data (Dai et al 2019). The NVG and HVG are the two types of visibility graphs (Luque et al 2009) as shown in figure 6. A visibility graph is created using a time series of length N, denoted as S(i), where i = 1, 2, 3,K,N. The i, jth sample in the time series is represented as S(i), S( j), with s(i) and s( j) corresponding to the magnitudes. When any further intermediate sample S(k), i < k < j with magnitude s(k) meets the following conditions for NVG (equation (3)) and HVG (equation (4)), an undirected edge is produced between nodes. The detailed procedure is as follows: ( ) s k < ( ) s i + ( ( ) s j - ( )) s i k j - - i i ( ) s i > ( ) s k , s ( ) j > ( ) s k ( ) 3 ( ) 4 Without any optimization, the construction cost of an NVG is generally O(N2), where N represents the length of the time series (Li et al 2018). On the other hand, the HVG is a simplified version of the NVG, with a computational complexity of O(N). The adjacency matrix serves as a representation of the graph structure, with each element indicating the presence or absence of edges between nodes (Stephen et al 2015). For each PPG window, the corresponding visibility graph is initially constructed using the aforementioned algorithm. Subsequently, the adjacency matrix of the visibility graph is extracted and utilized to train the models for BP estimation. In this manuscript, both the NVG and HVG generated from each PPG window in this study, as illustrated in figure 5, are employed. 2.3.3. Gramian angular field Wang et al (2015) first proposed the Gramian angular field (GAF), which consists of two types: the Gramian angular summary field (GASF) and the Gramian angular difference field (GADF). The GAF is derived from Gram’s matrix, which is constructed from the dot products of multiple vectors, revealing their temporal relationship. Converting a 1D time series into a 2D GAF preserves the temporal dependencies between individual samples. This property is advantageous for maintaining the temporal nature of the PPG signal. GAF achieves this by increasing time representation as positions move from the upper left to the lower right corner. Additionally, GAF employs polar coordinates to represent the time series, using the triangular sum or difference between samples to identify time dependencies at various intervals. The calculations for GAF are as follows: S = { s following equation: , ( ) 1 , ( ) 2 s   , s ( ) i , , s ( N } ) is a time series of length N. First, S is normalized to the range of [ - ] 1, 1 by the ( s ( ) i - ˜ s ( ) i = max max ( )) S ( ) S + - ( s ( ) i min - min ( )) S ( ) S , i = 1, 2,  , N ( ) 5 The normalized time series, denoted as coordinates using the following equation: ˜ S = {˜ s , ( ) 1 ˜ , s ( ) 2   ˜ , s ( ) i , , ˜ } s ) ( N , is subsequently converted to polar 7 Physiol. Meas. 44 (2023) 125004 Y Liu et al Figure 7. The encoder part of the image fusion method. Figure 8. The decoder part of the image fusion method. f = arccos (˜ ) s ( ) i , - 1 r =  i n , ⎧ ⎨ ⎩ ˜ s ( ) i  1, ˜ s ( ) i Î ˜ S ( ) 6 , the range of f is 0,[ ( ) Î - ] i˜ Since s 1, 1 f one, because cos( )f is monotonic on [ ]p . The input and output values in the polar coordinates are one-to- Î p 0,[ ] . Finally, the transformed angle and radius features undergo Gram-like matrix operations to obtain the final feature matrix. This matrix can be expressed as either the cosine of the angle sum or the sine of the difference, accounting for the time dependence. The mathematical definitions of GASF and GADF are provided below. GASF = [ cos ( f i + f j )] cos ( f 1  + f ) 1  cos ( + f n ) f 1  cos ( f n + f ) 1  cos ( f n + f n ) ⎤ ⎥ ⎥ ⎦ ⎡ ⎢ ⎢ ⎣ 8 ( ) 7 Physiol. Meas. 44 (2023) 125004 GADF = [ sin ( f i - f j )] sin ( f 1  - f ) 1  sin ( - f n ) f 1  sin ( f n - f ) 1  sin ( f n - f n ) ⎤ ⎥ ⎥ ⎦ ⎡ ⎢ ⎢ ⎣ Y Liu et al ( ) 8 2.4. Image fusion The image feature fusion method proposed in this paper consists of two parts: the encoder (shown in figure 7) and the decoder (shown in figure 8). The encoder employs three different network models to extract the high- dimensional features from PPG image encodings. On the other hand, the decoder incorporates channel dimension summation, averaging, convolutional layers, and fully connected (FC) layers to fuse the extracted features from the encoder and generate the SBP and DBP values. 2.4.1. Encoder The field of computer vision has witnessed remarkable advancements in artificial intelligence techniques, giving rise to various outstanding neural networks like ResNet, MobileNet, and ViT. In the biomedical domain, several studies (Wang et al 2021, Salem et al 2018, Wu et al 2018) have demonstrated the successful transfer of models pre-trained on ImageNet, resulting in a favorable performance. Transfer learning enables us to apply empirical knowledge gained from one problem to a related but different task. Pre-training provides the advantage of establishing optimal initial weights, facilitating faster and better learning of the target task by leveraging prior understanding of the related task. In this manuscript, we select three pre-trained models (Resnet18 (He et al 2016), MobileNetv3 (Howard et al 2019), and Vit (Dosovitskiy et al 2020)) from ImageNet (Deng 2009) as encoders. Additionally, the input and output sizes of each model are modified from 3 and 1000, to 1 and 2, respectively. The models mentioned below are all modified. Given that ImageNet significantly differs from the task of BP estimation, utilizing pre-trained weights directly for feature extraction would inevitably result in increased errors in BP estimation. In this manuscript, we utilize the pre-trained weights as the initialization weights for our models. Subsequently, we continue training the models on our dataset until convergence is achieved. Finally, we utilize the input of the final FC layer of the models as the representation of high-dimensional features for the image input. The subsequent mathematical formulation describes this process. As previously mentioned, we utilize five types of image encodings from the = i{ X i, PPG window as input, which can be denoted as X = = j{ F j, The models of the encoder can be denoted as F = { V 1, 2, 3, i represented as V ji = } 1, 2, 3 . Finally, the output of the encoder is } 1, 2, 3, 4, 5 . . Each Xi has a shape of (1, 250, 250). } 1, 2, 3, 4, 5 F X j i j = = = ( ) , Algorithm 1. Image feature extraction. Input: Current image encoding from PPG, Xi; Current model of encoder, Fj; Output: Image features on the current input and model, Vji; 1: Resize the input Xi and get X ;i¢ 2: Train the model Fj on X ;i¢ 3: Extract the input of last FC layer of the modelFj as Vji; 4: return Vji; In particular, we applied a fixed model with one image encoding as input for each feature extraction process. In this paper, we employ three models and five image encodings to extract features a total of 15 times. It is important to highlight that each operation is conducted independently from the others. Specifically, we initialized the model parameters by utilizing pre-trained weights on ImageNet and subsequently retrained the model using our dataset until convergence. Finally, we evaluated the converged model on our dataset and use the final FC layer’s input as the feature output. 2.4.2. Decoder Each unique feature typically corresponds to a specific component of the data. As an example, the visibility graph solely takes into account the amplitude difference between samples. Given that different feature vectors capture distinct aspects of the pattern, the advantage of feature fusion becomes evident. The optimal combination of these features preserves valuable information while eliminating redundancy among the vectors. Therefore, the design of the decoder plays a crucial role. Our proposed image feature fusion method primarily relies on the decoder to fuse the high-dimensional features extracted by the encoder and calculate the SBP and DBP values. Multiple fusion blocks are employed to ensure thorough fusion of the high-dimensional encoder 9 Physiol. Meas. 44 (2023) 125004 Y Liu et al Table 1. RMSE and MAE when predicting SBP and DBP values using three models trained with five image encodings as input. Furthermore, both non-mixed and mixed PPG datasets were evaluated and analyzed. Dataset Input Model SBP DBP RMSE (mmHg) MAE (mmHg) RMSE (mmHg) MAE (mmHg) Non-mixed Plot NVG HVG GASF GADF Mixed Plot NVG HVG GASF GADF ResNet18 MobileNetv3 Vit ResNet18 MobileNetv3 Vit ResNet18 MobileNetv3 Vit ResNet18 MobileNetv3 Vit ResNet18 MobileNetv3 Vit ResNet18 MobileNetv3 Vit ResNet18 MobileNetv3 Vit ResNet18 MobileNetv3 Vit ResNet18 MobileNetv3 Vit ResNet18 MobileNetv3 Vit 13.973 15.153 14.626 13.524 15.637 15.189 14.938 15.646 14.506 13.155 13.494 14.545 13.302 14.588 14.014 5.149 7.676 7.213 5.161 8.342 7.027 6.850 11.514 9.085 4.692 5.631 5.313 4.668 7.162 5.798 9.726 11.561 11.355 9.520 12.026 11.217 10.926 12.181 11.227 9.119 9.780 10.515 9.330 10.543 10.888 3.348 5.634 4.397 3.439 6.225 4.665 4.505 8.656 5.751 3.085 3.904 3.373 3.056 5.215 3.536 5.322 6.017 5.614 5.381 5.862 5.595 5.562 5.976 5.831 5.087 5.353 5.573 5.082 5.528 5.471 2.556 3.411 3.278 2.590 3.823 3.303 3.134 4.773 3.907 2.349 2.774 2.673 2.364 3.340 2.844 3.962 4.641 4.363 4.077 4.554 4.357 4.294 4.742 4.619 3.791 4.090 4.256 3.788 4.124 4.377 1.721 2.566 2.167 1.789 2.903 2.301 2.179 3.677 2.685 1.608 1.976 1.766 1.601 2.494 1.846 features. As shown in figure 8, the decoder consists of three main parts. The first part encompasses the summation and averaging operation among the features. The second part consists of three fusion blocks connected in series, with each block comprising two residual-connected branches. The first branch involves a 1D convolutional layer with a kernel size of 3, a stride of 2, and a padding of 1. The second branch consists of an average pooling layer and a convolutional layer. The convolutional layer has a kernel size of 1 and a stride of 1. The average pooling layer is configured with a pool size of 3, a stride of 2, and a padding of 1. Lastly, the third part is composed of a two-layer FC layer with 512 and 64 neurons, respectively. Algorithm 2. Image feature fusion. Input: Features of a model from encoder, V; Output: SBP and DBP values; 1: Calculate the mean and sum of the input then concatenate them as V ;¢ 2: Train the Conv and FC layers on V ;¢ 3: Evaluate the Conv and FC layers on V ;¢ 4: return SBP and DBP; Similarly to the encoder, we utilize the features from a single model as the decoder’s input every time. In this work, we perform feature fusion procedures three times. Importantly, each feature fusion is independent of the others. Due to the usage of three pre-trained models in the encoder, the size of the feature vectors produced by each model varies. As a result, the input V has three distinct input shapes: (5,512), (5,1024), and (5,192). The following is a ResNet18 example that explains the procedure of the decoder. First, we concatenate the features before passing them to the decoder. Specifically, we concatenate five features with a shape of (1,512) to obtain a consolidated V of shape (5,512). Subsequently, on the channel dimension, we conduct summation and averaging operations on the multi-channel V, and concatenate the resulting values, resulting in a V¢ of shape 10 Physiol. Meas. 44 (2023) 125004 Y Liu et al Figure 9. Pearson correlation and Bland–Altman analysis for BP-IEF-ResNet18 on the non-mixed dataset. Table 2. RMSE and MAE when using image fusion methods on mixed and non-mixed datasets. The X in BP-IEF-X represents the model used in the encoder. Dataset Model SBP DBP RMSE (mmHg) MAE (mmHg) RMSE (mmHg) MAE (mmHg) Non-mixed dataset Mixed dataset BP-IEF-ResNet18 BP-IEF-MobileNetv3 BP-IEF-Vit BP-IEF-ResNet18 BP-IEF-MobileNetv3 BP-IEF-Vit 13.031 14.062 13.287 4.623 5.535 5.083 9.187 10.155 9.613 3.058 3.680 3.456 5.049 5.408 5.058 2.350 2.707 2.598 3.810 4.182 3.883 1.608 1.856 1.843 (2,512). The V¢ is then inputted into the three serially connected fusion blocks to complete the feature fusion operation. Finally, it is fed into the two FC layers, which calculate the SBP and DBP values for output. 3. Results and discussion In this section, we present a brief overview of the experimental settings. Subsequently, we present and analyze the BP estimation results obtained using our BP-IEF approach. Finally, we compare the performance of our method with existing approaches. 3.1. Experiment settings In section 2, we have provided the details of the dataset. The training set to test set ratio is set at 8:2 for both mixed and non-mixed datasets. Additionally, the sizes of the training set and the test set are 54,000 and 13,500, respectively. For each experiment, we set the number of training epochs to 300. Early stopping (Prechelt 1998) is implemented for every training epoch to save the model weights once the error on the test set begins to increase. To train both the encoder and decoder, a learning rate of 0.001 and a batch size of 512 are employed. The Adam 11 Physiol. Meas. 44 (2023) 125004 Y Liu et al Figure 10. Pearson correlation and Bland–Altman analysis for BP-IEF-ResNet18 on the mixed dataset. optimizer (Kingma and Ba 2014) is used to train all of the models. We utilized two measurements to assess the performance of the proposed BP-IEF: the MAE and the root mean square error (RMSE). The definitions for the two measurements are as follows, where yˆ and y are the estimated and reference BP values, respectively: MAE n 1 å= n = i 1 ∣ ˆ y i - ∣ y i RMSE n 1 å= n = 1 i ( ˆ y i - 2 ) y i ( ) 9 ( ) 10 3.2. BP estimation results We used five types of image encodings (PPG figure, NVG, HVG, GASF and GADF) as inputs in this manuscript. Additionally, we utilized three pre-trained models (resnet18, MobileNetv3, and Vit) in the encoder for image feature extraction. Thus, there are a total of 15 combinations for the mixed and non-mixed datasets, respectively. The performance of these combinations is summarized in table 1. On the non-mixed dataset, the best results in terms of the MAE and RMSE were obtained when utilizing pre-trained ResNet18 with GASF as input. Under this optimal setting, the RMSE and MAE for SBP were 13.155 mmHg and 9.119 mmHg, respectively, while for DBP they were 5.087 mmHg and 3.791 mmHg. On the mixed dataset, the best results in terms of the MAE and RMSE were obtained when using pre-trained ResNet18 with GASF as input. Under this configuration, the RMSE and MAE for SBP were 4.692 mmHg and 3.085 mmHg, respectively, while for DBP they were 2.349 mmHg and 1.608 mmHg. Furthermore, when five image encodings are utilized as input for both the non-mixed and mixed datasets, the results in table 1 reveal that ResNet18 outperforms the other two models in terms of feature extraction. Additionally, GASF yields the best performance when utilized as input for both the non-mixed and mixed datasets. This suggests that the GAF is superior to other image encodings for predicting BP. To verify the effectiveness of image feature fusion, we fed the features derived from encoders into the decoder for both the non-mixed and mixed datasets. Table 2 summarizes their overall performance. When using BP-IEF-ResNet18 on the non-mixed dataset, the best results in terms of MAE and RMSE were obtained. Under this setting, the RMSE and MAE for SBP were 13.031 mmHg and 9.187 mmHg, respectively, while for DBP they were 5.049 mmHg and 3.810 mmHg. Similarly, the use of BP-IEF-ResNet18 yields the best results on the mixed 12 P h y s i o l . M e a s . 4 4 ( 2 0 2 3 ) Table 3. Comparison of our method to the BHS standard. 1 3 Dataset Method „ 5 mmHg „ 10 mmHg „ 15 mmHg Grade „ 5 mmHg „ 10 mmHg „ 15 mmHg Grade SBP DBP Non-mixed Non-mixed Non-mixed Mixed Mixed Mixed BP-IEF-ResNet18 BP-IEF-MobileNetv3 BP-IEF-Vit BP-IEF-ResNet18 BP-IEF-MobileNetv3 BP-IEF-Vit 46.18% 42.41% 39.73% 81.93% 78.66% 76.47% 66.76% 65.33% 63.66% 94.82% 94.11% 92.92% 77.67% 77.23% 75.79% 98.52% 98.09% 97.56% — — — A A A 70.67% 70.48% 67.28% 95.49% 94.16% 93.52% 93.73% 94.16% 92.30% 99.40% 99.27% 99.02% 99.53% 99.46% 99.20% 99.96% 99.95% 99.89% A A A A A A 1 2 5 0 0 4 Y L i u e t a l P h y s i o l . M e a s . 4 4 ( 2 0 2 3 ) Table 4. Performance comparison between the proposed method and prior works. Citation Number of subjects Length of PPG segment Range of SBP/DBP (mmHg) 1 4 BP-IEF-ResNet18 (non-mixed dataset) BP-IEF-ResNet18 (mixed dataset) Wang et al (2021) Wang et al (2021) Baek et al (2019) Leitner et al (2021) Zhong et al (2023) Wang et al (2023) Qin et al (2023) Meng et al (2022) Yan et al (2019) 450 450 169 348 942 100 N/A N/A N/A 50 604 2 s 2 s 3 peaks 3 peaks 4.096 s 5 s 8 s 1 beat 3 s 6 s 10 s [75,165] / [40,80] [75,165] / [40,80] N/A N/A [90,180] / [60,120] [16,180] / [30,110] N/A [86,194] / [30,100] [70,180] / [50,100] [69,190] / [36,102] [80,180] / [50,130] SBP DBP RMSE (mmHg) MAE (mmHg) RMSE (mmHg) MAE (mmHg) 13.031 4.623 7.458 8.46 N/A N/A N/A 5.27 N/A 3.69 N/A 9.187 3.058 4.673 6.17 10.86 3.52 3.49 4.46 5.98 3.21 3.09 5.049 2.350 4.079 5.36 N/A N/A N/A 4.70 N/A 2.23 N/A 3.810 1.608 2.476 3.66 5.95 2.2 2.11 4.20 3.24 1.80 2.11 1 2 5 0 0 4 Y L i u e t a l Physiol. Meas. 44 (2023) 125004 Y Liu et al dataset in terms of MAE and RMSE. Under this setting, the RMSE and MAE for SBP were 4.623 mmHg and 3.058 mmHg, respectively, while for DBP they were 2.350 mmHg and 1.608 mmHg. Additionally, BP-IEF- ResNet18 outperformed the other models on both the non-mixed and mixed datasets, providing further evidence that ResNet18 is better suited for extracting BP-related features. Furthermore, the results in table 2 were often better than those in table 1. This demonstrates the effectiveness of feature fusion in the decoder compared to relying solely on a specific image encoding. The Pearson-R correlation coefficient is a metric utilized to assess the linear correlation between two sets of data (Benesty et al 2009). To evaluate the performance of the proposed BP-IEF, Pearson correlation analyses were conducted between the estimated BP and the reference BP for both the non-mixed and mixed datasets. The Pearson-R coefficients for SBP and DBP in the optimal configuration of the non-mixed dataset were 0.587 and 0.599, respectively. (See figures 9(a) and (b)). Similarly, figures 9(c) and (d) show the Bland–Altman analysis for SBP and DBP, respectively. Under the optimal configuration of the mixed dataset, the Pearson-R coefficients for SBP and DBP were 0.959 and 0.932, respectively (see figures 10(a) and (b)). Similarly, figures 10(c) and (d) show the Bland–Altman analysis for SBP and DBP, respectively. The British Hypertension Society (BHS) (O’Brien et al 2001) and the Association for the Advancement of Medical Instrumentation (AAMI) (Mukkamala et al 2021) standard were used to evaluate the proposed BP-IEF for BP measurement. The BHS standards determine performance levels by calculating the percentage difference between the estimated BP sample and the corresponding reference BP, using thresholds of 5, 10, and 15 mmHg. The comparison of BP-IEF with the BHS standard is summarized in table 3. On the non-mixed dataset, three types of BP-IEF all achieved grade A in DBP prediction. On the mixed dataset, all three types of BP-IEF received grade A for both SBP and DBP prediction. The AAMI standard for BP measurement requires that the mean error (ME) and the standard deviation of error (SDE) between the estimated and reference BPs must be within ± 5 mmHg and 8 mmHg, respectively. On the non-mixed dataset, the BP-IEF-ResNet18 approach achieved an ME and SDE of −1.05 ± 12.989 mmHg and −0.665 ± 5.005 for SBP and DBP, respectively. On the mixed dataset, the BP-IEF-ResNet18 approach achieved an ME and SDE of 0.068 ± 4.543 mmHg and 0.036 ± 2.35 for SBP and DBP, respectively. On the non-mixed dataset, the performance of DBP meets the AAMI standard. On the mixed dataset, both SBP and DBP match the AAMI standard, with an ME close to 0 mmHg. Based on the above results, we can see that our proposed BP-IEF is effective in the BP estimation task. This further proves that the image perspective helps to extract features associated with BP. Moreover, the results obtained from mixed datasets are usually better than those from non-mixed datasets. The variations in accuracy can be attributed to the differences in cardiovascular relationships between PPG and BP among individuals. 3.3. Performance comparison To further validate the effectiveness of our proposed BP-IEF, we conducted a comprehensive comparison with numerous existing PPG-based BP estimation methods. Table 4 compares our proposed method’s optimal settings with recent studies on PPG-based BP estimation published within the last five years. The comparison is based on the length of the PPG segment, the number of subjects, the range of SBP/DBP, and the estimation performance. However, due to differences in the datasets across studies, a fair quantitative comparison is not viable. In what follows, we discuss the possible loopholes of AI methods in PPG-based cuffless BP measurement. The data in table 4 indicate that longer PPG segments tend to result in an improved estimation performance. This may be attributed to the presence of more time-dependent relationships within longer PPG segments, thereby facilitating network modeling. However, our method converts PPG to 2D images for BP estimation, utilizing shorter PPG segments to achieve a comparable performance. Moreover, wider intervals of BP range result in a reduced performance. This may be due to the fact that the network considers higher BP values as outliers, impairing network convergence and potentially affecting the performance. Furthermore, increasing the number of patients did not result in an improved performance, likely attributed to the substantial variations in BP distribution among patients, which do not provide the network with additional valuable information to enhance its generalization capabilities. 4. Conclusion In this manuscript, an end-to-end BP estimation method named BP-IEF is proposed. The proposed BP-IEF consists of two main components: the encoder and the decoder. To accomplish this, five image encodings, namely the PPG figure, NVG, HVG, GASF, and GADF, were employed as inputs. Firstly, we employed five image encodings to transform PPG signals from the 1D time domain into 2D images. Subsequently, these encodings were utilized as inputs for encoders composed of three distinct neural networks, enabling the extraction of high-dimensional features. Following this, the extracted features from each encoding were 15 Physiol. Meas. 44 (2023) 125004 Y Liu et al inputted into the decoder to fuse the high-dimensional features effectively. Finally, the decoder outputs a pair of BP values, SBP and DBP, thus enabling the end-to-end estimation of BP. The numerical results show that our method is very promising. Considering realistic scenarios where BP data for target patients is usually not available, in future research we will focus on improving the generalization capability of our method to achieve a better BP estimation performance on non-mixed datasets. Acknowledgments This work was supported by the BUPT Excellent Ph.D. Students Foundation grant no. CX2022116. Data availability statement The data cannot be made publicly available upon publication because they are not available in a format that is sufficiently accessible or reusable by other researchers. The data that support the findings of this study are available upon reasonable request from the authors. ORCID iDs Yinsong Liu https://orcid.org/0000-0003-1073-3147 References Baek S, Jang J and Yoon S 2019 End-to-end BP prediction via fully convolutional networks IEEE Access 7 185458–68 Benesty J, Chen J, Huang Y and Cohen I 2009 Pearson correlation coefficient Noise reduction in speech processing (Springer) pp 1–4 Chen S, Gallagher M J, Hogg F, Papadopoulos M C and Saadoun S 2018 Visibility graph analysis of intraspinal pressure signal predicts functional outcome in spinal cord injured patients J. Neurotrauma 35 2947–56 Chua C P and Heneghan C 2006 Continuous BP monitoring using ecg and finger photoplethysmogram 2006 International Conference of the IEEE Engineering in Medicine and Biology Society, IEEE pp 5117–20 Dai P-F, Xiong X and Zhou W-X 2019 Visibility graph analysis of economy policy uncertainty indices Physica A 531 121748 Deng J 2009 A large-scale hierarchical image database Proc. of IEEE Computer Vision and Pattern Recognition 2009 Dosovitskiy A et al 2020 An image is worth 16 × 16 words: Transformers for image recognition at scale arXiv:2010.11929 Duan K, Qian Z, Atef M and Wang G 2016 A feature exploration methodology for learning based cuffless BP measurement using photoplethysmography 2016 38 Annual International Conference of The IEEE Engineering In Medicine and Biology Society (EMBC) IEEE pp 6385–8 Elgendi M, Fletcher R, Liang Y, Howard N, Lovell N H, Abbott D, Lim K and Ward R 2019 The use of photoplethysmography for assessing hypertension NPJ Digit. Med. 2 1–11 Elgendi M, Norton I, Brearley M, Abbott D and Schuurmans D 2013 Systolic peak detection in acceleration photoplethysmograms measured from emergency responders in tropical conditions PLoS One 8 e76585 Eom H, Lee D, Han S, Hariyani Y S, Lim Y, Sohn I, Park K and Park C 2020 End-to-end deep learning architecture for continuous BP estimation using attention mechanism Sensors 20 2338 Goldberger A L, Amaral L A, Glass L, Hausdorff J M, Ivanov P C, Mark R G, Mietus J E, Moody G B, Peng C-K and Stanley H E 2000 Physiobank, physiotoolkit, and physionet: components of a new research resource for complex physiologic signals Circulation 101 e215–20 Griffin M J, Letson H L and Dobson G P 2014 Adenosine, lidocaine and mg2+ (alm) induces a reversible hypotensive state, reduces lung edema and prevents coagulopathy in the rat model of polymicrobial sepsis J. Trauma Acute Care Surg. 77 471–8 He K, Zhang X, Ren S and Sun J 2016 Deep residual learning for image recognition Proceedings of The IEEE Conference On Computer Vision and Pattern Recognition 770–8 Hosanee M et al 2020 Cuffless single-site photoplethysmography for BP monitoring J. Clin. Med. 9 723 Howard A et al 2019 Searching for mobilenetv3 Proceedings Of The IEEE/CVF International Conference on Computer Vision pp 1314–24 Hu Q, Wang D and Yang C 2022 Ppg-based BP estimation can benefit from scalable multi-scale fusion neural networks and multi-task learning Biomed. Signal Process. Control 78 103891 Ji H, Xu T, Wu W and Wang J 2016 Visibility graph analysis on eeg signal 2016 IX International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI) IEEE 1557–61 Kachuee M, Kiani M M, Mohammadzade H and Shabany M 2015 Cuff-less high-accuracy calibration-free BP estimation using pulse transit time 2015 IEEE International Symposium on Circuits and Systems (ISCAS) IEEE pp 1006–9 Kaniusas E et al 2006 Method for continuous nondisturbing monitoring of BP by magnetoelastic skin curvature sensor and ecg IEEE Sensors J. 6 819–28 Kingma D P and Ba J 2014 Adam: A method for stochastic optimization arXiv: 1412.6980 Kurylyak Y, Lamonaca F and Grimaldi D 2013 A neural network-based method for continuous BP estimation from a PPG signal 2013 IEEE International Instrumentation And Measurement Technology Conference (I2MTC) IEEE pp 280–3 Lacasa L, Luque B, Ballesteros F, Luque J and Nuno J C 2008 From time series to complex networks: the visibility graph Proc. Natl. Acad. Sci. 105 4972–5 Leitner J, Chiang P-H and Dey S 2021 Personalized BP estimation using photoplethysmography: a transfer learning approach IEEE J. Biomed. Health Inform. 26 218–28 Li D, Lin J, Bissyande T F D A, Klein J and Traon Y Le 2018 Extracting statistical graph features for accurate and efficient time series classification XXI International Conference on Extending Database Technology 16 Physiol. Meas. 44 (2023) 125004 Y Liu et al Luque B, Lacasa L, Ballesteros F and Luque J 2009 Horizontal visibility graphs: Exact results for random time series Phys. Rev. E 80 046103 Ma Y et al 2018 Relation between BP and pulse wave velocity for human arteries Proceedings of the National Academy of Sciences 115 pp 11 144–11 149 Meng Z, Yang X, Liu X, Wang D and Han X 2022 Non-invasive BP estimation combining deep neural networks with pre-training and partial fine-tuning Physiol. Meas. 43 11NT01 Mou H and Yu J 2021 Cnn-lstm prediction method for BP based on pulse wave Electronics 10 1664 Mukkamala R, Yavarimanesh M, Natarajan K, Hahn J-O, Kyriakoulis K G, Avolio A P and Stergiou G S 2021 Evaluation of the accuracy of cuffless BP measurement devices: challenges and proposals Hypertension 78 1161–7 O’brien E, Waeber B, Parati G, Staessen J and Myers M G 2001 BP measuring devices: recommendations of the european society of hypertension Bmj 322 531–6 Paviglianiti A, Randazzo V, Villata S, Cirrincione G and Pasero E 2021 A comparison of deep learning techniques for arterial BP prediction Cognitive Computation 14 1689–1710 Prechelt L 1998 Early stopping-but when? Neural Networks: Tricks of the trade (Springer) pp 55–69 Qin C, Li Y, Liu C and Ma X 2023 Cuff-less BP prediction based on photoplethysmography and modified resnet Bioengineering 10 400 Rundo F, Ortis A, Battiato S and Conoci S 2018 Advanced bio-inspired system for noninvasive cuff-less BP estimation from physiological signal analysis Computation 6 46 Salem M, Taheri S and Yuan J-S 2018 Ecg arrhythmia classification using transfer learning from 2-dimensional deep cnn features 2018 IEEE Biomedical Circuits and Systems Conference (BioCAS) IEEE pp 1–4 Schrumpf F, Frenzel P, Aust C, Osterhoff G and Fuchs M 2021 Assessment of non-invasive BP prediction from ppg and rppg signals using deep learning Sensors 21 6022 Shao Z-G 2010 Network analysis of human heartbeat dynamics Appl. Phys. Lett. 96 073703 Stephen M, Gu C and Yang H 2015 Visibility graph based time series analysis PLoS One 10 e0143015 Wang L-H, Sun K-K, Xie C-X, Fan M-H, Abu P A R and Huang P-C 2023 Cuffless BP estimation using dual physiological signal and its morphological features IEEE Sensors J. 23 11956–67 Wang W, Mohseni P, Kilgore K and Najafizadeh L 2021 Cuff-less BP estimation via small convolutional neural networks 2021 XLIII Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC) IEEE pp 1031–4 Wang W, Mohseni P, Kilgore K L and Najafizadeh L 2021 Cuff-less BP estimation from photoplethysmography via visibility graph and transfer learning IEEE J. Biomed. Health Inform. 26 2075–85 Wang Z et al 2015 Encoding time series as images for visual inspection and classification using tiled convolutional neural networks Workshops at the Twenty-IX AAAI Conference on Artificial Intelligence (Menlo Park, CA, USA: AAAI) vol 1 Wu D, Xu L, Zhang R, Zhang H, Ren L and Zhang Y-T 2018 Continuous cuff-less BP estimation based on combined information using deep learning approach J. Med. Imaging & Health Infor. 8 1290–9 Wu Y, Yang F, Liu Y, Zha X and Yuan S 2018 A comparison of 1-d and 2-d deep convolutional neural networks in ecg classification arXiv:1810.07088 Yan C, Li Z, Zhao W, Hu J, Jia D, Wang H and You T 2019 Novel deep convolutional neural network for cuff-less BP measurement using ecg and ppg signals 2019 XLI Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) IEEE pp 1917–20 Yang S, Zaki W S W, Morgan S P, Cho S-Y, Correia R and Zhang Y 2020 BP estimation with complexity features from electrocardiogram and photoplethysmogram signals Opt. Quantum Electron. 52 1–16 Zhang G, Gao M, Xu D, Olivier N B and Mukkamala R 2011 Pulse arrival time is not an adequate surrogate for pulse transit time as a marker of BP J. Appl. Physiol. 111 1681–6 Zhong Y, Chen Y, Zhang D, Xu Y and Karimi H R 2023 A mixed attention-gated u-net for continuous cuffless BP estimation Signal Image Video Process 1–9 17
null
10.1007_s00122-023-04372-4.pdf
Data availability Electronic supplementary material The online version of this article (https:// doi. org/ xxxxx) contains supplementary material, which is available to authorized users.
Data availability Electronic supplementary material The online version of this article ( https:// doi. org/ xxxxx ) contains supplementary material, which is available to authorized users.
Theoretical and Applied Genetics (2023) 136:138 https://doi.org/10.1007/s00122-023-04372-4 ORIGINAL ARTICLE The putative vacuolar processing enzyme gene TaVPE3cB is a candidate gene for wheat stem pith‑thickness Qier Liu1,4 · Yun Zhao1,2 · Shanjida Rahman1 · Maoyun She1 · Jingjuan Zhang1 · Rongchang Yang1 · Shahidul Islam1 · Graham O’Hara1 · Rajeev K. Varshney1 · Hang Liu1 · Hongxiang Ma4 · Wujun Ma1,3 Received: 25 October 2022 / Accepted: 27 April 2023 / Published online: 26 May 2023 © The Author(s) 2023 Abstract Key message The vacuolar processing enzyme gene TaVPE3cB is identified as a candidate gene for a QTL of wheat pith-thickness on chromosome 3B by BSR-seq and differential expression analyses. Abstract The high pith-thickness (PT) of the wheat stem could greatly enhance stem mechanical strength, especially the basal internodes which support the heavier upper part, such as upper stems, leaves and spikes. A QTL for PT in wheat was previously discovered on 3BL in a double haploid population of ‘Westonia’ × ‘Kauz’. Here, a bulked segregant RNA-seq analysis was applied to identify candidate genes and develop associated SNP markers for PT. In this study, we aimed at screening differentially expressed genes (DEGs) and SNPs in the 3BL QTL interval. Sixteen DEGs were obtained based on BSR-seq and differential expression analyses. Twenty-four high-probability SNPs in eight genes were identified by compar- ing the allelic polymorphism in mRNA sequences between the high PT and low PT samples. Among them, six genes were confirmed to be associated with PT by qRT-PCR and sequencing. A putative vacuolar processing enzyme gene TaVPE3cB was screened out as a potential PT candidate gene in Australian wheat ‘Westonia’. A robust SNP marker associated with TaVPE3cB was developed, which can assist in the introgression of TaVPE3cB.b in wheat breeding programs. In addition, we also discussed the function of other DEGs which may be related to pith development and programmed cell death (PCD). A five-level hierarchical regulation mechanism of stem pith PCD in wheat was proposed. Introduction Wheat is the most widely grown crop in the world, account- ing for 220 million hectares with annual global production of ∼772 million tonnes (FAOSTAT 2022). By 2050, global demand for wheat is predicted to grow sharply as the world’s Qier Liu and Yun Zhao authors contributed equally to this work. * Wujun Ma w.ma@murdoch.edu.au 1 Centre for Crop and Food Innovation, Food Futures Institute and College of Science, Health, Engineering and Education, Murdoch University, Perth, WA 6150, Australia 2 Institute of Cereal and Oil Crops, Hebei Academy of Agriculture and Forestry Sciences, Shijiazhuang 050035, People’s Republic of China 3 College of Agronomy, Qingdao Agriculture University, Qingdao 266109, People’s Republic of China 4 Provincial Key Laboratory of Agrobiology, and Institute of Food Crops, Jiangsu Academy of Agricultural Sciences, Nanjing 210014, People’s Republic of China population is expected to exceed 9 billion (Keating et al. 2014). The Green Revolution led to tremendous increases in wheat yield by providing excellent growing conditions and improving crop varieties. With advances in molecular genet- ics technology, some yield-related genes that control plant height and tiller number have been cloned, such as wheat reduced-height genes (Rht) (Appleford et al. 2007) and semi- dwarf gene (Sd1) (Monna et al. 2002). The semi-dwarf cul- tivars are inherently more stable mechanically, reducing the leverage on the stem base and anchorage system in wheat, thereby increasing the lodging resistance under nitrogen application and achieving maximum yield potential (Hed- den 2003). However, severe dwarfism causes inadequate bio- mass accumulation, eventually, lower yield potential (Hirano et al. 2017). Therefore, breeding wheat varieties with strong stem phenotypes is a breeding strategy for enhancing lodg- ing resistance and yield (Reynolds et al. 2010). The wheat stem plays an important role in providing mechanical support for leaves and spikes (Kirby 2002), transporting water and mineral nutrients, storing water- soluble carbohydrates (WSC) and starch (Scofield et al. Vol.:(0123456789)1 3 138 Page 2 of 21 Theoretical and Applied Genetics (2023) 136:138 2009), and remobilizing nutrients during grain filling (Blum 1998). Stem lodging is mainly occurring at the 2nd internode (Peng et al. 2014), and the density of the basal 2nd internode has been proven to correlate with stem mechanical strength (Li et al. 2022). The central part of the young stem is occupied by pith tissues, which are com- posed of undifferentiated parenchyma cells. Parenchyma- tous pith cells store a large amount of water and WSCs, such as sucrose, glucose, fructose and fructan (Ruuska et al. 2006). In mature wheat stems, the majority of pith cells die and collapse, which leads to the formation of a central cavity and hollow stem. The death of the pith cells has been regarded as programmed cell death (PCD), but the molecular mechanism of pith death remains unex- plained (Fujimoto et al. 2018). Stem pith thickness is an important agronomic trait of durum and bread wheat that provides resistance to the wheat pest (Hayat et al. 1995), lodging (Kong et al. 2013) and drought (Saint Pierre et al. 2010). Adopting forward genetic strategies, many stem- strength-related QTLs have been identified on 1A and 2D for culm wall thickness (Hai et al. 2005; Liu et al. 2017; Pan et al. 2017); 3B and 2D for culm diameter (Hai et al. 2005; Song et al. 2021); 1A, 3A and 4B for stem internode length (Berry et al. 2008; Piñera‐Chavez et al., 2021); 1B, 2D, 3A, 3B, 4B and 4D for stem internode wall width (Berry et al. 2008; Piñera‐Chavez et al., 2021; Verma et al. 2005). The major genetic factor related to stem solidness has been mapped on chromosome 3B in durum (SSt1) and bread wheat (Qss.msub-3BL), conferring solid stems with thick sclerenchyma tissues and a strong culm phenotype (Cook et al. 2004; Nilsen et al. 2017). Recently, a putative Dof transcription factor, TdDof, was cloned as the SSt1 causal gene (Nilsen et al. 2020). BSR-seq approach that combines bulked segreant analysis (BSA) with RNA sequencing provides an efficient method to rapidly identify candidate genes of QTLs. It uses RNA sequencing data to call SNPs and filter out SNPs linked to the candidate genomic region through BSA, thus the hot spot region of genetic variation associated with the pheno- type could be identified (Liu et al. 2012). In addition, RNA- seq reveals DEGs between two bulked sample pools in the mapping interval and provides the necessary information for gene screening. Recently, several genes have been identi- fied through BSR-seq in different plant species, including the genes related to powdery mildew resistance (Xie et al. 2020; Zhan et al. 2021), pest resistance (Hao et al. 2019), male sterile (Tan et al. 2019) and waterlogging-tolerance (Du et al. 2017). The objectives of this study were to: (i) identify genome- wide mRNA variants related to PT through BSR-seq; (ii) determine the physical location of Qpt-3B through BSR-seq; (iv) identify the candidate gene for Qpt-3B; (v) develop SNP marker linked to Qpt-3B for marker-assisted selection. Materials and methods The overall experimental procesure is outlined in Fig. 1. Plant materials and growth conditions A doubled haploid (DH) population ‘Westonia’ (high PT) x ‘Kauz’ (low PT) with 225 lines were used for the PT can- didate gene identification (Butler et al. 2005; Rajaram et al. 2002; Zhang et al. 2013). A set of Australian historical wheat cultivar collections (171 varieties) spanning approximately 125 years (1890–2015) was selected for marker validation (Table S1). The genetic resource information can be found in the CIMMYT-Wheat Germplasm Bank (https:// wgb. cimmyt. org/ gring lobal/ search). The DH population and historical cultivars were grown repeatedly in a glasshouse at Murdoch University, Western Australia, Australia, from 2018 to 2020. Pots were placed following a complete randomized block design (RCBD) and three seeds from each cultivar were planted in a 4 L free- draining pot filled with soil mix. Plants were grown under controlled temperature with 25/15 °C (day/night) and sun- light conditions and equipped with an automatic watering system. Evaluation of stem pith thickness Pith-thickness data of each line was obtained at the fully matured stage (Pluta et al. 2021) by evaluating the average rating of the stem: the main stem was cross-sectioned in the center of the second basal internode, and the stem diameter and pith thickness were measured by the Vernier caliper with three biological replicates. The pith filling of internode was rated using a five-grade system according to the methodol- ogy developed by PAUW and Read (1982). Pith-thickness index was calculated with the formula: PI = (2 × pith thick- ness)/stem diameter (Wallace et al. 1973). RNA isolation, library construction and sequencing The selected DH lines (listed in Table S2) and two parental lines were used for RNA-seq analysis. For each line, approx- imately 0.5 cm of the middle of the second basal internode of main stem was sampled at the early internode elonga- tion stage (Z32) (Zadoks et al. 1974). RNA was extracted using the TRIzol reagent (Invitrogen Canada, catalogue No. 15596026) and subsequently treated with Qiagen DNase set (Catalog No. 79254) to remove genomic DNA. For bulked sample pooling, equal amounts of RNA from each of the 20 selected DH lines were mixed to construct solid bulk (Sbulk, the high PT lines) or hollow bulk (Hbulk, the low PT lines). 1 3 Theoretical and Applied Genetics (2023) 136:138 Page 3 of 21 138 Fig. 1 Schematic flowchart of the experimental procedure. W: ‘Westonia’; K: ‘Kauz’; S: Solid bulks; H: Hollow bulks Two parents and two bulked samples with three replicates each were prepared and submitted to Singapore Novogene company for sequencing. SNP calling and ΔSNP‑index analysis The data were ordered and assembled using SAMtools v-1.14 (http:// www. htslib. org/ downl oad/). The sequenc- ing data of three biological replicates per sample were analyzed together. The initial SNP calling was performed using Genome Analysis Toolkit (GATK, v4.2.3.0) package (McKenna et al. 2010), and SnpEff was used for SNP anno- tation (Cingolani et al. 2012). The high-quality SNPs were filtered according to Liu et al. (2012) with the following criteria: sequencing depth for each SNP ≥ 5; Quality of vari- ation detection ≥ 50; the minimum quality score of 20, and only homozygous SNPs between parental lines were used for SNP-index analysis. After filtration, for each genomic position, the SNP-index of two bulks were estimated using a MutMap method, with SNPs in ‘Kauz’ as a reference. SNP- index calculates the proportion of short reads that cover a particular site sharing an SNP (Abe et al. 2012). Then, the ΔSNP index was calculated by subtracting the SNP index of the Hbulk from the Sbulk (Takagi et al. 2013). The average value of ΔSNP index in the corresponding window was plot- ted by calculating in a 5 Mb window size and 50 kb window step size. PT-associated loci were identified when the fitted values of ΔSNP index were higher than the 99% confidence threshold. A Circos graph (Krzywinski et al. 2009) includ- ing chromosomes, genes and SNP density was generated by CIRCOS software (http:// circos. ca/). Differentially expressed gene analysis The high-quality reads were mapped against the latest Chinese spring genome (IWGSC RefSeq v2.1) using the HISAT2 software (Kim et al. 2019), and the expression level was calculated with fragments per kilobase of tran- script per million fragments mapped (FPKM) (Trapnell et al. 2010). The fold change was calculated based on the normal- ized expression values between the high PT sample and the low PT sample. Genes with more than two-fold differential 1 3 138 Page 4 of 21 Theoretical and Applied Genetics (2023) 136:138 expression (|fold change| ≧ 2) and false discovery rate (FDR) < 0.001 for the groups of ‘Westonia vs Kauz’ and ‘Sbulk vs Hbulk’ were classified as significant DEGs. Only the DEGs coexisting between parent comparison and bulks comparison group were considered as pith-thickness-related genes. Then, those coexisting DEGs were classified into two types, Hcluster and Scluster. Hcluster contains genes which were highly expressed in low PT samples, while Scluster consists of the genes which were highly expressed in high PT samples. the normalization of gene expression studies (Wang et al. 2013). The processing for the 3-step cycling qRT-PCR was as follows: 95 °C for 2 min, followed by 40 cycles of 95 °C for 5 s, 60 °C for 15 s, 72 °C for 15 s. Reac- tion specificities were assessed by melting curve analysis. Gene relative expression level was calculated using 2−ΔΔCt method with three technical repeats (Livak and Schmittgen 2001). A one-way ANOVA followed by a Tukey’s test was performed to identify significant differences. Go and KEGG pathway analyses Cloning, sequencing and phylogenetic analysis Gene ontology (GO) and KEGG pathway enrichment analy- sis of the DEGs was performed according to the method described by Hao et al. (2019). The GO enrichment analysis was performed using an R package based on hypergeometric distribution test to find the significantly enriched terms in DEGs. For KEGG pathway enrichment analysis, the meta- bolic pathway annotation was performed using KOBAS soft- ware against the KEGG database (http:// www. genome. jp/ kegg/). GO terms and KEGG pathway with FDR corrected p value ≤ 0.05 was regarded as significantly enriched. NBT staining The production of ROS in stems was detected by nitro- tetrazolium blue chloride (NBT) staining as described by Wohlgemuth et al. (2002) with minor modifications. Stems at three different stages (Z30, Z32 and Z65) were harvested, and immersed in 50 mm PBS buffer (pH 7.8) containing 0.1 mg  ml−1 NBT and 10 mm sodium azide. Samples were vacuum-infiltrated for 2 min, and subsequently incubated at 25 °C for 2 h in the darkness and then the stained samples were immersed in 80% (v/v) ethanol for 1 h to remove the chlorophyll. ROS production was visualised as a dark blue formazan deposit in stem tissues. qRT‑PCR validation qRT-PCR was performed to evaluate the reliability of the sequencing results and reveal expression profiles of DEGs. For the evaluation of sequencing results, the same RNA samples were used for qRT-PCR as for RNA-seq. In addi- tion, the stems on three different Zadoks stages (Z30, Z32, Z65) and the leaves at the stage of Z32 were collected from parents. For TaVPEcB gene expression analysis of Chinese spring and the selected historical lines, the stems were collected on the stage of Z32 only. The first strand cDNA was synthesised using the SensiFAST cDNA Syn- thesis Kit (Bioline, UK) and the qRT-PCR amplification was performed using SensiFAST SYBR No-ROX Kit (Bio- line, UK). Taactin was used as an internal control gene for The synthesized cDNA, gDNA and genome-specific primers (Table  S3) were used for the amplification of full-length CDS and promoter region in both parents. The PCR reaction was conducted by using Q5 High-fidelity DNA Polymerase (NEB) according to product instructions. The target fragments were separated and purified using a Gel Extraction Kit (Promega). Then, the purified prod- ucts were amplified using BigDye Terminator V3.1 Cycle Sequencing Kits (Applied Biosystems) and sequenced by Applied Biosystems 3730 DNA Analyzers. Protein sequences of VPE family of Arabidopsis, brach- ypodium, rice and wheat were gathered from the published database, and corresponding accession numbers of used sequences are provided in Table S4. The alignment of pro- tein sequences was performed by the ClustalX program. The phylogenetic tree was constructed by the neighbour- joining method with 1000 bootstraps in the MEGA11 soft- ware (Tamura et al. 2007). The sequence alignment and the GC content were analyzed using DNAMAN software, and their cis-acting elements were predicted by PLACE and PlantCARE. Maker development and linkage map construction Co-dominant markers were designed based on the SNPs within the candidate DEGs and genotyped the 225 DH lines for being mapped onto an existing linkage map. PCR reactions were carried out in 10 μL reaction mixture consisting of GoTaq® Green Master Mix 2X (Promega), primer sets (0.5 µM), and genomic DNA (50 ng). The pro- cedure of PCR was as follows: 95 °C for 2 min; 30 cycles of 95 °C for 10 s, and 56 °C for 15 s, 72 °C for 1 min; 72 °C for 7 min. Based on our previous mapping results of major pith thickness QTL on 3BL (Zhao 2019), the developed makers as well as the previous makers were used for a new linkage map construction with the soft- ware IciMapping software V4.1 (Meng et al. 2015). The graphical presentation of linkage maps and QTLs were conducted by MapChart V2.3.2. 1 3 Theoretical and Applied Genetics (2023) 136:138 Page 5 of 21 138 Results Pith‑thickness evaluation of wheat lines The wheat stem pith-thickness index scores varied from 0.19 to 1 (p < 0.05) in the DH population. Its distribution pattern was similar to a bimodal distribution (Fig. S1), con- sistent with single gene inheritance for stem pith thickness. Wheat parent ‘Westonia’ was classified as high PT stem (PI = 0.55~0.61) and ‘Kauz’ was classified as low PT stem (PI = 0.22~0.27) across three test years. PT index for solid bulks (Sbulks) ranged from 1.00 to 0.75, indicating that these bulked samples belong to completely solid or high PT grade, while PT index for hollow bulks (Hbulks) ranged from 0.19 to 0.25, belonging to hollow or low PT grade (Fig. 2 and Table S2). Differentially expressed gene identification and GO/ KEGG pathway analysis The transcript profiles of solid and hollow stem pools were built by comparing their gene expression levels based on FPKM (fragments per kilobase of transcript per million fragments mapped). A total of 20,493 DEGs were revealed among ‘Westonia’ versus ‘Kauz’. Among them, 8209 genes were highly expressed in Westonia with 878 genes up-regu- lated by over 50 folds. The number of highly expressed genes in Kauz was 12,284, among which 709 genes were down- regulated by over 50 folds. However, only 5453 DEGs were identified between Sbulk versus Hbulk. Among them, 1613 genes were up-regulated and 3840 were down-regulated, with 93 and 256 genes up-regulated and down-regulated by 20–50 folds, respectively. In addition, in this comparison group, no DEGs with more than 50-fold difference has been found. The number of DEGs in the two comparison groups differed significantly, but 2424 common DEGs were identi- fied (Fig. 3A). Among the coexisting DEGs, 765 genes were classified as hollow cluster (Hcluster) genes which were highly expressed in all low PT samples (Fig. 3B) and 213 genes were classified as solid cluster (Scluster) genes which were highly expressed in all high PT samples. Results from GO analysis on the DEGs showed that in the biological process, Scluster DEGs were mainly enriched in cellular carbohydrate metabolic process, protein transport and ATP biosynthetic process (Fig. 3C). However, Hclus- ter DEGs were mainly involved in the protein modification process, response to oxidative stress, metal ion transport and cell wall organization or biogenesis. In addition, the top enriched molecule functions in Scluster were associated with cytoskeletal protein binding, ATP binding, hydrolase activity and transferase activity (such as glucosyltransferase and O-methyltransferase activity; while the most enriched GO terms in Hcluster were ATP binding, oxidoreductase activity, peroxidase activity and endopeptidase activity. The most enriched cellular components in Scluster belonged to the membrane protein complex, endomembrane system and vesicle membrane; while in Hcluster, the extracellular region and cell wall were significantly enriched. Fig. 2 Morphological characteristics of different wheat genotypes. A and C: High pith-thickness stems ‘Westonia’ and DH line 209. B and D: hollow stem internodes of Kauz and DH line 156. Numbers 1, 2, 3 and 4 are the second, third, fourth and fifth stem internodes (from the bottom to top), respectively. E and G: Wiesner staining of the 2nd basal internode stem sections of ‘Kauz’; F and H: Wiesner staining of the 2nd basal internode stem sections of ‘Westonia’. P: pith; SV: small vascular bundle; LV: large vascular bundle 1 3 138 Page 6 of 21 Theoretical and Applied Genetics (2023) 136:138 1 3 Theoretical and Applied Genetics (2023) 136:138 Page 7 of 21 138 ◂ Fig. 3 Transcriptional changes in solid stem samples (Westonia and Sbulks) and hollow stem samples (Kauz and Hbulks). A Venn dia- gram showing a total of 2424 coexisting in the comparison of ‘Westo- nia vs Kauz’ and ‘Sbulks vs Hbulks’; B Heat map showing the DEGs with the same expression profiles. Gene expression was normalized and transformed by log10 (FPKM + 1) values. Red and Green lines represent genes showing high and low expression levels, respectively. Hcluster: hollow cluster; Scluster: solid cluster; C GO term enrich- ment analysis of DEGs in two clusters; D Enriched KEGG pathway scatterplots for DEGs in two cluster (colour figure online) KEGG pathway analysis showed that the significantly enriched metabolic pathways in the Scluster including carbon metabolism, biosynthesis of amino acids and pro- tein processing (Fig. 3D). While in the Hcluster, four sig- nificantly enriched pathways were found, including the plant-pathogen interaction pathway, plant hormone signal transduction, phenylpropanoid biosynthesis and MAPK signalling pathway. Among them, phenylpropanoid biosyn- thesis is an important metabolic pathway to scavenge the over-accumulated reactive oxygen species (ROS) which may cause oxidative damage to proteins, lipids, and DNA, ulti- mately resulting in PCD (Sharma et al. 2012). Histochemical detection of superoxide anion accumulation during stem development As reflected by the GO and KEGG analyses, significantly enriched DEGs were found related to oxidative stress and ROS scavenging pathway in low PT samples. We suspected that ROS metabolism of which might be more active than that in high PT samples. This was consistent with the ROS accumulation in stems detected by NBT staining. At Z30 stage, the dark blue deposit was found in the pith cells and xylem vessel elements in the stem of ‘Kauz’ (Fig. 4A), but no obvious staining was found in ‘Westonia’ (Fig.  4D). At Z32 stage, only xylem vessel elements of ‘Kauz’ were intensely stained by NBT (Fig. 4B), while, at Z65 stage, neither ‘Kauz’ nor ‘Westonia’ showed strong staining signals (Fig. 4C, D). This result confirmed the accumulation of ROS in ‘Kauz’ was higher than that in ‘Westonia’, and the active ROS metabolism in ‘Kauz’ may be involved in pith PCD, as ROS can induce cell death. SNP calling and DEG discovery via BSR‑seq We further identified 72,301 expressed genes from four sequencing libraries and 352,388 SNPs were called through GATK in total. The genome-wide SNPs distribution was shown in Fig. 5A. After filtering, 11,331 high-quality SNPs were obtained (Table S5). The average density of SNPs on all chromosomes was 0.74 SNPs per Mb, with the high- est density on chromosome 5B (1.47 SNPs per Mb) and the lowest density on chromosome 4D (0.11 SNPs per Mb) (Table S5). The sequencing depth of the four samples ranged from 6.84 × to 9.93 × ; 2275 of the SNPs were non- synonymous. Under the threshold of 99% confidence, only one putative candidate region has been revealed (Table S6, Fig.  5B). This region contains the Qpt-3B QTL with a genomic size of 6.83 Mb (819, 897, 386–826, 725, 912 bp). Using ∆SNP > 0.74 as a threshold (Hao et al. 2019), a total of 25 SNPs with high confidence were identified in this region, and 24 SNPs were in the exon region (Table S7). The variants in the Qpt-3B QTL region were examined using the Integrative Genomics Viewer (IGV) (Thorvaldsdóttir et al. 2013). One gene (TraesCS3B02G597900) showed consistent frequencies of DNA variants with corresponding parents at about 0% in Hbulks and 100% in Sbulks (Fig. S2). By searching the candidate region and adjacent region, a total of 143 high confidence genes were included, among which 44 genes were detected in at least one comparison group through RNA-seq. The expression analysis revealed that all these expressed genes could be divided into four categories. The first two categories include the genes only expressed in one of the samples. In the other two catego- ries, genes are expressed in both samples but up-regulated in one of the samples (Table S8 and Table S9). We found 16 coexisting DEGs with the same expression pattern in two comparison groups. Among them, 12 DEGs were highly expressed (> twofold) in both ‘Kauz’ and Hbulk, while 4 DEGs (TraesCS3B02G597800, TraesCS3B02G597900, TraesCS3B02G603900 and TraesCS3B02G608500) were highly expressed in both ‘Westonia’ and Sbulk (Fig. 5C). These four genes have potentially deleterious SNPs and Indels related to high PT phenotype (Table S7). The expres- sion levels of TraesCS3B02G597800, TraesCS3B02G597900 and TraesCS3B02G603900 were higher in low PT samples ‘Kauz’ and ‘Hbulk’, while TraesCS3B02G608500 was higher in high PT samples ‘Westonia’ and ‘Sbulk’. qRT‑PCR Twelve DEGs in the 3BL candidate region were validated through real-time qRT-PCR to verify the authenticity of RNA-seq results. The designed primer set was listed (Table S3). The qRT-PCR results confirmed the direction of regulation (positive or negative) between high and low PT samples for selected genes. The log2 fold-change (log2FC) value was also similar for the majority of genes, with a cor- relation coefficient of 0.85 and 0.68 between RNA-seq and qRT-PCR data sets derived from ‘Westonia vs Kauz’ and ‘Sbulks vs Hbulks’, respectively (Fig. 6A). In addition, six genes were selected for further investiga- tion of gene expression profiles at three stem developmen- tal stages in two different tissues (Fig. 6B). Among them, four DEGs with selected based on SNPs, while the other two DEGs were selected based on gene annotation results. 1 3 138 Page 8 of 21 Theoretical and Applied Genetics (2023) 136:138 Fig. 4 NBT staining for ROS in wheat stems at Z30, Z32 and Z65 stage. NBT reacts with ROS to form a dark blue insoluble formazan compound. ‘Kauz’ stem section A–C, ‘Westonia’ stem section (D–F). V, xylem vessel elements; P, pith parenchyma cells. The staining dif- ferences are mark by arrows (colour figure online) TraesCS3B02G608800 was annotated as a Dof transcription factor, which might be involved in regulating stem pith cell apoptosis. TraesCS3B02G612000 was annotated as Caf- feic acid 3-O-methyltransferase (COMT), which might be involved in the stem lignin synthesis pathway. Except for the GATA transcription factor gene (TraesCS3B02G603900) and the aquaporin gene (TraesCS3B02G608500), the other four genes showed significant expression differences at three developmen- tal stages. The expression profiles of two VPE genes (TraesCS3B02G597800 and TraesCS3B02G597900) were similar, displaying high expressions in stems than in leaves. Their expression levels were significantly higher in low PT ‘Kauz’, with the highest transcript abundance at Z32 stage then showed a downward trend at Z65 stage. COMT gene (TraesCS3B02G612000) also maintained a higher relative expression level in ‘Kauz’, with the gene transcript abun- dance increasing gradually and reaching peak at the flower- ing stage. In addition, the Dof gene (TraesCS3B02G608800) was highly expressed in ‘Westonia’ and the transcript abundance was the highest at the early stage of stem tissue in both par- ents and then gradually decreased during development. It can be seen that the expression levels of VPE, DOF and COMT genes varied between developing stages. VPE and DOF were highly expressed at the stem elongation stage (Z32) when the stem pith cells were undergoing autoly- sis; while COMT accumulated significantly at the flower- ing stage (Z65). Considering that VPE is a cysteine-type endopeptidase and plays an important role in regulating the programmed death of plant cells, we concluded that TraesCS3B02G597800 and TraesCS3B02G597900 are more likely to be the genes responsible for the phenotypic differ- ences in pith thickness. Sequencing and phylogenetic analysis of TraesCS3B02G597900 In the candidate region, the CDS of six genes were amplified in the parents. Primers for six genes (TraesCS3B02G597800, Tra e s C S 3 B 0 2 G 5 9 7 9 0 0 , Tra e s C S 3 B 0 2 G 6 0 3 9 0 0, TraesCS3B02G608500, TraesCS3B02G608800 and TraesCS3B02G612000) are shown in the Table S3. The cor- responding PCR products were sequenced and aligned with the reference genome (IWGSC v2.1). TraesCS3B02G597800 and TraesCS3B02G597900 from low PT stem parent ‘Kauz’ shared the same sequence as the reference. The sequence of TraesCS3B02G597800 in high PT parent ‘Westonia’ contains only one missense SNP, while TraesCS3B02G597900 has not only several point mutations but also a 9-bp deletion in the first exon, result- ing in a 3-aa deletion and 14 amino acid substitutions in ‘Westonia’ (Fig. 7A, Fig. S3). Of the 14 amino acid substi- tutions, M465T displayed an extremely low SIFT (Sorting Intolerant from Tolerant) score of 0.01, implying that the substitution could affect the protein function according to Sim et al. (2012). Phylogenetic analysis was performed on putative VPE amino acid sequences from common wheat, Brachypodium distachyon (a relative of the wheat), the distant relative rice 1 3 Theoretical and Applied Genetics (2023) 136:138 Page 9 of 21 138 (Oryza sativa) and model plant Arabidopsis thaliana. The phylogenetic tree showed that wheat VPEs can be clustered into five clades with one Brachypodium VPE in each clade, including the endosperm-specific VPE1, the pericarp-spe- cific VPE4, and vegetative tissue-specific VPE3 and VPE5. For the VPE3 subfamilies, the wheat genome harbours three copies (VPE3a, 3b and 3c) with one from each of the three sub-genomes. TraesCS3B02G597900 belongs to VPE3 fam- ily (Fig. 7B). Therefore, we named it TaVPE3cB. TaVPE3 contains two conserved domains, peptidase C13 domain and legumain C domain. The N-terminal catalytic domain is a caspase-like from C13 family and the C-terminal is involved in legumain stabilization and activity modulation (LSAM). In high PT parent ‘Westonia’, TaVPE3cB contains three amino acid substitutions in the catalytic domain, one in the activation peptide and eight in the LSAM domain (Fig. S4). Nevertheless, there was no substitution of the key amino acids in the substrate pocket and catalytic dyad site (Hara-Nishimura et al. 2005). The sequence similarity analysis in ten common wheat varieties (‘Westonia’, ‘Lancer’, ‘CDC Landmark’, ‘Claire’, ‘Janz’, ‘Chinese Spring’, ‘Julius’, ‘SyMattis’, ‘Weebill’, ‘Kauz’) revealed that TaVPE3cB has two natural allelic vari- ations. High pith-thickness cultivars, such as ‘Lancer’, ‘CDC Landmark’ and ‘Janz’, shared the same allele as ‘Westonia’, which was named TaVPE3cB.b; while low pith-thickness ‘Kauz’ carries the same allele as ‘Chinese spring’, ‘Julius’, ‘SyMattis’ and ‘Weebill’, which was named TaVPE3cB.a (Fig. S4). Next, we cloned the 1  kb promoter of TaVPE3cB.a and TaVPE3cB.b. Sequence alignment revealed 73.83% sequence identity and a 309 bp insertion at 284 bp upstream of the start codon in the promoter of TaVPE3cB.b in Westo- nia (Fig. S5). We also carried out predictive analysis of the cis-acting elements in the promoters of TaVPE3cB.a and TaVPE3cB.b using PlantCARE. The analysis revealed vari- ous possible cis-acting elements in the two promoters that were mostly related to phytohormone response and stress induction, implying that TaVPE3cB may participate in the regulation of multiple phytohormones and environmental signalling pathways. Within the 309 bp insertion in the pro- moter of TaVPE3cB.b, 21 cis-acting elements exist includ- ing one unique cis-acting element MBS (MYB binding site involved in drought-inducibility, CAA CTG ) motif. SNP marker development and linkage analysis for a major pith‑thickness locus on 3BL To facilitate the use of TaVPE3cB.b in wheat-breeding pro- grams and confirm that all high pith-thickness accessions contain TaVPE3cB.b allele with the 309-bp insertion in the promoter, we developed two allele-specific PCR mark- ers (Table S3), Qpt3B-F1/R1 (dominant SNP marker) and Qpt3B-F2/R2 (codominant Indel marker for the promoter). The PCR products of Qpt3B-F1/R1 were 1097 bp in size for the varieties carrying TaVPE3cB.b allele, whereas no bands were amplified for the varieties carrying TaVPE3cB.a. The PCR products of Qpt3B-F2/R2 displayed a 634-bp band from TaVPE3cB.b, whereas a 325 bp band was observed in the varieties containing TaVPE3cB.a (Fig. S6). The genotyp- ing results obtained from these two pairs of markers were the same. Using these two pairs of allele-specific PCR markers to screen the DH population and historical varieties, we found that the TaVPE3cB.b genotype is closely linked to the high PT phenotype, while TaVPE3cB.a was associated with the low pith-thickness phenotype (Fig. 8A). The discrimination rate of this marker in high PT (PI > 0.6) DH lines was 100%, and 87.64% in low PT DH lines (PI < 0.4). A Spearman’s correlation coefficient of 0.782 (P < 0.01) between the PT index and TaVPE3cB.b gene demonstrated that TaVPE3cB.b significantly increase wheat stem pith-thickness in the DH population. However, the marker discrimination rate was low to 74.41% in high pith-thickness historical varieties (PI > 0.6), demonstrating the presence of other PT-related loci in some wheat varieties. When extreme phenotype vari- eties with solid stem (PI > 0.8) were tested, the detection rate was 89.47%. The discrimination rate of low PT (PI < 0.4) varieties was 89.79%, which was consistent with that in low PT DH lines, and the Spearman’s correlation coefficient was 0.501 (P < 0.01) between the PT index and TaVPE3cB.b gene. Therefore, this marker can be useful to identify the allele of QTL-3B for wheat varieties. The newly developed makers of Qpt3B were integrated into the previous QTL-3B linkage map developed by Zhao (2019). Finally, this marker was mapped in the genetic region of 302.5 cM. The QTL linked to high pith-thickness was detected under three individual environments using DH lines. This QTL was confined to an interval of 3.41 cM flanked by markers Qpt3B and NM3950 (Fig.  8B) and explained 68.72% and 13.85% of the phenotypic variance with the LOD value of 40.59 and 12.13, respectively. Expression analysis of TaVPE3cB Promoter structural analysis has wide implications in the prediction of gene expression profiles. We analyzed the GC- content in the 1.5 kb fragment upstream from the transla- tion initiation codon of the TaVPE3cB allelic variations and found that the AT-content of TaVPE3cB.a and TaVPE3cB.b promoter was 56% and 53%, respectively, higher than the corresponding GC content, which is characteristic of an AT- rich plant gene-promoter element. To clarify the contribu- tion of the 309 bp indel to gene expression, we performed qRT-PCR using the stems of ‘Chinese spring’ (CS) and 15 varieties containing TaVPE3cB.a genotype and 15 varieties 1 3 138 Page 10 of 21 Theoretical and Applied Genetics (2023) 136:138 1 3 Theoretical and Applied Genetics (2023) 136:138 Page 11 of 21 138 Fig. 5 SNP calling and DEG discovery via BSR-Seq. A Circos graph ◂ of genome-wide genes and SNPs distribution. The outer circle rep- resents chromosomes, the middle circle represents gene distribu- tion, and the inner circle represents the SNP density distribution. B: BSR-seq mapping of pith-thickness based on ∆SNP index value. ‘Bulk1’ represent the Hbulk SNP index, ‘Bulk2’ represent the Sbulk SNP index. The x-axis represents the position of chromosomes and the y-axis represents the SNP index or ∆SNP index value. The blue line represents the average value of ∆SNP index which was computed in a 5  Mb interval using a 50  kb sliding window. The red line and purple line represent the 99% and 95% confidence level threshold, respectively. C Gene expression comparison within the pith-thick- ness interval in the Chinese spring chromosome 3B as the reference. Physical positions are shown to the left of the map in Mb. Genes that contained SNPs between two parents are highlighted in green font. Gene expression differences between ‘Westonia vs Kauz’ and ‘Sbulk vs Hbulk’ comparisons are shown as a heatmap on the right. Posi- tive fold changes shown in red shading indicate higher expression in the solid sample. Negative fold changes shown in blue shading indi- cate higher expression in the hollow sample. Expression values are expressed as log2 FC. The red dot highlights the DEGs with upreg- ulated expression in both ‘Westonia’ and solid bulk; the green dot highlights DEGs with downregulated expression in both ‘Kauz’ and hollow bulk (colour figure online) containing TaVPE3cB.b genotype, which were selected from 171 historical varieties. The expression in all varieties con- taining TaVPE3cB.b (with 309 bp insertion) was lower than that in those containing TaVPE3cB.a (without the 309 bp insertion) (Fig. 9). This indicates that this 309 bp insertion can downregulate the expression of TaVPE3cB.b, which fur- ther inhibits pith death to induce a high PT stem phenotype. Discussion Insights of pith thickness formation mechanism Wheat stems are generally solid in the nodal region at the initial developmental stage, followed by an internodal cav- ity formation due to the death of pith cells during internode elongation. The genes responsible for stem pith-thickness are probably involved in the death of pith cells and cell wall composition. For example, pith thickness can be modulated by activating or inhibiting PCD (Fujimoto et al. 2018), or changing the cell wall composition, increasing stem cell wall thickness and lignin content (Kong et al. 2013). Plant cysteine proteases and PCD of stem Previous studies have revealed the roles of cysteine pro- teases in plant development as PCD initiators and executors (Rustgi et al. 2017; Sueldo and van der Hoorn 2017; Zhang et al. 2014). VPEs are cysteine proteinases and have impor- tant functions in the processing and maturation of proteins and PCD in the plant (Hara-Nishimura et al. 1993). Four functional VPE isoforms (α, β, γ, and δ-VPE) have been identified in Arabidopsis (Shimada et al. 2003). They can be divided into two subfamilies: seed type (β and δ- VPE) and vegetative type (α and γ- VPE), which are expressed primarily in seeds and vegetative organs, respectively. Seed type VPEs involve in the processing and maturation of seed storage proteins (Gruis et al. 2004), while vegetative type VPEs are found in lytic vacuoles and have been confirmed to involve in plant PCD and may act as functional substitution of caspases (Hatsugai et al. 2015). Arabidopsis γvpe mutants revealed that hypersensitive response related to PCD is reduced and the susceptibility to pathogens is increased in the absence of γVPE, as cell death can be blocked (Rojo et al. 2004). In this study, the GO enrichment analysis revealed that the Hcluster contains 35 up-regulated genes encoding for enzymes with aspartic, serine, and cysteine endopeptidase activity. We also found that TaVPE3cB was highly expressed in stem tissues rather than leaves at the elongation stage. The results are consistent with the finding of Kinoshita et al. (1995) in that γVPE is expressed predominantly in the stem of wheat. Recently, Cheng et al. (2019) demonstrated that γVPE regulates xylem fiber cell PCD by activating cysteine endopeptidase 1 (CEP1) during stem development, and CEP1 can function as an executor in clearing cellular con- tents during PCD in xylem development. Moreover, the mutation of γVPE exhibited a similar phenotype as cep1 mutant, such as incomplete degradation of the cellular con- tents and thickening secondary cell walls, which was caused by the prolonged PCD in xylem cells (Han et al. 2019). It was concluded that γVPE is not only involved in the matura- tion of CEP1, and also plays an important role in regulating the degradation of cellular content and the thickening of the secondary cell wall (Cheng et al. 2019). And more notably, Fujimoto et al. (2018) identified a NAC transcription fac- tor and referred as D gene (Sobic.006G147400), which up- regulated the expression of CEP1 and VPEs, thus triggering pith parenchyma cell PCD in sorghum. Therefore, we can speculate that the downregulated expression of TaVPE3cB blocked pith cell apoptosis, leading to thicker pith tissue. Transcriptional regulatory factors and PCD of stem A putative Dof zinc finger protein (TraesCS3B02G608800) was found in the adjacent region of PT QTL. It is the ortholog gene of TdDof (TRITD3Bv1G280530). Nilsen et  al. (2020) demonstrated that multiple copy numbers of TdDof increased the accumulation of gene transcripts and eventually inhibited PCD in pith parenchyma cells of solid-stemmed durum wheat. However, an exception was found in an Australia common wheat cultivar ‘Janz’, which only had a single copy of Dof but still exhibited the solid stemmed phenotype (Beres et al. 2013; Nilsen 2017). In addition, the relative copy number of TaDof gene was 1 3 138 Page 12 of 21 Theoretical and Applied Genetics (2023) 136:138 Fig. 6 Real-time quantitative PCR of candidate genes at three devel- opment stage in two parental lines. A Validation of RNA-seq data for differential gene expression by qRT-PCR. Inset: simple correlation plot of the log2 FC in expression obtained by RNA-Seq (x-axis) and qRT-PCR (y-axis). B Expression profiles of two VPEs, Dof, GATA, COMT and Aquaporin in each corresponding period of two parents in leaves and stems, respectively (colour figure online) estimated with the ∆∆CT method using a single copy gene, TraesCS3B02G612200, as the endogenous control gene. No copy number difference was found between the two paren- tal lines and among the DH lines. In addition, the gene fell outside of our defined QTL interval, suggesting that it is not a strong candidate gene of the QTL identified in the ‘Westo- nia’/ ‘Kauz’ DH population. In the current study, the PT QTL interval also harbours a GATA 17 transaction factor gene (TraesCS3B02G603900). GATA 12 has been reported to be involved in regulating the processes of plant xylem vessel differentiation and PCD (Cubría-Radío and Nowack 2019). For example, overexpres- sion of AtGATA12 in Arabidopsis can induce the formation of ectopic xylem vessel-like elements by manipulating the expression of VND7 transcription factor (Endo et al. 2015). Similarly, overexpression of PtrGATA12 in poplar resulted in increased contents of lignin and secondary cell wall (SCW) thickness by controlling the expressions of some master TFs and pathway genes involved in SCW formation and PCD. Moreover, the PtrGATA12 transgenic lines exhib- ited significantly increased stem diameter (Ren et al. 2021). Besides, GATA19 has been proved to be involved in regulat- ing plant growth rate. For instance, PdGATA19 transgenic lines exhibited increased biomass accumulation, stem height and photosynthetic rate; while CRISPR/Cas9-mediated mutant plants showed severe developmental retardation and increased formation of secondary xylem (An et al. 2020). In this study, three polymorphic SNPs with higher ΔSNP index values (> 0.75) were found in this gene. Both RNA- seq and qRT-PCR showed that the expression level of this gene is also different between the two parents as well as the two bulked samples. 1 3 Theoretical and Applied Genetics (2023) 136:138 Page 13 of 21 138 Fig. 7 Amino acid sequence alignment and phylogenetic tree of TraesCS3B02G597900. A Alignment of the deduced amino acid sequences of TraesCS3B02G597900 from ‘Westonia’ and ‘Kauz’. The red rectangle represents a 9 bp Indel. The red triangle represents an amino acid substitution M465T with SIFT = 0.01; B Phylogenetic tree of vacuolar processing enzyme family. This tree illustrates the VPE1-5 family groups and includes VPE proteins from Arabidopsis (red font), rice (blue font), Brachypodium (orange font) and wheat (black font). Red star represents TraesCS3B02G597900 (colour figure online) Cell wall modification and cell expansion COMT is considered as an important gene that functions in lignin biosynthesis, and it is positively correlated with lignin content in wheat stems (Bi et al. 2011). Lignin dep- osition could reinforce the cell wall to provide mechanical support to the stem which makes it possible to modify stem strength and lodging resistance by affecting lignin content (Ma 2009; Tu et al. 2010). A p u t a t i v e O - m e t h y l t r a n s f e r a s e g e n e (TraesCS3B02G612000) which contains the O-MeTrfase_ COMT domain was located on the adjacent region of the target QTL. This gene was suggested as a promising candidate gene for stem pith production based on strong 1 3 138 Page 14 of 21 Theoretical and Applied Genetics (2023) 136:138 Fig. 8 TaVPE3cB marker development and linkage map analysis. A Pith-thickness index in TaVPE3cB.a and TaVPE3cB.b genotypes of DH population (a) and historical lines (b). B Location of QTL for PT on 3B under three individual environments (2018-2020) differential expression between solid and hollow cultivars (Oiestad et al. 2017). In the current study, no SNP for COMT was found between the two parental lines. How- ever, it was significantly up-regulated in low PT ‘Kauz’ and hollow bulked samples with log2 FC of 2.25 and 1.64, respectively. In addition, GO analysis also identified sig- nificant functional enrichment of O-methyltransferase in Hcluster. This suggests that the activity of cellular ligni- fication by COMT was lower in high PT Wheat, which is consistent with the result reported by Nilsen (2017). 1 3 Theoretical and Applied Genetics (2023) 136:138 Page 15 of 21 138 Fig. 9 Expression analysis of TaVPE3cB.a and TaVPE3cB.b in historical wheat varieties. The red line shows the relative expres- sion levels of TaVPE3cB in CS which contains TaVPE3cB.a (con- trol). + , −  Denote TaVPE3cB.b containing the 309  bp insertion and TaVPE3cB.a missing the 309  bp insertion in the promoter, respec- tively. ∗, ∗∗Denotes significant differences at 5% and 1% probability levels, respectively (colour figure online) Taking together, TraesCS3B02G612000 affects PT is probably through lignifying the stem pith cell wall. An aquaporin gene (TraesCS3B02G608500) was also observed adjacent to the PT QTL interval. Aquaporins are universal membrane integrated water channel proteins which play an important role in cell expansion and cell division by controlling water uptake (Maurel et al. 2015). Fujimoto et al. (2018) found high PT stems were filled with plump pith cells which enhanced stem water con- tent. In this study, a tonoplast intrinsic protein aquaporin gene (TIP1) with two nonsynonymous substitutions in the coding region showed significant expression-level differ- ences. It was significantly upregulated in high PT sam- ples but with low expression abundance (FPKM < 0.8). Previous studies showed a correlation between TIP1 expression level with cell elongation and differentiation in the vascular tissue of Arabidopsis thaliana (Ludevid et al. 1992). In addition to TIP, the plasma membrane intrinsic aquaporin (PIP) has been implicated in plant stem growth. For example, increasing the expression of PIP1b in transgenic tobacco can improve water transport and transgenic plants with thicker stem diameters more than those of wild-type plants (Aharon et al. 2003). Yu et al. (2005) found that PIP1 antisense transgenic tobacco plants displayed thicker and shorter stems than wild-type plants. Therefore, the differential expression of aquaporin genes may be the cause of differences in pith cell expan- sion and water uptake between the two stem phenotypes. Metallothionein and PCD of stem Plant metallothionein (MT) is a small and functionally, cysteine-rich protein that plays multiple roles in reactive oxygen species (ROS) scavenging. Metallothionein protein functions as a cytosol ROS scavenger, it can stall the signal transduction of ROS-mediated PCD, which is a widespread regulatory mechanism for eliminating unwanted cells in normal plant growth and development. For example, the OsMT2b gene in rice and the MT3a gene in cotton exhib- ited strong antioxidative activities against ROS (Xue et al. 2009). Knocking out OsMT2b caused excessive epider- mal cell death in stems (Steffens and Sauter 2009). Beers (1997) proposed that PCD is essential for eliminating pith parenchyma cells and forming aerenchyma to facilitate gas exchange. In addition, previous studies have shown that pith PCD activation is inhibited in solid stemmed wheat during stem elongation (Nilsen et al. 2020). Therefore, MT could be involved in the normal PCD of pith cells. In this study, we observed significant up-regulation of MT genes in low PT samples according to RNA-seq results, which may be due to the antioxidant defence mechanism increasing the expres- sion of ROS scavenger genes, thus mitigating the damage caused by ROS. Based on the above results, we proposed a regulation model for the formation of wheat hollow stems (Fig. 10). The topmost level of regulation of the differences between high PT and low PT stem was correlated with a hor- monal signalling pathway. Many DEGs involved in auxin, 1 3 138 Page 16 of 21 Fig. 10 Five-level hierarchy diagram of stem pith cells PCD. Dotted lines indicated indirect regulation; solid lines indicate direct regulation Theoretical and Applied Genetics (2023) 136:138 cytokine and brassinosteroid plant hormone signal transduc- tion were revealed by KEGG analysis, and the role of this pathway in regulating stem elongation has been reviewed by Haruta and Sussman (2017). In the downstream of hormonal signalling, multiple transcription factors such as NAC, Dof and GATA have been involved in cell differentiation includ- ing PCD as its final destination. In this pathway, numerous proteases participate in PCD execution. The most common executor is cysteine proteases, such as CEP1, VPE and meta- caspases, which have been shown to contribute to cellular autolysis before and after PCD. In addition, ROS can act as the cell death signal in the MAPK signalling pathway together with hormones to activate protease to initiate cell death (Biswas and Mano 2016; Li et al. 2012; Overmyer et al. 2003). After hydrolytic enzymes are released, the cell wall breakdown and cell wall components recombination occurs, as many DEGs are related to polysaccharides and cellulose metabolic, lignin biosynthesis, glycosyltransferase and aquaporins, which can modify the cell wall component (Gunawardena et al. 2007). Eventually, the pith cell vacuolar ruptures and triggers chromatin degradation (Obara et al. 2001). Comparison of genes/QTLs for stem‑related traits In the wheat breeding program (Dreccer et al. 2020), the design of wheat varieties with wider stem diameter, high culm wall thickness, small pith cavity and high stem sol- idness is desirable for enhancing stem strength and lodg- ing resistance. Using forward genetic approaches, several strong stem phenotype-related QTLs have been identified. SSt1 in durum wheat and Qss.msub-3BL in common wheat are the earliest mapped stem solidness loci on chromosome 3B. Later, loci on 1A, 2D, 3B and 4B for culm wall thick- ness and pith diameter were identified. However, only the loci on 1A for culm wall thickness and the 3B locus for stem solidness have been identified through map-based cloning. The Csl is the candidate gene from chromosome 1A, which altered carbon partitioning throughout the plant and increased the cell wall thickness (Hyles et al. 2017). TdDof (TRITD3Bv1G280530) was cloned as the most likely SSt1 candidate gene due to different copy numbers in solid- stemmed and hollow-stemmed durum wheat lines. Likewise, TaDof (TraesCS3B01G608800) has been reported as the can- didate gene for Qss.msub-3BL. In previous studies of our group, Zhao (2019) detected the major QTLs for wheat pith thickness and stem diameter on 3BL by using DH population from ‘Westonia’/ ‘Kauz’. The 3B QTL for PT was saturated to a 3.0 cM interval which corresponded to a 1.43 Mb (820,760,675–822,192,510 bp) physical region and was stably expressed in five differ- ent environments. In this study, one PT-related candi- date interval was identified at a 6.83 Mb physical interval (819,897,386–826,725,912 bp) of 3BL using BSR-seq data. 1 3 Theoretical and Applied Genetics (2023) 136:138 Page 17 of 21 138 However, the reported TaDof in Qss.msub-3BL is situated at 828,110,748–828,112,481 bp, which is different from the QTL region in this study. Meanwhile, the copy number estimation using the ∆∆CT method also excluded the pos- sibility of TaDof being the candidate gene for the QTL in our study (see below for details). These indicate that there are other candidate genes present in ‘Westonia’ for the PT phenotype. The novel maker developed in this study was mapped to the adjacent region reported by Cook et  al. (2004) (gwm247, gwm340, gwm547, and BARC77), Pan et  al. (2017) (gwm547–gwm247) and Piñera‐Chavez et al. (2021) (gwm285 and gwm547). Furthermore, the physical region mapped with MutMap in the current overlapped with the region from those of Zhao (2019) (NM1756–NM3950). Six genes were selected based on BSR-seq analysis, differen- tial expression analysis and gene functional annotation. The gene effect of TraesCS3B02G597900 on PT phenotype was verified through an SNP maker in 171 historical wheat cul- tivars. These results demonstrate that TraesCS3B02G597900 is the candidate gene underlying the 3BL QTL in our study. TraesCS3B02G597900 is a putative candidate gene for pith‑thickness Based on sequencing results, TraesCS3B02G597900 (TaVPE3cB) showed a 9 bp Indel and multiple nonsyn- onymous SNPs between the two parents. RNA-seq results showed that this gene has significant differences between the two parents (log2FC = −3.77) and between the two extreme pools (log2FC = −2.55). In addition, qRT-PCR results con- firmed that this gene was highly expressed in low PT sam- ples during the stem elongation stage, which is a critical time for the central pith cells to initiate apoptosis and form the pith cavity (Nilsen et al. 2020). The negative correlation between gene expression level and pith thickness extent was confirmed in the current study. In addition, the expression profiles of TaVPE3cB in CS and 30 wheat varieties indicated that a 309 bp insertion in the promoter might inhibit the gene expression. This insertion contains a unique MBS motif that is related to drought inducibility. Several studies confirmed that gene expressions can be affected by large Indels located in the promoter region. For example, a 160-bp insertion in the promoter of Rht-B1i-1 significantly enhances the gene expression and significantly increased the plant height of wheat (Lou et al. 2016). Recently, Mao et al. (2022) con- firmed that a 108 bp insertion in the promoter of TaNAC071- A increases its gene transcription level and drought toler- ance. However, further functional validation of this 309 bp indel in the promoter of TaVPE3cB is required. We designed a dominant SNP marker Qpt3B-F1/R1 and re-mapped it in the DH population and found the phenotype was co-segregated with it. When genotyping 171 historical Australian wheat cultivars, TaVPE3cB.b is significantly related to low pith thickness with a correlation coeffi- cient of 0.501 (P < 0.01), and the ratio of TaVPE3cB.a to TaVPE3cB.b is about 2:1, suggesting that the percentage of high PT wheat variety is less than the low one. This finding is consistent with the fact that most wheat cultivars grown worldwide have a hollow stem with a thin pith, only a small number of varieties have fully developed stem pith cells and exhibit solid stemmed phenotype (Pluta et al. 2021). Improving the pith thickness of wheat stem is a way to increase the ability of wheat to resist lodging (Kong et al. 2013), wheat stem sawfly (Beres et al. 2011), and drought stress (Monneveux et al. 2012). Several studies have shown that the diameter and the wall thickness of the basal stems are positively related to lodging and stem mechanical strength (Pinera-Chavez et al. 2016; Zuber et al. 1999). In addition, the thickness of the pith parenchyma also posi- tively affects the mechanical resistance against stem bend- ing. For instance, wheat cultivars with solidness-stems tend to have higher resistance against stem bending than hollow- stem wheat cultivars (Kong et al. 2013). However, stem wall thickness can lead to increasing stem material per unit of strength which can be biomass costly. Berry et al. (2007) suggested that the ideal strategy to enhance lodging resist- ance with the minimum biomass investment in winter wheat would be to increase internode width and internode material strength instead of stem wall thickness. Therefore, it might be a possible strategy of breeding lodging tolerance wheat with higher biomass through mutating TaVPE3, which might have some effects on the biosynthesis of the secondary cell wall and regulating pith thickness. Conclusion The present study identified mRNA variants in com- mon wheat for stem pith thickness through BSR-seq. One pith thickness-related candidate region was located on a 6.83 Mb physical interval of Qpt-3B using BSR-seq data. A total of sixteen genes were found differentially expressed, among them four DEGs, TraesCS3B02G597800, TraesCS3B02G597900, TraesCS3B02G603900 and TraesCS3B02G608500, exhibited both differential expres- sion levels and polymorphic SNPs between high PT and low PT samples. Finally, TaVPE3cB was identified as a high- confidence candidate gene for PT. The SNP makers for the candidate gene were developed and successfully separated TaVPE3cB.a and TaVPE3cB.b alleles. It was further applied to screen historical wheat cultivars of different pith thick- nesses. In addition, an insertion in the promoter region of TaVPE3cB has been found related to the downregulation of this gene expression in wheat. 1 3 138 Page 18 of 21 Theoretical and Applied Genetics (2023) 136:138 Supplementary Information The online version contains supplemen- tary material available at https:// doi. org/ 10. 1007/ s00122- 023- 04372-4. Acknowledgements This work was supported by Murdoch Univer- sity and the Australia Grains Research & Development Corporation (GRDC) (grant number UMU00048), the Department of Primary Industries and Regional Development (DPIRD), Western Australia. We thank InterGrain, Western Australia, for providing the Westonia and Kauz DH population. Author contribution statement WM, HM and SI conceived the pro- ject and designed the study; QL, YZ, SI, SRHL and RY caried out field experiments; QL, JZ, YR and YZ performed the gene sequenc- ing, molecular marker development and data analysis; GO, JZ, RKV and WM provided the resources for the study; QL and YZ wrote the original draft of the manuscript; WM, JZ, and MS provided extensive revision and editing; WM, SI and WY supervised and managed the project. All authors have read and agreed to the published version of the manuscript. Funding Open Access funding enabled and organized by CAUL and its Member Institutions. This research is financially support by GRDC project UMU00048. Data availability Electronic supplementary material The online version of this article (https:// doi. org/ xxxxx) contains supplementary material, which is available to authorized users. Declarations Conflict of interest The authors declare that there is no conflict of in- terest. Open Access This article is licensed under a Creative Commons Attri- bution 4.0 International License, which permits use, sharing, adapta- tion, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/. References Abe A, Kosugi S, Yoshida K, Natsume S, Takagi H, Kanzaki H, Mat- sumura H, Yoshida K, Mitsuoka C, Tamiru M (2012) Genome sequencing reveals agronomically important loci in rice using mutmap. Nat Biotechnol 30:174–178 Aharon R, Shahak Y, Wininger S, Bendov R, Kapulnik Y, Galili G (2003) Overexpression of a plasma membrane aquaporin in transgenic tobacco improves plant vigor under favorable growth conditions but not under drought or salt stress. Plant Cell 15:439–447 An Y, Zhou Y, Han X, Shen C, Wang S, Liu C, Yin W, Xia X (2020) The GATA transcription factor GNC plays an important role in photosynthesis and growth in poplar. J Exp Bot 71:1969–1984 Appleford NE, Wilkinson MD, Ma Q, Evans DJ, Stone MC, Pearce SP, Powers SJ, Thomas SG, Jones HD, Phillips AL (2007) Decreased shoot stature and grain α-amylase activity following ectopic expression of a gibberellin 2-oxidase gene in transgenic wheat. J Exp Bot 58:3213–3226 Beers EP (1997) Programmed cell death during plant growth and devel- opment. Cell Death Differ 4:649–661 Beres BL, Dosdall LM, Weaver DK, Cárcamo HA, Spaner DM (2011) Biology and integrated management of wheat stem sawfly and the need for continuing research. Can Entomol 143:105–125 Beres B, Cárcamo H, Byers J, Clarke F, Pozniak C, Basu S, DePauw R (2013) Host plant interactions between wheat germplasm source and wheat stem sawfly cephus cinctus norton (hymenoptera: cephidae) I. Commercial cultivars. Can J Plant Sci 93:607–617 Berry PM, Sylvester-Bradley R, Berry S (2007) Ideotype design for lodging-resistant wheat. Euphytica 154:165–179 Berry PM, Berry S, Spink J (2008) Identification of genetic markers for lodging resistance in wheat. Hgca Project Report Bi C, Chen F, Jackson L, Gill BS, Li W (2011) Expression of lignin biosynthetic genes in wheat during development and upon infec- tion by fungal pathogens. Plant Mol Biol Report 29:149–161 Biswas MS, Mano JI (2016) Reactive carbonyl species activate cas- pase-3-like protease to initiate programmed cell death in plants. Plant Cell Physiol 57:1432–1442 Blum A (1998) Improving wheat grain filling under stress by stem reserve mobilisation. Euphytica 100:77–83 Butler JD, Byrne PF, Mohammadi V, Chapman PL, Haley SD (2005) Agronomic performance of Rht alleles in a spring wheat popu- lation across a range of moisture levels. Crop Sci 45:939–947 Cheng Z, Zhang J, Yin B, Liu Y, Wang B, Li H, Lu H (2019) γVPE plays an important role in programmed cell death for xylem fiber cells by activating protease CEP1 maturation in arabidopsis thali- ana. Int J Biol Macromol 137:703–711 Cingolani P, Platts A, Wang LL, Coon M, Nguyen T, Wang L, Land SJ, Lu X, Ruden DM (2012) A program for annotating and pre- dicting the effects of single nucleotide polymorphisms, SnpEff: SNPs in the genome of drosophila melanogaster strain w1118; iso-2; iso-3. Fly 6:80–92 Cook J, Wichman D, Martin J, Bruckner P, Talbert L (2004) Identifica- tion of microsatellite markers associated with a stem solidness locus in wheat. Crop Sci 44:1397–1402 Cubría-Radío M, Nowack MK (2019) Transcriptional networks orches- trating programmed cell death during plant development. Curr Top Dev Biol 131:161–184 Dreccer MF, Condon AG, Macdonald B, Rebetzke GJ, Awasi M-A, Borgognone MG, Peake A, Piñera-Chavez FJ, Hundt A, Jack- way P (2020) Genotypic variation for lodging tolerance in spring wheat: wider and deeper root plates, a feature of low lodging, high yielding germplasm. Field Crop Res 258:107942 Du H, Zhu J, Su H, Huang M, Wang H, Ding S, Zhang B, Luo A, Wei S, Tian X (2017) Bulked segregant RNA-seq reveals differential expression and SNPs of candidate genes associated with water- logging tolerance in maize. Front Plant Sci 8:1022 Endo H, Yamaguchi M, Tamura T, Nakano Y, Nishikubo N, Yoneda A, Kato K, Kubo M, Kajita S, Katayama Y (2015) Multiple classes of transcription factors regulate the expression of vascular-related NAC-DOMAIN7, a master switch of xylem vessel differentia- tion. Plant Cell Physiol 56:242–254 FAOSTAT (2022) Food and agricultural organization of the united nations, FAOSTAT. https:// www. fao. org/ faost at/ en/. Accessed 28 January 2022 Fujimoto M, Sazuka T, Oda Y, Kawahigashi H, Wu J, Takanashi H, Ohnishi T, Yoneda J-I, Ishimori M, Kajiya-Kanegae H (2018) Transcriptional switch for programmed cell death in pith paren- chyma of sorghum stems. Proc Natl Acad Sci 115:E8783–E8792 Gruis D, Schulze J, Jung R (2004) Storage protein accumulation in the absence of the vacuolar processing enzyme family of cysteine proteases. Plant Cell 16:270–290 1 3 Theoretical and Applied Genetics (2023) 136:138 Page 19 of 21 138 Gunawardena AH, Greenwood JS, Dengler NG (2007) Cell wall deg- radation and modification during programmed cell death in lace plant, aponogeton madagascariensis (aponogetonaceae). Am J Bot 94:1116–1128 Hai L, Guo HH, Xiao SH, Jiang GL, Zhang XY, Yan CS, Xin ZY, Jia JZ (2005) Quantitative trait loci (QTL) of stem strength and related traits in a doubled-haploid population of wheat (triticum aestivum L.). Euphytica 141:1–9 Han J, Li H, Yin B, Zhang Y, Liu Y, Cheng Z, Liu D, Lu H (2019) The papain-like cysteine protease CEP1 is involved in programmed cell death and secondary wall thickening during xylem develop- ment in arabidopsis. J Exp Bot 70:205–215 Hao Z, Geng M, Hao Y, Zhang Y, Zhang L, Wen S, Wang R, Liu G (2019) Screening for differential expression of genes for resist- ance to sitodiplosis mosellana in bread wheat via BSR-seq analy- sis. Theor Appl Genet 132:3201–3221 Hara-Nishimura I, Takeuchi Y, Nishimura M (1993) Molecular char- acterization of a vacuolar processing enzyme related to a puta- tive cysteine proteinase of schistosoma mansoni. Plant Cell 5:1651–1659 Hara-Nishimura I, Hatsugai N, Nakaune S, Kuroyanagi M, Nishimura M (2005) Vacuolar processing enzyme: an executor of plant cell death. Curr Opin Plant Biol 8:404–408 Haruta M, Sussman MR (2017) Ligand receptor-mediated regulation of growth in plants. Curr Top Dev Biol 123:331–363 Hatsugai N, Yamada K, Goto-Yamada S, Hara-Nishimura I (2015) Vacuolar processing enzyme in plant programmed cell death. Front Plant Sci 6:234 Hayat M, Martin J, Lanning S, McGuire C, Talbert L (1995) Variation for stem solidness and its association with agronomic traits in spring wheat. Can J Plant Sci 75:775–780 Hedden P (2003) The genes of the green revolution. Trends Genet 19:5–9 Hirano K, Ordonio RL, Matsuoka M (2017) Engineering the lodg- ing resistance mechanism of post-green revolution rice to meet future demands. Proceedings of the Japan academy. Ser B Phys biol sci. 93: 220–233 Hyles J, Vautrin S, Pettolino F, MacMillan C, Stachurski Z, Breen J, Berges H, Wicker T, Spielmeyer W (2017) Repeat-length vari- ation in a wheat cellulose synthase-like gene is associated with altered tiller number and stem cell wall composition. J Exp Bot 68:1519–1529 Keating BA, Herrero M, Carberry PS, Gardner J, Cole MB (2014) Food wedges: framing the global food demand and supply challenge towards 2050. Glob Food Sec 3:125–132 Kim D, Paggi JM, Park C, Bennett C, Salzberg SL (2019) Graph-based genome alignment and genotyping with HISAT2 and HISAT- genotype. Nat Biotechnol 37:907–915 Kinoshita T, Nishimura M, Hara-Nishimura I (1995) The sequence and expression of the γ-VPE gene, one member of a family of three genes for vacuolar processing enzymes in arabidopsis thaliana. Plant Cell Physiol 36:1555–1562 Kirby EJM (2002) Botany of the wheat plant Kong E, Liu D, Guo X, Yang W, Sun J, Li X, Zhan K, Cui D, Lin J, Zhang A (2013) Anatomical and chemical characteristics associ- ated with lodging resistance in wheat. Crop J 1:43–49 Krzywinski M, Schein J, Birol I, Connors J, Gascoyne R, Horsman D, Jones SJ, Marra MA (2009) Circos: an information aesthetic for comparative genomics. Genome Res 19:1639–1645 Li Z, Yue H, Xing D (2012) Map kinase 6-mediated activation of vacuolar processing enzyme modulates heat shock-induced pro- grammed cell death in arabidopsis. New Phytol 195:85–96 Li W-Q, Han M-M, Pang D-W, Jin C, WANG YY, DONG H-H, CHANG Y-L, Min J, LUO Y-L, Yong L (2022) Characteris- tics of lodging resistance of high-yield winter wheat as affected by nitrogen rate and irrigation managements. J Integr Agric 21:1290–1309 Liu S, Yeh C-T, Tang HM, Nettleton D, Schnable PS (2012) Gene mapping via bulked segregant RNA-seq (BSR-seq). PLoS ONE 7:e36406 Liu K, Deng Z, Zhang Y, Wang F, Liu T, Li Q, Shao W, Zhao B, Tian J, Chen J (2017) Linkage analysis and genome-wide association study of QTLs controlling stem-breaking-strength-related traits in wheat. Acta Agron Sin 43:483–495 Livak KJ, Schmittgen TD (2001) Analysis of relative gene expression data using real-time quantitative PCR and the 2−ΔΔCT method. Methods 25:402–408 Lou X, Li X, Li A, Pu M, Shoaib M, Liu D, Sun J, Zhang A, Yang W (2016) The 160 bp insertion in the promoter of Rht-B1i plays a vital role in increasing wheat height. Front Plant Sci 7:307 Ludevid D, Hofte H, Himelblau E, Chrispeels MJ (1992) The expres- sion pattern of the tonoplast intrinsic protein γ-TIP in arabidop- sis thaliana is correlated with cell enlargement. Plant Physiol 100:1633–1639 Ma Q-H (2009) The expression of caffeic acid 3-O-methyltransferase in two wheat genotypes differing in lodging resistance. J Exp Bot 60:2763–2771 Mao H, Li S, Chen B, Jian C, Mei F, Zhang Y, Li F, Chen N, Li T, Du L (2022) Variation in cis-regulation of a NAC transcription factor contributes to drought tolerance in wheat. Mol Plant 15:276–292 Maurel C, Boursiac Y, Luu D-T, Santoni V, Shahzad Z, Verdoucq L (2015) Aquaporins in plants. Physiol Rev 95:1321–1358 McKenna A, Hanna M, Banks E, Sivachenko A, Cibulskis K, Kerny- tsky A, Garimella K, Altshuler D, Gabriel S, Daly M (2010) The genome analysis toolkit: a mapreduce framework for ana- lyzing next-generation DNA sequencing data. Genome Res 20:1297–1303 Meng L, Li H, Zhang L, Wang J (2015) QTL IciMapping: integrated software for genetic linkage map construction and quantitative trait locus mapping in biparental populations. Crop J 3:269–283 Monna L, Kitazawa N, Yoshino R, Suzuki J, Masuda H, Maehara Y, Tanji M, Sato M, Nasu S, Minobe Y (2002) Positional cloning of rice semidwarfing gene, sd-1: rice “green revolution gene” encodes a mutant enzyme involved in gibberellin synthesis. DNA Res 9:11–17 Monneveux P, Jing R, Misra SC (2012) Phenotyping for drought adap- tation in wheat using physiological traits. Front Physiol 3:429 Nilsen KT, Walkowiak S, Xiang D, Gao P, Quilichini TD, Willick IR, Byrns B, N’Diaye A, Ens J, Wiebe K et al (2020) Copy number variation of TdDof controls solid-stemmed architecture in wheat. Proc Natl Acad Sci USA 117:28708–28718 Nilsen KT, N'Diaye A, MacLachlan PR, Clarke JM, Ruan YF, Cuthbert RD, Knox RE, Wiebe K, Cory AT, Walkowiak S, et al (2017) High density mapping and haplotype analysis of the major stem- solidness locus SSt1 in durum and common wheat. Plos One 12(4):e0175285. Nilsen KT (2017) A study of solid-stem expression in durum and com- mon wheat: University of Saskatchewan Graduate Theses and Dissertations. https:// hdl. handle. net/ 10388/ 8360 Obara K, Kuriyama H, Fukuda H (2001) Direct evidence of active and rapid nuclear degradation triggered by vacuole rupture during programmed cell death in Zinnia. Plant Physiol 125:615–626 Oiestad AJ, Martin JM, Cook J, Varella AC, Giroux MJ (2017) Iden- tification of candidate genes responsible for stem pith produc- tion using expression analysis in solid‐stemmed wheat. Plant Genome. 10: plantgenome2017.2002.0008 Overmyer K, Brosché M, Kangasjärvi J (2003) Reactive oxygen species and hormonal control of cell death. Trends Plant Sci 8:335–342 Pan T, Hu W, Li D, Cheng X, Wu R, Cheng S (2017) Influence of stem solidness on stem strength and stem solidness associated QTLs in bread wheat. Acta Agron Sin 43:9–18 1 3 138 Page 20 of 21 Theoretical and Applied Genetics (2023) 136:138 PAUWRead RDD (1982) The effect of nitrogen and phosphorus on the expression of stem solidness in Canuck wheat at four locations in southwestern Saskatchewan. Can J Plant Sci 62:593–598 Peng D, Chen X, Yin Y, Lu K, Yang W, Tang Y, Wang Z (2014) Lodging resistance of winter wheat (triticum aestivum L.): lignin accumulation and its related enzymes activities due to the application of paclobutrazol or gibberellin acid. Field Crop Res 157:1–7 Pinera-Chavez FJ, Berry PM, Foulkes MJ, Jesson MA, Reynolds MP (2016) Avoiding lodging in irrigated spring wheat. I. Stem and root structural requirements. Field Crop Res 196:325–336 Piñera-Chavez FJ, Berry PM, Foulkes MJ, Sukumaran S, Reynolds MP (2021) Identifying quantitative trait loci for lodging-associated traits in the wheat doubled-haploid population Avalon×Cadenza. Crop Sci 61:2371–2386 Pluta M, Kurasiak-Popowska D, Nawracała J, Bocianowski J, Mikołajczyk S (2021) Estimation of stem-solidness and yield components in selected spring wheat genotypes. Agronomy 11:1640 Rajaram S, Borlaug N, Van Ginkel M (2002) CIMMYT international wheat breeding. Bread wheat improvement and production. FAO. Rome. 103–117 Ren M, Zhang Y, Liu C, Liu Y, Tian S, Cheng H, Zhang H, Wei H, Wei Z (2021) Characterization of a high hierarchical regulator, ptrgata12, functioning in differentially regulating secondary wall component biosynthesis in populus trichocarpa. Front Plant Sci 12:726 Reynolds M, Bonnett D, Chapman SC, Furbank RT, Manès Y, Mather DE, Parry MAJ (2010) Raising yield potential of wheat. I. Overview of a consortium approach and breeding strategies. J Exp Bot 62:439–452 Rojo E, Martın R, Carter C, Zouhar J, Pan S, Plotnikova J, Jin H, Paneque M, Sánchez-Serrano JJ, Baker B (2004) VPEγ exhib- its a caspase-like activity that contributes to defense against pathogens. Curr Biol 14:1897–1906 Rustgi S, Boex-Fontvieille E, Reinbothe C, von Wettstein D, Reinbothe S (2017) Serpin1 and WSCP differentially regu- late the activity of the cysteine protease RD21 during plant development in arabidopsis thaliana. Proc Natl Acad Sci 114:2212–2217 Ruuska SA, Rebetzke GJ, Van Herwaarden AF, Richards RA, Fettell NA, Tabe L, Jenkins CL (2006) Genotypic variation in water- soluble carbohydrate accumulation in wheat. Funct Plant Biol 33:799–809 Saint Pierre C, Trethowan R, Reynolds M (2010) Stem solidness and its relationship to water-soluble carbohydrates: associa- tion with wheat yield under water deficit. Funct Plant Biol 37:166–174 Scofield GN, Ruuska SA, Aoki N, Lewis DC, Tabe LM, Jenkins CL (2009) Starch storage in the stems of wheat plants: localization and temporal changes. Ann Bot 103:859–868 Sharma P, Jha AB, Dubey RS, Pessarakli M (2012) Reactive oxygen species, oxidative damage, and antioxidative defense mecha- nism in plants under stressful conditions. J Bot. https:// doi. org/ 10. 1155/ 2012/ 217037 Shimada T, Yamada K, Kataoka M, Nakaune S, Koumoto Y, Kuroy- anagi M, Tabata S, Kato T, Shinozaki K, Seki M (2003) Vacu- olar processing enzymes are essential for proper processing of seed storage proteins in arabidopsis thaliana. J Biol Chem 278:32292–32299 Sim N-L, Kumar P, Hu J, Henikoff S, Schneider G, Ng PC (2012) SIFT web server: predicting effects of amino acid substitutions on proteins. Nucleic Acids Res 40:W452–W457 Song P, Wang X, Wang X, Zhou F, Xu X, Wu B, Yao J, Lv D, Yang M, Song X (2021) Application of 50K chip-based genetic map to QTL mapping of stem-related traits in wheat. Crop Pasture Sci 72:105–112 Steffens B, Sauter M (2009) Epidermal cell death in rice is confined to cells with a distinct molecular identity and is mediated by ethylene and H2O2 through an autoamplified signal pathway. Plant Cell 21:184–196 Sueldo DJ, van der Hoorn RA (2017) Plant life needs cell death, but does plant cell death need Cys proteases? FEBS J 284:1577–1585 Takagi H, Abe A, Yoshida K, Kosugi S, Natsume S, Mitsuoka C, Uemura A, Utsushi H, Tamiru M, Takuno S (2013) QTL- seq: rapid mapping of quantitative trait loci in rice by whole genome resequencing of DNA from two bulked populations. Plant J 74:174–183 Tamura K, Dudley J, Nei M, Kumar S (2007) MEGA4: molecular evolutionary genetics analysis (MEGA) software version 4.0. Mol Biol Evol 24:1596–1599 Tan C, Liu Z, Huang S, Feng H (2019) Mapping of the male sterile mutant gene ftms in brassica rapa L. ssp. pekinensis via BSR- seq combined with whole-genome resequencing. Theor Appl Genet 132:355–370 Thorvaldsdóttir H, Robinson JT, Mesirov JP (2013) Integrative genomics viewer (IGV): high-performance genomics data visualization and exploration. Brief Bioinform 14:178–192 Trapnell C, Williams BA, Pertea G, Mortazavi A, Kwan G, Van Baren MJ, Salzberg SL, Wold BJ, Pachter L (2010) Transcript assembly and quantification by RNA-seq reveals unannotated transcripts and isoform switching during cell differentiation. Nat Biotechnol 28:511–515 Tu Y, Rochfort S, Liu Z, Ran Y, Griffith M, Badenhorst P, Louie GV, Bowman ME, Smith KF, Noel JP (2010) Functional analyses of caffeic acid O-methyltransferase and cinnamoyl-CoA-reductase genes from perennial ryegrass (lolium perenne). Plant Cell 22:3357–3373 Verma V, Worland A, Savers E, Fish L, Caligari P, Snape J (2005) Identification and characterization of quantitative trait loci related to lodging resistance and associated traits in bread wheat. Plant Breed 124:234–241 Wallace L, McNeal F, Berg M (1973) Minimum stem solidness required in wheat for resistance to the wheat stem sawfly. J Econ Entomol 66:1121–1124 Wang C, Deng P, Chen L, Wang X, Ma H, Hu W, Yao N, Feng Y, Chai R, Yang G (2013) A wheat WRKY transcription factor TaWRKY10 confers tolerance to multiple abiotic stresses in transgenic tobacco. PLoS ONE 8:e65120 Wohlgemuth H, Mittelstrass K, Kschieschan S, Bender J, Weigel HJ, Overmyer K, Kangasjärvi J, Sandermann H, Langebartels C (2002) Activation of an oxidative burst is a general feature of sensitive plants exposed to the air pollutant ozone. Plant Cell Environ 25:717–726 Xie J, Guo G, Wang Y, Hu T, Wang L, Li J, Qiu D, Li Y, Wu Q, Lu P (2020) A rare single nucleotide variant in Pm5e confers powdery mildew resistance in common wheat. New Phytol 228:1011–1026 Xue T, Li X, Zhu W, Wu C, Yang G, Zheng C (2009) Cotton met- allothionein GhMT3a, a reactive oxygen species scavenger, increased tolerance against abiotic stress in transgenic tobacco and yeast. J Exp Bot 60:339–349 Yu Q, Hu Y, Li J, Wu Q, Lin Z (2005) Sense and antisense expres- sion of plasma membrane aquaporin BnPIP1 from brassica napus in tobacco and its effects on plant drought resistance. Plant Sci 169:647–656 Zadoks JC, Chang TT, Konzak CF (1974) A decimal code for the growth stages of cereals. Weed Res 14:415–421 Zhan H, Wang Y, Zhang D, Du C, Zhang X, Liu X, Wang G, Zhang S (2021) RNA-seq bulked segregant analysis combined with 1 3 Theoretical and Applied Genetics (2023) 136:138 Page 21 of 21 138 KASP genotyping rapidly identified PmCH7087 as responsi- ble for powdery mildew resistance in wheat. Plant Genome 14:e20120 Zhang J, Dell B, Biddulph B, Drake-Brockman F, Walker E, Khan N, Wong D, Hayden M, Appels R (2013) Wild-type alleles of Rht-B1 and Rht-D1 as independent determinants of thousand- grain weight and kernel number per spike in wheat. Mol Breed 32:771–783 Zhang D, Liu D, Lv X, Wang Y, Xun Z, Liu Z, Li F, Lu H (2014) The cysteine protease CEP1, a key executor involved in tapetal programmed cell death, regulates pollen development in arabi- dopsis. Plant Cell 26:2939–2961 Zhao Y (2019) Genetic dissection of wheat nitrogen use efficiency related traits: Murdoch University PhD thesis.https:// resea rchpo rtal. murdo ch. edu. au/ esplo ro/ Zuber U, Winzeler H, Messmer M, Keller M, Keller B, Schmid J, Stamp P (1999) Morphological traits associated with lodging resistance of spring wheat (triticum aestivum L.). J Agron Crop Sci 182:17–24 Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. 1 3
null
10.1017_etds.2021.165.pdf
null
null
Ergod. Th. & Dynam. Sys., (2023), 43, 729–793 © The Author(s), 2022. Published by Cambridge University Press. This is an Open Access article, distributed under the terms of the Creative Commons Attribution licence (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted re-use, distribution and reproduction, provided the original article is properly cited. doi:10.1017/etds.2021.165 729 Thermodynamic metrics on outer space TARIK AOUGAB†, MATT CLAY ‡ and YO’AV RIECK‡ † Department of Mathematics, Haverford College, 370 Lancaster Avenue, Haverford, PA 19041, USA (e-mail: taougab@haverford.edu) ‡ Department of Mathematics, University of Arkansas, Fayetteville, AR 72701, USA (e-mail: mattclay@uark.edu, yoav@uark.edu) (Received 27 January 2021 and accepted in revised form 29 November 2021) Abstract. In this paper we consider two piecewise Riemannian metrics defined on the Culler–Vogtmann outer space which we call the entropy metric and the pressure metric. As a result of work of McMullen, these metrics can be seen as analogs of the Weil–Petersson metric on the Teichmüller space of a closed surface. We show that while the geometric analysis of these metrics is similar to that of the Weil–Petersson metric, from the point of view of geometric group theory, these metrics behave very differently than the Weil–Petersson metric. Specifically, we show that when the rank r is at least 4, the action of Out(Fr ) on the completion of the Culler–Vogtmann outer space using the entropy metric has a fixed point. A similar statement also holds for the pressure metric. Key words: outer space, automorphisms of free groups, thermodynamic formalism, Weill–Petersson metric 2020 Mathematics Subject Classification: 20F65 (Primary); 20E05, 57-XX (Secondary) Contents 1 Introduction 1.1 Metrics on outer space 1.2 Thermodynamic metrics Incompletion of the metric 1.3 1.4 The moduli space of the r-rose 1.5 A fixed point in the completion 1.6 Analogous statements for pressure metric 1.7 Further discussion and questions 2 Graphs and outer space 2.1 Graphs 2.2 Outer space 730 731 732 733 734 735 737 737 738 738 739 https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press T. Aougab et al 730 3 Thermodynamic metrics 3.1 Entropy 3.2 Pressure 3.3 Thermodynamic metrics 4 A determinant-defining equation for M1(G) 5 6 7 8 9 4.1 Determinant equation 4.2 A simplification The topology induced by the entropy metric The entropy metric on X1(F2) 6.1 The 2-rose 6.2 The barbell graph 6.3 The theta graph 6.4 (X1(F2), dh) is complete The moduli space of the rose 7.1 M1(Rr ) as a zero locus 7.2 Finite-length paths in M1(Rr ) for r ≥ 3 7.3 The diameter of M1(Rr ) is infinite Proof of Theorem 1.1 The completion of (M1(Rr ), dh,Rr ) 9.1 The model space (cid:2)M1 9.2 Proof of Theorem 1.2 9.3 The thin part of M1(Rr ) (Rr ) 10 The moduli space of a graph with a separating edge 10.1 Finite-length paths in M1(G) 10.2 The model space (cid:2)M1 (G) 11 X1(Rr , id) has bounded diameter in X1(Fr ) 12 Proof of Theorem 1.3 Acknowledgements References 742 742 744 745 748 748 752 754 757 757 759 760 761 762 763 765 768 773 774 774 777 778 781 782 785 788 790 791 791 Introduction 1. The purpose of this paper is to introduce and examine two piecewise Riemannian metrics, called the entropy metric and the pressure metric, on the Culler–Vogtmann outer space CV (Fr ). The Culler–Vogtmann outer space is the moduli space of unit-volume marked metric graphs and as such it is often viewed as the analog of the Teichmüller space of an orientable surface Sg. Both the Culler–Vogtmann outer space and the Teichmüller space admit a natural properly discontinuous action by a group. For the Culler–Vogtmann outer space, the group is the outer automorphism group of a free group Out(Fr ) = Aut(Fr )/ Inn(Fr ). For the Teichmüller space, the group is the mapping class group of the surface MCG(Sg) = π0(Homeo+(Sg)). Strengthening the connection between these spaces and groups are the facts that (i) Out(Fr ) is isomorphic to the group of homotopy equivalences of a graph whose fundamental group is isomorphic to Fr , that is, Out(Fr ) can be thought of as the mapping class group of a graph, and (ii) the Dehn–Nielsen–Baer https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 731 theorem which states that the extended mapping class group MCG±(Sg) (which also includes isotopy classes of orientation-reversing homeomorphisms) is isomorphic to Out(π1(Sg)) [17]. This analogy has led to much fruitful research on the outer automor- phism group of a free group Out(Fr ). The metrics on the Culler–Vogtmann outer space we consider in this paper are analogs to the classical Weil–Petersson metric on the Teichmüller space of an orientable surface. The Weil–Petersson metric has been studied extensively from the point of view of both geometric analysis and geometric group theory. On the one hand, it enjoys many important analytic properties which can be expressed naturally in terms of hyperbolic geometry on Sg. Its utility in geometric group theory then stems from the fact that every isometry of the Weil–Petersson metric is induced by a mapping class [25]. Thus, the action of MCG(Sg) on the Teichmüller space equipped with the Weil–Petersson metric encodes information about useful invariants for mapping classes. As the piecewise Riemannian metrics on the Culler–Vogtmann outer space that we study in this paper are motivated by the classical Weil–Petersson metric on the Teichmüller space of a closed surface, it is natural to ask to what extent they are true analogs of the Weil–Petersson metric. A major takeaway from the work in this paper is that they should be seen as natural analogs from the geometric analysis point of view, but not from the geometric group theory perspective. Specifically, while we highlight some similarities between these metrics and the Weil–Petersson metric as seen from the analytic point of view (Theorems 1.1 and 1.2) the main result (Theorem 1.3) of this paper shows that from the geometric group-theoretic perspective, these metrics are not useful (except possibly when r = 3). The content of this theorem is summarized as follows: the action of Out(Fr ) on the metric completion of the Culler–Vogtmann outer space has a fixed point for r ≥ 4. The remainder of this introduction discusses these metrics more thoroughly and provides context for the main results. 1.1. Metrics on outer space. The topology of CV (Fr ) has been well studied; see, for instance, the survey papers of Bestvina [5] and Vogtmann [35]. The metric theory of CV (Fr ) has been steadily developing over the past decade. What is desired is a theory that reflects the dynamical properties of the natural action by Out(Fr ), that further elucidates the connection between Out(Fr ) and MCG(S), and that leads to useful new discoveries. The metric that has received the most attention to date is the Lipschitz metric. Points in the Culler–Vogtmann outer space are represented by triples (G, ρ, (cid:4)) where G is a finite connected graph, ρ : Rr → G is a homotopy equivalence where Rr is the r-rose, and (cid:4) is a function from the edges of G to (0, ∞) for which the sum of (cid:4)(e) over all edges of G is equal to 1. (See §2.2 for complete details.) We think of the function (cid:4) as specifying the length of each edge and as such (cid:4) determines a metric on G where the interior of each edge e is locally isometric to the interval (0, (cid:4)(e)). The Lipschitz distance between two unit-volume marked metric graphs (G1, ρ1, (cid:4)1) and (G2, ρ2, (cid:4)2) in CV (Fr ) is defined by dLip((G1, ρ1, (cid:4)1), (G2, ρ2, (cid:4)2)) = log inf{Lip(f ) | f : G1 → G2, ρ2 (cid:5) f ◦ ρ1}, (1.1) where Lip(f ) is the Lipschitz constant of the function f : G1 → G2 using the metrics induced by (cid:4)1 and (cid:4)2. respectively. In general the function dLip is not symmetric. As such, https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 732 T. Aougab et al dLip((cid:2), (cid:2)) is not a true metric, but an asymmetric metric. See [1, 2, 20] for more on the asymmetric aspects of the Lipschitz metric. Regardless, the Lipschitz metric has been essential in several recent developments for Out(Fr ). This is in part due to the fact that the Lipschitz metric connects the dynamical properties of an outer automorphism of Fr acting on CV (Fr ) to its action on conjugacy classes—of elements and of free factors—in Fr . Notable are the ‘Bers-like proof’ of the existence of train-tracks by Bestvina [6], the proof of hyperbolicity of the free factor complex by Bestvina and Feighn [8], and the proof of hyperbolicity of certain free group extensions by Dowdall and Taylor [18]. In this way, the Lipschitz metric is akin to the Teichmüller metric on Teichmüller space which was used to prove the corresponding statements for the mapping class group [4, 19, 26]. One can also define the Lipschitz metric on Teichmüller space using the same idea as in (1.1), and in this setting it is oftentimes called Thurston’s asymmetric metric [34]. This metric has seen renewed attention lately, in part due to the usefulness of the Lipschitz metric on CV (Fr ). As a result of McMullen’s interpretation of the Weil–Petersson metric on Teichmüller space via tools from the thermodynamic formalism applied to the geodesic flow on the hyperbolic surface [27, Theorem 1.12], there exists a natural candidate for the Weil–Petersson metric on the Culler–Vogtmann outer space. This idea was originally pursued by Pollicott and Sharp [31]. 1.2. Thermodynamic metrics. The metrics we consider in this paper arise from the tools of the thermodynamic formalism as developed by Bowen [9], Parry and Pollicott [29], Ruelle [32] and others. The central objects involved are the notions of entropy and pressure. For a graph G, these notions define functions h (1.2) G : Rn → R G : M(G) → R and P where n is the number of (geometric) edges in G and M(G) = Rn >0—this space parametrizes the length functions on G. The entropy and pressure functions are real analytic, strictly convex and are related by h G(−(cid:4)) = 0 (see Theorem 3.7). As these functions are smooth and strictly convex, their Hessians induce an inner product on the tangent space of the unit-entropy subspace M1(G) = {(cid:4) ∈ M(G) | G((cid:4)) = 1} at a length function (see Definition 3.10). Hence the notions of entropy and h pressure induce Riemannian metrics on M1(G) which we call the entropy metric and pressure metric, respectively. By dh,G and dP,G we denote the induced distance functions on M1(G). We caution the reader that these metrics have been considered by others with conflicting terminology. Throughout this introduction, we will use the above terminology even when referencing the work of others. See Remark 3.11 for a further discussion. G((cid:4)) = 1 if and only if P Pollicott and Sharp initiated the study of the thermodynamic metrics when they first defined the pressure metric on M1(G) [31]. They proved that the pressure metric is not complete for the 2-rose R2 and they derived formulas for the sectional curvature for the theta graph (cid:5)2 and barbell graph B2 (see Figure 2 for these graphs). Additionally, Pollicott and Sharp produce a dynamical characterization of the entropy metric in terms of generic geodesics similar to Wolpert’s result for the Weil–Petersson metric [37] (see Remark https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 733 3.11). Kao furthered these results by showing that the pressure metric is incomplete for (cid:5)2, B2 and the 3-rose R3, and by showing that the entropy metric is complete for R2 [22]. Additionally, he derives formulas for the sectional curvature with respect to both the entropy and the pressure metric for (cid:5)2, B2 and R3. Xu shows that for certain graphs, the moduli space M1(G) equipped with the entropy metric arises in the completion of the Teichmüller space of a surface with boundary using the pressure metric [39]. In this paper we will investigate the entropy metric not only on the moduli space of a single graph, but on the full moduli space of all marked graphs. Let X(Fr ) be the space of marked metric graphs so that contained in X(Fr ) is the Culler–Vogtmann outer space CV (Fr ). The notion of entropy extends to X(Fr ) by h([(G, ρ, (cid:4))]) = h G((cid:4)) and we set X1(Fr ) = {[(G, ρ, (cid:4))] ∈ X(Fr ) | h([(G, ρ, (cid:4))]) = 1}. (1.3) There is a homeomorphism between CV (Fr ) and X1(Fr ) defined by scaling the length function (see §3.1). Fixing a graph G and a marking ρ : Rr → G, the map M1(G) → X1(Fr ) that sends a length function (cid:4) in M1(G) to the point determined by (G, ρ, (cid:4)) in X1(Fr ) is an embedding whose image we denote by X1(G, ρ). Considering all marked graphs individually, this induces a piecewise Riemannian metric on X1(Fr ). See §3.3 for complete details. We denote the induced distance function on X1(Fr ) by dh. For a closed orientable surface Sg, one can repeat the above discussion using the moduli space of marked Riemannian metrics with constant curvature X(Sg). In this case, the entropy and the area of the Riemann surface are directly related. In particular, the unit-entropy, constant-curvature metrics correspond to the hyperbolic metrics, that is, those with constant curvature equal to −1, and hence to those with area equal to 2π(2g − 2). In other words, the entropy and area normalizations on X(Sg) result in the same subspace, the Teichmüller space of the surface. McMullen proved that the ensuing entropy metric on the Teichmüller space is proportional to the Weil–Petersson metric [27, Theorem 1.12]. It is this connection between the entropy metric and the Weil–Petersson metric that drives the research in this paper. After introducing the framework for both the entropy and the pressure metrics in §3, we specialize the discussion to the entropy metric because of this connection to the Weil–Petersson metric. All of the main results of this paper have analogous statements for the pressure metric and the proofs are similar, and in most cases substantially easier. The statements for the pressure metric are given in §1.6. It is not necessary for this paper, but we mention that building on work of McMullen, Bridgeman [10] and Bridgeman et al [11] used these same ideas to define a metric on the space of conjugacy classes of regular irreducible representations of a hyperbolic group into a special linear group. In the next three subsections, we explain our main results on the entropy metric on X1(Fr ) and their relation to the Weil–Petersson metric on Teichmüller space. Incompletion of the metric. Our first main result concerns the completion of the 1.3. entropy metric on X1(Fr ). Wolpert showed that the Weil–Petersson metric on Teichmüller space is not complete [36]. Our first theorem shows that when r ≥ 3, the same holds for the entropy metric on X1(Fr ). https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 734 T. Aougab et al THEOREM 1.1. The metric space (X1(Fr ), dh) is complete if r = 2 and incomplete if r ≥ 3. For r ≥ 3, this theorem is proved by exhibiting a finite-length path in M1(Rr ) that exits every compact subset. This path is defined by sending the length of one edge in Rr to infinity, while shrinking the others to maintain unit entropy (Proposition 7.8). This strategy—also used by Pollicott and Sharp [31] and Kao [22]—shows that (M1(Rr ), dh,Rr ) is incomplete. We further show in §8 that this path also exits every compact set in X1(Fr ). As dh,Rr is an upper bound to dh (when defined), this path still has finite length when considered in X1(Fr ) and thus Theorem 1.1 follows. The path described above illustrates a general method for producing paths that exit every compact subset and that have finite length in the entropy metric: deform the metric by sending the length of some collection of edges to infinity while shrinking the others to maintain unit entropy. So long as the complement of the collection supports a unit-entropy metric, this path will have finite length. This explains why (X1(F2), dh) is complete: any metric on a graph where every component has rank at most 1 has entropy equal to zero. In §6 we demonstrate the calculations required to prove that (X1(F2), dh) is complete. This is completely analogous to the setting of the Weil–Petersson metric on Teichmüller space. In that setting, deforming a hyperbolic metric on Sg by pinching a simple closed curve results in a path with finite length that exits every compact set. Moreover, the geometric analysis agrees. For the path in M1(Rr ) described above, if we parametrize the long edge by − log(t) as t → 0, then the entropy norm along this path is O(t −1/2), as shown in Proposition 7.8. For the path in the Teichmüller space of Sg, if we parametrize the curve which is being pinched by t as t → 0, then the Weil–Petersson norm along this path is O(t −1/2) [38, §7]. Note that in this case, the length of the shortest curve that intersects the pinched one has length approximately − log(t). 1.4. The moduli space of the r-rose. Our second main result is concerned with the entropy metric on the moduli space of the r-rose Rr . THEOREM 1.2. The completion of (M1(Rr ), dh,Rr ) is homeomorphic to the complement of the vertices of an (r − 1)-simplex. The space M1(Rr ) is homeomorphic to the interior of an (r − 1)-simplex. The faces added in the completion for dh,Rr correspond to unit-entropy metrics on subroses. Such a metric is obtained as the limit of a sequence of length functions on Rr by sending the length of a collection of edges to infinity and scaling the others to maintain unit entropy. Specifically, a (k − 1)-dimensional face of the completion corresponds to the moduli space of unit-entropy metrics on a sub-k-rose. As before, intuitively, the vertices of the (r − 1)-simplex are missing in the completion as there does not exist a unit-entropy metric on R1. That unit-entropy metrics on subroses arise as points in the completion follows from the calculations provided for the proof of incompleteness in Theorem 1.1 and some continuity arguments. This is shown in §9.1. The difficult part of the proof of Theorem 1.2 is showing that any path in (M1(Rr ), dh,Rr ) that sends the length of one edge to 0 (and hence the https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 735 lengths of the other edges to infinity) necessarily has infinite length. This argument appears in Lemma 7.10 and Proposition 7.14. In §9.2 we combine these two facts to complete the proof of Theorem 1.2. In Example 9.7 we compare the completion of (M1(R3), dh,R3) to the closure of the unit-volume metrics on R3 in the axes topology on CV (F3) (see §2.2 for definitions). By Theorem 1.2, the completion in the entropy metric is a 2-simplex without vertices, whereas the closure in the axes topology is a 2-simplex. More interestingly, the newly added edges and vertices are dual: edges in the entropy completion correspond to vertices in the axes closure and the missing vertices in the entropy completion correspond to the edges in the axes closure. This is explained in detail in Example 9.7 and illustrated in Figure 6. While it is not necessary for Theorem 1.2, we mention that in §9.3 we prove that the diameter of a cross-section of the (r − 1)-simplex goes to 0 as the length of one of the edges goes to 0, that is, as the cross-section moves out toward one of the missing vertices. In other words, the completion of (M1(Rr ), dh,Rr ) is geometrically akin to an ideal hyperbolic (r − 1)-simplex; see Lemma 9.9. 1.5. A fixed point in the completion. Whereas the first two main results demonstrate the similarity between the geometric analysis for the Weil–Petersson metric on the Teichmüller space and the entropy metric on the Culler–Vogtmann outer space, our final main result provides a stark contrast between these two metrics with respect to geometric group theory. THEOREM 1.3. The subspace (X1(Rr , id) · Out(Fr ), dh) ⊂ (X1(Fr ), dh) is bounded if r ≥ 4. Moreover, the action of Out(Fr ) on the completion of (X1(Fr ), dh) has a fixed point. This subspace consists of the unit-entropy metrics on every marked r-rose. To illustrate the difference with respect to the setting of the Weil–Petersson metric on Teichmüller space, we mention the fact due to Daskalopoulos and Wentworth that pseudo-Anosov mapping classes have positive translation length in their action on the Teichmüller space [16]. In particular, the action of the mapping class group does not have a fixed point in the completion of Teichmüller space with the Weil–Petersson metric. The first step in the proof of Theorem 1.3 is to show that the image of the inclusion map M1(Rr ) → X1(Rr , id) ⊂ X1(Fr ) has bounded diameter for r ≥ 4. This result is particularly striking in contrast to Theorem 1.2, since implicit in that theorem is that the space (M1(Rr ), dh,Rr ) has infinite diameter. Boundedness of the image of M1(Rr ) is achieved by showing that the map induced via Theorem 1.2, (cid:6) : (cid:7)r−1 − V → (cid:3)X1 (Fr ), extends to (cid:7)r−1, where (cid:7)r−1 is an (r − 1)-simplex, V ⊂ (cid:7)r−1 is the set of vertices, and (cid:3)X1 (Fr ) is the completion of X1(Fr ) for dh. The existence of this extension shows that X1(Rr , id) lies in a compact set and hence is bounded. In order to show that (cid:6) : (cid:7)r−1 − V → (cid:3)X1 (Fr ) extends to the set V, we show that (cid:6) maps every (k − 1)-dimensional face of (cid:7)r−1 − V to a single point when 1 < k < r − 1. This is shown in §11. This fact, together with the previously mentioned fact about the diameter of the cross-sections going to 0, gives that (cid:6) extends to the set V and that the entire (r − 3)-skeleton of (cid:7)r−1 is mapped to a single point in (cid:3)X1 (Fr ). The collapse of a https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 736 T. Aougab et al G2,2 = a b c d a c d a, b constant c, d → ∞ linearly b unit entropy a, b constant c, d → ∞ linearly a b c d M1(R2) a b a b b, c, d constant a grows linearly a b c d a, c, d constant b grows linearly FIGURE 1. Illustration of a path with length 0 in the completion of (M1(G2,2), dh,G2,2 ). (k − 1)-dimensional face of (cid:7)r−1 − V for 1 < k < r − 1 arises from paths in X1(Fr ) connecting points in X1(Rr , id) whose length is much shorter than paths in M1(Rr ) connecting the same points. In other words, there are shortcuts present in X1(Fr ) that are not present in M1(Rr ). These shortcuts are most easily understood in terms of unit-entropy metrics on marked subgraphs, that is, points in the completion of (X1(Fr ), dh). Pathologies arise when the subgraph is not connected. In this case, the entropy of the metric on the subgraph is the maximum of the entropy—in the previous sense—on a component of the subgraph. Hence, by holding the length function constant on a component of the subgraph with unit entropy, we are free to modify the length function on the other components at will, so long as the entropy is never greater than 1 on any of these components. In Proposition 3.12 we show that the entropy and pressure metrics can be computed using the second derivatives of the lengths of edges along a path. Hence the length of a path that changes the length of the edges in a component with entropy less than 1 linearly has zero length in either of these metrics. Figure 1 illustrates the central idea that is exploited in §10 to show that many paths have zero length. This figure is taking place in the completion of M1(G2,2) using the metric dh,G2,2. This space has M1(R4) as a face in X1(F4) corresponding to the collapse of the separating edge. The completion of M1(R4) has an edge corresponding to unit-entropy metrics on two of the edges (denoted a and b in Figure 1). This edge also corresponds to a subset in the completion of M1(G2,2). Illustrated in Figure 1 is a path through unit-entropy length functions on subgraphs of G2,2, that is, points in the completion. As all edge lengths are changing linearly, this path has length 0 and hence all of these length functions correspond to the same point in the completion. This shows that the edge corresponding to M1(R2) is mapped by (cid:6) to a point. The same idea works for any sub-k-rose of Rr so long as 1 < k < r − 1: it is necessary to separate https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 737 two subroses, each of which supports a unit-entropy metric. This is the reason why we require r ≥ 4 in Theorem 1.3. Once we know that the entire (r − 3)-skeleton of (cid:7)r−1 is mapped by (cid:6) to a point in (cid:3)X1 (Fr ), we utilize the structure of the Culler–Vogtmann outer space to conclude in §12 that this point is independent of the marking ρ : Rr → Rr used to define the inclusion M1(Rr ) → X1(Fr ). This completes the proof of Theorem 1.3. 1.6. Analogous statements for pressure metric. For the pressure metric on X1(Fr ) we have the following analogs of Theorems 1.1–1.3. By dP we denote the induced distance function. (1) The space (X1(Fr ), dP) is incomplete for r ≥ 2. (2) The completion of (M1(Rr ), dP,Rr ) is homeomorphic to an (r − 1)-simplex. (3) The space (X1(Fr ), dP) is bounded if r ≥ 2; moreover, the action of Out(Fr ) on the completion of (X1(Fr ), dP) has a fixed point. These can be shown using techniques similar—and simpler—to those in this paper. The key source of the distinction between the entropy and pressure metrics is that the length function that assigns 0 to the unique edge on R1 has pressure equal to 0 even through the entropy is not defined. Hence the path in M1(R2) that sends the length of one edge to infinity while shrinking the length of the other (necessarily to 0) to maintain unit entropy has finite length in the pressure metric, whereas the length in the entropy metric is infinite. 1.7. Further discussion and questions. This work raises a number of questions. Our proof that the action of Out(Fr ) on the completion of (X1(Fr ), dh) has a fixed point relies heavily on the assumption that r ≥ 4: the key construction uses an edge that separates a given graph into two subgraphs, each with rank at least 2. This leaves the door open to a negative answer for the following question, which would allow for interesting applications specifically for F3. Question 1.4. Does (X1(F3), dh) admit an Out(F3)-invariant bounded subcomplex? Theorem 1.3 demonstrates the existence of an Out(Fr ) orbit in (X1(Fr ), dh) with bounded diameter but we do not yet know that the entire space has bounded diameter. We therefore ask the following question. Question 1.5. Is (X1(Fr ), dh) bounded for r ≥ 4? We believe the answer to this question is yes. Indeed, the only way (X1(Fr ), dh) could fail to be bounded is if the subspace (X1(G, ρ), dh) has infinite diameter for some marked graph ρ : Rr → G. As the diameter of (X1(Rr , id) · Out(Fr ), dh) is bounded, to answer the question in the affirmative, it would suffice to find a bound (in terms of r) on the distance from any point X1(Fr ) to a point in X1(Rr , id) · φ for some φ ∈ Out(Fr ). Another approach to answer Question 1.5 in the affirmative would be to show the existence of a bound (in terms of r) on distance from any point in X1(G, ρ) to a completion point represented by a unit-entropy metric on a proper subgraph (with the goal of getting to a point in the completion of a marked rose via induction). This led us to the following question, which is of independent interest and we pose here as a conjecture. https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 738 T. Aougab et al Conjecture 1.6. For any r ≥ 3 there exists C > 0 so that any metric graph of rank r with unit entropy contains a proper subgraph with entropy at least C. It suffices to show the conjecture for a fixed topological type of graph since, for a given rank r, there are only finitely many topological types of graph of rank r. One can also define the notion of the entropy metric on the Teichmüller space of a surface with boundary. In [39], Xu shows that this metric is incomplete. As mentioned previously, McMullen proved that for closed surfaces, the entropy metric is a constant multiple of the Weil–Petersson metric. However, by partially characterizing the completion of the entropy metric in the bordered setting, Xu is able to show that this is not true in the presence of boundary. Concretely, Xu identifies certain graphs G so that, in the notation of this paper, (M1(G), dh) isometrically embeds in the completion of the Teichmüller space of the surface equipped with the entropy metric. We therefore ask if the work in this paper can be used to fully understand the completion of the Teichmüller space of a bordered surface equipped with the entropy metric. Problem 1.7. Fully characterize the completion of (M1(G), dh,G) for an arbitrary graph G and use this to study the completion of the Teichmüller space of a bordered surface, equipped with the entropy metric. The pathology exhibited by Theorem 1.3 relies on the existence of a sequence of unit-entropy length functions whose limiting metric is supported on a subgraph with multiple components where the metric on some component need not have entropy equal to 1. This behavior does not occur in the Teichmüller space of a closed surface since a unit-entropy metric on a constant-curvature surface is a hyperbolic metric and vice versa, and thus for the subsurface supporting the limit of a sequence of unit-entropy metrics, the metric on each component also has entropy equal to 1. One can also consider an entropy function defined over the moduli space of singular flat metrics on a closed surface. This setting appears more similar to the situation of Theorem 1.3 in that the unit-entropy condition is not encoded by the local geometry. It appears likely that some version of Theorem 1.3 holds for singular flat metrics, and so we therefore ask our final question of this introduction. Question 1.8. Can the techniques used in this paper in the setting of metric graphs apply to the study of an entropy metric on the moduli space of singular flat metrics on a closed surface? 2. Graphs and outer space In this section we introduce some concepts that are necessary for the sequel. First, we set some notation for dealing with graphs. Then we define the Culler–Vogtmann outer space—including its topology—and the Out(Fr ) action on this space. 2.1. Graphs. We use Serre’s convention for graphs [33]. That is, an (undirected) graph is a tuple G = (V , E, o, τ , ¯) where: (1) V and E are sets, called the vertices and the directed edges (we think of E as containing two copies, with opposite orientations, of each undirected edge); https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 739 (2) o, τ : E → V are functions that specify the originating and terminating vertices of an edge; ¯ : E → E is a fixed point free involution such that o(e) = τ ( ¯e) (¯ flips edges). (3) We fix an orientation on G, that is, a subset E+ ⊂ E that contains exactly one edge from each pair {e, ¯e}. Since we consider the pair {e, ¯e} to be a single edge, the number of edges of G is |E+| = |E|/2. The valance of a vertex v is the number of edges from e ∈ E+ with o(e) = v plus the number of edges from e ∈ E+ with τ (e) = v (an edge e for which o(e) = τ (e) = v contributes 2 to the valance). Oftentimes when defining a graph we only specify the edges in E+ (together with the restrictions of o and τ to E+). The complete set of edges is then defined as E = E+ ∪ E+, where E+ is a copy of E+, and o, τ , and ¯ are defined in the obvious way. We blur the distinction between the tuple (V , E, o, τ , ¯) and the corresponding one-dimensional CW-complex with 0-cells V and 1-cells E+. The space of length functions on G is the open convex cone M(G) = {(cid:4) : E+ → R>0}. (2.1) We consider this set as a subset of R|E+|. A length function (cid:4) : E+ → R>0 extends to a function (cid:4) : E → R>0 by (cid:4)(e) = (cid:4)( ¯e) if e /∈ E+. By 1 ∈ M(G) we denote the constant function with value 1. An edge path is a sequence of edges (e1, . . . , en) in E such that τ (ei) = o(ei+1) for i = 1, . . . , n − 1. A function f : E → R (in particular, a length function) extends to a function on edge paths γ = (e1, . . . , en) by f (γ ) = n(cid:4) i=1 f (ei). (2.2) 2.2. Outer space. We will introduce some definitions and notation for the Culler– Vogtmann outer space. This space was originally defined by Culler and Vogtmann [14]. For more information, see for example the survey papers by Vogtmann [35] or Bestvina [5]. Let Rr be the r-rose. That is, Rr the graph with a unique vertex v and r edges. Fix ∼= π1(Rr , v). A marked metric graph (of rank r) is a triple (G, ρ, (cid:4)) an isomorphism Fr where: (1) G is a finite connected graph without vertices of valence 1 or 2; (2) (3) There is an equivalence relation on the set of marked metric graphs defined by (G1, ρ1, (cid:4)1) ∼ (G2, ρ2, (cid:4)2) if there exists a graph automorphism α : G1 → G2 such that (cid:4)1 = (cid:4)2 ◦ α and such that the following diagram commutes up to homotopy: ρ : Rr → G is a homotopy equivalence; and (cid:4) is a length function on G. Rr ρ1 (cid:3)(cid:2)(cid:2)(cid:2)(cid:2)(cid:2)(cid:2) (cid:4)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3) ρ2 G1 α G2 https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press (cid:2) (cid:2) (cid:3) (cid:4) 740 T. Aougab et al We let X(Fr ) denote the set of equivalence classes of marked metric graphs of rank r. The group Out(Fr ) acts on X(Fr ) on the right by precomposing the marking. Specifically, for any outer automorphism φ ∈ Out(Fr ), there is a homotopy equivalence ∼= gφ : Rr → Rr that induces φ on π1(Rr ) via the aforementioned fixed isomorphism Fr π1(Rr , ∗). Moreover, this homotopy equivalence is unique up to homotopy. With this, we define (G, ρ, (cid:4)) · φ = (G, ρ ◦ gφ, (cid:4)). (2.3) This action respects the equivalence relation on marked metric graphs and so defines an action on X(Fr ) as claimed. Let Gr denote the set of finite connected graphs without vertices of valence 1 or 2 whose fundamental group has rank r. We observe that this is a finite set. Given a graph G ∈ Gr and homotopy equivalence ρ : Rr → G, we set X(G, ρ) = {[(G0, ρ0, (cid:4)0)] ∈ X(Fr ) | G0 = G and ρ0 (cid:5) ρ}. There is a bijection X(G, ρ) → M(G) defined by [(G0, ρ0, (cid:4)0)] (cid:12)→ (cid:4)0. These sets partition the set X(Fr ) and are permuted under the action by Out(Fr ). Specifically, for each G ∈ Gr we fix a marking ρG : Rr → G. Then (cid:5) (cid:5) X(Fr ) = X(G, ρG) · φ. G∈Gr φ∈Out(Fr ) There is a topology on X(Fr ) that is often defined in three different ways. We will need to use the first two and for completeness we explain all three here. The weak topology. The notion of a collapse induces a partial order on the set of marked graphs. Specifically, for two graphs G and G0, we say that G collapses to G0 if there is a surjection c : G → G0 such that the image of any edge in G is either a vertex or an edge of G0 and such that c−1(x) is a contractible subgraph of G for each point x of G0. The map c is a called a collapse. Observe that if the map c : G → G0 is a collapse, then a length function (cid:4) ∈ M(G0) can be considered as a degenerate length function (cid:4)G0 on G by (cid:6) (cid:4)G0(e) = (cid:4)(c(e)) 0 if c(e) is an edge in G0, otherwise. (2.4) This defines a map c∗ : M(G0) → R|E+| subset of R|E+| ≥0 : ≥0 by c∗((cid:4)) = (cid:4)G0. We now define the following M(G) = (cid:5) c : G→G0 ∗ c (M(G0)). (2.5) We note the M(G) is a subset of M(G) as the identity map id : G → G is a collapse. Next, given two marked graphs ρ : Rr → G and ρ0 : Rr → G0, we say that (G, ρ) collapses to (G0, ρ0) if there is a collapse c : G → G0 such that ρ0 (cid:5) c ◦ ρ. In this case https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space we write (G0, ρ0) ≤ (G,ρ). We now define the following subset of X(Fr ): (cid:5) X(G, ρ) = X(G0, ρ0). 741 (2.6) (G0,ρ0)≤(G,ρ) The bijection X(G, ρ) → M(G) extends in a natural way to a bijection X(G, ρ) → M(G) and allows us to consider X(G, ρ) as a subset of R|E+| ≥0 . The weak topology is defined using this collection of subsets. Specifically, a set U ⊆ X(Fr ) is open if U ∩ X(G, ρ) is open as a subset of R|E+| ≥0 The axes topology. Given a marked metric graph (G, ρ, (cid:4)) and an element g ∈ Fr , we denoted by (cid:4)([g]) the (cid:4)-length of the shortest loop in G representing the conjugacy class [ρ(g)]. This induces a function Len : X(Fr ) → RFr ≥0 where Len([(G, ρ, (cid:4))]) : Fr → R≥0 is the function defined by for all marked graphs (G, ρ). Len([(G, ρ, (cid:4))])(g) = (cid:4)([g]). Culler and Morgan proved that the map Len is injective [13, 3.7 Theorem]. The resulting subspace topology on Len(X(Fr )) ⊂ RFr ≥0 is called the axes topology. It is known that this topology agrees with the weak topology. (See [14, §1.1] or [21, Proposition 5.4].) The equivariant Gromov–Hausdorff topology. We will not need this definition, and we only remark that Paulin showed that it is equivalent to the axes topology [30, Main Theorem]. There is an action of R>0 on X(Fr ) given by scaling the length function. Specifically, a · (G, ρ, (cid:4)) = (G, ρ, a · (cid:4)). The quotient of X(Fr ) is denoted PX(Fr ). There are many continuous sections of the quotient map X(Fr ) → PX(Fr ). An often used choice uses the notion of the volume of a length function, vol((cid:4)), that we define now. For a length function (cid:4) ∈ M(G), we define the volume of (cid:4) by vol((cid:4)) = e∈E+ (cid:4)(e). There is a section V : PX(Fr ) → X(Fr ) defined by (cid:7) V([[(G, ρ, (cid:4))]]) = G, ρ, (cid:8)(cid:9) (cid:10)(cid:11) . 1 vol((cid:4)) (cid:4) We denote the image of this section by CV (Fr ); it is known as the Culler–Vogtmann outer space. Further, given a marked graph ρ : Rr → G, we set CV (G, ρ) = X(G, ρ) ∩ CV (Fr ). This set is homeomorphic to an open simplex of dimension |E+| − 1. Example 2.1. There are three graphs in G2: the 2-rose R2, the theta graph (cid:5)2 and the barbell graph B2; see Figure 2. Figure 3 shows a portion of CV (F2) and how these simplices piece together. The homotopy equivalences used for Figure 3 are as follows: ρ(cid:5) : R2 → (cid:5)2 : e1 (cid:12)→ e1 ¯e3, e2 (cid:12)→ e2 ¯e3; ρB : R2 → B2 : e1 (cid:12)→ e1, e2 (cid:12)→ e3e2 ¯e3; a : R2 → R2 : e1 (cid:12)→ e1, e2 (cid:12)→ ¯e2; b : R2 → R2 : e1 (cid:12)→ ¯e2, e2 (cid:12)→ e1 ¯e2. Notice that ρ(cid:5) is the homotopy inverse to the map (cid:5)2 → R2 that collapses the edge e3. Likewise ρ(cid:5) ◦ b is the homotopy inverse to the collapse of e2 and ρ(cid:5) ◦ b2 is the homotopy https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 742 T. Aougab et al R2 v e1 e2 B2 e3 v e1 w v e2 Θ2 e1 e2 e3 w FIGURE 2. The three homeomorphism types of graphs in G2. CV (Θ2, ρΘ ◦ a) CV (Θ2, ρΘ ◦ a) CV (Θ2, ρΘ ◦ a) CV (Θ2, ρΘ ◦ a) C C C C V V V V (R (R (R (R 2, 2, 2, 2, id) id) id) id) b) b) b) b) , , , , (R2 (R2 (R2 (R2 V V V V C C C C CV (Θ2, ρΘ) CV (Θ2, ρΘ) CV (Θ2, ρΘ) CV (Θ2, ρΘ) CV (R2, b2) CV (R2, b2) CV (R2, b2) CV (R2, b2) C C C C V V V V (B (B (B (B 2, 2, 2, 2, ρ ρ ρ ρ B) B) B) B) C C C C V V V V (R (R (R (R 2, 2, 2, 2, id) id) id) id) FIGURE 3. A portion of the Culler–Vogtmann outer space CV (F2). inverse to the collapse of e1. Similarly, ρB is the homotopy inverse to the map B2 → R2 that collapses e3. One of the goals of this paper is to investigate a different continuous section of the G((cid:4)), defined in §3.1. Using this notion, quotient map. This uses the notion of entropy, h there is a section H : PX(Fr ) → X(Fr ) defined by H([[(G, ρ, (cid:4))]]) = [(G, ρ, h G((cid:4))(cid:4))]. We will denote the image of this section by X1(Fr ). 3. Thermodynamic metrics In this section we introduce the entropy and pressure of a length function in M(G), for a graph G as in §2.1. By normalizing the entropy to be equal to 1, we realize X1(Fr ) (as defined in §2.2) as a section of X(Fr ) → PX(Fr ); it will follow that X1(Fr ) is homeomorphic to CV (Fr ) (see Theorem 3.5). We use entropy and pressure to construct piecewise Riemannian metrics on X1(Fr ), which we call the thermodynamic metrics. Pollicott and Sharp were the first to consider one of these metrics [31]. Kao [22] and Xu [39] have also investigated these metrics. In these papers, the metric is only considered for a single marked graph and never on the entire outer space, as we will do here. 3.1. Entropy. Fix a finite connected graph G = (V , E, o, τ , ¯). An edge path (e1, . . . , en) in a graph G is reduced if ei (cid:16)= ¯ei+1 for i = 1, . . . , n − 1. A reduced edge path (e1, . . . , en) is a based circuit if τ (en) = o(e1) and en (cid:16)= ¯e1. The set of all based https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 743 circuits in G is denoted by C(G). For a length function (cid:4) ∈ M(G) and a real number t ≥ 0, define CG,(cid:4)(t) = {γ ∈ C(G) | (cid:4)(γ ) ≤ t}. Definition 3.1. The entropy of a length function (cid:4) ∈ M(G) is h G((cid:4)) = lim t→∞ 1 t log|CG,(cid:4)(t)|. Remark 3.2. We defined entropy as the growth rate of the number of reduced based circuits. In the literature, there exist many equivalent definitions of entropy. In particular, one can count the growth of reduced edge paths in G starting at a particular vertex and the adjective ‘based’ can be removed from the count of circuits [24, Proposition 2.3]. This shows that h G((cid:4)) equals the volume entropy of (G, (cid:4)). The volume entropy is defined as the exponential growth rate of the volume of balls in ((cid:12)G, g(cid:4)), where (cid:12)G is the universal cover of G and g(cid:4) is the piecewise Riemannian metric obtained by pulling back the length function (cid:4). That is, 1 t where B(x, t) is the ball of radius t centered at x ∈ (cid:12)G, which is an arbitrary basepoint. G((cid:4)) = lim t→∞ log volg(cid:4) B(x, t) h Example 3.3. The number of reduced edge paths in Rr with 1-length equal to n is exactly 2r(2r − 1)n−1. Thus for any vertex v ∈ (cid:12)Rr we have volg1 B(v, n) = r r − 1 ((2r − 1)n − 1). Hence hRr (1) = log(2r − 1). The next lemma shows that entropy is homogeneous of degree −1 and thus any length G(a · (cid:4)) = 1 if and function (cid:4) ∈ M(G) can be scaled to have unit entropy. Specifically, h only if a = h G((cid:4)). LEMMA 3.4. Let G be a finite connected graph and fix (cid:4) ∈ M(G). If a ∈ R>0, then G(a · (cid:4)) = 1 a G((cid:4)). h h Proof. We reparametrize the limit defining entropy by setting s = at. Then h G((cid:4)) = lim t→∞ = lim s→∞ 1 t a s log|{γ ∈ C(G) | (cid:4)(γ ) ≤ t}| log|{γ ∈ C(G) | a · (cid:4)(γ ) ≤ s}| = ah G(a · (cid:4)). Entropy defines an Out(Fr )-invariant function on X(Fr ) by h([(G, ρ, (cid:4))]) = h G((cid:4)). This function was investigated by Kapovich and Nagnibeda, who showed the following theorem. THEOREM 3.5. [23, Theorem A] The entropy function h : X(Fr ) → R is continuous. https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 744 T. Aougab et al In particular, the map H : PX(Fr ) → X(Fr ) defined by normalizing to have unit entropy, H([[(G, ρ, (cid:4))]]) = [(G, ρ, h G((cid:4))(cid:4))], is a section. Hence the image X1(Fr ) = {[(G, ρ, (cid:4))] ∈ X(Fr ) | h phic to CV (Fr ). G((cid:4)) = 1} is homeomor- 3.2. Pressure. Fix a finite connected graph G = (V , E, o, τ , ¯). We assume throughout this subsection and the next that χ(G) < 0 (where χ(G) = |V | − 1 |E| is the Euler 2 characteristic of G—note the 1 2 factor is present as E includes edges with both orientations) and that G has no vertices with valence equal to 1 or 2. Define AG ∈ Mat|E|(R) by AG(e, e (cid:17) ) = (cid:6) 1 0 if τ (e) = o(e(cid:17)) and ¯e (cid:16)= e(cid:17), otherwise. (3.1) G(e, e(cid:17)) is the number of reduced edge paths of the form It follows that the entry An (e1, . . . , en) where e1 = e, τ (en) = o(e(cid:17)) and ¯en (cid:16)= e(cid:17). In particular, tr(An G) is the number of based edge circuits with 1-length equal to n. Denoting the spectral radius of a matrix by spec((cid:2)), we get, from the definition of entropy, that h G(1) = log(spec(AG)). (3.2) We remark that the above assumptions on G ensure that AG is irreducible. In order to get a matrix that incorporates the metric and is related to entropy, we scale the rows of AG as follows: given a function f : E → R, we define AG,f ∈ Mat|E|(R) by As for AG, it follows that An form γ = (e1, . . . , en) where e1 = e, τ (en) = o(e(cid:17)) and ¯en (cid:16)= e(cid:17). (cid:17) (cid:17) ) = AG(e, e AG,f (e, e (3.3) G,f (e, e(cid:17)) is the sum of exp(−f (γ )) over all edge paths of the ) exp(−f (e)). Definition 3.6. The pressure of a function f : E → R is defined as P log spec(AG,−f ). G(f ) = By equation (3.2) we have that P G(0) = h G(1) as AG,−0 = AG, where 0 is the zero function. The connection between entropy and pressure is given by the following theorem. THEOREM 3.7. Suppose that G = (V , E, o, τ , ¯) is a finite connected graph. Then the following statements hold. (1) For any length function (cid:4) ∈ M(G), P (2) (3) G : R|E+| → R is real analytic and convex. G : M(G) → R is real analytic and strictly convex. The pressure function P The entropy function h G(−(cid:4)) = 0 if and only if h G((cid:4)) = 1. Proof. The first item appears in the work of Pollicott and Sharp [31, Lemma 3.1(2)]. The properties of pressure stated in the second item can be found in the work of Parry and Pollicott [29, Propositions 4.7 and 4.12]. https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 745 The properties of entropy stated in the third item can be found in the work of McMullen [28, Proposition A.4]. Kapovich and Nagnibeda gave an alternative proof of the real analyticity of h G [23]. Let M1(G) = {(cid:4) ∈ M(G) | h natively M1(G) = {(cid:4) ∈ M(G) | P submanifold of R|E+| we need to argue that 1 is a regular value of h the following lemma. We denote the standard Euclidean inner product on Rn by (cid:18)(cid:2), (cid:2)(cid:19). G((cid:4)) = 1}. By the first item above, we have that alter- G(−(cid:4)) = 0}. To see that M1(G) is a codimension-1 G. This follows from LEMMA 3.8. Let G be a finite connected graph and fix (cid:4) ∈ M(G). Then (cid:18)(cid:4), ∇h −h G((cid:4)). G((cid:4))(cid:19) = Proof. This follows from the homogeneity of the entropy function (Lemma 3.4). Indeed, (cid:18)(cid:4), ∇h G((cid:4))(cid:19) = lim s→0 h G((cid:4)) h = lim s→0 G((cid:4) + s(cid:4)) − h s G((cid:4)) − h s h 1 1+s G((cid:4)) = h G((cid:4)) lim s→0 G((cid:4)) G((1 + s)(cid:4)) − h s −s s(s + 1) = −h G((cid:4)). = lim s→0 We record the following properties of the partial derivatives and the gradient of the pressure function. Given a function f : R|E+| → R and an edge e ∈ E+, we denote the partial derivative of f with respect to the eth coordinate by ∂ef . Let (cid:21)(cid:2)(cid:21)1 denote the usual L1-norm on vectors in Rn. LEMMA 3.9. Let G = (V , E, o, τ , ¯) be a finite connected graph and fix (cid:4) ∈ M(G). Then the following statements hold. (1) (2) G((cid:4)) > 0 for any e ∈ E+. G((cid:4))(cid:21)1 = 1. ∂eP (cid:21)∇P Proof. By the Perron–Frobenius theorem, the spectral radius of AG,(cid:4) is realized by a positive, real, simple eigenvalue λ. Let v ∈ R|E| be a corresponding positive left eigenvector so that vAG,(cid:4) = λv. Consider the matrix QG,(cid:4) ∈ Mat|E|(R) defined by QG,(cid:4)(e, e(cid:17)) = (v(e)/λv(e(cid:17)))AG,(cid:4)(e, e(cid:17)). Again, by the Perron–Frobenius theorem, as this matrix is column stochastic, there is a positive vector p ∈ R|E| with QG,(cid:4)p = p and (cid:21)p(cid:21)1 = 1. As explained by Parry and Pollicott, we have that ∂eP G((cid:4)) = p(e) + p( ¯e) [29, Ch. 2, Remark 1 and Proposition 4.10]. Items (3.9) and (3.9) readily follow. 3.3. Thermodynamic metrics. Fix a finite connected graph G = (V , E, o, τ , ¯). As in the previous section, we assume that χ(G) < 0 and that G has no vertices with valence equal to 1 or 2. The tangent space T(cid:4)M1(G) at the length function (cid:4) ∈ M1(G) is the space of vectors v ∈ R|E+| such that (cid:18)v, ∇h G((cid:4))(cid:19) = 0. The tangent bundle T M1(G) is the subspace of M1(G) × R|E+| consisting of pairs ((cid:4), v) where v ∈ T(cid:4)M1(G). We now define two Riemannian metrics on M1(G). We denote the Hessian (that is, the matrix of second derivatives) of a smooth function f : Rn → R by H[f (x)]. https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 746 T. Aougab et al Definition 3.10. Given a length function (cid:4) ∈ M1(G) and tangent vectors v1, v2 ∈ T(cid:4)M1(G) we define the entropy metric by and the pressure metric by (cid:18)v1, v2(cid:19)h,G = (cid:18)v1, H[h G((cid:4))]v2(cid:19), (cid:18)v1, v2(cid:19)P,G = (cid:18)v1, H[P G(−(cid:4))]v2(cid:19). The associated norms on the tangent bundle T M1(G) are denoted by G((cid:4))]v(cid:19) and (cid:21)((cid:4), v)(cid:21)2 = (cid:18)v, H[h (cid:21)((cid:4), v)(cid:21)2 = (cid:18)v, H[P P,G h,G G(−(cid:4))]v(cid:19). By Theorem 3.7(3) we have that (cid:18)(cid:2), (cid:2)(cid:19)h,G is positive definite. Positive definiteness of (cid:18)(cid:2), (cid:2)(cid:19)P,G on T(cid:4)M1(G) has been noted by others, but also follows from the positive definiteness of (cid:18)(cid:2), (cid:2)(cid:19)h,G by Proposition 3.12. Remark 3.11. Other authors have considered these metrics with different and conflicting terminology. We discuss this now using the notation introduced above. Pollicott and Sharp defined (cid:21)(cid:2)(cid:21)P,G, calling it the Weil–Petersson metric [31]. Kao defined (cid:21)(cid:2)(cid:21)h,G, calling it the Weil–Petersson metric, and also studied (cid:21)(cid:2)(cid:21)P,G, calling it the pressure metric [22]. Xu considered (cid:21)(cid:2)(cid:21)h,G, calling it the pressure metric [39]. We use the terminology as stated in Definition 3.10 as it accurately reflects the functions on which the metrics are based. The definitions of these metrics in the literature are not those as given in Definition 3.10, but are equivalent as can be seen by Proposition 3.12. We note that Theorem 3 in the paper by Pollicott and Sharp [31] holds for the metric (cid:21)(cid:2)(cid:21)h,G and not for (cid:21)(cid:2)(cid:21)P,G as claimed. The following proposition shows that these metrics lie in the same conformal class and that they can be calculated using the second derivative along a path. This feature is essential particularly for the material in §10. PROPOSITION 3.12. Let G be a finite connected graph. If (cid:4)t : (−1, 1) → M1(G) is a smooth path, then (cid:21)((cid:4)t , ˙(cid:4)t )(cid:21)2 and (cid:21)((cid:4)t , ˙(cid:4)t )(cid:21)2 = −(cid:18) ¨(cid:4)t , ∇h = (cid:18) ¨(cid:4)t , ∇P G(−(cid:4)t )(cid:19). G((cid:4)t )(cid:19) P,G h,G Additionally, given a length function (cid:4) ∈ M1(G) and tangent vectors v1, v2 ∈ T(cid:4)M1(G), we have (cid:18)v1, v2(cid:19)h,G = (cid:18)v1, v2(cid:19)P,G (cid:18)(cid:4), ∇P G(−(cid:4))(cid:19) . G(−(cid:4)t ) = 0 with respect Proof. Differentiating the equation P ∇P G(−(cid:4)t )(cid:19) = 0. Differentiating again, we find that G(−(cid:4)t )(cid:19) − (cid:18) ˙(cid:4)t , H[P (cid:18) ¨(cid:4)t , ∇P G(−(cid:4)t )] ˙(cid:4)t (cid:19) = 0. to t, we have (cid:18) ˙(cid:4)t , Hence (cid:21)((cid:4)t , ˙(cid:4)t )(cid:21)2 P = (cid:18) ˙(cid:4)t , H[P(−(cid:4)t )] ˙(cid:4)t (cid:19) = (cid:18) ¨(cid:4)t , ∇P(−(cid:4)t )(cid:19) as claimed. The proof of the analogous statement for the entropy norm is similar, observing that G((cid:4)t ) = 1. h https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 747 By Lemma 3.8, we have that (cid:18)(cid:4), ∇h G(−(cid:4)) is non-zero by Lemma 3.9 and hence is parallel to ∇h G((cid:4))(cid:19) = −1 for any (cid:4) ∈ M1(G). Further, we have G((cid:4)) by Theorem that ∇P 3.7(1). Hence we find that (cid:21)((cid:4)t , ˙(cid:4)t )(cid:21)2 h,G = − (cid:18) ¨(cid:4)t , ∇h G((cid:4)t )(cid:19) = (cid:18) ¨(cid:4)t , ∇h (cid:18)(cid:4)t , ∇h G((cid:4)t )(cid:19) G((cid:4)t )(cid:19) = (cid:18) ¨(cid:4)t , ∇P (cid:18)(cid:4)t , ∇P G(−(cid:4)t )(cid:19) G(−(cid:4)t )(cid:19) = (cid:21)((cid:4)t , ˙(cid:4)t )(cid:21)2 (cid:18)(cid:4)t , ∇P G(−(cid:4)t )(cid:19). P,G By polarization, the norm determines the inner product and so the claim follows. Positive definiteness of the Hessians follows from strict convexity of h M(G) and R|E+|, respectively (Theorem 3.7). G and P G on Using these norms, we can define the entropy or pressure length of a piecewise smooth path (cid:4)t : [t0, t1] → M1(G) by Lh,G((cid:4)t |[t0, t1]) = LP,G((cid:4)t |[t0, t1]) = (cid:13) t1 t0 (cid:13) t1 t0 (cid:21)((cid:4)t , ˙(cid:4)t )(cid:21)h,G dt, (cid:21)((cid:4)t , ˙(cid:4)t )(cid:21)P,G dt. These induce the entropy and pressure distance functions on M1(G) by dh,G(x, y) = inf{Lh,G((cid:4)t |[0, 1]) | (cid:4)t : [0, 1] → M1(G), (cid:4)0 = x, (cid:4)1 = y}, dP,G(x, y) = inf{LP,G((cid:4)t |[0, 1]) | (cid:4)t : [0, 1] → M1(G), (cid:4)0 = x, (cid:4)1 = y}. Given a marked graph (G, ρ), we set X1(G, ρ) = X(G, ρ) ∩ X1(Fr ). Using the natural bijection X1(G, ρ) ↔ M1(G), we get metrics and distance functions on X1(G, ρ) that we denote using the same notation as above. Next, we explain how these fit together to get distance functions on X1(Fr ). Suppose αt : [0, 1] → X1(Fr ) is a piecewise smooth path and that there is a partition t1 = 0 < t2 < · · · < tn+1 = 1 and marked graphs (Gk, ρk) for k = 1, . . . , n such that αt ∈ X1(Gk, ρk) for t ∈ (tk, tk+1). We set Lh(αt ) = n(cid:4) k=1 Lh,Gk (αt |(tk, tk+1)) and LP(αt ) = n(cid:4) k=1 LP,Gk ((αt |(tk, tk+1)). These define distance functions on X1(Fr ) as usual—by taking the infimum of the lengths of paths—that we denote by dh and dP. It is obvious that the proposed distance functions are symmetric and satisfy the triangle inequality; positive definiteness of the Hessians implies non-degeneracy. However, it is not obvious that the distances they define are finite: a priori it may be possible that the length of a path that collapses an edge is infinite. This will be addressed in §5. Remark 3.13. For the remainder of the paper we will mainly be concerned with the entropy metric. As stated in §1.6, the main results of this paper also hold for the pressure metric with slightly altered hypotheses. https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 748 T. Aougab et al 4. A determinant-defining equation for M1(G) The purpose of this section is to derive formulas to assist in computing the metrics introduced in the previous section. The first of these appears in §4.1, specifically Proposition 4.6, where it is shown that these metrics can be computed using finite sums of exponential functions on M1(G). Next, in §4.2, we present a simplification for certain graphs that is useful in the sequel. 4.1. Determinant equation. Fix a finite connected graph G = (V , E, o, τ , ¯). We always assume that χ(G) < 0 and G has no vertices of valence 1 or 2. Let DG be the directed graph with adjacency matrix AG. Thus the vertex set for DG is the set E (recall our notation E = E+ ∪ E+) and there is an edge from e to e(cid:17) if AG(e, e(cid:17)) = 1, that is, if τ (e) = o(e(cid:17)) and ¯e (cid:16)= e(cid:17). The cycle complex of DG, denoted by CG, is the abstract simplicial complex with an n-simplex for each collection (cid:7) = {z1, . . . , zn+1} of pairwise disjoint simple cycles, that is, embedded loops, in DG. Example 4.1. Suppose that G is the barbell graph as shown below: a c b Order the edges of G by a, ¯a, b, ¯b, c, ¯c. The matrix AG and directed graph DG are as presented below: ⎡ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ ⎤ 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 0 1 0 0 0 1 0 1 0 0 1 1 0 0 1 1 0 0 0 0 ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦ a ¯a c ¯c b ¯b There are eight simple cycles in DG: γa = (a), γ ¯a = ( ¯a), γb = (b), γ ¯b = (a, c, ¯b, ¯c), γ ¯ab = ( ¯a, c, b, ¯c) and γ ¯a ¯b (a, c, b, ¯c), γa ¯b CG is the flag complex whose 1-skeleton is shown in Figure 4. = ( ¯b), γab = = ( ¯a, c, ¯b, ¯c). The cycle complex Given a function f : E → R (in particular, a length function) and a simple cycle z in (cid:7) n DG, we set f (z) = i=1 f (ei) where e1, . . . , en are the vertices in DG traversed by z (each corresponding to an oriented edge in G). Likewise, for a simplex (cid:7) = {z1, . . . , zn} (cid:7) n in CG we set f ((cid:7)) = k=1 f (zk). We consider the empty set as a simplex and define f (∅) = 0 for any function f : E → R. Lastly, for a simplex (cid:7) = {z1, . . . , zn} we set |(cid:7)| = n. Recall is defined by AG,(cid:4)(e, e(cid:17)) = exp(−(cid:4)(e))AG(e, e(cid:17)). We consider the function FG : M(G) → R that given a length function (cid:4) ∈ M(G), the matrix AG,(cid:4) https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 749 γ¯a¯b γa¯b γb γa γ¯a γ¯b γ¯ab γab FIGURE 4. The 1-skeleton of the cycle complex CG in Example 4.1. given by FG((cid:4)) = det(I − AG,(cid:4)). (4.1) This function can be expressed using the cycle complex CG as follows. THEOREM 4.2. Let G be a finite connected graph and fix (cid:4) ∈ M(G). Then (cid:4) FG((cid:4)) = (−1) |(cid:7)| exp(−(cid:4)((cid:7))). (cid:7)∈CG Proof. This follows from the coefficient theorem for digraphs. See, for instance, [15] and [3, Theorems 2.5 and 2.14]. Example 4.3. We apply Theorem 4.2 to the case when G is the barbell graph as in Example 4.1. Using the change of variables x = exp(−(cid:4)(a)), y = exp(−(cid:4)(b)) and z = exp(−(cid:4)(c)), we find that FG((cid:4)) = 1 − (2x + 2y + 4xyz2) + (x2 + y2 + 4xy + 4x2yz2 + 4xy2z2) − (2x2y + 2xy2 + 4x2y2z2) + x2y2. The following statements show how the function FG is related to M1(G). LEMMA 4.4. For (cid:4) ∈ M1(G) we have: (1) FG((cid:4)) = 0; (2) ∇FG((cid:4)) (cid:16)= 0; and moreover, ∂eFG((cid:4)) > 0 for any e ∈ E+. (3) Proof. Since (cid:4) lies in M1(G), Theorem 3.7(1) and the definition of pressure imply that spec(AG,(cid:4)) = 1. Above we remarked that the assumptions on G imply that AG is irreducible; hence so is AG,(cid:4). By the Perron–Frobenius theorem the spectral radius of AG,(cid:4) is realized by a positive, real, simple eigenvalue; (1) follows. https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 750 T. Aougab et al Now consider the function p : R → R defined by p(t) = det(I − tAG,(cid:4)). Let 1 = λ1, . . . , λ|E| be the roots of the characteristic polynomial of AG,(cid:4). Then we can write p(t) = (1 − t) |E|(cid:20) i=2 (1 − tλi). Therefore, taking the derivative, we find that (cid:17) p (1) = − |E|(cid:20) (1 − λi). i=2 For i = 2, . . . , |E| we have that |λi| ≤ 1 and λi (cid:16)= 1. Combining these observations with the fact that complex eigenvalues come in conjugate pairs, it follows that p(cid:17)(1) < 0. Observe that FG((cid:4) + s · 1) = det(I − exp(−s)AG,(cid:4)) = p(exp(−s)). Therefore we have that (cid:18)1, ∇FG((cid:4))(cid:19) = −p(cid:17)(1) exp(0) > 0, giving (2). We claim that ∇FG((cid:4)) is parallel to ∇P G((cid:4)) for (cid:4) ∈ M1(G). Indeed, this follows as (1), −1 (2) and Theorem 3.7 imply that P−1 G (0). By (2) and Lemma 3.9(2) the gradients of FG and P G are non-zero for this subset. As gradients are always orthogonal to level sets, the claim follows. Hence by Lemma 3.9(1), we have that either ∂eFG((cid:4)) > 0 or ∂eFG((cid:4)) < 0 for all e ∈ E+ and (cid:4) ∈ M1(G). Since (cid:18)1, ∇FG((cid:4))(cid:19) > 0, we must have the former, whence (3). G (0) = M1(G) is a connected component of F As a remarked in the proof of Lemma 4.4, we have the following corollary. COROLLARY 4.5. The unit-entropy moduli space M1(G) is a connected component of the level set {(cid:4) ∈ M(G) | FG((cid:4)) = 0}. Using these observations, we can compute the entropy and pressure norms using the function FG. PROPOSITION 4.6. If (cid:4)t : (−1, 1) → M1(G) is a smooth path, then = h,G (cid:21)((cid:4)t , ˙(cid:4)t )(cid:21)2 −(cid:18) ˙(cid:4)t , H[FG((cid:4)t )] ˙(cid:4)t (cid:19) (cid:18)(cid:4)t , ∇FG((cid:4)t )(cid:19) −(cid:18) ˙(cid:4)t , H[FG((cid:4)t )] ˙(cid:4)t (cid:19) (cid:21)∇FG((cid:4)t )(cid:21)1 Proof. The proof of the formula for the entropy norm is similar to that of Proposition 3.12 and left to the reader. (cid:18) ¨(cid:4)t , ∇FG((cid:4)t )(cid:19) (cid:18)(cid:4)t , ∇FG((cid:4)t )(cid:19) , (cid:18) ¨(cid:4)t , ∇FG((cid:4)t )(cid:19) (cid:21)∇FG((cid:4)t )(cid:21)1 (cid:21)((cid:4)t , ˙(cid:4)t )(cid:21)2 P,G = = = . The proof of the formula for the pressure norm is again similar, noting that (cid:21)∇P((cid:4))(cid:21)1 = 1 as stated in Lemma 3.9(2). Using Theorem 4.2, we can compute the partial derivatives of FG. We find for any edges e, e(cid:17) ∈ E+ that (cid:4) ∂eFG((cid:4)) = − (−1) |(cid:7)| (cid:7)(e) exp(−(cid:4)((cid:7))), ∂ee(cid:17)FG((cid:4)) = (cid:7)∈CG (cid:4) (−1) |(cid:7)| (cid:7)(e)(cid:7)(e (cid:17) ) exp(−(cid:4)((cid:7))), (4.2) (4.3) (cid:7)∈CG https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 751 (cid:7) where (cid:7)(e) ∈ {0, 1, 2} denotes the cardinality of the intersection {e, ¯e} ∩ (cid:7). Using this e∈E+ (cid:7)(e)(cid:4)(e) for a length function (cid:4) ∈ M(G) and notation, we remark that (cid:4)((cid:7)) = simplex (cid:7) ∈ CG. Given a vector v ∈ R|E+| and a simplex (cid:7) ∈ CG, we set v((cid:7)) = (cid:7) e∈E+ (cid:7)(e)v(e). Using these expressions, we can rewrite the dot products appearing in the formulas for the metrics in Proposition 4.6 as sums over simplices in CG rather than over the edges of G as follows. LEMMA 4.7. For (cid:4) ∈ M(G), we have (cid:18)(cid:4), ∇FG((cid:4))(cid:19) = − (cid:4) (cid:7)∈CG (−1) |(cid:7)| (cid:4)((cid:7)) exp(−(cid:4)((cid:7))). Proof. We compute: (cid:18)(cid:4), ∇FG((cid:4))(cid:19) = (cid:9) (cid:4)(e) − (cid:4) e∈E+ (cid:4) (cid:7)∈CG (cid:10) (cid:10) (−1) |(cid:7)| (cid:7)(e) exp(−(cid:4)((cid:7))) (cid:9) (cid:4) (−1) |(cid:7)| exp(−(cid:4)((cid:7))) (cid:4)(e)(cid:7)(e) e∈E+ (cid:4)((cid:7)) exp(−(cid:4)((cid:7))). (−1) |(cid:7)| = − = − (cid:4) (cid:7)∈CG (cid:4) (cid:7)∈CG LEMMA 4.8. Let G be a finite connected graph. If ((cid:4), v) ∈ T M1(G), then (cid:18)v, H[FG((cid:4))]v(cid:19) = (−1) |(cid:7)|v((cid:7))2 exp(−(cid:4)((cid:7))). (cid:4) (cid:7)∈CG Proof. This is similar to Lemma 4.7. The eth component of H[FG((cid:4))]v is (cid:4) (cid:9) (cid:4) (cid:4) (cid:10) (−1) |(cid:7)| (cid:7)(e)(cid:7)(e ) exp(−(cid:4)((cid:7))) (cid:17) (cid:17) v(e ) = (−1) |(cid:7)| (cid:7)(e)v((cid:7)) exp(−(cid:4)((cid:7))). e(cid:17)∈E+ (cid:7)∈CG Hence (cid:18)v, H[∇FG((cid:4))]v(cid:19) = (cid:7)∈CG (cid:4) (cid:9) (cid:4) v(e) (−1) |(cid:7)| (cid:7)(e)v((cid:7)) exp(−(cid:4)((cid:7))) (cid:10) e∈E+ (cid:4) (cid:7)∈CG |(cid:7)|v((cid:7))2 exp(−(cid:4)((cid:7))). (−1) = (cid:7)∈CG By Lemma 4.4, if h G((cid:4)) = 1 then FG((cid:4)) = 0. The next lemma gives a partial converse. LEMMA 4.9. If h G((cid:4)) < 1, then FG((cid:4)) > 0. Proof. We begin by showing that if h if h G((cid:4)) < 1, then P G((cid:4)) < 1, then FG((cid:4)) (cid:16)= 0. To begin, we observe that G(−(cid:4)) < 0. Indeed, let (cid:4)t : [0, 1] → M(G) be the path defined by (cid:4)t = (1 − t)(cid:4) + th G(−(cid:4)t ), − ˙(cid:4)t (cid:19) > 0 as each component of ∇P G((cid:4))(cid:4). We have that (cid:18)∇P Lemma 3.9(1) and each component of − ˙(cid:4)t G(−(cid:4)t ) is positive by is positive by construction. Notice that https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 752 T. Aougab et al P G(−(cid:4)1) = 0 by Theorem 3.7(1) since h G((cid:4)1) = 1. Therefore, we find that −P G(−(cid:4)0) = (cid:13) 1 0 (cid:18)∇P G(−(cid:4)t ), − ˙(cid:4)t (cid:19) dt > 0 G(−(cid:4)) = P We now complete the proof of the lemma. Suppose that h G(−(cid:4)0) is negative as claimed. Therefore we have that and hence P spec(AG,(cid:4)) < 1 and, in particular, 1 is not an eigenvalue of AG,(cid:4). Hence FG((cid:4)) (cid:16)= 0. This completes the claim that FG((cid:4)) (cid:16)= 0 for any (cid:4) ∈ M(G) where h G((cid:4)) < 1. G((cid:4)) < 1 and consider the con- G((cid:4) + t · 1) < 1 t ∈ [0, ∞), by the above claim we have that p(t) (cid:16)= 0. Since p(t) → 1 t ∈ [0, ∞). In particular, we have that tinuous function p : [0, ∞) → R defined by p(t) = FG((cid:4) + t · 1). As h for all as t → ∞, we have that p(t) > 0 for all FG((cid:4)) = p(0) > 0. 4.2. A simplification. The function FG factors non-trivially in special cases as a result of certain aspects of the graph G. In such a case, we can replace FG with one of these factors and simplify the expressions for the entropy and pressure norm from Proposition 4.6. For instance, one factorization occurs if e is a loop edge. When (cid:4)(e) = 0 the vector v ∈ R|E|, where v(e) = 1, v( ¯e) = −1, and the rest of the entries are equal to 0, is an eigenvector of AG,(cid:4) with eigenvalue 1. This means 1 − exp(−(cid:4)(e)) is a factor of FG. Example 4.10. Using the notation from Example 4.3, we have that both 1 − x and 1 − y are factors of FG. Factoring, we have FG((cid:4)) = (1 − x)(1 − y)(1 − x − y + xy − 4xyz2). Another case where there is a factorization of FG is when the edge involution e ↔ ¯e is a graph automorphism of G. There are only two types of graphs for which such an automorphism exists: the r-rose, Rr ; (1) the graph (cid:5)r with vertices, v and w, and edges e1, . . . , er+1 where o(ei) = v and (2) τ (ei) = w for i = 1, . . . , r + 1. In this case ordering the edges in E+ first and then ordering the edges in E − E+ accordingly, we have that AG = (cid:11) (cid:8) BG B(cid:17) G B(cid:17) G BG for two matrices BG, B(cid:17) G ∈ Mat|E+|(R). Thus (cid:21) FG((cid:4)) = det(I − AG,(cid:4)) = det (cid:22) I − BG,(cid:4) −B(cid:17) −B(cid:17) G,(cid:4) I − BG,(cid:4) = det(I − BG,(cid:4) − B G,(cid:4) (cid:17) G,(cid:4)) det(I − BG,(cid:4) + B G corresponds to an edge e ∈ E+ the notation BG,(cid:4) and B(cid:17) (cid:17) G,(cid:4)). G,(cid:4) Since each row of BG or B(cid:17) still makes sense. https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 753 For Rr , we have that BRr is the r × r matrix consisting of all 1s and B(cid:17) Rr this case = BRr − I . In det(I − BRr ,(cid:4) + B (cid:17) Rr ,(cid:4)) = (cid:20) e∈E+ (1 − exp(−(cid:4)(e))). These are precisely the factors which were observed above for loop edges. For (cid:5)r , we have that B(cid:5)r is the (r + 1) × (r + 1) matrix consisting of all 0s and B(cid:17) is (cid:5)r the (r + 1) × (r + 1) matrix where all diagonal entries are 0 and all non-diagonal entries are 1. In this case we have F(cid:5)r ((cid:4)) = det(I − B (cid:17) (cid:5)r ,(cid:4)) det(I + B (cid:17) (cid:5)r ,(cid:4)). In general, we now construct a graph quotient DG → DG that identifies certain edge pairs {e, ¯e} resulting in a new matrix AG, which selects the appropriate factor. In this new matrix, every row corresponds to either an edge e ∈ E or an edge pair {e, ¯e}, and so we can still make sense of AG,f for a function f : E+ → R. When G is Rr or (cid:5)r , we take D(cid:5)r to be the quotient of D(cid:5)r by the orientation-reversing automorphism e (cid:12)→ ¯e. In this case AG,(cid:4) = BG,(cid:4) + B(cid:17) G,(cid:4). Otherwise, for each pair {e, ¯e} that is a loop edge of G, we identify the vertices of DG corresponding to e and ¯e, now denoted e ¯e, keep the incoming edges and identify the outgoing edges that have the same terminal vertex. We call the resulting graph DG. Algebraically, we add together the columns corresponding to e and ¯e and delete one of the rows corresponding to e and ¯e. We define F G : M(G) → R by F G((cid:4)) = det(I − AG,(cid:4)). (4.4) The formula in Theorem 4.2, the formulas for the partial derivatives in (4.2) and (4.3), and the inner products in Lemmas 4.7 and 4.8 hold for F G using the complex CG, which is the cycle complex of the directed graph DG. Example 4.11. For G equal to the barbell graph as in Example 4.1 we have AG and DG as shown below (columns of the matrix are ordered as a ¯a, b ¯b, c, ¯c): ⎡ ⎢ ⎢ ⎣ ⎤ ⎥ ⎥ ⎦ 1 0 1 0 0 1 0 1 0 2 0 0 2 0 0 0 a ¯a c ¯c b ¯b The two directed edges from ¯c to a ¯a are identified with the set {a, ¯a} so that we think of the sequence c, a or c, ¯a as specifying which of the two edges between ¯c and a ¯a to = (b ¯b), γab = (a, c, b, ¯c), traverse in a cycle. There are six simple cycles: γa ¯a = (a ¯a), γb ¯b = ( ¯a, c, ¯b, ¯c). The cycle complex CG is γa ¯b = (a, c, ¯b, ¯c), γ ¯ab = ( ¯a, c, b, ¯c) and γ ¯a ¯b https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 754 shown below: T. Aougab et al γa ¯a γb ¯b γab γ ¯ab γa ¯b γ ¯a ¯b Using Theorem 4.2, we find (with x = exp(−(cid:4)(a)), y = exp(−(cid:4)(b)) and z = exp(−(cid:4)(c))) that F G((cid:4)) = (cid:4) (cid:7)∈CG (−1) |(cid:7)| exp(−(cid:4)((cid:7))) = 1 − (x + y + 4xyz2) + xy. The astute reader will notice the comparison with Example 4.10. LEMMA 4.12. With the above setup, spec(AG,(cid:4)) = spec(AG,(cid:4)). Proof. Each circuit in DG lifts to at most two circuits of the same length in DG. Thus n G,(cid:4)) ≤ tr(An tr(A n G,(cid:4)) for all n ∈ N and so the lemma follows. G,(cid:4)) ≤ 2 tr(A In particular, we have that P G(−(cid:4)) = log spec(AG,(cid:4)). As in Corollary 4.5, we have the following statement. This follows for the same reasons as in Lemma 4.4 as F G((cid:4)) = 0 and ∇F G((cid:4)) (cid:16)= 0 for (cid:4) ∈ M1(G). PROPOSITION 4.13. The unit-entropy moduli space M1(G) is a connected component of the level set {(cid:4) ∈ M(G) | F G((cid:4)) = 0}. We also observe that the formulas for the metrics in Proposition 4.6 hold for F G. PROPOSITION 4.14. If (cid:4)t : (−1, 1) → M1(G) is a smooth path, then (cid:21)((cid:4)t , ˙(cid:4)t )(cid:21)2 h,G = (cid:21)((cid:4)t , ˙(cid:4)t )(cid:21)2 P,G = −(cid:18) ˙(cid:4)t , H[F G((cid:4)t )] ˙(cid:4)t (cid:19) (cid:18)(cid:4)t , ∇F G((cid:4)t )(cid:19) −(cid:18) ˙(cid:4)t , H[F G((cid:4)t )] ˙(cid:4)t (cid:19) (cid:21)∇F G((cid:4)t )(cid:21)1 = = (cid:18) ¨(cid:4)t , ∇F G((cid:4)t )(cid:19) (cid:18)(cid:4)t , ∇F G((cid:4)t )(cid:19) (cid:18) ¨(cid:4)t , ∇F G((cid:4)t )(cid:19) (cid:21)∇F G((cid:4)t )(cid:21)1 , . 5. The topology induced by the entropy metric The purpose of this section is to show that the metric topology induced by dh on X1(Fr ) is the same as the weak topology on X1(Fr ). We do so using the formulas for the entropy metric derived in §4 and seeing that they behave as one might anticipate with regards to collapses. We refer the reader back to §2.2 for the notation used in this section. By Theorem 3.5 we have that h M(G). Indeed, if c : G → G0 is a collapse, then for (cid:4) ∈ M(G0) we have h G : M(G) → R extends to a continuous function on G(c∗((cid:4))) = https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 755 h G0((cid:4)). We set M1 (G) = {(cid:4) ∈ M(G) | h G((cid:4)) = 1} and observe that we have (cid:5) M1 (G) = ∗ (M1(G0)) c c : G→G0 as well. This set is homeomorphic to the closure of CV (G, ρ) in CV (Fr ) for any marking ρ : Rr → G. Given a graph G, we observe that FG : M(G) → R admits an extension (still denoted FG) to R|E+|. In particular, FG is defined on M(G) ⊂ R|E+|. The next result shows that this function behaves as expected with respect to collapses. LEMMA 5.1. If c : G → G0 is a collapse, then FG ◦ c∗ = FG0. Proof. It suffices to consider the case when c : G → G0 is the collapse of a single edge e ∈ E+. Order the edges in E starting with e and ¯e. Since e can be collapsed, it is not a loop and so we have that AG(e, e) = AG( ¯e, ¯e) = 0. By definition, AG(e, ¯e) = AG( ¯e, e) = 0. Thus the top-leftmost 2 × 2 block of the matrix I − AG is the 2 × 2 identity matrix. Let (cid:4) ∈ M(G0). For an edge e(cid:17) /∈ {e, ¯e}, the entry [I − AG,c∗((cid:4))](e, e(cid:17)) is either −1 or 0 depending on whether or not e(cid:17) can follow e. Likewise for [I − AG,c∗((cid:4))]( ¯e, e(cid:17)). Again, as e is not a loop, for any edge e(cid:17) /∈ {e, ¯e}, at most one of these entries is non-zero. For each edge e(cid:17) /∈ {e, ¯e}, where [I − AG,c∗((cid:4))](e, e(cid:17)) = −1, we consider the column operation that adds the column for e to the column for e(cid:17). This zeros the (e, e(cid:17)) entry. The ( ¯e, e(cid:17)) entry was previously 0 and is unaffected by this operation. We next see what effect this has on the remaining rows. In the row for e(cid:17)(cid:17) /∈ {e, ¯e}, this adds − exp(−(cid:4)(e(cid:17)(cid:17))) to AG,c∗((cid:4))(e(cid:17)(cid:17), e(cid:17)) if e can follow e(cid:17)(cid:17) and 0 otherwise. In the former case, the previous entry was either 0 (e(cid:17) (cid:16)= e(cid:17)(cid:17)) or 1 (e(cid:17) = e(cid:17)(cid:17)) as e is not a loop edge. In other words, this adds − exp(−(cid:4)(e(cid:17)(cid:17))) whenever e(cid:17) can follow e(cid:17)(cid:17) in G0. Therefore, the remaining entries in the column for e(cid:17) agree with the corresponding entries in the column of I − AG0,(cid:4) for e(cid:17). Hence, after performing column operations to I − AG,c∗((cid:4)) with the column for e to clear out the rest of the row for e and column operations with the column for ¯e to clear out the rest of the row for ¯e, the resulting matrix has lower block triangular form. The top-leftmost 2 × 2 block is still the 2 × 2 identity matrix and the bottom-rightmost (|E| − 2) × (|E| − 2) block is I − AG0,(cid:4). As these column operations do not change the determinant, we have for (cid:4) ∈ M(G0) that ∗ FG(c ((cid:4))) = det(I − AG,c∗((cid:4))) = det(I − AG0,(cid:4)) = FG0((cid:4)). As a consequence of Lemma 5.1 we deduce the following result. If c : G → G0 is a collapse and e ∈ E+ is an edge such that c(e) is not a vertex, then ∂c(e)FG0((cid:4)) = ∂eFG(c∗((cid:4))) for all (cid:4) ∈ M(G0). Similarly, in this same setting, if additionally c(e(cid:17)) is not a vertex for an edge e(cid:17) ∈ E+, then ∂c(e)c(e(cid:17))FG0((cid:4)) = ∂ee(cid:17)FG(c∗((cid:4))). The tangent bundle T M1 (G) is the subspace of ((cid:4), v) ∈ R|E+| × R|E+| such that (G) and (cid:18)v, ∇FG((cid:4))(cid:19) = 0. Given a collapse, we let c∗ : R|(E0)+| → R|E+| be the (G) be the (cid:4) ∈ M1 derivative of the map c∗ : M(G0) → M(G) and T c∗ : T M1(G0) → T M1 map given by T c∗((cid:4), v) = (c∗((cid:4)), c∗(v)). https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 756 T. Aougab et al Using this notation, we see that (cid:18)c∗ (v), H[FG(c ∗ (cid:18)c ((cid:4)), ∇FG(c ∗ ((cid:4)))]c∗ ∗ (v)(cid:19) = (cid:18)v, H[FG0 ((cid:4))]v(cid:19), ((cid:4)))(cid:19) = (cid:18)(cid:4), ∇FG0((cid:4))(cid:19). Hence, by Proposition 4.6, we get the following proposition. PROPOSITION 5.2. Let G be a finite connected graph. The entropy norm (cid:21)(cid:2)(cid:21)h,G : T M1(G) → R extends to a continuous semi-norm (cid:21)(cid:2)(cid:21)h,G : T M1 (G) → R. Specifically, if c : G → G0 is a collapse and ((cid:4), v) = T c∗((cid:4)0, v0), then the extension satisfies (cid:21)((cid:4), v)(cid:21)h,G = (cid:21)((cid:4)0, v0)(cid:21)h,G0. Consequently, we see that the distance function dh,G extends to a distance function on (G) and that the induced topology is the same as the subspace topology. As X1(Fr ) is M1 locally finite, the metric topology agrees with the weak topology, as we now show. THEOREM 5.3. The metric topology on (X1(Fr ), dh) is the same as the weak topology on X1(Fr ). Proof. Let U ⊆ X1(Fr ) be an open set in the weak topology and fix a marked metric graph x = [(G, ρ, (cid:4))] ∈ U . There are finitely many marked graphs (G1, ρ1), . . . , (Gn, ρn) such that (G, ρ) ≤ (Gi, ρi). By definition of the weak topology, the set U ∩ X1 (Gi, ρi) is open in X1 , where Ei is the set of edges for Gi. As remarked above after Proposition 5.2, this set is also open in the metric topology induced by dh,Gi . Hence there is an (cid:15)i > 0 such that (Gi, ρi) in the subspace topology inherited from R|(Ei )+| ≥0 {y ∈ X1 (Gi, ρi) | dh,Gi (x, y) < (cid:15)i} ⊆ U ∩ X1 (Gi, ρi). Let (cid:15) = min{(cid:15)i | 1 ≤ i ≤ n}. As dh(x, y) ≤ dh,Gi (x, y) for any y ∈ X1 (Gi, ρi) we have {y ∈ X1 (Fr ) | dh(x, y) < (cid:15)} ⊆ n(cid:5) i=1 U ∩ X1 (Gi, ρi) ⊆ U . Hence the metric topology is finer than the subspace topology. Next, fix a marked metric graph x = [(G, ρ, (cid:4))] ∈ X1(Fr ) and let (cid:15) > 0. Enumerate the finitely many marked graphs (G1, ρ1), . . . , (Gn, ρn) such that (G, ρ) ≤ (Gi, ρi) and such that Gi is trivalent. In other words, (Gi, ρi) are maximal elements in the partial order on marked graphs. As the norm varies continuously by Proposition 5.2, there is an L and an open neighborhood V ⊆ (Gi, ρi) of x in the weak topology such that ≤ L whenever y ∈ X1 (cid:21)(y, v)(cid:21)h,Gi (Gi) ∩ V and (cid:18)v, v(cid:19) = 1. Therefore, there is an open neighborhood U of x in the weak topology such that n i=1 X1 (cid:23) U ⊆ {y ∈ X1 (Fr ) | dh(x, y) < (cid:15)}. Hence the subspace topology is finer than the metric topology. https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 757 6. The entropy metric on X1(F2) The goal of this section is to show that (X1(F2), dh) is complete. This appears as Proposition 6.8 in §6.4. The results in this section are not necessary for the remainder of the paper and can safely be skipped by a reader primarily interested in Theorems 1.2 and 1.3. However, the calculations can serve as a good introduction to estimating lengths with the entropy metric. In particular, the main strategy in each of Lemmas 6.1, 6.3 and 6.5 is very similar to the main strategy of Lemma 7.10 in §7.3 which is the key tool used to show that (M1(Rr ), dh,Rr ) has infinite diameter. To begin, we analyze the metric for each of the three topological types of graphs that appear in rank 2: the 2-rose R2, the barbell graph B2 and the theta graph (cid:5)2. We refer the reader back to Figure 2 for these graphs. To this end, we define a continuous function m : X1(F2) → R which is a slight variation of the volume function in that it counts separating edges twice. In particular, it does not depend on the marking. The exact definition appears in §6.4. The strategy is to show that for any path (cid:4)t : [0, 1] → M1(G) for G ∈ {R2, B2, (cid:5)2}, if m((cid:4)0) and m((cid:4)1) are large enough, then the length of (cid:4)t is bounded below by (cid:24) ( 1√ 5 m((cid:4)1) − (cid:24) m((cid:4)0)). These calculations appear in the next three sections (Propositions 6.2, 6.4, and 6.6). Using these estimates, it is not too hard to see that if (xn)n∈N ⊂ X1(F2) is Cauchy, then there is an L such that m(xn) ≤ L for all n (Lemma 6.7). From here, using local finiteness of X1(F2) and a compactness argument, the completeness of (X1(F2), dh) follows. In the calculations, we make use of Lemmas 4.7 and 4.8. 6.1. The 2-rose. Denote the edges of R2 by e1 and e2. To make the calculations in this subsection easier to read, given a length function (cid:4) ∈ M(R2), we set a = (cid:4)(e1), b = (cid:4)(e2) and m = (cid:4)(e1) + (cid:4)(e2). Applying the definition of F G from §4.2 to R2, we find the formula F R2((cid:4)) = 1 − exp(−a) − exp(−b) − 3 exp(−m). (6.1) LEMMA 6.1. Suppose (cid:4)t : [0, 1] → M1(R2) is a smooth path such that for all t ∈ [0, 1] we have ˙mt > 0. If m0 ≥ 4, then Lh,R2((cid:4)t |[0, 1]) ≥ √ m1 − √ m0. Proof. Suppose that (cid:4)t : [0, 1] → M1(R2) is a path where ˙mt > 0 as in the statement of the lemma and assume that m0 ≥ 4. We reparametrize the path (cid:4)t so that mt = t. Let nt = min{at , bt }. As F R2((cid:4)t ) = 0, we have 1 − 2 exp(−nt ) ≤ 1 − exp(−at ) − exp(−bt ) = 3 exp(−mt ) = 3 exp(−t). In particular, 2 exp(−nt ) ≥ 1 − 3 exp(−t) ≥ 2 exp(−1) as t ≥ 4 and so nt < 1 for all t. Setting pt = max{at , bt }, we find that pt = t − nt ≥ t − 1. Therefore, as t − 1 ≥ 1 for t ≥ 4, the edge that realizes the minimum of {at , bt } does not depend on t and hence https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 758 T. Aougab et al we may assume that bt = nt and that at = pt ≥ t − 1. This gives us that exp(−at ) ≤ exp(−t + 1). Hence, again as F R2((cid:4)t ) = 0, we have 1 − exp(−bt ) = exp(−at ) + 3 exp(−mt ) ≤ (exp(1) + 3) exp(−t) ≤ 8 exp(−t). This enables us to give an upper bound on the denominator in the expression for the entropy norm. Specifically, using the fact that x exp(−x) ≤ 1 − exp(−x) for x ≥ 0, we have (cid:18)(cid:4)t , ∇F R2((cid:4)t )(cid:19) = at exp(−at ) + bt exp(−bt ) + 3mt exp(−mt ) ≤ t exp(−t + 1) + (1 − exp(−bt )) + 3t exp(−t) ≤ 8t exp(−t) + 8 exp(−t) ≤ 12t exp(−t). In the final inequality we used that fact that t ≥ 4. The expression for (cid:18)(cid:4)t , ∇F R2((cid:4)t )(cid:19) can be computed either directly from (6.1) or via Lemma 4.7. Next, we get an upper bound on the numerator in the expression for the entropy norm by just using mt . Specifically, −(cid:18) ˙(cid:4)t , H[F R2((cid:4)t )] ˙(cid:4)t (cid:19) = ( ˙at )2 exp(−at ) + ( ˙bt )2 exp(−bt ) + 3( ˙mt )2 exp(−mt ) ≥ 3 exp(−t). As above, the expression for −(cid:18) ˙(cid:4)t , H[F R2((cid:4)t )] ˙(cid:4)t (cid:19) can be computed either directly from (6.1) or via Lemma 4.8. Hence we find that the entropy norm along this path is bounded below by (cid:21)((cid:4)t , ˙(cid:4)t )(cid:21)2 h,R2 = −(cid:18) ˙(cid:4)t , H[F R2((cid:4)t )] ˙(cid:4)t (cid:19) (cid:18)(cid:4)t , ∇F R2((cid:4)t )(cid:19) ≥ 1 4t . Therefore the length of this path in the entropy metric is at least (cid:25) (cid:13) m1 m0 dt = √ m1 − √ m0. 1 4t Using this lemma, we can get a lower bound on the distance between length functions in M1(R2) in terms of the sum of the lengths of edges, so long as they are sufficiently large. PROPOSITION 6.2. Suppose (cid:4) and (cid:4)(cid:17) are length functions in M1(R2) where m = (cid:4)(e1) + (cid:4)(e2) and m(cid:17) = (cid:4)(cid:17)(e1) + (cid:4)(cid:17)(e2) are at least 4. Then dh,R2((cid:4), (cid:4) (cid:17) ) ≥ √ √ m(cid:17) − m. [0, 1] → M1(R2) be a piecewise smooth path such that (cid:4)0 = (cid:4) and Proof. Let (cid:4)t : (cid:4)1 = (cid:4)(cid:17). Let δ ∈ [0, 1] be the minimal value such that mt ≥ 4 for t ∈ [δ, 1]. In particular, mδ ≤ m. By only considering the smooth subpaths of (cid:4)t |[δ, 1] for which ˙mt > 0, by Lemma 6.1, we find that Lh,R2((cid:4)t |[δ, 1]) ≥ √ m(cid:17) − √ mδ ≥ √ m(cid:17) − √ m. https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 759 This also provides a lower bound on Lh,R2((cid:4)t |[0, 1]). Since the path was arbitrary this also is a lower bound on the distance between (cid:4) and (cid:4)(cid:17). 6.2. The barbell graph. Let B2 denote the graph with vertices v and w, and edges e1, e2 and e3 where o(e1) = τ (e1) = v, o(e2) = τ (e2) = w, and o(e3) = v and τ (e3) = w. To make the calculations in this section easier to read, given a length function (cid:4) ∈ M(B2), we set a = (cid:4)(e1), b = (cid:4)(e2) and m = (cid:4)(e1) + (cid:4)(e2) + 2(cid:4)(e3). Applying the definition of F G from §4.2 to B2, we find the formula F B2((cid:4)) = (1 − exp(−a))(1 − exp(−b)) − 4 exp(−m). LEMMA 6.3. Suppose (cid:4)t : [0, 1] → M1(B2) is a smooth path such that for all t ∈ [0, 1] we have ˙mt > 0. If m0 ≥ 4, then Lh,B2((cid:4)t |[0, 1]) ≥ 1√ 2 √ ( m1 − √ m0). Proof. Suppose that (cid:4)t : [0, 1] → M1(B2) is a path where ˙mt > 0 as in the statement of the lemma and assume that m0 ≥ 4. We reparametrize the path (cid:4)t so that mt = t. As F B2((cid:4)t ) = 0, we have (1 − exp(−at ))(1 − exp(−bt )) = 4 exp(−mt ) = 4 exp(−t). (6.2) This enables us to give an upper bound on the denominator in the expression for the entropy norm. Specifically, using the fact that x exp(−x) ≤ 1 − exp(−x) for x ≥ 0, we have (cid:18)(cid:4)t , ∇F B2((cid:4)t )(cid:19) = at exp(−at )(1 − exp(−bt )) + bt exp(−bt )(1 − exp(−at )) + 4mt exp(−mt ) ≤ 2(1 − exp(−at ))(1 − exp(−bt )) + 4t exp(−t) = 4t exp(−t) + 8 exp(−t) ≤ 8t exp(−t). The penultimate line follows from (6.2), and in the final inequality we use the fact that t ≥ 4. Next, we get a lower bound on the numerator in the expression for the entropy norm. We claim that −(cid:18) ˙(cid:4)t , H[F B2((cid:4)t )] ˙(cid:4)t (cid:19) ≥ ( ˙mt )2 exp(−mt ). We have that −(cid:18) ˙(cid:4)t , H[F B2((cid:4)t )] ˙(cid:4)t (cid:19) = ( ˙at )2 exp(−at )(1 − exp(−bt )) + ( ˙bt )2 exp(−bt )(1 − exp(−at )) − 2 ˙at ˙bt exp(−at − bt ) + 4( ˙mt )2 exp(−mt ). Therefore if ˙at assume that ˙at and ˙bt have the same sign. As (cid:18) ˙(cid:4)t , ∇F B2((cid:4)t )(cid:19) = 0, we have that ˙bt < 0 then each term is positive and so the claim holds. Therefore, we 4 ˙mt exp(−mt ) = − ˙at exp(−at )(1 − exp(−bt )) − ˙bt exp(−bt )(1 − exp(−at )). We write this equation as w = u + v. As F B2((cid:4)t ) = 0, we find that 2uv = 2 ˙at = 2 ˙at ˙bt exp(−at ) exp(−bt )(1 − exp(−at ))(1 − exp(−bt )) ˙bt exp(−at − bt )(4 exp(−mt )). https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 760 T. Aougab et al As 2xy ≤ 3 4 (x + y)2 for all x and y, we find that 2 ˙at ˙bt exp(−at − bt )(4 exp(−mt )) ≤ 3 4 (4 ˙mt exp(−mt ))2 = 3( ˙mt )2(4 exp(−2mt )). Therefore 2 ˙at furthermore that ˙bt exp(−at − bt ) ≤ 3( ˙mt )2 exp(−mt ). From this the claim now follows, and −(cid:18) ˙(cid:4)t , H[F B2((cid:4)t )] ˙(cid:4)t (cid:19) ≥ ( ˙mt )2 exp(−mt ) = exp(−t). Hence we find that the entropy norm along this path is bounded below by (cid:21)((cid:4)t , ˙(cid:4)t )(cid:21)2 h,B2 = −(cid:18) ˙(cid:4)t , H[F B2((cid:4)t )] ˙(cid:4)t (cid:19) (cid:18)(cid:4)t , ∇F B2 ((cid:4)t )(cid:19) ≥ 1 8t . Therefore the length of this path in the entropy metric is at least (cid:25) (cid:13) m1 m0 1 8t dt = 1√ 2 √ ( m1 − √ m0). As for the 2-rose, we obtain the following proposition. PROPOSITION 6.4. Suppose (cid:4) and (cid:4)(cid:17) are length functions in M1(B2) where m = (cid:4)(e1) + (cid:4)(e2) + 2(cid:4)(e3) and m(cid:17) = (cid:4)(cid:17)(e1) + (cid:4)(cid:17)(e2) + 2(cid:4)(cid:17)(e3) are at least 4. Then dh,B2((cid:4), (cid:4) (cid:17) ) ≥ 1√ 2 √ ( m(cid:17) − √ m). 6.3. The theta graph. Let (cid:5)2 denote the graph with vertices v and w, and edges e1, e2 and e3 where o(ei) = v and τ (ei) = w for i ∈ {1, 2, 3}. To make the calculations in this section easier to read, given a length function (cid:4) ∈ M((cid:5)2), we set a = (cid:4)(e1) + (cid:4)(e2), b = (cid:4)(e2) + (cid:4)(e3), c = (cid:4)(e3) + (cid:4)(e1) and m = (cid:4)(e1) + (cid:4)(e2) + (cid:4)(e3). Applying the definition of F G from §4.2 to (cid:5)2, we find the formula F (cid:5)2((cid:4)) = 1 − exp(−a) − exp(−b) − exp(−c) − 2 exp(−m). LEMMA 6.5. Suppose (cid:4)t : [0, 1] → M1((cid:5)2) is a smooth path such that for all t ∈ [0, 1] we have ˙mt > 0. If m0 ≥ 4, then Lh,(cid:5)2((cid:4)t |[0, 1]) ≥ 1√ 5 √ ( m1 − √ m0). Proof. Suppose that (cid:4)t : [0, 1] → M1((cid:5)2) is a path where ˙mt > 0 as in the statement of the lemma and assume that m0 ≥ 4. We reparametrize the path (cid:4)t so that mt = t. Let nt = min{at , bt , ct }. As F (cid:5)2((cid:4)t ) = 0, we have 1 − 3 exp(−nt ) ≤ 1 − exp(−at ) − exp(−bt ) − exp(−ct ) = 2 exp(−mt ) = 2 exp(−t). In particular, 3 exp(−nt ) ≥ 1 − 2 exp(−t) > 3 exp(−2) as t ≥ 4 and thus nt < 2 for all t. Setting pt = max{at , bt , ct } and qt = at + bt + ct − pt − nt so that {pt , qt , nt } = {at , bt , ct } for all t, we find that pt , qt ≥ t − 2 as pt + qt = 2t − nt ≥ 2t − 2 and pt , qt ≤ t. Therefore, as t − 2 ≥ 2 for t ≥ 4, the cycle that realizes the minimum of {at , bt ct } does not depend on t and therefore we may assume that ct = min{at , bt , ct } https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 761 and thus at , bt ≥ t − 2. Therefore exp(−at ), exp(−bt ) ≤ exp(−t + 2). Hence, again as F (cid:5)2((cid:4)t ) = 0, we have 1 − exp(−ct ) = exp(−at ) + exp(−bt ) + 2 exp(−mt ) ≤ (2 exp(2) + 2) exp(−t) ≤ 20 exp(−t). This enables us to give an upper bound on the denominator in the expression for the entropy norm. Specifically, using the fact that x exp(−x) ≤ 1 − exp(−x) for x ≥ 0, we have (cid:18)(cid:4)t , ∇F (cid:5)2 ((cid:4)t )(cid:19) = at exp(−at ) + bt exp(−bt ) + ct exp(−ct ) + 2mt exp(−mt ) ≤ at exp(−at ) + bt exp(−bt ) + 1 − exp(−ct ) + 2mt exp(−mt ) ≤ t exp(−t + 2) + t exp(−t + 2) + 20 exp(−t) + 2t exp(−t) ≤ 20t exp(−t) + 20 exp(−t) ≤ 40t exp(−t). In the final inequality we used the fact that t ≥ 4. Next, we get a lower bound on the numerator in the expression for the entropy norm by just using mt . Specifically, −(cid:18) ˙(cid:4)t , H[F (cid:5)2((cid:4)t )] ˙(cid:4)t (cid:19) = ( ˙at )2 exp(−at ) + ( ˙bt )2 exp(−bt ) + ( ˙ct )2 exp(−ct ) + 2( ˙mt )2 exp(−mt ) ≥ 2 exp(−t). Hence we find that the entropy norm along this path is bounded below by (cid:21)((cid:4)t , ˙(cid:4)t )(cid:21)2 h,(cid:5)2 = −(cid:18) ˙(cid:4)t , H[F (cid:5)2((cid:4)t )] ˙(cid:4)t (cid:19) (cid:18)(cid:4)t , ∇F (cid:5)2 ((cid:4)t )(cid:19) ≥ 1 20t . Therefore the length of this path in the entropy metric is at least (cid:25) (cid:13) m1 m0 1 20t dt = 1√ 5 √ ( m1 − √ m0). Again, as for the 2-rose, we obtain the following proposition. PROPOSITION 6.6. Suppose (cid:4) and (cid:4)(cid:17) are length functions in M1((cid:5)2) where m = (cid:4)(e1) + (cid:4)(e2) + (cid:4)(e3) and m(cid:17) = (cid:4)(cid:17)(e1) + (cid:4)(cid:17)(e2) + (cid:4)(cid:17)(e3) are at least 4. Then dh,(cid:5)2 ((cid:4), (cid:4) (cid:17) ) ≥ 1√ 5 √ ( m(cid:17) − √ m). 6.4. (X1(F2), dh) is complete. We can now prove the main result of this section that the entropy metric on X1(F2) is complete. Given a marked metric graph x = [(G, ρ, (cid:4))] in X1(F2), we let m(x) = ⎧ ⎪⎪⎨ (cid:4)(e1) + (cid:4)(e2) (cid:4)(e1) + (cid:4)(e2) + 2(cid:4)(e3) ⎪⎪⎩ (cid:4)(e1) + (cid:4)(e2) + (cid:4)(e3) if G = R2, if G = B2, if G = (cid:5)2 We remark that m : X1(F2) → R is an Out(F2)-invariant continuous function. https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 762 T. Aougab et al LEMMA 6.7. Let (xn)n∈N be a Cauchy sequence in (X1(F2), dh). Then there is an L such that m(xn) ≤ L for all n. Proof. Suppose that (xn)n∈N is a sequence in X1(F2) such that m(xn) → ∞. We will show that (xn)n∈N is not Cauchy by showing that lim sup dh(x1, xn) = ∞. m(xn) ≥ m. Consider a path αt : [0, 1] → X1(F2) such that α0 = x1 and α1 = xn. Let δ ∈ [0, 1] be the minimal value such that m(αt ) ≥ 4 for all t ∈ [δ, 1]. In particular, m(αδ) ≤ m. Let m = max{m(x1), 4}. Given N ≥ 0, we let n be such that 5N + √ √ √ Combining Propositions 6.2, 6.4 and 6.6, we find that (cid:24) m(αδ)) ≥ 1√ 5 Lh(αt |[δ, 1]) ≥ 1√ 5 m(xn) − (cid:24) ( (cid:24) ( m(xn) − √ m) ≥ N. This also provides a lower bound on Lh(αt |[0, 1]). Since the path was arbitrary, this also is a lower bound on dh(x1, xn). Therefore dh(x1, xn) ≥ N, showing that lim sup dh(x1, xn) = ∞ as claimed. Given L ≥ 0, we set X1 L(F2) = {x ∈ X1(F2) | m(x) ≤ L}, and additionally, given a L(F2). We remark that L(G, ρ) = X1(G, ρ) ∩ X1 marked graph ρ : R2 → G, we set X1 the closure of X1 L(G, ρ) is compact. As X1(F2) is locally finite and there are only finitely many topological types of graphs, the following statement holds. For all L, D ≥ 0, there is an N such that if x ∈ X1 L(F2) then there is a collection of marked graphs (G1, ρ1), . . . , (GN , ρN ) such that {x (cid:17) ∈ X1 L(F2) | dh(x, x (cid:17) ) ≤ D} ⊆ N(cid:5) k=1 X1 L(Gk, ρk). PROPOSITION 6.8. The metric space (X1(F2), dh) is complete. Proof. Let (xn)n∈N be a Cauchy sequence in (X1(F2), dh). By Lemma 6.7, there is an L L(F2). Let n0 be such that dh(xn, xm) ≤ 1 if n, m ≥ n0. By the above such that (xn) ⊂ X1 remark, there are finitely marked graphs (G1, ρ1), . . . , (GN , ρN ) such that {xn|n ≥ n0} ⊂ {x (cid:17) ∈ X1 L(F2) | dh(xn0, x (cid:17) ) ≤ 1} ⊆ N(cid:5) k=1 X1 L(Gk, ρk). As the closure of this set in X1(F2) is compact, the sequence (xn)n∈N converges. 7. The moduli space of the rose The purpose of this section is to examine the entropy metric on the moduli space of an r-rose M1(Rr ). We begin in §7.1 by computing the function F Rr introduced in §4.2. For the r-rose, we can strengthen Proposition 4.13 and conclude that M1(Rr ) equals the set of length functions (cid:4) for which F Rr ((cid:4)) = 0. Next, in §7.2 we show that (M1(Rr ), dh,Rr ) for r ≥ 3 is not complete by producing paths that have finite length yet no accumulation point. Lastly, in §7.3 we show that (M1(Rr ), dh,Rr ) has infinite diameter. Specifically, a path that shrinks the length of an edge to zero necessarily has infinite length. https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 763 7.1. M1(Rr ) as a zero locus. In this section we compute the function F Rr . This appears as Proposition 7.3. In Proposition 7.5 we prove that M1(Rr ) = {(cid:4) ∈ M(Rr ) | F Rr ((cid:4)) = 0}; this strengthens Proposition 4.13 in this setting. First we set some notation for working with the graph Rr . We identify the unoriented edges of Rr with the set [r] = {1, 2, . . . , r}. To simplify the expressions, we will use r as the identifying subscript rather than Rr and we will use variables (cid:4) = ((cid:4)1, . . . , (cid:4)r ) to denote the length of the unoriented edges. The matrix Ar,(cid:4) ∈ Matr (R) has rows and columns indexed by [r], and we have Ar,(cid:4)(i, j ) = exp(−(cid:4)i)(2 − δ(i, j )) (7.1) where δ((cid:2), (cid:2)) is the Kronecker delta function. For the calculations in this section, we need the following combinatorial identities. LEMMA 7.1. For any r ≥ 1 and any x ∈ R, the following equations hold: (1 + x)r−1(x − (2r − 1)) = xr (1 + x)r−1(x + (2r + 1)) = xr (cid:9) r(cid:4) k=0 r(cid:4) (cid:9) (cid:10) (cid:10) r k r k (1 − 2k)x −k, (1 + 2k)x −k. (7.2) (7.3) Proof. Differentiate the equation (1 + x)r = (cid:31) (cid:30) (cid:7) the equality rx(1 + x)r−1 = r r k=0(r − k) k xr−k and multiply it by x to obtain k=0 (cid:30) (cid:31) (cid:7) r k=0 r k xr−k. Therefore (cid:9) (cid:10) r(cid:4) k=0 − 2r r(cid:4) k=0 r k (cid:9) (cid:10) r k (cid:9) r(cid:4) k=0 xr−k + (cid:10) r k = (1 − 2k) xr−k = xr r(cid:4) k=0 (cid:9) 2k (cid:10) r k (cid:10) (cid:9) r k r(cid:4) k=0 xr−k xr−k (cid:9) (1 − 2k) (cid:10) r k −k. x r(cid:4) k=0 2rx(1 + x)r−1 − (2r − 1)(1 + x)r = 2r xr−k − The left-hand side in the above equation simplifies to (1 + x)r−1(x − (2r − 1)). This shows (7.2). In a similar manner one can derive (7.3). COROLLARY 7.2. For any r ≥ 1, (cid:9) r(cid:4) k=0 (cid:10) r k (1 − 2k)(2r − 1) −k = 0. (7.4) Proof. Evaluate equation (7.2) with x = 2r − 1. The left-hand side becomes 0. Dividing the resulting equation by (2r − 1)r , we obtain (7.4). Given a length function (cid:4) = ((cid:4)1, . . . , (cid:4)n) ∈ M(Rr ) and a subset S ⊆ [r], we define (cid:7) (cid:4)S = k∈S (cid:4)k. In particular, we have (cid:4)∅ = 0. https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 764 T. Aougab et al PROPOSITION 7.3. For any r ≥ 2 and any length function (cid:4) ∈ M(Rr ), F r ((cid:4)) = (1 − 2|S|) exp(−(cid:4)S). (7.5) (cid:4) S⊆[r] Proof. Using the expansion of the determinant via permutations of [r], we can express F r ((cid:4)) = det(I − Ar,(cid:4)) as (cid:4) F r ((cid:4)) = cS,r exp(−(cid:4)S) S⊆[r] for some coefficients cS,r ∈ R depending on the subset S ⊆ [r] and the rank r. Further, it is apparent that the coefficient cS,r only depends on the cardinality of S. It remains to determine these coefficients. We will do so by induction. For r = 2, we compute F 2((cid:4)1, (cid:4)2) = det (cid:8) 1 − exp(−(cid:4)1) −2 exp(−(cid:4)1) −2 exp(−(cid:4)2) 1 − exp(−(cid:4)2) (cid:11) = 1 − exp(−(cid:4)1) − exp(−(cid:4)2) − 3 exp(−(cid:4)1 − (cid:4)2). This shows the proposition for r = 2. Suppose r ≥ 3 and that the proposition holds for r − 1. That is, we assume that cS,r−1 = 1 − 2|S| for any S ⊆ [r − 1]. Since cS,r only depends on the cardinality of S, this implies that cS,r = 1 − 2|S| for any S ⊆ [r] where |S| < r as well. Hence it only remains to compute c[r],r . h To compute c[r],r , we make use of Corollary 7.2. Indeed, by Example 3.3, we have r (log(2r − 1) · 1) = 1. Therefore, by Proposition 4.13, we obtain F r (log(2r − 1) · 1) = 0. Hence 0 = F r (log(2r − 1) · 1) (cid:4) (1 − 2|S|) exp(−|S| log(2r − 1)) + c[r],r exp(−r log(2r − 1)) = = S⊂[r] (cid:9) r−1(cid:4) k=0 (cid:10) r k (1 − 2k)(2r − 1) −k + c[r],r (2r − 1) −r . By Corollary 7.2, we find that c[r],r = 1 − 2r as desired. Example 7.4. For r = 2 and r = 3, using the coordinates x = exp(−(cid:4)1), y = exp(−(cid:4)2) and z = exp(−(cid:4)3), we find F 2((cid:4)1, (cid:4)2) = 1 − x − y − 3xy, F 3((cid:4)1, (cid:4)2, (cid:4)3) = 1 − x − y − z − 3xy − 3xz − 3yz − 5xyz. Figure 5 shows M1(Rr ) as a subset of M(Rr ) for r = 2 and r = 3. Using Proposition 7.3, we can provide a strengthening of Proposition 4.13 for the r-rose. PROPOSITION 7.5. For any r ≥ 2, the unit-entropy moduli space M1(Rr ) equals the level set {(cid:4) ∈ M(Rr ) | F r ((cid:4)) = 0}. https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 765 8 6 b 4 2 0 0 2 4 a 6 8 FIGURE 5. The hypersurfaces M1(Rr ) for the roses with 2 and 3 petals. Proof. By Proposition 4.13, we have that M1(Rr ) ⊆ {(cid:4) ∈ M(Rr ) | F r ((cid:4)) = 0}. Suppose that F r ((cid:4)) = 0 for some (cid:4) ∈ M(Rr ). Set h = h r ((cid:4)). We need to show that h = 1. Consider the function p : R>0 → R defined by p(t) = F r (t · (cid:4)). We have p(1) = r (h · (cid:4)) = 1, we have p(h) = F r (h · (cid:4)) = 0 as well by Proposition 4.13. F r ((cid:4)) = 0. As h Using the expression for F r ((cid:4)) derived in Proposition 7.3, we compute that (cid:4) (cid:17) p (t) = (2|S| − 1)(cid:4)S exp(−t · (cid:4)S). S⊆[r] S(cid:16)=∅ Therefore p(cid:17)(t) > 0 for all t ∈ R>0. As p(h · (cid:4)) = 0 = p((cid:4)), we must have that h · (cid:4) = (cid:4) and hence h = 1. 7.2. Finite-length paths in M1(Rr ) for r ≥ 3. Using the computation of F r , in Proposition 7.8 we will compute the length of the path in M1(Rr ) starting at log(2r − 1) · 1 that blows up the length of one edge while shrinking the lengths of the others at the same rate. As we will show, when r is at least 3, this path has finite length and thus the moduli space (M1(Rr ), dh,Rr ) is not complete for r ≥ 3. Before we begin, it is useful to introduce the following functions Xi, Yi : M(Rr ) → R for each i ∈ [r]: Xi((cid:4)) = Yi((cid:4)) = (cid:4) (1 − 2|S|) exp(−(cid:4)S), S⊆[r]−{i} (cid:4) (1 + 2|S|) exp(−(cid:4)S). S⊆[r]−{i} (7.6) (7.7) Both Xi and Yi are constant with respect to (cid:4)i. Using these functions, we can isolate the terms in F r ((cid:4)) in which (cid:4)i appears and write F r ((cid:4)) = Xi((cid:4)) − exp(−(cid:4)i)Yi((cid:4)). (7.8) https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 766 T. Aougab et al Hence for (cid:4) ∈ M1(Rr ), as F r ((cid:4)) = 0 we can solve for (cid:4)i and write (cid:4)i = log (cid:9) (cid:10) Yi((cid:4)) Xi((cid:4)) . (7.9) Further, we find the following expression for the partial derivative of F r ((cid:4)) with respect to (cid:4)i: ∂iF r ((cid:4)) = exp(−(cid:4)i)Yi((cid:4)). We observe the following inequalities for any (cid:4) ∈ M1(Rr ). LEMMA 7.6. Let r ≥ 2 and let (cid:4) ∈ M1(Rr ). Then 0 < Xi((cid:4)) < 1, 1 < Yi((cid:4)) < 4. (7.10) (7.11) (7.12) Proof. For (7.11), we first note that Xi((cid:4)) = exp(−(cid:4)i)Yi((cid:4)) for all (cid:4) ∈ M1(Rr ) by Proposition 4.13 and (7.8). Since every term in Yi((cid:4)) has a positive coefficient, we find that 0 < Xi((cid:4)). As the term in Xi((cid:4)) corresponding to S = ∅ is 1 and all other terms have negative coefficients, we find Xi((cid:4)) < 1. For (7.12), we have that the term in Yi((cid:4)) corresponding to S = ∅ is 1 and all other terms have positive coefficients, thus 1 < Yi((cid:4)). The terms in 1 − Xi((cid:4)) and Yi((cid:4)) − 1 correspond to the non-empty subsets S ⊆ [r] − {i}. The coefficient for the term corresponding to S in 1 − Xi((cid:4)) is 2|S| + 1 2|S| − 1 times the coefficient for the same term in Yi((cid:4)) − 1. As this ratio is bounded by 3, we find that Yi((cid:4)) − 1 ≤ 3(1 − Xi((cid:4))). Hence, as 1 − Xi((cid:4)) < 1 by (7.11), we have Yi((cid:4)) − 1 < 3 and so Yi((cid:4)) < 4. We record the following calculation. LEMMA 7.7. Let r ≥ 2, and let (cid:4) ∈ M1(Rr ) be such that (cid:4)i = log(L) for i ∈ [r − 1] for some L > 2r − 3. Then (cid:4)r = log (cid:9) (cid:10) L + (2r − 1) L − (2r − 3) . (7.13) Proof. For any S ⊆ [r − 1] we have exp(−(cid:4)S) = exp(−|S| log L) = L−|S|. Hence, by Lemma 7.1, we have that (cid:9) (cid:10) r − 1 k Xr ((cid:4)) = Yr ((cid:4)) = r−1(cid:4) k=0 r−1(cid:4) (cid:9) k=0 (1 − 2k)L −k = L −r+1(L + 1)r−2(L − (2r − 3)), (cid:10) r − 1 k (1 + 2k)L −k = L −r+1(L + 1)r−2(L + (2r − 1)). https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 767 Therefore, by (7.9), we find that (cid:4)r = log (cid:9) Yr ((cid:4)) Xr ((cid:4)) (cid:10) (cid:9) = log L + (2r − 1) L − (2r − 3) (cid:10) . For any r ≥ 3, we will construct a path (cid:4)t : [0, 1) → M1(Rr ) that has finite length and the property that (cid:4)r t → ∞ as t → 1−. PROPOSITION 7.8. Fix r ≥ 3 and let Nt = 2(r − t) − 1. Let (cid:4)t : [0, 1) → M1(Rr ) be the → ∞ smooth path where (cid:4)i t as t → 1−. = log(Nt ) for i ∈ [r − 1]. Then Lh,r ((cid:4)t |[0, 1)) is finite and (cid:4)r t Proof. Let (cid:4)t : [0, 1) → M1(Rr ) be as in the statement. Using Lemma 7.7, we find that (cid:9) (cid:10) (cid:4)r t = log 2r − 1 − t 1 − t . In particular, we have (cid:4)r t → ∞ when t → 1− as claimed. We first provide a lower bound on (cid:18)(cid:4)t , ∇F r ((cid:4)t )(cid:19). This is the denominator of the expression for the entropy norm in Proposition 4.14. Using the expressions for the partial derivatives for F r ((cid:4)t ) in (7.10), the fact that 1 < Yi((cid:4)t ) from (7.12) and that t exp(−(cid:4)r (cid:4)r t )Yr ((cid:4)t ) > 0, we have that (cid:18)(cid:4)t , ∇F r ((cid:4)t )(cid:19) = r(cid:4) i=1 t exp(−(cid:4)i (cid:4)i t )Yi((cid:4)t ) > r−1(cid:4) i=1 t exp(−(cid:4)i (cid:4)i t ) = (r − 1) log(Nt ) Nt . As log(Nt ) ≥ log(N1) and Nt ≤ N0 for all t ∈ [0, 1], we conclude that (cid:18)(cid:4)t , ∇F r ((cid:4)t )(cid:19) > (r − 1) log(Nt ) Nt ≥ (r − 1) log(N1) N0 = (r − 1) log(2r − 3) 2r − 1 . (7.14) Next, we provide an upper bound on (cid:18) ¨(cid:4)t , ∇F r ((cid:4)t )(cid:19). This is the numerator of the expression for the entropy norm in Proposition 4.14. To do so, we compute that ¨(cid:4)i t = − 4 N 2 t , for i ∈ [r − 1], and ¨(cid:4)r t = − 1 (2r − 1 − t)2 + 1 (1 − t)2 . t exp(−(cid:4)i In particular, ¨(cid:4)i t < 1/(1 − t)2. Combining these with the expressions for the partial derivatives for F r ((cid:4)t ) in (7.10) and the fact that Yi((cid:4)t ) < 4 (7.12), we have that t )Yi((cid:4)t ) < 0 for i ∈ [r − 1] and ¨(cid:4)r (cid:18) ¨(cid:4)t , ∇F r ((cid:4)t )(cid:19) = r(cid:4) ¨(cid:4)i t exp(−(cid:4)i t )Yi((cid:4)t ) i=1 < ¨(cid:4)r t exp(−(cid:4)r 1 (1 − t)2 < · t )Yi((cid:4)t ) 1 − t 2r − 1 − t · 4 ≤ 2 r − 1 · 1 1 − t . (7.15) https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 768 T. Aougab et al Proposition 4.14, together with the bounds appearing in (7.14) and (7.15), implies that (cid:21)((cid:4)t , ˙(cid:4)t )(cid:21)2 h,Rr = (cid:18) ¨(cid:4)t , ∇F r ((cid:4)t )(cid:19) (cid:18)(cid:4)t , ∇F r ((cid:4)t )(cid:19) ≤ 2(2r − 1) (r − 1)2 log(2r − 3) · 1 1 − t . Therefore, the entropy length of the path (cid:4)t : [0, 1) → M1(Rr ) is finite as claimed. As a consequence, we obtain that M1(Rr ) is not complete when r ≥ 3. PROPOSITION 7.9. For any r ≥ 3, the moduli space (M1(Rr ), dh,Rr ) is not complete. In §8 we will use Proposition 7.8 to show that (X1(Fr ), dh) is not complete as well when r ≥ 3. 7.3. The diameter of M1(Rr ) is infinite. In this subsection we show that (M1(Rr ), dh,Rr ) has infinite diameter by showing that any path that shrinks an edge to 0 has infinite length. Before we begin, it is useful to introduce the following functions. For distinct i, j ∈ [r], we define (cid:4) Xij ((cid:4)) = Yij ((cid:4)) = (1 + 2|S|) exp(−(cid:4)S), S⊆[r]−{i,j } (cid:4) (3 + 2|S|) exp(−(cid:4)S). S⊆[r]−{i,j } (7.16) (7.17) As in Lemma 7.6, we observe that, for any (cid:4) ∈ M1(Rr ), we have 3 < Yij ((cid:4)) < 3Yi((cid:4)) < 12. Notice that both Xij and Yij are constant with respect to both (cid:4)i and (cid:4)j . For any distinct i, j ∈ [r], these functions allow us to write (7.18) Yi((cid:4)) = Xij ((cid:4)) + exp(−(cid:4)j )Yij ((cid:4)). (7.19) Using (7.19) plus the expressions for the partial derivatives for F r ((cid:4)) in (7.10), we find the following expressions for the second partial derivatives of F r ((cid:4)): ∂iiF r ((cid:4)) = − exp(−(cid:4)i)Yi((cid:4)), ∂ij F r ((cid:4)) = − exp(−(cid:4)i − (cid:4)j )Yij ((cid:4)) for i (cid:16)= j . (7.20) (7.21) The following technical lemma is the main tool for estimating length. Intuitively, it says that when one of the edge lengths—(cid:4)r in the statement—is short, the length of a path is bounded below by the difference in the square roots of the lengths of second shortest edge—(cid:4)1 in the statement—at the endpoints of the path. In the statement below, shortness of (cid:4)r is guaranteed by taking (cid:4)1 large enough. LEMMA 7.10. Let r ≥ 2. There is an Lr with the following property. Suppose (cid:4)t : [0, 1] → M1(Rr ) is a piecewise smooth path such that, for all t ∈ [0, 1]: (1) (2) (3) = min{(cid:4)i t ≥ Lr ; and | i ∈ [r − 1]}; (cid:4)1 t (cid:4)1 0 ˙(cid:4)1 t > 0. https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 769 Then Lh,Rr ((cid:4)t |[0, 1]) ≥ (cid:30) B1(cid:4)1 1 1 √ 2B1 2 + B2 − B1(cid:4)1 0 + B2 (cid:31) where B1 = 4(r − 1) and B2 = 2r+3(2r − 1). Proof. Let (cid:4)t : [0, 1] → M1(Rr ) be as in the statement. By (3) we may reparametrize the path so that (cid:4)1 t = t. Let Lr be large enough so that max{2r (2r − 3) exp(−Lr ), 288r exp(−Lr )} ≤ 1. The method of proof is similar to the calculations performed in §6. Specifically, using the expression (cid:21)((cid:4)t , ˙(cid:4)t )(cid:21)2 h,Rr = −(cid:18) ˙(cid:4)t , H[F r ((cid:4)t )] ˙(cid:4)t (cid:19) (cid:18)(cid:4)t , ∇F r ((cid:4)t )(cid:19) (7.22) from Proposition 4.14, we show that the square of the entropy norm along this path is bounded from below by 1/2(B1t + B2). This is done by showing that the denominator is bounded from above by exp(−t)(B1t + B2) in (7.27), and that the numerator is bounded from below by 1 2 exp(−t) in (7.33). We first provide the upper bound on (cid:18)(cid:4)t , ∇F r ((cid:4)t )(cid:19). As (cid:4)i t = t for all i ∈ [r − 1], we have that exp(−(cid:4)S t ) ≤ exp(−t) for all non-empty subsets S ⊆ [r − 1]. Since 1 − 2|S| ≥ −(2r − 3) for any non-empty subset S ⊆ [r − 1], using the definition of Xi((cid:4)t ) in (7.6) we have that ≥ (cid:4)1 t (1 − 2|S|) exp(−(cid:4)S t ) ≥ 1 − 2r−1(2r − 3) exp(−t). (7.23) (cid:4) S⊆[r−1] Xr ((cid:4)t ) = Therefore 1 − Xr ((cid:4)t ) ≤ 2r−1(2r − 3) exp(−t). As t ≥ Lr we additionally find that Xr ((cid:4)t ) ≥ 1 2 . (7.24) (7.25) As 0 < Xr ((cid:4)t ) < 1 by (7.11) and − log(1 − x) ≤ x/(1 − x) for all 0 < x < 1, using (7.24) and (7.25), we find that − log(Xr ((cid:4)t )) = − log(1 − (1 − Xr ((cid:4)t ))) ≤ 1 − Xr ((cid:4)t ) Xr ((cid:4)t ) ≤ 2r (2r − 3) exp(−t). Similarly, as 1 < Yr ((cid:4)t ) by (7.12) and log(x) ≤ x − 1 for all x ≥ 1, using the definition of Yi((cid:4)t ) from (7.7), we have that (1 + 2|S|) exp(−(cid:4)S t ) ≤ 2r−1(2r − 1) exp(−t). log(Yr ((cid:4)t )) ≤ Yr ((cid:4)t ) − 1 = Thus by (7.9), we find (cid:4) S⊆[r−1] S(cid:16)=∅ (cid:4)r t = log(Yr ((cid:4)t )) − log(Xr ((cid:4)t )) ≤ 2r+1(2r − 1) exp(−t). (7.26) https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 770 T. Aougab et al As x exp(−x) is decreasing for x > 1, we have that (cid:4)i t ) = t exp(−t) for i ∈ [r − 1]. Using the expressions for the partial derivatives of F r ((cid:4)t ) in (7.10) and the fact that Yi((cid:4)t ) < 4 from (7.12), we have that t exp(−(cid:4)1 t exp(−(cid:4)i t ) ≤ (cid:4)1 (cid:18)(cid:4)t , ∇F r ((cid:4)t )(cid:19) = r(cid:4) i=1 t exp(−(cid:4)i (cid:4)i t )Yi((cid:4)t ) < 4((r − 1)t exp(−t) + (cid:4)r t )) ≤ 4 exp(−t)((r − 1)t + 2r+1(2r − 1)). t exp(−(cid:4)r As defined above, we have that B1 = 4(r − 1) and B2 = 2r+3(2r − 1). Hence (cid:18)(cid:4)t , ∇F r ((cid:4)t )(cid:19) ≤ exp(−t)(B1t + B2). (7.27) Next we provide a lower bound on −(cid:18) ˙(cid:4)t , H[F r ((cid:4)t )] ˙(cid:4)t (cid:19). Using the expressions for the second partial derivatives of F r ((cid:4)t ) in (7.20) and (7.21), we have that −(cid:18) ˙(cid:4)t , H[F r ((cid:4)t )] ˙(cid:4)t (cid:19) = r(cid:4) ( ˙(cid:4)i t )2 exp(−(cid:4)i t )Yi((cid:4)t ) + i=1 r−1(cid:4) r(cid:4) i=1 j =i+1 2 ˙(cid:4)i t ˙(cid:4)j t exp(−(cid:4)i t − (cid:4)j t )Yij ((cid:4)t ). (7.28) The following claim says that the diagonal terms in H[F r ((cid:4)t )] dominate in the current setting, that is, when one of the edge lengths is small. CLAIM 7.11. 1 2 (cid:18)( ˙(cid:4)t )2, ∇F r ((cid:4)t )(cid:19) ≤ −(cid:18) ˙(cid:4)t , H[F r ((cid:4)t )] ˙(cid:4)t (cid:19) ≤ 3 2 (cid:18)( ˙(cid:4)t )2, ∇F r ((cid:4)t )(cid:19). Proof of Claim 7.11. We observe that is exactly (cid:18)( ˙(cid:4)t )2, ∇F r ((cid:4))(cid:19). The claim is thus proved by showing that the second summand has (cid:18)( ˙(cid:4)t )2, ∇F r ((cid:4)t )(cid:19). We accomplish this by breaking this absolute value bounded above by 1 2 summand into various pieces. the first summand in (7.28) To begin, we focus on the terms in this summand where j = r. Let Kr ⊆ [r − 1] be the set of indices where |2 ˙(cid:4)i t exp(−(cid:4)i t )Yir ((cid:4)t )| ≤ (1/2r)| ˙(cid:4)r t Yr ((cid:4)t )|. Summing over the elements in Kr , we find that (cid:4) ! ! ! ! i∈Kr 2 ˙(cid:4)i t ˙(cid:4)r t exp(−(cid:4)i t − (cid:4)r t )Yir ((cid:4)t ) ! ! ! ≤ 1 ! 2 ( ˙(cid:4)r t )2 exp(−(cid:4)r t )Yr ((cid:4)t ). (7.29) From the definition of Lr we have 24r exp(−Lr ) ≤ 1/12. Thus if i < r and i /∈ Kr as (cid:4)i t 2| ˙(cid:4)r ≥ Lr and Yir ((cid:4)t ) < 3 max{Yi((cid:4)t ), Yr ((cid:4)t )} from (7.18), we have that t Yir ((cid:4)t )| ≤ 6| ˙(cid:4)r t )Yir ((cid:4)t )| ≤ 1/12| ˙(cid:4)i t Yir ((cid:4)t )| ≤ 1/4| ˙(cid:4)i t Yr ((cid:4)t )| ≤ |24r ˙(cid:4)i t exp(−(cid:4)i t Yi((cid:4)t )|. (7.30) Thus for i < r and i /∈ Kr we have that t )Yir ((cid:4)t )| ≤ |2 ˙(cid:4)i ˙(cid:4)r t exp(−(cid:4)i t |2 ˙(cid:4)i t − (cid:4)r t ˙(cid:4)r t exp(−(cid:4)i t )Yir ((cid:4)t )| ≤ 1 4 ( ˙(cid:4)i t )2 exp(−(cid:4)i t )Yi((cid:4)t ). (7.31) Next we turn our attention to the terms where j < r. https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 771 | = | ˙(cid:4)j t For i ∈ [r − 1] we let Ki ⊆ [r − 1] be the set of indices where | ˙(cid:4)i | or where | ˙(cid:4)i | and j > i. We observe that for any distinct pair of indices i, j ∈ t [r − 1] either j ∈ Ki and i /∈ Kj or i ∈ Kj and j /∈ Ki. From the definition of Lr we have 2 exp(−Lr ) ≤ 1/12r. Hence as Yij ((cid:4)t ) < 3Yi((cid:4)t ) from (7.18), we find that 2 exp(−(cid:4)j t )Yij ((cid:4)t ) ≤ (1/4r)Yi((cid:4)t ) for j ∈ [r − 1]. Therefore, summing over the indices in Ki we find that (cid:4) | > | ˙(cid:4)j t t 2 ˙(cid:4)i t ˙(cid:4)j t exp(−(cid:4)i t − (cid:4)j ( ˙(cid:4)i t )2 exp(−(cid:4)i t )Yi((cid:4)t ). (7.32) ! ! ! ≤ 1 ! t )Yij ((cid:4)t ) 4 ! ! ! ! j ∈Ki Rearranging the terms and using (7.29), (7.31) and (7.32), we find that ! r−1(cid:4) ! ! ! ! ! ! ! ≤ t )Yij ((cid:4)t ) ˙(cid:4)j t exp(−(cid:4)i t ˙(cid:4)j t exp(−(cid:4)i t ! r−1(cid:4) ! ! ! − (cid:4)j 2 ˙(cid:4)i t 2 ˙(cid:4)i t r(cid:4) (cid:4) − (cid:4)j i=1 j =i+1 ≤ ! ! ! t )Yij ((cid:4)t ) ! ! ! ! ! 2 ˙(cid:4)i t ˙(cid:4)r t exp(−(cid:4)i t − (cid:4)r t )Yir ((cid:4)t ) 2 ˙(cid:4)i t ˙(cid:4)r t exp(−(cid:4)i t − (cid:4)r t )Yir ((cid:4)t ) ! ! ! ! j ∈Ki (cid:4) i=1 ! ! ! ! i∈Kr ! (cid:4) ! ! ! i /∈Kr + + r−1(cid:4) i=1 + 1 1 4 2 ( ˙(cid:4)r (cid:4) + ( ˙(cid:4)i t )2 exp(−(cid:4)i t )Yi((cid:4)t ) t )Yr ((cid:4)t ) t )2 exp(−(cid:4)i t )Yi((cid:4)t ) t )2 exp(−(cid:4)r 1 ( ˙(cid:4)i 4 i /∈Kr (cid:18)( ˙(cid:4)t )2, ∇F r ((cid:4)t )(cid:19). As explained above, the claim now follows. ≤ 1 2 Thus, applying Claim 7.11 and by focusing on the term in (cid:18)( ˙(cid:4)t )2, ∇F r ((cid:4)t )(cid:19) corre- t and tossing out the rest—which are all non-negative—we get our desired sponding to (cid:4)1 bound: −(cid:18) ˙(cid:4)t , H[F r ((cid:4)t )] ˙(cid:4)t (cid:19) ≥ 1 2 (cid:18)( ˙(cid:4)t )2, ∇F r ((cid:4)t )(cid:19) ≥ 1 2 ( ˙(cid:4)1 t )2 exp(−(cid:4)1 t )Y1((cid:4)t ) ≥ 1 2 exp(−t). (7.33) For the last inequality, recall from (7.12) that 1 < Y1((cid:4)t ). Combining (7.33) with our previous bound on (cid:18)(cid:4)t , ∇F r ((cid:4)t )(cid:19) from (7.27), we see that (cid:21)((cid:4)t , ˙(cid:4)t )(cid:21)2 h,Rr = −(cid:18) ˙(cid:4)t , H[F r ((cid:4)t )] ˙(cid:4)t (cid:19) (cid:18)(cid:4)t , ∇F r ((cid:4)t )(cid:19) ≥ 1 2(B1t + B2) . Hence the length of this path in the entropy metric is at least " (cid:13) (cid:4)1 1 (cid:4)1 0 1 2(B1t + B2) dt = 1 √ 2B1 2 ( B1(cid:4)1 1 + B2 − B1(cid:4)1 0 + B2). https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 772 T. Aougab et al Before we can apply Lemma 7.10 to show that (M1(Rr ), dh,Rr ) has infinite diameter, we require two more estimates. The first states that for a length function in M1(Rr ) when (cid:4)r is bounded from below, there is an upper bound on the length of the shortest edge that is not r. LEMMA 7.12. Let r ≥ 2. If (cid:4) ∈ M1(Rr ) where (cid:4)r ≥ log(3), then min{(cid:4)i | i ∈ [r − 1]} ≤ log(4r − 5). Proof. We first prove the lemma under the additional assumption that (cid:4)i = (cid:4)1 for any i ∈ [r − 1]. In this case, we have that (cid:4)i = log(L) for i ∈ [r − 1] and some L > 2r − 3. By Lemma 7.7, we have log(3) ≤ (cid:4)r = log (cid:9) L + (2r − 1) L − (2r − 3) (cid:10) . Hence we have that 3(L − (2r − 3)) ≤ L + (2r − 1), which implies that L ≤ 4r − 5. Next we prove the general case. Let (cid:4) ∈ M1(Rr ) be such that (cid:4)r ≥ log(3). Without loss of generality, we assume that (cid:4)1 = min{(cid:4)i | i ∈ [r − 1]}. If (cid:4)1 ≤ log(2r − 3), then we are done. Otherwise, we may decrease the lengths (cid:4)2, . . . , (cid:4)r−1 to be equal to (cid:4)1 while increasing (cid:4)r to maintain the fact that the metric has unit entropy. The assumption that (cid:4)1 > log(2r − 3) ensures that (cid:4)r is finite. Denote the resulting metric by ˆ(cid:4). Observe that ˆ(cid:4)r ≥ (cid:4)r ≥ log(3). By the special case considered above, ˆ(cid:4)i ≤ log(4r − 5) for each i ∈ [r − 1]. As (cid:4)1 = ˆ(cid:4)1, this completes the proof of the lemma. The second estimate shows that when the length of an edge is small for a length function in M1(Rr ), the lengths of the other edges must be very large. LEMMA 7.13. Let r ≥ 2 and let (cid:4) ∈ M1(Rr ). For any (cid:15) > 0, if (cid:4)i ≤ (cid:15) for some i ∈ [r], then for any j ∈ [r] − {i} we have (cid:4)j > − log(exp((cid:15)) − 1). Proof. The subrose consisting of the edges i and j has entropy less than or equal to 1 (strictly less than 1 when r ≥ 3). By Lemma 7.7, this implies that (cid:10) (cid:9) (cid:4)j ≥ log exp((cid:4)i) + 3 exp((cid:4)i) − 1 > − log(exp((cid:4)i) − 1) ≥ − log(exp((cid:15)) − 1). We can now prove the main inequality in this section that shows that any path that shrinks the length of an edge to zero must have infinite length. PROPOSITION 7.14. Let r ≥ 2. For any D > 0, there is an (cid:15) > 0 such that for any (cid:4) ∈ M1(Rr ) with min{(cid:4)i | i ∈ [r]} ≤ (cid:15) we have dh,Rr (log(2r − 1) · 1, (cid:4)) ≥ D. Proof. Let L0 = max{log(4r − 5), Lr }, where Lr is the constant from Lemma 7.10. Fix an (cid:15) > 0 such that (cid:24) ( 1 √ 2B1 2 −B1 log(exp((cid:15)) − 1) + B2 − (cid:24) B1L0 + B2) ≥ D. https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 773 Since − log(x − 1) → ∞ as x → 1+, such an (cid:15) exists. Observe that for this (cid:15) we have that − log(exp((cid:15)) − 1) ≥ L0. Let (cid:4) ∈ M1(Rr ) be such that min{(cid:4)i | i ∈ [r]} ≤ (cid:15) and let (cid:4)t : [0, 1] → M1(Rr ) be a piecewise smooth path where (cid:4)0 = log(2r − 1) · 1 and (cid:4)1 = (cid:4). We will show that the entropy length of this path is at least D. As the path is arbitrary, this shows that dh,Rr (log(2r − 1) · 1, (cid:4)) ≥ D as desired. Without loss of generality, assume that (cid:4)r = min{(cid:4)i | i ∈ [r]}. Let δ0 ∈ [0, 1] be the ≥ log(3). minimal value so that (cid:4)r t Indeed, there is another index i ∈ [r − 1] such that (cid:4)i < log(3), then the δ0 entropy of the subgraph consisting of the ith and rth edges is greater than 1. This contradicts the fact that the entropy of (cid:4)δ0 is equal to 1. | i ∈ [r], t ∈ [δ0, 1]}. We observe that (cid:4)r δ0 = min{(cid:4)i t . If (cid:4)r δ0 = (cid:4)r δ0 = min{(cid:4)i t As there is an automorphism of the r-rose permuting any two edges, by redefining (cid:4)t if | i ∈ [r − 1]} for each t ∈ [δ0, 1]. By Lemma ≤ (cid:15), we ≤ log(4r − 5) ≤ L0. By Lemma 7.13, as (cid:4)r 1 necessary, we may assume that (cid:4)1 t 7.12, as (cid:4)r δ0 have that (cid:4)1 ≥ log(3), we have that (cid:4)1 δ0 1 > − log(exp((cid:15)) − 1) ≥ L0. Let δ1 ∈ [δ0, 1] be the minimal value so that (cid:4)1 t t ∈ [δ1, 1]. As Lh,Rr ((cid:4)t |[0, 1]) ≥ Lh,Rr ((cid:4)t |[δ1, 1]), it suffices to show that the latter is bounded below by D. ≥ L0 for all By only considering the portion of (cid:4)t along the subintervals of [δ1, 1] with ˙(cid:4)1 t > 0, we find by Lemma 7.10 that Lh,Rr ((cid:4)t |[δ1, 1]) ≥ 1 √ 2B1 2 ( B1(cid:4)1 1 + B2 − B1(cid:4)1 δ1 + B2). ≤ (cid:15), we have (cid:4)1 1 As (cid:4)r 1 Therefore ≥ − log(exp((cid:15)) − 1) by Lemma 7.13. By definition (cid:4)1 δ1 = L0. Lh,Rr ((cid:4)t |[δ1, 1]) ≥ (cid:24) ( 1 √ 2B1 2 −B1 log(exp((cid:15)) − 1) + B2 − (cid:24) B1L0 + B2) ≥ D. 8. Proof of Theorem 1.1 In this section we give the proof of the first main result of this paper. Theorem 1.1 states that (X1(Fr ), dh) is complete when r = 2 and not complete if r ≥ 3. Proof of Theorem 1.1. In §6 we showed that (X1(F2), dh) is complete (Proposition 6.8), and so it remains to show that (X1(Fr ), dh) is not complete when r ≥ 3. This is a simple consequence of Proposition 7.8, as we now explain. Let r ≥ 3 and let (cid:4)t : [0, 1) → M1(Rr ) be the path described in Proposition 7.8. Using the natural homeomorphism M1(Rr ) ↔ X1(Rr , id), we can consider (cid:4)t as a path in X1(Fr ). The sequence of length functions ((cid:4)1−1/n)n∈N is a Cauchy sequence in (X1(Fr ), dh) as the entropy distance on M1(Rr ) is an upper bound on the entropy distance on X1(Fr ). We claim that this sequence does not have a limit. Indeed, any length function (cid:4) that does not lie in X1(Rr , id) has an open neighborhood in the weak topology that does not intersect X1(Rr , id). As the metric topology and the weak topology agree, any possible → ∞ as n → ∞, we limit of this sequence must lie in X1(Rr , id). However, as (cid:4)r 1−1/n https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 774 T. Aougab et al see that for any (cid:4) ∈ X1(Rr , id), there is an open neighborhood of (cid:4) in the weak topology U ⊂ X1(Fr ) such that (cid:4)1−1/n ∈ U for only finitely many n. Hence, again as the metric topology and the weak topology agree, we see that this sequence does not have a limit in X1(Rr , id). 9. The completion of (M1(Rr ), dh,Rr ) The goal of this section is to prove Theorem 1.2 which states that the completion of (M1(Rr ), dh,Rr ) is homeomorphic to the complement of the vertices in an (r − 1)-simplex. Intuitively, the newly added completion points correspond to unit-entropy metrics on the subroses of Rr consisting of at least two edges. Specifically, a face of dimension k − 1 corresponds to unit-entropy metrics on a sub-k-rose. We observe that R1 does not possess a metric with unit entropy. This accounts for the missing vertices in the completion. There are two steps to the proof. First, in §9.1 we introduce a model space (cid:2)M1 (Rr ) for the completion of M1(Rr ) with respect to the entropy metric. This model considers M(Rr ) as a subset of [0, ∞]r and adds the faces where at most r − 2 of the coordinates are equal to ∞. It is apparent from the construction that (cid:2)M1 (Rr ) is homeomorphic to the complement of the vertices in an (r − 1)-simplex. Proposition 9.6 shows that the distance function dh,Rr on M1(Rr ) extends to a distance function on (cid:2)M1 (Rr ). It is clear that M1(Rr ) is dense in (cid:2)M1 (Rr ). In §9.2 we complete the proof of Theorem 1.2 by showing that ((cid:2)M1 (R3) and contrasts this with the closure of CV (R3, id) considered as a subset of RF3 in the axis topology. (Rr ), dh,Rr ) is complete. Example 9.7 illustrates (cid:2)M1 Finally, in §9.3 we show that the diameter of cross-section of M1(Rr ) consisting of length functions with (cid:4)i = (cid:15) for some fixed i ∈ [r] and (cid:15) > 0 goes to zero as (cid:15) → 0+. This is important for §11 where we show that X1(Rr , id) is a bounded subset of (X1(Fr ), dh). 9.1. The model space (cid:2)M1 In this section we introduce a model for the completion of (M1(Rr ), dh,Rr ). As mentioned above, we add the faces to M1(Rr ) corresponding to subroses consisting of at least two edges where the rest of the edge lengths are infinite. (Rr ). Topologize [0, ∞] as a closed interval. The natural inclusion M1(Rr ) ⊂ (0, ∞)r ⊂ [0, ∞]r is an embedding. By setting x + ∞ = ∞ and exp (−∞) = 0 we get that the functions F r from (7.5), Xi from (7.2), Yi from (7.7), Xij from (7.16) and Yij from (7.17) extend to continuous functions on [0, ∞]r , and the entropy function h r ((cid:4)) extends to (0, ∞]r . Given a subset S ⊆ [r], we identify the subsets MS = {(cid:4) ∈ (0, ∞]r | (cid:4)i < ∞ if i ∈ S and (cid:4)i = ∞ if i /∈ S}, M1 S = {(cid:4) ∈ MS | F r ((cid:4)) = 0}. and Notice that M[r] = M(Rr ) and that MS ∩ MS(cid:17) = ∅ if S (cid:16)= S(cid:17). We further observe that M1 {i} = ∅ for any i ∈ [r]. For the latter, note that Xi((cid:4)) = 1 and Yi((cid:4)) = 1 for any (cid:4) ∈ M{i}. Thus for (cid:4) ∈ M{i} we have that F r ((cid:4)) = 1 − exp(−(cid:4)i) > 0. ∅ = ∅ and M1 https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 775 Fix S ⊆ [r] with |S| > 1 and let ιS : S → {1, 2, . . . , |S|} be the order-preserving bijection and let εS : [0, ∞]|S| → [0, ∞]r be the embedding defined by εS((cid:4)1, . . . , (cid:4) |S| )i = (cid:6) if i ∈ S, (cid:4)ιS (i) ∞ otherwise. The function εS allows us to consider a length function on R|S| as a length function on Rr where the edges not in S have infinite length. Indeed, with these definitions we have εS(M(R|S|)) = MS. The following lemma is immediate from Proposition 7.3 and the definitions. LEMMA 9.1. Let r ≥ 2 and let S ⊆ [r] have |S| > 1. Then the following statements are true. (1) F r ◦ εS = F |S|. (2) For (cid:4) ∈ M(R|S|), we have h (3) εS restricts to a homeomorphism M1(R|S|) → M1 S. r (εS((cid:4))) = h|S|((cid:4)). Next we define the sets (cid:2)M(Rr ) = (0, ∞]r = (cid:5) S⊆[r] MS, (cid:2)M1 (Rr ) = {(cid:4) ∈ (cid:2)M(Rr ) | F r ((cid:4)) = 0} = (cid:5) S⊆[r] M1 S. The set (cid:2)M1 (Rr ) is homeomorphic to the complement of the set of vertices in an (r − 1)-simplex. Applying Proposition 7.5, Lemma 9.1, and the above definitions, we get the following result. PROPOSITION 9.2. Let r ≥ 2. A length function (cid:4) ∈ (cid:2)M(Rr ) lies in (cid:2)M1 it has entropy equal to 1. (Rr ) if and only if As Yi((cid:4)) and Yij ((cid:4)) extend to continuous functions on (cid:2)M(Rr ), using the expressions for the partial derivatives of F r ((cid:4)) in (7.10), (7.20) and (7.21), we see that ∂iF r ((cid:4)), ∂iiF r ((cid:4)) and ∂ij F r ((cid:4)) extend to continuous functions on (cid:2)M(Rr ). Using these formulas, we see that extensions satisfy the following properties. LEMMA 9.3. Let r ≥ 2, let S ⊆ [r] have |S| > 1 and fix (cid:4) ∈ M(R|S|). Then for i, j ∈ [r] the following hold: (cid:6) ∂iF r (εS((cid:4))) = (cid:6) ∂ij F r (εS((cid:4))) = ∂ιS (i)F |S|((cid:4)) 0 if i ∈ S, otherwise; ∂ιS (i)ιS (j )F |S|((cid:4)) 0 if i, j ∈ S, otherwise. Hence both ∇F r ((cid:4)) and H[F r ((cid:4))] are well defined for (cid:4) ∈ (cid:2)M(Rr ). Additionally, the inner product (cid:18)(cid:4), ∇F r ((cid:4))(cid:19) extends continuously as we next show. https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 776 T. Aougab et al LEMMA 9.4. Let r ≥ 2. The function (cid:4) (cid:12)→ (cid:18)(cid:4), ∇F r ((cid:4))(cid:19) has a continuous extension to (cid:2)M1 (Rr ). Moreover, if S ⊆ [r] has |S| > 1 and (cid:4) ∈ M|S|, then (cid:18)εS((cid:4)), ∇F r (εS((cid:4)))(cid:19) = (cid:18)(cid:4), ∇F |S|((cid:4))(cid:19). Proof. Both of these statements follow from the expressions for the partial derivatives of F r ((cid:4)) in (7.10) as x exp(−x) → 0 when x → ∞. We define the tangent bundle T (cid:2)M1 where (cid:18)v, ∇F r ((cid:4))(cid:19) = 0. The subset of ((cid:4), v) ∈ T (cid:2)M1 S. For consistency, we denote T M1 by T M1 εS : R|S| → Rr defined by (Rr ) as the subspace of ((cid:4), v) ∈ (cid:2)M1 (Rr ) × Rr (Rr ) where (cid:4) ∈ M1 S is denoted [r] by T M1(Rr ). There is an embedding εS(v1, . . . , v|S| )i = (cid:6) vιS (i) 0 if i ∈ S, otherwise. (Rr ) whose image is S by TS((cid:4), v) = (εS((cid:4)), εS(v)). Proposition 4.14, together with Lemmas This allows us to define an embedding TS : T M1(R|S|) → T (cid:2)M1 contained in T M1 9.3 and 9.4, has the following implication. PROPOSITION 9.5. Let r ≥ 2. The entropy norm (cid:21)(cid:2)(cid:21)h,Rr : T M1(Rr ) → R extends to a continuous semi-norm (cid:21)(cid:2)(cid:21)h,Rr : T (cid:2)M1 (Rr ) → R. Moreover, the embedding maps TS : T M1(R|S|) → T (cid:2)M1 (Rr ) are norm-preserving. Specifically, if S ⊆ [r] has |S| > 1, and ((cid:4), v) = TS((cid:4)S, vS), then the extension satisfies (cid:21)((cid:4), v)(cid:21)h,Rr = (cid:21)((cid:4)S, vS)(cid:21)h,R|S| . (9.1) The reason why the extension fails to be a norm is as follows. If ((cid:4), v) ∈ M1 S vi = 0 for i ∈ S, then ((cid:4), v) ∈ T M1 whenever i /∈ S or j /∈ S. Thus −(cid:18)v, H[F r ]((cid:4))v(cid:19) = 0 and hence (cid:21)((cid:4), v)(cid:21)h,Rr points. × Rr and S as ∂iF r ((cid:4)) = 0 whenever i /∈ S. Further, ∂ij F r ((cid:4)) = 0 = 0 for such The following proposition is the main result of this section. It shows that (cid:2)M1 (Rr ) is contained in the completion of (M1(Rr ), dh,Rr ). PROPOSITION 9.6. Let r ≥ 2. The following statements hold. (1) The extension of the entropy norm defines a distance function dh,Rr on (cid:2)M1 The inclusion (M1(Rr ), dh,Rr ) → ((cid:2)M1 The topology induced by dh,Rr equals the subspace topology (cid:2)M1 (Rr ). (Rr ), dh,Rr ) is an isometric embedding. (Rr ) ⊆ [0, ∞]r . (2) (3) Proof. By definition (cid:2)M1 assume from now on that r ≥ 3. (R2) = M1(R2). Hence the proposition is obvious for r = 2. We First we need to show that for each (cid:4), (cid:4)(cid:17) ∈ (cid:2)M1 (Rr ) with (cid:4)0 = (cid:4) and (cid:4)1 = (cid:4)(cid:17) that has finite length so that dh,Rr ((cid:4), (cid:4)(cid:17)) is defined. Notice that by Proposition 9.5 for each S ⊆ [r] with |S| > 1, any two points in M1 S can be joined by a path of finite length. Hence it suffices to show that, for any S ⊆ [r] with |S| > 1, there is a path of finite length joining log(2r − 1) · 1 ∈ M1(Rr ) to εS(log(2|S| − 1) · 1) ∈ M1 S. (Rr ) there is a path (cid:4)t : [0, 1] → (cid:2)M1 https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 777 Recall that in Proposition 7.8 we showed that there is a path (cid:4)t : [0, 1] → (cid:2)M1 where (cid:4)0 = log(2r − 1) · 1 and (cid:4)1 = ε[r−1](log(2r − 3) · 1) ∈ M1 length. [r−1] (Rr ) that has finite Si+1 Now, given S ⊆ [r] with |S| > 1, we inductively define subsets Si for i = 0, . . . , r − |S| by S0 = [r], and Si+1 = Si − {max(Si − S)} so that Sr−|S| = S. The calculation (Rr ) in Proposition 7.8 shows that there is a finite-length path ((cid:4)i)t : with ((cid:4)i)0 = εSi (log(2(r − i) − 1) · 1) ∈ M1 and ((cid:4)i)1 = εSi+1(log(2(r − (i + 1)) − Si 1) · 1) ∈ M1 . The concatenation of these paths is a path with finite length from S. Therefore dh,Rr ((cid:4), (cid:4)(cid:17)) is defined log(2r − 1) · 1 ∈ M1(Rr ) to εS(log(2|S| − 1) · 1) ∈ M1 for all (cid:4), (cid:4)(cid:17) ∈ (cid:2)M1 [0, 1] → (cid:2)M1 Next, we show that dh,Rr defines a distance function on (cid:2)M1 (Rr ). Symmetry and the triangle inequality obviously hold. What needs to be checked is that if (cid:4) and (cid:4)(cid:17) are distinct points in (cid:2)M1 (Rr ), then there is an (cid:15) such that any path from (cid:4) to (cid:4)(cid:17) has length at least (cid:15). (Rr ). This is clear if at least one of (cid:4) and (cid:4)(cid:17) lie in M1(Rr ). Indeed, say (cid:4) lies in M1(Rr ). Then there is an open set U ⊂ M1(Rr ) containing (cid:4) with compact closure U such that (cid:4)(cid:17) /∈ U . Further, there is an (cid:15) > 0 such that if dh,Rr ((cid:4), (cid:4)(cid:17)(cid:17)) < (cid:15) then (cid:4)(cid:17)(cid:17) ∈ U . Hence any path from (cid:4) to (cid:4)(cid:17) must have length at least (cid:15) and therefore dh,Rr ((cid:4), (cid:4)(cid:17)) ≥ (cid:15) > 0. It remains to consider the case where neither (cid:4) nor (cid:4)(cid:17) lies in M1(Rr ). Suppose that there is a sequence of paths ((cid:4)n)t : [0, 1] → (cid:2)M1 (Rr ) − M1(Rr ) where Lh,Rr (((cid:4)n)t |[0, 1]) → 0. As the lengths of the paths ((cid:4)n)t go to 0, by Proposition 7.14, there is an (cid:15) > 0 such that (((cid:4)n)t )i ≥ (cid:15) for all t ∈ [0, 1] and all n. Hence the images of the paths lie in a compact set of (cid:2)M1 (Rr ) and, by the Arzelà–Ascoli theorem, there is a path (cid:4)t : [0, 1] → (cid:2)M1 (Rr ) from (cid:4) to (cid:4)(cid:17) with length 0. (Rr ) from (cid:4), (cid:4)(cid:17) ∈ (cid:2)M1 The image of such a path must be contained in (cid:2)M1 (Rr ) − M1(Rr ). As (cid:5) (cid:2)M1 (Rr ) − M1(Rr ) = MS, we must have that Proposition 9.5 and hence (cid:4) = (cid:4)(cid:17). ˙(cid:4)t = 0 since the semi-norm is non-degenerate on any T M1 S by S⊂[r] This shows (1). Item (2) follows as any path in (cid:2)M1 (Rr ) with endpoints in M1(Rr ) is close to a path entirely contained in M1(Rr ) by continuity of the semi-norm. Item (3) now follows by continuity of the semi-norm. 9.2. Proof of Theorem 1.2. We can now complete the proof of Theorem 1.2 which states that the completion of M1(Rr ) with respect to the entropy metric is homeomorphic to the complement of the vertices of an (r − 1)-simplex. We accomplish this by showing that ((cid:2)M1 (Rr ), dh,Rr ) is the completion as we have already observed that this space is homeomorphic to the complement of the vertices of an (r − 1)-simplex. https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 778 T. Aougab et al Proof of Theorem 1.2. As the inclusion map (M1(Rr ), dh,Rr ) → ((cid:2)M1 (Rr ), dh,Rr ) is an isometric embedding by Proposition 9.6(2) and the image is clearly dense, it remains to show that ((cid:2)M1 (Rr ), dh,Rr ) is complete. To this end, let ((cid:4)n)n∈N ⊂ (cid:2)M1 (Rr ) be a Cauchy sequence. Then for each 1 ≤ i ≤ r, we ∞) = 0. What remains ∞ ∈ [0, ∞]r where F r ((cid:4)1 have that ((cid:4)i)n limits to some (cid:4)i ∞, . . . , (cid:4)r to be shown is that such a limiting length function (cid:4)∞ belongs to (cid:2)M1 ∞ (cid:16)= ∞}. If (cid:4)i (cid:16)= 0 for all i ∈ [r], then (cid:4) ∈ M1 S Let S = {i ∈ [r] | (cid:4)i ⊂ (cid:2)M1 (Rr ). are done. This is indeed always the case as by Proposition 7.14, the limiting length (cid:4)i bounded away from zero for all i since the sequence is Cauchy. (Rr ) and we ∞ is Example 9.7. In this example, we compare the completion (cid:2)M1 (R3) with the closure of CV (R3, id) considered as a subset of RF3. Recall that CV (R3, id) ⊂ CV (F3) is the space of length functions on R3 with unit volume, that is, the sum of the lengths of the edges is equal to 1. For the current discussion, the marking is irrelevant. For more information about the closure of CV (Fr ) in RFr we refer the reader to the papers by Bestvina and Feighn [7] and Cohen and Lustig [12]. We again consider Figure 5 in §7.1, which shows M1(Rr ) for r = 2 and r = 3. These images suggest that as the length of one of the edges goes to infinity, the moduli space M1(R3) limits to the moduli space M1(R2) for the subgraph consisting of the other two edges. There are three such R2 subgraphs in R3, each contributing a one-dimensional face to (cid:2)M1 (R3) contrasted with CV (R3, id), the closure of CV (R3, id) considered as a subset of RF3. The spaces are not homeo- morphic; (cid:2)M1 (R3) is a 2-simplex without vertices, whereas CV (R3, id) is a 2-simplex. A more striking difference comes from the duality between the newly added edges and vertices. (cid:129) (R3). Figure 6 shows a schematic of (cid:2)M1 In (cid:2)M1 (R3), as c → ∞ we obtain a copy of M1(R2) for the subgraph on a and b. In CV (R3, id), the corresponding sequence would send a, b → 0, c → 1 and the result is a single point corresponding to the graph of groups decomposition of (cid:18)a, b, c(cid:19) with underlying graph R1 where (cid:18)a, b(cid:19) is the vertex group and the edge group is trivial. In (cid:2)M1 (R3), as b, c → ∞ there is no limit. This is a missing vertex of the 2-simplex; this stems from the fact that R1 cannot be scaled to have unit entropy. In CV (R3, id), the corresponding sequence would send a → 0 and we obtain a one-dimensional face in the closure corresponding to unit-volume length functions on the graph of groups decomposition of (cid:18)a, b, c(cid:19) with underlying graph R2 where (cid:18)a(cid:19) is the vertex group and both edge groups are trivial. (cid:129) 9.3. The thin part of M1(Rr ). For (cid:15) > 0 and i ∈ [r], we define Si (cid:15) = {(cid:4) ∈ M1(Rr ) | (cid:4)i = (cid:15)}. We use the letter ‘S’ as we think of this subset as a slice of the moduli space. The goal of this section is to prove the following proposition. https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space (cid:5)b, c(cid:6) a c (cid:5)b(cid:6) 779 (cid:5)a, b(cid:6) c b a = ∞ a b c = ∞ a b (cid:5)c(cid:6) b c (cid:5)a(cid:6) a c b = ∞ (cid:5)a, c(cid:6) FIGURE 6. The completion of entropy normalization (cid:2)M1 (R3) contrasted with the closure of the volume normalization CV (R3, id) in RF3 . PROPOSITION 9.8. Let r ≥ 2 and let i ∈ [r]. Then lim (cid:15)→0+ diam(Si (cid:15)) = 0. Topologically, we have seen that (cid:2)M1 (Rr ) is homeomorphic to a simplex with its vertices removed. Proposition 9.8 shows that geometrically (cid:2)M1 (Rr ) is similar to an ideal hyperbolic simplex, with cross-sections whose diameter shrinks to zero as we move toward an ideal vertex. Given distinct i, j ∈ [r] and (cid:15) > 0, we let (cid:4)i,j ((cid:15)) denote the length function in M1 {i,j } where ((cid:4)i,j ((cid:15)))i = (cid:15). As a result, we get that ((cid:4)i,j ((cid:15)))j = log (cid:9) (cid:10) exp(−(cid:15)) + 3 exp(−(cid:15)) − 1 (9.2) by Lemma 7.7. We begin with a technical lemma that bounds the length of a path in Si (cid:15) to such a point. LEMMA 9.9. Let r ≥ 2. There is a constant D with the following property. Let 0 < (cid:15) < log(2) and let i ∈ [r]. Suppose (cid:4) ∈ Si (cid:15) and that j ∈ [r] − {i} is such that (cid:4)j = min{(cid:4)k | k ∈ [r] − {i}}. Then dh,Rr ((cid:4), (cid:4)i,j ((cid:15))) ≤ D − log(exp((cid:15)) − 1) . Proof. Using the notation of the lemma, we consider the path (cid:4)t : [0, ∞) → Si by (cid:15) defined (cid:4)k t = (cid:4)k + t, k (cid:16)= i, j ; (cid:4)i t = (cid:15). (9.3) Note that (cid:4)j t Notice that (cid:4)t extends to a path (cid:4)t : [0, ∞] → (cid:2)M1 is not specified; its value is determined by the constraint that h (Rr ) and (cid:4)∞ = (cid:4)i,j ((cid:15)) ∈ M1 r ((cid:4)t ) = 1. {i,j }. We https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 780 T. Aougab et al t is increasing for k (cid:16)= i, j and as (cid:4)i t is constant, we have observe that since the length of (cid:4)k (cid:4)j t ≤ (cid:4)j for all t. Given a subset S ⊆ [r], let |S|i = |S| − 1 if i ∈ S and let |S|i = |S| otherwise. With = (cid:4)S + |S|it. Therefore, using this definition, for a subset S ⊆ [r] − {j } we have that (cid:4)S t the definition of Xj ((cid:4)t ) from (7.6) and Yj ((cid:4)t ) from (7.7), we find that (cid:4) Xj ((cid:4)t ) = (1 − 2|S|) exp(−(cid:4)S − |S|it), S⊆[r]−{j } (cid:4) (1 + 2|S|) exp(−(cid:4)S − |S|it). Yj ((cid:4)t ) = S⊆[r]−{j } (9.4) (9.5) Let p(t) = log(Yj ((cid:4)t )) and q(t) = − log(Xj ((cid:4)t )), and so (cid:4)j (t) = p(t) + q(t) by (7.9). The next two claims establish bounds on the second derivatives of p(t) and q(t). CLAIM 9.10. There is a constant C1 such that | ¨p(t)| ≤ C1 exp(−t). Proof of Claim 9.10. Using that fact that 1 < Yj ((cid:4)t ) from (7.12), we have ! ! ! ! ≤ ¨Yj ((cid:4)t )Yj ((cid:4)t ) − ˙Yj ((cid:4)t )2 Yj ((cid:4)t )2 ˙Yj ((cid:4)t ) Yj ((cid:4)t ) ¨Yj ((cid:4)t ) Yj ((cid:4)t ) | ¨p(t)| = ! ! ! ! + ! ! 2 ! ! ! ! ! ! ! ! ! ! ! ! ! ! ≤ | ¨Yj ((cid:4)t )| + | ˙Yj ((cid:4)t )|2. The summands in | ¨Yj ((cid:4)t )| have the form |S|2 i (1 + 2|S|) exp(−(cid:4)S − |S|it). (9.6) The summands in | ˙Yj ((cid:4)t )|2 have the form (cid:17)|i(1 + 2|S|)(1 + 2|S |S|i|S (cid:17)|) exp(−(cid:4)S − (cid:4)S(cid:17) − (|S|i + |S (cid:17)|i)t). (9.7) Each non-zero term in (9.6) and (9.7) has the form A exp(−B − Ct) where A, B ≥ 0 and C ≥ 1. Hence each term is bounded by A exp(−t) for some A ≥ 0. The existence of C1 is now clear. CLAIM 9.11. There is a constant C2 such that | ¨q(t)| ≤ C2 exp(−t). Proof of Claim 9.11. Using the facts that 1 < Yj ((cid:4)t ) from (7.12) and exp(−(cid:4)j Xj ((cid:4)t ) from (7.8), we find that exp(−(cid:4)j t )Yj ((cid:4)t ) = Xj ((cid:4)t ). Hence ! ! ! ! ! ! ! ! t )Xj ((cid:4)t ) − ˙Xj ((cid:4)t )2 t ) ≤ exp(−(cid:4)j | ¨q(t)| = ! ! ! ! ≤ ! ! ! ! + ¨Xj ((cid:4)j ! ! 2 ! ! ! ! ! ! ˙Xj ((cid:4)t ) Xj ((cid:4)t ) Xj ((cid:4)t )2 t )Yj ((cid:4)t ) = ¨Xj ((cid:4)t ) Xj ((cid:4)t ) t )| ˙Xj ((cid:4)t )|2. The summands in exp((cid:4)j The summands in exp(2(cid:4)j ≤ exp((cid:4)j t )| ¨Xj ((cid:4)t )| + exp(2(cid:4)j t )| ¨Xj ((cid:4)t )| have the form i (1 + 2|S|) exp((cid:4)j |S|2 t )| ˙Xj ((cid:4)t )|2 have the form t − (cid:4)S − |S|it). (9.8) |S|i|S (cid:17)|i(1 + 2|S|)(1 + 2|S (cid:17)|) exp(2(cid:4)j t − (cid:4)S − (cid:4)S(cid:17) − (|S|i + |S (cid:17)|i)t). (9.9) https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 781 ≤ min{(cid:4)k | k ∈ [r] − {i}}, we find that (cid:4)j t As (cid:4)j − (cid:4)S ≤ 0 for all S ⊆ [r] − {j } when t |S|i (cid:16)= 0. Hence, as above, each non-zero term in (9.8) and (9.9) has the form A exp(−B − Ct) where A, B ≥ 0 and C ≥ 1. The existence of C2 is now clear. We can now bound the entropy norm of ((cid:4)t , ˙(cid:4)t ) using Proposition 4.14. As ¨(cid:4)k t t )Yk((cid:4)t ) > 0 for all k from (7.10), we find that t ∂kF r ((cid:4)t ) = (cid:4)k t exp(−(cid:4)k k (cid:16)= j and (cid:4)k = 0 for (cid:21)((cid:4)t , ˙(cid:4)t )(cid:21)2 h,Rr = (cid:18) ¨(cid:4)t , ∇F r ((cid:4)t )(cid:19) (cid:18)(cid:4)t , ∇F r ((cid:4)t )(cid:19) ≤ ¨(cid:4)j t ∂j F r ((cid:4)t ) (cid:4)j t ∂j F r ((cid:4)t ) = ¨(cid:4)j t (cid:4)j t ≤ ¨(cid:4)j t (cid:4)j . Thus we can bound the length of the path (cid:4)t by (cid:13) ∞ 0 √ Lh,Rr ((cid:4)t |[0, ∞)) = ≤ (cid:13) ∞ (cid:21)((cid:4)t , ˙(cid:4)t )(cid:21)h,Rr dt ≤ 1 (cid:4)j (cid:13) ∞ exp(−t/2) dt = 2 √ C1 + C2 (cid:4)j 0 0 ¨(cid:4)j t dt √ C1 + C2 (cid:4)j C1 + C2 we complete the proof of the . As (cid:4)j ≥ − log(exp((cid:15)) − 1) (9.2), setting D = 2 lemma. Proof of Proposition 9.8. Let r ≥ 2 and let D be the constant from Lemma 9.9. Fix i ∈ [r] = (cid:15) and such that all other (cid:4)j and let (cid:4)(cid:15) ∈ Si (cid:15) are equal. By Lemma 9.9, we have that (cid:15), defined by (cid:4)i (cid:15) dh,Rr ((cid:4)(cid:15), (cid:4)i,j ((cid:15))) ≤ D − log(exp((cid:15)) − 1) for all j ∈ [r] − {i}. In particular, the set {(cid:4)i,j ((cid:15)) | j ∈ [r] − {i}} has diameter at most 2D/(− log(exp((cid:15)) − 1)). Again, by Lemma 9.9, each (cid:4) ∈ Si (cid:15) has distance at most D/(− log(exp((cid:15)) − 1)) from some point in {(cid:4)i,j ((cid:15)) | j ∈ [r] − {i}}. Hence diam(Si (cid:15)) ≤ 3D − log(exp((cid:15)) − 1) . As − log(exp((cid:15)) − 1) → ∞ as (cid:15) → 0+, the proof of the proposition is complete. 1, . . . , e1 n1 , and G2, with edges e2 10. The moduli space of a graph with a separating edge The purpose of this section is to introduce tools to analyze the entropy metric on the moduli space of a graph with a separating edge. Throughout this section, let G = (V , E, o, τ , ¯) be a finite connected graph which consists of two disjoint connected subgraphs G1, with edges e1 1, . . . , e2 connected by an edge e0. We assume n2 that both G1 and G2 have rank at least 2. We begin our analysis in §10.1 by showing that there exist paths of finite length limiting to any unit-entropy metric on either G1 or G2. Using this, in §10.2 we construct a space (cid:2)M1 (G) that is similar to the construction of (cid:2)M1 (Rr ) from §9. The main difference is that in this section, we do not bother to construct the entire completion of (M1(G), dh,G), rather we just add the points that correspond to a length function of entropy 1 on G1 ∪ e0 or G2 ∪ e0 or G1 ∪ G2. This is sufficient for our purposes. Proposition 10.6 is the culmination of this analysis where we show that there is https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 782 T. Aougab et al a map from (cid:2)M1 length functions to a single point. (G) to the completion of (M1(G), dh,G) that collapses these newly added 10.1. Finite-length paths in M1(G). We seek to show that there is a finite-length path in (M1(G), dh,G) that limits onto an arbitrary unit-entropy metric on either G1 or G2. This calculation appears in Proposition 10.2. Given a length function (cid:4) ∈ M(G) we denote )) and (cid:4)2 = ((cid:4)(e2 (cid:4)0 = (cid:4)(e0), (cid:4)1 = ((cid:4)(e1 )). Given a simplex (cid:7) ∈ CG and an edge e ∈ E of G, we recall that (cid:7)(e) denotes the number of times e or ¯e appears as a vertex in a simple cycle contained in (cid:7). By the construction of CG we have that (cid:7)(e) ∈ {0, 1, 2} for any edge. Further, (cid:7)(e0) ∈ {0, 2} as e0 is separating. 1), . . . , (cid:4)(e1 n1 1), . . . , (cid:4)(e2 n2 Analogous to the functions Yi defined in §7.2 that allowed us to isolate the variable (cid:4)i for the r-rose, we define a function Y : M(G) → R by Y ((cid:4)) = − (−1) |(cid:7)| exp(−(cid:4)((cid:7)) + 2(cid:4)0). (cid:4) (cid:7)∈CG (cid:7)(e0)=2 Notice that this function is constant with respect to (cid:4)0 as we may write (cid:4) (cid:4)((cid:7)) = (cid:7)(e)(cid:4)(e). e∈E+ Hence if (cid:7)(e0) = 2, then (cid:4)((cid:7)) − 2(cid:4)0 has no dependence on (cid:4)0. Also we remark that the function Y : M(G) → R has a continuous extension to [0, ∞]|E+| and is bounded on [0, ∞]|E+|. With this notation we have the following expression for FG. LEMMA 10.1. Let (cid:4) ∈ M(G). Then FG((cid:4)) = FG1((cid:4)1)FG2((cid:4)2) − exp(−2(cid:4)0)Y ((cid:4)). Proof. Let (cid:7) be a simplex in CG. If (cid:7)(e0) = 0, then (cid:7) is the join of two (possibly empty) simplices (cid:7)1 ∈ CG1 and (cid:7)2 ∈ CG2. Indeed, if (cid:7) = {γ 1 }, 1 , . . . , γ 1 m1 } for i = 1, 2. We have |(cid:7)| = m1 + m2 = then (cid:7) = (cid:7)1 ∗ (cid:7)2 where (cid:7)i = {γ i |(cid:7)1| + |(cid:7)2| and thus 1 , . . . , γ 2 m2 1 , . . . , γ i mi , γ 2 (−1) |(cid:7)| exp(−(cid:4)((cid:7))) = ((−1) |(cid:7)1| exp(−(cid:4)1((cid:7)1)))((−1) |(cid:7)2| exp(−(cid:4)2((cid:7)2))). Therefore, by Theorem 4.2, we find that (cid:4) (−1) |(cid:7)| exp(−(cid:4)((cid:7))) = (cid:9) (cid:4) (−1) |(cid:7)1| exp(−(cid:4)1((cid:7)1)) (cid:10) (cid:7)∈CG (cid:7)(e0)=0 (cid:7)1∈CG1 (cid:9) (cid:4) × (−1) |(cid:7)2| exp(−(cid:4)2((cid:7)2)) (cid:10) (cid:7)2∈CG2 = FG1((cid:4)1)FG2((cid:4)2). If (cid:7)(e0) = 2, then |(cid:7)| (−1) exp(−(cid:4)((cid:7))) = exp(−2(cid:4)0)(−1) |(cid:7)| exp(−(cid:4)((cid:7)) + 2(cid:4)0). https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 783 Hence, by the definition of Y ((cid:4)), we have (cid:4) (−1) |(cid:7)| exp(−(cid:4)((cid:7))) = − exp(−2(cid:4)0)Y ((cid:4)). (cid:7)∈CG (cid:7)(e0)=2 As e0 is separating, there are no simplices in CG for which (cid:7)(e0) = 1. By Theorem 4.2 again, the lemma follows. In particular, for (cid:4) ∈ M1(G) we have FG((cid:4)) = 0 by Lemma 4.4, and thus (cid:9) (cid:10) (cid:4)0 = 1 2 log Y ((cid:4)) FG1((cid:4)1)FG2((cid:4)2) . (10.1) Using the above expression for FG and (cid:4)0, we will give a method of a finite-length path in G for which (cid:4)0 → ∞. PROPOSITION 10.2. Fix a length function (cid:4)(cid:17) ∈ M1(G1) and let 1 < x1, . . . , xn1 < ∞ be such that (cid:4)(cid:17)(e1 i ) = log(xi) for 1 ≤ i ≤ n1. Let (cid:4) ∈ M1(G) be such that (cid:4)(e1 i ) = log(xi + 1) for 1 ≤ i ≤ n1 and let (cid:4)t : [0, 1) → (cid:2)M1 i ) = log(xi + 1 − t), (cid:4)2 t (G) be the path defined by (cid:4)t (e1 = (cid:4)2 and (cid:9) (cid:10) (cid:4)0 t = 1 2 Then Lh,G((cid:4)t |[0, 1)) is finite and (cid:4)0 t Y ((cid:4)t ) t )FG2((cid:4)2 t ) log FG1((cid:4)1 → ∞ as t → 1−. . (10.2) Proof. We will use the notation as in the statement of the proposition. As h we must have that (cid:4)0 t unit entropy, and with this length function the subgraph G1 also has unit entropy. G1((cid:4)(cid:17)) = 1, → ∞ at t → 1−. Indeed, if not then the limiting length function has Notice that since (cid:4)t (e)∂eFG((cid:4)t ) > 0 for all e ∈ E+ by Lemma 4.4(3), we have that (cid:18)(cid:4)t , ∇FG((cid:4)t )(cid:19) = (cid:4) e∈E+ (cid:4)t (e)∂eFG((cid:4)t ) ≥ n1(cid:4) i=1 By Lemma 10.1, we compute that (cid:4)t (e1 i )∂e1 i FG((cid:4)t ). (10.3) ∂e1 i FG((cid:4)t ) = FG2((cid:4)2 t )∂e1 i FG1((cid:4)1 t ) − exp(−2(cid:4)0 t )∂e1 i Y ((cid:4)t ). (10.4) i Y ((cid:4)) is bounded on M(G), we see that (cid:18)(cid:4)t , ∇FG((cid:4)t )(cid:19) has a limit, as t → 1−, Thus since ∂e1 that is bounded below by FG2((cid:4)2)(cid:18)(cid:4)(cid:17), ∇FG1((cid:4)(cid:17))(cid:19), which is positive by Lemma 4.4(3) and Lemma 4.9. (We will see in Lemma 10.4(2) that the limit is in fact exactly equal to FG2((cid:4)2)(cid:18)(cid:4)(cid:17), ∇FG1((cid:4)(cid:17))(cid:19).) Hence there is an (cid:15) > 0 such that (cid:18)(cid:4)t , ∇FG((cid:4)t )(cid:19) ≥ (cid:15) (10.5) for all t ∈ [0, 1). We compute that ¨(cid:4)t (e1 i ) = (−1)/(xi + 1 − t) < 0, hence ¨(cid:4)t (e1 FG((cid:4)t ) < 0 for all i )∂e1 FG((cid:4)t ) = 0 for all 1 ≤ i ≤ n2 and 0 < t < 1. i 0 < t < 1 and 1 ≤ i ≤ n1. Clearly ¨(cid:4)t (e2 i )∂e2 i https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 784 Thus T. Aougab et al (cid:18) ¨(cid:4)t , ∇FG((cid:4)t )(cid:19) ≤ ¨(cid:4)0 t ∂e0FG((cid:4)t ). (10.6) To deal with this term, we need the following claim. CLAIM 10.3. There are polynomials p, q ∈ R[t] where p(t), q(t) (cid:16)= 0 for t ∈ [0, 1] such that exp(−2(cid:4)0 t ) = (1 − t)p(t) . q(t) Proof of Claim 10.3. As FG((cid:4)t ) = 0, we have that t ) = FG1((cid:4)1 exp(−2(cid:4)0 t )FG2((cid:4)2 t ) Y ((cid:4)t ) . (cid:7) Let (cid:4)(Ei) = ni j =1 (cid:4)(ei j ) for i = 1, 2. Notice that we can write FG1((cid:4)1 n1(cid:20) (cid:4) t ) as FG1((cid:4)1 t ) = (−1) |(cid:7)| (xi + 1 − t) −(cid:7)(e1 i ). (cid:7)∈CG1 i=1 Hence exp(2(cid:4)t (E1))FG1 ((cid:4)1 that exp(2(cid:4)t (E2))FG2((cid:4)2 we have that FG1((cid:4)1 t ) is a polynomial in t with real coefficients. Also, we observe t ) is a non-zero constant with respect to t. By the definition of (cid:4)t 1) = 0. Hence we can write exp(2(cid:4)t (E1) + 2(cid:4)t (E2))FG1 ((cid:4)1 t )FG2((cid:4)2 t ) = (1 − t)p(t) where p(t) ∈ R[t]. As the left-hand side of this equation does not vanish for t ∈ [0, 1) by Lemma 4.9, we see that p(t) (cid:16)= 0 for t ∈ [0, 1). To show that p(1) (cid:16)= 0, we see that p(1) = d dt (exp(2(cid:4)t (E1) + 2(cid:4)t (E2))FG1 ((cid:4)1 1)(cid:18) ˙(cid:4)1 = exp(2(cid:4)1(E1) + 2(cid:4)1(E2))FG2 ((cid:4)2 ! ! t )FG2((cid:4)2 t )) 1, ∇FG1((cid:4)1 t=1 1)(cid:19). ∈ M1(G1), we have that ∇FG1((cid:4)1 As (cid:4)1 1 Lemma 4.4. Since h (past t = 1 as well), we have that (cid:18) ˙(cid:4)1 G1((cid:4)1 1) by t ) is increasing with respect to t as every edge length is decreasing 1) is non-zero and parallel to ∇h G1((cid:4)1 1)(cid:19) (cid:16)= 0. Hence p(1) (cid:16)= 0 as well. 1, ∇h G1((cid:4)1 In a similar way, we observe that we can write exp(2(cid:4)t (E1) + 2(cid:4)t (E2))Y ((cid:4)t ) = q(t) for some q(t) ∈ R[t]. As Y ((cid:4)t ) = exp(2(cid:4)0)FG1((cid:4)1 t ) by Lemma 10.1, we see that Y ((cid:4)t ) (cid:16)= 0 for t ∈ [0, 1) by Lemma 4.9 and hence q(t) (cid:16)= 0 for t ∈ [0, 1) as well. As t → 1−, we have that (cid:4)0 → ∞ and thus (1 − t)p(t)/q(t) → 0 as t → 1−. As p(1) (cid:16)= 0, we t must have that q(1) (cid:16)= 0 as well. t )FG2((cid:4)2 By Claim 10.3, we compute that (cid:9) ¨(cid:4)0 t = 1 2 1 (1 − t)2 − ¨p(t)p(t) + ( ˙p(t))2 (p(t))2 + ¨q(t)q(t) + ( ˙q(t))2 (q(t))2 (cid:10) . https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 785 Using Lemma 10.1 and Claim 10.3, we find that ∂e0FG((cid:4)) = 2 exp(−2(cid:4)0)Y ((cid:4)) = 2Y ((cid:4)) (1 − t)p(t) q(t) . Hence we see that there exists a constant C > 0 such that t ∂e0FG((cid:4)t ) ≤ C ¨(cid:4)0 1 − t . (10.7) Therefore, by combining Proposition 4.6 with (10.5), (10.6) and (10.7), we have that (cid:21)((cid:4)t , ˙(cid:4)t )(cid:21)2 h,G = (cid:18) ¨(cid:4)t , ∇FG((cid:4)t )(cid:19) (cid:18)(cid:4)t , ∇FG((cid:4)t )(cid:19) ≤ ¨(cid:4)0 t ∂e0FG((cid:4)t ) (cid:18)(cid:4)t , ∇FG((cid:4)t )(cid:19) ≤ C (cid:15)(1 − t) . Hence the length of (cid:4)t is finite. 10.2. The model space (cid:2)M1 (G). The previous example shows that we should expect some points in the completion of (M1(G), dh,G) to correspond to unit-entropy metrics on G1 or G2. For the model, we add these points to M1(G) as well as points that correspond to unit-entropy metrics on G1 ∪ G2. To this end, we set M1 = {(cid:4) ∈ (0, ∞] M2 = {(cid:4) ∈ (0, ∞] M1,2 = {(cid:4) ∈ (0, ∞] |E+| | (cid:4)1 ∈ M(G1) and (cid:4)2 = ∞ · 1}, |E+| | (cid:4)1 = ∞ · 1 and (cid:4)2 ∈ M(G2)}, |E+| | (cid:4)0 = ∞, (cid:4)1 ∈ M(G1) and (cid:4)2 ∈ M(G2)}. We consider their union (cid:2)M(G) = M(G) ∪ M1 ∪ M2 ∪ M1,2 as a subset of (0, ∞]|E+|, endowed with the subspace topology as in §9. There are obvious embeddings εi : M(Gi) → Mi for i = 1, 2 where we set εi((cid:4))0 = ∞, and an obvious embedding ε1,2 : M(G1) × M(G2) → M1,2 as well. Next, we define M1 1 M1 2 M1 1,2 = {(cid:4) ∈ M1 | h = {(cid:4) ∈ M2 | h = {(cid:4) ∈ M1,2 | max{h G1((cid:4)1) = 1}, G2((cid:4)2) = 1}, G1((cid:4)1), h G2((cid:4)2)} = 1}. Our model space is the union of these sets. Specifically, we define (cid:2)M1 (G) = M1(G) ∪ M1 1 ∪ M1 2 ∪ M1 1,2. (10.8) Using (4.2), we see that each partial derivative of FG extends to a bounded continuous function on (cid:2)M(G). The naturality of these extensions is the same as in Lemma 9.3. Even more, the inner product (cid:18)(cid:4), ∇FG((cid:4))(cid:19) has a continuous extension as was the case in Lemma 9.4. LEMMA 10.4. The function (cid:4) (cid:12)→ (cid:18)(cid:4), ∇FG((cid:4))(cid:19) has a continuous extension to (cid:2)M(G). This extension is such that the following statements hold. (1) If i ∈ {1, 2} and (cid:4) ∈ Mi, then (cid:18)(cid:4), ∇FG((cid:4))(cid:19) = (cid:18)(cid:4)i, ∇FGi ((cid:4)i)(cid:19). https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 786 T. Aougab et al (2) If (cid:4) ∈ M1,2, then (cid:18)(cid:4), ∇FG((cid:4))(cid:19) = FG2((cid:4)2)(cid:18)(cid:4)1, ∇FG1((cid:4)1)(cid:19) + FG1((cid:4)1)(cid:18)(cid:4)2, ∇FG2((cid:4)2)(cid:19). (3) For all (cid:4) ∈ (cid:2)M(G), we have (cid:18)(cid:4), ∇FG((cid:4))(cid:19) ≥ 0 with equality if and only if h G1((cid:4)1) = h G2((cid:4)2) = 1. Proof. From Lemma 4.4(3) and the expression for ∂eFG((cid:4)) in (4.2), we see that there is a constant A > 0 such that 0 < ∂eFG((cid:4)) ≤ A exp(−(cid:4)(e)) for any edge e ∈ E+. The existence of the continuous extension now follows for the same reason as for Lemma 9.4. If (cid:4)1 = ∞ · 1, then Y ((cid:4)) = 0. Indeed, this follows as every simple cycle in G that contains e0 must also contain an edge in G1 as e0 is separating. Likewise, if (cid:4)2 = ∞ · 1, then Y ((cid:4)) = 0 as well. Hence ∂e0FG((cid:4)) = 2 exp(−2(cid:4)0)Y ((cid:4)) = 0 for (cid:4) ∈ Mi when i = 1, 2. This, together with the paragraph above, shows (1). Using Lemma 10.1, we compute the following expression for (cid:18)(cid:4), ∇FG((cid:4))(cid:19): (cid:18)(cid:4), ∇FG((cid:4))(cid:19) = FG2((cid:4)2)(cid:18)(cid:4)1, ∇FG1((cid:4)1)(cid:19) + FG1((cid:4)1)(cid:18)(cid:4)2, ∇FG2((cid:4)2)(cid:19) − exp(−2(cid:4)0)((cid:18) ˆ(cid:4), ∇Y ( ˆ(cid:4))(cid:19) − 2(cid:4)0Y ( ˆ(cid:4))) From this (2) is apparent. As (cid:18)(cid:4), ∇FG((cid:4))(cid:19) > 0 for all (cid:4) ∈ M1(G) by Lemma 4.4(3), the extension is clearly non-negative. Statement (3) now follows from (1) and (2) as (cid:18)(cid:4), ∇FGi ((cid:4))(cid:19) > 0 for any (cid:4) ∈ M1(Gi) again by Lemma 4.4(3) and FGi ((cid:4)) > 0 for any (cid:4) ∈ M1(Gi) if h(Gi)((cid:4)) < 1 by Lemma 4.9. Next we partition (cid:2)M1 points, respectively: (G) into two subsets that we call the singular points and regular (cid:2)M1 (cid:2)M1 (G)sing = {(cid:4) ∈ (cid:2)M1 (G)reg = (cid:2)M1 (G) | h (G) − (cid:2)M1 G1((cid:4)1) = h (G)sing. G2((cid:4)2) = 1}, Notice that (cid:2)M1 (G)sing is a subset of M1 1,2. As in §9.1, we also define the tangent bundle T (cid:2)M1 (G) to be the subspace (G) × R|E+| where (cid:18)v, ∇FG((cid:4))(cid:19) = 0. There are obvious embeddings (G) for i = 1, 2 defined using εi : M1(Gi) → Mi as in (G)reg. of ((cid:4), v) ∈ (cid:2)M1 T εi : T M1(Gi) → T (cid:2)M1 §9.1. We define T (cid:2)M1 Proposition 4.6 together with Lemma 10.4 implies the following proposition. reg(G) to be the subset of ((cid:4), v) ∈ T (cid:2)M1 (G) where (cid:4) ∈ (cid:2)M1 PROPOSITION 10.5. The entropy norm (cid:21)(cid:2)(cid:21)h,G : T M1(G) → R extends to a continuous semi-norm (cid:21)(cid:2)(cid:21)h,G : T (cid:2)M1 (G)reg → R. Moreover, the embedding maps T εi : T M1(Gi) → T (cid:2)M1 (G) are norm-preserving. As in §9.1, we have the following proposition that shows us that there is a map from (G) to the completion of M1(G) with respect to the entropy metric. (cid:2)M1 https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 787 PROPOSITION 10.6. The following statements hold. The entropy norm defines a pseudo-metric dh,G on (cid:2)M1 ∪ M1 (1) (2) We have diam(M1 1 1,2) = 0. The inclusion (M1(G), dh,G) → ((cid:2)M1 ∪ M1 2 (3) (G). (G), dh,G) is an isometric embedding. Proof. As in Proposition 9.6, we need to show that for any (cid:4), (cid:4)(cid:17) ∈ (cid:2)M1 (cid:4)t : [0, 1] → (cid:2)M(G) with (cid:4)0 = (cid:4) and (cid:4)1 = (cid:4)(cid:17) that has finite length. (G) there is a path M1 1,2. 2 or (cid:4)(cid:17) ∈ M1 To this end, fix a point (cid:4) ∈ M1(G). There are several cases depending on whether (cid:4)(cid:17) ∈ 1, (cid:4)(cid:17) ∈ M1 We first deal with the case that (cid:4)(cid:17) ∈ M1 G1(((cid:4)(cid:17))1) = 1. In Proposition 10.2 we produced a path (cid:4)t : [0, 1] → (cid:2)M1 h M1(G) and (cid:4)1 ∈ M1,2 is such that (cid:4)1 1 ˜(cid:4)t : [0, 1] → M1 1,2 from (cid:4)1 to (cid:4)(cid:17) in M1 ˜(cid:4)t = (∞, (cid:4)1 1,2. Without loss of generality we assume that (G) where (cid:4)0 ∈ = ((cid:4)(cid:17))1. We can concatenate the path (cid:4)t with a path 1,2 as follows. We define the path by 1, (1 − t) · (cid:4)2 1 and we observe that ˜(cid:4)0 = (cid:4)1 to ˜(cid:4)1 = (cid:4)(cid:17). Note that by the convexity of entropy we have that (cid:17) + t · ((cid:4) )2) h G2((1 − t) · (cid:4)2 1 + t · ((cid:4) (cid:17) )2) ≤ 1 with equality only possible at the endpoints. Hence the interior of the path ˜(cid:4)t lies in (cid:2)M1 (G)reg. Further, (cid:21)( ˜(cid:4)t , ˙˜(cid:4)t )(cid:21)h,G = 0 as the length of edges in G2 is changing linearly. Hence there is a path of finite length from a length function in M1(G) to any length function in M1 Next, we deal with the case that (cid:4)(cid:17) ∈ M1; the case of (cid:4)(cid:17) ∈ M2 is symmetric. We will show that we can connect (cid:4)(cid:17) to a length function in M1,2 with a concatenation of two paths that have finite length—in fact each has length 0. Let (cid:4)(cid:17)(cid:17) ∈ M(G2) have entropy less than 1. The paths ((cid:4)1)t and ((cid:4)2)t are as follows: 1,2. ((cid:4)1)t : [0, ∞] → (cid:2)M1 ((cid:4)2)t : [0, ∞] → (cid:2)M1 (G) (G) (cid:17) ((cid:4)1)t = (((cid:4) ((cid:4)2)t = (∞, ((cid:4) (cid:17) )0 + t, ((cid:4) (cid:17) (cid:17) )1, ((cid:4) )1, ∞ · 1), )2 + t · 1). The concatenation of ((cid:4)1)t |[0, ∞] and ((cid:4)2)t |[∞, 0] is a path from (cid:4)(cid:17) to (cid:4)(cid:17)(cid:17). We observe that ( ¨(cid:4)1)t = 0 and ( ¨(cid:4)2)t = 0 as edge lengths are changing linearly. Also, we observe that the interiors of these paths lie in (cid:2)M1 (G)reg. Hence (cid:21)(((cid:4)k)t , ( ˙(cid:4)k)t )(cid:21)h,G = 0 for k = 1, 2 showing that the paths have finite—in fact zero—length. This shows (10.6). The previous argument shows that for any (cid:4) ∈ M1 1,2 such that dh,G((cid:4), (cid:4)(cid:17)) = 0. Likewise the analogous statement holds for (cid:4) ∈ M1 2. Given, (cid:4), (cid:4)(cid:17) ∈ M1,2, we will show that dh,G((cid:4), (cid:4)(cid:17)) = 0, completing the proof of (10.6). There are four cases depending on the entropies of the length functions (cid:4)1, (cid:4)2, ((cid:4)(cid:17))1 and ((cid:4)(cid:17))2. 1, there is an (cid:4)(cid:17) ∈ M1 https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 788 T. Aougab et al The first case we consider is when h G1((cid:4)1) = 1 and h G2(((cid:4)(cid:17))2) = 1. In this case we consider the concatenation of the two paths ((cid:4)1)t and ((cid:4)2)t that are defined as follows: ((cid:4)1)t : [0, 1] → (cid:2)M1 ((cid:4)2)t : [0, 1] → (cid:2)M1 (G) (G) ((cid:4)1)t = (∞, (cid:4)1, (1 − t) · (cid:4)2 + t · ((cid:4) ((cid:4)2)t = (∞, (1 − t) · (cid:4)1 + t · ((cid:4) (cid:17) (cid:17) )2), (cid:17) )1, ((cid:4) )2). As above, the interiors of these paths lie in (cid:2)M1 lengths of edges are changing linearly. This completes this case. (G)reg and they have length 0 since the The case where h G2((cid:4)2) = 1 and h Next we consider the case where h G1(((cid:4)(cid:17))1) = 1 is similar. G1((cid:4)1) = 1 and h G1(((cid:4)(cid:17))1) = 1. Fix a length function (cid:4)(cid:17)(cid:17) ∈ M1 G2(((cid:4)(cid:17)(cid:17))2) = 1. By the first argument, we can connect both (cid:4) and (cid:4)(cid:17) to (cid:4)(cid:17)(cid:17) with paths of length 0. Concatenating these two paths shows that this case holds as well. 1,2 with h G2((cid:4)2) = 1 and h The case where h This completes the proof of (10.6). We observe that any path in (cid:2)M1 G2(((cid:4)(cid:17))2) = 1 is similar. 10.5, we have that the inclusion (M1(G), dh,G) → ((cid:2)M1 ding, hence (10.6) holds. (G) is close to a path in M1(G). Thus by Proposition (G), dh,G) is an isometric embed- 11. X1(Rr , id) has bounded diameter in X1(Fr ) In this section we make use of the collapsing phenomena witnessed in the previous section to show that even though (M1(Rr ), dh,Rr ) has infinite diameter (Proposition 7.14), the subspace (X1(Rr , id), dh) ⊂ (X1(Fr ), dh) has bounded diameter. The idea is as follows. Using the natural bijection M1(Rr ) ↔ X1(Rr , id), since (cid:2)M1 (Rr ) is the completion of (M1(Rr ), dh,Rr ) (§9) there is a map (cid:6) : (cid:2)M1 (Fr ) is the completion of (X1(Fr ), dh). Indeed, if a sequence ((cid:4)n) ⊂ (M1(Rr ), dh,Rr ) is Cauchy, then so is its image under (cid:6) in (X1(Fr ), dh) as (cid:6) is 1-Lipschitz. As (cid:2)M1 (Rr ) is homeomorphic to the complement of the vertices of an (r − 1)-simplex, we can consider (cid:6) as the map (cid:6) : (cid:7)r−1 − V → (cid:3)X1 (Fr ) where (cid:7)r−1 is the standard (r − 1)-dimensional simplex and V ⊂ (cid:7)r−1 is the set of vertices. We will show that the map (cid:6) extends to the vertex set V. Since the image (cid:6)((cid:7)r−1) is compact and contains X1(Rr , id), it follows that (X1(Rr , id), dh) has bounded diameter. (Fr ) where (cid:3)X1 (Rr ) → (cid:3)X1 (cid:23) This is accomplished by considering M1(Rr ) as the face of M1(G) for a particular graph G that has a separating edge and using the tools developed in §10. Lemma 11.1 {MS | 1 < |S| < r − 1} ⊂ (cid:2)M1 (Rr ) is collapsed to a single establishes that the subset ⊂ (cid:2)M1 point in the completion of (X1(Fr ), dh). We recall that M1 (Rr ) is the subset of S unit-entropy length functions on the subrose R|S| ⊆ Rr utilizing the edges in S ⊆ [r]; the length of an edge in [r] − S is ∞. The subset M1 S corresponds to an (|S| − 1)-dimensional face of (cid:7)r−1. Thus Lemma 11.1 shows that the entire (r − 3)-skeleton of (cid:7)r−1 is collapsed to a point by (cid:6) in (cid:3)X1 (Fr ). LEMMA 11.1. For r ≥ 4, (cid:6) maps the subset single point in (cid:3)X1 (Fr ). (cid:23) {MS | 1 < |S| < r − 1} ⊂ (cid:2)M1 (Rr ) to a https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 789 e1 1 e1 2 e2 1 e2 2 v1 e0 v2 e1 n1 e2 n2 FIGURE 7. The graph Gn1,n2 : there are n1 loop edges attached to v1 and n2 loop edges attached to v2. Proof. Fix r ≥ 4 and let S be a subset of [r] with 1 < |S| < r − 1. To begin, we claim that the image of M1 (Fr ). To this end, we set n1 = |S| and n2 = r − |S|. Notice that n1, n2 ≥ 2. Let Gn1,n2 be the graph that consists of two vertices v1 and v2, and edges e0, e1 . The edges are attached via the following table. S is a single point in (cid:3)X1 and e2 1, . . . , e1 n1 1, . . . , e2 n2 o v1 v1 v2 τ v2 v1 v2 e0 e1 i e2 i See Figure 7. We adopt the notation introduced in §10 for Gn1,n2. Let c : Gn1,n2 → Rr be the map induced by collapsing the edge e0 and let ρ : Rr → Gn1,n2 be a map so that c ◦ ρ is homotopic to id : Rr → Rr . Thus (cid:17) X1(Rr , id) ⊂ X1 Specifically, viewing X1 is the image of the embedding ε : M1(Rr ) → [0, ∞)1+n1+n2 where ε((cid:4))0 = 0. (Gn1,n2 , ρ) = {[(G, ρ (Gn1,n2, ρ) as a subset of [0, ∞)1+n1+n2, we see that X1(Rr , id) , (cid:4))] ∈ X(Gn1,n2, ρ) | h G((cid:4)) = 1}. Moreover, ε extends to an embedding (cid:2)M1 S) is the face of M1 1 S. By Proposition 10.6(2), the set M1 1 is homeomorphic to (0, ∞] × M1 (Fr ). Hence so does M1 S, completing the proof of our claim. (Rr ) → [0, ∞]1+n1+n2 in the same way. ⊂ (cid:2)M1 (Gn1,n2) ⊂ [0, ∞]1+n1+n2. Indeed, 1 maps Under this embedding, ε(M1 the set M1 to a single point in (cid:3)X1 As the diameter of the thin part goes to zero as the short edge goes to zero (Proposition 9.8), the point that M1 S is sent to is independent of S. The main result of this section now follows easily. PROPOSITION 11.2. For r ≥ 4, bounded diameter. the subspace (X1(Rr , id), dh) ⊂ (X1(Fr ), dh) has Proof. As explained above in the introduction to this section, by Theorem 1.2, there is a map (cid:6) : (cid:7)r−1 − V → (cid:3)X1 (Fr ). By Lemma 11.1, the map (cid:6) extends to V and the entire https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 790 e1 2 e1 1 e1 r−2 Γr e2 0 e2 1 e2 2 v w T. Aougab et al (cid:2)Γr v1 v2 e0 e1 1 e1 2 e1 r−2 e2 0 e2 1 e2 2 w FIGURE 8. The graphs (cid:19)r and (cid:3)(cid:19)r . In (cid:19)r there are r − 2 loop edges attached to v and three edges connecting v to w. In (cid:3)(cid:19)r , there are r − 2 loop edges attached to v1, three edges connecting v2 to w and a separating edge connecting v1 to v2. (r − 3)-skeleton of (cid:7)r−1 is mapped to a single point. As (cid:7)r−1 is compact, (cid:6)((cid:7)r−1) is compact and hence has bounded diameter. Thus X1(Rr , id) ⊂ (cid:6)((cid:7)r−1) ⊂ X1(F) has bounded diameter too. 12. Proof of Theorem 1.3 The goal of this final section is the proof of the last of the main results. Theorem 1.3 states that the Out(Fr )-invariant subcomplex (X1(Rr , id) · Out(Fr ), dh) has bounded diameter and, moreover, that the action of Out(Fr ) on the completion of (X1(Fr ), dh) has a global {MS | 1 < |S| < r − 1} for any marking of the fixed point. This point is the image of rose. We show that the image of this point in the completion is independent of the marking. This is done by showing that it is independent for markings that differ by a single simple move—we call such markings Nielsen adjacent. This is accomplished again by making use of a graph with a separating edge and the analysis in §10. This simple move suffices to connect any two markings and the theorem easily follows. (cid:23) Proof of Theorem 1.3. Given a marked rose (Rr , ρ), there is an embedding (cid:6)ρ : M1 (Rr )→X1(Fr ) whose image is X1(Rr , ρ). As in §11, this map extends to (cid:6)ρ : (cid:2)M1 (Rr ) → (cid:3)X1 (Fr ) is the completion of X1(Fr ) with the entropy metric. By | 1 < |S| < r − 1} to a single point in (cid:3)X1 (Fr ). Let xρ (cid:23) (Fr ) where (cid:3)X1 Lemma 11.1, (cid:6)ρ maps denote this point in X(F). {M1 S Given an integer r > 2, we define a graph (cid:19)r that has two vertices v and w, and edges 1, . . . , e1 e1 3. The edges are attached via the following table. r−2 and e2 1, e2 0, e2 o v τ v v w e1 i e2 i See Figure 8. We call such a graph a rose-theta graph. https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 791 Given i ∈ {0, 1, 2}, collapsing the edge e2 i induces a map ci : (cid:19)r → Rr . We say two marked roses (Rr , ρ1) and (Rr , ρ2) are Nielsen adjacent if there is a marked rose-theta graph ((cid:19)r , ρ) such that ρi (cid:5) ci ◦ ρ for i = 1, 2 for some collapses ci : (cid:19)r → Rr . Given any two marked roses, (Rr , ρ) and (Rr , ρ(cid:17)), it is well known that there is a sequence of markings ρ = ρ1, . . . , ρn = ρ(cid:17) such that (Rr , ρi−1) and (Rr , ρi) are Nielsen adjacent for i = 2, . . . , n. For instance, see [14]. We will prove the theorem by showing that if (Rr , ρ1) and (Rr , ρ2) are Nielsen adjacent, = xρ2. Notice that the collection {xid·φ} is invariant under the action by Out(Fr ). then xρ1 Hence this also shows that the action of Out(Fr ) on (cid:3)X1 (Fr ) has a global fixed point. To this end, let ((cid:19)r , ρ) be the marked rose-theta graph such that ρi (cid:5) ci ◦ ρ. We need to introduce a separating edge to take advantage of the shortcuts utilized in §10. Let (cid:3)(cid:19)r be the graph obtained from blowing up the vertex v in (cid:19)r . Specifically, in (cid:3)(cid:19)r there are three vertices v1, v2 and w, and edges e0, e1 1, . . . , e1 3. The edges are attached via the following table. r−2 and e2 1, e2 0, e2 o v1 v1 τ v2 v1 v2 w e0 e1 i e2 i See Figure 8. We adopt the notation from §10 for (cid:3)(cid:19)r . Let c : (cid:3)(cid:19)r → (cid:19)r be the map that collapses the edge e0. There is a marking ˆρ : Rr → ((cid:3)(cid:19)r , ˆρ) as a subset of [0, ∞]r+2, there are two (Rr ) → [0, ∞]r+2 where (cid:3)(cid:19)r such that c ◦ ˆρ (cid:5) ρ. Viewing X1 embeddings corresponding to ρ1 and ρ2 denoted ε1, ε2 : (cid:2)M1 εi((cid:4))0 = εi((cid:4))(e2 i ) = 0 for i = 1, 2. Let S ⊂ [r] denote the set of edges {ci(e1 S) and ε2(M1 = xρ2. is independent of i. Both ε1(M1 Proposition 10.6(2), we have that xρ1 1), . . . , ci(e1 S) are faces of M1 1,2 r−2)} in Rr . Notice that this set ((cid:3)(cid:19)r ). Hence by ⊂ (cid:2)M1 Acknowledgements. The authors thank the Institute for Computational and Experimental Research in Mathematics (ICERM) for hosting the workshop Effective and Algorithmic Methods in Hyperbolic Geometry and Free Groups, at which work on this project began. We also thank Jing Tao for suggesting Question 1.8 and the referee for a careful reading and for helpful comments. The first author is supported by NSF grant DMS-1807319. The second author is supported by Simons Foundation Grant No. 316383. The third author is supported by Simons Foundation Grant No. 637880. REFERENCES [1] Y. Algom-Kfir. Strongly contracting geodesics in outer space. Geom. Topol. 15(4) (2011), 2181–2233. [2] Y. Algom-Kfir and M. Bestvina. Asymmetry of outer space. Geom. Dedicata 156 (2012), 81–92. https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press 792 T. Aougab et al [3] Y. Algom-Kfir, E. Hironaka and K. Rafi. Digraphs and cycle polynomials for free-by-cyclic groups. Geom. [4] L. Bers. An extremal problem for quasiconformal mappings and a theorem by Thurston. Acta Math. Topol. 19(2) (2015), 1111–1154. 141(1–2) (1978), 73–98. [5] M. Bestvina. The topology of Out(Fn). Proc. Int. Congr. of Mathematicians, Vol. II (Beijing, 2002). Ed. T. Li. Higher Education Press, Beijing, 2002, pp. 373–384. [6] M. Bestvina. A Bers-like proof of the existence of train tracks for free group automorphisms. Fund. Math. 214(1) (2011), 1–12. [7] M. Bestvina and M. Feighn. Outer limits. Preprint, 1992, https://www.math.utah.edu/~bestvina/eprints/ bestvina.feighn..outer_limits.pdf. [8] M. Bestvina and M. Feighn. Hyperbolicity of the complex of free factors. Adv. Math. 256 (2014), 104–155. [9] R. Bowen. Equilibrium States and the Ergodic Theory of Anosov Diffeomorphisms (Lecture Notes in Mathematics, 470), revised edition. Springer-Verlag, Berlin, 2008. [10] M. Bridgeman. Hausdorff dimension and the Weil–Petersson extension to quasifuchsian space. Geom. [11] M. Bridgeman, R. Canary, F. Labourie and A. Samborino. The pressure metric for Anosov representations. Topol. 14(2) (2010), 799–831. Geom. Funct. Anal. 25(4) (2015), 1089–1179. [12] M. M. Cohen and M. Lustig. Very small group actions on R-trees and Dehn twist automorphisms. Topology [13] M. Culler and J. W. Morgan. Group actions on R-trees. Proc. Lond. Math. Soc. (3) 55(3) (1987), 571–604. [14] M. Culler and K. Vogtmann. Moduli of graphs and automorphisms of free groups. Invent. Math. 84(1) [15] D. Cvetkovi´c and P. Rowlinson. The largest eigenvalue of a graph: a survey. Linear Multilinear Algebra 34(3) (1995), 575–617. (1986), 91–119. 28(1–2) (1990), 3–33. [16] G. Daskalopoulos and R. Wentworth. Classification of Weil–Petersson isometries. Amer. J. Math. 125(4) (2003), 941–975. [17] M. Dehn. Papers on Group Theory and Topology. Springer-Verlag, New York, 1987, translated from the German and with introductions and an appendix by J. Stillwell, with an appendix by O. Schreier. [18] S. Dowdall and S. J. Taylor. Hyperbolic extensions of free groups. Geom. Topol. 22(1) (2018), 517–570. [19] B. Farb and L. Mosher. Convex cocompact subgroups of mapping class groups. Geom. Topol. 6 (2002), 91–152. [20] S. Francaviglia and A. Martino. Metric properties of outer space. Publ. Mat. 55(2) (2011), 433–473. [21] V. Guirardel and G. Levitt. Deformation spaces of trees. Groups Geom. Dyn. 1(2) (2007), 135–181. [22] L.-Y. Kao. Pressure type metrics on spaces of metric graphs. Geom. Dedicata 187 (2017), 151–177. [23] I. Kapovich and T. Nagnibeda. The Patterson–Sullivan embedding and minimal volume entropy for outer space. Geom. Funct. Anal. 17(4) (2007), 1201–1236. [24] I. Kapovich and I. Rivin. On the absence of McShane-type identities for the outer space. J. Algebra 320(10) (2008), 3659–3670. [25] H. Masur and M. Wolf. The Weil–Petersson isometry group. Geom. Dedicata 93 (2002), 177–190. [26] H. A. Masur and Y. N. Minsky. Geometry of the complex of curves. I. Hyperbolicity. Invent. Math. 138(1) (1999), 103–149. [27] C. T. McMullen. Thermodynamics, dimension and the Weil–Petersson metric. Invent. Math. 173(2) (2008), [28] C. T. McMullen. Entropy and the clique polynomial. J. Topol. 8(1) (2015), 184–212. [29] W. Parry and M. Pollicott. Zeta functions and the periodic orbit structure of hyperbolic dynamics. Astérisque 365–425. 187–188 (1990), 268. [30] F. Paulin. The Gromov topology on R-trees. Topology Appl. 32(3) (1989), 197–221. [31] M. Pollicott and R. Sharp. A Weil–Petersson type metric on spaces of metric graphs. Geom. Dedicata 172 (2014), 229–244. [32] D. Ruelle. Thermodynamic Formalism: The Mathematical Structures of Classical Equilibrium Statistical Mechanics (Encyclopedia of Mathematics and Its Applications, 5). Addison-Wesley, Reading, MA, 1978, with a foreword by G. Gallavotti and G.-C. Rota. [33] J.-P. Serre. Trees (Springer Monographs in Mathematics). Springer-Verlag, Berlin, 2003, translated from the French original by J. Stillwell, corrected 2nd printing of the 1980 English translation. [34] W. P. Thurston. Minimal stretch maps between hyperbolic surfaces. Preprint, 1998, arXiv:math/9801039. [35] K. Vogtmann. Automorphisms of free groups and outer space. Geom. Dedicata 94 (2002), 1–31. [36] S. Wolpert. Noncompleteness of the Weil–Petersson metric for Teichmüller space. Pacific J. Math. 61(2) (1975), 573–577. https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press Thermodynamic metrics on outer space 793 [37] S. Wolpert. Thurston’s Riemannian metric for Teichmüller space. J. Differential Geom. 23(2) (1986), 143–174. [38] S. A. Wolpert. The Weil–Petersson metric geometry. Handbook of Teichmüller Theory, Volume II (EMS IRMA Lectures in Mathematics and Theoretical Physics, 13). Ed. A. Papadopoulos. European Mathematical Society, Zürich, 2009, pp. 47–64. [39] B. Xu. Incompleteness of the pressure metric on the Teichmüller space of a bordered surface. Ergod. Th. & Dynam. Sys. 39(6) (2019), 1710–1728. https://doi.org/10.1017/etds.2021.165 Published online by Cambridge University Press
null
10.1371_journal.pone.0227230.pdf
Data Availability Statement: All relevant data are within the manuscript and its Supporting Information files.
All relevant data are within the manuscript and its Supporting Information files.
RESEARCH ARTICLE Optogenetically induced cellular habituation in non-neuronal cells Mattia BonzanniID 1, Nicolas Rouleau1, Michael Levin2, David L. KaplanID 1* a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Bonzanni M, Rouleau N, Levin M, Kaplan DL (2020) Optogenetically induced cellular habituation in non-neuronal cells. PLoS ONE 15(1): e0227230. https://doi.org/10.1371/journal. pone.0227230 Editor: Mark S. Shapiro, University of Texas Health Science Center, UNITED STATES Received: September 5, 2019 Accepted: December 13, 2019 Published: January 17, 2020 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0227230 Copyright: © 2020 Bonzanni et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the manuscript and its Supporting Information files. Funding: Funded by 1) ML and DK, No. 12171, Paul G. Allen Frontiers Group, https://alleninstitute. 1 Department of Biomedical Engineering, Allen Discovery Center, Tufts University, Medford, United States of America, 2 Department of Biology, Allen Discovery Center, Tufts University, Medford, United States of America * david.kaplan@tufts.edu Abstract Habituation, defined as the reversible decrement of a response during repetitive stimulation, is widely established as a form of non-associative learning. Though more commonly ascribed to neural cells and systems, habituation has also been observed in single aneural cells, although evidence is limited. Considering the generalizability of the habituation pro- cess, we tested the degree to which organism-level behavioral and single cell manifesta- tions were similar. Human embryonic kidney (HEK) cells that overexpressed an optogenetic actuator were photostimulated to test the effect of different stimulation protocols on cell responses. Depolarization induced by the photocurrent decreased successively over the stimulation protocol and the effect was reversible upon withdrawal of the stimulus. In addi- tion to frequency- and intensity-dependent effects, the history of stimulations on the cells impacted subsequent depolarization in response to further stimulation. We identified tetra- ethylammonium (TEA)-sensitive native K+ channels as one of the mediators of this habitua- tion phenotype. Finally, we used a theoretical model of habituation to elucidate some mechanistic aspects of the habituation response. In conclusion, we affirm that habituation is a time- and state-dependent biological strategy that can be adopted also by individual non- neuronal cells in response to repetitive stimuli. Introduction The behavioral manifestation of habituation is intuitive and can be simplified as a reversible asymptotic response decrement after repeated stimulations [1]. The seminal work of Thomp- son and Spencer [2] delineated the original characteristics of habituation which remain largely unchanged today [1]. The principal features, which are now succinctly summarized in ten points by Rankin and colleagues [1], represent the gold standard for the definition of behav- ioral habituation in organisms. Briefly, the habituation profile is, in most cases, an exponen- tial-like curve and, most importantly, the decremental response is reversible–a condition that distinguishes habituation from fatigue. The dependence of the habituation profile upon the parameters of the stimulus cannot be overstated. Indeed, they are affected by both the intensity and frequency of stimulation as well as by the stimulation history (i.e., series of stimulation]. A PLOS ONE | https://doi.org/10.1371/journal.pone.0227230 January 17, 2020 1 / 14 org/what-we-do/frontiers-group/. 2) M.L, No. TWCF0089/AB55, Templeton World Charity Foundation, https://www.templetonworldcharity. org/. 3) D.K., P41EB002520, National Institutes of Health, https://www.nih.gov/. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Habituation in single non-excitable cells generalizable mechanism for this phenomenon, however, is still lacking. So far, the dual pro- cess theory, proposed by Groves and Thompson [3], the stimulus-model comparator by Soko- lov [4] and the “negative-image model” by Ramaswami [5] are the most prominent theories which offer explanatory value. The formulation of a general hypothesis that explains the pro- cess is challenging, mainly due to the multivariate cellular mechanisms that underlie the pro- cesses. In order to overcome this difficulty, we recently proposed a model of habituation that does not require a priori knowledge of the system’s biological components [6]. Interestingly, some features of habituation can also be detected in non-neuronal system, [7] [8] [9] [10]. The evolutionary and cell-biological origins of learning are nowadays the focus of an emerging field—basal cognition; recent and classic work has sought to identify and mechanistically char- acterize primitive forms of learning in non-neural biological systems[11, 12]. So far, a clear understanding of the potential general nature of the habituation process has not been achieved. We took advantage of the overexpression of channelrodopsin2 (ChR2) to optogenetically stim- ulate human embryonic kidney (HEK) cells to highlight, if present, the fundamental similari- ties between behavioral and cellular manifestations of the habituation response and to potentially reveal new findings that can lead to a mechanistic understanding of the process itself. Here, we explored the first five of the ten points listed in the paper by Rankin and col- leagues (as the last five points refer to special cases or instances with more than one stimulus) in the in vitro aneural system. We found that the system responded to the repetitive stimula- tion with a reversible asymptotical, exponential-like profile; moreover, the cell system response was stimulation-dependent. This indicates that responses associated with single non-neuronal cells share a high degree of similarity with behavioral manifestations of habituation. Material and methods Cell culture and transfection For electrophysiological recordings, human embryonic kidney (HEK) cells were maintained in DMEM high glucose (Thermofisher) supplemented with 10% of fetal bovine serum (FBS; Gibco) and 2 mM of L-Glutamine (Sigma) at 37 C in a 5% C02 incubator. HEK were plated in 35 mm dishes and transfected with 1.5 μg of the pcDNA3.1/hChR2(H134R)-mCherry plasmid (Addgene #20938) using LipofectamineTM 3000 (Thermofisher) accordingly manufacturer instructions. After 24–36 hours, mCherry-expressing cells were selected for patch clamp analysis. Electrophysiology and optogenetic stimulation Patch clamp experiments in the whole-cell configuration were carried out 24–36 hours post- transfection on mCherry-expressing cells at room temperature. HEK cells were superfused with an extracellular-like solution containing (mM): NaCl 140, KCl 5.4, CaCl2 1.8, MgCl2 1, Hepes-NaOH 10, Glucose 5.5, pH = 7.4. The pipette (7–9 MO) were filled with an intracellu- lar-like solution containing (mM): K-Asp 130, NaCl 10, EGTA-KOH 5, MgCl2 2, CaCl2 2, ATP (Na2-salt) 2, creatine phosphate 5, GTP 0.1, Hepes-KOH 10; pH 7.2. Optogenetic stimula- tion was delivered by the OptoPatcher system using LSD-1 light stimulation device (ALA Sci- entific Instruments) as previously described [13]. Data acquisition and light triggering were controlled with pCLAMP software via DigiData 1440A interfaces (Molecular Devices). The channelrodopsin (ChR2) photocurrent was measured under voltage-clamp conditions from a holding potential of 0 mV applying concomitantly hyperpolarizing test steps in the range 0/-90 mV and high-intensity illumination for 2600 ms. Peak and stationary currents were normal- ized by cell capacitance. Patch-clamp currents were acquired with a sampling rate of 4 kHz without lowpass filter. Neither series resistance compensation nor leak or liquid junctional PLOS ONE | https://doi.org/10.1371/journal.pone.0227230 January 17, 2020 2 / 14 Habituation in single non-excitable cells potential corrections were applied. The light stimulation was delivered for 20 s as pulse train (or cosine wave) in I/0 configuration at three different frequencies (in Hz: 0.5; 1; 2) and three intensities (Low: 1V; Middle: 2V; High: 5V. Voltage values referrers to the LSD-1 light stimula- tion device). The mono-exponential decay fitting was used to calculate the percentage of depo- larization at the steady state and the tau of habituation (τH), defined as the number of events/ time necessary to reach the 37% of the percentage of depolarization at the steady state. The probability of habituation (p(H)) was defined as 1 if the cell response fitted or 0 if the cell response did not fit with a mono-exponential fitting. Statistical analysis Data were analyzed with Clampfit10 (Axon) and Origin Pro 9. To test the impact of the stimu- lation features on the habituation profile, we compared the mean percentage of depolarization at the steady state and the mean tau of habituation (τH) at different conditions. These two parameters are sufficient to uniquely describe a mono-exponential profile. Data were com- pared using either One-Way ANOVA followed by Fisher’s LSD post-hoc test or Student’s T- test; significance level was set to p = 0.05. Data outliers were excluded using Tukey’s method. Data were collected from different transfection experiments ranging from a minimum of four to a maximum of twelve. Results Optogenetically-induced depolarizations are reduced by repetitive stimulation To explore the habituation process in single aneural cells, human embryonic kidney (HEK) cells were transfected with a Channelrodopsin2 (ChR2)-expressing plasmid and the functional expression of the photocurrent was assessed in mCherry-positive cells (S1 Fig). Subsequently, ChR2-expressing cells were photostimulated (pulse train) and the membrane potential (Vmem) was simultaneously recorded using a patch clamp approach in the whole cell configuration. A representative stretch of the Vmem profile during 1Hz/5V light stimulation is shown in Fig 1A, in which the depolarization induced by the photocurrent (hν, blue lines) is visibly reduced over time. A similar reduction is also observed when the stimulation was given as cosine waves rather the pulse train (S2 Fig) suggesting the independence of the cell’s response from the shape of the delivered stimulation. In the absence of the ChR2 channel expression, the light stimulation did not induce any change in the Vmem (S3 Fig). The decremental reduction of the depolarization is summarized in Fig 1B. All data points were normalized by the magnitude of depolarization of the first event, obtaining the percentage of depolarization (y-axis, Fig 1B); data were plotted against either the number of events or time. For each profile, the percentage of depolarization at the steady state and tau of habituation (% of dep. at s.s. and τH, respec- tively) are computed using a monoexponential decay fitting and used to define the magnitude (% of dep. at s.s.) and kinetic (τH) characteristics of habituation. By definition, τH is the num- ber of events or time necessary to reach 37% of the amplitude value (Fig 1B). The observed asymptotical response reduction during repetitive stimulation is a key feature necessary to define any habituation profile. The frequency and intensity of stimulation affects both magnitude and kinetic of habituation The frequency and intensity characteristics of the stimulation are well-known modulators of the habituation. First, we explored the impact of the frequency of stimulation on the PLOS ONE | https://doi.org/10.1371/journal.pone.0227230 January 17, 2020 3 / 14 Habituation in single non-excitable cells Fig 1. Definition of the habituation profile. A) Representative trace of voltage recorded in the I/0 configuration during a light stimulation at 465 nm (blue lines). B) Normalized values of depolarization during 20 s of 1Hz/5V stimulation protocol. Monoexponential fitting curve of the plotted data (circle) is shown in red. Percentage of depolarization at the steady state (% of dep. at s.s.) and tau of habituation (τH) are indicated. https://doi.org/10.1371/journal.pone.0227230.g001 habituation profile. In Fig 2 (top panel), HEK cells were stimulated at 5V for 20 s at three different frequencies as indicated (top panel, in Hz: 0.5, 1, 2; black square, purple circle and green triangle, respectively). The resulting mean traces are shown either superimposed (Fig 2A) or divided (Fig 2B) plotting the number of events on the x-axis. Mean τH and % of dep. at s.s. values are summarized in Fig 2C and 2D in a frequency-dependent fashion. When we considered the number of events, other things being equal, higher stimulation frequencies were associated with a slower kinetic (Fig 2C; p<0.05 among groups) and more pronounced amplitude (Fig 2D and 2H; p<0.05). On the other hand, when we considered time rather than events as displayed on the x-axis (S4 Fig), higher stimulation frequency was associated with a faster kinetic (S4 Fig). From these results, the frequency of stimulation clearly affects both the kinetic and magnitude of the habituation profile indicating a frequency-dependent response. We also explored the impact of different intensities of stimulation on the habituation profile (bottom panel). HEK cells were stimulated at 1 Hz for 20 s at three different intensities: Low: 1V; Medium:2V; High:5V (bottom panel: black square, purple circle and green triangle, respectively). The resulting mean traces were shown superimposed (Fig 2E) or separated (Fig 2F) plotting the number of events on the x-axis; mean τH and % of dep. at s.s. values are sum- marized in Fig 2G and 2H in an intensity-dependent fashion. Other factors being equal, at 1V the kinetic is significantly slower when compared to both 2V and 5V stimulations (Fig 2G). Moreover, at 1V the magnitude of habituation is less pronounced (p<0.05) than both 2V and 5V conditions (Fig 2H). Taken together, these results highlight both frequency- and intensity- dependent behavior of the cellular system. The recovery profile is frequency-dependent A hallmark of habituation is the reversibility of the decremental response. We thus explored the recovery profile from the steady state condition (filled symbols, Fig 3) increasing the recov- ery time between consecutive series of stimulations. We evaluated the recovery profile in a PLOS ONE | https://doi.org/10.1371/journal.pone.0227230 January 17, 2020 4 / 14 Habituation in single non-excitable cells Fig 2. The impact of the stimulation features on the habituation profile. HEK cells were stimulated at 5V at three different frequencies as indicated (in Hz: 0.5, black square; 1, purple circle; 2 green triangle). A) Superimposed (solid line is the mean and colored area the S.E.M.) and B) separated mean profiles are shown plotting the number of events. C) Mean τH (in events: 0.5Hz: 2.66±1.00, n = 21; 1Hz: 3.43±0.12, n = 43; 2Hz: 4.70±0.16, n = 43) and D) mean % of dep. at s.s. (0.5Hz: 19.87±1.00, n = 21; 1Hz: 23.00±0.23, n = 43; 2Hz: 29.84±0.19, n = 43) are shown. HEK cells were also stimulated at 1Hz at three different intensities as indicated (Low: 1V black square; Medium: 2V purple circle; High: 5V green triangle). E) Superimposed and F) separated mean profiles are shown, plotting the number of events. G) Mean τH (in events: Low: 10.53±3.69, n = 12; Medium: 3.74±0.18, n = 9; High: 3.43±0.12, n = 43) and H) mean % of dep. at s.s. (Low: 16.00±2.96, n = 12; Medium: 21.11±0.32, n = 9; High: 23.00±0.23, n = 43) are shown in the event-domain. One-way Anova, �p<0.05 vs 0.5Hz or 1V; #p<0.05 vs 1Hz. https://doi.org/10.1371/journal.pone.0227230.g002 frequency-dependent manner. After reaching the steady state of the habituation profile, we normalized the following stimulation profile based on the first event of the first stimulation (filled symbol) and reported on the graph the mean % of depolarization after increasing recov- ery times (unfilled symbols). In Fig 3A, the mean recovery profiles are shown for 0.5, 1 and 2Hz (square, circle and triangle, respectively). It is clear that the time necessary to reach again the 100% of the response is frequency-independent (26.7 s). On the other hand, the recovery trajectory appeared to be conserved at 1Hz and 2Hz and different at 0.5 Hz, suggesting poten- tial different frequency-dependent mechanisms. We then analyzed both τH (Fig 3B) and the % of dep. at s.s. (Fig 3C) of the profiles during the consecutive series of stimulation; the x-axis indicates the resting period between consecutive stimulations and the dotted line represent the value of the descriptor during the first stimulation (filled symbols). Both descriptors displayed a frequency signature; it is also interesting to notice that at 3.5 s and 4.2 s (1Hz stimulation) the kinetic is slower. We also reported the probability to generate a habituation profile (p(H)) (Fig 3D); we found that in all conditions, when examining cases where recovery time is below 2.3s, the probability to generate the habituation profile is null. Taken together, the results indi- cate that the decremental response was reversible and that, based on the recovery time, the kinetic and magnitude of the profiles have complex frequency-dependent behavior. Moreover, the probability to generate a habituation profile during consecutive stimulations is not an assumption that can be made a priori. PLOS ONE | https://doi.org/10.1371/journal.pone.0227230 January 17, 2020 5 / 14 Habituation in single non-excitable cells Fig 3. Frequency-dependent recovery profile. HEK cells were stimulated at 5V at three different frequencies (in Hz: 0.5, square, top; 1, circle, middle; 2 triangle, bottom) for 20s and, after a recovery time, the same frequency protocol was applied. A) Mean profiles and normalized values of the first event (filled symbols) after different resting periods (unfilled symbols). B) Mean τH (in events) and C) % of dep. at s.s. of the profiles at different recovery times (dot lines indicate the values of the initial profile). D) Mean bar graphs indicating the probability of habituation profile (p(H)) at different recovery times. Mean values are reported in S1 Table. https://doi.org/10.1371/journal.pone.0227230.g003 Frequency transitions influence the kinetics of the habituation profile We then explored what would happen to the cell’s output if the photostimulation suddenly changed frequency without an intervening rest period. Our aim was to simulate the rhythmic transition changes that could occur in quasi-periodic biological systems. The mean profile dur- ing the 1Hz-2Hz-1Hz transition is shown in Fig 4A (1Hz purple; 2Hz green). The mean τH and % of dep. at s.s. values are summarized in Fig 4B and 4C, respectively. Both the kinetic profile and magnitude at 2Hz are not affected by the previous 1Hz stimulation; indeed, the val- ues are not different from the 2Hz stimulation alone (Fig 2). However, after the 2Hz stimula- tion, the 1Hz profile is faster whereas the magnitude is invariant with respect to the 1Hz condition alone (Fig 2). Moreover, after the first stimulation, the change of frequency reduces the probability of generating a habituation profile to 50% (Fig 4D). The mean profile during the 2Hz-1Hz-2Hz transition is shown in Fig 4E. The first 2Hz stimulation influences the 1Hz kinetic profile during the 2Hz-1Hz transition as shown in Fig 4F; particularly, the τH is signifi- cantly slower compared to the 1Hz stimulation alone but, again, reached the same magnitude with a p(H) of about 60% (Fig 4H). The following 1Hz-2Hz transition did not produce any habituation profile (Fig 4H). Collectively, these results indicated that the frequency transitions without resting periods in between affect the kinetic profile but did not affect the magnitude of the habituation. Native channels participate in the habituation response Since habituation and desensitization share the same decremental response over time, we ana- lyzed the ChR2 photocurrent profile upon stimulation to address any channel-related PLOS ONE | https://doi.org/10.1371/journal.pone.0227230 January 17, 2020 6 / 14 Habituation in single non-excitable cells Fig 4. Intra-protocol frequency transitions influence the habituation profile. HEK cells were stimulated at 5V at either 1Hz (purple) or 2Hz (green) without a resting period in between. A) Mean profiles at 1Hz-2Hz-1Hz transition (solid line is the mean and colored area the S.E.M.). B) Mean τH (in events: Alone: 1Hz: 3.43±0.12, n = 43; 2Hz: 4.70±0.16, n = 43. Transitions: First 1Hz: 3.52±0.71; 2Hz: 6.17±1.02; Second 1Hz: 1.55±0.62, n = 12), C) mean % of dep. at s.s. (Alone: 1Hz: 23.00±0.23, n = 43; 2Hz: 29.84±0.19, n = 43. Transitions: First 1Hz: 26.16±1.33; 2Hz: 33.56±3.66; Second 1Hz: 20.28±2.61, n = 12) and D) mean bar graphs indicating the probability of habituation profile (p(H): First 1Hz: 100±0; 2Hz: 55.56±17.57; Second 1Hz: 57.14±20.20, n = 8) are shown. E) Mean profiles at 2Hz-1Hz-2Hz transition (solid line is the mean and colored area the S.E.M.). F) Mean τH (in events: Alone: 1Hz: 3.43±0.12, n = 43; 2Hz: 4.70±0.16, n = 43. Transitions: First 2Hz: 5.24±0.60; 1Hz: 17.59±7.0, n = 12) and G) mean % of dep. at s.s. (Alone: 1Hz: 23.00±0.23, n = 43; 2Hz: 29.84±0.19, n = 43. Transitions: First 2Hz: 29.73±1.79; 1Hz: 23.30±1.45, n = 12) and H) mean bar graphs indicating the probability of habituation profile (p(H): First 1Hz: 100±0; 2Hz: 66.67±21.08; Second 1Hz: 0, n = 12) are shown. Student’s T-test �p<0.05 vs Alone condition. https://doi.org/10.1371/journal.pone.0227230.g004 desensitization effect. In Fig 5A, representative traces of the photocurrent at -30, -40 and -50 mV (square, circle and triangle, respectively) are shown during the application of the 1Hz,5V stimulation protocol for 10 seconds (blue lines); we chose three voltage values near the mean value of the resting potential of HEK cells (-40.75±1.38 mV; n = 58). The steady current was then analyzed in an event- and voltage-dependent manner. The graph in Fig 5B shows the mean density current values of the photocurrent during the applied stimulations. No signifi- cant decrement of the density current appeared during repetitive stimulation. In order to address any active cell-autonomous processes, we explored the impact of native potassium channels in the habituation process; we thus blocked them using 10 μM of TEA, as previously reported[14]. After confirming that TEA does not influence the photocurrent (Fig 5B), we ana- lyzed the effect of the drug on the habituation profile at 1Hz, 5V. The mean profile is shown in Fig 5D and mean τH and % of dep. at s.s. values are summarized in Fig 5E and 5F indicating a significantly slower and more pronounced profile in the presence of TEA. This result high- lights that the TEA-sensitive native potassium channels actively participate in defining the photocurrent-induced habituation process. PLOS ONE | https://doi.org/10.1371/journal.pone.0227230 January 17, 2020 7 / 14 Habituation in single non-excitable cells Fig 5. ChR2-independent and ion-dependent habituation profile. A) Representative traces of the photocurrent at -30, -40 and -50 mV (square, circle and triangle, respectively) during a 1Hz,5V repetitive stimulation. B) Mean current density/event plot of the photocurrent. C) Mean photocurrent density currents with or without TEA (filled circle, empty square, respectively; n = 8 each). D) Mean habituation profiles with TEA 10 μM in the extracellular solution. E) Mean τH (in events: CTRL: 3.43 ±0.27, n = 43; TEA: 4.27±0.30, n = 18) and F) amplitude (CTRL: 23.38±0.89, n = 43; TEA: 28.03±1.63, n = 18). Student’s T-test �p<0.05 vs CTRL. https://doi.org/10.1371/journal.pone.0227230.g005 Mathematical modeling of habituation in HEK cells We recently proposed a generalization of the habituation process which could be applied inde- pendently of the biological details of the given system [6]. As outlined in the paper, the habitu- ation process was described as the dynamic interplay between different elements, namely the stimulation, transducer, habituation, receiver and background elements. Each element is described by a variable and, overall, the process is described by the following equation: Rn ¼ T0 n þ H0 ðnsÞ0 � s i¼0 Di þ B Pn(cid:0) 1 ð1Þ where Rn is the output of the receiver element (the element that we monitor during the stimu- lation), T0 n is the output of the transducer elements (influenced by the frequency (t(s)) and the intensity of stimulation and the nature of the modules composing the element itself), H0 (ns)0 is an index of the initial state of the habituation element and thus the output of the habituation element before the stimulation, sigma (σ) is the stimulation factor, delta (Δ) is the spontaneous decay factor during the recovery phase from the stimulation, B is the output of the background elements (stimulation invariant elements) and n is the number of events delivered to the sys- tem. Through a mathematical manipulation of the Eq 1 (S1 File), we computed from the raw data Δ, σ and A (where A = T0 (ns)0+B) associated with some conditions tested throughout the paper. Each parameter, as detailed in the S1 File, is influenced either by the stimulation fea- tures (t(ns), t(s) and intensity) or by the nature/composition of the habituation system (T’, B n+H0 PLOS ONE | https://doi.org/10.1371/journal.pone.0227230 January 17, 2020 8 / 14 Habituation in single non-excitable cells Table 1. Relationship between the different combinations of parameters and the variables. In the table are indi- cated the variables when more than one parameter is different among conditions. B is the output of the background element, H(ns)0 is the output of the habituation element before the stimulation, T’ is the output of the transducer ele- ments, int is the intensity of the stimulation, t(s) is the time of stimulation, t(ns) is the time of non-stimulation between two events and H’ is the output of the habituation element. AND σ AND A Δ AND A Δ AND σ Δ All σ All A All https://doi.org/10.1371/journal.pone.0227230.t001 �Δ B, H(ns)0, T’, int, t(s) �σ B, H(ns)0, t(ns) �A H’, t(ns) and H(ns)0). The detailed relationships between the parameters (Δ, σ and A), the variables (B, H(ns)0, T’, H’, t(ns)) are reported in the S1 File. In Table 1 we summarize the variables that can influence all the possible combinations of the significant (i.e. Δ) and not significant (i.e. �D) parameters. In Table 2 we report the signifi- cant parameters among the indicated conditions, and the variables that can be neglected are also listed (S1 File); with this eliminative procedure, we then obtained the significant variables that can explain the observed parameter combinations (for numerical details, see S1 Table). It emerges that the differences among 1Hz and 2Hz stimulations (Fig 2A) arose just from the dif- ferent stimulation protocol (t(ns)), whereas during the 0.5 Hz condition the differences must also be related to a different nature/composition of the habituation system (T’, H’). When we compare 2V vs 5V (Fig 2B), we can see that a different response of T’ is the explanation of the different output (in particular, reflecting the different intensities of stimulation). Upon TEA application at 1Hz 5V stimulation (Fig 5D), we can conclude that native K+-channels partici- pate either in the composition of the translator (T’) or habituation element (H’ and/or H(ns)0). Finally, during the frequency transitions, the first 1Hz stimulation and the second 1Hz stimu- lation after the 2Hz stimulation (Fig 4A) differs because of either a difference in the pre-stimu- lation habituation elements (H(ns)0) or a difference in the nature of the translator element (T’). In conclusion, the previously proposed model could be instrumental in narrowing the biologi- cal processes involved in the different responses through an experimentally-driven eliminative procedure. Limitations In the present work, two main limitations are present: the non-physiological source of stimula- tion (the photostimulation of the ChR2) and the use of just one cellular type. Indeed, the over- expression of the ChR2 channels is an implausible physiological situation driven by the experimental need to fine-tune the stimulation features, which practically limited the use of Table 2. Experimental-driven eliminative procedure. After the computation of Δ, σ and A in each group, we identified the statistically significant parameters and using Table 1 we highlight the significant variables. Moreover, based on each specific group comparison, we could also identify the variables which are invariant based on the applied stimulation. Experimental feature Figure Group Comparison Statistically significant parameters Neglectable Variables Significant Variables Frequency Intensity Native Channels Frequency transitions Fig 2A Fig 2A Fig 2A Fig 2E Fig 5D Fig 4A 0.5 vs 1 Hz 0.5 vs 2 Hz 1 vs 2 Hz 2 vs 5 V (-)TEA vs (+)TEA First vs Second 1Hz Δ AND A AND σ Δ AND A AND σ Δ AND A A AND σ Δ AND A AND σ A AND σ https://doi.org/10.1371/journal.pone.0227230.t002 B, mag, H(ns)0, t(s) B, mag, H(ns)0 B, mag, H(ns)0 B, H(ns)0, t(s), t(ns) B, mag, t(s), t(ns) B, mag, t(s), t(ns) H’, T’, t(ns) H’, T’, t(ns), t(s) t(ns) T’, mag T’, H’, H(ns)0 T’, H(ns)0 PLOS ONE | https://doi.org/10.1371/journal.pone.0227230 January 17, 2020 9 / 14 Habituation in single non-excitable cells more biologically relevant stimulation sources. On the other hand, the ionic currents gener- ated by the opening of the channels (from which the depolarization arose) is a universal lan- guage for cells. Nonetheless, it is important to mention that the use of a single type of channelrodopsin prevents us to conclude which ionic current-dependent phenomena (namely the depolarization of the membrane or any other ion-dependent mechanisms) is responsible for the habituation. Moreover, we explored the process only in HEK cells (since it is a well- established heterologous system in electrophysiology); this limits any robust claim of generali- zation of the presented results to other non-neuronal system. Finally, we only explored the non-associative aspect of the habituation, namely using one and only one form of stimulation and, because of the intrinsic instability of the whole cell configuration over long recording periods (more than an hour), we did not explore any potential long-term effects of the stimula- tion. In light of these limitations, the present work should be seen as a proof of concept of the ability of non-neuronal cells to habituate rather than an indication for habituation as a biologi- cally universal process with defined features and rules; more data must be collected to prove this claim. Discussion Whether they are self-generated by the body (i.e. heartbeat, brain waves, circadian rhythms, hormone release, etc.) or delivered from environmental sources (new drug treatment, training, routine behaviors, etc.), repetitive stimulations are ubiquitous and essential to the adaptive behavior and physiology of living organisms. A common behavioral strategy to deal with repetitive stimulations is to reversibly reduce the output of the system; a process which is termed habituation [2]. Over the last 50 years, an extensive characterization of the behavioral manifestation of habituation has been performed [1] mostly confirming the characteristics previously identified [2]. So far, the list of features reported by Rankin and colleagues [1] rep- resents the most up-to-date guideline to correctly classify behavioral habituation. Habituation is considered within an exclusively neural-based framework even though some experiments demonstrate the process clearly emerges within aneural systems [7] [8] [9] [10]. While data continue to accumulate to broaden our view of the gradual evolution of learning capacities from basal taxa, it is essential to develop platforms that facilitate the study of universal cellular mechanisms for computation and optimization of behavior. While single-cell habituation is apparently robust, a deeper characterization has not yet been achieved. A proper comparison between the cellular and behavioral manifestations of habituation could reveal a more general process that is not restricted to neuronal substrates. In the present work, we explored the habituation process in ChR2-expressing HEK cells. The main advantage of using the ChR2 is to uniquely stimulate a singular element of the cell (indeed, the blue light stimulation did not affect the resting membrane potential of the cells, the output that we monitored throughout the study). The impact of ChR2-mediated depolari- zation on the voltage profile of the cells was studied, defining three descriptors: percentage of depolarization at the steady state (% of dep. at s.s.) and τH to describe the magnitude and kinetic of habituation, respectively, and p(H), the probability to generate an exponential-like profile. From Fig 1A, the repetitive series of stimulations decreased the amplitude of the photo- current-induced depolarization within the protocol with an asymptotic profile (Fig 1B). It is also important to notice that the photocurrent amplitude was invariant throughout the stimu- lation (Fig 5B), demonstrating that the decrement was ChR2 independent. In support of this hypothesis, the blockage of native potassium channels with TEA changed the profile’s features (Fig 5D) indicating that the cell was actively responding to the repetitive stimulation; it is also relevant to mention that TEA dosage did not influence the photocurrent characteristics (Fig PLOS ONE | https://doi.org/10.1371/journal.pone.0227230 January 17, 2020 10 / 14 Habituation in single non-excitable cells 5C). Finally, the recovery of the output after a resting period (Fig 3) led us to exclude any dele- terious effects of the stimulation on the cell output. Taken together, these results point toward a robust indication of habituation in the analyzed cell system. As previously described from a behavioral standpoint [1], the stimulation characteristics must affect the response. We thus tested the impact of different frequencies of stimulation (Fig 2) finding that increasing the stimulation frequency produced a more pronounced profile (Fig 2D). Plotting the time on the x-axis, higher stimulation frequencies were associated with faster profiles (S4 Fig), which is in line with the behavioral data; we observed the opposite effect when plotting the number of events (Fig 2C). This apparent contradiction highlights the neces- sity to always clarify if the analysis of the kinetic is made with respect to either time or events. We then manipulated the intensity of the stimulation and found a less pronounced and slower profile at 1V (p<0.05) and no differences between 2V and 5V. Taking into consideration the limited range of intensities that we explored, our results are clearly in opposition with the behavioral data. So far, we discussed the response of the cell system to a novel stimulation; in Fig 3, however, we explored the profile after consecutive stimulations. We found that the kinetic profile was not necessarily faster after consecutive stimulations, as it was framed for the behavioral habituation; a similar contradiction between behavioral and cellular data can also be highlighted when considering the magnitude. Most importantly, it emerged that habitua- tion cannot be considered granted without satisfying certain temporal criteria; indeed, below a recovery period of 2.3 s, it seems that the cells cannot generate any habituation profile (Fig 3D). An absolute habituation refractory period emerged below which the habituation itself could not occur; in other words, the habituation elements in the system are not responsive during the absolute habituation refractory period. Moreover, in Fig 4 we explored systematic changes in rhythmicity without deliberate recov- ery. This protocol was designed to mimic physiological changes in the frequency of biological periodic stimulation: actually, considering stimulations that arise inside the body, it is more common that the system experiences a modification in the rhythmic event rather than a new type of stimulation. It appears that the kinetic, but not the magnitude, was affected by the sequence of the frequency transitions. It follows that the magnitude of habituation can be con- sidered the only invariant frequency-dependent signature during the frequency transitions. Most importantly perhaps, it highlights that the same stimulation (1Hz) can lead to either a habituation or sensitization profile based on the pre-1Hz stimulation state (Novel vs 2Hz vs 1Hz-2Hz). The evidence that habituation and sensitization arise from the same protocol of stimulation suggests that the state of the system before the stimulation is a crucial factor, more so than the features of the stimulation itself in defining the ultimate phenotype. In particular, we can speculate that a habituation profile emerges if the % of dep. at s.s. of the previous state is smaller than the one associated with the frequency of the second stimulation; on the other hand, if it is greater, a sensitization profile emerges. It also leads to the speculation that habitu- ation and sensitization are two facets of the same process. In other words, the system seems to achieve a defined frequency-dependent steady state using either habituation or sensitization phenomena accordingly to the previous state of the system. The determinant of whether one emerges over the other would be the pre-stimulation state of the system; however, any robust conclusion cannot be irrefutable considering the limited data presented here. Most impor- tantly, perhaps, this establishes the experimental basis to explore the effect, if any, of patho- physiological changes of rhythmic processes generated by excitable cells (i.e. cardiomyocytes, neurons) on non-excitable cells (i.e. endothelial cells, fibroblasts, macrophages, microglia). Even if we confirmed the habituation process in HEK cells, those results reveal little about any mechanistic explanation. PLOS ONE | https://doi.org/10.1371/journal.pone.0227230 January 17, 2020 11 / 14 Habituation in single non-excitable cells Using a mathematical generalization of the habituation process [6], we narrowed some potential mechanisms of the habituation in the present system. In particular, we can see that any difference between 1Hz and 2Hz is due to just the different frequency but not because of recruitment/dismissal of elements in the HEK system: in other words, the system that reacts to the stimulation is, in both activity and composition, identical. A similar picture arises when we compared 2V vs 5V. On the contrary, when we analyzed the 0.5Hz vs 1Hz (or 2Hz) stimula- tion, we realized that the differences were not only because of the different stimulation proto- cols, but also because of a different activity/composition of the HEK system reacting at those frequencies. In other words, different frequencies are processed differently by the system because of a change in its state. This hypothesis also seems to be reflected in the different pro- file of the recovery in Fig 3. Taken together, our data show that both the behavioral and our cellular model share a dec- remental decrease during repetitive stimulation that, after a resting period, is reversible. More- over, they both showed a frequency and intensity dependence of the habituation profile; however, it is critical to report that the similar changes in the stimulation features do not nec- essarily lead to the same habituation profile changes in the behavioral vs cellular comparison. The authors suggest that this is due to the fact that the specific response to stimulation changes are not amenable to generalization. Namely, the responses lie on the peculiar composition of the system that we are monitoring and must be tested de novo for any new system. To summa- rize, the behavioral and cellular habituation processes shares 1) an asymptotical decrement of the output during repetitive stimulation, 2) the reversibility of the profile after a resting period and 3) a dependence on both frequency and intensity of stimulation. Based on these findings, we propose to consider and define habituation as a time- and state-dependent process which could occur if and only if 1) the time between two consecutive stimulations is smaller than the time necessary to the system to achieve a pre-stimulation state and larger than the absolute habituation refractory period, 2) satisfy the three points above-mentioned. Future experiments using many more cell substrates will test the solidity of our definition and clarify any claim as to the universality of the habituation process. Conclusions Bearing in mind the aforementioned limitations, the present work: 1) demonstrates that non- neuronal cells can habituate in a stimulation-dependent manner, 2) highlights similarity and discrepancies between the behavioral rules and our model responses, 3) gives defined descrip- tors to analyze the process (% of effect at s.s., τH and probability of habituation), 4) shows that systems respond differently in case of preceding history of stimulation and 5) guides the explo- ration of mechanistic information using an experimental-driven shortcut approach based on a mathematical generalization of the habituation process. Supporting information S1 File. Description of the mathematical model. (DOCX) S1 Fig. Photocurrent current density-voltage plot. A) Representative photocurrent density traces (holding potential: 0 mV) recorded in the range 0/-90 mV (ΔV = 10 mV). B) Current density-voltage plot analyzed at the peak (square) or steady state (circle). n = 14. (TIF) PLOS ONE | https://doi.org/10.1371/journal.pone.0227230 January 17, 2020 12 / 14 Habituation in single non-excitable cells S2 Fig. Cosine wave-induced habituation profile. A) Representative voltage trace upon the application of B) a 1Hz,5V cosine wave light stimulation. (TIF) S3 Fig. Non-transfected HEK cell does not respond to light. Representative voltage profile of non-transfected HEK cells (in black) in response to the light stimulation protocol (in blue). (TIF) S4 Fig. The stimulation’s features impact the habituation profile. HEK cells were stimulated at 5V at three different frequencies as indicated (in Hz: 0.5, black square; 1, purple circle; 2 green triangle). A) Superimposed and B) separated mean profiles are shown plotting the time pf stimulation. C) Mean τH (in s: 0.5Hz: 6.11±0.81, n = 21; 1Hz: 3.43±0.12, n = 43; 2Hz: 2.32 ±0.11, n = 43) and D) mean amplitude (in % of depolarization: 0.5Hz: 19.78±1.00, n = 21; 1Hz: 23.00±0.23, n = 43; 2Hz: 29.84±0.19, n = 43) are shown. One-way Anova �p<0.05 vs 0.5Hz; #p<0.05 vs 1Hz. (TIF) S1 Table. Fig 3 parameters details. (TIF) Author Contributions Conceptualization: Mattia Bonzanni, Nicolas Rouleau. Data curation: Mattia Bonzanni. Formal analysis: Mattia Bonzanni. Funding acquisition: Michael Levin, David L. Kaplan. Investigation: Mattia Bonzanni. Methodology: Mattia Bonzanni. Project administration: Mattia Bonzanni. Resources: Michael Levin, David L. Kaplan. Supervision: Mattia Bonzanni. Validation: Mattia Bonzanni. Visualization: Mattia Bonzanni. Writing – original draft: Mattia Bonzanni, Nicolas Rouleau. Writing – review & editing: Mattia Bonzanni, Nicolas Rouleau, Michael Levin, David L. Kaplan. References 1. Rankin CH, Abrams T, Barry RJ, Bhatnagar S, Clayton D, Colombo J et al. (2009) Habituation revisited: an updated and revised description of the behavioral characteristics of habituation. Neurobiol Learn Mem 92(2):135–138. https://doi.org/10.1016/j.nlm.2008.09.012 PMID: 18854219 2. Thompson RF & Spencer WA (1966) Habituation: a model phenomenon for the study of neuronal sub- strates of behavior. Psychol Rev 73(1):16–43. https://doi.org/10.1037/h0022681 PMID: 5324565 3. Groves PM & Thompson RF (1970) Habituation: a dual-process theory. Psychol Rev 77(5):419–450. https://doi.org/10.1037/h0029810 PMID: 4319167 4. Sokolov EN (1963) Higher nervous functions; the orienting reflex. Annu Rev Physiol 25:545–580. https://doi.org/10.1146/annurev.ph.25.030163.002553 PMID: 13977960 PLOS ONE | https://doi.org/10.1371/journal.pone.0227230 January 17, 2020 13 / 14 Habituation in single non-excitable cells 5. Ramaswami M (2014) Network plasticity in adaptive filtering and behavioral habituation. Neuron 82 (6):1216–1229. https://doi.org/10.1016/j.neuron.2014.04.035 PMID: 24945768 6. Bonzanni M, Rouleau N, Levin M, & Kaplan DL (2019) On the Generalization of Habituation: How Dis- crete Biological Systems Respond to Repetitive Stimuli: A Novel Model of Habituation That Is Indepen- dent of Any Biological System. Bioessays 41(7):e1900028. https://doi.org/10.1002/bies.201900028 PMID: 31222777 7. Boisseau RP, Vogel D, & Dussutour A (2016) Habituation in non-neural organisms: evidence from slime moulds. Proc. R. Soc. B 283(1829):20160446. https://doi.org/10.1098/rspb.2016.0446 PMID: 27122563 8. Eisenstein E, Brunder D, & Blair H (1982) Habituation and sensitization in an aneural cell: Some com- parative and theoretical considerations. Neuroscience & Biobehavioral Reviews 6(2):183–194. 9. Meins F Jr & Lutz J (1979) Tissue-specific variation in the cytokinin habituation of cultured tobacco cells. Differentiation 15(1–3):1–6. 10. Pischke MS, Huttlin EL, Hegeman AD, & Sussman MR (2006) A transcriptome-based characterization of habituation in plant tissue culture. Plant Physiology 140(4):1255–1278. https://doi.org/10.1104/pp. 105.076059 PMID: 16489130 11. Lyon P (2006) The biogenic approach to cognition. Cogn Process 7(1):11–29. https://doi.org/10.1007/ s10339-005-0016-8 PMID: 16628463 12. Baluska F & Levin M (2016) On Having No Head: Cognition throughout Biological Systems. Front Psy- chol 7:902. https://doi.org/10.3389/fpsyg.2016.00902 PMID: 27445884 13. Katz Y, Yizhar O, Staiger J, & Lampl I (2013) Optopatcher—an electrode holder for simultaneous intra- cellular patch-clamp recording and optical manipulation. J Neurosci Methods 214(1):113–117. https:// doi.org/10.1016/j.jneumeth.2013.01.017 PMID: 23370312 14. Ponce A, Castillo A, Hinojosa L, Martinez-Rendon J, & Cereijido M (2018) The expression of endoge- nous voltage-gated potassium channels in HEK293 cells is affected by culture conditions. Physiol Rep 6(8):e13663. https://doi.org/10.14814/phy2.13663 PMID: 29665277 PLOS ONE | https://doi.org/10.1371/journal.pone.0227230 January 17, 2020 14 / 14
null
10.1088_2634-4386_ad046d.pdf
Data availability statement The data that support the findings of this study are openly available (Kaiser et al 2023). The experiment code is available on https://github.com/electronicvisions/model-paper-mc-sbi.
Data availability statement The data that support the findings of this study are openly available (Kaiser et al 2023) . The experiment code is available on https://github.com/electronicvisions/model-paper-mc-sbi .
OPEN ACCESS RECEIVED 28 April 2023 REVISED 5 October 2023 ACCEPTED FOR PUBLICATION 18 October 2023 PUBLISHED 7 November 2023 Original Content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. Neuromorph. Comput. Eng. 3 (2023) 044006 https://doi.org/10.1088/2634-4386/ad046d PAPER Simulation-based inference for model parameterization on analog neuromorphic hardware Jakob Kaiser1,∗, Raphael Stock1, Eric Müller1, Johannes Schemmel1,∗ and Sebastian Schmitt2 1 Kirchhoff-Institute for Physics (European Institute for Neuromorphic Computing), Heidelberg University, Heidelberg, Germany 2 Department for Neuro- and Sensory Physiology, University Medical Center Göttingen, Göttingen, Germany ∗ Authors to whom any correspondence should be addressed. E-mail: jakob.kaiser@kip.uni-heidelberg.de and schemmel@kip.uni-heidelberg.de Keywords: analog, neuromorphic, simulation-based inference, multi-compartment Abstract The BrainScaleS-2 (BSS-2) system implements physical models of neurons as well as synapses and aims for an energy-efficient and fast emulation of biological neurons. When replicating neuroscientific experiments on BSS-2, a major challenge is finding suitable model parameters. This study investigates the suitability of the sequential neural posterior estimation (SNPE) algorithm for parameterizing a multi-compartmental neuron model emulated on the BSS-2 analog neuromorphic system. The SNPE algorithm belongs to the class of simulation-based inference methods and estimates the posterior distribution of the model parameters; access to the posterior allows quantifying the confidence in parameter estimations and unveiling correlation between model parameters. For our multi-compartmental model, we show that the approximated posterior agrees with experimental observations and that the identified correlation between parameters fits theoretical expectations. Furthermore, as already shown for software simulations, the algorithm can deal with high-dimensional observations and parameter spaces when the data is generated by emulations on BSS-2. These results suggest that the SNPE algorithm is a promising approach for automating the parameterization and the analyzation of complex models, especially when dealing with characteristic properties of analog neuromorphic substrates, such as trial-to-trial variations or limited parameter ranges. 1. Introduction Mechanistic models, which try to explain the causality between inputs and outputs, are integral to scientific research. On the one hand they can increase the understanding of the mechanisms which cause the phenomena and on the other make predictions about new outcomes which can then be tested experimentally (Baker et al 2018). After a mechanistic model has been formulated, one of the remaining challenges is to find suitable model parameters which lead to a close agreement between model behavior and experimental observations. Several approaches such as the hand-tuning of parameters, grid searches, random/stochastic searches, evolutionary algorithms, simulated annealing and particle swarm algorithms have been used in neuroscience to find appropriate model parameters (Vanier and Bower 1999, Van Geit et al 2008). Drawbacks of these methods are that they rely on a score which represents how close the results of a simulated model are to the target observation and that they in general only yield the best performing set of parameters. Furthermore, these algorithms are often computationally expensive since they require many simulations to find suitable parameters (Gonçalves et al 2020). The class of simulation-based inference (SBI) algorithms makes statistical inference methods available for models where the likelihood is not tractable and provides an approximation of the posterior distribution of the model parameters. Advantages of deriving an approximation of the posterior include the possibility to find correlations between model parameters and to evaluate the confidence in the estimated parameters. Early SBI approaches rely on defining a score and are computationally inefficient since they disregard many simulation which have a low score (Sisson et al 2018). © 2023 The Author(s). Published by IOP Publishing Ltd Neuromorph. Comput. Eng. 3 (2023) 044006 J Kaiser et al Recent advances in machine learning lead to a new class of SBI algorithms which promise to be computationally more efficient and do not depend on a score function (Papamakarios and Murray 2016, Lueckmann et al 2017, Greenberg et al 2019, Cranmer et al 2020, Deistler et al 2022). In this paper we will focus on the sequential neural posterior estimation (SNPE) algorithm which was already applied to infer parameters for different neuroscientific models (Lueckmann et al 2017, Gonçalves et al 2020). More specifically, we want to investigate if this algorithm is suitable to parameterize neuron models which are emulated on the BrainScaleS-2 (BSS-2) analog neuromorphic hardware system (Pehle et al 2022). Neuromorphic computation draws inspiration from the brain to find time and energy efficient computing architectures as well as algorithms (Indiveri et al 2011). The BSS-2 system emulates the behavior of neurons and synapses on analog circuits in continuous time (Billaudelle et al 2022) and does not solve the model equations mathematically like digital neuromorphic hardware (Furber et al 2013, Davies et al 2018, Mayr et al 2019). In previous experiments on the BSS-2 system, hardware parameters were set by calibration routines, grid searches, gradient-based optimization or by hand-tuning (Aamir et al 2018, Wunderlich et al 2019, Billaudelle et al 2022, Cramer et al 2022, Kaiser et al 2022, Pehle et al 2023). The hand-tuning of parameters can be tedious and relies on the domain-specific knowledge of the experimenter such that automated parameter-search methods are inevitable for complex problems (Vanier and Bower 1999). Similarly, a calibration routine can only be formulated if the relationship between parameters and observations is known. Depending on the dimensionality of the parameter space, grid searches and random searches can be computationally too expensive. The SNPE algorithm promises to find approximations of the posterior even if the parameter space is high-dimensional and the relationship between the parameters and the observation is unknown (Lueckmann et al 2017, Greenberg et al 2019, Gonçalves et al 2020). Furthermore, the SNPE algorithm is designed for probabilistic models. This makes it a suitable choice for models which deal with intrinsic probabilistic behavior such as analog neuromorphic hardware which is subject to temporal noise. In the present study we emulated a passive multi-compartmental neuron model on BSS-2 and investigated whether the SNPE algorithm can find suitable model parameters to reproduce previously recorded target observations. For a two-dimensional (2D) parameter space, we show that the approximated posterior derived with the SNPE algorithm agreed with a grid search over the whole parameter space and that the correlations between model parameters are in agreement with theoretical predictions. Finally, we extended the problem to a higher-dimensional (7) parameter space and examined the approximated posteriors with posterior-predictive checks (PPCs). The correlations between parameters of this high-dimensional model did agree with the model equations. All in all, our results indicate that the SNPE algorithm is able to deal with the intrinsic trial-to-trial variations of analog neuromorphic hardware and is able to approximate posterior distributions which are in agreement with the given target observations. 2. Methods This section starts by introducing the BSS-2 neuromorphic system. We chose the attenuation of post-synaptic potentials (PSPs) in a passive chain of compartments to test if the SNPE algorithm is capable to parameterize experiments on BSS-2. Therefore, we introduce the attenuation experiment before we describe the SNPE algorithm. We conclude this section by introducing methods which we used to validate our posterior approximations. 2.1. BrainScaleS-2 BSS-2 is a mixed-signal analog neuromorphic system; neuron and synapse dynamics are emulated by analog circuits while spike events and configuration data rely on digital communication, figure 1(a). More specifically, the dynamics of the analog neuron circuits are designed to resemble the dynamics of the adaptive exponential integrate-and-fire (AdEx) neuron model (Brette and Gerstner 2005, Billaudelle et al 2022). Voltages and currents on these analog circuits directly represent the state of the emulated neuron. 2.1.1. Neuron dynamics The AdEx neuron model extends the leaky integrate-and-fire (LIF) neuron model by introducing an exponential and an adaptation current (Brette and Gerstner 2005). The high configurability, see below, of the BSS-2 system allows disabling these currents to model LIF neurons. Furthermore, several neuron circuits can be connected to form multi-compartmental neuron models (Kaiser et al 2022). 2 Neuromorph. Comput. Eng. 3 (2023) 044006 J Kaiser et al Figure 1. The BrainScaleS-2 (BSS-2) system and the sequential neural posterior estimation (SNPE) algorithm. (a) Photograph of the BSS-2 neuromorphic chip bonded to a carrier board. Reproduced from (Pehle et al 2022). CC BY 4.0. (b) Visualization of the SNPE algorithm (Papamakarios and Murray 2016, Lueckmann et al 2017, Greenberg et al 2019). Reproduced from (Gonçalves et al 2020). CC BY 4.0. This algorithm can be used to find an approximation for the posterior distribution p(θ | x∗) of parameters θ which recreate a target observation x∗. The target observation x∗, a prior belief about the parameter distribution p(θ) and a model which gives implicit access to the likelihood p(x | θ) are given as inputs to the algorithm. In step (cid:192), we sample parameters θ ′ from the prior distribution and the model is evaluated with these parameters to obtain observations x ′. This implicitly allows us to sample form the likelihood p(x | θ ′). In the following step `, the set of parameters and the corresponding observations are used to train a neural density estimator (NDE). The NDE serves as a surrogate for the posterior distribution p(θ | x). Frequently, we are interested in a single observation x∗ and we can restrict the NDE to this observation, step ´. We can now use samples drawn from the posterior θ ′ ∼ p(θ | x∗) to generate new samples and retrain the NDE, repeating steps ` and ´. Steps `–ˆ can be repeated several times to improve the estimate of the posterior. In this publication, we will consider multi-compartmental neuron models for which the membrane potentials in the different compartments Vm adhere to the dynamics of the LIF neuron model, Cm dVm (t) dt = gleak · (Vleak − Vm (t)) + Isyn (t) + Iaxial (t) , where Cm is the membrane capacitance, gleak the leak conductance and Vleak the leak potential. The two currents in equation (1) arise due to synaptic input, Isyn, and connections to neighboring compartments, Iaxial. The synaptic current Isyn models current-based synapses with an exponential kernel. The current Iaxial,i(t) on compartment i 3 due to neighboring compartments is given by Iaxial,i (t) = ∑ j gi ↔j axial · ( ) Vm,j (t) − Vm,i (t) , (1) (2) where the sum runs over all neighboring compartments {j}, gi ↔j compartments and Vj is the membrane potential of the neighboring compartment. axial represents the conductance between these Once the membrane potential Vm crosses a threshold potential Vthres a spike is generated and the membrane potential is reset to the reset potential Vreset and the membrane potential Vm continues to adhere to the dynamics of equation (1). 4. After the refractory time τ ref the reset is released 2.1.2. Configurability The behavior of the neuron circuits on BSS-2 can be controlled by several digital and analog parameters. Digital parameters, for example, control if the adaptation or exponential currents are connected to the membrane (Billaudelle et al 2022) and how different neuron circuits are connected to each other to form multi-compartmental neuron models (Kaiser et al 2022). Analog reference voltages and currents control quantities such as the leak conductance gleak, leak potential Vleak or the axial conductance gaxial between neuron circuits. These analog references are provided by an analog on-chip memory array which converts digital 10 bit values to currents and voltages (Hock et al 2013). Since the last value is reserved, reference currents and voltages can be adjusted digitally from 0 to 1022. This large configuration range allows tuning the neuron circuits to a variety of different operating regimes and to compensate manufacturing-induced mismatch between different neuron circuits (Billaudelle et al 2022). 3 Since all variables in equation (1) refer to compartment i, we omitted the subscript i in equation (1) for easier readability. 4 These digital spikes can be used as inputs to other neurons on the chip or can be recorded as observables. 3 Neuromorph. Comput. Eng. 3 (2023) 044006 J Kaiser et al leak and the axial conductance between compartments gi ↔i +1 Figure 2. Model of a passive compartment chain and grid search results. (a) The parameters of the model are given by the leak conductance in each compartment gi the propagation of post-synaptic potentials (PSPs). Here we show membrane traces of neurons which were emulated on the BrainScaleS-2 system. We inject a synaptic input (vertical lines) in one compartment after another and record the membrane potential in each compartment (different rows). From these traces we extract the heights of the PSPs hij. We use the matrix of all heights H, the heights resulting from an input to the first compartment F = [h00, h10, h20, h30] or the decay constant τ from an exponential fit to F as observables. The scale bar in the lower right corner indicates the voltage and time in the hardware domain. (b) Grid search on BrainScaleS-2 of the decay constant τ ; the decay constant is given in units of ‘compartments’ and calculated by fitting an exponential to the PSPs which result from an input to the first compartment, compare panel (a). We divided the parameter space in an evenly spaced grid with 40 values in each dimension, recorded the resulting PSP heights in each compartment and extracted the decay constant τ ; figure A1 shows the exponential fits for some exemplary measurements. The decay constant τ decreases as the leak conductance gleak is increased or the axial conductance gaxial is reduced. The white contour lines mark regions with equal decay constant and show a correlation between leak and axial conductance. Traces recorded at the numbered points are displayed in figure 3. . In our experiment we observe axial In the current publication, we use the latest revision of the BSS-2 system (Billaudelle et al 2022, Pehle et al 2022). The PyNN domain-specific language (Davison et al 2009) was used to formulate the experiments and the BSS-2 OS to define as well as to control the experiments (Müller et al 2020). 2.2. Experiment description—a linear chain of compartments In order to test the capabilities of the SNPE algorithm, we considered a multi-compartmental model which consisted of a chain of passive compartments, see figure 2(a). Such multi-compartmental models have been used to model dendrites and axons (Fatt and Katz 1951, Rall 1962). Each compartment i was connected to a leak potential Vleak via a leak conductance gi conductance gi ↔i +1 parameters were fixed. , compare equation (2). These conductances served as our parameters θ, all other leak and to the neighboring compartment via an axial axial We injected synaptic inputs in the different compartments and observed how the PSPs propagate along the chain. More specifically, we looked at the heights of PSPs; in the following we will use the notation hij to describe the PSP height which was observed in compartment i after an input to compartment j, figure 2(a). Since we were only interested in the passive propagation, we disabled the spiking threshold, this is equivalent to Vthres → ∞. Due to the low-pass properties of the passive chain, the response in the first compartment broadened and its height decreased as the synaptic input was injected further away from the first compartment, compare first row in figure 2(a). A similar behavior was visible when we looked at the voltage traces in the second compartment: the PSPs broadened and flattened for inputs further away from the recording site. Since we considered a finite chain, we saw that an input at the end of the chain affected the membrane potential more strongly, for example h10 > h12. The height of the PSPs depended on the leak and axial conductance (Fatt and Katz 1951). A higher leak or axial conductance resulted in lower PSP heights at the injection site as less charge can be accumulated on the compartment, figure 2(b). Therefore, the PSP heights H or quantities derived form them were suitable observations x that could be used to infer parameters θ. Besides the full matrix of PSP heights, we used the PSP heights which resulted from an input to the first compartment F = [h00, h10, h20, h30] and the decay constant τ from an exponential fit to F as observables. 2.3. Sequential neural posterior estimation algorithm The SNPE algorithm (Papamakarios and Murray 2016, Lueckmann et al 2017, Greenberg et al 2019) belongs to the class of SBI algorithms and allows finding an approximation of the posterior distribution p (θ | x∗) in cases where the likelihood p (x | θ) is intractable. Here θ are the parameters of a mechanistic model for which we try to find parameters which reproduce a target observation x∗. The main idea is to evaluate the model for different parameters {θi }, extract the observations {xi } and fit a flexible probability distribution as a 4 Neuromorph. Comput. Eng. 3 (2023) 044006 J Kaiser et al posterior to this set of parameters and observations. As the name suggests the parameters of these probability distributions are determined by neural networks. The algorithm takes a target observation x∗, prior p(θ) and a model for which suitable parameters should be found as an input, figure 1(b). The prior is used to draw random parameters θ ′ ∼ p(θ). By executing the model with the given parameters θ ′ our case the evaluation of the model is the emulation on the BSS-2 system. we implicitly sample from the likelihood x ′ ∼ p(x | θ ′). In In the second step, a neural density estimator (NDE) is trained to approximate the posterior distribution p(θ | x). The NDE is a flexible set of probability distributions which are parameterized by a neural network. Typical choices are mixture-density networks (Bishop 1994, Papamakarios and Murray 2016, Lueckmann et al 2017, Greenberg et al 2019) or masked autoregressive flows (MAFs) (Papamakarios et al 2017, 2019, 2021, Gonçalves et al 2020). The NDE is commonly trained by minimizing the negative log-likelihood of the previously drawn samples. Therefore, unlike traditional SBI algorithms the SNPE algorithm does not depend on a user-defined score function. After successful training, the NDE approximates the posterior distribution of the parameters for any observation x. If we are only interested in a single target observation x∗, we can use the estimated posterior distribution in the following rounds as a proposal prior (Papamakarios and Murray 2016, Lueckmann et al 2017, Greenberg et al 2019). While this sequential approach can increase sample efficiency, the obtained approximation of the posterior is no longer amortized, i.e. it can only be used to infer parameters for the target observation x∗ and not any arbitrary observation x. In our experiments we applied the algorithm presented in Greenberg et al (2019) which is implemented in the Python package sbi5 (Tejero-Cantero et al 2020). The structure of the NDE as well as other hyperparameters of the SNPE algorithm can be found in appendix A.2. 2.4. Validation In order to validate the approximated posteriors we used PPCs and calculated the expected coverage for each posterior (Hermans et al 2022). 2.4.1. Predictive posterior check We performed PPCs to check if an approximated posterior p (θ | x∗) yielded parameters θ which are in agreement with the original observation x∗. As discussed in Lueckmann et al (2021), PPCs do not measure the similarity of the approximated and true posterior and should just be used as a check rather than a metric. Nevertheless, we found that PPC were sensitive enough to highlight posterior approximations which did not agree with our expectation of the posterior based on grid search results. In the appendix, we illustrate examples of mismatching posteriors, figure A5, and show how we used PPCs to adjust the hyperparameters of the NDE, figure A6. For all PPCs we drew 1000 random parameters {θi} from the approximated posterior p (θ | x∗), emulated the chain model with these parameters on BSS-2 and recorded the observables {xi}. We used the mean Euclidean distance between these observations and the target observation x∗ as an indicator for an successful approximation. 2.4.2. Expected coverage Recent publications indicate that the posteriors approximated with the SNPE algorithm tend to yield overconfident posterior approximations, i.e. the posterior distribution is to narrow (Deistler et al 2022, Hermans et al 2022). To test the confidence of our posteriors we calculated the expected coverage as suggested in Hermans et al (2022). We calculated the expected coverage as follows. First we drew 1000 random samples from the prior distribution, {θ∗ i observations {x∗ i (θ∗, x∗)i and averaged over them to get the expected coverage. } ∼ p(θ). We then performed the experiment with these parameters on BSS-2 to obtain }, yielding pairs {(θ∗, x∗)i } ∼ p(θ, x). Finally, we calculated the coverage of each pair The coverage of a single pair was calculated as follows. We drew 10 000 samples for from the amortized } ∼ p(θ | x∗ posterior {θ ′ j approximations of the posterior and not the final approximations6). Next, we used the posterior probability of the original parameter p(θ∗ | x∗ i ) for each pair (i.e. we performed the coverage tests with the first round i ) and of the drawn samples {p(θ ′ i )}j to estimate the coverage. | x∗ j 5 https://github.com/mackelab/sbi, we used version 0.21.0. 6 As the first round approximation is amortized, we could condition it on arbitrary observations. Approximations in later rounds are restricted to a single target observation. 5 Neuromorph. Comput. Eng. 3 (2023) 044006 J Kaiser et al 3. Results To simplify the problem, we started by considering a 2D parameter space. This was achieved by setting the leak and axial conductance globally. The low dimensionality of the parameter space allowed us to perform a grid search in a reasonable amount of time and to easily visualize the results. The grid search result can give an intuition about the behavior of the chain and was used as a comparison to the approximated posterior obtained with the SNPE algorithm. We also executed the SNPE for a high-dimensional (7) parameter space and performed PPCs. For both, the 2D and high-dimensional parameter space we looked at different kind of observations and how these influence the approximated posterior. Furthermore, we analyzed the correlation for each posterior and performed coverage tests. 3.1. 2D parameter space We reduced the dimensions of the parameter space to two by setting the leak and axial conductance for all compartments and connections to the same digital value7; gi axial = gaxial ∀i ∈ {0, 1, 2}. leak = gleak ∀i ∈ {0, 1, 2, 3} and gi ↔i +1 3.1.1. Grid search In order to obtain an overview of the model behavior, we performed a grid search over the 2D parameter space. We created a grid of parameters by choosing equally spaced values of the leak and axial conductance which span the whole parameter range. The model was then emulated with these parameters on the BSS-2 system and the membrane traces in the different compartments were recorded. In order to easily visualize the results, we selected a one-dimensional (1D) observable. Exponential fits to the maximal height of propagating PSPs were used in other publications to classify the attenuation of PSPs in apical dendrites (Berger et al 2001). Similarly, we fitted an exponential to the PSP heights which resulted from an input to the first compartment F = [h00, h10, h20, h30] and analyzed the exponential decay constant τ , figure 2(b). The decay constant increased with increasing axial conductance gaxial and decreasing leak conductance gleak. Even though the exponential is just an approximation for the attenuation of transient inputs in multi-compartmental models, a correlation between leak and axial conductance is expected (Fatt and Katz 1951, Rall 1962). This behavior can also be understood with equations (1) and (2): a lower leak conductance gleak leads to less charge leaking from the membrane and consequently a larger charge transfer to the neighboring compartments, which can be counterbalanced by a lower axial conductance gaxial. The responses of the membrane potentials to a synaptic input in the first compartment are displayed in figure 3(b). For a low leak and a large axial conductance, 0⃝, the attenuation was the weakest and the PSP was still clearly visible in the last compartment. Parameters on the same contour line showed, as expected, similar attenuation, (cid:192) and `, even though the exact shape of the PSPs differed. For a large leak and a low axial conductance, ´, the PSP decayed quickly and almost vanished in the third compartment. 3.1.2. Simulation-based inference We used the SNPE algorithm to infer possible parameters θ = [gleak, gaxial] which reproduce a target observation x∗ = [τ ∗]. Furthermore, we investigated how the posterior distribution changed when a more informative observation x∗ = F∗ = [h∗ 30] was used, compare figure 2(b). In the case where a target observation x∗ is given by an experiment, the true posterior and the optimal model parameters which replicate the observation are typically unknown. This makes it hard to assess the quality of the posterior approximated by the SNPE algorithm. Therefore, we explicitly chose target parameters θ∗ observation x∗ = τ ∗. This allowed us to perform a closure test and check whether the SNPE algorithm was able to estimate a posterior which agreed with the initial observation. , emulated our model with these parameters on BSS-2 and measured an ‘artificial’ target 10, h∗ 20, h∗ 00, h∗ We picked a target parameter θ∗ at the center of the parameter space and executed the model with this parameter 100 times to account for trial-to-trial variations due to temporal noise. From the full matrix of PSP heights H we extracted different target observations such as the decay constant τ . The mean of the observed decay constants was our target observation x∗ = [τ ∗] = 1.17 ± 0.04; the decay constant is in units of ‘compartments’. In contrast, while running the SNPE algorithm we executed the model just once for each parameter and did not average over several trials. We used a uniform distribution over all possible parameters as a prior distribution p(θ) and executed the SNPE algorithm to obtain an approximation of the posterior distribution p (θ | x∗). The uniform 7 Due to the production induced mismatch between analog circuits, the same digital values lead to different conductances on the BSS-2 system. 6 Neuromorph. Comput. Eng. 3 (2023) 044006 J Kaiser et al Figure 3. Propagation of post-synaptic potentials (PSPs) in a passive chain of four compartments emulated on the BrainScaleS-2 system. Leak and axial conductance were set to the same value for all compartments and connections between compartments. (a) Grid search result illustrated as the difference of the measured decay constant τ , compare figure 2(b), to the target decay constant τ ∗: |τ − τ ∗|. Traces recorded at the numbered points are displayed in panels (b) and (e). (b) Example traces recorded at different locations in the parameter space, compare panel (a). The colors of the traces indicate in which compartment the trace was recorded, compare figure 2(b). The traces are scaled relative to the height in the first compartment h00. Due to the faster emulation of the neural dynamics on BSS-2, the time scales are in the microsecond rather than in the millisecond range. (c) Posterior obtained with the sequential neural posterior estimation (SNPE) algorithm. The posterior shows a high density in the parameter region where the target decay constant τ ∗ was recorded, `. As expected from the grid search result in panel (a), a correlation between the leak and axial conductance is visible. Points where the decay constant is significantly lower/higher than the target observation show a low probability density, 0⃝ and ´. (d) 500 random samples drawn from the approximated posteriors for two different types of observations. The green points represent samples drawn from the posterior which is shown in panel (c). The samples show a correlation between both parameters. If the absolute heights of the PSP which resulted from an input to the first compartment F = [h00, h10, h20, h30] was chosen as observations (blue), the samples scatter around point ` where the original target F∗ was recorded. The histograms at the top and right of the scatter plot show histograms of the parameter distribution in one dimension. (e) Same traces as in panel (b) but shown on an absolute scale. While traces (cid:192) and ` share a similar decay constant τ , compare panels (a) and (b), their absolute heights differs. distribution covered the whole adjustable range of the leak and axial conductance which ranges from 0 to 1022, see section 2.1.2. For a number of problems the SNPE algorithm was reported to be overconfident and ensembles made up of several posteriors were used to retrieve a more conservative posterior approximation (Deistler et al 2022, Hermans et al 2022). Since some of your posteriors were also overconfident, see later section, we combined five posterior to a posterior ensemble. In order to facilitate the comparison of the grid search results and the approximated posterior, we display the difference between the target decay constant τ ∗ and the measured decay constant τ during the grid search in figure 3(a). As expected form the grid search, a correlation between the leak gleak and the axial conductance gaxial is clearly visible in the approximated posterior, figure 3(c). The posterior distribution shows high densities for parameters θ which reproduced observations near the target observation during the grid search. In order to retrieve a narrow posterior around the original parameters θ∗ , a more informative observations was needed. While the PSP heights showed a similar decay for different sets of leak and axial conductance, figure 3(b), the absolute heights of the PSPs differed, figure 3(e). We therefore used the PSPs heights which resulted from an input to the first compartment F as a target observation, x∗ = F∗, to further constrain possible parameters. The heights in the first compartment F were extracted from the same 100 trials as the decay constant τ . We ran the SNPE algorithm once again to retrieve another approximation of the posterior. Samples {θi} drawn from this posterior were now scattered around the original parameter θ∗ in the parameter space and the parameters were uncorrelated, figure 3(d); the Pearson correlation coefficient decreased from 0.92 to 0.004. The marginal distribution of the leak and axial conductance were bell-shaped and showed a high density near the target parameter θ∗ . 3.1.2.1. Validation In order to perform a PPC, we drew samples {θi} from the posterior distribution, figure 3(c), configured our model with them and compared the observations {xi} with the target observation x∗. We measured a mean decay constant of τ = 1.18 ± 0.08 which agrees with the target τ ∗ = 1.17 ± 0.04. Therefore, we conclude that the approximated posterior is in agreement with the target observation τ ∗. The uncertainty of the posterior predictive increased compared to the target observation since it contains the aleatoric uncertainty, 7 Neuromorph. Comput. Eng. 3 (2023) 044006 J Kaiser et al Figure 4. Validation of the approximated posteriors found with the sequential neural posterior estimation (SNPE) for a compartment chain of four compartments and setting parameters globally, compare figure 3. The gray lines mark the expected coverage (Deistler et al 2022, Hermans et al 2022) of posterior approximations found with the SNPE algorithm, the black line marks the expected overage of the posterior ensemble which is made up of those posteriors. Left: using the decay constant τ as an observable. Several posteriors have an expected coverage below the diagonal which indicates an overconfident probability distribution. When combining several posteriors to an ensemble, the expected coverage follows the diagonal which is a sign of a well calibrated posterior. Right: post-synaptic potential heights resulting from an input to the first compartment as an observation. All posteriors and their ensemble appear well calibrated as they closely follow the diagonal. due to the inherent trial-to-trial variations, as well as the epistemic uncertainty which stems from the width of the posterior distribution. To test the calibration of the approximated posterior, we calculated the expected coverage, compare section 2.4. When we used the decay constant τ as a target, three out of five posteriors were overconfident, figure 4. We followed the methods presented in Deistler et al (2022) and Hermans et al (2022) and combined five posteriors to form an ensemble. The expected coverage of this ensemble closely follows the diagonal and indicates a well calibrated posterior. In case of the heights F as a target, the single posteriors were already well calibrated. And consequently, the ensemble of five posteriors was also well calibrated. In the appendix , we compare the results from the emulation on BSS-2 with computer simulations performed in the simulation library Arbor (Abi Akar et al 2019), figure A8. 3.2. Multidimensional parameter space In order to increase the problem complexity, we set the leak and axial conductance for each compartment and connection individually. For four compartments this resulted in a total of seven parameters; four leak conductances gi leak (i = 0, 1, 2, 3) and three axial conductances gi ↔i +1 (i = 0, 1, 2). axial As in the previous section we used a uniform prior and the PSP heights caused by an input to the first compartment as a target (x∗ = F∗). We then executed the SNPE algorithm, combined five approximated posteriors to an ensemble and drew samples from this posterior p(θ | x∗). The marginal distribution of the sampled leak conductance in the first compartment g0 leak was bell-shaped and peaked near the target parameter, figure 5(a). The almost uniform distributions of the leak conductances in the other compartments indicated that they were not relevant for the chosen observation. In contrast, the marginal distribution of all axial conductances were bell-shaped with a high density around the original parameters. The distributions of the axial conductance became broader for conductances later in the chain, suggesting that the influence of these conductances on the observable was weaker. Similar to the 2D case, we considered a higher-dimensional observation as a target to retrieve narrower posterior distributions, i.e. we chose all PSP heights as a target (x∗ = H∗), figure 5(a). Now the 1D marginals of all parameters were bell-shaped. The marginals of the axial conductance showed a narrower distribution than these of the leak conductance, indicating that the given observation was more sensitive to the axial conductance. 3.2.1. Correlation In figure 5(b) we display the correlation between posterior samples, the 1D and 2D marginals of posterior samples can be found in figures A3 and A4. When we considered the PSP heights which resulted from an input to the first compartment F as an observable we saw strong negative correlations between the leak conductance in the first compartment g0 as well as the axial conductance between both compartments g1↔2 leak and the leak conductance in the neighboring compartment g1 axial . This can be explained with equations (1) leak 8 Neuromorph. Comput. Eng. 3 (2023) 044006 J Kaiser et al Figure 5. Results of the sequential neural posterior estimation algorithm for a compartment chain of four compartments and setting parameters individually for each compartment and connection between them. Emulations were performed on the neuromorphic BrainScaleS-2 system. (a) Histograms of 10 000 parameters drawn from the approximated posterior. For the heights F of the post-synaptic potentials (PSPs) which resulted from an input to the first compartment as a target observation (blue), the distribution of the leak conductance in the first compartments is bell-shaped and peaks near the target parameter (dotted line). The leak conductance is roughly uniformly distributed in later compartments. The distributions of the axial conductance are bell-shaped and broaden for later compartments. Choosing all heights H as a target (orange) leads to narrower distributions. All histograms are now bell-shaped with a peak near the target (dotted line). (b) Pearson correlations between different parameters. The color denotes the value of the correlation while the radius of the circle encodes the absolute value of the correlation. Left: PSP heights resulting from an input to the first compartment F as a target. The strongest correlations can be observed for the leak conductance in the first compartment g0 compartment g1↔2 Overall the correlations decrease for this more informative target. Only between neighboring leak conductances a high negative correlation can be observed. axial ; for parameters later in the chain the correlations shows lower values. Right: all PSP heights H as a target. leak and the axial conductance between the first and second and (2) and considering the PSP height in the first compartment: when the leak conductance g0 compartment increases, a higher current leaks from the membrane which would result in a smaller PSP height; to counter this effect the charge which flows to the neighboring compartment has to be minimized by reducing the axial conductance g1↔2 neighboring compartment. The leak conductance g0 axial between the compartments or the leak conductance g1 leak was also negatively correlated to the other leak and axial conductances, leak in the first leak of the compare first column in figure 5(b). The magnitude of the correlation decreased for parameters further away from the first compartment. Apart from the correlation with the leak conductance g0 compartment, the correlation between the other leak conductances was low. Interestingly, the correlations between the axial conductances gi ↔i+1 and the other leak conductance leak of the first axial gi leak, i > 0 was positive. As mentioned above a higher leak conductance leads to a larger leak current which results in a smaller PSP height. Since we only considered an input to the first compartment, this increased leak conductance gi leak could be counteracted by increasing the charge which is injected from the previous compartments and therefore increasing the conductance gj−1↔j ; j ⩽ i, i > 0 to compartments earlier in the ; j ⩾ i, i > 0 were still chain. The correlation of the leak conductance gi positive but significantly lower. leak to later axial conductances gj ↔j+1 axial axial As expected, all axial conductances were correlated positively. This can be explained when considering one compartment i, i > 0. An increase in the axial conductance gi−1↔i compartment i; to prevent an accumulation of charge and therefore a larger PSP height, the conductance to the next compartment gi ↔i+1 leads to a stronger current on has to increase as well. axial axial leak and gi+1 When taking all heights H as a target observation, the correlation between the different parameters leak rather high negative correlations decreased. Only between neighboring leak conductances gi could be observed. To understand this correlation, we can consider two cases. We look at one compartment i, increase its leak conductance gi neighboring compartment i ± 1. First, input in the same compartment: as before an increased leak leak results in a lower PSP height which can be compensated by a smaller leak conductance gi±1 conductance gi in the neighboring compartments. Similarly, in the second case when the input is injected in a neighboring compartment, an increased leak conductance gi decreased PSP height. Consequently, the leak conductance gi±1 reduced such that more charge can flow on compartment i. leak and consider once an input to the same compartment i and once to a leak would once again lead to an increased leakage and a leak in the neighboring compartment should be leak 3.2.2. Validation We once again used PPCs to check if samples drawn from the approximated posterior {θi} reproduce the target observation. The mean difference between observations {Hi} obtained with these parameters and the 9 Neuromorph. Comput. Eng. 3 (2023) 044006 J Kaiser et al Figure 6. Validation of the approximated posteriors found with the sequential neural posterior estimation (SNPE) for a compartment chain of four compartments and setting parameters individually for each compartment and connection between them, compare figure 5. Emulations were performed on the neuromorphic BrainScaleS-2 system. (a) Posterior-predictive check. The passive chain was configured with 1000 of the parameters {θi} drawn in figure 5(a) and the post-synaptic potential (PSP) heights in all compartments {Hi} were measured on the BrainScaleS-2 system. These PSP heights were compared to the observation H∗ which represents the measurement with the target parameters θ∗. The vertical lines show the mean deviation of the observations {Hi} from this target H∗ while the horizontal bars illustrate the standard deviation of this deviation. As mentioned in the introduction, analog hardware is subject to temporal noise. Therefore, the hardware was configured to the target parameters θ∗ 100 times and the mean PSP heights were chosen as a target H∗; the deviations in this panel are scaled by the standard deviation σ∗ of these 100 measurements (each height deviation hij is divided by the standard deviation of the height σ∗ ij ). For all PSP heights the mean observation is within 1 to 2 standard deviations of the initial target. When a more informative observation H is chosen, the standard deviations decreases. A prior-predictive check can be found in the appendix, figure A2. (b) Coverage tests. The gray lines mark the expected coverage (Deistler et al 2022, Hermans et al 2022) of posterior approximations found with the SNPE algorithm, the black line marks the expected overage of the posterior ensemble which is made up of these posteriors. Left: PSP heights resulting from an input to the first compartment F as an observation. The expected coverage is for all confidence levels below the diagonal which suggests that the posteriors are overconfident. Even an ensemble made up of five posteriors is not well calibrated. Right: all PSP heights H as a target. The individual posteriors are overconfident but the ensemble of them is well calibrated. target observation H∗ are displayed in figure 6(a). H describes all observed PSP heights and the target observation F∗ was extracted from H∗, see figure 2(a). The mean of the PSP heights for an input to the first compartment (first column) was near the initial target values; the standard deviation was in the range of 1–2 σ∗ where σ∗ is the standard deviation of the measurements which were used to extract the target observation H∗. For responses in the first compartment (first row) a similar standard deviation could be observed, but the mean observation showed a slightly higher deviation from the target observation. For the other PSP heights the mean was still in the one-sigma range of the initial target observation, but the standard deviation of the observations was significantly higher. The small deviation of the mean observations can be explained by our target parameter which is located at the center of the parameter space; a prior predictive check also yielded mean observations near the target observations, compare figure A2. The higher standard deviations are expected since these PSP heights have not been part of the observation and can be attributed to the broad posterior distribution of the leak and axial conductance in later compartments, compare figure 5(a). The sharpening of the posterior distribution was also visible in the results of the PPC, figure 6(a). Here the standard deviation of the observations decreased to the range of 1–2 σ∗ for all PSP heights. As for the 2D case, we calculated the expected coverage for the approximated posteriors and their ensembles, figure 6(b). With the heights which resulted from an input to the first compartment F as an observable, all five posteriors were overconfident and also an ensemble made up of these five posteriors was still overconfident. When using all heights H as an observation, the expected coverage of the individual posteriors were similar to the case before. However, the expected coverage of the ensemble was near the diagonal; this suggests that the posterior was well calibrated. 4. Discussion We have shown that the SNPE algorithm can be used to parameterize the analog neuromorphic BSS-2 system. To be able to investigate the posteriors approximated by the SNPE algorithm, we selected a multi-compartmental model which takes the form of a chain of passive compartments. We chose the leak conductance as well as the axial conductance between compartments as parameters and observed how PSPs propagated along the chain. This model allowed us to easily change the dimensionality of the parameter space as well as the choice of observable and evaluate how this influences the approximated posteriors. 10 Neuromorph. Comput. Eng. 3 (2023) 044006 J Kaiser et al In all our experiments, we picked a set of target parameters, extracted an observation with these parameters and then used the SNPE algorithm to approximate the posterior distribution of the parameters which reproduce this given observation. As a first step, we considered a 2D parameter space where we set all leak conductances and axial conductances to the same value. The low dimensionality of the parameter space allowed us to perform a grid search in a reasonable amount of time. The posterior approximated by the SNPE algorithm agreed with the results from this grid search. In both cases we found a correlation between the leak and axial conductance when looking at the attenuation of PSPs; this agrees with theoretical expectations (Fatt and Katz 1951, Rall 1962). To be able to find such correlation is one of the advantages of a posterior approximation over traditional parameter search algorithms which usually only yield a set of parameters which reproduce the given observation but do not illustrate the relation between different parameters. When we chose a more informative observation, specifically the height of the PSPs which result from an input to the first compartment, the posterior distribution of the parameters narrowed and the correlation between leak and axial conductance vanished. We further showed that the algorithm is capable of finding appropriate posterior approximations for several, random values of the target parameters. The approximations were even in agreement with the target parameters if they lie at the edges of the parameter space. This indicates that the algorithm is able to deal with the hard parameter limits which are dictated by the neuromorphic hardware. Furthermore, we performed coverage tests to assess the calibration of the posterior approximations. The posteriors produced with the less informative observation were overconfident, requiring an ensemble of five posteriors to retrieve a well calibrated posterior. In contrast, all posteriors approximated for the more informative observation were well calibrated. Next, we increased the dimensionality of the parameter space by adjusting each leak and axial conductance individually; resulting in a seven-dimensional parameter space. We showed that the marginal distributions of samples drawn from the posterior approximations have a high density around the target parameters. In addition, we analyzed the correlation between the different parameters and showed that they agree with the model equations. Furthermore, we conducted PPCs to verify that the parameters drawn from the approximated posterior yield emulated results which align with the target observation. Similar to the 2D case, increasing the dimensionality of the observable resulted in a narrower posterior distribution. When using the height of the PSPs which resulted from an input to the first compartment as an observable, we did not find well calibrated posteriors even when combining multiple posteriors into an ensemble. After increasing the dimensionality of the observable, the individual posteriors remained overconfident but the ensemble made up of five of them was well calibrated. 5. Conclusion The SNPE algorithm has previously only been utilized to identify suitable parameters for numerical simulations (Lueckmann et al 2017, Greenberg et al 2019, Gonçalves et al 2020, Deistler et al 2022). In the current work we show that the algorithm can also be employed to parameterize a physical system, namely the BSS-2 neuromorphic system. In contrast to other search algorithms such as random search, genetic algorithms or gradient-based algorithms, the SNPE algorithm provides an approximation of the full posterior and therefore allows to identify correlations between parameters and to quantify the confidence of the parameter estimation. Additionally, the SNPE algorithm is agnostic to the internal dynamics of the experiment and does not require the calculation of gradients. Compared to traditional SBI methods the SNPE algorithm offers a higher simulation efficiency (Papamakarios and Murray 2016, Cranmer et al 2020). As a result, SNPE is a viable alternative to traditional optimization methods. When one simply optimizes for a single objective and is not concerned with the correlations between parameters, gradient based methods can offer a more directed optimization approach and are potentially faster in recovering suitable parameters; they have successfully been used to find parameters for BSS-2 (Cramer et al 2022, Arnold et al 2023, Pehle et al 2023). However, having access to an approximated posterior distribution and the correlations between different parameters can give valuable insight in the dynamics of the underlying model as shown in the current study. To evaluate the quality of the approximated posteriors, we generated the target observation from our model. As a result, we knew the true parameters of the target observation and were certain that our model can reproduce the given observation. In subsequent studies, we will use the SNPE algorithm to replicate observations which are generated by another model such as numerical simulations or by physiological experiments. 11 Neuromorph. Comput. Eng. 3 (2023) 044006 J Kaiser et al Furthermore, we only considered passive neuron properties in our current experiments. As seen in the grid search results, this lead to a rather smooth parameter space, where the observations change gradually with the model parameters. More complex neuron models of interest include non-linear behavior such as somatic or dendritic spikes and will potentially have high-dimensional parameter spaces. Gonçalves et al (2020) and Deistler et al (2022) have previously shown that the SNPE algorithm and derivatives of it can deal with such high-dimensional parameter spaces and non-linear behavior and it will be interesting if this success can be transferred to emulations on neuromorphic hardware. In summary, we demonstrated that the SNPE algorithm is able to find posterior approximations for parameters of the analog neuromorphic BSS-2 system. Data availability statement The data that support the findings of this study are openly available (Kaiser et al 2023). The experiment code is available on https://github.com/electronicvisions/model-paper-mc-sbi. Acknowledgments We thank the lead of HBP’s FIPPA project Christian Tetzlaff for scientific input; A Baumbach, S Billaudelle, A Grübl, J Ilmberger, C Mauch, C Pehle, Y Stradmann, P Spilger and J Weis for their contributions to BSS-2; as well as all present and former members of the Electronic Vision(s) research group. We thank the anonymous reviewers for their thorough evaluation and valuable suggestions. This work has received funding from the EU ([FP7/2007–2013], [H2020/2014–2020]) under Grant Agreements 604102 (HBP), 720270 (HBP SGA1), 785907 (HBP SGA2) and 945539 (HBP SGA3); the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany’s Excellence Strategy EXC 2181/1-390900948 (the Heidelberg STRUCTURES Excellence Cluster) as well as from the Manfred Stärk Foundation. Contributions J K, J S and S S designed research; J K and R S performed research; J K, J S, R S and S S analyzed data; J K and S S wrote the paper; all authors edited the paper; E M, J K, R S and S S contributed software; and J S designed the BrainScaleS-2 neuromorphic system. Appendix In the following appendix, we state which neuron parameters were used during the experiment, how the hyperparameters of the SNPE algorithm influenced the approximated posterior and compare our results to simulations in Arbor (Abi Akar et al 2019). Furthermore, we attach figures which extend the results presented in the paper; figures A1–A4. A.1. Neuron parameters In order to ensure a similar behavior of the different compartments, the leak potential and the synaptic properties were calibrated. The synaptic time constant was calibrated to a value of 10 µs. As can be extracted from figure 2(b), the decay constant varied in our experiments between 0.16 to 4.08 compartments. When varying the leak conductance gleak over the full range specified in figure 3, the membrane time constant τm = Cm gleak varies in the range of 12–30 µs. A.2. Sequential neural posterior estimation algorithm We adjusted the number of simulations as well as the properties of the NDE and used PPCs to check how these hyperparameters influence the approximated posterior. For each set of hyperparameters we executed the SNPE algorithm ten times with different seeds. The seeds influence the initial weights as well as the parameters θ which are drawn from the prior in the first round. Different sets of hyperparameters shared the same seeds. A.2.1. Number of simulations and rounds For the 2D parameter space and the decay constant τ as an observable, three times 50 emulations were sufficient to recover a posterior which is in agreement with the target observation. Retrieving the observation 12 Neuromorph. Comput. Eng. 3 (2023) 044006 J Kaiser et al Figure A1. Exponential fit to the traces displayed in figures 3(b) and (e). The heights of the post-synaptic potentials (PSPs) are extracted from the recorded membrane traces, compare figure 2(a), and exponentials (solid lines) are fitted to the measurement points. The numbering is the same as in figure 3. The x-axis label mark the compartment in which the height of the PSP was measured and in brackets the variable name as defined in figure 2(a). Figure A2. Comparison between a prior-predictive check and the posterior-predictive checks (PPCs) in figure 6(a). The data for the posterior-predictive checks (PPCs) are copied from figure 6(a), the PPC was performed for two different observations: post-synaptic potential (PSP) heights which resulted from an input to the first compartment F and all PSP heights H. The prior-predictive check was performed similar to the PPCs but the samples were drawn from the prior distribution p(θ). The vertical lines show the mean deviation of the observations {Hi} from this target H∗ while the horizontal bars illustrate the standard deviation of this deviation. The hardware was configured to the target parameters θ∗ 100 times and the mean PSP heights were chosen as a target H∗; the deviations in this figure are scaled by the standard deviation σ∗ of these 100 measurements (each height deviation hij is divided by the standard deviation of the height σ∗ so far off from the target observation in most cases, the standard deviation is significantly higher than for the PPCs. ij ). While the mean observation is not of a single emulation (including hardware configuration, experiment execution, data retrieval and evaluation) took about took about 420 ms. When the observable is changed to the height of the PSPs which result from an input to the first compartment F, the SNPE algorithm failed to find a suitable approximation if the number of emulations was too low. This was due to a poor approximation in the first round from which the algorithm needed some time to recover or may not recover in the given emulation budget, figure A5. We observed that a higher number of emulations in the first round reduced the number of cases where the posterior was approximated poorly. Therefore, we chose 500 emulations in the first round followed by ten rounds of 50 emulations for a 2D parameter space with F as an observable. We used two times 1000 emulations for the multidimensional parameter space, section 3.2. A.2.2. Neural density estimator Based on the results in Lueckmann et al (2021) we use MAFs as NDEs (Papamakarios et al 2017). MAFs transform normal distributions in other probability distributions. We used the values provided by the sbi package (Tejero-Cantero et al 2020) as defaults; similar values have also been used in previous publications (Gonçalves et al 2020, Lueckmann et al 2021). Here the MAF is made up of five transformations which are chained together. Each of these transformation consists of two blocks with 50 hidden units per block. For more information see Papamakarios et al (2017, 2021). In case of a 2D parameter space and the decay constant τ as a target, section 3.1, a single transformation with two blocks of ten hidden units each was sufficient. If we selected the heights which result from an input to the first compartment F as a target, a single transformation was not sufficient to recover a meaningful posterior, figure A6. Starting from two transformations and 30 hidden units, the best value of the PPC were 13 Neuromorph. Comput. Eng. 3 (2023) 044006 J Kaiser et al Figure A3. One- and two-dimensional marginal distributions of 1000 samples which were drawn from the posterior ensembles displayed in figures 5 and 6, the target observations were the PSP heights F which resulted from an input to the first compartment. obtained. The only exception is the network with three transformations and 20 hidden units for which the algorithm could not recover from a poor approximation in the first round. An MAF with 1 transformation and 50 hidden units is made up of 3764 trainable parameters and fails to approximate the true posterior. On the other hand an MAF with 5 transformations and 10 hidden units in each block offers just 1720 trainable parameters but is able to find approximations which agree with the target observation. We conclude, that a high number of transformations was more important for a good posterior approximation than a high number of trainable parameters. For the results reported in figure 3(d) we used the NDE with five transformations, two blocks and ten hidden units. A.3. Choice of the target parameters We chose a target parameter θ∗ at the center of the parameter space to measure target observations x∗. For the experiment with the 2D parameter space and the PSP heights for an input to the first compartment, we want to show that the approximated posterior is also appropriate for other choices of the target parameter θ∗ . As mentioned in the introduction, the posterior estimation is amortized after the first round of SNPE and can therefore be used to infer parameters θ for any observation x. We draw five random parameters {θ∗ } from the uniform prior and emulate the model on BSS-2 with the given parameters to record i observations {x∗ i θ ∼ p(θ | x∗ }. For each of these observations, we draw samples from the amortized posterior estimation i ), figure A7. For each of the randomly selected observations x∗ which were used to obtain the given observation θ∗ i the drawn samples cluster around the parameters i . Even if the target parameters are at the edge of the 14 Neuromorph. Comput. Eng. 3 (2023) 044006 J Kaiser et al Figure A4. One- and two-dimensional marginal distributions of 1000 samples which were drawn from the posterior ensembles displayed in figures 5 and 6, the target observation were all PSP heights H. parameter space, the approximated posterior returns samples near these target parameters. Therefore, we conclude that the SNPE algorithm is suitable to find parameters for observations which were obtained for parameters at arbitrary locations in the parameter space and that our choice of target parameters θ∗ at the center of the parameter space does not affect the generality of the reported results. A.4. Simulations We used the Arbor simulation library (version 0.8.1) to compare our results to computer simulations (Abi Akar et al 2019). Arbor is a high-performance simulator which supports multi-compartmental neuron models. As Arbor solves the model equations numerically, it does not suffer from trial-to-trial variations and thus we expect the posterior distributions to be narrower. As in the main part of the paper, we simulated a chain with four compartments. The length of a single compartment was set to lcomp = 1 mm, its diameter to dcomp = 4 µm and its capacitance to C = 125 pF. While the length and diameter were chosen arbitrarily, the capacitance reflected the capacitance of the compartments used during the emulation on BSS-2. The range of the leak conductance gleak was selected such that the membrane time constant of the simulated neurons was in agreement with the emulated neurons on BSS-2. Similarly, the range of the axial conductance gaxial was chosen such that the axial conductance along a simulated compartment is comparable to the conductance between compartments on BSS-2. The results from the grid searches were comparable, figures A8 and 3, but the chosen parameter ranges led to a slightly higher dynamic range of the length constant. In both cases a correlation between the leak and axial conductance was observed. 15 Neuromorph. Comput. Eng. 3 (2023) 044006 J Kaiser et al Figure A5. Evolution of the approximated posterior over several rounds of the sequential neural posterior estimation (SNPE) algorithm. Results are shown for emulations executed on the BrainScaleS-2 system. (a) Posterior-predictive check (PPC) for a emulation budget of 10 rounds with 50 emulations in each round (the PPC was executed with 1000 parameters sampled from the posteriors). The SNPE algorithm was executed ten times with different seeds. For some executions of the SNPE algorithm, the approximated posterior in the first round poorly replicates observations which are similar to x∗; this is evident in a high mean distance E. In all displayed cases the SNPE algorithm is able to recover a meaningful posterior. (b) Examples for one case where the SNPE algorithm is able to approximate a meaningful posterior and one case in which the algorithm fails to find a good approximation in the first three rounds. In both cases, the approximation in the first round does not agree with the true posterior. In the top row, the algorithm is able to quickly recover from the poor approximation while in the bottom row more rounds are needed to obtain a meaningful approximation. The parameter ranges are the same as in figure 3(c). Figure A6. Influence of the parameterization of the neural density estimator (NDE) on the approximation of the posterior. We used masked autoregressive flows (MAFs) as NDEs. MAFs transform normal distributions in other distributions (Papamakarios et al 2017). We used transformations which are made up of two blocks and change the number of hidden units which are used in each block (Gonçalves et al 2020). Furthermore, we changed the number of transformations which are chained together. As in figure A5(a) we performed a posterior-predictive check and used the mean distance between these samples and the target as a measure to decide if the approximation agreed with the target observation x∗. Again, we used the post-synaptic potential heights resulting from an input to the first compartment as an observable and repeated the sequential neural posterior estimation algorithm with ten different seeds for each set of hyperparameters. At least two transformation were needed to recover a meaningful posterior. The number of experiments in which a meaningful posterior could be recovered seemed to increase with the number of transformations. The total number of trainable parameters was not an indicator how well the NDE was able to approximate the true posterior. The shapes of the approximated posteriors also agreed with the results obtained for emulation on BSS-2. As expected, the approximated posterior distribution for the simulation was narrower than the approximation for BSS-2 due to temporal noise. 16 Neuromorph. Comput. Eng. 3 (2023) 044006 J Kaiser et al Figure A7. Posterior samples {θj}i ∼ p(θ | x∗ i ) for different observations x∗ i . We drew five random parameters θi from a uniform prior and one parameter at the center of the parameter space (marked by black crosses). The target observations {x∗ } were obtained by emulating the model 100 times for each parameter on BrainScaleS-2 and taking the mean height of the i post-synaptic potential obtained from an input to the first compartment, compare figure 3(d). As a posterior approximation we used the first round posterior obtained while executing the sequential neural posterior estimation algorithm in section 3.1.2. The samples drawn from the approximated posterior (small dots) are in the vicinity of the parameters which were used to create the target observations (black crosses). Figure A8. Propagation of post-synaptic potentials (PSPs) in a passive chain of four compartments simulated in Arbor. We performed the same experiments as in figure 3 and we follow the structure of this figure. (a) Grid search of the decay constant τ . The dependency on the leak conductance gleak and the axial conductance gaxial is comparable to figure 3(a). (b) Example traces recorded at different locations in the parameter space, compare panel (a). The traces are scaled relative to the height in the first compartment h00. (c) Posterior obtained with the sequential neural posterior estimation (SNPE) algorithm. While the shape of the approximated posterior is comparable to the one in figure 3(c), the approximated posterior for the simulations is narrower. (d) 500 random samples drawn from the approximated posteriors for two different types of observations. The distribution of the random samples is comparable to the results in figure 3(d), but in agreement with the narrower posterior in panel (c), the distribution of the samples is more narrow. (e) Same traces as in panel (b) but shown on an absolute scale. ORCID iDs Jakob Kaiser  https://orcid.org/0000-0002-3586-2634 Raphael Stock  https://orcid.org/0009-0008-5531-1072 Eric Müller  https://orcid.org/0000-0001-5880-2012 Johannes Schemmel  https://orcid.org/0000-0003-1440-4375 Sebastian Schmitt  https://orcid.org/0000-0002-7935-0470 References Aamir S A, Müller P, Kiene G, Kriener L, Stradmann Y, Grübl A, Schemmel J and Meier K 2018 A mixed-signal structured AdEx neuron for accelerated neuromorphic cores IEEE Trans. Biomed. Circuits Syst. 12 1027–37 Abi Akar N, Cumming B, Karakasis V, Küsters A, Klijn W, Peyser A and Yates S 2019 Arbor—a morphologically-detailed neural network simulation library for contemporary high-performance computing architectures 2019 27th Euromicro Int. Conf. on Parallel, Distributed and Network-Based Processing (PDP) (IEEE) pp 274–82 Arnold E, Böcherer G, Strasser F, Müller E, Spilger P, Billaudelle S, Weis J, Schemmel J, Calabr`o S and Kuschnerov M 2023 Spiking neural network nonlinear demapping on neuromorphic hardware for IM/DD optical communication J. Lightwave Technol. 41 1–8 Baker R E, Pe˜na J-M, Jayamohan J and Jérusalem A 2018 Mechanistic models versus machine learning, a fight worth fighting for the biological community? Biol. Lett. 14 20170660 17 Neuromorph. Comput. Eng. 3 (2023) 044006 J Kaiser et al Berger T, Larkum M E and Lüscher H-R 2001 High I(h) channel density in the distal apical dendrite of layer V pyramidal cells increases bidirectional attenuation of EPSPs J. Neurophysiol. 85 855–68 Billaudelle S, Weis J, Dauer P and Schemmel J 2022 An accurate and flexible analog emulation of AdEx neuron dynamics in silicon 2022 29th IEEE Int. Conf. on Electronics, Circuits and Systems (ICECS) pp 1–4 Bishop C M 1994 Mixture density networks Technical Report (Aston University) (available at: https://research.aston.ac.uk/en/ publications/mixture-density-networks) Brette R and Gerstner W 2005 Adaptive exponential integrate-and-fire model as an effective description of neuronal activity J. Neurophysiol. 94 3637–42 Cramer B, Billaudelle S, Kanya S, Leibfried A, Grübl A, Karasenko V, Pehle C, Schreiber K, Stradmann Y and Weis J 2022 Surrogate gradients for analog neuromorphic computing Proc. Natl Acad. Sci. 119 e2109194119 Cranmer K, Brehmer J and Louppe G 2020 The frontier of simulation-based inference Proc. Natl Acad. Sci. 117 30055–62 Davies M, Srinivasa N, Lin T-H, Chinya G, Cao Y, Choday S H, Dimou G, Joshi P, Imam N and Jain S 2018 Loihi: a neuromorphic manycore processor with on-chip learning IEEE Micro 38 82–99 Davison A P, Brüderle D, Eppler J, Kremkow J, Muller E, Pecevski D, Perrinet L and Yger P 2009 PyNN: a common interface for neuronal network simulators Front. Neuroinform. 2 11 Deistler M, Goncalves P J and Macke J H 2022 Truncated proposals for scalable and hassle-free simulation-based inference (arXiv:2210.04815) Fatt P and Katz B 1951 An analysis of the end-plate potential recorded with an intra-cellular electrode J. Physiol. 115 320–70 Furber S B, Lester D R, Plana L A, Garside J D, Painkras E, Temple S and Brown A D 2013 Overview of the SpiNNaker system architecture IEEE Trans. Comput. 62 2454–67 Gonçalves P J et al 2020 Training deep neural density estimators to identify mechanistic models of neural dynamics eLife 9 e56261 Greenberg D, Nonnenmacher M and Macke J 2019 Automatic posterior transformation for likelihood-free inference Proc. 36th Int. Conf. on Machine Learning vol 97 (PMLR) pp 2404–14 (available at: https://proceedings.mlr.press/v97/greenberg19a.html) Hermans J, Delaunoy A, Rozet F, Wehenkel A and Louppe G 2022 A crisis in simulation-based inference? Beware, your posterior approximations can be unfaithful Trans. Mach. Learn. Res. (available at: https://openreview.net/forum?id=LHAbHkt6Aq) Hock M, Hartel A, Schemmel J and Meier K 2013 An analog dynamic memory array for neuromorphic hardware 2013 European Conf. on Circuit Theory and Design (ECCTD) pp 1–4 Indiveri G et al 2011 Neuromorphic silicon neuron circuits Front. Neurosci. 5 73 Kaiser J, Billaudelle S, Müller E, Tetzlaff C, Schemmel J and Schmitt S 2022 Emulating dendritic computing paradigms on analog neuromorphic hardware Neuroscience 489 290–300 Kaiser J, Stock R, Müller E, Schemmel J and Schmitt S 2023 Simulation-based inference for model parameterization on analog neuromorphic hardware [data] (https://doi.org/10.11588/data/AVFF2E) Lueckmann J M, Boelts J, Greenberg D, Goncalves P and Macke J 2021 Benchmarking simulation-based inference Proc. 24th Int. Conf. on Artificial Intelligence and Statistics (AISTATS) vol 130 (PMLR) pp 343–51 (available at: https://proceedings.mlr.press/v130/ lueckmann21a.html) Lueckmann J M, Goncalves P, Bassetto G, Öcal K, Nonnenmacher M and Macke J 2017 Flexible statistical inference for mechanistic models of neural dynamics Advances in Neural Information Processing Systems vol 30 (available at: https://proceedings.neurips.cc/ paper_files/paper/2017/hash/addfa9b7e234254d26e9c7f2af1005cb-Abstract.html) Mayr C, Hoeppner S and Furber S 2019 SpiNNaker 2: a 10 million core processor system for brain simulation and machine learning (arXiv:1911.02385) Müller E, Mauch C, Spilger P, Breitwieser O J, Klähn J, Stöckel D, Wunderlich T and Schemmel J 2020 Extending BrainScaleS OS for BrainScaleS-2 Technical Report (Electronic Vision(s), Kirchhoff Institute for Physics, Heidelberg University) (arXiv:2003.13750) Papamakarios G and Murray I 2016 Fast ε-free inference of simulation models with Bayesian conditional density estimation Advances in Neural Information Processing Systems vol 29 (Curran Associates Inc.) pp 1036–44 Papamakarios G, Nalisnick E, Rezende D J, Mohamed S and Lakshminarayanan B 2021 Normalizing flows for probabilistic modeling and inference J. Mach. Learn. Res. 22 1–64 (available at: https://jmlr.org/papers/v22/19-1028.html) Papamakarios G, Pavlakou T and Murray I 2017 Masked autoregressive flow for density estimation Advances in Neural Information Processing Systems vol 30 (available at: https://proceedings.neurips.cc/paper_files/paper/2016/hash/6aca97005c68f12068238 15f66102863-Abstract.html) Papamakarios G, Sterratt D and Murray I 2019 Sequential neural likelihood: fast likelihood-free inference with autoregressive flows Proc. 22nd Int. Conf. on Artificial Intelligence and Statistics (Proc. Machine Learning Research) vol 89 (PMLR) pp 837–48 (available at: https://proceedings.mlr.press/v89/papamakarios19a.html) Pehle C, Billaudelle S, Cramer B, Kaiser J, Schreiber K, Stradmann Y, Weis J, Leibfried A, Müller E and Schemmel J 2022 The BrainScaleS-2 accelerated neuromorphic system with hybrid plasticity Front. Neurosci. 16 795876 Pehle C, Blessing L, Arnold E, Müller E and Schemmel J 2023 Event-based backpropagation for analog neuromorphic hardware (arXiv:2302.07141) Rall W 1962 Electrophysiology of a dendritic neuron model Biophys. J. 2 145 Sisson S A, Fan Y and Beaumont M 2018 Handbook of Approximate Bayesian Computation (CRC Press) (available at: https://books. google.de/books/about/Handbook_of_Approximate_Bayesian_Computa.html?id=9QhpDwAAQBAJ& source=kp_book_description&redir_esc=y) Tejero-Cantero A, Boelts J, Deistler M, Lueckmann J-M, Durkan C, Gonçalves P J, Greenberg D S and Macke J H 2020 SBI: a toolkit for simulation-based inference J. Open Source Softw. 5 2505 Van Geit W, De Schutter E and Achard P 2008 Automated neuron model optimization techniques: a review Biol. Cybern. 99 241–51 Vanier M C and Bower J M 1999 A comparative survey of automated parameter-search methods for compartmental neural models J. Comput. Neurosci. 7 149–71 Wunderlich T et al 2019 Demonstrating advantages of neuromorphic computation: a pilot study Front. Neurosci. 13 260 18
null
10.1103_physrevresearch.5.013045.pdf
null
null
PHYSICAL REVIEW RESEARCH 5, 013045 (2023) Superconductivity of non-Fermi liquids described by Sachdev-Ye-Kitaev models Chenyuan Li , Subir Sachdev , and Darshan G. Joshi Department of Physics, Harvard University, Cambridge, Massachusetts 02138, USA (Received 16 August 2022; revised 9 November 2022; accepted 20 December 2022; published 25 January 2023) We investigate models of electrons in the Sachdev-Ye-Kitaev class with random and all-to-all electron hop- ping, electron spin exchange, and Cooper-pair hopping. An attractive on-site interaction between electrons leads to superconductivity at low temperatures. Depending on the relative strengths of the hopping and spin exchange, the normal state at the critical temperature is either a Fermi-liquid or a non-Fermi liquid. We present a large-M [where spin symmetry is enlarged to SU(M )] study of the normal state to superconductor phase transition. We describe the transition temperature, the superconducting order parameter, and the electron spectral functions. We contrast between Fermi liquid and non-Fermi liquid normal states: we find that for weaker attractive on-site interaction there is a relative enhancement of Tc when the normal state is a non-Fermi liquid, and correspondingly a strong deviation from BCS limit. Also, the phase transition in this case becomes a first-order transition for strong non-Fermi liquids. On the other hand, for stronger on-site interaction, there is no appreciable difference in Tc between whether the superconductivity emerges from a Fermi liquid or a non-Fermi liquid. Notable features of superconductivity emerging from a non-Fermi liquid are that the superconducting electron spectral function is different from the Fermi-liquid case, with additional peaks at higher energies, and there is no Hebel-Slichter peak in the NMR relaxation rate in the non-Fermi liquid case. DOI: 10.1103/PhysRevResearch.5.013045 I. INTRODUCTION The classic BCS theory provides a highly successful de- scription of the onset of superconductivity (SC) from a Fermi liquid (FL). However, in modern correlated electron materials, the normal state at the onset of higher temperature super- conductivity is usually not a Fermi liquid. Below the critical temperature, basic aspects of the BCS superconducting state [such as the breaking of U(1) gauge symmetry by an electron pair condensate] continue to hold, but numerous quantitative details on the critical temperature, superconducting gap am- plitude, and electron spectral function are not described by BCS theory. A popular class of theories for the onset of superconduc- tivity from a non-Fermi liquid (NFL) focus on a normal state which has a Fermi surface coupled to a critical boson [1–6]. The boson could represent a symmetry breaking order pa- rameter at a quantum critical point, or an emergent excitation associated with spin liquid physics. This critical boson plays a dual role—it leads to the breakdown of quasiparticles in the normal state, and it also leads to superconductivity at low temperature (T ) by inducing pairing between the underlying electrons. The precise manner in which the non-Fermi liquid gives way to superconductivity at low T is not well under- stood, and remains a topic of great interest. Published by the American Physical Society under the terms of the Creative Commons Attribution 4.0 International license. Further distribution of this work must maintain attribution to the author(s) and the published article’s title, journal citation, and DOI. In this paper, we will address the interplay between the non-Fermi liquid and superconductivity using a different class of simpler and more tractable models. These models do not have much spatial structure because of the presence of all-to- all hopping and interactions. However, they have the virtue of being exactly solvable, and so can describe the compe- tition between the different energy scales in a quantitative manner. We consider the Sachdev-Ye-Kitaev (SYK) type of models [7,8], which are a rare class of solvable models leading to non-Fermi liquid phases [9]. Models in this class have been recently studied in different contexts of strongly correlated systems. In this work, we consider a model of electrons with an attractive on-site interaction. In the spirit of SYK models, we consider random and all-to-all hopping, exchange inter- action, and Cooper-pair hopping. This model was previously considered by us and for weak interaction an anomalous metal phase (or a Bose metal) was shown to exist [10] in the prox- imity of superconducting phase. In this work, our focus is on the superconducting phase, and the associated thermal phase transition. Depending on the relative strength of the hopping amplitude and exchange interaction, the normal state at higher temperatures is either a FL or a NFL. Thus our model allows us to systematically investigate the emergence of supercon- ductivity by continuously tuning between FL and NFL normal states. Moreover, we show that SC emerging from a NFL has certain unique features in the spectral function that are absent in the case of a FL-SC transition. There have been previous studies of superconductivity in SYK models [11–19]. However, our model is distinct from the previously considered models. In our model in Eq. (18), we start with a SU(2) spin symmetry [see HJ in Eq. (20)], just as in the original Sachdev-Ye (SY) model [7]. In previous 2643-1564/2023/5(1)/013045(14) 013045-1 Published by the American Physical Society LI, SACHDEV, AND JOSHI PHYSICAL REVIEW RESEARCH 5, 013045 (2023) models the random and all-to-all SYK term is in general not SU(2) symmetric: in Refs. [11,12] a general Hamiltonian of two coupled SYK models is considered, which has a SU(2) symmetry only at a special point (α = 1/4 in the notation used in Ref. [11]), and it corresponds to the zero hopping limit with U = t = L = 0 in our model. However, it is shown in Refs. [11,12] that at this SU(2) symmetric point there is no superconductivity, which is consistent with our results. Refer- ence [14] also examined models without any hopping, but did examine finite N corrections. The models of Refs. [16,17] are related to the one examined here, but with lattice rather than random matrix hopping: the lattice dispersions and all-to-all random hopping for electrons lead to equations with similar solutions [9]. Because of the simpler form of our equations, we are able to present spectral functions within the supercon- ducting phase across the full range of the crossover between the FL and NFL cases. The plan of the paper is as follows. In Sec. II, we first study SC in a simple model of attractive Hubbard model with random and all-to-all hopping. Then we introduce our model in Sec. III and discuss the saddle-point equations. These equa- tions are solved to obtain the normal state and SC solutions in Sec. IV. Therein we discuss several observables. Finally we conclude in Sec. V. Technical details are provided in Appendices. II. RANDOM MATRIX BOGOLIUBOV-DE GENNES THEORY Before we dive into the actual model and its detailed analysis, let us first consider a simpler case. In this section, we present a BCS theory of superconductivity for a Hub- bard model with attractive on-site interaction U along with a random and all-to-all hopping. Our main purpose here is to introduce the formalism in a more familiar setting. Curiously, the spectral functions in the superconducting state in this simple model do not appear to have been obtained earlier, although there have been results for other quantities for finite N [20,21]. We consider a model of electrons ciα, with i = 1, . . . , N a site index, and α = 1, . . . , M a USp(M) index. We have thus enlarged the usual SU(2) spin symmetry. The USp(M) group, M even, is defined by the set of M × M unitary matrices U such that U T J U = J , where 1 ⎛ −1 ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎝ Jαβ = J αβ = 1 −1 . . . . . . (1) (2) ⎞ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎠ is the generalization of the ε tensor to M > 2. It is clear that USp(M ) ⊂ SU(M) for M > 2, while USp(2) ∼= SU(2). We will consider SYK-like models on N sites with USp(M ) symmetry, and take the N → ∞ limit followed by the M → ∞ limit. We don’t expect the large M limit to significantly modify the results, as discussed in Ref. [9]; the large N limit is more significant, and there can additional phases at finite N, as discussed in Refs. [14,18]. We shall calculate the electron spectral density using a set of saddle-point equations, which we derive below. We consider an attractive Hubbard model on a random hopping matrix with the Hamiltonian, HtU = − 1√ N (cid:9) iαcα c† j ti j (cid:8) i< j + c† jαcα i (cid:11) (cid:8) (cid:10) + i −μc† iαcα i + U 2M |J αβc† iαc† iβ |2 (cid:12) , (3) where ti j is a real random number with zero mean and root-mean-square value t, N is the number of sites, μ is the chemical potential and U < 0 is the attractive on-site interaction. In terms of the electron annihilation (creation) operator, cα (c† α ), the number operator nα = c† αcα. We perform a disorder average to obtain the following action: (cid:11) (cid:14) (cid:15) − μ S = (cid:13) (cid:8) i + t 2 2N iα (τ ) c† ⎡ dτ (cid:13) dτ dτ (cid:7) ⎣ ∂ ∂τ (cid:18) (cid:18) (cid:18) (cid:18) (cid:18) (cid:8) i iα (τ )c c† i (τ ) + U cα 2M (cid:18) (cid:18) (cid:18) (cid:18) (cid:8) (cid:18) (cid:18) (cid:18) (cid:18) (cid:18) (cid:18) i (τ (cid:7)) − β 2 i |J αβc† iα (τ )c† (cid:12) iβ (τ )|2 ⎤ 2 (cid:18) (cid:18) (cid:18) (cid:18) (cid:18) iα (τ )c† c† iβ (τ (cid:7)) ⎦, (4) where τ is the imaginary time. Note that we have ignored here the replica indices as they are not significant for the present discussion. Next, we proceed by the G-(cid:7) method used for SYK models. We introduce the normal and anomalous Green’s functions G and F , respectively, as well as the normal and anomalous self-energies (cid:7) and (cid:8), respectively. We can then write the path integral as (cid:13) ZtU = DGDF D(cid:7)D(cid:8)Dc exp(−S0 − S1), (5) 013045-2 SUPERCONDUCTIVITY OF NON-FERMI LIQUIDS … PHYSICAL REVIEW RESEARCH 5, 013045 (2023) where, initially, the role of the self-energies is to impose delta functions which define the Green’s functions as two-point fermion correlators. Let us now look at the two contributions in the action. First we have (cid:14) (cid:15) (cid:22) (cid:21) (cid:13) (cid:13) iα (τ ) c† − μ cα i (τ ) + dτ dτ (cid:7)(cid:7)(τ, τ (cid:7)) (cid:8) i (τ (cid:7)) − NMG(τ (cid:7), τ ) iα (τ )cα c† (cid:22) i iα (τ )c† c† iβ (τ (cid:7)) + NMF ∗(τ, τ (cid:7)) (cid:22) ciα (τ )c β i (τ (cid:7)) − NMF (τ, τ (cid:7)) . (6) S0 = dτ (cid:13) (cid:13) + − (cid:8) i ∂ ∂τ (cid:21) dτ dτ (cid:7) (cid:8)(τ, τ (cid:7)) 2 dτ dτ (cid:7) (cid:8)∗(τ, τ (cid:7)) 2 J αβ (cid:21) Jαβ (cid:8) i (cid:8) i For the interaction terms in (4), we need to introduce additional Hubbard-Stratonovich terms which decouple the quartic fermion interactions, and then use the large M limit to replace these fields by their saddle-point values. This procedure has been carried out explicitly for a related model in Ref. [22], and we do not display the intermediate steps here. Assuming the saddle-point has USp(M) symmetry, we can obtain the final answer more directly simply by the following identifications in the interaction terms: α (τ )cβ (τ (cid:7)) ⇒ δβ c† α G(τ (cid:7), τ ), cα (τ )cβ (τ (cid:7)) ⇒ −J αβF (τ, τ (cid:7)). In this manner, we obtain the second contribution in the action, = U 2 dτ |F (τ, τ )|2 + t 2 2 S1 NM (cid:13) (cid:13) dτ dτ (cid:7)[G(τ, τ (cid:7))G(τ (cid:7), τ ) − F (τ, τ (cid:7))F ∗(τ (cid:7), τ )]. Now we take the variational derivative of the action with respect to G and F ∗, and obtain the saddle-point equations, (cid:7)(τ, τ (cid:7)) = t 2G(τ, τ (cid:7)), (cid:8)(τ, τ (cid:7)) = −U F (τ, τ )δ(τ − τ (cid:7)) + t 2F (τ, τ (cid:7)). (7) (8) (9) These equations have to be supplemented by the Dyson equations obtained from the single-site action for the fermions, which follows from the first 2 spin components of the action S0, (cid:8) (cid:24)(cid:23) (cid:24) (cid:23) Sc = T (c† ↑(iω), c↓(−iω)) ω −iω − μ + (cid:7)(iω) (cid:8)∗(iω) (cid:8)(iω) −iω + μ − (cid:7)(−iω) c↑(iω) ↓(−iω) c† , where T is the temperature. We can now write down the combined saddle point equations: G(cid:7) (iω) ≡ 1 iω + μ − (cid:7)(iω) , (cid:7)(iω) = t 2G(iω) = t 2 [G(cid:7) (−iω)]−1 |(cid:8)(iω)|2 + [G(cid:7) (iω)G(cid:7) (−iω)]−1 , (cid:11) = −U T (cid:8) ω (cid:8)(iω) |(cid:8)(iω)|2 + [G(cid:7) (iω)G(cid:7) (−iω)]−1 , F (iω) = (cid:8)(iω) |(cid:8)(iω)|2 + [G(cid:7) (iω)G(cid:7) (−iω)]−1 , (cid:8)(iω) = (cid:11) + t 2F (iω). (10) (11) The normal and anomalous Green’s function in the super- conducting state are G(iω) and F (iω) along the Matsubara frequency axis, while G(cid:7) (iω) is an intermediate quantity de- fined for notational convenience; G(iω) = G(cid:7) (iω) only in the normal state where (cid:11) = F (iω) = 0. It is useful to first solve these equations in the normal state solution by setting (cid:11) = F (iω) = 0, which yields for μ < 2t G(iω) ≡ G0(iω) = iω + μ 2t 2 4t 2 + (ω − iμ)2, sgn(ω) 2t 2 (12) where the sign in front of the square-root is discontinuous across the real frequency axis, and is chosen so that G0(z) ∼ 1/z as |z| → ∞. This yields the expected semicircle density of states. − i (cid:25) Next, we can linearize Eqs. (11) in (cid:11) at T > 0 and so obtain the superconducting critical temperature Tc. We find the condition 1 = −U T (cid:8) ωn G0(iωn)G0(−iωn) 1 − t 2G0(iωn)G0(−iωn) , (13) with ωn a Matsubara frequency. At small |ωn| we obtain from Eq. (12) that t 2G0(iωn)G0(−iωn) = 1 − 2|ωn| (cid:25) 4t 2 − μ2 + O (cid:9) ω2 n (cid:10) . (14) We can now observe that the denominator in Eq. (13) has a singularity at ωn = 0, which yields the BCS log divergence. 013045-3 LI, SACHDEV, AND JOSHI PHYSICAL REVIEW RESEARCH 5, 013045 (2023) (a) (c) (b) (d) FIG. 1. (a) The spectral function, A(ω), of the normal Green’s function in the SC phase at a fixed (cid:11) for the particle-hole symmetric case (μ = 0). The solid line is exact solution to the saddle point equations (11), and the yellow bars are obtained by averaging exact diagonalizations of random instances of Eq. (3). (b) Same as (a) but μ = 5. (c) Imaginary part of the anomalous Green’s function F (ω) in the SC phase at a fixed (cid:11) and μ = 0. (d) Same as (c) with μ = 5. In all the plots, t = 10 and (cid:11) = 5. This implies that there is superconductivity at T = 0 for in- finitesimal negative U . We can analytically solve Eqs. (11) at T = 0 to linear order in (cid:11) for general μ. Such a solution will be valid for (cid:25) |ω|, 4t 2 − μ2 (cid:13) (cid:11). We find F (iω) = (cid:11) ( (cid:25) 4t 2 + (ω − iμ)2 + (cid:25) 4t 2 + (ω + iμ)2 − 2|ω|) 4|ω| + O((cid:11)3), G(iω) = G0(iω) + O((cid:11)2). (15) Note that F (iω) is a real and even function of ω along the imaginary frequency axis. However, neither F nor G are ana- lytic at ω = 0. Similarly, we can see that G(−iω) = G∗(iω), and for μ = 0 G(iω) is purely imaginary, with G(−iω) = −G(iω). At μ = 0, the exact solution of the saddle-point equa- tions in (11) is G(iω) = − iω 2t 2 F (iω) = (cid:11) 2t 2 (cid:23) √ ω2 + 4t 2 + (cid:11)2 √ ω2 + (cid:11)2 (cid:24) − 1 , (cid:23) √ ω2 + 4t 2 + (cid:11)2 √ ω2 + (cid:11)2 (cid:24) − 1 . Analytic continuation gives the spectral function, A(ω) ≡ − 1 π ImG(ω + iδ), A(ω) = √ |ω| 2πt 2 4t 2 + (cid:11)2 − ω2 √ ω2 − (cid:11)2 , (cid:11) < |ω| < (cid:25) (cid:11)2 + 4t 2. (17) The spectral function is plotted in Fig. 1(a), along with the numerical results obtained by exact diagonalization of random realizations of the Hamiltonian in Eq. (3). As expected, the gap is centered at ω = 0, between (cid:11) and −(cid:11). It is also straightforward to obtain the imaginary part of the retarded anomalous Green’s function, which is shown in Fig. 1(c). For μ (cid:14)= 0 an analytic solution is no longer possible, and we show numerical results in Figs. 1(b) and 1(d). III. MODEL Having discussed the basic set-up we are now ready to discuss our model. To the random Hubbard model considered in the previous section, we will now add random and all-to-all spin exchange and Cooper-pair hopping terms. So the full Hamiltonian is (16) 013045-4 H = HtU + HJ + HL, (18) SUPERCONDUCTIVITY OF NON-FERMI LIQUIDS … PHYSICAL REVIEW RESEARCH 5, 013045 (2023) (cid:9) iαcα c† j + c† jαcα i (cid:10) (cid:8) ti j HtU = − 1√ N (cid:11) (cid:8) −μc† i< j + iαcα + U 2M i (cid:8) |J αβc† iαc† iβ |2 Ji j c† iαc β i c† jβ cα j , (cid:12) , (19) (20) i HJ = 1√ (cid:8) NM i< j HL = − 1 √ 2 NM Li jJ αβJγ δ (cid:26) iαc† c† iβ c γ j cδ j + c† jαc† jβ c γ i cδ i (cid:27) . i< j (21) Recall that we have solved HtU in Sec. II. HJ describes the exchange interaction of the original SY model [7], while HL describes the random Cooper-pair hopping. In the above Hamiltonian, Ji j are real random numbers with zero mean value and root-mean-square value of J. Similarly, Li j can be either real or complex random numbers with zero mean value and root-mean-square value of L. For clarity, let us consider the contribution of individual terms in the Hamiltonian in Eq. (18). The first term, HtU , in Eq. (19) was already dealt with in Sec. II. Next, let us consider the contribution of HJ in Eq. (20) to the action of the full Hamiltonian. After averaging over Gaussian random variable Ji j the resulting action is S J = − J 2 4NM (cid:13) (cid:18) (cid:18) (cid:18) (cid:18) (cid:18) dτ dτ (cid:7) i (cid:8) iα (τ )c c† β i (τ )c† iγ (τ (cid:7))cδ i (τ (cid:7)) (cid:18) (cid:18) (cid:18) (cid:18) (cid:18) 2 . (22) In the large M limit, we can use an identity analogous to Eq. (7), α (τ )cβ (τ )c† c† γ (τ (cid:7))cδ (τ (cid:7)) ⇒ δδ αδβ γ G(τ, τ (cid:7))G(τ (cid:7), τ ) + J βδJαγ F ∗(τ, τ (cid:7))F (τ, τ (cid:7)). (23) Here we have dropped factorizations associated with equal-time Green’s functions. Then the contribution to the action from the HJ term is SJ with (cid:13) SJ NM = − J 2 4 dτ dτ (cid:7)([G(τ, τ (cid:7))G(τ (cid:7), τ )]2 + |F (τ, τ (cid:7))F (τ (cid:7), τ )|2). (24) Finally, let us consider the contribution from the random Cooper-pair hopping term, HL, in Eq. (21). Averaging over real Gaussian random variable Li j yields the action (cid:13) S L = − L2 8NM dτ dτ (cid:7)J αβJ μνJγ δJρσ (cid:23) (cid:8) ⎡ ⎣ iα (τ )c† c† iβ (τ )c ρ i (τ (cid:7))cσ i (τ (cid:7)) (cid:24)⎛ ⎝ (cid:8) jμ(τ (cid:7))c† c† jν (τ (cid:7))c γ j (τ )cδ j (τ ) (cid:23) (cid:8) + i iα (τ )c† c† iβ (τ )c† iμ(τ (cid:7))c† iν (τ (cid:7)) i (cid:8) (cid:24)⎛ ⎝ j ⎞ ⎤ ⎠ ⎦. ρ j (τ (cid:7))cσ j (τ (cid:7)) γ j (τ )cδ j (τ )c c j ⎞ ⎠ (25) Note that the last term would be absent for complex Li j. Now, we use large M identities similar to Eqs. (7) and (23), again dropping equal-time factorizations, (cid:9) α δρ δσ β (τ )cρ (τ (cid:7))cσ (τ (cid:7)) ⇒ ν (τ (cid:7)) ⇒ (JανJβμ − JαμJβν )[F ∗(τ, τ (cid:7))]2. μ(τ (cid:7))c† α (τ )c† c† α (τ )c† c† β (τ )c† α δσ β (26) (cid:10) [G(τ, τ (cid:7))]2, The contribution of the HL term to the action is SL with β − δρ dτ dτ (cid:7)([G(τ, τ (cid:7))G(τ (cid:7), τ )]2 + |F (τ, τ (cid:7))F (τ (cid:7), τ )|2), (27) SL NM = − L2 4 having the same form as SJ in Eq. (24). (cid:13) So finally, the action corresponding to the full Hamiltonian in Eq. (18) is S = S0 + S1 + SJ + SL, (28) with the terms S0 and S1 quoted in Eqs. (6) and (8), respec- tively, while the terms SJ and SL are shown in Eqs. (24) and (27), respectively. Putting everything together, the final saddle-point equa- tions for the normal and anomalous equations are G(cid:7) (iω) ≡ 1 iω + μ − (cid:7)(iω) , (29) (cid:7)(τ, τ (cid:7)) = t 2G(τ, τ (cid:7)) − (J 2 + L2)G2(τ, τ (cid:7))G(τ (cid:7), τ ), (30) (cid:11) = −U T G(iω) = (cid:8) ω F (iω) = , [G(cid:7) (−iω)]−1 |(cid:8)(iω)|2 + [G(cid:7) (iω)G(cid:7) (−iω)]−1 (cid:8)(iω) |(cid:8)(iω)|2 + [G(cid:7) (iω)G(cid:7) (−iω)]−1 (cid:8)(iω) |(cid:8)(iω)|2 + [G(cid:7) (iω)G(cid:7) (−iω)]−1 , (31) , (32) (33) (cid:8)(τ, τ (cid:7)) = − U F (τ, τ )δ(τ − τ (cid:7)) + t 2F (τ, τ (cid:7)) + (J 2 + L2)F 2(τ, τ (cid:7))F ∗(τ (cid:7), τ ). (34) Note that Eqs. (30) and (34) generalize the expressions in Eq. (9) upon the inclusion of the spin exchange and Cooper- pair hopping terms. 013045-5 LI, SACHDEV, AND JOSHI PHYSICAL REVIEW RESEARCH 5, 013045 (2023) (a) (b) FIG. 2. (a) The normal-state spectral function A(ω) for different values of θ at μ = 0, and R/|U | = 2. The dashed line is the exact semicircle solution for θ = 0, as obtained in Eq. (12). (b) Effective spin exponent as a function of θ in the normal state at R/|U | = 2. The spin exponent takes the expected FL value for low θ , while it approaches the SYK value for larger θ. IV. NUMERICAL SOLUTIONS We shall now solve the saddle-point equations [Eqs. (29)– (34)] at finite temperature and obtain the normal-state as well as SC solutions. For simplicity and clarity, we will focus on the μ = 0 half-filling case, but the results are qualitatively similar for nonzero μ as we show at the end of this section. Hence, unless otherwise stated μ = 0 throughout this section. We introduce the notation (cid:28)J = J 2 + L2 since the interac- tions J and L are on equal footing in the large-M limit, as seen from Eqs. (30) and (34). Furthermore, we will parametrize the hopping t and interaction (cid:28)J as √ t = R cos θ , (cid:28)J = R sin θ , (35) (cid:25) t 2 + (cid:28)J 2, and the parameter θ ∈ [0, π /2] tunes where R = between FL (θ = 0) and SYK-NFL (θ = π /2) limits. We will discuss results for different relative strengths with respect to U , i.e., different ratios R/|U |. We solve the saddle-point equations, Eqs. (29)–(34), on the imaginary (Matsubara) frequency axis at finite temperature. The strategy is as follows. We first start with a free fermion normal Green’s function, G(iωn) = (iωn + μ)−1, and a ran- domly chosen real function F (iωn), and iterate until we find a converged solution for the normal and anomalous Green’s functions. The SC order parameter, (cid:11)(T ) = −U Jαβ (cid:16)cαcβ (cid:17), is then determined as a function of temperature. It is finite at low temperatures in the superconducting phase, and it vanishes in the normal state at higher temperature. The su- perconducting critical temperature Tsc is thus determined nu- merically using (cid:11)(T → T − sc ) → 0. We will use the notation (cid:11)0 ≡ (cid:11)(T → 0). In both the normal and SC phases, we also compute the spectral function. The spectral function is obtained by nu- merical analytic continuation of Matsubara Green’s functions to the real frequency axis. More details regarding numerical analytic continuation are discussed in Appendix A. A. Normal state The normal-state equations with (cid:11) = 0 and F = 0 are the same as those in Refs. [23,24]. As stated earlier, in our model we tune the parameter θ , defined in Eq. (35), to go from FL to NFL normal states. At any given temperature T , the normal state is FL like for θ (cid:2) θcoh and NFL-like for θ (cid:3) θcoh, where θcoh is defined by T ∼ Tcoh = t 2/(cid:28)J = R cos θcoh cot θcoh. In Fig. 2(a), we show the spectral function in the normal state. For the FL-like phase (smaller θ ), we see the expected semicircular spectral function, whereas for a NFL-like phase (larger θ ) a pronounced peak at ω = 0 is seen. This is consis- tent with earlier results obtained for a similar random model in Ref. [23]. Also, note that the FL-like normal state (θ < θcoh) has the usual T 2 dependence of resistivity, while the NFL state (θ > θcoh) has a linear-in-T resistivity. This is similar to the results obtained in Refs. [23,24]. The cross-over between the FL and NFL normal states can be further characterized by looking at the effective spin exponent (ηs), which is shown in Fig. 2(b). This exponent is extracted from the dynamical susceptibility, χ (cid:7)(cid:7)(ω), which is the imaginary part of the spin correlation. In Appendix B, we discuss the details related to the evaluation of the spin exponent ηs. Clearly, for lower values of θ the spin exponent takes the value ηs = 2 expected for a disordered FL, while in the limit θ → π /2, it takes the value ηs = 1 corresponding to the marginal NFL. In the intermediate θ region ηs smoothly interpolates between these extreme values. As expected, this crossover is roughly around θcoh. B. Superconducting state Before we discuss the numerical results, we first show analytically that SC phase exists at zero temperature for any infinitesimal attractive on-site interaction. The analysis is sim- ilar to that presented in Sec. II. We determine the instability to the superconducting state by expanding the action to second order in F (iω). This leads to the same condition for the in- stability as Eq. (13). However, the important difference is that the Green’s function now also contains contribution from the exchange interaction terms and satisfies the equations: G0(iω) = 1 iω + μ − (cid:7)el (iω) − (cid:7)in(iω) , (cid:7)el (iω) = t 2G0(iω), (cid:7)in(τ ) = −(J 2 + L2)[G0(τ )]2G0(−τ ). (36) 013045-6 SUPERCONDUCTIVITY OF NON-FERMI LIQUIDS … PHYSICAL REVIEW RESEARCH 5, 013045 (2023) FIG. 3. SC order parameter, (cid:11), as a function of temperature, T . (a) SYK-NFL (θ = π /2) case with varying R/|U |. Note that for larger values of R/|U |, the phase transition becomes first order instead of a continuous transition. (b) Here R = |U | is fixed and θ is varied. Note that we have separated the self-energy into an ‘elastic’ part (cid:7)el , and an ‘inelastic’ part (cid:7)in. This is useful because Im (cid:7)in(ω → 0) = 0 at T = 0, and that is not true for the elastic part. From Eq. (36), we can write a quadratic equation for G0(0): t 2[G0(0)]2 − (μ − (cid:7)in(0))G0 + 1 = 0. An important point is that (cid:7)in(0) is real, and so it can be absorbed into μ. This quadratic equation has two roots, and they correspond to G0(i0+) and G0(i0−). From the formula for the product of the roots of a quadratic equation, we can therefore conclude that at T = 0, (37) lim ω→0 G0(iω)G0(−iω) = 1 t 2 . (38) So this equation holds even when J or L are nonzero, and the denominator in Eq. (13) vanishes. Thus indicating the presence of SC at T = 0. Let us now discuss the numerical results obtained by solv- ing the saddle-point equations. For low enough temperature, we find a SC solution with a nonzero (cid:11) and F (iω). In Fig. 3, we have shown the variation of SC order parameter, (cid:11), with temperature. It turns out that for small values of θ , i.e., FL-like normal state, the SC-normal state transition is continuous. However, at larger values of θ , the phase transition (SC to NFL) becomes first order for larger values of R/|U | [as seen in Fig. 3(a)]. Note that although the absolute value of (cid:11) and Tsc depends on the value of U , the variation of (cid:11)/|U | as a function of T /|U | depends only on the ratio R/|U |. In Fig. 4, we show the variation of SC transition temperature (Tsc) as a function of θ for different values of R/|U |. For very large on- site interaction, i.e., for very small R/|U | there is no difference between SC emerging from FL or NFL. This is because in this case both hopping as well as exchange interaction are subdominant. However, at larger values of the ratio R/|U |, i.e., weaker on-site interaction the SC transition temperature Tsc strongly depends on the nature of the normal state or θ . It is larger for NFL-SC transition (larger θ ) as compared to the FL-SC transition (smaller θ ). The same trend applies to the SC order parameter in the limit of zero temperature, (cid:11)0, and the SC gap (as obtained from the spectral function) in the T → 0 limit, (cid:28)(cid:11)0, as seen in Fig. 5. Recall that in our model SC phase corresponds to the condensation of doublon, i.e., the Cooper pairs are on the same site. A single-particle hopping tends to break these pairs and destroy SC. The exchange interaction and Cooper-pair hopping have a very weak effect in destruction of SC. Therefore, Tsc, (cid:11)0, and (cid:28)(cid:11)0 have very weak dependence on θ for larger on-site interaction (smaller R/|U |), as in this case the relative strength of hopping and spin-exchange is unimportant. On the other hand, for weaker on-site interaction the relative strength of hopping, t, com- pared to (cid:28)J is important. Hence for larger θ (weaker t) SC is more stable leading to a higher Tsc. This is also the reason why the SC-NFL transition becomes first order in nature for larger R/|U |. We have also calculated the ratio 2(cid:11)0/Tsc and 2(cid:28)(cid:11)0/Tsc, which is 3.53 for the BCS superconductivity (for FL-SC there is no difference between (cid:11)0 and (cid:28)(cid:11)0 as discussed below). This is shown in Figs. 5(c) and 5(d). We find that in our case, this ratio approaches the BCS value for smaller θ (FL normal state) and weaker on-site interaction. For SC emerging from NFL normal state this ratio deviates strongly from the BCS value. The value of this ratio first increases with θ as long as the transition is continuous, and then tends to decrease as the transition changes its nature to first order. This trend follows from the observation that the transition temperature increases very sharply for large values of θ and R/|U | compared to the much gradual increase in (cid:11)0. In the FL case (smaller θ ), both (cid:11)0 and Tsc are suppressed exponentially as a function of R/|U | such that their ratio is a constant. However, in the NFL case (larger θ ) this is not true anymore. Both (cid:11)0 and Tsc appear to decrease with different power-laws with respect to R/|U |, and in particular for larger values of θ the transition temperature Tsc saturates quickly for large θ . This is shown in Fig. 13 in Appendix. We have also computed the spectral function for the SC phase. This is shown in Fig. 6. As ex- pected, we clearly see the SC gap in the spectral function. For θ = 0 (FL normal state), we see the expected square-root divergence near ω = (cid:11). The form of this divergence seems to be modified for θ away from zero. In particular, for θ = π /2 (SYK-NFL normal state), we see very narrow peaks. We also note that the SC gap ((cid:28)(cid:11)) observed in the spectral function may not be the same as SC order parameter (cid:11) calculated above, as is shown in Figs. 5(a) and 5(b). The two quantities are same 013045-7 LI, SACHDEV, AND JOSHI PHYSICAL REVIEW RESEARCH 5, 013045 (2023) FIG. 4. (a) The SC transition temperature Tsc as a function of θ for different values of R/|U |. Qualitatively, the phase transition becomes first order (indicated by open circles) at larger values of R/|U | and θ instead of a continuous transition (indicated by filled circles). (b) Comparison of Tsc and Tcoh/3 = t 2/3J = (1/3)R cos θ cot θ at R/|U | = 3. For larger values of R/|U |, the transition becomes first order for θ (cid:3) θcoh. (c) Same as (b) but R/|U | = 5. for SC emerging from FL (smaller θ ), but may deviate from each other for the SC emerging from a NFL phase (larger θ ). In particular, the deviation between (cid:11) and (cid:28)(cid:11) is strongest for larger θ and larger values of R/|U | (where the transition is of first order). In Fig. 7(a), we show the variation of the ratio of these two quantities in the limit of zero temperature, FIG. 5. (a) The variation of SC order parameter in the zero temperature limit, (cid:11)0, with θ for different values of the ratio R/|U |. (b) The SC gap observed in the spectral function in the zero temperature limit, (cid:28)(cid:11)0, as a function of θ. (c) The ratio 2(cid:11)0/Tsc. (d) The ratio 2(cid:28)(cid:11)0/Tsc. For larger values of θ, these ratios deviate strongly away from the BCS value of 3.53. As θ → 0 and R/|U | (cid:13) 1, both 2(cid:11)0/Tsc and 2(cid:28)(cid:11)0/Tsc tend to the BCS result. 013045-8 SUPERCONDUCTIVITY OF NON-FERMI LIQUIDS … PHYSICAL REVIEW RESEARCH 5, 013045 (2023) FIG. 6. The spectral functions in the superconducting phase at R = |U | for different values of θ. i.e., (cid:28)(cid:11)0/(cid:11)0, with respect to θ and R/|U |. We do not have an analytic expression for the gap in the spectral function, (cid:28)(cid:11). But numerically we find that (cid:11)0 + (cid:28)(cid:11)0 (cid:18) |U | at θ = π /2, independent of the ratio R/|U |. This relation does not hold for other values of θ . This is shown in Fig. 7(b). A noticeable new feature for SC emerging from NFL (larger values of θ ) is the presence of peaks at higher energies compared to the SC gap [see Figs. 6(c) and 6(d). In the limit of T → 0 the first higher-order peak appears at ∼3(cid:28)(cid:11). A dom- inant all-to-all exchange interaction (large θ ) means strongly interacting Cooper pairs, which may be the reason for these additional peaks. For smaller values of θ the Cooper pairs are weakly interacting. Note that such high energy features in the spectral function have also been reported for SYK-like electron-phonon model for SC [15]. We have so far focused on the particle-hole symmetric point, μ = 0, for clarity. However, it is straightforward to also do the same analysis for a nonzero chemical potential. The re- sults are qualitatively the same as discussed above. The main difference seen is the particle-hole asymmetric distribution of the spectral weights at positive and negative frequencies, as shown in Fig. 8. The gap is however symmetric around ω = 0. In the remainder of the paper, we again focus only on μ = 0. We further also compute the spin correlation in the SC phase. In Fig. 9, we plot χ (cid:7)(cid:7)(ω) for different values of θ in the SC phase. The features essentially follow from what was discussed for the electron spectral function earlier. The high-energy peak present in the electron spectral function at larger values of θ is also seen in χ (cid:7)(cid:7)(ω). FIG. 7. (a) In the limit of zero temperature, the ratio of the SC gap, (cid:28)(cid:11)0, (as obtained from the spectral function) and the SC order parameter, (cid:11)0, as a function of R/|U | at θ = π /2 (blue) and as a function of θ at R/|U | = 1 (red) is shown. The two quantities are in general different away from the FL limit and for small on-site interaction the deviation between the two quantities is strongest. (b) The sum (cid:28)(cid:11)0 and (cid:11)0 as a function of R/|U | at θ = π /2 (blue) and as a function of θ at R/|U | = 1 (red) is shown. The sum is a constant for θ = π /2. However, this is not the case for other values of θ . 013045-9 LI, SACHDEV, AND JOSHI PHYSICAL REVIEW RESEARCH 5, 013045 (2023) FIG. 8. The spectral functions in the superconducting phase at R = |U |, μ/|U | = 0.2 for different values of θ. Recall that for the standard BCS superconductor one expects a peak (often referred to as the “Hebel-Slichter” peak) around the critical temperature as a consequence of the square-root divergence in the spectral function [25]. However, one of the signatures of the unconventional superconductivity is the absence of Hebel-Slichter peak, for instance, as observed in cuprates [26] and Fe-based superconductors [27]. We find that for a fixed R/|U | when θ (cid:2) θcoh there is a well distinguished Using χ (cid:7)(cid:7)(ω) we can also evaluate the temperature depen- dence of the NMR relaxation rate, 1/T1, which we show in Fig. 10. The NMR relaxation rate is given by the following relation [23]: 1 T1 = T (cid:18) (cid:18) (cid:18) (cid:18) χ (cid:7)(cid:7)(ω) ω . ω=0 0.04 0.03 0.02 0.01 ) ω ( (cid:2) (cid:2) χ (39) θ = 0 0.06 0.04 0.02 ) ω ( (cid:2) (cid:2) χ 0 0 1 2 3 ω/|U | (a) 4 5 0 0 1 0.3 0.2 0.1 ) ω ( (cid:2) (cid:2) χ θ = 3π 8 0 0 1 4 5 2 3 ω/|U | (c) ) ω ( (cid:2) (cid:2) χ 0.8 0.6 0.4 0.2 0 0 θ = π 4 2 3 ω/|U | (b) 4 5 θ = π 2 1 2 3 ω/|U | (d) 4 5 FIG. 9. Plot of imaginary part of spin correlation χ (cid:7)(cid:7)(ω) in the SC phase as a function of real frequency for different values of θ at R = |U | and T = 0.01. 013045-10 SUPERCONDUCTIVITY OF NON-FERMI LIQUIDS … PHYSICAL REVIEW RESEARCH 5, 013045 (2023) 1 T1 1 T1 θ = 0 0.0001 0.00008 0.00006 0.00004 0.00002 0 0 0.005 0.01 T /|U | (a) θ = π 4 1 T1 0.0006 0.0004 0.0002 0 0 0.01 0.04 0.03 0.02 T /|U | (b) 0.005 0.004 0.003 0.002 0.001 0 0 θ = 3π 8 0.08 0.10 0.02 0.04 0.06 T /|U | (c) 1 T1 0.02 0.015 0.01 0.005 0 0 θ = π 2 0.15 0.20 0.05 0.10 T /|U | (d) FIG. 10. Temperature dependence of the NMR relaxation rate, 1/T1, for different values of θ at R = 2|U |. Note that for smaller values of θ , where the normal state is FL-like, there is a Hebel-Slichter peak around the SC transition temperature. This can be seen in (a) around T /|U | ∼ 0.006 and in (b) around T /|U | ∼ 0.018, which have FL normal state. Note that the peak height diminishes as we increase θ and go closer to NFL case. The Hebel-Slichter peak is absent (and replaced by a kink) in case of NFL normal state, shown in (c) and (d). Hebel-Slichter peak whose strength diminishes with increas- ing θ [see Figs. 10(a) and 10(b)]. After the crossover into the NFL regime for larger θ the peak is absent and there is only a kink around the critical temperature [see Figs. 10(c) and 10(d)]. This is another distinguishing feature between the FL and NFL cases. We also see that the relaxation rate is higher for the NFL case compared to the FL case. In the normal state this trend easily follows from the fact that the critical temperature is much higher in the NFL case. Also note that the height of the Hebel-Slichter peak present for smaller θ is roughly inversely proportional to R/|U |, i.e., for smaller Hubbard interaction the peak is smaller. V. DISCUSSION We have investigated the emergence of SC in a SYK-like model of interacting electrons, Eq. (18). The model is solved in the large-M limit, where we generalize the spin symmetry from SU(2) to SU(M ). The solution of the large-M saddle- point equations can be viewed as a dynamical mean-field solution. We have shown the contrast between the emergence of SC from a NFL as opposed to a FL normal state. Several distinguishing features are found for SC emerging from a NFL and we summarize below the salient features of our work. (1) Even in the presence of all-to-all and random exchange interaction and Cooper-pair hopping, we show that BCS- type superconducting instability is present, thus ensuring SC ground state at zero temperature for any infinitesimal attrac- tive Hubbard interaction. (2) The SC transition temperature Tsc is shown to be strongly enhanced for NFL normal state as compared to a FL normal state. This is an important highlight of our results. This is understood physically by realizing that the most dominant mechanism to break Cooper pairs is single- particle hopping. However, for the NFL case, the Cooper pairs are strongly interacting, and single-particle hopping is sub- dominant, thus leading to a higher Tsc. This also renders the transition in case of NFL to be first order for weaker Hubbard interaction. (3) While for the FL (BCS-like) case both Tsc and (cid:11) are exponentially suppressed with respect to R/|U |, we show that for NFL case they decay with different power-laws. Conse- quently, the ratio 2(cid:11)/Tsc strongly deviates from the BCS value for SC arising from NFL. (4) We have presented a detailed study of the local electron spectral function in the SC as well as the normal states (FL and NFL). This is an observable in photoemission experi- ments like ARPES. We discuss how the SC gap closes upon approaching Tsc. In the case of a FL normal state, the transi- tion is continuous and BCS like, and the spectral function in SC phase features the well-known square-root divergence at ω = (cid:11). We show that this is not the case when the normal state is a NFL. (5) We show that for SC emerging from a NFL there is a distinct new feature in the local electron spectral function— peaks at higher energy at ω ∼ 3(cid:11). This is a consequence of strong interactions between Cooper pairs (which is absent in case of FL normal state). We believe that this is a generic feature of SC emerging from a NFL, and could be a relevant observation in many materials. How generic and model in- dependent is this feature is an interesting open question for future. (6) In the normal state, as a function of the parameter θ , there is a crossover between FL and NFL phase for a fixed 013045-11 LI, SACHDEV, AND JOSHI PHYSICAL REVIEW RESEARCH 5, 013045 (2023) 0.04 0.03 0.02 0.01 ) ω ( (cid:2) (cid:2) χ θ = 0 θ = 0.11π θ = 0.22π θ = 0.33π θ = 0.44π 0.04 0.03 0.02 0.01 ) ω ( (cid:2) (cid:2) χ T /|U | = 0.13 T /|U | = 0.135 T /|U | = 0.14 T /|U | = 0.145 T /|U | = 0.15 0 0 1 2 3 4 ω/|U | (a) 5 6 7 8 0 0 1 2 3 5 6 7 8 4 ω/|U | (b) FIG. 11. (a) Plot of χ (cid:7)(cid:7)(ω) for different values of θ at a fixed temperature T /|U | = 0.13 in the normal state. (b) Plot of χ (cid:7)(cid:7)(ω) for different values of temperature at a fixed value of θ = 0.388π . In both the plots, R/|U | = 2. temperature, which we characterize using the effective local spin correlation exponent, ηS. The exponent deviates from FL value for θ (cid:3) θcoh. We hope that our work motivates the ob- servation of this exponent in neutron scattering experiments. (7) We also note that NFL phase, i.e., the normal state for θ > θcoh has a linear-in-temperature resistivity. Thus SC emerging from this state may have some relevance to the situation in correlated systems. (8) We have evaluated the local dynamic structure factor, χ (cid:7)(cid:7)(ω), an observable in neutron scattering experiments. In the SC phase emerging from NFL, χ (cid:7)(cid:7)(ω) shows distinct peaks at high energies akin to that discussed for the spectral function. (9) Further we have also calculated the NMR relaxation rate, 1/T1, as a function of temperature. Here we show that for FL normal state there is a Hebel-Slichter peak near Tsc, which is a hallmark of BCS SC. However, for NFL case, this peak disappears and the transition temperature is marked by a kink. Such observations have been reported in experiments on unconventional SC in cuprates and pnictides. Our work clearly shows the mechanism for the disappearance of the Hebel- Slichter peak in the case of a NFL normal state. This may be of general relevance to the NMR experiments in unconventional SC materials. We believe that our work will further motivate and pro- vide a pathway to investigate SC emerging from NFL. Our work also motivates numerical investigation of the model in Eq. (18) at M = 2 to further elucidate the SC-NFL phase transition. We hope that our work may also provide a good starting point for constructing more realistic lattice models. While this work was being completed, we learnt of the study Ref. [18] of essentially the same model, but with a focus on the finite N behavior. ACKNOWLEDGMENTS We thank G. Tarnopolsky for valuable discussions. This research was supported by the National Science Foundation under Grant No. DMR-2002850. This work was also sup- ported by the Simons Collaboration on Ultra-Quantum Matter, which is a grant from the Simons Foundation (651440, S.S.). D.G.J. acknowledges support from the Leopoldina fellowship by the German National Academy of Sciences through Grant No. LPDS 2020-01. APPENDIX A: NUMERICAL ANALYTIC CONTINUATION We also perform numerical analytic continuation to real frequency. In general, performing analytic continuation is an ill-posed problem if the function on the imaginary axis is known only at a finite number of points. There are several techniques to do analytic continuation. However, for simplic- ity, we use the Pade approximation method. This technique parametrizes the function on imaginary axis as a ratio of two polynomials or by terminating a continued fraction. There are several ways for implementing Pade approximation. We adopt the simple strategy outlined in Ref. [28] of evaluating the coefficients of the two polynomials recursively, which is based on Thiele’s reciprocal difference method. Details of the algorithm can be found in the Appendix of Ref. [28]. Briefly, we first solve the saddle-point equations on the imaginary- frequency axis to obtain the required Green’s function, say G(iω), at non-negative Matsubara frequencies. The number of Matsubara frequencies used in our calculation is 105. Then we evaluate the required polynomials, An(z) and Bn(z), to approx- imate the imaginary-frequency function, G(z) = An(z)/Bn(z). The accuracy of these polynomials depends on the number of Pade points, n, and in our calculation we find that n = 200 points are sufficient to obtain accurate results. We have checked our results by increasing or decreasing n and it does not result in any significant improvement. The resulting ra- tio of polynomials then corresponds to the retarded Green’s function on real-frequency axis, once we identify z = ω + −4 −5 −6 −7 ) 0 ω / ) 0 ω ( (cid:2) (cid:2) χ ( g o l θ = 0 θ = 0.055π θ = 0.111π θ = 0.166π θ = 0.222π θ = 0.277π θ = 0.333π θ = 0.388π θ = 0.444π θ = 0.50π fit 0.25 0.30 0.35 0.40 log(T ) FIG. 12. Plot of ln(χ (cid:7)(cid:7)(ω0)/ω0) vs ln(T ) at ω0 = 0.2, R/|U | = 2, and in the temperature range between T /|U | = 0.13 and T /|U | = 0.15, for different values of θ. The slope of the linear fit (black dashed lines) gives ηs − 2, from Eq. (B3). This is used to plot the curve of ηs as a function of θ in Fig. 2(b). 013045-12 SUPERCONDUCTIVITY OF NON-FERMI LIQUIDS … PHYSICAL REVIEW RESEARCH 5, 013045 (2023) FIG. 13. Plots of transition temperature Tsc, the SC order parameter (cid:11)0 and the SC gap (cid:28)(cid:11)0 vs R for different values of θ. Note that for smaller values of θ when the normal state is FL-like, there is an exponential decay with respect to R/|U |. In contrast, in the case of NFL normal state these are replaced by different power laws. i0+. Imaginary part of this function then gives the spectral function. APPENDIX B: EFFECTIVE SPIN EXPONENT In this Appendix, we discuss the evaluation of the effec- tive spin exponent (ηs) in the normal state. To start with, we first evaluate the spin correlation, χ (τ ) = (cid:16)(cid:19)S(τ ) · (cid:19)S(0)(cid:17) ∼ −G(τ )G(−τ ), which is straightforward to obtain from the imaginary frequency numerics. We then Fourier transform to obtain χ (iω), and then perform numerical analytic continu- ation to obtain χ (ω) whose imaginary part is the dynamical susceptibility, χ (cid:7)(cid:7)(ω). This is shown in Fig. 11 for the normal state and in Fig. 9 in the SC phase. At temperature above the SC transition temperature, the normal state solution is one of the SYK-type conformal so- lutions at low energy (ω (cid:20) ˜J). For such a solution, the spin susceptibility follows the scaling relation [23,29,30], χ (cid:7)(cid:7)(ω) ∼ T ηs−1(cid:8)ηs (cid:14) (cid:15) , ¯hω kBT where (cid:8)ηs (y) = sinh (cid:29) (cid:30)(cid:18) (cid:18) (cid:18)(cid:20) (cid:29) ηs 2 y 2 + i y 2π (cid:30)(cid:18) (cid:18) (cid:18) 2 . For ¯hω (cid:20) kBT , we have χ (cid:7)(cid:7)(ω) ∼ ω T ηs−2, (B1) (B2) (B3) while in the limit of ¯hω (cid:13) kBT , the result is similar to the zero temperature form, χ (cid:7)(cid:7)(ω) ∼ sgn (ω)|ω|ηs−1. We can thus use Eq. (B3) to extract the effective spin exponent (ηs) from the slope of plot of ln(χ (cid:7)(cid:7)(ω0)/ω0) versus ln(T ), where ω0 is a fixed small frequency. In Fig. 12, we present the data for such a procedure for R/|U | = 2 in the temperature range of T /|U | = 0.13 and T /|U | = 0.15, with ω0 = 0.2. We have also checked our results for two other small frequency points, and the results are unchanged. The resulting ηs as a function of θ is plotted in Fig. 2(b). Similar procedure can be done at other values of R/|U |. This works well for larger values of R/|U |. At smaller values of R/|U |, the transition temperature is relatively high, where our numerical analytic continuation is not very reliable, and so extracting ηs there is difficult. APPENDIX C: ADDITIONAL PLOTS FOR Tsc AND (cid:2) As noted earlier in the main text, the SC transition temperature (Tsc) and the SC order parameter ((cid:11)0) decay exponentially with R/|U | in the case of SC emerging from FL normal state. This results in the expected BCS value for the ratio of Tsc and (cid:11)0. However, we observe that in the case when SC emerges from a NFL normal state the transition tempera- ture and the order parameter decay with different power-laws with respect to R/|U |. This is shown in Fig. 13. Deducing these power-laws analytically is an interesting problem for future work. The consequence of these power-law decay in place of an exponential is that the ratio of Tsc and (cid:11)0 is no longer a constant. [1] S.-S. Zhang, Y.-M. Wu, A. Abanov, and A. V. Chubukov, In- terplay between superconductivity and non-Fermi liquid at a quantum critical point in a metal. VI. The γ model and its phase diagram at 2 < γ < 3, Phys. Rev. B 104, 144509 (2021). [2] D. Pimenov and A. V. Chubukov, Twists and turns of super- conductivity from a repulsive dynamical interaction, Ann. Phys. 447, 169049 (2022). [3] C. Bauer, Y. Schattner, S. Trebst, and E. Berg, Hierarchy of energy scales in an O(3) symmetric antiferromagnetic quantum critical metal: A Monte Carlo study, Phys. Rev. Res. 2, 023008 (2020). [4] M. A. Metlitski, D. F. Mross, S. Sachdev, and T. Senthil, Cooper pairing in non-Fermi liquids, Phys. Rev. B 91, 115111 (2015). 013045-13 LI, SACHDEV, AND JOSHI PHYSICAL REVIEW RESEARCH 5, 013045 (2023) [5] I. Mandal, Superconducting instability in non-Fermi liquids, Phys. Rev. B 94, 115138 (2016). [6] I. Esterlis, H. Guo, A. A. Patel, and S. Sachdev, Large N theory of critical Fermi surfaces, Phys. Rev. B 103, 235129 (2021). [7] S. Sachdev and J. Ye, Gapless Spin-Fluid Ground State in a Random Quantum Heisenberg Magnet, Phys. Rev. Lett. 70, 3339 (1993). [8] A. Y. Kitaev, Talks at KITP, University of California, Santa Barbara, Entanglement in Strongly-Correlated Quantum Matter (2015). [9] D. Chowdhury, A. Georges, O. Parcollet, and S. Sachdev, Sachdev-Ye-Kitaev models and beyond: A window into non- fermi liquids, Rev. Modern Phys. 94, 035004 (2022). [10] C. Li, D. G. Joshi, and S. Sachdev, Critical anomalous met- als near superconductivity in models with random interactions, Phys. Rev. B 103, 115147 (2021). [11] I. R. Klebanov, A. Milekhin, G. Tarnopolsky, and W. Zhao, Spontaneous breaking of U(1) symmetry in coupled complex SYK models, J. High Energy Phys. 11 (2020) 162. [12] É. Lantagne-Hurtubise, V. Pathak, S. Sahoo, and M. Franz, Superconducting instabilities in a spinful Sachdev-Ye-Kitaev model, Phys. Rev. B 104, L020509 (2021). [13] Y. Wang, Solvable Strong-Coupling Quantum-Dot Model with a Non-Fermi-Liquid Pairing Transition, Phys. Rev. Lett. 124, 017002 (2020). [14] H. Wang, A. L. Chudnovskiy, A. Gorsky, and A. Kamenev, Sachdev-Ye-Kitaev superconductivity: Quantum Kuramoto and generalized Richardson models, Phys. Rev. Res. 2, 033025 (2020). [15] I. Esterlis and J. Schmalian, Cooper pairing of incoherent elec- trons: An electron-phonon version of the Sachdev-Ye-Kitaev model, Phys. Rev. B 100, 115132 (2019). [16] A. A. Patel, M. J. Lawler, and E.-A. Kim, Coherent Supercon- ductivity with a Large Gap Ratio from Incoherent Metals, Phys. Rev. Lett. 121, 187001 (2018). [17] D. Chowdhury and E. Berg, Intrinsic superconducting instabili- ties of a solvable model for an incoherent metal, Phys. Rev. Res. 2, 013301 (2020). [18] A. L. Chudnovskiy and A. Kamenev, Superconductor–insulator transition in a non-fermi liquid, arXiv:2207.12307. [19] S. Sahoo, É. Lantagne-Hurtubise, S. Plugge, and M. Franz, Traversable wormhole and Hawking-Page transition in cou- pled complex SYK models, Phys. Rev. Res. 2, 043049 (2020). [20] E. A. Yuzbashyan, A. A. Baytin, and B. L. Altshuler, Strong- coupling expansion for the pairing Hamiltonian for small superconducting metallic grains, Phys. Rev. B 68, 214509 (2003). [21] E. A. Yuzbashyan, A. A. Baytin, and B. L. Altshuler, Finite- size corrections for the pairing Hamiltonian, Phys. Rev. B 71, 094505 (2005). [22] M. Christos, F. M. Haehl, and S. Sachdev, Spin liquid to spin glass crossover in the random quantum Heisenberg magnet, Phys. Rev. B 105, 085120 (2022). [23] O. Parcollet and A. Georges, Non-Fermi-liquid regime insulator, Phys. Rev. B 59, 5341 a doped Mott of (1999). [24] X.-Y. Song, C.-M. Jian, and L. Balents, Strongly Correlated Metal Built from Sachdev-Ye-Kitaev Models, Phys. Rev. Lett. 119, 216601 (2017). [25] L. C. Hebel and C. P. Slichter, Nuclear spin relaxation in normal and superconducting aluminum, Phys. Rev. 113, 1504 (1959). [26] M. Takigawa, A. P. Reyes, P. C. Hammel, J. D. Thompson, R. H. Heffner, Z. Fisk, and K. C. Ott, Cu and O NMR studies of the magnetic properties of YBa2Cu3O6.63 (tc = 62 k), Phys. Rev. B 43, 247 (1991). [27] H. Fukazawa, Y. Yamada, K. Kondo, T. Saito, Y. Kohori, K. Kuga, Y. Matsumoto, S. Nakatsuji, H. Kito, P. M. Shirage, K. Kihou, N. Takeshita, C.-H. Lee, A. Iyo, and H. Eisaki, Possi- ble multiple gap superconductivity with line nodes in heavily hole-doped superconductor KFe2As2 studied by 75As nuclear quadrupole resonance and specific heat, J. Phys. Soc. Jpn. 78, 083712 (2009). [28] H. J. Vidberg and J. W. Serene, Solving the eliashberg equations by means ofn-point padé approximants, J. Low Temp. Phys. 29, 179 (1977). [29] S. Sachdev, T. Senthil, and R. Shankar, Finite-temperature properties of quantum antiferromagnets in a uniform mag- netic field in one and two dimensions, Phys. Rev. B 50, 258 (1994). [30] O. Parcollet, A. Georges, G. Kotliar, and A. Sengupta, Overscreened multichannel SU(N) Kondo model: Large-N so- lution and conformal field theory, Phys. Rev. B 58, 3794 (1998). 013045-14
null
10.3390_math10224246.pdf
Data Availability Statement: Data sharing is not applicable to this article as no datasets were generated or analyzed during the current study.
Data Availability Statement: Data sharing is not applicable to this article as no datasets were generated or analyzed during the current study.
Article Speed of Convergence of Time Euler Schemes for a Stochastic 2D Boussinesq Model Hakima Bessaih 1 and Annie Millet 2,3,* 1 Mathematics and Statistics Department, Florida International University, 11200 SW 8th Street, 2 3 Miami, FL 33199, USA Statistique, Analyse et Modélisation Multidisciplinaire, EA 4543, Université Paris 1 Panthéon Sorbonne, Centre Pierre Mendès France, 90 Rue de Tolbiac, CEDEX, 75634 Paris, France Laboratoire de Probabilités, Statistique et Modélisation, UMR 8001, Universités Paris 6-Paris 7, Place Aurélie Nemours, 75013 Paris, France * Correspondence: amillet@univ-paris1.fr Abstract: We prove that an implicit time Euler scheme for the 2D Boussinesq model on the torus D converges. The various moments of the W1,2-norms of the velocity and temperature, as well as their discretizations, were computed. We obtained the optimal speed of convergence in probability, and a logarithmic speed of convergence in L2(Ω). These results were deduced from a time regularity of the solution both in L2(D) and W1,2(D), and from an L2(Ω) convergence restricted to a subset where the W1,2-norms of the solutions are bounded. Keywords: Boussinesq model; implicit time Euler schemes; convergence in probability; strong convergence MSC: Primary 60H15; 60H35; 65M12; Secondary 76D03; 76M35 1. Introduction The Boussinesq equations have been used as a model in many geophysical applications. They have been widely studied in both deterministic and stochastic settings. We take random forces into account and formulate the Bénard convection problem as a system of stochastic partial differential equations (SPDEs). The need to take stochastic effects into account for modeling complex systems has now become widely recognized. Stochastic partial differential equations (SPDEs) arise naturally as mathematical models for nonlinear macroscopic dynamics under random influences. The Navier–Stokes equations are coupled with a transport equation for the temperature and with diffusion. The system is subjected to a multiplicative random perturbation, which will be defined later. Here, u describes the fluid velocity field, whereas θ describes the temperature of the buoyancy-driven fluid, and π is the fluid’s pressure. We study the multiplicative stochastic Boussinesq equations ∂tu − ν∆u + (u · ∇)u + ∇π = θ + G(u) dW in ∂tθ − κ∆θ + (u · ∇θ) = ˜G(θ) d ˜W in (0, T) × D, (0, T) × D, (1) (2) div u = 0 in (0, T) × D, where T > 0. The processes u : Ω × (0, T) × D → R2 and θ : Ω × (0, T) × D → R have initial conditions u0 and θ0 in D, respectively. The parameter ν > 0 denotes the kinematic viscosity of the fluid, and κ > 0 denotes its thermal diffusivity. These fields satisfy periodic boundary conditions u(t, x + Lvi) = u(t, x), θ(t, x + Lvi) = θ(t, x) on (0, T) × ∂D, where vi, i = 1, 2 denotes the canonical basis of R2, and π : Ω × (0, T) × D → R is the pressure. Citation: Bessaih, H.; Millet, A. Speed of Convergence of Time Euler Schemes for a Stochastic 2D Boussinesq Model. Mathematics 2022, 10, 4246. https://doi.org/10.3390/ math10224246 Academic Editors: Vicente Martínez and Pablo Gregori Received: 12 October 2022 Accepted: 8 November 2022 Published: 13 November 2022 Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affil- iations. Copyright: © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/). Mathematics 2022, 10, 4246. https://doi.org/10.3390/math10224246 https://www.mdpi.com/journal/mathematics mathematics Mathematics 2022, 10, 4246 2 of 39 In dimension 2 without any stochastic perturbation, this system has been extensively studied with a complete picture about its well-posedness and long-time behavior. In the deterministic setting, more investigations have been extended to the cases where ν = 0 and/or κ = 0, with some partial results. If the (L2)2 (resp., L2) norms of u0 and θ0 are square integrable, it is known that the random system (1)–(2) is well-posed, and that there exists a unique solution (u × θ) in C([0, T]; (L2)2 × L2) ∩ L2(Ω; (H1)2 × H1); see, e.g., [1,2]. Numerical schemes and algorithms have been introduced to best approximate the solution to non-linear PDEs. The time approximation is either an implicit Euler or a time- splitting scheme coupled with a Galerkin approximation or finite elements to approximate the space variable. The literature on numerical analysis for SPDEs is now very extensive. In many papers, the models are either linear, have global Lipschitz properties, or, more generally, have some monotonicity property. In this case, the convergence was proven to be in mean square. When nonlinearities are involved that are not of Lipschitz or monotone type, then a rate of convergence in mean square is more difficult to obtain. Indeed, because of the stochastic perturbation, one may not use the Gronwall lemma after taking the expectation of the error bound, since it involves a nonlinear term that is often quadratic; such a nonlinearity requires some localization. In a random setting, the discretization of the Navier–Stokes equations on the torus has been intensively investigated. Various space–time numerical schemes have been studied for the stochastic Navier–Stokes equations with a multiplicative or an additive noise, where, in the right hand side of (1) (with no θ), we have either G(u) dW or dW. We refer to [3–7], where the convergence in probability is stated with various rates of convergence in a multiplicative setting for a time implicit Euler scheme, and [8] for a time splitting scheme. As stated previously, the main tool used to obtain the convergence in probability is the localization of the nonlinear term over a space of large probability. We studied the strong (that is, L2(Ω)) rate of convergence of the time-implicit Euler scheme (resp., space–time- implicit Euler scheme coupled with finite element space discretization) in our previous papers [9] (resp., [10]) for an H1-valued initial condition. The method is based on the fact that the solution (and the scheme) have finite moments (bounded uniformly on the mesh). For a general multiplicative noise, the rate is logarithmic. When the diffusion coefficient is bounded (which is a slight extension of an additive noise), the supremum of the H1-norm of the solution has exponential moments; we used this property in [9,10] to obtain an explicit polynomial strong rate of convergence. However, this rate depends on the viscosity and the strength of the noise, and is strictly less than 1/2 for the time parameter (resp., less than 1 for the spatial one). For a given viscosity, the time rates on convergence increase to 1/2 when the strength of the noise converges to 0. For an additive noise, if the strength of the noise is not too large, the strong (L2(Ω)) rate of convergence in time is the optimal one, and is almost 1/2 (see [11]). Once more, this is based on exponential moments of the supremum of the H1-norm of the solution (and of its scheme for the space discretization); this enabled us to have strong polynomial time rates. In the current paper, we study the time approximation of the Boussinesq Equations (1) and (2) in a multiplicative setting. To the best of our knowledge, it is the first result where a time-numerical scheme is implemented for a more general hydrodynamical model with a multiplicative noise. We use a fully implicit time Euler scheme and once more assume that the initial conditions u0 and θ0 belong to H1(D) in order to prove a rate of convergence in L2(D) uniformly in time. We prove the existence of finite moments of the H1-norms of the velocity and the temperature uniformly in time. Since we are on the torus, this is quite easy for the velocity. However, for the temperature, due to the presence of the velocity in the bilinear term, the argument is more involved and has to be carried out in two steps. It requires higher moments on the H1-norm of the initial condition. The time regularity of the solutions u, θ is the same as that of u in the Navier–Stokes equations. We then study rates of convergence in probability and in L2(Ω). The rate of convergence in probability is optimal (almost 1/2); we have to impose higher moments on the initial conditions than Mathematics 2022, 10, 4246 3 of 39 what is needed for the velocity described by stochastic Navier–Stokes equations. Once more, we first obtain an L2(Ω) convergence on a set where we bound the L2 norm of the gradients of both the velocity and the temperature. We deduce an optimal rate of convergence in probability that is strictly less than 1/2. When the H1-norm of the initial condition has all moments (for example, it is a Gaussian H1-valued random variable), the rate of convergence in L2(Ω) is any negative exponent of the logarithm of the number of time steps. These results extend those established for the Navier–Stokes equations subject to a multiplicative stochastic perturbation. The paper is organized as follows. In Section 2, we describe the model and the assumptions on the noise and the diffusion coefficients, and describe the fully implicit time Euler scheme. In Section 3, we state the global well-posedeness of the solution to (1)–(2), moment estimates of the gradient of u and θ uniformly in time and the existence of the scheme. We then formulate the main results of the paper about the rates of the convergence in probability and in L2(Ω) of the scheme to the solution. In Section 4, we prove moment estimates in H1 of u and θ uniformly on the time interval [0, T] if we start with more regular (H1) initial conditions. This is essential in order to be able to deduce a rate of convergence from the localized result. Section 5 states the time regularity results of the solution (u, θ) both in L2(D) and H1(D); this a crucial ingredient of the final results. In Section 6, we prove that the time Euler scheme is well-defined and prove its moment estimates in L2 and H1. Section 7 deals with the localized convergence of the scheme in L2(Ω). This preliminary step is necessary due to the bilinear term, which requires some control of the H1 norm of u and θ. In Section 8, we prove the rate of convergence in probability and in L2(Ω). Finally, Section 9 summarizes the interest of the model and describes some further necessary/possible extensions of this work. As usual, except if specified otherwise, C denotes a positive constant that may change throughout the paper, and C(a) denotes a positive constant depending on some parameter a. 2. Preliminaries and Assumptions In this section, we describe the functional framework, the driving noise, the evolution equations, and the fully implicit time Euler scheme. 2.1. The Functional Framework Let D = [0, L]2 with periodic boundary conditions Lp := Lp(D)2 (resp., Wk,p := Wk,p(D)2) be the usual Lebesgue and Sobolev spaces of vector-valued functions endowed with the norms (cid:107) · (cid:107)Lp (resp., (cid:107) · (cid:107)Wk,p ). Let V0 := {u ∈ L2 : div(u) = 0 on D}. Let Π : L2 → V0 denote the Leray projection, and let A = −Π∆ denote the Stokes operator, with domain Dom(A) = W2,2 ∩ V0. Let ˜A = −∆ acting on L2(D). For any non-negative real number k, let Hk = Dom(cid:0) ˜A k 2 (cid:1), Vk = Dom(cid:0)A k 2 (cid:1), endowed with the norms (cid:107) · (cid:107)Hk and (cid:107) · (cid:107)Vk . Thus, H0 = L2(D) and Hk = Wk,2. Moreover, let V−1 be the dual space of V1 with respect to the pivot space V0, and (cid:104)·, ·(cid:105) denote the duality between V1 and V−1. Let b : (V1)3 → R denote the trilinear map defined by b(u1, u2, u3) := (cid:90) D (cid:0)(cid:2)u1(x) · ∇(cid:3)u2(x)(cid:1) · u3(x) dx. The incompressibility condition implies that b(u1, u2, u3) = −b(u1, u3, u2) for ui ∈ V1, i = 1, 2, 3. There exists a continuous bilinear map B : V1 × V1 (cid:55)→ V−1 such that (cid:104)B(u1, u2), u3(cid:105) = b(u1, u2, u3), for all ui ∈ V1, i = 1, 2, 3. Mathematics 2022, 10, 4246 4 of 39 Therefore, the map B satisfies the following antisymmetry relations: (cid:104)B(u1, u2), u3(cid:105) = −(cid:104)B(u1, u3), u2(cid:105), For u, v ∈ V1, we have B(u, v) := Π(cid:0)(cid:2)u · ∇(cid:3)v(cid:1). Furthermore, since D = [0, L]2 with periodic boundary conditions, we have (see e.g., [12]) (cid:104)B(u1, u2), u2(cid:105) = 0 for all (3) ui ∈ V1. (cid:104)B(u, u), Au(cid:105) = 0, ∀u ∈ V2. Note that, for u ∈ V1 and θ1, θ2 ∈ H1, if (u.∇)θ = ∑i=1,2 ui∂iθ, we have (cid:104)[u.∇]θ1 , θ2(cid:105) = −(cid:104)[u.∇]θ2 , θ1(cid:105), (4) (5) so that (cid:104)[u.∇]θ , θ(cid:105) = 0 for u ∈ V1 and θ ∈ H1. In dimension 2, the inclusions H1 ⊂ Lp and V1 ⊂ Lp for p ∈ [2, ∞) follow from the Sobolev embedding theorem. More precisely, the following Gagliardo–Nirenberg inequality is true for some constant ¯Cp: (cid:107)u(cid:107)Lp ≤ ¯Cp (cid:107)A 1 2 u(cid:107)α L2 (cid:107)u(cid:107)1−α L2 for α = 1 − 2 p , ∀u ∈ V1. (6) Finally, let us recall the following estimate of the bilinear terms (u.∇)v and (u.∇)θ. Lemma 1. Let α, ρ be positive numbers and δ ∈ [0, 1) be such that δ + ρ > 1 Let u ∈ Vα, v ∈ Vρ and θ ∈ Hρ; then, 2 and α + δ + ρ ≥ 1. (cid:107)A−δΠ[(u.∇)v](cid:107)V0 ≤ C(cid:107)Aαu(cid:107)V0 (cid:107)Aρv(cid:107)V0, (cid:107) ˜A−δ[(u.∇)θ](cid:107)H0 ≤ C(cid:107)Aαu(cid:107)V0 (cid:107) ˜Aρθ(cid:107)H0, (7) (8) for some positive constant C := C(α, δ, ρ). Proof. The upper estimate (7) is Lemma 2.2 in [13]. The argument, which is based on the Sobolev embedding theorem and Hölder’s inequality, clearly proves (8). 2.2. The Stochastic Perturbation Let K (resp., ˜K) be a Hilbert space and let (W(t), t ≥ 0) (resp., ( ˜W(t), t ≥ 0) ) be a K-valued (resp., ˜K-valued) Brownian motion with covariance Q (resp., ˜Q), which is a trace-class operator of K (resp., ˜K) such that Qζ j = qjζ j (resp., ˜Q ˜ζ j = ˜qj ˜ζ j), where {ζ j}j≥0 (resp., { ˜ζ j}j≥0) is a complete orthonormal system of K (resp., ˜K), qj, ˜qj > 0, and Tr(Q) = ∑j≥0 qj < ∞ (resp., Tr( ˜Q) = ∑j≥0 ˜qj < ∞). Let {βj}j≥0 (resp., { ˜βj}j≥0) be a sequence of independent one-dimensional Brownian motions on the same filtered probability space (Ω, F , (Ft, t ≥ 0), P). Then, W(t) = ∑ j≥0 (cid:112)qj βj(t) ζ j, ˜W(t) = ∑ j≥0 (cid:113) ˜qj ˜βj ˜ζ j. For details concerning these Wiener processes, we refer to [14]. Projecting the velocity on divergence-free fields, we consider the following SPDEs for processes modeling the velocity u(t) and the temperature θ(t). The initial conditions u0 and θ0 are F0-measurable, taking values in V0 and H0, respectively, and ∂tu(t) + (cid:2)ν Au(t) + B(u(t), u(t))(cid:3)dt = Π(θ(t)v2) + G(u(t)) dW(t), ∂tθ(t) + (cid:2)κ ˜Aθ(t) + (u(t).∇)θ(t)(cid:3)dt = ˜G(θ(t)) d ˜W(t), (9) (10) where ν, κ are strictly positive constants, and v2 = (0, 1) ∈ R2. Mathematics 2022, 10, 4246 5 of 39 We make the following classical linear growth and Lipschitz assumptions on the diffusion coefficients G and ˜G. For technical reasons, we will have to require u0 ∈ V1 and θ ∈ H1 and prove estimates similar to (19) and (20), raising the space regularity of the processes by one step in the scale of Sobolev spaces. Therefore, we have to strengthen the regularity of the diffusion coefficients. Condition (C-u) (i) Let G : V0 → L(K; V0) be such that (cid:107)G(u)(cid:107)2 L(K,V0) ≤ K0 + K1(cid:107)u(cid:107)2 L(K,V0) ≤ L1(cid:107)u1 − u2|2 V0 , V0, ∀u ∈ V0, ∀u1, u2 ∈ V0. (cid:107)G(u1) − G(u2)(cid:107)2 (ii) Let also G : V1 → L(K; V1) satisfy the growth condition (cid:107)G(u)(cid:107)2 L(K;V1) ≤ K2 + K3(cid:107)u(cid:107)2 V1, ∀u ∈ V1. and Condition (C-θ) (i) Let ˜G : H0 → L( ˜K; H0) be such that (cid:107) ˜G(θ)(cid:107)2 L( ˜K,H0) ≤ ˜K0 + ˜K1(cid:107)θ(cid:107)2 L( ˜K,H0) ≤ ˜L1(cid:107)θ1 − θ2(cid:107)2 H0, H0, ∀θ ∈ H0, ∀θ1, θ2 ∈ H0. (cid:107) ˜G(θ1) − ˜G(θ2)(cid:107)2 (ii) Let also ˜G : H1 → L(K; H1) satisfy the growth condition (cid:107) ˜G(θ)(cid:107)2 L( ˜K;H1) ≤ ˜K2 + ˜K3(cid:107)θ(cid:107)2 H1, ∀θ ∈ H1. (11) (12) (13) (14) (15) (16) 2.3. The Fully Implicit Time Euler Scheme Fix N ∈ {1, 2, ...}, let h := T N denote the time mesh, and, for j = 0, 1, . . . , N, set tj := j T N . The fully implicit time Euler scheme {uk; k = 0, 1, ..., N} and {θk; k = 0, 1, ..., N} is defined by u0 = u0, θ0 = θ0, and, for ϕ ∈ V1, ψ ∈ H1 and l = 1, ..., N, (cid:16) ul − ul−1 + hνAul + hB(cid:0)ul, ul(cid:1), ϕ (cid:16) θl − θl−1 + hκ ˜Aθl + h[ul−1.∇]θl, ψ (cid:17) (cid:17) =(cid:0)Πθl−1v2, ϕ)h + (cid:0)G(ul−1)[W(tl) − W(tl−1)] , ϕ), =(cid:0) ˜G(θl−1)[ ˜W(tl) − ˜W(tl−1)] , ϕ). (17) (18) 3. Main Results In this section, we state the main results about the well-posedness of the solutions (u, θ), the scheme {uk; k = 0, 1, ..., N} and the rate of the convergence of the scheme {(uk, θk); k = 0, 1, ..., N} to (u, θ). 3.1. Global Well-Posedness and Moment Estimates of (u, θ) The first results state the existence and uniqueness of a weak pathwise solution (that is a strong probabilistic solution in the weak deterministic sense) of (9) and (10). It is proven in [1] (see also [2]). Theorem 1. Let u0 ∈ L2p(Ω; V0) and θ0 ∈ L2p(Ω; H0) for p = 1 or p ∈ [2, ∞). Let the coefficients G and ˜G satisfy the conditions (C-u)(i) and (C-θ)(i), respectively. Then, Equations (9) and (10) have a unique pathwise solution, i.e., • u (resp., θ) is an adapted V0-valued (resp., H0-valued) process that belongs a.s. to L2(0, T; V1) (resp., to L2(0, T; H1)); Mathematics 2022, 10, 4246 6 of 39 • P a.s. we have u ∈ C([0, T]; V0), θ ∈ C(0, T]; H0) and (cid:0)u(t), ϕ(cid:1)+ν (cid:90) t 0 (cid:0)A 1 2 u(s), A 1 2 ϕ(cid:1)ds + (cid:90) t 0 (cid:10)[u(s) · ∇]u(s), ϕ(cid:11)ds = (cid:0)u0, ϕ) + (cid:90) t (cid:90) t 0 (cid:0)Πθ(t)v2, ϕ)ds + (cid:90) t (cid:90) t 0 (cid:0)ϕ, G(u(s))dW(s)(cid:1), (cid:0)θ(t), ψ(cid:1)+κ 1 (cid:0) ˜A 2 θ(s), ˜A 1 2 ψ(cid:1)ds + (cid:10)[u(s) · ∇]θ(s), ψ(cid:11)ds 0 = (cid:0)θ0, ψ) + (cid:90) t 0 0 (cid:0)φ, ˜G(θ(s))d ˜W(s)(cid:1), for every t ∈ [0, T] and every ϕ ∈ V1 and ψ ∈ H1. Furthermore, E(cid:16) E(cid:16) sup t∈[0,T] sup t∈[0,T] (cid:107)u(t)(cid:107)2p V0 + (cid:107)θ(t)(cid:107)2p H0 + (cid:90) T 0 (cid:90) T 0 (cid:107)A 1 2 u(t)(cid:107)2 V0 (cid:2)1 + (cid:107)u(t)(cid:107)2(p−1) V0 (cid:107) ˜A 1 2 θ(t)(cid:107)2 H0 (cid:2)1 + (cid:107)θ(t)(cid:107)2(p−1) H0 (cid:17) (cid:3)dt (cid:17) (cid:3)dt ≤ C(cid:0)1 + E((cid:107)u0(cid:107)2p V0 (cid:1), ≤ C(cid:0)1 + E((cid:107)θ0(cid:107)2p H0 (cid:1). (19) (20) The following result proves that, if u0 ∈ V1, the solution u to (9) and (10) is more regular. Proposition 1. Let u0 ∈ L2p(Ω; V1) and θ0 ∈ L2p(Ω; H0) for p = 1 or some p ∈ [2, ∞), and let G satisfy condition (C-u) and ˜G satisfy condition (C-θ). Then, the solution u to (9) and (10) belongs a.s. to C([0, T]; V1) ∩ L2([0, T]; V2). Moreover, for some constant C, E(cid:16) sup t∈[0,T] (cid:107)u(t)(cid:107)2p V1 + (cid:90) T 0 (cid:107)Au(t)(cid:107)2 V0 (cid:2)1 + (cid:107)A 1 2 u(t)(cid:107)2(p−1) V0 (cid:3)dt (cid:17) ≤ C(cid:2)1 + E(cid:0)(cid:107)u0(cid:107)2p V1 + (cid:107)θ0(cid:107)2p H0 (cid:1)(cid:3). (21) The next result proves similar bounds for moments of the gradient of the temperature uniformly in time. Proposition 2. Let u0 ∈ L8p+(cid:101)(Ω; V1) and θ0 ∈ L8p+(cid:101)(Ω; H1) for some (cid:101) > 0 and p = 1 or p ∈ [2, +∞). Suppose that the coefficients G and ˜G satisfy the conditions (C-u) and (C-θ). There exists a constant C such that E(cid:104) (cid:107) ˜A sup t≤T 1 2 θ(t)(cid:107)2p H0 + (cid:90) T 0 (cid:107) ˜Aθ(s)(cid:107)2 H0 (cid:107) ˜A 1 2 θ(s)(cid:107)2(p−1) H0 ds (cid:105) ≤ C. (22) 3.2. Global Well-Posedness of the Time Euler Scheme The following proposition states the existence and uniqueness of the sequences {uk}k=0,...,N and {θk}k=0,...,N. Proposition 3. Let condition (G-u)(i) and (C-θ)(i) be satisfied, u0 ∈ V0 and θ0 ∈ H0 a.s. The time fully implicit scheme (17) and (18) has a unique solution {ul}l=1,...,N ∈ V1, {θl}l=1,...,N ∈ H1. 3.3. Rates of Convergence in Probability and in L2(Ω) The following theorem states that the implicit time Euler scheme converges to the pair (u, θ) in probability with the “optimal” rate “almost 1/2”. It is the main result of the paper. For j = 0, ..., N, set ej := u(tj) − uj and ˜ej := θ(tj) − θ j; then, e0 = ˜e0 = 0. Mathematics 2022, 10, 4246 7 of 39 Theorem 2. Suppose that the conditions (C-u) and (C-θ) hold. Let u0 ∈ L32+(cid:101)(Ω; V1) and θ0 ∈ L32+(cid:101)(Ω; H1) for some (cid:101) > 0, u, θ be the solution to (9) and (10) and {uj, θ j}j=0,...,N be the solution to (17) and (18). Then, for every η ∈ (0, 1), we have (cid:16) lim N→∞ P max 1≤J≤N (cid:2)(cid:107)eJ(cid:107)2 V0 + (cid:107) ˜eJ(cid:107)2 H0 (cid:3) + T N N ∑ j=1 (cid:2)(cid:107)A 1 2 ej(cid:107)2 V0 + (cid:107) ˜A 1 2 ˜ej(cid:107)2 H0 (cid:3) ≥ N−η(cid:17) = 0. (23) We finally state that the strong (i.e., in L2(Ω)) rate of convergence of the implicit time Euler scheme is some negative exponent of ln N. Note that, if the initial conditions u0 and θ0 are deterministic, or if their V1 and H1-norms have moments of all orders (for example, if u0 and θ0 are Gaussian random variables), the strong rate of convergence is any negative exponent of ln N. More precisely, we have the following result. Theorem 3. Suppose that the conditions (C-u) and (C-θ)(i) hold. Let u0 ∈ L2q+(cid:101)(Ω; V1) and θ0 ∈ L2q+(cid:101)(Ω; H1) for q ∈ [5, ∞) and some (cid:101) > 0. Then, for some constant C such that 1 2 ej(cid:107)2 V0 + (cid:107) ˜A (cid:17) 1 2 ˜ej(cid:107)2 H0 ≤ C(cid:0) ln(N)(cid:1)−(2q−1+1) (24) E(cid:16) max 1≤J≤N (cid:2)(cid:107)eJ(cid:107)2 V0 + (cid:107) ˜eJ(cid:107)2 H0 (cid:3) + T N N ∑ j=1 (cid:2)(cid:107)A for large enough N. 4. More Regularity of the Solution 4.1. Moments of u in L ∞(0, T; V1) In this section, we prove that, if u0 ∈ V1 and θ0 ∈ H0, the H1-norm of the velocity has bounded moments uniformly in time. Proof of Proposition 1. Apply the operator A the square of the (cid:107).(cid:107)V0-norm of A 1 2 u(t). Then, using (4), we obtain 1 2 to (9) and use (formally) Itô’s formula for 1 (cid:107)A 2 u(t)(cid:107)2 (cid:107)Au(s)(cid:107)2 V0 ds = (cid:107)A 1 2 u0(cid:107)2 (cid:0)A 1 2 Πθ(s)v2, A 1 2 u(s)(cid:1)ds (25) (cid:90) t V0 + 2ν (cid:90) t + 2 0 (cid:0)A 0 (cid:90) t V0 + 2 (cid:90) t 0 0 1 2 G(u(s))dW(s), A 1 2 u(s)(cid:1) + 1 (cid:107)A 2 G(u(s))(cid:107)2 L(K;V0) Tr(Q)ds. Let τM := inf{t : (cid:107)u(t)(cid:107)V1 ≥ M}; using (13), integration by parts and the Cauchy–Schwarz and Young inequalities, we deduce, for M > 0 and t ∈ [0, T], E(cid:16) (cid:107)A 1 (cid:90) t∧τM 2 u(t ∧ τM)(cid:107)2 + 2E(cid:16) (cid:90) t∧τM V0 + 2ν (cid:107)Au(s)(cid:107)2 V0 ds (cid:107)θ(s)(cid:107)H0 (cid:107)Au(s)(cid:107)V0 ds + Tr(Q)E(cid:16) (cid:90) t∧τN ≤ E(cid:0)(cid:107)u0(cid:107)2 V0 (cid:1) 0 (cid:17) ≤E(cid:16) (cid:107)u0(cid:107)2 + K3TE(cid:16) 0 V0 + ν (cid:90) t∧τM 0 (cid:107)Au(s)(cid:107)2 V0 ds (cid:90) t + K3 0 (cid:17) (cid:107)u(t)(cid:107)2 V0 sup t∈[0,T] (cid:17) + 1 ν E(cid:16) (cid:90) t∧τM 0 E(cid:16) (cid:107)A 1 2 u(s ∧ τM)(cid:107)2 V0 H0 ds (cid:17) ds. 0 (cid:2)K2 + K3(cid:107)u(s)(cid:107)2 V1 (cid:17) + K2T (cid:107)θ(s)(cid:107)2 (cid:17) (cid:3)ds Indeed the stochastic integral in the right hand side of (25) is a square integrable, and hence a centered martingale. Neglecting the time integral in the left hand side, using (19) and the Gronwall lemma, we deduce E(cid:16) A 1 2 (cid:107)u(t ∧ τM)(cid:107)2 V0 (cid:17) ≤ C < ∞. sup M sup t∈[0,T] (26) Mathematics 2022, 10, 4246 8 of 39 As M → ∞, this implies that E(cid:0) (cid:82) T Furthermore, the Davis inequality and Young’s inequality imply V0 ds(cid:1) < ∞. 0 (cid:107)Au(s)(cid:107)2 E(cid:16) sup s≤t (cid:90) s∧τM 0 (cid:0)A 1 2 G(u(r))dW(r), A 1 2 u(r)(cid:1)(cid:17) ≤ 3E(cid:16)(cid:110) (cid:90) t 0 (cid:107)A ≤ 3E(cid:16) (cid:107)A sup s≤t 1 1 2 u(r ∧ τM)(cid:107)2 2 u(s ∧ τM)(cid:107)V0 V0 Tr(Q) (cid:107)A (cid:90) t (cid:110) Tr(Q) 0 1 2 G(u(r ∧ τM))(cid:107)2 L(K;V0)dr (cid:111) 1 2 (cid:17) [K2 + K3(cid:107)u(s ∧ τM)(cid:107)2 V1 ]ds (cid:111) 1 2 (cid:17) ≤ E(cid:16) 1 2 sup s≤t (cid:107)A 1 2 u(s ∧ τM)(cid:107)2 V0 (cid:17) + 9Tr(Q)E(cid:16) (cid:90) t 0 [K2 + K3(cid:107)u(r ∧ τM)(cid:107)2 V1 ds (cid:17) . The upper estimates (19), (20), (25) and (26) imply that, for some constant C depending on E((cid:82) T 0 V0 + (cid:107)θ(t)(cid:107)2 H0 (cid:3)ds(cid:1) < ∞, (cid:2)(cid:107)u(t)(cid:107)2 V0 + (cid:107)A 1 2 u(t)(cid:107)2 E(cid:16) 1 2 sup M (cid:107)A sup t≤T 1 2 u(t ∧ τM)(cid:107)2 V0 + (cid:90) T∧τM 0 (cid:107)Au(s)(cid:107)2 V0 ds (cid:17) ≤ C + CE(cid:16) (cid:90) T 0 (cid:2)(cid:107)A As M → ∞, we deduce 1 2 u(t)(cid:107)2 V1 + (cid:107)θ(t)(cid:107)2 H0 (cid:17) (cid:3)ds < ∞. E(cid:16) sup t∈[0,T] (cid:107)A 1 2 u(t)(cid:107)2 V0 (cid:17) + E(cid:16) (cid:90) T 0 (cid:107)Au(s)(cid:107)2 V0 ds (cid:17) ≤ C < ∞. This proves (21) for p = 1. Given p ∈ [2, ∞) and using Itô’s formula for the map x (cid:55)→ xp in (25), we obtain (cid:90) t∧τM (cid:107)A 1 2 u(t ∧ τM)(cid:107)2p (cid:90) t∧τM V0 + 2pν 0 2 Πθ(s)v2, A 1 (cid:0)A + 2p + 2p 0 (cid:90) t∧τM 0 (cid:107)Au(s)(cid:107)2 V0 (cid:107)A 1 2 u(s)(cid:107)2(p−1) V0 ds = (cid:107)A 1 2 u0(cid:107)2p V0 1 2 u(s)(cid:1) (cid:107)A 1 2 u(s)(cid:107)2(p−1) V0 ds (cid:0)A 1 2 G(u(s))dW(s), A 1 2 u(s)(cid:1) (cid:107)A 1 2 u(s)(cid:107)2(p−1) V0 + pTr(Q) (cid:90) t∧τM 0 (cid:107)G(u(s)(cid:107)2 L(K;V1)(cid:107)A 1 2 u(s)(cid:107)2(p−1) V0 ds + 2p(p − 1)Tr(Q) (cid:90) t∧τM 0 (cid:107)(cid:0)A 1 2 G(cid:1)∗(u(s))(cid:0)A 1 2 u(s)(cid:1)(cid:107)2 K(cid:107)A 1 2 u(s)(cid:107)2(p−2) V0 ds. (27) Integration by parts and the Cauchy–Schwarz, Hölder and Young inequalities imply that (cid:12) (cid:12) (cid:12) (cid:90) t 0 (cid:0)A 1 2 Πθ(s)v2, A 1 2 u(s)(cid:1)(cid:107)A 1 2 u(s)(cid:107)2(p−1) V0 ds (cid:107)Au(s)(cid:107)V0 (cid:107)θ(s)(cid:107)H0 (cid:107)A 1 2 u(s)(cid:107)2(p−1) V0 ds (cid:90) t (cid:12) (cid:12) (cid:12) ≤ 0 2 (cid:110) (cid:90) t (cid:111) 1 (cid:107)Au(s)(cid:107)2 V0 (cid:107)A 1 2 u(s)(cid:107)2(p−1) V0 ds (cid:107)Au(s)(cid:107)2 V0 (cid:107)A 1 2 u(s)(cid:107)2(p−1) V0 ds + ≤ ≤ (cid:110) (cid:90) t 0 (cid:90) t pν 2 (cid:90) t 0 ≤ (cid:101) 0 1 2 u(s)(cid:107)2(p−1) V0 ds (cid:111) 1 2 1 2 u(s)(cid:107)2(p−1) V0 ds (cid:107)θ(s)(cid:107)2 H0 (cid:107)A (cid:90) t (cid:107)θ(s)(cid:107)2 0 1 2pν (cid:90) t 0 0 (cid:107)θ(s)(cid:107)2p H0 (cid:107)A (cid:90) t (cid:107)Au(s)(cid:107)2 V0 (cid:107)A 1 2 u(s)(cid:107)2(p−1) V0 ds + C H0 ds + C 0 (cid:107)A 1 2 u(s)(cid:107)2p V0 ds. (28) Mathematics 2022, 10, 4246 9 of 39 Since ap−1 ≤ 1 + ap for any a ≥ 0, the growth condition (13) implies that (cid:90) t 0 (cid:107)A 1 2 G(u(s))(cid:107)2 L(K,V0)(cid:107)A 1 2 u(s)(cid:107)2(p−1) V0 ds (cid:90) t ≤ 0 (cid:16) ≤ C (cid:2)K2 + K3(cid:107)u(s)(cid:107)2 1 2 u(s)(cid:107)2 V0 (cid:3)(cid:107)A 1 2 u(s)(cid:107)2(p−1) V0 ds V0 + K3(cid:107)A (cid:90) t T + (cid:90) T 0 (cid:107)u(s)(cid:107)2p V0 + (cid:107)A 1 2 u(s)(cid:107)2p V0 ds 0 (cid:17) . (29) 1 2 G(u(s))(cid:1)∗ Furthermore, since (cid:0)(cid:107)A V0, the upper estimate of the corresponding integral is similar to that of (29). Since the stochastic in- tegral (cid:82) t∧τM 2 u(s)(cid:1)(cid:107)u(s)(cid:107)2(p−1) (cid:1) is square integrable, it is centered. Therefore, (27) and the above upper estimates (28) and (29) imply that V0 ≤ [K2 + K3(cid:107)u(s)(cid:107)2 1 2 G(u(s))dW(s), A 1 2 u(s)(cid:107)2 1 2 u(s)(cid:107)2 V1 ](cid:107)A (cid:0)A V0 A 0 1 E(cid:16) (cid:107)A sup M 1 2 u(t ∧ τM)(cid:107)2p V0 + pν (cid:90) t∧τM (cid:107)Au(s)(cid:107)2 V0 (cid:107)A 1 2 u(s)(cid:107)2(p−1) V0 (cid:17) (cid:16) ≤ C T + E(cid:16) (cid:90) t 0 V0 + (cid:107)θ(s)(cid:107)2p H0 (cid:17) (cid:3)ds + (cid:90) t 0 E(cid:0)(cid:107)A 1 2 u(s ∧ τM)(cid:107)2p V0 (cid:1)ds. 0 (cid:2)(cid:107)u(s)(cid:107)2p Using Gronwall’s lemma we obtain sup M sup M sup t∈[0,T] E(cid:16) (cid:90) T∧τM 0 E(cid:0)(cid:107)A 1 2 u(s ∧ τM)(cid:107)2p V0 (cid:1) = C < ∞, (cid:107)Au(s)(cid:107)2 V0 (cid:107)A 1 2 u(s)(cid:107)2(p−1) V0 ds (cid:17) = C < ∞. (30) (31) Finally, using the Davis inequality, the Hölder and Young inequalities, we deduce E(cid:16) (cid:12) (cid:12) (cid:12) 2p sup s∈[0,t] (cid:90) s∧τM 0 (cid:0)A 1 2 G(u(r))dW(r), A 1 2 u(r)(cid:107)A 1 2 u(r)(cid:107)2(p−1) V0 (cid:17) (cid:12) (cid:12) (cid:12) Tr(Q)(cid:107)A 1 2 G(u(s))(cid:107)2 L(K;V0)(cid:107)A 1 2 u(s)(cid:107)4p−2 V0 (cid:111) 1 2 (cid:17) ds ≤ 6p E(cid:16)(cid:110) (cid:90) t∧τM 2 E(cid:16) ≤ 6p (cid:0)Tr(Q)(cid:1) 1 0 (cid:107)A 1 2 u(s)(cid:107)p V0 sup s≤t∧τM (cid:110) (cid:90) t × 0 ≤ E(cid:16) 1 2 sup s∈[0,t∧τM] (cid:107)A (cid:107)A 1 2 G(u(s ∧ τN))(cid:107)2 + CE(cid:16) 2 u(s)(cid:107)2p V0 (cid:17) 1 L(K;V0)(cid:107)A (cid:90) t 1 + 0 1 2 u(s ∧ τM)(cid:107)2p−2 (cid:90) t V0 (cid:107)u(s)(cid:107)2p V0 ds + 0 (cid:111) 1 2 (cid:17) ds (cid:107)A 1 2 u(s)(cid:107)2p V0 ds (cid:17) . (32) The upper estimates (27), (19) and (32) imply that E(cid:16) sup M sup s∈[0,T∧τM] (cid:107)A 1 2 u(s)(cid:107)2p V0 (cid:17) (cid:104) ≤ C 1 + sup M E(cid:16) (cid:90) T 0 (cid:2)(cid:107)θ(s ∧ τM)(cid:107)2p H0 + (cid:107)u(s ∧ τM)(cid:107)2p V1 (cid:17)(cid:105) (cid:3)ds < ∞. As M → ∞ in this inequality and in (31), the monotone convergence theorem concludes the proof of (21). 4.2. Moment Estimates of θ in L ∞(0, T; H1) We next give upper estimates for moments of supt∈[0,T] (cid:107) ˜A 1 2 θ(t)(cid:107)H0, i.e., prove Propo- sition 2. Mathematics 2022, 10, 4246 10 of 39 However, since (cid:104)[u(s).∇]θ(s), ˜Aθ(s)(cid:105) (cid:54)= 0, unlike what we have in the proof of the previous result, we keep the bilinear term. This creates technical problems and we proceed in two steps. First, using the mild formulation of the weak solution θ of (10), we prove that the gradient of the temperature has finite moments. Then, going back to the weak form, we prove the desired result. Let {S(t)}t≥0 be the semi-group generated by −νA, { ˜S(t)}t≥0 be the semi-group generated by −κ ˜A, which is S(t) = exp(−νtA), and ˜S(t) = exp(−κt ˜A) for every t ≥ 0. Note that, for every α > 0, (cid:107)AαS(t)(cid:107)L(V0;V0) ≤ Ct−α, (cid:107)A−α(cid:2)Id − S(t)(cid:3)(cid:107)L(V0;V0) ≤ Ctα, ∀t > 0 ∀t > 0. (33) (34) Similar upper estimates are valid when we replace A with ˜A, S(t) with ˜S(t) and V0 with H0. Note that if u0 ∈ L2(Ω; V1) and θ0 ∈ L2(Ω; H0), we deduce u ∈ L2(Ω; C([0, T]; V0) ∩ ∞([0, T]; V1)) and θ ∈ L2(Ω; C([0, T]; H0)) ∩ L2(Ω × [0, T]; H1). We can write the solu- L tions of (9) and (10) in the following mild form: u(t) = S(t)u0 − (cid:90) t 0 S(t − s)B(u(s), u(s)) ds + (cid:90) t 0 S(t − s)(cid:0)Πθ(t)v2 (cid:1) ds + (cid:90) t 0 S(t − s)G(u(s))dW(s), θ(t) = ˜S(t)θ0 − (cid:90) t 0 ˜S(t − s)(cid:0)[u(s).∇]θ(s)(cid:1) ds + (cid:90) t 0 ˜S(t − s) ˜G(θ(s))d ˜W(s), (35) (36) where the first equality holds a.s. in V0 and the second one in H0. Indeed, since (cid:107)Aαu(cid:107)V0 ≤ C(cid:107)A 1 2 u(cid:107)2α V0 (cid:107)u(s)(cid:107)1−2α V0 , the upper estimate (7) for δ + ρ > 1 2 , δ + α + ρ = 1 and the Minkowski inequality imply that (cid:13) (cid:13) (cid:13) (cid:90) t 0 S(t − s)B(u(s), u(s))ds (cid:13) (cid:13) (cid:13)V0 ≤ (cid:90) t 0 (cid:107)Aδ A−δB(u(s), u(s))(cid:107)V0 ds ≤ C (cid:90) t 0 (t − s)−δ(cid:107)Aαu(s)(cid:107)V0 (cid:107)Aρu(s)(cid:107)V0 ds ≤ C sup s∈[0,t] (cid:107)u(s)(cid:107)2 V1 (cid:90) t 0 (t − s)−δds Since (cid:107)S(t)(cid:107)L(V0;V0) ≤ 1, it is easy to see that (cid:13) (cid:13) (cid:13) (cid:90) t 0 S(t − s)Πθ(t)v2ds (cid:13) (cid:13) (cid:13)V0 ≤ C (cid:90) t 0 (cid:107)θ(t)(cid:107)H0 ds. Furthermore, E(cid:16)(cid:13) (cid:13) (cid:13) (cid:90) t 0 S(t − s)G(u(s))dW(s) (cid:13) (cid:13) (cid:13) 2 (cid:17) V0 ≤ Tr(Q)E(cid:16) (cid:90) t 0 [K0 + K1(cid:107)u(t)(cid:107)2 V0 (cid:17) (cid:3)ds < ∞. Therefore, the stochastic integral (cid:82) t is true a.s. in V0. 0 S(t − s)G(u(s))dW(s) ∈ V0 a.s., and the identity (35) A similar argument shows that (36) holds a.s. in H0. We only show that the convolution involving the bilinear term belongs to H0. Using the Minkowski inequality and the upper Mathematics 2022, 10, 4246 11 of 39 estimate (8) with positive constants δ, α, ρ such that α, ρ ∈ (0, 1 1, we obtain 2 ), δ + ρ > 1 2 and δ + α + ρ = (cid:13) (cid:13) (cid:13) (cid:90) t 0 ˜S(t − s)[(u(s).∇)θ(s)]ds (cid:13) (cid:13) (cid:13)H0 ≤ (cid:90) t 0 (cid:107) ˜Aδ ˜S(t − s) ˜A−δ[(u(s).∇)θ(s)](cid:107)H0 ds ≤ C (cid:90) t 0 (t − s)−δ (cid:107)Aαu(s)(cid:107)V0 (cid:107) ˜Aρθ(s)(cid:107)H0 d s ≤ C sup s∈[0,t] (cid:107)u(s)(cid:107)V1 sup s∈[0,t] (cid:107)θ(s)(cid:107)1−2ρ H0 (cid:16) (cid:90) t 0 (t − s)− δ 1−ρ ds (cid:17)1−ρ(cid:16) (cid:90) t 0 (cid:107) ˜A 1 2 θ(s)(cid:107)2 H0 ds (cid:17)ρ < ∞, where the last upper estimate is deduced from Hölder’s inequality and δ 1−ρ < 1. The following result shows that, for fixed t, the L2-norm of the gradient of θ(t) has finite moments. Lemma 2. Let p ∈ [0, +∞), u0 ∈ L4p+(cid:101)(Ω; V1) and θ0 ∈ L4p+(cid:101)(Ω; H1) for some (cid:101) ∈ (0, 1 2 ). Let the diffusion coefficient G and ˜G satisfy the condition (C) and ( ˜C), respectively. For every N, let ˜τN := inf{t ≥ 0 : (cid:107) ˜A 2 θ(t)(cid:107)H0 ≥ N} ∧ T; then, 1 sup N>0 sup t∈[0,T] E(cid:0)(cid:107) ˜A 1 2 θ(t ∧ ˜τN)(cid:107)2p H0 (cid:1) < ∞. (37) Proof. Write θ(t) using (36); then, (cid:107) ˜A 1 2 θ(t)(cid:107)H0 ≤ ∑3 i=1 Ti(t), where T1(t) = (cid:107) ˜A 2 ˜S(t)θ0(cid:107)H0 , T2(t) = (cid:90) t 1 ˜A 2 ˜S(t − s) ˜G(θ(s))d ˜W(s) T3(t) = (cid:13) (cid:13) (cid:13) 1 0 (cid:90) t 1 (cid:13) (cid:13) (cid:13) 0 (cid:13) (cid:13) (cid:13)H0 . ˜A 2 ˜S(t − s)[(u(s).∇)θ(s)]ds (cid:13) (cid:13) (cid:13)H0 , The Minkowski inequality implies that, for β ∈ (0, 1 2 ), (cid:90) t 0 ≤ T2(t) ≤ (cid:107) ˜A 1 2 ˜S(t − s)[(u(s).∇)θ(s)](cid:107)H0 ds (cid:90) t 0 (cid:107) ˜A1−β ˜S(t − s)(cid:107)L(H0;H0)(cid:107) ˜A−( 1 2 −β)[(u(s).∇)θ(s)(cid:107)H0 ds. Apply (8) with δ = 1 (cid:107) ˜Aρ f (cid:107)H0 ≤ (cid:107) ˜A 2 f (cid:107)2ρ 1 2 − β, α = 1 2 and ρ ∈ (β, 1 for any f ∈ H1. Therefore, H0 (cid:107) f (cid:107)1−2ρ H0 2 ). A simple computation proves that (cid:107) ˜A−( 1 2 −β)[(u(s).∇)θ(s)](cid:107)H0 ≤ C(cid:107)A ≤ C(cid:107)A This upper estimate and (33) imply that 1 2 u(s)(cid:107)V0 (cid:107) ˜Aρθ(s)(cid:107)H0 2 θ(s)(cid:107)2ρ 2 u(s)(cid:107)V0 (cid:107) ˜A 1 1 H0 (cid:107)θ(s)(cid:107)1−2ρ H0 . T2(t) ≤ C sup s∈[0,T] (cid:107)A 1 2 u(s)(cid:107)V0 sup s∈[0,t] (cid:107)θ(s)(cid:107)1−2ρ H0 (cid:90) t 0 (t − s)−1+β(cid:107) ˜A 1 2 θ(s)(cid:107)2ρ H0 ds. For p ∈ [1, ∞), Hölder’s inequality with respect to the measure (t − s)−(1−β)1[0,t)(s)ds implies that (cid:107)A 1 2 u(s)(cid:107)2p V0 sup s∈[0,t] (cid:107)θ(s)(cid:107)2p(1−2ρ) H0 (cid:16) (cid:90) t 0 (t − s)−(1−β)ds (cid:17)2p−1 T2(t)2p ≤ C sup s∈[0,t] (cid:16) (cid:90) t × (t − s)−(1−β)(cid:107) ˜A 0 1 2 θ(s)(cid:107)4pρ H0 ds (cid:17) . Mathematics 2022, 10, 4246 12 of 39 Let p1 = 2(1−ρ) (1−2ρ)2 and p3 = 1 pp1 = p(1 − 2ρ)p2 := ˜p. Young’s and Hölder’s inequalities imply that 1−2ρ , p2 = 2(1−ρ) 2ρ . Then, 1 p1 + 1 p3 + 1 p2 = 1, 4ρpp3 = 2p and T2(t)2p ≤ C (cid:104) 1 p1 (cid:107)A 1 2 u(s)(cid:107)2 ˜p V0 + 1 p2 sup s∈[0,t] (cid:107)θ(s)(cid:107)2 ˜p H0 sup s∈[0,t] (cid:16) (cid:90) t + 1 p3 (t − s)−1+β(cid:107) ˜A 1 2 θ(s)(cid:107)2p H0 ds (cid:17)(cid:16) (cid:90) t 0 (t − s)−1+βds (cid:17)p3−1(cid:105) . 0 Note that the continuous function ρ ∈ (0, 1 Given (cid:101) > 0, choose ρ ∈ (0, 1 choose β ∈ (0, ρ). The above computations yield 2 ) (cid:55)→ 2(1−ρ) 2 ) close enough to 0 to have 2 ˜p = 2p 2(1−ρ) 1−2ρ increases with limρ→0 2(1−ρ) 1−2ρ = 2. 1−2ρ = 4p + (cid:101), and then T2(t)2p ≤ C (cid:104) (cid:107)A 1 2 u(s)(cid:107)4p+(cid:101) V0 + sup s∈[0,t] (cid:107)θ(s)(cid:107)4p+(cid:101) H0 (cid:105) + C (t − s)−1+β(cid:107) ˜A 1 2 θ(s)(cid:107)2p H0 ds. (38) sup s∈[0,t] (cid:90) t 0 Finally, Burhholder’s inequality, the growth condition (16) and Hölder’s inequality imply that, for t ∈ [0, T], E(cid:16)(cid:13) (cid:13) (cid:13) (cid:90) t∧τN ˜A 1 0 (cid:13) (cid:13) (cid:13) 2p (cid:17) H0 ≤ Cp (cid:90) t∧τN 2 ˜S(t − s) ˜G(θ(s))d ˜W(s) (cid:0)Tr(Q)(cid:1)pE(cid:16)(cid:12) (cid:12) (cid:12) (cid:0)Tr(Q)(cid:1)pE(cid:16)(cid:12) (cid:12) ≤ Cp (cid:12) ≤ C(p, ˜K2, ˜K3, Tr(Q))T p(cid:104) 0 (cid:90) t∧τN 0 1 (cid:107) ˜A 2 ˜G(θ(s))(cid:107)2 L( ˜K;H0)ds p(cid:17) (cid:12) (cid:12) (cid:12) [ ˜K2 + ˜K3(cid:107)θ(s)(cid:107)2 1 + E(cid:16) sup s∈[0,T] (cid:105) (cid:107)θ(s)(cid:107)2p H0 H0 + ˜K3(cid:107) ˜A 1 2 θ(s)(cid:107)2 H0 (cid:3)ds p(cid:17) (cid:12) (cid:12) (cid:12) + Cp (cid:0)Tr(Q)(cid:1)p ˜K3 p T p−1 (cid:90) t 0 E(cid:0)(cid:107) ˜A 1 2 θ(s ∧ τN)(cid:107)2p H0 (cid:1)ds. (39) The upper estimates (38), (39) and T1(t) ≤ (cid:107)A instead of t imply that, for every t ∈ [0, T], 1 2 θ0(cid:107)H0 ≤ (cid:107)θ0(cid:107)H1 used with t ∧ ˜τN E(cid:0)(cid:107) ˜A 1 2 θ(t ∧ ˜τN)(cid:107)2p H0 (cid:1) ≤ Cp (cid:104) 1 + E(cid:16) (cid:107) ˜A 1 2 θ0(cid:107)2p H0 + sup s∈[0,T] (cid:107)A 1 2 u(s)(cid:107)4p+(cid:101) V0 + sup s∈[0,T] (cid:107)θ(s)(cid:107)4p+(cid:101) H0 (cid:17)(cid:105) + Cp (cid:90) t 0 (cid:2)(t − s)−1+β + ˜K3T p−1(cid:3)E(cid:0)(cid:107) ˜A 1 2 θ(s ∧ ˜τN)(cid:107)2p H0 (cid:1)ds, where the constant Cp does not depend on t and N. Theorem 1, Proposition 1 and the version of Gronwall’s lemma proved in the following lemma 3 imply that (37) for some ) and E((cid:107)θ0(cid:107)4p+(cid:101) constant C depending on E((cid:107)u0(cid:107)4p+(cid:101) ). The proof of the Lemma is com- plete. H0 V1 The following lemma is an extension of Lemma 3.3, p. 316 in [15]. For the sake of completeness, its proof is given at the end of this section. Lemma 3. Let (cid:101) ∈ (0, 1), a, b, c be positive constants and ϕ be a bounded non-negative function such that ϕ(t) ≤ a + (cid:2)b + c(t − s)−1+(cid:101)(cid:3) ϕ(s) ds, ∀t ∈ [0, T]. (40) (cid:90) t 0 Then, supt∈[0,T] ϕ(t) ≤ C for some constant C depending on a, b, c, T and (cid:101). Proof of Proposition 2. We next prove that the gradient of the temperature has bounded moments uniformly in time. Mathematics 2022, 10, 4246 13 of 39 We only prove (22) for p ∈ [2, +∞); the other argument is similar and easier. Applying the operator ˜A 1 2 to Equation (10), and writing Itô’s formula for the square of the corresponding H0-norm, we obtain 1 (cid:107) ˜A 2 θ(t)(cid:107)2 (cid:107) ˜Aθ(s)(cid:107)2 H0 ds = (cid:107) ˜A 1 2 θ0(cid:107)2 H0 − 2 (cid:104)(u(s).∇)θ(s), ˜Aθ(s)(cid:105)ds 2 ˜G(θ(s))d ˜W(s), ˜A 1 2 θ(s)(cid:1) + Tr(Q) 1 (cid:107) ˜A 2 ˜G(θ(s))(cid:107)2 H0 ds. (cid:90) t H0 + 2κ (cid:90) t + 2 0 0 (cid:0) ˜A 1 (cid:90) t 0 (cid:90) t 0 Then, apply Itô’s formula for the map x (cid:55)→ xp. This yields, using integration by parts, (cid:107) ˜A 1 2 θ(t)(cid:107)2p H0 + 2pκ (cid:90) t (cid:90) t 0 (cid:107) ˜Aθ(s)(cid:107)2 H0 (cid:107) ˜A 1 2 θ(s)(cid:107)2(p−1) H0 ds = (cid:107) ˜A 1 2 θ0(cid:107)2p H0 − 2p + 2p 0 (cid:90) t 0 (cid:104)(u(s).∇)θ(s), ˜Aθ(s)(cid:105)(cid:107) ˜A 1 2 θ(s)(cid:107)2(p−1) H0 ds (cid:0) ˜A 1 2 ˜G(θ(s))d ˜W(s), ˜A 1 2 θ(s)(cid:1)(cid:107) ˜A 1 2 θ(s)(cid:107)2(p−1) H0 + pTr( ˜Q) (cid:90) t 0 (cid:107) ˜A 1 2 ˜G(θ(s))(cid:107)2 H0 (cid:107) ˜A 1 2 θ(s)(cid:107)2(p−1) H0 ds + 2p(p − 1)Tr( ˜Q) (cid:90) t 0 (cid:107)(cid:0) ˜A 1 2 ˜G(cid:1)∗(θ(s))(cid:0) ˜A 1 2 θ(s)(cid:1)(cid:107)2 K(cid:107) ˜A 1 2 θ(s)(cid:107)2(p−2) H0 ds. (41) The Gagliardo–Nirenberg inequality (6) and the inclusion V1 ⊂ L4 implies that (cid:90) t 0 (cid:12)(cid:104)(u(s).∇)θ(s), ˜Aθ(s)(cid:105)(cid:12) (cid:12) (cid:12) (cid:107) ˜A 1 2 θ(s)(cid:107)2(p−1) H0 ds ≤ C ≤ C (cid:90) t 0 (cid:90) t 0 (cid:107) ˜Aθ(s)(cid:107)H0 (cid:107)u(s)(cid:107)L4 (cid:107) ˜A 1 2 θ(s)(cid:107)L4 (cid:107) ˜A 1 2 θ(s)(cid:107)2(p−1) H0 ds (cid:107) ˜Aθ(s)(cid:107) 3 2 H0 (cid:107)u(s)(cid:107)V1 (cid:107) ˜A 1 2 θ(s)(cid:107)2p− 3 2 H0 ds. Then, using the Hölder and Young’s inequalities, we deduce 2p (cid:90) t 0 (cid:12)(cid:104)(u(s).∇)θ(s), ˜Aθ(s)(cid:105)(cid:12) (cid:12) (cid:90) t ≤ (2p − 1) κ (cid:107) ˜A(θ(s))(cid:107)2 0 (cid:12)(cid:107) ˜A 1 2 θ(s)(cid:107)2(p−1) H0 ds 1 2 θ(s)(cid:107)2(p−1) H0 ds H0 (cid:107) ˜A (cid:90) t + C(κ, p) sup s∈[0,T] (cid:107)u(s)(cid:107)4 V1 0 (cid:107) ˜A 1 2 θ(s)(cid:107)2p H0 ds. (42) The growth condition (16) and Hölder’s and Young inequalities imply that (cid:90) t 0 (cid:107) ˜A 1 2 ˜G(θ(s))(cid:107)2 H0 (cid:107) ˜A 1 2 θ(s)(cid:107)2(p−1) H0 ds ≤ C (cid:90) t 0 (cid:2)1 + (cid:107)θ(s)(cid:107)2p H0 + (cid:107) ˜A 1 2 θ(s)(cid:107)2p H0 (cid:3)ds, (43) and a similar computation yields (cid:90) t 0 (cid:13) (cid:13) (cid:0) ˜A 1 2 ˜G(cid:1)∗(θ(s))(cid:0) ˜A 1 2 θ(s)(cid:1)(cid:107)2 K(cid:107) ˜A 1 2 θ(s)(cid:107)2(p−2) H0 ds ≤ C (cid:90) t 0 (cid:2)1 + (cid:107)θ(s)(cid:107)2p H0 + (cid:107) ˜A 1 2 θ(s)(cid:107)2p H0 (cid:3)ds. (44) Let ˜τN := inf{t ≥ 0 : (cid:107) ˜A t ∧ ˜τN instead of t imply 1 2 θ(t)(cid:107)H0 ≥ N}. The upper estimates (41)–(44) written for Mathematics 2022, 10, 4246 14 of 39 (cid:107) ˜A sup t∈[0,T] 1 2 θ(t ∧ ˜τN)(cid:107)2p H0 + κ (cid:90) T∧ ˜τN 0 (cid:107) ˜Aθ(s)(cid:107)2 H0 (cid:107) ˜A 1 2 θ(s)(cid:107)2(p−1) H0 ds ≤ (cid:107) ˜A 1 2 θ0(cid:107)2p H0 + C sup s∈[0,T] (cid:90) T∧ ˜τN (cid:107)u(s)(cid:107)4 V1 0 (cid:107) ˜A 1 2 θ(s)(cid:107)2p H0 ds + C 0 (cid:90) T∧ ˜τN (cid:0)1 + (cid:107)θ(s)(cid:107)2p H0 + (cid:107) ˜A 1 2 θ(s)(cid:107)2p H0 (cid:1)ds + 2p sup t∈[0,T] (cid:90) t∧ ˜τN 0 (cid:0) ˜A 1 2 ˜G(θ(s))d ˜W(s), ˜A 1 2 θ(s)(cid:1)(cid:107) ˜A 1 2 θ(s)(cid:107)2(p−1) H0 . Using the Cauchy–Schwarz inequality, Fubini’s theorem, (21) and (37), we deduce E(cid:16) (cid:107)u(s)(cid:107)4 V1 sup s∈[0,T] (cid:90) T∧ ˜τN 0 (cid:107) ˜A 1 2 θ(s)(cid:107)2p H0 ds (cid:17) (cid:110)E(cid:16) ≤ (cid:107)u(s)(cid:107)8 V1 sup s∈[0,T] (cid:17)(cid:111) 1 2 (cid:110) (cid:90) T 0 E(cid:0)(cid:107) ˜A 1 2 θ(s ∧ ˜τN)(cid:107)4p H0 (cid:1)ds (cid:111) 1 2 ≤ C. (45) The Davis inequality, the growth condition (16) and the Cauchy–Schwarz, Young and Hölder inequalities imply that E(cid:16) (cid:12) (cid:12) (cid:12) sup t∈[0,T] (cid:90) t∧ ˜τN 0 ≤ C E(cid:16)(cid:110) (cid:90) T ≤ C E(cid:104)(cid:0) sup 0 s≤T (cid:0) ˜A 1 2 ˜G(θ(s))d ˜W(s), ˜A 1 2 θ(s)(cid:1)(cid:107) ˜A 1 2 θ(s)(cid:107)2(p−1) H0 (cid:17) Tr( ˜Q)(cid:2) ˜K2 + ˜K3(cid:107)θ(s ∧ ˜τN)(cid:107)2 H1 (cid:1)(Tr( ˜Q)) (cid:0)(cid:107) ˜A 2 θ(s ∧ ˜τN)(cid:107)p H0 1 2 1 (cid:3)(cid:107) ˜A 1 2 θ(s ∧ ˜τN)(cid:107)4p−2 H0 ds (cid:111) 1 2 (cid:17) (cid:2) ˜K2 + ˜K3(cid:107)θ(s ∧ ˜τN)(cid:107)2 H0 + ˜K3(cid:107) ˜A (cid:17) 1 (cid:0)(cid:107) ˜A 2 θ(s ∧ ˜τN)(cid:107)2p H0 ≤ (cid:110) (cid:90) T × 0 E(cid:16) sup s≤T 1 4p + CE(cid:16) (cid:90) T 0 (cid:2)1 + (cid:107)θ(s ∧ ˜τN)(cid:107)2p H0 + (cid:107) ˜A 1 2 θ(s ∧ ˜τN)(cid:107)2p H0 (cid:3)ds (cid:17) . 1 2 θ(s ∧ ˜τN)(cid:107)2 H0 (cid:3)(cid:107) ˜A 1 2 θ(s ∧ ˜τN)(cid:107)2(p−1) H0 (cid:111) 1 2 (cid:17) ds Therefore, the upper estimates (20), (37) and (45) imply that E(cid:16) 1 2 sup s≤T (cid:107) ˜A 1 2 θ(s ∧ ˜τN)(cid:107)2p H0 (cid:17) + κ E(cid:16) (cid:90) T∧ ˜τN 0 (cid:107) ˜Aθ(s ∧ ˜τN)(cid:107)2 H0 (cid:107) ˜A 1 2 θ(s)(cid:107)2(p−1) H0 ds (cid:17) ≤ C for some constant C independent of N. As N → +∞, we deduce (22); this completes the proof of Proposition 3. We conclude this section with the proof of an extension of the Gronwall lemma. Mathematics 2022, 10, 4246 15 of 39 Proof of Lemma 3. For t ∈ [0, T], iterating (40) and using the Fubini theorem, we obtain ϕ(t) ≤ a + (cid:2)b + c(t − s)−1+(cid:101)] (cid:104) a + (cid:90) s 0 (cid:0)b + c(s − r)−1+(cid:101)(cid:1)ϕ(r)dr (cid:105) ds (cid:16) ≤ a 1 + [b + c(t − s)−1+(cid:101)]ds (cid:17) (cid:2)b + c(t − s)−1+(cid:101)][b + c(s − r)−1+(cid:101)]ds (cid:17) ϕ(r)dr (cid:90) t 0 (cid:90) t + 0 (cid:16)(cid:90) t (cid:90) t r 0 (cid:90) t (cid:104) ≤ A1 + ≤A1 + 0 (cid:90) t (cid:104) 0 b2(t − r) + 2bc (cid:101) (t − r)(cid:101) + c2 (cid:90) t r (t − s)−1+(cid:101)(s − r)−1+(cid:101)ds (cid:105) ϕ(r)dr B1 + C1(t − r)−1+2(cid:101) (cid:90) 1 0 λ−1+(cid:101)(1 − λ)−1+(cid:101)dλ (cid:105) ϕ(r)dr, for positive constants A1 (depending on a, b, c, T, (cid:101)), B1 (depending on b, c, T, (cid:101)) and C1 (depending on c and (cid:101)). One easily proves by induction on k that, for every integer k ≥ 1, ϕ(t) ≤ Ak + ≤ Ak + 0 (cid:90) t 0 (cid:90) t (cid:104) Bk + c Ck−1 (cid:90) t r (t − s)−1+k(cid:101)(s − r)−1+(cid:101)ds (cid:105) ϕ(r)dr (cid:2)Bk + Ck(t − r)−1+(k+1)(cid:101)(cid:3)ϕ(r)dr, for some positive constants Ak, Bk and Ck depending on a, b, c, T and (cid:101). Indeed, a change in variables implies that (cid:90) t r (t − s)−1+k(cid:101)(s − r)−1+(cid:101)ds = (t − r)−1+(k+1)(cid:101) λ−1+k(cid:101)(1 − λ)−1+(cid:101)dλ (cid:90) 1 0 = ˜Ck(t − r)−1+(k+1)(cid:101), for some constant ˜Ck depending on k and (cid:101). Let k∗ be the largest integer such that k(cid:101) < 1; that is, k∗(cid:101) < 1 ≤ (k∗ + 1)(cid:101). Then, since (t − r)−1+(k∗+1)(cid:101) ≤ T−1+(k∗+1)(cid:101), we deduce that ϕ(t) ≤ A + (cid:90) T 0 B ϕ(r)dr, for some positive constants A and B depending on the parameters a, b, c, T and (cid:101). The classical Gronwall lemma concludes the proof of the lemma. 5. Moment Estimates of Time Increments of the Solution In this section ,we prove moment estimates for various norms of time increments of the solution to (9) and (10). This will be crucial for deducing the speed of the convergence of numerical schemes. We first prove the time regularity of the velocity and temperature in L2. Proposition 4. Let u0, θ0 be F0-measurable; suppose that G and ˜G satisfy (C-u) and (C-θ), respectively. (i) Let u0 ∈ L4p(Ω; V1) and θ0 ∈ L2p(Ω; H0). Then for 0 ≤ τ1 < τ2 ≤ T, E(cid:0)(cid:107)u(τ2) − u(τ1)(cid:107)2p V0 (cid:1) ≤ C |τ2 − τ1|p. (46) (ii) Let u0 ∈ L8p+(cid:101)(Ω; V1), θ0 ∈ L8p+(cid:101)(Ω; H1) for some (cid:101) > 0. Then, for 0 ≤ τ1 < τ2 ≤ T, E(cid:0)(cid:107)θ(τ2) − θ(τ1)(cid:107)2p H0 (cid:1) ≤ C |τ2 − τ1|p. (47) Proof. Recall that S(t) = e−νtA is the analytic semi group generated by the Stokes operator A multiplied by the viscosity ν and that ˜S(t) = e−κt ˜A is the semi group generated by ˜A = −∆. We use the mild formulation of the solutions stated in (35) and (36). Mathematics 2022, 10, 4246 16 of 39 (i) Let 0 ≤ τ1 < τ2 ≤ T; then, u(τ2) − u(τ1) = ∑4 i=1 Ti(τ1, τ2), where T1(τ1, τ2) = S(τ2)u0 − S(τ1)u0 = (cid:2)S(τ2) − S(τ1)(cid:3)S(τ1)u0, T2(τ1, τ2) = T3(τ1, τ2) = T4(τ1, τ2) = (cid:90) τ2 0 (cid:90) τ2 0 (cid:90) τ2 0 S(τ2 − s)B(u(s), u(s))ds − (cid:90) τ1 S(τ2 − s)Πθ(s)v2ds − (cid:90) τ1 0 S(τ1 − s)B(u(s), u(s))ds, 0 S(τ1 − s)Πθ(s)v2ds (cid:90) τ1 0 S(τ2 − s)G(u(s))dW(s) − S(τ1 − s)G(u(s))dW(s). (48) The arguments used in the proof of Lemma 2.1 [11], using (7), (33), (34) and (21) yield E(cid:0)(cid:107)T1(τ1, τ2)(cid:107)2p V0 + (cid:107)T2(τ1, τ2)(cid:107)2p V0 (cid:1) ≤ C(cid:2)1 + E((cid:107)u0(cid:107)4p V1 )]|τ2 − τ1|p. (49) Let T3(τ1, τ2) = T3,1(τ1, τ2) + T3,2(τ1, τ2), where T3,1(τ1, τ2) = T3,2(τ1, τ2) = (cid:90) τ1 0 (cid:90) τ2 τ1 [S(τ2 − τ1) − Id]S(τ1 − s)(cid:2)Πθ(s)v2 (cid:3)ds, S(τ2 − s)(cid:2)Πθ(s)v2 (cid:3)ds. Since the family of sets {A(t, M)}t is decreasing, using the Minkowski inequality, (33) and (34), we obtain (cid:107)T3,1(τ1, τ2)(cid:107)V0 ≤ (cid:107)A 1 2 S(τ1 − s)(cid:107)L(V0;V0) (cid:107)A− 1 2 [S(τ2 − τ1) − Id](cid:107)L(V0;V0)(cid:107)Πθ(s)v2(cid:107)V0 ds (cid:90) τ1 0 (cid:12) (cid:12)τ2 − τ1 ≤ C 1 2 (cid:12) (cid:12) sup s∈[0,T] (cid:107)θ(s)(cid:107)H0 , and (cid:107)T3,2(τ1, τ2)(cid:107)V0 ≤ (cid:90) τ2 τ1 (cid:107)S(τ − s)(cid:2)Πθ(s)v2 (cid:3)(cid:107)V0 ds ≤ (cid:12) (cid:12)τ2 − τ1 (cid:12) (cid:12) sup s∈[0,T] (cid:107)θ(s)(cid:107)H0. The inequality (20) implies that E(cid:16) (cid:107)T3(τ1, τ2)(cid:107)2p V0 (cid:17) ≤ C (cid:12) (cid:12)τ2 − τ1|p E((cid:107)θ0(cid:107)2p H0 ). (50) Finally, decompose the stochastic integral as follows: T4,1(τ1, τ2) = [S(τ2 − τ1) − Id]S(τ1 − s)G(s)dW(s), T4,2(τ1, τ2) = (cid:90) τ1 0 (cid:90) τ2 τ1 S(τ2 − s)G(s)dW(s). The Burkholder inequality, (34), Hölder’s inequality and the growth condition (13) yield E(cid:16) (cid:107)[S(τ2 − τ1) − Id]S(τ1 − s)G(u(s))(cid:107)2 (cid:90) τ1 p(cid:17) (cid:17) V0Tr(Q)ds (cid:107)T4,1(cid:107)2p V0 (cid:12) (cid:12) (cid:12) 0 ≤ CpE(cid:16)(cid:12) (cid:12) (cid:12) ≤ C(Tr(Q))pE(cid:16)(cid:12) (cid:12) (cid:12) (cid:90) τ1 ≤ CE(cid:16)(cid:12) (cid:12) (cid:12) ≤ C(cid:2)1 + E((cid:107)u0(cid:107)2p 0 (cid:90) τ1 0 (cid:107)A− 1 2 [S(τ2 − τ1) − Id](cid:107)2 L(V0;V0)(cid:107)A p(cid:17) (cid:12) (cid:12)τ2 − τ1 (cid:12) (cid:12) (cid:2)K2 + K3(cid:107)u(s)(cid:107)2 V1 (cid:3)ds (cid:12) (cid:12) (cid:12) V1 )(cid:3)|τ2 − τ1|p, 1 2 G(u(s))(cid:107)2 V0 ds p(cid:17) (cid:12) (cid:12) (cid:12) (51) Mathematics 2022, 10, 4246 17 of 39 where the last upper estimate is a consequence of (19) and (21). A similar easier argument implies that E(cid:16) (cid:107)T4,2(cid:107)2p V0 (cid:17) (cid:107)S(τ2 − s)G(u(s))(cid:107)2 V0Tr(Q)ds p(cid:17) (cid:12) (cid:12) (cid:12) (cid:90) τ2 ≤ CpE(cid:16)(cid:12) (cid:12) (cid:12) ≤ C(cid:2)1 + E((cid:107)u0(cid:107)2p τ1 V0 )(cid:3) (cid:12) (cid:12)τ2 − τ1 p (cid:12) (cid:12) . (52) The inequalities (49)–(52) complete the proof of (46). (53) (54) (ii) As in the proof of (i), for 0 ≤ τ1 < τ2 ≤ T, let θ(τ2) − θ(τ1) = ∑3 ˜T1(τ1, τ2) = (cid:2) ˜S(τ2 − τ1) − Id(cid:3) ˜S(τ1)θ0, ˜T2(τ1, τ2) = − (cid:90) τ1 (cid:90) τ2 ˜S(τ2 − s)(cid:0)[u(s).∇]θ(s)(cid:1)ds + (cid:90) τ1 0 ˜S(τ2 − s) ˜G(θ(s))d ˜W(s) − 0 ˜T3(τ1, τ2) = (cid:90) τ2 0 ˜S(τ1 − s) ˜G(θ(s))d ˜W(s). 0 ˜S(τ1 − s)(cid:0)[u(s).∇]θ(s)(cid:1)ds ˜Ti(τ1, τ2), where i=1 The inequality (34) implies that (cid:107) ˜T1(τ1, τ2)(cid:107)H0 = (cid:107) ˜A− 1 2 (cid:2) ˜S(τ2 − τ1) − Id(cid:3) ˜S(τ1) ˜A 1 2 θ0(cid:107)H0 ≤ C (cid:12) (cid:12)τ2 − τ1| 1 2 (cid:107)θ0(cid:107)H1. Decompose ˜T2(τ1, τ2) = ˜T2,1(τ1, τ2) + ˜T2,2(τ1, τ2), where ˜T2,1(τ1, τ2) = − ˜T2,2(τ1, τ2) = − (cid:90) τ1 0 (cid:90) τ2 τ1 (cid:2) ˜S(τ2 − τ1) − Id] ˜S(τ1 − s) (cid:0)[u(s).∇]θ(s)(cid:3)ds, ˜S(τ2 − s) (cid:0)[u(s).∇]θ(s)(cid:1)ds. Let δ ∈ (0, 1 that 2 ); the Minkowski inequality, (33), (34) and (8) applied with α = ρ = 1 2 imply (cid:107) ˜T2,1(τ1, τ2)(cid:107)H0 ≤ (cid:90) τ1 0 (cid:107) ˜S(τ1 − s) (cid:2) ˜S(τ2 − τ1) − Id(cid:3)(cid:0)[u(s).∇]θ(s)(cid:1)(cid:107)H0 ds ≤ (cid:90) τ1 0 ≤ C (cid:90) τ1 0 1 2 +δ ˜S(τ1 − s)(cid:107)L(H0;H0)(cid:107) ˜A− 1 (cid:107) ˜A × (cid:107) ˜A−δ(cid:0)[u(s).∇]θ(s)(cid:107)H0 ds 1 (τ1 − s)−( 1 2 (cid:107)A 2 +δ) |τ2 − τ1| 2 [ ˜S(τ2 − τ1) − Id](cid:107)L(H0;H0) 1 2 u(s)(cid:107)V0 (cid:107) ˜A 1 2 θ(s)(cid:107)H0 ds ≤ C|τ2 − τ1| 1 2 (cid:107)A 1 2 u(s)(cid:107)V0 sup s∈[0,T] (cid:90) τ1 0 (τ1 − s)−( 1 2 +δ) (cid:107) ˜A 1 2 θ(s)(cid:107)H0 ds. (cid:1) and let δ ∈ (cid:0)0, 1 Let p1 ∈ (cid:0)2, 2 + (cid:101) 2 − 1 p1 2 + δ)p2 < 1. Thus, Hölder’s inequality for the finite measure (τ1 − s)−( 1 ( 1 with exponents 2p and 2p (cid:1). Let p2 be the conjugate exponent of p1; we have 2 +δ)1[0,τ1)(s)ds 2p−1 , and then, ds with conjugate exponents p1 and p2 imply 4p (cid:107) ˜T2,1(τ1, τ2)(cid:107)2p H0 ≤ C (cid:12) (cid:12)τ2 − τ1 (cid:12) (cid:12) p 1 2 u(s)(cid:107)2p V0 (cid:90) τ1 0 (τ1 − s)−( 1 2 +δ)(cid:107) ˜A 1 2 θ(s)(cid:107)2p H0 ds (cid:107)A sup s∈[0,T] (τ1 − s)−( 1 (cid:110) (cid:90) τ1 × 0 (cid:111)2p−1 2 +δ)ds ≤ C (cid:12) (cid:12)τ2 − τ1 (cid:12) (cid:12) p (cid:107)A 1 2 u(s)(cid:107)2p V0 sup s∈[0,T] (cid:110) (cid:90) τ1 × 0 (τ1 − s)−( 1 2 +δ)p2 ds (cid:107) ˜A (cid:110) (cid:90) τ1 0 (cid:111) 1 p2 . 1 2 θ(s)(cid:107)2pp1 H0 ds (cid:111) 1 p1 Mathematics 2022, 10, 4246 18 of 39 Since 2pp1 < 4p + (cid:101) with the upper estimates (21) and (37), imply that 2 and 2pp2 < 4p, Hölder’s inequality and Fubini’s theorem, together E(cid:0)(cid:107) ˜T2,1(τ1, τ2)(cid:107)2p H0 (cid:1) ≤ C (cid:12)τ2 − τ1|p(cid:110)E(cid:16) (cid:12) (cid:107)A 1 2 u(s)(cid:107)2pp2 V0 (cid:17)(cid:111) 1 p2 sup s∈[0,T] (cid:110) (cid:90) τ1 × 0 E (cid:0)(cid:107) ˜A 1 2 θ(s)(cid:107)2pp1 H0 (cid:1)ds (cid:111) 1 p1 ≤ C (cid:12) (cid:12)τ2 − τ1|p. (55) A similar argument proves that for η ∈ (0, 1), (cid:107) ˜T2,2(τ1, τ2)(cid:107)H0 ≤ (cid:107) ˜A1−η ˜S(τ2 − s)(cid:107)L(H0;H0)(cid:107) ˜A−(1−η)(cid:0)[u(s).∇]θ(s)(cid:1)(cid:107)H0 ds (cid:90) τ2 τ1 (cid:90) τ2 ≤ C ≤ C (τ2 − s)−1+η (cid:107)A τ1 (cid:12) (cid:12)τ2 − τ1|η sup s∈[0,T] 1 1 2 u(s)(cid:107)V0 (cid:107) ˜A (cid:90) τ2 1 2 θ(s)(cid:107)H0 ds (cid:107)A 2 u(s)(cid:107)V0 (τ2 − s)−1+η (cid:107) ˜A 1 2 θ(s)(cid:107)H0 ds. τ1 Let η ∈ (cid:0) p (cid:1) ∨ (cid:0) 8p+(cid:101) (cid:0) 1 4p+(cid:101) η 2p−1 , 1(cid:1); for (cid:101) > 0, let p1, p2 ∈ (1, +∞) be conjugate exponents such that (cid:1) < p1 < 2; then (1 − η)p2 < 1. Hölder’s inequality implies that (cid:107) ˜T2,2(τ1, τ2)(cid:107)2p H0 ≤ C (cid:12) (cid:12)τ2 − τ1|(2p−1)η sup s∈[0,T] (cid:107)A 1 2 u(s)(cid:107)2p V0 (cid:90) τ2 τ1 (τ2 − s)−1+η (cid:107) ˜A 1 2 θ(s)(cid:107)2p H0 ds ≤ C (cid:12) (cid:12)τ2 − τ1|(2p−1)η sup s∈[0,T] (cid:107)A 1 2 u(s)(cid:107)2p V0 (cid:110) (cid:90) τ2 τ1 (τ2 − s)−(1−η)p2 ds (cid:111) 1 p2 (cid:110) (cid:90) τ2 × τ1 (cid:107) ˜A 1 2 θ(s)(cid:107)2pp1 H0 (cid:111) 1 p1 . ds Since (2p − 1)η > p, 1 inequality together with the upper estimates (21) and (22) imply that η < 2; furthermore, 2pp2 < 4p + (cid:101) 2 and 2pp1 ≤ 4p. Hölder’s E(cid:0)(cid:107) ˜T2,2(τ1, τ2)(cid:107)2p H0 (cid:1) ≤ C (cid:12)τ2 − τ1|p (cid:110)E(cid:16) (cid:12) (cid:107)A 1 2 u(s)(cid:107)2pp2 V0 (cid:17)(cid:111) 1 p2 sup s∈[0,T] (cid:111) 1 p1 ≤ C (cid:12) (cid:12)τ2 − τ1|p. (cid:110) (cid:90) τ2 × τ1 E(cid:0)(cid:107) ˜A 1 2 θ(s)(cid:107)2pp1 H0 (cid:1)ds This inequality and (55) yield E(cid:0)(cid:107) ˜T2(τ1, τ2)(cid:107)2p H0 (cid:1) ≤ C (cid:12) (cid:12)τ2 − τ1|p. (56) (57) Finally, an argument similar to that used to prove (51) and (52), using the growth condition (16) and (20), implies that E(cid:0)(cid:107) ˜T3(τ1, τ2)(cid:107)2p H0 (cid:1) ≤ C (cid:12) (cid:12)τ2 − τ1 (cid:12) (cid:12) p . (58) The upper estimates (54), (57) and (58) complete the proof of (47). We next prove some time regularity for the gradient of the velocity and the temperature. Proposition 5. Let N ≥ 1 be an integer and, for k = 0, · · · , N, set tk = kT satisfy conditions (C-u) and (C-θ), respectively, and let η ∈ (0, 1). N , where G and ˜G Mathematics 2022, 10, 4246 19 of 39 (i) Let p ∈ [2, ∞), u0 ∈ L4p(Ω; V1) and θ0 ∈ L2p(Ω; H0). Then, there exists a positive constant C (independent of N) such that E(cid:16)(cid:12) (cid:12) (cid:12) N ∑ j=1 (cid:90) tj tj−1 (cid:2)(cid:107)u(s) − u(tj)(cid:107)2 V1 + (cid:107)u(s) − u(tj−1)(cid:107)2 V1 p(cid:17) (cid:3)ds (cid:12) (cid:12) (cid:12) ≤ C (cid:16) T N (cid:17)η p (ii) Let p ∈ [2, ∞), u0 ∈ L16p+(cid:101)(Ω; V1) and θ0 ∈ L16p+(cid:101)(Ω; H0) for some (cid:101) > 0. Then, E(cid:16)(cid:12) (cid:12) (cid:12) N ∑ j=1 (cid:90) tj tj−1 (cid:2)(cid:107)θ(s) − θ(tj)(cid:107)2 H1 + (cid:107)θ(s) − θ(tj−1)(cid:107)2 H1 p(cid:17) (cid:3)ds (cid:12) (cid:12) (cid:12) ≤ C (cid:16) T N (cid:17)η p (59) (60) Proof. (i) For j = 1, ..., N, write the decomposition (48) of u(tj) − u(s) used in the proof 1 of Lemma 4 (that is, τ1 = s, τ2 = tj), and apply A 2 . The upper estimates of the sum of 2 T2(s, tj) obtained in the proof of Lemma 2.2 in [11] imply that, for terms A η ∈ (0, 1), 2 T1(s, tj) and A 1 1 E(cid:16)(cid:12) (cid:12) (cid:12) N ∑ j=1 (cid:90) tj tj−1 (cid:2)(cid:107)A 1 2 T1(s, tj)(cid:107)2 V0 + (cid:107)A 1 2 T2(s, tj)(cid:107)2 V0 (cid:3)ds p(cid:17) (cid:12) (cid:12) (cid:12) ≤ C(E((cid:107)u0(cid:107)4p V1 )) (cid:17)η p . (cid:16) T N (61) The Minkowski inequality and the upper estimates (33) and (34) imply, for δ ∈ (0, 1 2 ) (cid:107)A 1 2 T3,1(s, tj)(cid:107)V0 ≤ (cid:90) tj 0 (cid:107)A 1 2 +δS(tj − s)(cid:107)L(V0;V0) (cid:107)A−δ(cid:2)S(tj − s) − Id (cid:3)(cid:107)L(V0;V0) × (cid:107)Πθ(s)v2(cid:107)V0 ds ≤ C|tj − s|δ sup s∈[0,tj] (cid:107)θ(s)(cid:107)H0 (cid:90) tj 0 (t1 − s)−( 1 2 +δ)ds, Hence, we deduce (cid:12) (cid:12) (cid:12) N ∑ j=1 (cid:90) tj tj−1 (cid:107)A 1 2 T3,1(s, tj)(cid:107)2 V0 ds p (cid:12) (cid:12) (cid:12) ≤ C (cid:16) T N (cid:17)2pδ (cid:107)θ(s)(cid:107)2p H0 sup s∈[0,T] (cid:90) tj tj−1 (cid:12) (cid:12) (cid:12) (cid:90) s 0 (s − r)−( 1 2 +δ)dr 2 (cid:12) (cid:12) (cid:12) p (cid:12) (cid:12) (cid:12) ds N ∑ j=1 (cid:17)2pδ × (cid:12) (cid:12) (cid:12) ≤ C (cid:16) T N (cid:107)θ(s)(cid:107)2p H0 (cid:12) (cid:12) (cid:12) (cid:90) T 0 s1−2δds p (cid:12) (cid:12) (cid:12) ≤ C (cid:16) T N (cid:17)2pδ sup s∈[0,T] (cid:107)θ(s)(cid:107)2p H0. sup s∈[0,T] Using the Minkowski inequality and (33) once more, we obtain (cid:12) (cid:12) (cid:12) N ∑ j=1 (cid:90) tj tj−1 1 (cid:107)A 2 T3,2(s, tj)(cid:107)2 V0 ds p (cid:12) (cid:12) (cid:12) ≤ (cid:12) (cid:12) (cid:12) N ∑ j=1 (cid:90) tj+1 tj (cid:12) (cid:12) (cid:12) (cid:90) tj s (cid:107)A 1 2 S(tj − r)Πθ(r)v2(cid:107)V0 dr 2 (cid:12) (cid:12) (cid:12) p (cid:12) (cid:12) (cid:12) ds ≤ C sup r∈[0,T] (cid:107)θ(r)(cid:107)2p H0 (cid:12) (cid:12) (cid:12) N ∑ j=1 (cid:90) tj tj−1 (cid:12) (cid:12) (cid:12) (cid:90) tj s (tj − s)− 1 2 dr 2 (cid:12) (cid:12) (cid:12) p (cid:12) (cid:12) (cid:12) ds ≤ C sup r∈[0,T] (cid:107)θ(r)(cid:107)2p H0 (cid:17)p . (cid:16) T N The above estimates of T3,1 and T3,2, together with (20), imply, for η ∈ (0, 1), that E(cid:16)(cid:12) (cid:12) (cid:12) N ∑ j=1 (cid:90) tj tj−1 (cid:107)A 1 2 T3(s, tj)(cid:107)2 V0 ds p(cid:17) (cid:12) (cid:12) (cid:12) ≤ C (cid:17)η p . (cid:16) T N (62) Mathematics 2022, 10, 4246 20 of 39 We next study the stochastic integrals. Using Hölder’s inequality, the Burkholder inequality, (33), (34) and the growth condition (13) twice, we obtain for δ ∈ (0, 1 2 ) E(cid:16)(cid:12) (cid:12) (cid:12) 2 T4,1(s, tj)(cid:107)2 2 T4,1(s, tj)(cid:107)2 ≤ N p−1 E(cid:16)(cid:12) (cid:12) (cid:12) V0 ds V0 ds (cid:107)A (cid:107)A (cid:90) tj p(cid:17) p(cid:17) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) 1 1 N ∑ j=1 tj−1 N ∑ j=1 ≤N p−1(cid:16) T N ≤ CpT p−1 (cid:90) tj (cid:17)p−1 N ∑ j=1 E(cid:16)(cid:12) (cid:12) (cid:12) (cid:90) tj tj−1 tj−1 E(cid:16) (cid:90) s N ∑ j=1 (cid:17)2δp (cid:90) T E(cid:16)(cid:13) (cid:13) (cid:13) (cid:90) s 0 AδS(s − r)A−δ[S(tj − s) − Id]A 1 2 G(u(r))dW(r) (cid:13) (cid:13) (cid:13) 2p (cid:17) V0 ds (cid:90) s 0 (s − r)−2δ(tj − s)2δ(cid:107)A 1 2 G(u(r))(cid:107)2 L(K,V0)Tr(Q)dr p(cid:17) (cid:12) (cid:12) (cid:12) ds ≤ C ≤ C (cid:16) T N (cid:16) T N (s − r)−2δ(cid:2)K2 + K3(cid:107)u(r)(cid:107)2 V1 (cid:16) (cid:90) T (s − r)−2δds (cid:107)u(r)(cid:107)2p V1 r 0 0 1 + E(cid:16) (cid:90) T 0 (cid:17)2δp(cid:104) p(cid:17) (cid:3)dr (cid:12) (cid:12) (cid:12) ds (cid:17)(cid:105) (cid:17) dr ≤ C (cid:17)2δp , (cid:16) T N (63) where the last upper estimates are deduced from the Fubini theorem, and from the upper estimates (19) and (21). A similar argument proves that E(cid:16)(cid:12) (cid:12) (cid:12) N ∑ j=1 (cid:107)A 1 2 T4,2(s, tj)(cid:107)2 V0 ds p(cid:17) (cid:12) (cid:12) (cid:12) ≤ T p−1 N ∑ j=1 (cid:90) tj tj−1 E(cid:16)(cid:13) (cid:13) (cid:13) (cid:90) tj s S(tj − s)A 1 2 G(u(r))dW(r) (cid:13) (cid:13) (cid:13) 2p (cid:17) V0 ds ≤ Cp T p−1Tr(Q)p N ∑ j=1 (cid:90) tj tj−1 E(cid:16)(cid:12) (cid:12) (cid:12) (cid:90) tj s (cid:2)K2 + K3(cid:107)u(r)(cid:107)2 V1 (cid:3)dr p(cid:17) (cid:12) (cid:12) (cid:12) ds ≤ C N ∑ j=1 (cid:90) tj tj−1 (cid:16) T N (cid:17)p−1 (cid:90) tj s (cid:2)K p 2 + K p 3 E((cid:107)A 1 2 u(r)(cid:107)2p V0 )(cid:3)drds ≤ C (cid:16) T N (cid:17)p−1 N ∑ j=1 (cid:90) tj tj−1 (cid:2)K p 2 + K p 3 E((cid:107)A 1 2 u(r)(cid:107)2p V0 )(cid:3)(cid:16) (cid:90) tj r ≤ C (cid:16) T N (cid:17)p(cid:104) 1 + (cid:90) T 0 E((cid:107)A 1 2 u(s)(cid:107)2p V0 )ds (cid:105) ≤ C (cid:17)p . (cid:16) T N The inequalities (63) and (64) imply that, for η ∈ (0, 1), E(cid:16)(cid:12) (cid:12) (cid:12) N ∑ j=1 (cid:90) tj tj−1 (cid:107)A 1 2 T4(s, tj)(cid:107)2 V0 ds p(cid:17) (cid:12) (cid:12) (cid:12) ≤ C (cid:17)η p . (cid:16) T N (cid:17) dr ds (64) (65) The above arguments (61), (62) and (65) prove similar inequalities when replacing Ti(s, tj) with Ti(tj−1, s) for i = 1, ..., 4 and j = 1, ..., N. Using (46), this concludes the proof of (59). Mathematics 2022, 10, 4246 21 of 39 (ii) As above, we apply ˜A 1 θ(tj) − θ(s) introduced in the proof of Proposition 4 (ii). For δ ∈ (0, 1 and (34) imply that 2 to the terms ˜Ti(s, tj), i = 1, 2, 3 of the decomposition (53) of 2 ), the inequalities (33) (cid:12) (cid:12) (cid:12) N ∑ j=1 (cid:90) tj tj−1 (cid:107) ˜A 1 2 ˜S(s)(cid:2) ˜S(τj − s) − Id(cid:3)θ0(cid:107)2 H0 p (cid:12) (cid:12) (cid:12) ≤ (cid:12) (cid:12) (cid:12) N ∑ j=1 (cid:90) tj tj−1 (cid:107) ˜Aδ ˜S(s) ˜A−δ(cid:2) ˜S(τj − s) − Id(cid:3) ˜A 1 2 θ0(cid:107)2 H0 p (cid:12) (cid:12) (cid:12) ≤ C ≤ C Hence, for η ∈ (0, 1), (cid:12) (cid:12) (cid:12) N ∑ j=1 (cid:16) T N (cid:90) tj tj−1 s−2δ(cid:16) T N (cid:17)2δ (cid:107) ˜A 1 2 θ0(cid:107)2 H0 ds p (cid:12) (cid:12) (cid:12) (cid:17)2δp (cid:107) ˜A 1 2 θ0(cid:107)2p H0 (cid:12) (cid:12) (cid:12) (cid:90) T 0 s−2δds p . (cid:12) (cid:12) (cid:12) E(cid:16)(cid:12) (cid:12) (cid:12) N ∑ j=1 (cid:90) tj tj−1 (cid:107) ˜A 1 2 ˜S(s)(cid:2) ˜S(τj − s) − Id(cid:3)θ0(cid:107)2 H0 p(cid:17) (cid:12) (cid:12) (cid:12) ≤ CE(cid:0)(cid:107)θ0(cid:107)2p H0 (cid:1) (cid:16) T N (cid:17)η p . (66) Let β ∈ (0, 1 2 ) and δ ∈ (0, 1 2 − δ). The Minkowski inequality, (33), (34) and (8) applied with α = ρ = 1 2 imply that, for s ∈ [tj−1, tj], (cid:13) (cid:13) (cid:13) (cid:90) s 0 ˜A 1 2 ˜S(s − r)(cid:2) ˜S(tj − s) − Id(cid:3)(cid:0)[u(r).∇]θ(r)(cid:1)dr (cid:13) (cid:13) (cid:13)H0 (cid:90) s ≤ 0 (cid:90) s 0 ≤ C (cid:107) ˜A 1 2 +β+δ ˜S(s − r) ˜A−β(cid:2) ˜S(tj − s) − Id(cid:3) ˜A−δ(cid:0)[u(r).∇]θ(r)(cid:1) (cid:107)H0 dr (s − r)−( 1 2 +β+δ)(cid:16) T N (cid:1)β(cid:107)A 1 2 u(r)(cid:107)V0 (cid:107) ˜A 1 2 θ(r)(cid:107)H0 dr. Therefore, using the Cauchy–Schwarz inequality and Fubini’s theorem, we obtain E(cid:16)(cid:12) (cid:12) (cid:12) N ∑ j=1 (cid:90) tj tj−1 (cid:107) ˜A 1 2 ˜T2,1(s, tj)(cid:107)2 H0 ds p(cid:17) (cid:12) (cid:12) (cid:12) ≤ C (cid:16) T N (cid:17)2βp E(cid:104) sup s∈[0,T] (cid:107)A 1 2 u(s)(cid:107)2p V0 (cid:12) (cid:12) (cid:12) (cid:16) (cid:90) s × (cid:17)2βp 0 E(cid:104) (s − r)−( 1 2 +β+δ)dr (cid:17) ds (cid:107)A 1 2 u(s)(cid:107)2p V0 (cid:12) (cid:12) (cid:12) sup s∈[0,T] tj N ∑ j=1 (cid:12) p(cid:105) (cid:12) (cid:12) (cid:16) (cid:90) T 0 ≤ C ≤ C (cid:16) T N (cid:16) T N (cid:90) tj+1 (cid:16) (cid:90) s 0 (s − r)−( 1 2 +β+δ)(cid:107) ˜A (cid:17) 1 2 θ(r)(cid:107)2 H0 1 (cid:107) ˜A 2 θ(r)(cid:107)2 H0 ds (cid:17)(cid:16) (cid:90) T r (s − r)−( 1 2 +β+δ)ds p(cid:105) (cid:17) (cid:12) (cid:12) (cid:12) dr (cid:17)2βp (cid:110)E(cid:16) (cid:107)A 1 2 u(s)(cid:107)4p V0 sup s∈[0,T] (cid:17)(cid:111) 1 2 (cid:110) (cid:90) T 0 E(cid:0)(cid:107) ˜A 1 2 θ(r)(cid:107)4p H0 (cid:1)dr (cid:111) 1 2 The upper estimates (21) and (37) imply, for η ∈ (0, 1), that E(cid:16)(cid:12) (cid:12) (cid:12) N ∑ j=1 (cid:90) tj tj−1 (cid:107) ˜A 1 2 ˜T2,1(s, tj)(cid:107)2 H0 ds p(cid:17) (cid:12) (cid:12) (cid:12) ≤ C (cid:17)η p . (cid:16) T N (67) Using the Minkowski inequality, (33) and (8) with α = ρ = 1 obtain, for δ ∈ (0, 1 2 ), 2 , and Fubini’s theorem, we Mathematics 2022, 10, 4246 22 of 39 (cid:12) (cid:12) (cid:12) N ∑ j=1 (cid:90) tj tj−1 (cid:107) ˜A 1 2 ˜T2,2(s, tj)(cid:107)2 H0 ds p (cid:12) (cid:12) (cid:12) ≤ (cid:12) (cid:12) (cid:12) N ∑ j=1 (cid:90) tj tj−1 (cid:12) (cid:12) (cid:12) (cid:90) tj s (tj − r)−( 1 2 +δ)(cid:107)A 1 2 u(r)(cid:107)V0 (cid:107) ˜A 1 2 θ(r)(cid:107)H0 dr 2 (cid:12) (cid:12) (cid:12) p (cid:12) (cid:12) (cid:12) ds ≤ C sup r∈[0,T] (cid:107)A 1 2 u(r)(cid:107)2p V0 (cid:12) (cid:12) (cid:12) N ∑ j=1 (cid:90) tj (cid:16) (cid:90) tj tj−1 s (tj − r)−( 1 2 +δ)(cid:107) ˜A (cid:16) (cid:90) tj × s (tj − r)−( 1 2 +δ)dr (cid:17) p (cid:12) (cid:12) (cid:12) ds 1 2 θ(s)(cid:107)2 H0 dr ≤ C sup r∈[0,T] (cid:107)A 1 2 u(r)(cid:107)2p V0 ≤ C sup r∈[0,T] (cid:107)A 1 2 u(r)(cid:107)2p V0 (cid:90) tj tj−1 (cid:17)p (cid:90) T (cid:12) (cid:12) (cid:12) N ∑ j=1 (cid:16) T N 0 (cid:107) ˜A 1 2 θ(r)(cid:107)2 H0 (cid:16) (cid:90) tj r (cid:17) p (cid:12) (cid:12) (cid:12) dr ds (cid:107) ˜A 1 2 θ(s)(cid:107)2p H0 dr Using the Cauchy–Schwarz inequality, (21) and (37), we obtain E(cid:16)(cid:12) (cid:12) (cid:12) N ∑ j=1 (cid:90) tj tj−1 (cid:107) ˜A 1 2 ˜T2,2(s, tj)(cid:107)2 H0 ds p(cid:17) (cid:12) (cid:12) (cid:12) ≤ C (cid:17)p . (cid:16) T N Finally, arguments similar to those used to prove (65) imply, for η ∈ (0, 1), that E(cid:16)(cid:12) (cid:12) (cid:12) N ∑ j=1 (cid:90) tj tj−1 (cid:107) ˜A 1 2 ˜T3(s, tj)(cid:107)2 V0 ds p(cid:17) (cid:12) (cid:12) (cid:12) ≤ C (cid:17)η p . (cid:16) T N (cid:17) (68) (69) The upper estimates (66)–(69) conclude the proof of E(cid:16)(cid:12) (cid:12) (cid:12) N ∑ j=1 (cid:90) tj tj−1 (cid:107) ˜A 1 2 (cid:2)θ(tj) − θ(s)(cid:3)(cid:107)2 H0 ds p(cid:17) (cid:12) (cid:12) (cid:12) ≤ C (cid:16) T N (cid:17)η p , η ∈ (0, 1). Using (47), a similar argument completes the proof of (60). 6. The Implicit Time Euler Scheme We first prove the existence of the fully time-implicit time Euler scheme {uk; k = lW := W(tl) − W(tl−1) 0, 1, ..., N} and {θk; k = 0, 1, ..., N} defined by (17) and (18). Set ∆ and ∆ l ˜W = ˜W(tl) − ˜W(tl−1), l = 1, ..., N. 6.1. Existence of the Scheme Proof of Proposition 3. The proof is divided into two steps. Step 1 For technical reasons, we consider a Galerkin approximation. Let {el}l denote an orthonormal basis of V0 made of elements of V2 that are orthogonal in V1 (resp., let { ˜el}l denote an orthonormal basis of H0 made of elements of H2 that are orthogonal in H1). For m = 1, 2, ..., let Vm = span (e1, . . . , em) ⊂ V2 and let Pm : V0 → Vm denote the projection from V0 to Vm. Similarly, let ˜Hm = span ( ˜e1, ..., ˜em) ⊂ H2 and let ˜Pm : H0 → ˜Hm denote the projection from H0 to ˜Hm. Mathematics 2022, 10, 4246 23 of 39 In order to find a solution to (17) and (18), we project these equations onto Vm and ˜Hm, respectively, which we define by induction as {uk(m)}k=0,...,N ∈ Vm and {θk(m)}k=0,...,N ∈ ˜Hm such that u0(m) = Pm(u0), θ0(m) = ˜Pm(θ0), and, for k = 1, ..., N, ϕ ∈ Vm and ψ ∈ ˜Hm, (cid:0)uk(m) − uk−1(m), ϕ(cid:1) + h (cid:104) ν(cid:0)A 1 2 uk(m), A 1 2 ϕ) + (cid:10)B(cid:0)uk(m), uk(m)(cid:1), ϕ(cid:11) = h(cid:0)Πθk−1v2, ϕ(cid:1) + (cid:0)G(uk−1(m))∆ kW , ϕ(cid:1) (cid:0)θk(m) − θk−1(m), ψ(cid:1) + h (cid:104) 1 1 κ(cid:0) ˜A 2 θk(m), ˜A = (cid:0) ˜G(θk−1(m))∆ 2 ψ) + (cid:10)[uk−1k(m).∇]θk(m)(cid:1), ψ(cid:11) k ˜W , ψ(cid:1) (70) (71) For almost every ω set, R(0, ω) = (cid:107)u0(ω)(cid:107)V0 and ˜R(0, ω) = (cid:107)θ0(ω)(cid:107)H0. Fix k = 1, ..., N and suppose that, for j = 0, . . . , k − 1, the Ftj - measurable random variables uj(m)and θ j(m) have been defined, and that R(j, ω) := sup m≥1 (cid:107)uj(m, ω)(cid:107)L2 < ∞ and ˜R(j, ω) := sup m≥1 (cid:107)θ j(m, ω)(cid:107)L2 < ∞ for almost every ω. We prove that uk(m) and θk(m) exist and satisfy supm≥1 (cid:107)uk(m, ω)(cid:107)V0 < ∞ and supm≥1 (cid:107)θk(m, ω)(cid:107)H0 < ∞ a.s. For ω ∈ Ω, let Φk m,ω : Vm → Vm (resp., ˜Φk m,ω) be defined for f ∈ Vm (resp., for ˜f ∈ ˜Hm) as the solution of (cid:104) ν(cid:0)A 1 2 f , A (cid:0)Φk m,ω( f ), ϕ(cid:1) = (cid:0) f − uk−1(m, ω), ϕ(cid:1) + h − (cid:0)Πθk−1(m)v2, ϕ(cid:1)(cid:105) m,ω( ˜f ), ψ(cid:1) = (cid:0) ˜f − θk−1(m, ω), ψ(cid:1) + h − (cid:0) ˜Pm ˜G(θk−1(m, ω))∆ 1 κ(cid:0) ˜A 2 ˜f , ˜A k ˜W(ω), ψ(cid:1), − (cid:0)PmG(uk−1(m, ω))∆ (cid:104) 1 (cid:0) ˜Φk ∀ψ ∈ ˜Hm. 2 ψ(cid:1) + (cid:10)[uk−1(m). ˜A 1 2 ] ˜f ], ψ(cid:11) 1 2 ϕ(cid:1) + (cid:10)PmB( f , f ), ϕ(cid:11) kW(ω), ϕ(cid:1), ∀ϕ ∈ Vm, Then, the Cauchy–Schwarz and Young inequalities imply (cid:12) (cid:12) (cid:0)uk−1(m, ω), f (cid:1)(cid:12) (cid:12) ≤ (cid:12) (cid:12) (cid:0)θk−1(m, ω), ˜f (cid:1)(cid:12) (cid:12) ≤ (cid:12) (cid:12) (cid:0)Πθk−1(m, ω), f (cid:1)(cid:12) (cid:12) ≤ (cid:12) (cid:12) (cid:0)G(uk−1(m, ω))∆ kW, f (cid:1) ≤ ≤ (cid:12) (cid:12) (cid:0) ˜G(θk−1(m, ω))∆ k ˜W, ˜f (cid:1) ≤ ≤ 1 4 1 4 1 4 1 4 1 4 1 4 1 4 (cid:107) f (cid:107)2 V0 + (cid:107)uk−1(m, ω)(cid:107)2 V0, (cid:107) ˜f (cid:107)2 H0 + (cid:107)θk−1(m, ω)(cid:107)2 H0, (cid:107) f (cid:107)2 V0 + (cid:107)θk−1(m, ω)(cid:107)2 H0, kW(cid:107)2 K (cid:107) f (cid:107)2 L(K,V0)(cid:107)∆ (cid:3)(cid:107)∆ V0 V0 + (cid:107)G(uk−1(m, ω))(cid:107)2 V0 + (cid:2)K0 + K1(cid:107)uk−1(m, ω)(cid:107)2 H0 + (cid:107) ˜G(θk−1(m, ω))(cid:107)2 H0 + (cid:2) ˜K0 + ˜K1(cid:107)uk−1(m, ω)(cid:107)2 (cid:107) f (cid:107)2 (cid:107) ˜f (cid:107)2 (cid:107) ˜f (cid:107)2 L( ˜K,H0)(cid:107)∆ k ˜W(cid:107)2 K (cid:3)(cid:107)∆ k ˜W(cid:107)2 ˜K. H0 kW(cid:107)2 K, If (cid:107) f (cid:107)2 V0 = R2(k, ω) := 4 (cid:107) ˜f (cid:107)2 H0 = ˜R2(k, ω) := 2 (cid:104) (cid:104) R2(k − 1, ω) + (cid:0)h ˜R(k − 1, ω)(cid:1)2 + (cid:2)K0 + K1 R2(k − 1, ω)(cid:3)(cid:107)∆ kW(ω)(cid:107)2 K ˜R2(k − 1, ω) + (cid:0) ˜K0 + ˜K1 ˜R2(k − 1, ω)(cid:1)(cid:107)∆ (cid:17)(cid:105) , k ˜W(ω)(cid:107)2 ˜K (cid:105) , Mathematics 2022, 10, 4246 24 of 39 we deduce (cid:0)Φk m,ω( f ), f (cid:1) ≥ (cid:0) ˜Φk m,ω( f ), ˜f (cid:1) ≥ 1 4 1 2 1 2 f (cid:107)2 V0 − h2(cid:107)θk−1(m, ω)(cid:107)2 H0 L2 + hν(cid:107)A (cid:3)(cid:107)∆ L2 − (cid:107)uk−1(m, ω)(cid:107)2 (cid:107) f (cid:107)2 − (cid:2)K0 + K1(cid:107)uk−1(m, ω)(cid:107)2 V0 H0 − (cid:107)θk−1(m, ω)(cid:107)2 (cid:107) ˜f (cid:107)2 H0 + h(cid:107) ˜A (cid:3)(cid:107)∆ − (cid:2) ˜K0 + ˜K1(cid:107)θk−1(m, ω)(cid:107)2 H0 kW(ω)(cid:107)2 K ≥ 0 1 2 ˜f (cid:107)2 H0 k ˜W(ω)(cid:107)2 ˜K ≥ 0. Using ([16], Cor 1.1) page 279, which can be deduced from Brouwer’s theorem, we deduce the existence of an element uk(m, ω) ∈ V(m) (resp., θk(m, ω) ∈ ˜H(m)), such that Φk(m, ω)(uk(m, ω)) = 0 (resp., ˜Φk(m, ω)(θk(m, ω)) = 0) and (cid:107)uk(m, ω)(cid:107)2 V0 ≤ R2(k, ω) (resp., (cid:107)θk(m, ω)(cid:107)H0 ≤ ˜R2(k, ω)) a.s. Note that these elements uk(m, ω) and θk(m, ω) need not be unique. Furthermore, the random variables uk(m) and θk(ω) are Ftk -measurable. The definition of uk(m) (resp., θk(m)) implies that it is a solution to (70) (resp., (71)). Taking ϕ = uk(m) in (70), using the antisymmetry property (3) and the Young inequality, we obtain (cid:107)uk(m)(cid:107)2 V0 + h ν(cid:107)A + (cid:0)G(uk−1(m)∆ kW, uk(m)(cid:1) V0 + (cid:107)uk−1(m)(cid:107)2 (cid:107)uk(m)(cid:107)2 ≤ 3 4 Hence, a.s., 1 2 uk(m)(cid:107)2 V0 = (cid:0)uk−1(m), uk(m)(cid:1) + h (cid:0)Πθk−1(m)v2, uk(m)(cid:1) V0 + (cid:107)θk−1(m)(cid:107)2 H0 + (cid:2)K0 + K1(cid:107)uk−1(m)(cid:107)2 V0 (cid:3)(cid:107)∆ kW(cid:107)2 K. (cid:104) 1 4 sup m≥1 (cid:107)uk(m, ω)(cid:107)2 V0 + h ν(cid:107)A 1 2 uk(m, ω)(cid:107)2 V0 (cid:105) ≤ R2(k − 1, ω) + ˜R2(k − 1, ω) + (cid:2)K0 + K1R2(k − 1, ω)(cid:3)(cid:107)∆ kW(ω)(cid:107)2 K. A similar computation using ψ = θk(m) in (71) implies that (cid:104) 1 2 sup m≥1 (cid:107)θk(m, ω)(cid:107)2 H0 + h κ(cid:107) ˜A 1 2 θk(m, ω)(cid:107)2 H0 (cid:105) ≤ ˜R2(k − 1) + (cid:2) ˜K0 + ˜K1 ˜R2(k − 1)(cid:3)(cid:107)∆ k ˜W(cid:107) ˜K. Therefore, for fixed k and almost every ω, the sequence {uk(m, ω)}m is bounded in V1; it has a sub-sequence (still denoted as {uk(m, ω)}m) that converges weakly in V1 to φk(ω). The random variable φk is Ftk -measurable. Similarly, for fixed k and almost every ω, the sequence {θk(m, ω)}m is bounded in H1; it has a sub-sequence (still denoted as {θk(m, ω)}m) that converges weakly in H1 to ˜φk(ω), which is Ftk -measurable. Since D is bounded, the embedding of V1 in V0 (resp., of H1 in H0) is compact; hence, the sub-sequence {uk(m, ω)}m converges strongly to φk(ω) in V0 (resp., {θk(m, ω)}m con- verges strongly to ˜φk(ω) in H0). Step 2 We next prove that the pair (φk, ˜φk) is a solution to (17) and (18). By definition, u0(m) converges strongly to u0 in V0, and θ0(m) converges strongly to θ0 in H0. We next prove by induction on k that the pair (φk, ˜φk) solves (17) and (18). Fix a positive integer m0 and consider the Equation (70) for k = 1, . . . , N, ϕ ∈ Vm0 and m ≥ m0. As m → ∞, we have, a.s., (cid:0)uk(m) − uk−1(m), ϕ) → (cid:0)φk − φk−1, ϕ), (cid:0)Πθk−1(m)v2, ϕ(cid:1) = (cid:0)θk−1(m)v2, ϕ(cid:1) → (cid:0) ˜φkv2, ϕ). (cid:0)A 1 2 uk(m), A 1 2 φ(cid:1) → (cid:0)A 1 2 φk, A 1 2 φ(cid:1), Mathematics 2022, 10, 4246 25 of 39 Furthermore, the antisymmetry of B (3) and the Gagliardo–Nirenberg inequality (6) yield, a.s., (cid:12) (cid:12) (cid:10)B(cid:0)uk(m), uk(m)(cid:1) − B(φk, φk), ϕ(cid:11)(cid:12) (cid:12) (cid:10)B(cid:0)uk(m) − φk, ϕ(cid:1), uk(m)(cid:11)(cid:12) 2 ϕ(cid:107)V0 (cid:107)uk(m) − φk(cid:107)L4 ≤ (cid:107)A ≤ (cid:12) (cid:12) 1 (cid:12) + (cid:12) (cid:12) (cid:10)B(cid:0)φk, ϕ(cid:1), uk(m) − φk(cid:11)(cid:12) (cid:12) (cid:2)(cid:107)uk(m)(cid:107)L4 + (cid:107)φk(cid:107)L4 (cid:3)](cid:107)A 1 (cid:3) 2 uk(m) − φk(cid:107) (cid:107)uk(m)(cid:107)V1 + (cid:107)φk(cid:107)V1 ≤ C (cid:107)ϕ(cid:107)V0 (cid:2) max m as m → ∞. 1 2 V0 (cid:107)uk(m) − φk(cid:107) 1 2 V0 → 0 Finally, the Cauchy–Schwarz inequality and the Lipschitz condition (12) imply that (cid:12) (cid:12) (cid:0)(cid:2)G(cid:0)uk−1(m)(cid:1) − G(cid:0)φk−1(cid:1)(cid:3)∆ kW, ϕ(cid:1)(cid:12) (cid:12) ≤ (cid:107)ϕ(cid:107)V0 (cid:107)G(uk−1(m) − G(φk−1)(cid:107)L(K;V0)(cid:107)∆ ≤ (cid:112) L1 (cid:107)ϕ(cid:107)L2 (cid:107)uk−1(m) − φk−1(cid:107)L2 (cid:107)∆ kW(cid:107)K kW(cid:107)K → 0 as m → ∞. Therefore, letting m → ∞ in (70), we deduce that (cid:16) φk − φk−1 + hνAφk + hB(cid:0)φk, φk(cid:1), ϕ (cid:17) = (cid:0)Πθk−1v2, ϕ(cid:1) + (cid:0)G(φk−1)∆ kW , ϕ), ∀ϕ ∈ Vm0. Since ∪m0 Vm0 is dense in V, we deduce that {φk}k=0,...,N is a solution to (17). A similar argument proves that ˜φk is a solution to (18). This concludes the proof. 6.2. Moments of the Euler Scheme We next prove the upper bounds of moments of uk and θk uniformly in k = 1, . . . , N. Proposition 6. Let G and ˜G satisfy the condition (C-u)(i) and (C-θ)(i), respectively. Let K ≥ 1 be an integer, and let u0 ∈ L2K (Ω; V0) and θ0 ∈ L2K (Ω; H0), respectively. Let {uk}k=0,...,N and {θk}k=0,...,N be the solution of (17) and (18), respectively. Then, E(cid:16) sup N≥1 max 0≤L≤N (cid:107)uL(cid:107)2K V0 + max 0≤L≤N (cid:107)θL(cid:107)2K H0 (cid:17) < ∞ E(cid:16) h N ∑ l=1 sup N≥1 (cid:107)A 1 2 ul(cid:107)2 V0 (cid:107)ul(cid:107)2K−2 V0 + h N ∑ l=1 (cid:107) ˜A 1 2 θl(cid:107)2 H0 (cid:107)θl(cid:107)2K−2 H0 < ∞, (72) (73) Proof. Write (17) with ϕ = ul, (18) with ψ = θl and use the identity ( f , f − g) = 1 2 (cid:107)g(cid:107)2 (3) and the growth condition (11) yields, for l = 1, . . . , N, (cid:2)(cid:107) f (cid:107)L2 − (cid:3). Using the Cauchy–Schwarz and Young inequalities, the antisymmetry L2 + (cid:107) f − g(cid:107)2 L2 1 2 1 2 (cid:2)(cid:107)ul(cid:107)2 V0 − (cid:107)ul−1(cid:107)2 (cid:2)(cid:107)θl(cid:107)2 H0 − (cid:107)θl−1(cid:107)2 1 (cid:3) + hν(cid:107)A V0 + (cid:107)ul − ul−1(cid:107)2 2 ul(cid:107)2 V0 V0 = h(Πθl−1e2, ul) + (cid:0)G(ul−1)∆ lW, ul), H0 = (cid:0) ˜G(θl−1)∆ 2 θl(cid:107)2 H0 + (cid:107)θl − θl−1(cid:107)2 H0 (cid:3) + hκ(cid:107) ˜A 1 (74) (75) l ˜W, θl). Mathematics 2022, 10, 4246 26 of 39 Fix L = 1, ..., N and add both equalities for l = 1, ..., L; this yields (cid:2)(cid:107)uL(cid:107)2 V0 − (cid:107)u0(cid:107)2 V0 + (cid:107)θL(cid:107)2 H0 − (cid:107)θ0(cid:107)2 H0 1 2 (cid:3) + 1 2 (cid:104) L ∑ l=1 (cid:107)ul − ul−1(cid:107)2 V0 + (cid:107)θl − θl−1(cid:107)2 H0 (cid:105) L ∑ l=1 + h L ∑ l=1 L ∑ l=1 L ∑ l=1 + + (cid:2)ν(cid:107)A 1 2 ul(cid:107)2 V0 + κ(cid:107) ˜A 1 2 θl(cid:107)2 H0 (cid:3) ≤ h 2 L−1 ∑ l=0 (cid:107)θl(cid:107)2 H0 + h (cid:2)(cid:107)G(ul−1)(cid:107)2 L(K;V0)(cid:107)∆ (cid:2)(cid:107) ˜G(θl−1)(cid:107)2 L( ˜K;H0)(cid:107)∆ lW(cid:107)2 K + l ˜W(cid:107)2 ˜K + 1 4 1 4 (cid:107)ul − ul−1(cid:107)2 V0 (cid:3) + (cid:107)θl − θl−1(cid:107)2 H0 (cid:3) + L−1 ∑ l=1 L ∑ l=1 L ∑ l=1 (cid:107)ul(cid:107)2 V0 + h(cid:107)uL(cid:107)2 V0 (cid:0)G(ul−1)∆ lW, ul−1(cid:1) (cid:0) ˜G(θl−1)∆ l ˜W, θl−1(cid:1). (76) Let N be large enough to have h = T N ≤ 1 8 . Taking the expected values, we obtain E(cid:16) (cid:107)uL(cid:107)2 V0 + (cid:107)θL(cid:107)2 H0 + 1 2 L ∑ l=1 (cid:2)(cid:107)ul − ul−1(cid:107)2 V0 + (cid:107)θl − θl−1(cid:107)2 V0 (cid:3) + 2h (cid:2)ν(cid:107)A N ∑ l=1 1 2 ul(cid:107)2 V0 + κ(cid:107) ˜A (cid:3)(cid:17) 1 2 θl(cid:107)2 H0 ≤ E(cid:0)(cid:107)u0(cid:107)2 V0 + (cid:107)θ0(cid:107)2 H0 (cid:1) + 2T(cid:2)K0Tr(Q) + ˜K0Tr( ˜Q)(cid:3) + h(cid:2)4 + 2 max(K1Tr(Q), ˜K1Tr( ˜Q)(cid:3) L−1 ∑ l=0 E(cid:0)(cid:107)ul(cid:107)2 V0 + (cid:107)θl(cid:107)2 H0 (cid:1). Neglecting both sums in the left hand side and using the discrete Gronwall lemma, we deduce that E(cid:0)(cid:107)uL(cid:107)2 V0 + (cid:107)θL(cid:107)2 H0 (cid:1) ≤ C, (77) sup 1≤L≤N where (cid:16) C = 2E(cid:0)(cid:107)u0(cid:107)2 V0 + (cid:107)θ0(cid:107)2 H0 (cid:1) + 2T(cid:2)K0Tr(Q) + ˜K0Tr( ˜Q)(cid:3)(cid:17) eT (cid:2) 4+2 max(K1Tr(Q), ˜K1Tr( ˜Q)(cid:3) is independent of N. This implies E(cid:16) N ∑ l=1 sup N≥1 (cid:2)(cid:107)Aul(cid:107)2 V0 + (cid:107) ˜Aθl(cid:107)2 H0 (cid:3) + (cid:107)ul − ul−1(cid:107)2 V0 + (cid:107)θl − θl−1(cid:107)2 H0; (cid:17) < ∞, which proves (73) for K = 1. For s ∈ [tj, tj+1), j = 0, . . . , N − 1, and set s = tj. The Davis inequality, and then the Cauchy-Schwarz and Young inequalities, imply that for any (cid:101) > 0, Mathematics 2022, 10, 4246 27 of 39 E(cid:16) max 1≤L≤N L ∑ l=1 (cid:2)(cid:0)G(ul−1))∆ lW, ul−1(cid:1) + (cid:0) ˜G(θl−1∆ l ˜W, θl−1(cid:1)(cid:3)(cid:17) ≤ E(cid:16) (cid:90) t 0 sup t∈[0,T] (cid:0)G(us)dW(s), us(cid:1)(cid:17) + E(cid:16) (cid:90) t 0 sup t∈[0,T] (cid:0) ˜G(θs)d ˜W(s), θs(cid:1)(cid:17) 0 (cid:90) T ≤ 3E(cid:16)(cid:12) (cid:12) (cid:12) + 3E(cid:16)(cid:12) (cid:12) (cid:12) 2 E(cid:104) ≤ 3Tr(Q) 1 (cid:90) T 0 (cid:107)G(us)(cid:107)2 L(K;V0)(cid:107)us(cid:107)2 V0Tr(Q)ds 1 2 (cid:17) (cid:12) (cid:12) (cid:12) (cid:107) ˜G(θs)(cid:107)2 L( ˜K;H0)(cid:107)θs(cid:107)2 H0Tr( ˜Q)ds 1 2 (cid:17) (cid:12) (cid:12) (cid:12) (cid:107)ul(cid:107)V0 (cid:16) h max 1≤l≤N [K0 + K1(cid:107)ul−1(cid:107)2 V0 (cid:3)(cid:17) 1 2 (cid:105) N ∑ l=0 (cid:16) + 3Tr( ˜Q) 1 2 E(cid:104) max 1≤l≤N (cid:107)θl(cid:107)H0 h [ ˜K0 + ˜K1(cid:107)θl−1(cid:107)2 H0 (cid:3)(cid:17) 1 2 (cid:105) N ∑ l=0 ≤ (cid:101)E(cid:16) max 1≤l≤N (cid:107)ul(cid:107)2 V0 (cid:17) + E(cid:0)(cid:107)u0(cid:107)2 V0 ) + 9 4(cid:101) Tr(Q) h N ∑ l=1 (cid:2)K0 + K1 E((cid:107)ul−1(cid:107)2 V0 )(cid:3) + (cid:101)E(cid:16) max 1≤l≤N (cid:107)θl(cid:107)2 H0 (cid:17) + E(cid:0)(cid:107)θ0(cid:107)2 H0 ) + 9 4(cid:101) Tr( ˜Q) h N ∑ l=1 (cid:2) ˜K0 + ˜K1 E((cid:107)θl−1(cid:107)2 H0 )(cid:3). (78) Taking the maximum over L in (76) and using (78), we deduce E(cid:16) max 1≤L≤N (cid:2)(cid:107)ul(cid:107)2 V0 + θl(cid:107)2 H0 (cid:3)(cid:17) ≤ 2E(cid:0)(cid:107)u(cid:107)2 V0 + (cid:107)θ0(cid:107)2 H0 (cid:1) + h (cid:0)(cid:107)θl−1(cid:107)2 H0 + (cid:107)ul−1(cid:107)2 V0 (cid:1) N ∑ l=1 (cid:2)K0 + K1 E((cid:107)ul−1(cid:107)2 V0 )(cid:3) + 2(cid:101)E(cid:16) max 1≤L≤N (cid:2)(cid:107)ul(cid:107)2 V0 + (cid:107)θL(cid:107)2 H0 (cid:3)(cid:17) + 9 4(cid:101) Tr(Q) h N ∑ l=1 + 9 4(cid:101) Tr( ˜Q) h N ∑ l=1 (cid:2) ˜K0 + ˜K1 E((cid:107)θl−1(cid:107)2 H0 )(cid:3). For (cid:101) = 1 4 , (77) proves that (cid:104)E(cid:16) sup N≥1 sup 1≤L≤N (cid:107)uL(cid:107)2 V0 (cid:17) + E(cid:16) sup 1≤L≤N (cid:107)θL(cid:107)2 H0 (cid:17)(cid:105) < ∞, which proves (72) for K = 1. We next prove (72) and (73) by induction on K. Multiply (74) by (cid:107)ul(cid:107)2 H0. Using the identity a(a − b) = 1 by (cid:107)θl(cid:107)2 H0) and b = (cid:107)ul−1(cid:107)2 a = (cid:107)θl(cid:107)2 (cid:104) 1 V0 − (cid:107)ul−1(cid:107)4 (cid:107)ul(cid:107)4 4 + (cid:12) (cid:12)(cid:107)θl(cid:107)2 V0 (resp., b = (cid:107)θk−1(cid:107)2 V0 + (cid:12) V0 − (cid:107)ul−1(cid:107)2 V0 2(cid:105) (cid:12) H0 − (cid:107)θl−1(cid:107)2 (cid:12) H0 (cid:12)(cid:107)ul(cid:107)2 2 V0 and (75) V0 (resp., (cid:2)a2 − b2 + |a − b|2(cid:3) for a = (cid:107)ul(cid:107)2 H0), we deduce, for l = 1, . . . , N, that H0 − (cid:107)θl−1(cid:107)4 H0 2 + (cid:107)θl(cid:107)4 (cid:12) (cid:12) + 1 2 (cid:2)(cid:107)ul − ul−1(cid:107)2 V0 (cid:107)ul(cid:107)2 V0 + (cid:107)θl − θl−1(cid:107)2 H0 (cid:107)θl(cid:107)2 H0 (cid:3) + hν(cid:107)A 1 2 ul(cid:107)2 V0 (cid:107)ul(cid:107)2 V0 + hκ(cid:107) ˜A where 1 2 θl(cid:107)2 H0 (cid:107)θl(cid:107)2 H0 = h(cid:0)Πθl−1v2, ul(cid:1)(cid:107)ul−1(cid:107)2 V0 + 4 ∑ i=1 Ti(l), (79) T1(l) =(cid:0)G(ul−1)∆ T3(l) =(cid:0) ˜G(θl−1)∆ lW, ul−1(cid:1)(cid:107)ul(cid:107)2 V0, l ˜W, θl−1(cid:1)(cid:107)θl−1(cid:107)2 H0, T2(l) = (cid:0)G(ul−1)∆ T4(l) = (cid:0) ˜G(θl−1)∆ lW, ul − ul−1(cid:1)(cid:107)ul(cid:107)2 l ˜W, θl − θl−1(cid:1)(cid:107)θl(cid:107)2 V0, H0. Mathematics 2022, 10, 4246 28 of 39 The Cauchy–Schwarz and Young inequalities imply that (cid:0)Πθl−1v2, ul(cid:1)(cid:107)ul(cid:107)2 V0 ≤ (cid:107)θl−1(cid:107)H0 (cid:107)ul(cid:107)3 V0 ≤ 1 4 (cid:107)θl−1(cid:107)4 H0 + 3 4 (cid:107)ul(cid:107)4 V0. (80) Using once more the Cauchy–Schwarz and Young inequalities, we deduce that for (cid:101), ¯(cid:101) > 0, |T2(l)| ≤ (cid:107)G(ul−1)(cid:107)L(K;V0)(cid:107)ul − ul−1(cid:107)2 V0 (cid:107)ul(cid:107)2 V0 ≤ (cid:101)(cid:107)ul − ul−1(cid:107)2 + 1 4(cid:101) (cid:107)G(ul−1)(cid:107)2 V0 (cid:107)ul(cid:107)2 V0 L(K;V0)(cid:107)∆ ≤ (cid:101)(cid:107)ul − ul−1(cid:107)2 V0 (cid:107)ul(cid:107)2 (cid:107)G(ul−1)(cid:107)2 lW(cid:107)2 K(cid:107)ul−1(cid:107)2 V0 (cid:2)(cid:107)ul−1(cid:107)2 V0 + (cid:0)(cid:107)ul(cid:107)2 L(K;V0)(cid:107)∆ V0 − (cid:107)ul−1(cid:107)2 V0 (cid:1)(cid:3) lW(cid:107)2 K 1 4(cid:101) 2 + V0 + (cid:12) (cid:12) + ¯(cid:101) (cid:12) (cid:12)(cid:107)ul(cid:107)2 V0 − (cid:107)ul−1(cid:107)2 V0 1 16(cid:101)2 1 4 ¯(cid:101) (cid:107)G(ul−1)(cid:107)4 L(K;V0)(cid:107)∆ lW(cid:107)4 K. (81) A similar argument proves, for (cid:101), ¯(cid:101) > 0, that |T4(l)| ≤ (cid:101)(cid:107)θl − θl−1(cid:107)2 H0 (cid:107)θl(cid:107)2 H0 + (cid:107) ˜G(θl−1)(cid:107)2 l ˜W(cid:107)2 ˜K (cid:107)θl−1(cid:107)2 H0 + ¯(cid:101) (cid:12) (cid:12)(cid:107)θl(cid:107)2 H0 − (cid:107)θl−1(cid:107)2 H0 L( ˜K;H0)(cid:107)∆ l ˜W(cid:107)4 ˜K. (82) 1 4(cid:101) (cid:12) 2 + (cid:12) L( ˜K;H0)(cid:107)∆ (cid:107) ˜G(θl−1)(cid:107)4 1 16(cid:101)2 1 4 ¯(cid:101) A similar argument shows, for ¯(cid:101) > 0, that lW(cid:107)V0 (cid:107)ul−1(cid:107)3 V0 |T1(l)| ≤ (cid:107)G(ul−1)∆ + (cid:107)G(ul−1)∆ ≤ 1 4 (cid:107)G(ul−1)(cid:107)4 L(K;V0)(cid:107)∆ + 1 4 ¯(cid:101) (cid:107)G(ul−1)(cid:107)2 lW(cid:107)V0 (cid:107)ul−1(cid:107)V0 3 lW(cid:107)4 4 lW(cid:107)2 L(K;V0)(cid:107)∆ K + K(cid:107)ul−1(cid:107)2 V0, (cid:2)(cid:107)ul(cid:107)2 (cid:107)ul−1(cid:107)4 (cid:3) V0 − (cid:107)ul−1(cid:107)2 V0 (cid:12) (cid:12)(cid:107)ul(cid:107)2 V0 + ¯(cid:101) V0 − (cid:107)ul−1(cid:107)V0 (cid:107)2(cid:12) (cid:12) 2 (83) and |T3(l)| ≤ 1 4 (cid:107) ˜G(θl−1)(cid:107)4 L( ˜K;H0)(cid:107)∆ l ˜W(cid:107)4 K + (cid:107)θl−1(cid:107)4 H0 + ¯(cid:101) (cid:12) (cid:12)(cid:107)θl(cid:107)2 H0 − (cid:107)θl−1(cid:107)H0 (cid:107)2(cid:12) (cid:12) 2 + 1 4 ¯(cid:101) (cid:107) ˜G(θl−1)(cid:107)2 L( ˜K;H0)(cid:107)∆ (cid:107)θl−1(cid:107)2 H0. (84) 3 4 l ˜W(cid:107)2 ˜K Add the inequalities (79)–(84) for l = 1 to L ≤ N, choose (cid:101) = 1 growth conditions (11) and (14). This yields 4 and ¯(cid:101) = 1 16 and use the (cid:107)uL(cid:107)4 V0 + (cid:107)θL(cid:107)4 1 2 H0 + L ∑ l=1 H0 − θl−1(cid:107)2 H0 (cid:2)ν(cid:107)A 2 ul(cid:107)2 1 + (cid:12) (cid:12)(cid:107)θl(cid:107)2 L ∑ l=1 + 4h (cid:2)(cid:12) (cid:12)(cid:107)ul(cid:107)2 V0 − (cid:107)ul−1(cid:107)2 V0 (cid:12) (cid:12) 2 + (cid:107)ul − ul−1(cid:107)2 V0 (cid:107)ul(cid:107)2 V0 (cid:12) (cid:12) 2 + (cid:107)θl − θl−1(cid:107)2 H0 (cid:107)θl(cid:107)2 H0 (cid:3) V0 (cid:107)ul(cid:107)2 V0 + κ(cid:107) ˜A 1 2 θl(cid:107)2 H0 (cid:107)θl(cid:107)2 H0 (cid:3) ≤(cid:107)u0(cid:107)4 V0 + (cid:107)θ0(cid:107)4 H0 + 1 4 h L−1 ∑ l=0 (cid:107)θl(cid:107)4 H0 + 3 4 h L ∑ l=1 (cid:107)ul(cid:107)4 V0 + C + C L ∑ l=0 L ∑ l=0 (cid:16)(cid:2)K0 + K1(cid:107)ul−1(cid:107)2 V0 (cid:3)(cid:107)ul−1(cid:107)2 V0 (cid:107)∆ lW(cid:107)2 K + (cid:2)K0 + K1(cid:107)ul−1(cid:107)2 V0 (cid:3)2(cid:107)∆ lW(cid:107)4 K (cid:17) (cid:16)(cid:2) ˜K0 + ˜K1(cid:107)θl−1(cid:107)2 H0 (cid:3)(cid:107)θl−1(cid:107)2 H0 (cid:107)∆ l ˜W(cid:107)2 ˜K + (cid:2) ˜K0 + ˜K1(cid:107)θl−1(cid:107)2 H0 (cid:3)2(cid:107)∆ l ˜W(cid:107)4 ˜K (cid:17) . (85) Mathematics 2022, 10, 4246 29 of 39 Taking expected values, we deduce, for every L = 1, . . . , N and h = T N ≤ 1, that E(cid:16) (cid:107)uL(cid:107)4 V0 + (cid:107)θL(cid:107)4 H0 + 1 2 H0 − θl−1(cid:107)2 H0 L ∑ l=1 (cid:12) (cid:12) + (cid:12) (cid:12)(cid:107)θl(cid:107)2 2 + (cid:107)θl − θl−1(cid:107)2 H0 (cid:107)θl(cid:107)2 H0 (cid:3)(cid:17) (cid:2)(cid:12) (cid:12)(cid:107)ul(cid:107)2 V0 − (cid:107)ul−1(cid:107)2 V0 (cid:12) (cid:12) 2 + (cid:107)ul − ul−1(cid:107)2 V0 (cid:107)ul(cid:107)2 V0 + E(cid:16) 4h (cid:2)ν(cid:107)A L ∑ l=1 1 2 ul(cid:107)2 V0 (cid:107)ul(cid:107)2 V0 + κ(cid:107) ˜A 1 2 θl(cid:107)2 H0 (cid:107)θl(cid:107)2 H0 (cid:3)(cid:17) ≤ E((cid:107)u0(cid:107)4 V0 + (cid:107)θ0(cid:107)4 H0 (cid:1) + 3 h E((cid:107)uL(cid:107)4 V0 ) + C + C h L−1 ∑ l=0 (cid:0)(cid:107)ul(cid:107)4 V0 + (cid:107)θl(cid:107)4 H0 (cid:1) for some constant C depending on Ki, ˜Ki, Tr(Q), Tr( ˜Q) and T that does not depend on N. Let N be large enough to have 3 h < 1 2 . Neglecting the sums in the left hand side and using the discrete Gronwall lemma, we deduce, for E(cid:0)(cid:107)u0(cid:107)4 V0 + (cid:107)θ0(cid:107)4 H0 (cid:17) (cid:1) < ∞, that E(cid:16) sup N≥1 max 0≤L≤N (cid:107)uL(cid:107)4 V0 + (cid:107)θL(cid:107)4 H0 < ∞. (86) This yields E(cid:16) h N ∑ l=1 sup N≥1 (cid:2)(cid:107)A 1 2 ul(cid:107)2 V0 (cid:107)ul(cid:107)2 V0 + (cid:107) ˜A 1 2 θl(cid:107)2 H0 (cid:107)θl(cid:107)2 H0 (cid:3)(cid:17) < ∞, (87) which proves (73) for K = 2. The argument used to prove (78) implies E(cid:16) max 1≤L≤N L ∑ l=1 (cid:0)G(ul−1)∆ lW, ul−1(cid:1)(cid:107)ul−1(cid:107)2 V0 (cid:17) ≤ (cid:101)E(cid:16) max 1≤L≤N (cid:107)uL(cid:107)4 V0 (cid:17) (cid:104) + C((cid:101)) 1 + max 1≤L≤N E((cid:107)uL(cid:107)4 V0 ) (cid:105) and E(cid:16) max 1≤L≤N L ∑ l=1 (cid:0) ˜G(θl−1)∆ l ˜W, θl−1(cid:1)(cid:107)uθl−1(cid:107)2 H0 (cid:17) ≤ (cid:101)E(cid:16) max 1≤L≤N (cid:107)θL(cid:107)4 H0 (cid:17) (cid:104) + C((cid:101)) 1 + max 1≤L≤N E((cid:107)θL(cid:107)4 H0 ) (cid:105) Taking the maximum for L = 1, ..., N and using (86), we deduce (72) for K = 2. The details of the induction step, similar to the proof in the case K = 2, are left to the reader. 7. Strong Convergence of the Localized Implicit Time Euler Scheme Due to the bilinear terms [u.∇]u and [u.∇]θ, we first prove an L2(Ω) convergence of the L2(D)-norm of the error, uniformly on the time grid, restricted to the set ΩM(N) defined below for some M > 0: ΩM(j) := (cid:110) (cid:107)A sup s∈[0,tj] 1 2 u(s)(cid:107)2 V0 ≤ M (cid:111) (cid:110) ∩ (cid:107) ˜A sup s∈[0,tj] 1 2 θ(s)(cid:107)2 H0 ≤ M (cid:111) , ∀j = 0, . . . , N, (88) Mathematics 2022, 10, 4246 30 of 39 and let ΩM := ΩM(N). Recall that, for j = 0, ..., N, set ej := u(tj) − uj and ˜ej := θ(tj) − θ j; then, e0 = ˜e0 = 0. Using (9), (10), (17) and (18), we deduce, for j = 1, ..., N, φ ∈ V1 and ψ ∈ H1, that (cid:0)ej − ej−1 , ϕ(cid:1) + ν (cid:90) tj tj−1 (cid:0)A 1 2 [u(s) − uj], A 1 2 ϕ(cid:1)ds + (cid:90) tj tj−1 (cid:10)B(u(s), u(s)) − B(uj, uj), ϕ(cid:11)ds (cid:90) tj+1 = tj (cid:0)Π[θ(s) − θ j−1]v2, ϕ(cid:1)ds + (cid:90) tj tj−1 (cid:0)[G(u(s)) − G(uj−1)]dW(s) , ϕ(cid:1), (89) and (cid:0) ˜ej − ˜ej−1 , ψ(cid:1) + κ (cid:90) tj tj−1 (cid:0) ˜A 1 2 [θ(s) − θ j], ˜A 1 2 ψ(cid:1)ds + (cid:90) tj tj−1 (cid:10)[u(s).∇]θ(s) − [u.∇]θ j], ψ(cid:11)ds (cid:90) tj = tj−1 (cid:0)[ ˜G(θ(s)) − ˜G(θ j−1)]d ˜W(s) , ψ(cid:1). (90) In this section, we will suppose that N is large enough to have h := T N ∈ (0, 1). The following result is a crucial step towards the rate of convergence of the implicit time Euler scheme. Proposition 7. Suppose that the conditions (C-u) and (C-θ) hold. Let u0 ∈ L32+(cid:101)(Ω; V1) and θ0 ∈ L32+(cid:101)(Ω; H1) for some (cid:101) > 0, u, θ be the solution to (9) and (10) and {uj, θ j}j=0,...,N be the solution to (17) and (18). Fix M > 0 and let ΩM = ΩM(N) be defined by (88). Then, for η ∈ (0, 1), there exists a positive constant C, independent of N, such that, for large enough N, E(cid:16) 1Ω M (cid:104) max 1≤j≤N (cid:0)(cid:107)u(tj) − uj(cid:107)2 + (cid:107) ˜A 1 2 [θ(tj) − θ j](cid:107)2 H0 (cid:105)(cid:17) (cid:1) + V0 + (cid:107)θ(tj) − θ j(cid:107)2 H0 T N ≤ C(1 + M)eC(M)T(cid:16) T N (cid:17)η N ∑ j=1 (cid:2)(cid:107)A 1 2 [u(tj) − uj](cid:107)2 V0 , (91) where C(M) = 9(1 + γ) ¯C2 4 8 max (cid:17) (cid:16) 5 ν , 1 κ M for some γ > 0, and ¯C4 is the constant in the right hand side of the Gagliardo–Nirenberg inequality (6). Proof. Write (89) with ϕ = ej and (90) with ψ = θ j; using the equality ( f , f − g) = (cid:3), we obtain for j = 1, . . . , N (cid:2)(cid:107) f (cid:107)2 1 2 L2 + (cid:107) f − g(cid:107)2 L2 L2 − (cid:107)g(cid:107)2 1 2 (cid:0)(cid:107)ej(cid:107)2 V0 − (cid:107)ej−1(cid:107)2 V0 (cid:1) + 1 2 (cid:0)(cid:107) ˜ej(cid:107)2 H0 − (cid:107) ˜ej−1(cid:107)2 H0 (cid:1) + 1 2 1 2 (cid:107)ej − ej−1(cid:107)2 V0 + νh(cid:107)A (cid:107) ˜ej − ˜ej−1(cid:107)2 H0 + κh(cid:107) ˜A 1 2 ej(cid:107)2 V0 ≤ 1 2 ˜ej(cid:107)2 H0 ≤ 7 ∑ l=1 6 ∑ l=1 Tj,l, ˜Tj,l, (92) (93) Mathematics 2022, 10, 4246 31 of 39 where, by the antisymmetry property (3), we have that Tj,1 = − Tj,2 = − Tj,3 = − (cid:90) tj tj−1 (cid:90) tj tj−1 (cid:90) tj tj−1 (cid:90) tj (cid:90) tj tj−1 (cid:90) tj tj−1 Tj,6 = Tj,7 = and ˜Tj,1 = − ˜Tj,2 = − ˜Tj,3 = − (cid:90) tj tj−1 (cid:90) tj tj−1 (cid:90) tj tj−1 (cid:90) tj ˜Tj,5 = ˜Tj,6 = (cid:90) tj tj−1 (cid:90) tj tj−1 (cid:90) tj (cid:10)B(cid:0)ej, uj(cid:1) , ej (cid:11)ds = − tj−1 (cid:10)B(cid:0)u(s) − u(tj) , u(tj)(cid:1) ej (cid:11)ds, (cid:10)B(cid:0)ej, u(tj)(cid:1) , ej (cid:11)ds, (cid:10)B(cid:0)u(s), u(s) − u(tj)(cid:1) , ej (cid:11)ds = (cid:10)B(cid:0)u(s), ej (cid:1) , u(s) − u(tj)(cid:11)ds, (cid:90) tj tj−1 (cid:1)ds, Tj,5 = 1 2 ej (cid:90) tj tj−1 (cid:0)Π[θ(s) − θ j−1]v2 , ej (cid:1)ds, Tj,4 = − ν 1 (cid:0)A 2 (u(s) − u(tj)), A tj−1 (cid:0)[G(u(s)) − G(uj−1)(cid:3)dW(s) , ej − ej−1 (cid:1), (cid:0)[G(u(s)) − G(uj−1)(cid:3)dW(s), ej−1 (cid:1), (cid:90) tj (cid:10)[ej−1.∇]θ j , ˜ej (cid:11)ds = − tj−1 (cid:10)[(u(s) − u(tj−1).∇]θ(tj) , ˜ej (cid:11)ds, (cid:10)[ej−1.∇]θ(tj) , ˜ej (cid:11)ds, (cid:10)[u(s).∇](θ(s) − θ(tj) , ˜ej (cid:11)ds = (cid:90) tj tj−1 (cid:10)[u(s).∇] ˜ej , (θ(s) − θ(tj)(cid:11)ds, ˜Tj,4 = − ν 1 (cid:0) ˜A 2 (θ(s) − θ(tj)), ˜A 1 2 ˜ej (cid:1)ds, tj−1 (cid:0)[ ˜G(θ(s)) − ˜G(θ j−1)d ˜W(s) , ˜ej − ˜ej−1 (cid:1), (cid:0)[G(u(s)) − G(uj−1)(cid:3)dW(s), ej−1 (cid:1), We next prove upper estimates of the terms Tj,l for l = 1, ..., 5 and ˜Tj,l for l = 1, . . . , 4, and of the expected value of Tj,6, Tj,7 ˜Tj,5 and ˜Tj,6. The Hölder and Young inequalities and the Gagliardo–Nirenberg inequality (6) imply, for δ1 > 0, that |Tj,1| ≤ ¯C4 h (cid:107)ej(cid:107)V0 (cid:107)A ≤ δ1 ν h (cid:107)A 1 2 ej(cid:107)2 h (cid:107)A 1 2 u(tj)(cid:107)2 V0 (cid:107)ej(cid:107)2 V0, 1 2 u(tj)(cid:107)V0 1 2 ej(cid:107)V0 (cid:107)A ¯C2 4 4δ1ν V0 + and, for ˜δ1, δ2 > 0, that | ˜Tj,1| ≤ ¯C4 h (cid:107)A 1 1 2 1 2 1 2 ej−1(cid:107) 2 ej−1(cid:107)2 V0 (cid:107) ˜A V0 (cid:107)ej−1(cid:107) V0 + ˜δ1hκ(cid:107) ˜A 2 ˜ej(cid:107) 2 ˜ej(cid:107)2 H0 1 1 1 2 H0 (cid:107) ˜ej(cid:107) 1 2 H0 (cid:107) ˜A 1 2 θ(tj)(cid:107)H0 h (cid:107) ˜A 1 2 θ(tj)(cid:107)2 H0 (cid:107)ej−1(cid:107)2 V0 + ¯C2 4 16 ˜δ1κ h (cid:107) ˜A 1 2 θ(tj)(cid:107)2 H0 (cid:107) ˜ej(cid:107)2 H0. ≤ δ2 νh(cid:107)A ¯C2 4 16δ2ν + (94) (95) Mathematics 2022, 10, 4246 32 of 39 Hölder’s inequality and the Sobolev embedding V1 ⊂ L4 imply, for δ3 > 0, that |Tj,2| ≤ C (cid:90) tj tj−1 (cid:107)u(s) − u(tj)(cid:107)V1 (cid:107)A 1 2 u(tj)(cid:107)V0 (cid:107)A 1 2 ej(cid:107) 1 2 1 2 V0 ds V0 (cid:107)ej(cid:107) (cid:90) tj ≤ δ3ν h (cid:107)A 1 2 ej(cid:107)2 V0 + h (cid:107)ej(cid:107)2 V0 + √ C νδ3 (cid:107)A 1 2 u(tj)(cid:107)2 V0 tj−1 (cid:107)u(tj) − u(s)(cid:107)2 V1 ds, whereas, for ˜δ2 > 0, | ˜Tj,2| ≤ ˜δ2κ h (cid:107) ˜A 1 2 ˜ej(cid:107)2 H0 + h (cid:107) ˜ej(cid:107)2 H0 + C κ ˜δ2 (cid:90) tj tj−1 (cid:107)A 1 2 (cid:2)u(s) − u(tj−1)(cid:3)(cid:107)2 V0 ds + C κ ˜δ2 (cid:107) ˜A 1 2 θ(tj)(cid:107)4 H0 (cid:90) tj tj−1 (cid:107)u(s) − u(tj−1)(cid:107)2 V0 ds. Similar arguments prove, for δ4, ˜δ3 > 0, that |Tj,3| ≤ δ4νh(cid:107)A 1 2 ej(cid:107)2 V0 + C νδ4 sup s∈[0,T] (cid:107)u(s)(cid:107)2 V1 | ˜Tj,3| ≤ ˜δ3 κ h (cid:107) ˜A 1 2 ˜ej(cid:107)2 H0 + C κ ˜δ3 sup s∈[0,T] (cid:107)u(s)(cid:107)2 V1 (cid:107)θ(s) − θ(tj)(cid:107)2 H1 ds. (cid:107)u(s) − u(tj)(cid:107)2 V1 ds, (cid:90) tj tj−1 (cid:90) tj tj−1 The Cauchy–Schwarz and Young inequalities imply, for δ5, ˜δ4 > 0, that |Tj,4| ≤ δ5ν h (cid:107)A 1 2 ej(cid:107)2 V0 + | ˜Tj,4| ≤ ˜δ4 κ h (cid:107) ˜A 1 2 ˜ej(cid:107)2 H0 + ν 4δ5 κ 4 ˜δ4 (cid:90) tj tj−1 (cid:90) tj tj−1 1 (cid:107)A 2 [u(s) − u(tj)](cid:107)2 V0 ds, 1 (cid:107) ˜A 2 [θ(s) − θ(tj)](cid:107)2 H0 ds. Using once more the Cauchy–Schwarz and Young inequalities, we deduce (96) (97) (98) (99) (100) (101) |Tj,5| ≤ ≤ (cid:90) tj tj−1 h 2 (cid:2)(cid:107)θ(s) − θ(tj−1)(cid:107)H0 + (cid:107) ˜ej−1(cid:107)H0 (cid:3) (cid:107)ej(cid:107)V0 ds (cid:107)ej(cid:107)2 V0 + h 2 (cid:107) ˜ej−1(cid:107)2 H0 + 1 2 (cid:90) tj tj−1 (cid:107)θ(s) − θ(tj−1)(cid:107)2 H0 ds. (102) Note that the sequence of subsets {ΩM(j)}0≤j≤N is decreasing. Therefore, since e0 = ˜e0 = 0, given L = 1, . . . , N, we obtain max 1≤J≤L J ∑ j=1 1Ω M(j−1) (cid:2)(cid:107)ej(cid:107)2 V0 − (cid:107)ej−1(cid:107)2 V0 + (cid:107) ˜ej(cid:107)2 H0 − (cid:107) ˜ej−1(cid:107)2 H0 (cid:3) = max 1≤J≤L (cid:16) L ∑ j=2 1Ω M(J−1) (cid:2)(cid:107)eJ(cid:107)2 V0 + (cid:107) ˜ej(cid:107)2 H0 (cid:3)(cid:17) + L ∑ j=2 (cid:0)1Ω M(j−2) − 1Ω M(j−1) (cid:1)(cid:2)(cid:107)ej−1(cid:107)2 V0 + (cid:107) ˜ej−1(cid:107)2 H0 (cid:3) ≥ max 1≤J≤L (cid:16) L ∑ j=2 1Ω M(J−1) (cid:2)(cid:107)eJ(cid:107)2 V0 + (cid:107) ˜ej(cid:107)2 H0 (cid:3)(cid:17) . Hence, for ∑5 α > 0, that j=1 δj ≤ 1 3 and ∑4 j=1 ˜δj < 1 3 , using Young’s inequality, we deduce, for every Mathematics 2022, 10, 4246 33 of 39 (cid:16) 1 6 max 1≤J≤L 1Ω M(J−1) (cid:2)(cid:107)eJ(cid:107)2 V0 + (cid:107) ˜ej(cid:107)2 H0 (cid:3)(cid:17) 1 6 L ∑ j=1 1Ω M(j−1) (cid:0)(cid:107)ej − ej−1(cid:107)2 V0 + (cid:107) ˜ej − ˜ej−1(cid:107)2 H0 (cid:1) 1Ω M(j−1) h (cid:104) ν (cid:16) 1 3 (cid:107)A 1 2 ej(cid:107)2 V0 + κ (cid:16) 1 3 − (cid:105) 1 2 ˜ej(cid:107)2 H0 (cid:17) δi − 5 ∑ i=1 (cid:16) (1 + α) ¯C2 4 4δ1ν (cid:17) ˜δi (cid:107) ˜A 4 ∑ i=1 (1 + α) ¯C2 4 16δ2ν 1Ω M(j−1)(cid:107)ej(cid:107)2 V0 (cid:107)A 1 2 u(tj−1)(cid:107)2 V0 + 1 (cid:107) ˜A 2 θ(tj−1)(cid:107)2 H0 + (cid:17) 3 2 + + ≤ h L ∑ j=1 L ∑ j=1 + h L ∑ j=1 1Ω M(j−1)(cid:107) ˜ej(cid:107)2 H0 (cid:16) (1 + α) ¯C2 4 16 ˜δ1κ (cid:107) ˜A 1 2 θ(tj−1)(cid:107)2 H0 + (cid:17) 3 2 + ZL + max 1≤J≤L J ∑ j=1 1Ω M(j−1) (cid:2)Tj,6 + ˜Tj,5 (cid:3) + max 1≤J≤L J ∑ j=1 1Ω M(j−1) (cid:2)Tj,7 + ˜Tj,6 (cid:3), (103) where ZL = C h L ∑ j=1 (cid:107)ej(cid:107)2 V0 (cid:0)(cid:107)A 1 2 [u(tj) − u(tj−1)](cid:107)2 V0 + (cid:107) ˜A 1 2 [θ(tj) − θ(tj−1)](cid:107)2 H0 (cid:1) + C h L ∑ j=1 (cid:107) ˜ej(cid:107)2 V0 (cid:107) ˜A 1 2 [θ(tj) − θ(tj−1)](cid:107)2 H0 + C + C + C L ∑ j=1 L ∑ j=1 L ∑ j=1 (cid:16) sup s∈[0,T] (cid:107)u(s)(cid:107)2 V1 + 1 (cid:17) (cid:90) tj tj−1 (cid:2)(cid:107)u(tj) − u(s)(cid:107)2 V1 + (cid:107)u(s) − u(tj−1(cid:107)2 V1 (cid:3)ds (cid:107) ˜A 1 2 θ(tj−1)(cid:107)4 H0 (cid:90) tj tj−1 (cid:107)u(s) − u(tj−1)(cid:107)2 V0 ds (cid:16) sup s∈[0,T] (cid:107)u(s)(cid:107)2 V1 + 1 (cid:17) (cid:90) tj tj−1 (cid:107)θ(s) − θ(tj)(cid:107)2 H1 ds. The Cauchy–Schwarz and Young inequalities imply that L ∑ j=1 L ∑ j=1 1Ω M(j−1)|Tj,6| ≤ + 3 2 L ∑ j=1 1Ω M(j−1) 1Ω M(j−1)| ˜Tj,5| ≤ + 3 2 L ∑ j=1 1Ω M(j−1) 1 6 (cid:13) (cid:13) (cid:13) 1 6 (cid:13) (cid:13) (cid:13) L ∑ j=1 (cid:90) tj L ∑ j=1 (cid:90) tj tj−1 tj−1 1Ω M(j−1)(cid:107)ej − ej−1(cid:107)2 V0 (cid:2)G(u(s)) − G(uj−1)(cid:3)dW(s) (cid:13) (cid:13) (cid:13) 2 V0 , 1Ω M(j−1)(cid:107) ˜ej − ˜ej−1(cid:107)2 H0 (cid:2) ˜G(θ(s)) − ˜G(θ j−1)(cid:3)d ˜W(s) (cid:13) (cid:13) (cid:13) 2 H0 . (104) (105) (106) Using the upper estimates (103)–(106), taking expected values and using the Cauchy– Schwarz and Young inequalities, as well as the inequalities (19), (20), (37), (46), (59), (60) and (72), we deduce that, for η ∈ (0, 1) and every L = 1, . . . , N, Mathematics 2022, 10, 4246 34 of 39 E(ZL) ≤ C (cid:110)E(cid:16) sup s∈[0,T] (cid:107)u(s)(cid:107)4 V0 + max 0≤j≤N (cid:107)uj(cid:107)4 V0 (cid:17)(cid:111) 1 2 × (cid:110)E(cid:16)(cid:12) (cid:12) (cid:12)h (cid:0)(cid:107)A N ∑ j=1 1 2 (cid:2)u(tj) − u(tj−1)(cid:3)(cid:107)2 V0 + (cid:107) ˜A 1 2 (cid:2)θ(tj) − θ(tj−1)(cid:3)(cid:107)2 V0 2(cid:17)(cid:111) 1 2 (cid:1)(cid:12) (cid:12) (cid:12) (cid:110)E(cid:16) + C sup s∈[0,T] (cid:107)θ(s)(cid:107)4 H0 + max 0≤j≤N (cid:107)θ j(cid:107)4 H0 (cid:17)(cid:111) 1 2 × (cid:110)E(cid:16)(cid:12) (cid:12) (cid:12)h N ∑ j=1 (cid:107) ˜A 1 2 (cid:2)θ(tj) − θ(tj−1)(cid:3)(cid:107)2 V0 2(cid:111) 1 2 (cid:1)(cid:12) (cid:12) (cid:12) (cid:110)E(cid:16) + C 1 + sup 0≤s≤T (cid:107)u(s)(cid:107)4 V1 (cid:17)(cid:111) 1 2 (cid:110)E(cid:16)(cid:12) (cid:12) (cid:12) N ∑ j=1 (cid:90) tj tj−1 (cid:2)(cid:107)u(s) − u(tj)(cid:107)2 V0 + (cid:107)u(s) − u(tj−1)(cid:107)2 V0 + (cid:107)θ(s) − θ(tj)(cid:107)2 H0 2(cid:17)(cid:111) 1 2 (cid:3)(cid:12) (cid:12) (cid:12) + C N ∑ j=1 (cid:90) tj tj−1 (cid:8)E(cid:0)(cid:107) ˜A 1 2 θ(tj)(cid:107)8 H0 (cid:1)(cid:9) 1 2 (cid:110)E(cid:0)(cid:107)u(s) − u(tj−1)(cid:107)4 V0 (cid:1)(cid:111) 1 2 ds ≤ C hη, (107) for some constant C independent of L and N. Furthermore, the Lipschitz conditions (12) and (15), the inclusion ΩM(j − 1) ⊂ ΩM(j − 2) for j = 2, ..., N and the upper estimates (46) and (47) imply that E(cid:16) L ∑ j=1 1ΩM(j−1) (cid:13) (cid:13) (cid:13) (cid:90) tj tj−1 (cid:2)G(u(s)) − G(uj−1)(cid:3)dW(s) (cid:13) (cid:13) (cid:13) 2 (cid:1) V0 ≤ L ∑ j=1 E(cid:16) (cid:90) tj tj−1 ≤ 2L1Tr(Q) h ≤ 2L1Tr(Q) h 1ΩM(j−1) L1(cid:107)u(s) − uj−1(cid:107)2 V0 Tr(Q)ds (cid:17) E(1ΩM(j−2)(cid:107)ej−1(cid:107)2 V0 (cid:1) + C L ∑ j=1 E(cid:16) (cid:90) tj tj−1 (cid:107)u(s) − u(tj−1)(cid:107)2 V0 ds (cid:17) E(1ΩM(j−2)(cid:107)ej−1(cid:107)2 V0 (cid:1) + Ch, L ∑ j=2 L ∑ j=2 E(cid:16) L ∑ j=1 1ΩM(j−1) (cid:13) (cid:13) (cid:13) (cid:90) tj tj−1 (cid:2) ˜G(θ(s)) − ˜G(θ j−1)(cid:3)d ˜W(s) (cid:13) (cid:13) (cid:13) 2 (cid:1) H0 ≤ 2 ˜L1Tr( ˜Q) h L ∑ j=2 E(1ΩM(j−2)(cid:107) ˜ej−1(cid:107)2 H0 (cid:1) + Ch. (108) (109) Mathematics 2022, 10, 4246 35 of 39 Finally, the Davis inequality, the inclusion ΩM(J − 1) ⊂ ΩM(j − 1) for j ≤ J, the local property of stochastic integrals, the Lipschitz condition (12), the Cauchy–Schwarz and Young inequalities and the upper estimate (46) imply, for λ > 0, that E(cid:16) max 1≤J≤L 1Ω M(J−1) (cid:17) Tj7 J ∑ j=1 ≤ 3 ≤ 3 L ∑ j=1 L ∑ j=1 ≤ λE(cid:16) ≤ λE(cid:16) E(cid:16)(cid:110) 1Ω M(j−1) (cid:90) tj tj−1 (cid:107)G(u(s)) − G(uj−1)(cid:107)2 L(K;V0)Tr(Q)(cid:107)ej−1(cid:107)2 V0 ds (cid:111) 1 2 (cid:17) E(cid:104)(cid:16) max 1≤j≤L 1Ω M(j−1)(cid:107)ej−1(cid:107)V0 (cid:17)(cid:110) (cid:90) tj tj−1 L1Tr(Q)(cid:107)u(s) − uj−1(cid:107)2 V0 ds (cid:111) 1 2 (cid:17) max 1≤j≤L 1Ω M(j−1)(cid:107)ej−1(cid:107)2 V0 max 1≤j≤L 1Ω M(j−2)(cid:107)ej−1(cid:107)2 V0 (cid:17) (cid:17) + CE(cid:16) L ∑ j=1 (cid:90) tj tj−1 L1Tr(Q)(cid:107)u(s) − uj−1(cid:107)2 V0 ds (cid:17) + Ch L ∑ j=1 E((cid:107)ej−1(cid:107)2 V0 ) + C h. (110) A similar argument, using the Lipschitz condition (15) and (47), yields, for λ > 0, E(cid:16) max 1≤J≤L 1Ω M(J−1) (cid:17) ˜Tj7 J ∑ j=1 ≤ λE(cid:16) max 1≤j≤L 1Ω M(j−2)(cid:107) ˜ej−1(cid:107)2 H0 (cid:17) + Ch L ∑ j=1 E((cid:107) ˜ej−1(cid:107)2 V0 ) + Ch. (111) Collecting the upper estimates (94)–(111), we obtain, for ∑5 and α, λ > 0, i=1 δi < 1 3 , ∑4 i=1 ˜δi < 1 3 , η ∈ (0, 1) E(cid:16) max 1≤J≤N 1ΩM(j−1) (cid:2)(cid:107)ej(cid:107)2 V0 + (cid:107) ˜ej(cid:107)2 H0 (cid:3)(cid:17) + E(cid:16) N ∑ j=1 1ΩM(j−1) (cid:16) (cid:104) ν ≤ h E(cid:16) N−1 ∑ j=1 1ΩM(j−1)(cid:107)ej(cid:107)2 V0 (cid:17) δi 2 − 6 5 ∑ i=1 (cid:104) 3(1 + α) ¯C2 4 2ν (cid:107)A 1 2 ej(cid:107)2 V0 + κ (cid:16) 2 − 6 (cid:16) 1 δ1 + (cid:17) 1 4δ2 M + C (cid:17) ˜δi (cid:107) ˜A 1 2 ˜ej(cid:107)2 V0 (cid:105)(cid:17) 4 ∑ i=1 (cid:105)(cid:17) + h E(cid:16) N−1 ∑ j=1 1ΩM(j−1)(cid:107) ˜ej(cid:107)2 H0 (cid:104) 3(1 + α) ¯C2 4 8 ˜δ1κ (cid:105)(cid:17) M + C + C(1 + M)hE(cid:16) sup t∈[0,T] (cid:2)(cid:107)u(t)(cid:107)2 V0 + (cid:107)θ(t)(cid:107)2 H0 (cid:3) + max 1≤j≤N (cid:2)(cid:107)uj(cid:107)2 V0 + (cid:107)θ j(cid:107)2 H0 (cid:1)(cid:3)(cid:17) + 12λE(cid:16) max 1≤j≤N 1ΩM(j−1) (cid:2)(cid:107)ej−1(cid:107)2 V0 + (cid:107) ˜ej(cid:107)2 H0 (cid:3)(cid:17) + Chη. (112) Therefore, given γ ∈ (0, 1), choosing λ ∈ (0, 1 1+α 1−12λ < 1 + γ, neglecting the sum in the left hand side and using the discrete Gronwall lemma, we deduce, for η ∈ (0, 1), that 12 ) and α > 0 such that E(cid:16) max 1≤J≤N 1Ω M(j−1) (cid:2)(cid:107)ej(cid:107)2 V0 + (cid:107) ˜ej(cid:107)2 H0 (cid:3)(cid:17) ≤ C(1 + M)eTC(M)hη, (113) where C(M) := 3(1 + γ) ¯C2 4 2 max (cid:16) 1 δ1ν + 1 4δ2ν , 1 4 ˜δ1κ (cid:17) M, Mathematics 2022, 10, 4246 36 of 39 for ∑2 and ∑4 i=1 δi < 1 ˜δi < 1 i=1 3 and ˜δ1 < 1 3 ). Let δ2 < 1 3 (and choosing δi, i = 3, 4, 5 and ˜δi, i = 2, 3, 4 such that ∑5 15 and δ1 = 4δ2. Then, for some γ > 0, we have that i=1 δi < 1 3 C(M) = 9(1 + γ) ¯C2 4 8 max (cid:17) (cid:16) 5 ν , 1 κ M. Plugging the upper estimate (113) in (112), we conclude the proof of (91). 8. Rate of Convergence in Probability and in L2(Ω) In this section, we deduce from Proposition 7 the convergence in probability of the implicit time Euler scheme with the “optimal” rate of convergence of “almost 1/2” and a logarithmic speed of convergence in L2(Ω). The presence of the bilinear term in the Itô formula for (cid:107) ˜A H0 does not enable us to prove exponential moments for this norm, which prevents us from using the general framework presented in [10] to prove a polynomial rate for the strong convergence. 1 2 θ(t)(cid:107)2 8.1. Rate of Convergence in Probability In this section, we deduce the rate of the convergence in probability (defined in [17]) from Propositions 1, 2, 6 and 7. Proof of Theorem 2. For N ≥ 1 and η ∈ (0, 1), let A(N, η) := (cid:110) max 1≤J≤N (cid:2)(cid:107)eJ(cid:107)2 V0 + (cid:107) ˜eJ(cid:107)2 H0 (cid:3) + T N N ∑ j=1 (cid:2)(cid:107)A Let ˜η ∈ (η, 1), M(N) = ln(ln N) for N ≥ 3. Then, 1 2 ej(cid:107)2 V0 + (cid:107) ˜A 1 2 ˜ej(cid:107)2 H0 (cid:3) ≥ N−η(cid:111) . P(cid:0)A(N, η)(cid:1) ≤ P(cid:0)A(N, η) ∩ Ω (cid:1) + P(cid:0)(Ω M(N))c(cid:1), M(N) where Ω M(N) = Ω M(N)(N) is defined in Proposition 7. The inequality (91) implies that P(cid:0)A(N, η) ∩ Ω ≤ Nη E(cid:16) 1Ω (cid:1) M(N) (cid:104) M(N) max 1≤J≤N (cid:2)(cid:107)eJ(cid:107)2 V0 + (cid:107) ˜eJ(cid:107)2 H0 (cid:3) + T N N ∑ j=1 (cid:2)(cid:107)A ≤ Nη C(cid:2)1 + ln(ln N)(cid:3)eT ˜C ln(ln N)(cid:16) T N ≤ C(cid:2)1 + ln(ln N)(cid:3)(cid:0) ln N(cid:1) ˜CT N− ˜η+η → 0 (cid:17) ˜η as N → ∞. 1 2 ej(cid:107)2 V0 + (cid:107) ˜A (cid:3)(cid:105)(cid:17) 1 2 ˜ej(cid:107)2 H0 The inequalities (20)–(22) imply that P(cid:0)(Ω M(N))c(cid:1) ≤ E(cid:16) 1 M(N) sup t∈[0,T] (cid:107)u(t)(cid:107)2 V1 + sup t∈[0,T] (cid:107)θ(t)(cid:107)2 H1 (cid:17) → 0 as N → ∞. The two above convergence results complete the proof of (23). 8.2. Rate of Convergence in L2(Ω) We finally prove the strong rate of convergence, which is also a consequence of Propositions 1, 2, 6 and 7. Mathematics 2022, 10, 4246 37 of 39 Proof of Theorem 3. For any integer N ≥ 1 and M ∈ [1, +∞), let ΩM = ΩM(N) be defined by (88). Let p be the conjugate exponent of 2q. Hölder’s inequality implies that (cid:2)(cid:107)eJ(cid:107)2 V0 + (cid:107) ˜eJ(cid:107)2 H0 (cid:3)(cid:17) (cid:110) P(cid:0)(ΩM)c(cid:1)(cid:111) 1 p ≤ sup s∈[0,T] (cid:107)u(s)(cid:107)2q V0 + sup s∈[0,T] (cid:107)θ(s)(cid:107)2q H0 + max 1≤j≤N (cid:107)uj(cid:107)2q V0 + max 1≤j≤N (cid:107)θ j(cid:107)2q H0 E(cid:16) 1(Ω M)c max 1≤J≤N (cid:110)E(cid:16) × (cid:110) P(cid:0)(ΩM)c(cid:1)(cid:111) 1 p , ≤ C (cid:17)(cid:111)2−q (114) where the last inequality is a consequence of (19), (20) and (72). Using (21) and (22), we deduce that P(cid:0)(ΩM)c(cid:1) ≤ M−2q−1E(cid:0) sup s∈[0,T] (cid:107)u(s)(cid:107)2q V1 + sup s∈[0,T] (cid:17) (cid:107)θ(s)(cid:107)2q H1 = CM−2q−1 . (115) Using (91), we choose M(N) → ∞ as N → ∞ such that, for η ∈ (0, 1) and γ > 0, N−η exp (cid:104) 9(1 + γ) ¯C2 4 T 8 (cid:16) 5 ν ∨ (cid:17) 1 κ (cid:105) M(N) M(N) (cid:16) M(N)−2q−1 which, taking logarithms, yields −η ln(N) + 9(1 + γ) ¯C2 8 4 T (cid:0) 5 ν ∨ 1 κ (cid:1)M(N) (cid:16) −2q−1 ln(M(N)). Set Then, M(N) = (cid:16) 8 9(1 + γ) ¯C2 4 8 9(1 + γ) ¯C2 4 (cid:0) 5 ν ∨ 1 κ (cid:1)T (cid:0) 5 ν ∨ 1 κ (cid:1)T (cid:2)η ln(N) − (cid:0)2q−1 + 1(cid:1) ln (cid:0) ln(N)(cid:1)(cid:3) η ln(N). −η ln(N) + 9(1 + γ) ¯C2 8 4 T (cid:0) 5 ν This implies that 1 κ (cid:1)M(N) + ln(M(N)) (cid:16) −(cid:0)2q−1 + 1(cid:1) ln (cid:0) ln(N)(cid:1) + 0(1), ∨ −(cid:0)2q−1 + 1(cid:1) ln (cid:0)M(N)(cid:1) (cid:16) −(cid:0)2q−1 + 1(cid:1) ln(N) + 0(1). E(cid:16) max 1≤J≤N (cid:2)(cid:107)eJ(cid:107)2 V0 + (cid:107) ˜eJ(cid:107)2 H0 (cid:3)(cid:17) ≤ C(cid:0) ln(N)(cid:1)−(cid:0) 2q−1+1) . The inequalities (21) and (22) for p = 1 and (73) for K = 1 imply E(cid:16) T N N ∑ j=1 sup N≥1 (cid:2)(cid:107)A 1 2 u(tj)(cid:107)2 V0 + (cid:107)A 1 2 uj(cid:107)2 V0 + (cid:107) ˜A 1 2 θ(tj)(cid:107)2 H0 + (cid:107) ˜A (cid:3)(cid:17) 1 2 θ j(cid:107)2 H0 < ∞. Using a similar argument, we obtain E(cid:16) T N N ∑ j=1 (cid:2)(cid:107)A 1 2 ej(cid:107)2 V0 + (cid:107) ˜A (cid:3)(cid:17) 1 2 ˜ej(cid:107)2 H0 ≤ C(cid:0) ln(N)(cid:1)−(2q−1+1) . This yields (24) and completes the proof. Mathematics 2022, 10, 4246 38 of 39 9. Conclusions This paper provides the first result on the rate of the convergence of a time discretiza- tion of the Navier–Stokes equations coupled with a transport equation for the temperature, driven by a random perturbation; this is the so-called Boussinesq/Bénard model. The perturbation may depend on both the velocity and temperature of the fluid. The rates of the convergence in probability and in L2(Ω) are similar to those obtained for the stochastic Navier–Stokes equations. The Boussinesq equations model a variety of phenomena in environmental, geophysical and climate systems (see, e.g., [18,19]). Even if the outline of the proof is similar to that used for the Navier–Stokes equations, the interplay between the velocity and the temperature is more delicate to deal with in many places. This interplay, which appears in Bénard systems, is crucial for describing more general hydrodynamical models. The presence of the velocity in the bilinear term describing the dynamics of the temperature makes it more difficult to prove bounds of moments for the H1-norm of the temperature uniformly in time and requires higher moments of the initial condition. Such bounds are crucial to deduce rates of convergence (in probability and in L2(Ω)) from the localized one. This localized version of the convergence is the usual first step in a non-linear (non- Lipschitz and non-monotonous) setting. Numerical simulations, which are the ultimate aim of this study since there is no other way to “produce” trajectories of the solution, would require a space discretization, such as finite elements. This is not dealt with in this paper and will be carried out in a forthcoming work. This new study is likely to provide results similar to those obtained for the 2D Navier–Stokes equations. In addition, note that another natural continuation of this work would be to consider a more general stochastic 2D magnetic Bénard model (as discussed in [1]) that describes the time evolution of the velocity, temperature and magnetic field of an incompressible fluid. It would also be interesting to study the variance of the L2(D)-norm of the error term, in both additive and multiplicative settings, for the Navier-0Stokes equations and more general Bénard systems. This would give some information about the accuracy of the approximation. Proving a.s. the convergence of the scheme for Bénard models is also a challenging question. Author Contributions: H.B. and A.M. contributed equally to this paper. Conceptualization, H.B. and A.M.; methodology, H.B. and A.M.; writing—original draft preparation, H.B. and A.M.; writ- ing—review and editing, H.B. and A.M. All authors have read and agreed to the published version of the manuscript. Funding: Hakima Bessaih was partially supported by Simons Foundation grant: 582264 and NSF grant DMS: 2147189. Data Availability Statement: Data sharing is not applicable to this article as no datasets were generated or analyzed during the current study. Acknowledgments: The authors thank anonymous referees for valuable remarks. Annie Millet’s research has been conducted within the FP2M federation (CNRS FR 2036). Conflicts of Interest: The authors have no conflict of interest to declare that are relevant to the content of this article. References 1. 2. 3. 4. Chueshov, I.; Millet, A. Stochastic 2D hydrodynamical type systems: Well posedness and large deviations. Appl. Math. Optim. 2010, 61, 379–420. [CrossRef] Duan, J.; Millet, A. Large deviations for the Boussinesq equations under random influences. Stoch. Process. Their Appl. 2009, 119, 2052–2081. [CrossRef] Breckner, H. Galerkin approximation and the strong solution of the Navier-Stokes equation. J. Appl. Math. Stoch. Anal. 2000, 13, 239–259. [CrossRef] Breit, D.; Dogson, A. Convergence rates for the numerical approximation of the 2D Navier-Stokes equations. Numer. Math. 2021, 147, 553–578. [CrossRef] Mathematics 2022, 10, 4246 39 of 39 5. 6. 7. 8. 9. Brze´zniak, Z.; Carelli, E.; Prohl, A. Finite element base discretizations of the incompressible Navier-Stokes equations with multiplicative random forcing. IMA J. Numer. Anal. 2013, 33, 771–824. [CrossRef] Carelli, E.; Prohl, A. Rates of convergence for discretizations of the stochastic incompressible Navier-Stokes equations. SIAM J. Numer. Anal. 2012, 50, 2467–2496. [CrossRef] Dörsek, P. Semigroup splitting and cubature approximations for the stochastic Navier-Stokes Equations. SIAM J. Numer. Anal. 2012, 50, 729–746. [CrossRef] Bessaih, H.; Brze´zniak, Z.; Millet, A. Splitting up method for the 2D stochastic Navier-Stokes equations. Stoch. PDE Anal. Comput. 2014, 2, 433–470. [CrossRef] Bessaih, H.; Millet, A. Strong L2 convergence of time numerical schemes for the stochastic two-dimensional Navier-Stokes equations. IMA J. Numer. Anal. 2019, 39, 2135–2167. [CrossRef] 10. Bessaih, H.; Millet, A. Space-time Euler discretization schemes for the stochastic 2D Navier-Stokes equations. Stoch. PDE Anal. Comput. 2021, 10, 1515–1558. [CrossRef] 11. Bessaih, H.; Millet, A. Strong rates of convergence of space-time discretization schemes for the 2D Navier-Stokes equations with additive noise. Stochastics Dyn. 2022, 22, 224005. [CrossRef] 12. Temam, R. Navier-Stokes Equations and Nonlinear Functional Analysis; CBMS-NSF Regional Conference Series in Applied Mathe- matics; 66. Society for Industrial and Applied Mathematics (SIAM): Philadelphia, PA, USA, 1995. 13. Giga, Y.; Miyakawa, T. Solutions in Lr of the Navier-Stokes Initial Value Problem. Arch. Ration. Anal. 1985, 89, 267–281. [CrossRef] 14. Da Prato, G.; Zabczyk, J. Stochastic Equations in Infinite Dimensions; Cambridge University Press: Cambridge, UK, 1992. 15. Walsh, J.B. An introduction to Stochastic Partial Differential Equations; In École d’Été de Probabilités de Saint-Flour XIV-1984; Lecture Notes in Mathematics 1180; Springer: Berlin/Heidelberg, Germany, 1986. 16. Girault, V.; Raviart, P.A. Finite Element Method for Navier-Stokes Equations: Theory and Algorithms; Springer: Berlin/Heidelberg, Germany; New York, NY, USA, 1981. 17. Printems, J. On the discretization in time of parabolic stochastic partial differential equations. M2AN Math. Model. Numer. Anal. 2001, 35, 1055–1078. [CrossRef] 18. Dijkstra, H.A. Nonlinear Physical Oceanography; Kluwer Academic Publishers: Boston, MA, USA, 2000. 19. Duan, J.; Gao, H.; Schmalfuss, B. Stochastic dynamics of a coupled atmosphere–ocean model. Stochastics Dyn. 2002, 2, 357–380. [CrossRef]
null
10.1186_s12889-023-16038-3.pdf
Availability of data and materials The datasets used and analyzed during this current study are available from the corresponding author on reasonable request.
Availability of data and materials The datasets used and analyzed during this current study are available from the corresponding author on reasonable request.
Chair et al. BMC Public Health (2023) 23:1081 https://doi.org/10.1186/s12889-023-16038-3 RESEARCH BMC Public Health Open Access Household air pollution from solid fuel use and depression among adults in rural China: evidence from the China Kadoorie Biobank data Sek Ying Chair1, Kai Chow Choi1, Mei Sin Chong1*, Ting Liu2 and Wai Tong Chien1 Abstract Background Solid fuels are still widely used for cooking in rural China, leading to various health implications. Yet, studies on household air pollution and its impact on depression remain scarce. Using baseline data from the China Kadoorie Biobank (CKB) study, we aimed to investigate the relationship between solid fuel use for cooking and depression among adults in rural China. Methods Data on exposure to household air pollution from cooking with solid fuels were collected and the Chinese version of the World Health Organization Composite International Diagnostic Interview short-form (CIDI-SF) was used to evaluate the status of major depressive episode. Logistic regression analysis was performed to investigate the asso- ciation between solid fuel use for cooking and depression. Results Amongst 283,170 participants, 68% of them used solid fuels for cooking. A total of 2,171 (0.8%) participants reported of having a major depressive episode in the past 12 months. Adjusted analysis showed that participants who had exposure to solid fuels used for cooking for up to 20 years, more than 20 to 35 years, and more than 35 years were 1.09 (95% CI: 0.94–1.27), 1.18 (95% CI: 1.01–1.38), and 1.19 (95% CI: 1.01–1.40) times greater odds of having a major depressive episode, respectively, compared with those who had no previous exposure to solid fuels used for cooking. Conclusion The findings highlight that longer exposure to solid fuels used for cooking would be associated with increased odds of major depressive episode. In spite of the uncertainty of causal relationship between them, using solid fuels for cooking can lead to undesirable household air pollution. Reducing the use of solid fuels for cooking by promoting the use of clean energy should be encouraged. Keywords Solid fuel, Cooking, Depression, Household air pollution Background Depression is one of the most common mental health disorders, affecting more than 280 million people glob- ally [1]. A recent systematic review and meta-analysis *Correspondence: Mei Sin Chong jomeisin@link.cuhk.edu.hk 1 The Nethersole School of Nursing, Faculty of Medicine, The Chinese University of Hong Kong, 6/F, Esther Lee Building, Horse Material Water, Shatin, New Territories, Hong Kong SAR, China 2 School of Nursing, Sun Yat Sen University, Guangzhou, China revealed that the 12-month and lifetime prevalence rates of major depressive disorder in China were 1.6% and 1.8%, respectively, and the percentages had been increas- ing over time [2]. If the population in China is estimated to be 1.426 billion in 2023 [3], the 12-month prevalence of major depressive disorder may reach over 22.8 million of individuals. A longitudinal population study in Aus- tralia suggested that the severity of depression is a major predictor for suicidal ideation and suicidal attempt [4]. Based on a recent meta-analysis on 15 studies, the preva- lence of suicidal attempt in a lifetime among individuals © The Author(s) 2023. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecom- mons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data. Chair et al. BMC Public Health (2023) 23:1081 Page 2 of 9 with major depressive disorder was 3.45 times higher than those without major depressive disorder [5]. A study conducted in mainland China reported that the prevalence of suicidal ideation was 16.7% among 1,916 patients (18–70  years old) with major depressive disor- der [6]. Symptoms of major depressive disorder, such as low mood, anhedonia and impaired cognition, are one of the key contributors to functional impairment [7], which could cause a great economic burden to the society. A study by Rayner et al. [8] reported that there was a signif- icant correlation between total healthcare costs (i.e., acci- dent and emergency department visits, hospitalizations, and visits to doctor) and depression. In addition, patients with multimorbidity and depression had more than twice the inpatient costs compared with those without depres- sion [9]. The estimated health burden from depression has been continuously increasing over the years. More recently, a systematic review reported that major depres- sive disorder was accounted for 49 million disability- adjusted life-years in 2020 [10]. Approximately 2.4 billion people still use solid fuels such as animal dung, wood, and coal for cooking glob- ally [11]. To date, solid fuels are extensively used in China, especially in rural households. There are half a billion people (40% of total population) in mainland China living in rural areas [12], with more than three- fourths of these rural households using solid fuels for cooking [13]. Incomplete combustion of solid fuels pro- duces compounds such as carbon monoxide, sulphur dioxide, black carbon and PM2.5(fine particulate matter with a diameter of less or equal to 2.5  μm) [14]. Con- centrations of PM2.5 from household air pollution due to cooking with solid fuels can be substantially high, causing up to 40% higher PM2.5exposure compared with the indoor and outdoor environments [15]. Previ- ous studies have confirmed that the exposure to solid fuel use contributes to adverse health effects such as sleep disturbance [16], chronic bronchitis and obstruc- tive pulmonary disease [17], hypertension [18] and an increased risk of cardiovascular disease hospitalization and stroke among rural population [19]. A nationwide prospective cohort study also reported significant asso- ciation between using solid fuels for cooking and car- diovascular mortality in China [20]. A longitudinal study conducted in the United States reported that individuals with previous 30-day exposure to ambient fine PM were 1.2 times more likely to have moderate to severe depressive symptoms [21]. Based on a national longitudinal survey in China, cooking with solid fuels was associated with a higher risk of depressive symptoms among individuals aged 60  years and above [22]. Similarly, a cohort study also reported positive association between solid fuel use and depression [21]. However, the study conducted by Pun et  al. [21] in the urban areas of the United States did not focus specifically on household use of solid fuels as the source of air pollu- tion. Meanwhile, the studies conducted by Li et  al. [22] and Shao et  al. [23] were restricted to middle-aged and older Chinese population living in urban or rural areas. Air pollutant emissions from solid fuels are associated with adverse health effects. In recent decades, improved cookstoves and combustion technologies have been implemented but a large number of individuals remain using solid fuels for cooking. Despite the important role of solid fuels in producing energy, its potential detrimen- tal impact on mental health demands urgent attention. There is a lack of studies focusing on the use of epidemio- logical data to investigate the association between solid fuel use and depression in developing countries; con- cerns arise as China is the second most populous coun- try where 40% of the population are living in rural areas and actively using solid fuels for cooking. It is crucial to investigate the relationship between these two variables specifically in rural China with a large-scale study. There- fore, this current study aimed to investigate the associa- tion between using solid fuels for cooking and depression in rural China. Methods Study design and population This study employed a secondary data analysis using the baseline data from the China Kadoorie Biobank (CKB) study. The original CKB study was conducted between 2004 and 2008, which recruited over 0.5 million adults from 10 regions across mainland China. After provid- ing informed written consent, each participant attended a face-to-face interview and a physical examination. A total of 512,681 adults aged between 30 and 79  years (without any major disability) with permanent residence were included in this baseline survey. A standardized electronic questionnaire was used to collect participant information including sociodemographic characteristics, lifestyle habits, exposure to passive smoking and domes- tic indoor air pollution, medical history, physical activity, and mental health status. The questionnaire used can be accessed via the official website of CKB (https:// www. ckbio bank. org/ study- resou rces/ survey- data). Each par- ticipant’s resting blood pressure (BP) was measured using the A&D digital BP monitor  (Model No.: UA-779). A body composition analyzer (Model No.: TBF-300GS) was used to measure body mass index (BMI), while a stand- ing height measuring instrument was used to measure weight and height. BMI is calculated by using this for- mula: the participant’s weight in kilograms (kg) divided by the square of height (H) in meters (m), (BMI = kg/ H2) [24]. BMI at 28  kg / m2is recommended as the cut-off Chair et al. BMC Public Health (2023) 23:1081 Page 3 of 9 point for obesity for the Chinese people [25, 26]. More detailed information about the original CKB study has been previously reported [27–29]. Ethics approvals for the CKB study were obtained from the Chinese Center for Disease Control and Prevention (Approval Notice 005/2004) and the Oxford Tropical Research Ethics Committee (OxTREC Ref: 025–04) of the University of Oxford. The study was conducted in line with the princi- ples outlined in the Declaration of Helsinki. Measurements Exposure to household use of solid fuels The approach of Yu et al. [20] was followed to calculate the durations of exposure to solid fuels used for cooking and heating separately. Participants were asked to pro- vide detailed information about their exposure to house- hold use of solid fuels for cooking and heating, including related information such as the duration (in years) they lived in their three most recent residences, frequency of cooking in each residence, types of fuels used for cook- ing and heating, and availability of cookstove ventila- tion (chimney or extractor). Participants who reported that they cooked less often than once a month in a resi- dence were considered as noncooking and regarded as having no exposure to solid fuel used for cooking. Par- ticipants who reported that they cooked at least once a month were then asked to provide additional infor- mation related to the types of primary fuels they used. There are two categories of primary fuels, namely “clean fuels” such as gas and electricity, and “solid fuels” such as wood and coal [30]. The total duration (in years) of household use of solid fuels for cooking was calculated by summing up the duration of using solid fuels as the primary cooking fuel in each residence. Likewise, par- ticipants who used solid fuels for heating in winter were asked further questions about the types of primary fuels they used, and the total duration (in years) of household use of solid fuels for heating was calculated by sum- ming up the corresponding duration in each residence. The level of exposure to solid fuels used for heating was estimated by multiplying a weight coefficient to years of solid fuels used for heating, of which the weight coef- ficient was calculated based on the average portion of years with temperature less than 8 degree Celsius in each of the residences from 1999 to 2013, ranging from 0.18 to 0.42, as detailed in Yu et al. [20]. Depression In this study, major depressive episode was evaluated by the Chinese version of the World Health Organization Composite International Diagnostic Interview short- form (CIDI-SF) [31]. As there is no gold standard for assessing mental disorders in the CIDI-SF, this version was calibrated rather than validated and produced simi- lar population estimates of major depressive episode to the Structured Clinical Interview for DSM-IV, which is a state-of-the-art clinical research diagnostic interview tool for mental disorders [32]. Participants were first asked whether they had any of the following symptoms lasting for ≥ 2  weeks in the past 12  months: a) feeling much saddened, or depressed than usual; b) loss of inter- est in most things like hobbies or activities that usually gave you pleasure; c) feeling so hopeless and loss of appe- tite even for your favorite food; d) feeling worthless and useless, everything that went wrong was your fault, and life was very difficult with no way out. If participants answered “yes” to any of the above-mentioned situations, they were further assessed for major depression using CIDI-SF through a face-to-face interview by trained health professionals. Participants who reported at least 3 out of 7 depression symptoms (i.e., 1) weight change, 2) difficulty in sleeping, 3) losing interest in things, 4) feeling tired or low on energy, 5) trouble concentrating, 6) feeling worthless, or 7) thoughts about death) in the CIDI-SF questionnaire were considered likely to have major depression [33]. Covariates Adjustment for covariates was performed in this analy- sis, including sociodemographic characteristics (i.e., age, gender, marital status, education level and annual house- hold income), lifestyle habits (i.e., smoking status, alcohol assumption, and physical activity), health status (i.e., BMI and blood pressure), stressful life events in the past two years, passive smoking, cookstove ventilation, and expo- sure to solid fuels used for heating. Smoking status was classified into four categories: 1) never smoke, 2) quit- ted, 3) occasional smoker, and 4) current smoker. Partici- pants were classified as a “regular alcohol drinker” if they reported that they drank alcohol “usually at least once a week.” Otherwise, they were classified as a “non-regular drinker.” Physical activity was estimated as metabolic equivalent task hours per day spent on activities related to occupation, commuting, housework, and non-sed- entary leisure-time activities. Exposure to stressful life events (Yes/No) was defined as the occurrence of com- mon major life events in the past two years, such as death of a spouse, marital separation/divorce, traffic accident and major natural disaster. Exposure to passive smok- ing was assessed by self-report responses to the question related to frequency of secondhand smoking exposure. The variable was categorized into 4 levels (none, > 0 to 2  h/week, > 2 to 12  h/week, > 12  h/week). The cut-off points were conventionally selected based on the tertile points among those who had exposure to passive smok- ing, with the three exposure categories being anticipated Chair et al. BMC Public Health (2023) 23:1081 Page 4 of 9 to reflect low, middle and high levels of exposure to pas- sive smoking. non-exposure group with over 80% power at 2-sided 5% level of significance. Statistical analysis All statistical analyses were conducted using the IBM SPSS 25.0 (IBM Corp., Armonk, NY). Data were sum- marized descriptively using statistics including means, standard deviations, frequencies and percentages. For continuous variables, skewness statistics and normal- ity probability plots were used to assess normality. In this study, the outcome of interest was status of major depressive episode in the past year (Yes/No). The pri- mary exposure of interest was duration of solid fuels used for cooking which was categorized into four lev- els. Specifically, those participants who had no previ- ous exposure to solid fuels used for cooking or always used clean fuels were categorized as the reference group. The remaining participants were conventionally stratified into three tertiles to characterize low, mid- dle and high levels of exposure with totally four levels for the exposure factor: (i) none, (ii) > 0 to 20  years, (iii) > 20 to ≤ 35 years, (iv) > 35 years. Likewise, the expo- sure to solid fuels used for heating was categorized into four levels: (i) none, (ii) > 0 to 8.2  years, (iii) > 8.2 to ≤ 13.5 years, (iv) > 13.5 years. The association between major depression in the past year and exposure to solid fuels used for cooking was examined by logistic regres- sion analysis. Unadjusted and adjusted logistic regres- sion analyses were conducted with adjustment for the covariates of sociodemographic characteristics and life- style habits, presence of stressful life events in the past two years, presence of cookstove ventilation, exposure to passive smoking, and level of exposure to solid fuels used for heating. As the time scope of the outcome of major depressive episode was the past 12 months from the time of survey, it was possible that some partici- pants might have a major depressive episode prior to exposure to solid fuel usage. A sensitivity analysis was therefore conducted by excluding those participants who had no more than one year of solid fuel usage before the survey. All tests involved were 2-sided at 5% level of significance. A total of 283,170 participants were included in this secondary data analysis study. Among them, 2,171 par- ticipants were classified as having major depressive episode in the past year, and there were totally 91,611 participants without exposure to solid fuels used for cooking and 61,873 to 65,612 participants with differ- ent levels of exposure to solid fuels used for cooking. Such a sample size is adequate to detect an odds ratio of having major depressive episode of as small as 1.17 when comparing anyone of the exposure groups with the Results Characteristics of the study population Amongst 283,170 participants who were included in the baseline survey of the CKB study, the average age was 51.4 (SD = 10.5) years, and 58.2% of them were female. About 68% of them used solid fuels for cook- ing, with a 27-year median. More than half of the study sample (67%) had at least some cookstove ventilation. Nearly 23% participants had exposure to passive smok- ing for more than 12 h per week. A total of 2,171 (0.8%) participants reported major depressive episode in the past year. Characteristics of the study population strati- fied by levels of exposure to solid fuels used for cooking are shown in Table 1. Association between household use of solid fuels for cooking and major depressive episode Based on their duration of exposure to solid fuels used for cooking, participants were categorized into four lev- els: (i) none, (ii) > 0 to 20  years, (iii) > 20 to ≤ 35  years, (iv) > 35  years. Those participants who had no previous exposure to solid fuels used for cooking or always used clean fuels for cooking were categorized as the refer- ence group (none exposure). The remaining partici- pants were conventionally stratified into three tertiles to characterize low, middle and high levels of exposure. Unadjusted logistic regression analysis showed that an increased level of exposure to solid fuels used for cooking was associated with an increased odds of hav- ing a major depressive episode (unadjusted model in Table  2). After adjusting for sociodemographic charac- teristics, obesity and lifestyle habits, presence of stress- ful life events, presence of cookstove ventilation, passive smoking exposure, and level of exposure to solid fuels used for heating, the pattern of association between an increased odds of having a major depressive epi- sode and an increased level of exposure was also noted. Participants who had exposure to solid fuels used for cooking for up to 20  years, more than 20 to 35  years, and more than 35  years were 1.09 (95% CI 0.94–1.27), 1.18 (95% CI: 1.01–1.38) and 1.19 (95% CI: 1.01–1.40) times greater odds of having a major depressive episode, respectively, compared with those who had no previous exposure to solid fuel used for cooking or always used clean fuels for cooking (adjusted model 1 in Table 2). A sensitivity analysis was conducted by excluding those participants who had no more than one year of solid fuel usage before the survey, the results were similar to the primary analysis one (adjusted model 2 in Table 2). Chair et al. BMC Public Health (2023) 23:1081 Page 5 of 9 Table 1 Characteristics of the study population by level of exposure to solid fuels used for cooking (N = 283,170) Level of exposure to solid fuels used for cooking All (N = 283,170) None (n = 91,611) > 0 to 20 years (n = 64,524) > 20 to 35 years (n = 65,162) > 35 years (n = 61,873) 46,577 (16.4%) 84,487 (29.8%) 88,697 (31.3%) 48,441 (17.1%) 14,968 (5.3%) 15,809 (17.3%) 20,298 (31.5%) 5053 (7.8%) 27,860 (30.4%) 21,798 (33.8%) 23,623 (36.3%) 27,153 (29.6%) 13,257 (20.5%) 28,678 (44.0%) 15,640 (17.1%) 6924 (10.7%) 5149 (5.6%) 2247 (3.5%) 6181 (9.5%) 1627 (2.5%) 118,260 (41.8%) 79,634 (86.9%) 20,576 (31.9%) 9111 (14.0%) 164,910 (58.2%) 11,977 (13.1%) 43,948 (68.1%) 56,051 (86.0%) 5417 (8.8%) 11,206 (18.1%) 19,609 (31.7%) 19,696 (31.8%) 5945 (9.6%) 8939 (14.4%) 52,934 (85.6%) 259,280 (91.6%) 87,231 (95.2%) 60,204 (93.3%) 60,226 (92.4%) 51,619 (83.4%) 21,671 (7.7%) 2219 (0.8%) 3666 (4.0%) 714 (0.8%) 3906 (6.1%) 414 (0.6%) 4551 (7.0%) 385 (0.6%) 9548 (15.4%) 706 (1.1%) Characteristics Demographics Age (years) 30 – < 40 40 – < 50 50 – < 60 60 – < 70 ≥ 70 Sex Male Female Marital status Married Widowed / separated / divorced Never married Highest education attainment No formal school Primary school 67,740 (23.9%) 13,190 (14.4%) 12,506 (19.4%) 18,147 (27.8%) 118,593 (41.9%) 37,707 (41.2%) 24,332 (37.7%) 28,739 (44.1%) Middle school / high school 93,744 (33.1%) 38,923 (42.5%) 26,722 (41.4%) 18,032 (27.7%) Technical school / college/ university 3093 (1.1%) 1791 (2.0%) 964 (1.5%) 244 (0.4%) Household income in last year (Yuan) < 5,000 5,000 – 9,999 10,000 – 19,999 20,000 – 34,999 ≥ 35,000 Obesity status and lifestyle characteristics Obesity status 41,918 (14.8%) 70,752 (25.0%) 81,352 (28.7%) 54,285 (19.2%) 34,863 (12.3%) 10,238 (11.2%) 6634 (10.3%) 8255 (12.7%) 20,794 (22.7%) 14,517 (22.5%) 16,663 (25.6%) 25,492 (27.8%) 19,283 (29.9%) 20,737 (31.8%) 19,522 (21.3%) 14,783 (22.9%) 12,725 (19.5%) 15,565 (17.0%) 9307 (14.4%) 6782 (10.4%) Normal weight (18.5 ≤ BMI < 23.9) Overweight (24.0 ≤ BMI < 27.9) Obese (BMI ≥ 28.0) Under weight (BMI < 18.5) 162,222 (57.3%) 56,475 (61.6%) 36,709 (56.9%) 34,507 (53.0%) 82,596 (29.2%) 24,858 (27.1%) 19,271 (29.9%) 20,821 (32.0%) 22,982 (8.1%) 15,368 (5.4%) 5442 (5.9%) 4835 (5.3%) 5379 (8.3%) 3165 (4.9%) 6813 (10.5%) 3021 (4.6%) 23,897 (38.6%) 27,815 (45.0%) 10,067 (16.3%) 94 (0.2%) 16,791 (27.1%) 18,778 (30.3%) 15,840 (25.6%) 7255 (11.7%) 3209 (5.2%) 34,531 (55.8%) 17,646 (28.5%) 5348 (8.6%) 4347 (7.0%) Smoking status Never smoke Quitted Occasional smoker Current smoker Regular alcohol drinker No Yes Physical activity, MET – hours/daya Stressful life event & sleep disturbance Stressful life event in the past two years 175,263 (61.9%) 23,540 (25.7%) 46,351 (71.8%) 55,238 (84.8%) 50,134 (81.0%) 16,919 (6.0%) 12,954 (4.6%) 9841 (10.7%) 7445 (8.1%) 3181 (4.9%) 2438 (3.8%) 1614 (2.5%) 1455 (2.2%) 78,034 (27.6%) 50,785 (55.4%) 12,554 (19.5%) 6855 (10.5%) 2283 (3.7%) 1616 (2.6%) 7840 (12.7%) 245,061 (86.5%) 69,241 (75.6%) 57,499 (89.1%) 60,967 (93.6%) 57,354 (92.7%) 38,109 (13.5%) 22,370 (24.4%) 7025 (10.9%) 23.2 (14.5) 24.8 (16.2) 24.4 (14.7) 4195 (6.4%) 22.2 (13.0) 4519 (7.3%) 20.7 (12.4) No Yes 261,388 (92.3%) 85,582 (93.4%) 59,660 (92.5%) 59,988 (92.1%) 56,158 (90.8%) 21,782 (7.7%) 6029 (6.6%) 4864 (7.5%) 5174 (7.9%) 5715 (9.2%) Cook stove ventilation & passive smoking exposure Had at least some cook stove ventilation No 94,242 (33.3%) 27,318 (29.9%) 23,802 (36.9%) 24,207 (37.2%) 18,915 (30.6%) Chair et al. BMC Public Health (2023) 23:1081 Page 6 of 9 Table 1 (continued) Characteristics Yes Passive smoking exposure None > 0 to 2 h/week > 2 to 12 h/week > 12 h/week Level of exposure to solid fuels used for cooking All (N = 283,170) None (n = 91,611) > 0 to 20 years (n = 64,524) > 20 to 35 years (n = 65,162) > 35 years (n = 61,873) 188,570 (66.7%) 64,034 (70.1%) 40,661 (63.1%) 40,934 (62.8%) 42,941 (69.4%) 91,358 (32.3%) 62,116 (21.9%) 65,490 (23.1%) 64,206 (22.7%) 28,650 (31.3%) 22,290 (34.5%) 19,727 (30.3%) 18,898 (20.6%) 14,138 (21.9%) 15,068 (23.1%) 22,473 (24.5%) 14,402 (22.3%) 14,921 (22.9%) 21,590 (23.6%) 13,694 (21.2%) 15,446 (23.7%) Solid fuels usage for heating Level of exposure to solid fuels used for heatingb None > 0 to 8.2 years > 8.2 to 13.5 years > 13.5 years Major depression Major depression episode in the past year 108,153 (38.5%) 38,143 (42.0%) 26,444 (41.3%) 22,064 (34.1%) 57,408 (20.4%) 59,153 (21.1%) 56,252 (20.0%) 13,163 (14.5%) 18,968 (29.6%) 16,993 (26.3%) 19,568 (21.5%) 11,307 (17.6%) 14,500 (22.4%) 19,975 (22.0%) 7385 (11.5%) 11,155 (17.2%) 20,691 (33.4%) 14,012 (22.6%) 13,694 (22.1%) 13,476 (21.8%) 21,502 (35.1%) 8284 (13.5%) 13,778 (22.5%) 17,737 (28.9%) No Yes 280,999 (99.2%) 91,161 (99.5%) 64,032 (99.2%) 64,548 (99.1%) 61,258 (99.0%) 2171 (0.8%) 450 (0.5%) 492 (0.8%) 614 (0.9%) 615 (1.0%) Variables with data marked with a are presented as mean (standard deviation), all others are presented as frequency (%) b There were less than 0.8% (n = 2204) of participants without detailed information about fuels used for heating Table 2 Risk of major depression episode by level of exposure to solid fuels used for cooking Sensitivity analysis Unadjusted model Adjusted model 1 Adjusted model 2 Exposure factor Odds ratio (95% CI) p-value Odds ratio (95% CI) p-value Odds ratio (95% CI) p-value Level of exposure to solid fuels used for cooking None (ref ) 1 > 0 to 20 years 1.56 (1.37 – 1.77) > 20 to 35 years 1.93 (1.71 – 2.18) > 35 years 2.03 (1.80 – 2.30) < 0.001 < 0.001 < 0.001 1 1.09 (0.94 – 1.27) 1.18 (1.01 – 1.38) 1.19 (1.01 – 1.40) 0.249 0.034 0.033 1 1.08 (0.93 – 1.25) 1.18 (1.01 – 1.38) 1.18 (1.01 – 1.39) 0.316 0.041 0.043 Model 1: with adjustment for demographic, obesity and lifestyle characteristics, presence of stressful life event, presence of cook stove ventilation, passive smoking exposure, and level of exposure to solid fuels used for heating as listed in Table 1 Model 2: with adjustment for covariates in model 1 and excluding those participants who used solid fuels for cooking but had no more than one year of solid fuel used for cooking into analysis (n = 622 excluded) ref reference category for calculating odds ratios of other comparison categories Discussion Approximately 46% of the population in China used solid fuels as a household energy source, leading to household air pollution; and the proportion was sub- stantially higher in rural areas [13, 23]. In fact, the pre- sent study found that 68% of rural residents used solid fuels for cooking. To the best of our knowledge, this is the largest national study to explore the relationship between solid fuel use and depression in rural China. The results revealed an association between household use of solid fuels for cooking and major depression, particularly for those who had used solid fuels for more than 20 years, after controlling for potential confound- ing covariates, including sociodemographic charac- teristics, lifestyle habits, health status, presence of stressful life events, presence of cookstove ventilation, passive smoking exposure, and exposure to solid fuels used for heating. Although participants with longer exposure generally associated with an increased odds of having a major depressive episode, the odds ratio of the longest exposure group (> 35 years, OR = 1.19) was unexpectedly similar to the second longest exposure Chair et al. BMC Public Health (2023) 23:1081 Page 7 of 9 group (> 20 to 35 years, OR = 1.18). A possible explana- tion may be owing to the fact that people with longer exposure were more likely subject to a competing risk of death, which may diminish the strength of associa- tion, particularly in the longest exposure group. There is a growing body of evidence that solid fuel use is associated with a high risk of depression [22, 23], which is consistent with the current findings. Individu- als (N= 8637) with exposure to solid fuel combustion for over 4 years had 1.12 times greater odds of having depres- sive symptoms [23]. Supported by the following longitu- dinal survey (N= 7005) [22], individuals using solid fuels in cooking for more than 7 years had 1.36 times greater odds of depression risk than those who always used clean fuels. This study, together with the aforementioned previous studies, provides evidence on the association between the exposure to solid fuels and the prevalence of depres- sion. However, only limited evidence exists on the mech- anisms linking the use of solid fuels for cooking with depression. The incomplete combustion of solid fuels generates various air pollutants including PM, carbon monoxide, sulfur oxides, and polycyclic aromatic hydro- carbons [34, 35]. One possible explanation may be that inhalation of air pollutants can trigger associated oxida- tive stress, cerebrovascular damage, neuroinflammation, and neurodegenerative pathology, which all might cause or exacerbate the risk of depression [36–38]. Animal experience revealed that PM might cause neurotoxic- ity by inducing microglia activation characterized by the release of TNFα, which damages the olfactory bulb and increases depression risk [39]. Moreover, studies indicated that PM causes elevated levels of cortisol [40], which has been related to the development of depres- sion [41]. Furthermore, domestic cooking with solid fuels could increase the risk of chronic diseases, such as cancer and cardiorespiratory diseases [19, 42], which are strongly associated with depression [43, 44]. In rural China, solid fuels are reported to be the domi- nant cooking fuel, with biomass and coal accounting for 47.6% [45] and 13.5% [46], respectively. Our study gives valuable insights into the potential hazardous effects of using solid fuels for cooking on mental health. It indicates household solid fuels used for cooking is a critical public health issue and that policy makers must take responsi- bility to make the needed policy changes. It is necessary to encourage people to switch to cleaner fuels and tech- nologies when cooking to reduce exposure to household air pollution. Moreover, in this study, depressive episode was more prevalent in those without cookstove ventila- tion. This result is in line with those of the previous stud- ies [22, 47], showing that cooking ventilation may weaken the relationship of cooking with solid fuel and long duration cooking with depressive symptoms,  suggest- ing that improvements in cooking ventilation should be strongly encouraged. As a remark, although people with longer exposure to solid fuels used for cooking generally associated with an increased odds of having a major depression episode, the odds ratio of the longest exposure group (> 35 years, OR = 1.19) was unexpectedly similar to the second long- est exposure group (> 20 to 35 years, OR = 1.18). A possi- ble explanation may be owing to the fact that people with longer exposure were more likely subject to a competing risk of death, which may diminish the strength of associa- tion, particularly in the longest exposure group. Despite the significance of the findings, there are sev- eral limitations in this study that may impact the gener- alisability of this study. First, the cross-sectional study design assesses both outcome of interest and exposure simultaneously. Therefore, it may not be able to estab- lish a cause-and-effect relationship between household solid fuels used for cooking and depression. In addi- tion, self-reported information is prone to recall bias when participants fail to accurately remember an event in the past. Nonetheless, the overestimation or underes- timation of association between cause and effect may be resolved through a longitudinal cohort study in which an event may be observed first, followed by the effects. On the other hand, different cooking practices and chemical properties of fuel such as density, volatility and thermal capacity which could affect the indoor air pollution were not examined in the CKB study. Hence, this could result in imprecision of actual exposure to solid fuels used for cooking. Although this study had controlled for poten- tial confounders (e.g., sociodemographic characteristics, obesity status and lifestyle habits, presence of stressful life events, presence of cookstove ventilation and passive smoking exposure), the results might be confounded by other unmeasured covariates. This is because our study was a secondary data analysis where the adjusted analysis was only able to be performed based on existing available variables. Conclusion This study demonstrates the significant association between the use of household solid fuels for cooking and the prevalence of depression in rural China; and the longer duration of exposure, the higher odds of having a depressive episode. Further studies are warranted to examine if there is a causal relationship between them. Nevertheless, reducing the use of solid fuels for cook- ing by promoting the use of clean energy should be encouraged. Acknowledgements Not applicable. Chair et al. BMC Public Health (2023) 23:1081 Page 8 of 9 Authors’ contributions Sek Ying Chair contributed to the conceptualization of the manuscript, writing of the original draft, reviewing and editing the manuscript. Kai Chow Choi contributed to the formal analysis, data curation, writing, reviewing and editing the manuscript. Mei Sin Chong contributed to writing, reviewing and editing the manuscript. Ting Liu contributed to writing, reviewing and editing the manuscript. Wai Tong Chien contributed to writing, reviewing and editing the manuscript. All authors read and approved the final manuscript. Funding This study was funded by grants from the National Key Research and Development Program of China (2016YFC0900500, 2016YFC0900501, 2016YFC0900504), the Kadoorie Charitable Foundation in Hong Kong, and Wellcome Trust (088158/Z/09/Z, 104085/Z/14/Z, 104085/Z/14/Z) in the UK. Availability of data and materials The datasets used and analyzed during this current study are available from the corresponding author on reasonable request. Declarations Ethics approval and consent to participate The CKB study was conducted in line with the principles outlined in the Declaration of Helsinki; ethics approvals were obtained from the Chinese Center for Disease Control and Prevention and the Oxford Tropical Research Ethics Committee of the University of Oxford; and an informed consent was obtained from each participant [26]. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Received: 16 March 2023 Accepted: 1 June 2023 References 1. World Health Organization. Depression. 2021. https:// www. who. int/ news- room/ fact- sheets/ detail/ depre ssion. Accessed 20 Nov 2022. 2. Zhao YJ, Jin Y, Rao WW, et al. Prevalence of major depressive disorder among adults in China: a systematic review and meta-analysis. Front Psychiatry. 2021;12:659470. https:// doi. org/ 10. 3389/ fpsyt. 2021. 659470. 3. United Nations. India overtakes China as the world’s most populous country. 2023. https:// www. un. org/ devel opment/ desa/ pd/ sites/ www. un. org. devel opment. desa. pd/ files/ undesa_ pd_ 2023_ policy- brief- 153. pdf . Accessed 3 May 2023. 4. Handley T, Rich J, Davies K, Lewin T, Kelly B. The challenges of predicting suicidal thoughts and behaviours in a sample of rural Australians with depression. Int J Environ Res Public Health. 2018;15(5):928. https:// doi. org/ 10. 3390/ ijerp h1505 0928. 5. Cai H, Xie XM, Zhang Q, Cui X, Lin JX, Sim K, Ungvari GS, Zhang L, Xiang YT. Prevalence of suicidality in major depressive disorder: a systematic review and meta-analysis of comparative studies. Front Psychiatry. 2021;12:690130. https:// doi. org/ 10. 3389/ fpsyt. 2021. 690130. 6. Ge F, Jiang J, Wang Y, Yuan C, Zhang W. Identifying suicidal ideation among Chinese patients with major depressive disorder: evidence from a real-world hospital-based study in China. Neuropsychiatr Dis Treat. 2020;16:665–72. https:// doi. org/ 10. 2147/ NDT. S2382 86. 7. Chow TK, Bowie CR, Morton M, Lalovic A, McInerney SJ, Rizvi SJ. Contribu- tors of functional impairment in major depressive disorder: a biopsycho- social approach. Curr Behav Neurosci Rep. 2022;9:59–72. https:// doi. org/ 10. 1007/ s40473- 022- 00247-y. 8. Rayner L, Hotopf M, Petkova H, Matcham F, Simpson A, McCracken LM. Depression in patients with chronic pain attending a specialised pain treatment centre: prevalence and impact on health care costs. Pain. 2016;157(7):1472–9. https:// doi. org/ 10. 1097/j. pain. 00000 00000 000542. 9. Bock JO, Luppa M, Brettschneider C, et al. Impact of depression on health care utilization and costs among multimorbid patients–from the Multi- Care Cohort Study. PLoS One. 2014;9(3):e91973. https:// doi. org/ 10. 1371/ journ al. pone. 00919 73. 10. COVID-19 Mental Disorders Collaborators. Global prevalence and burden of depressive and anxiety disorders in 204 countries and territories in 2020 due to the COVID-19 pandemic. Lancet (London, England). 2021;398:1700–12. https:// doi. org/ 10. 1016/ S0140- 6736(21) 02143-7. 11. World Health Organization. Household air pollution: key facts. 2022. https:// www. who. int/ news- room/ fact- sheets/ detail/ house hold- air- pollu tion- and- healt h#: ~: text= World wide% 2C% 20aro und% 202.4% 20bil lion% 20peo ple,and% 20i . Accessed 22 Jan 2023. 12. NBSC. China statistical yearbook-2019. Beijing: China Statistics Press; 2019. 13. Tang X, Liao H. Energy poverty and solid fuels use in rural China: analysis based on national population census. Energy Sustain Dev. 2014;23:122–9. https:// doi. org/ 10. 1016/J. ESD. 2014. 08. 006. 14. Shen G, Xing R, Zhou Y, et al. Revisiting the proportion of clean house- hold energy users in rural China by accounting for energy stacking. Sustainable Horizons. 2022;1:100010. https:// doi. org/ 10. 1016/j. horiz. 2022. 100010. 15 Xu H, Li Y, Guinot B, et al. Personal exposure of PM2. 5 emitted from solid fuels combustion for household heating and cooking in rural Guanzhong Plain, northwestern China. Atmos Environ. 2018;185:196–206. https:// doi. org/ 10. 1016/j. atmos env. 2018. 05. 018. 16. Chair SY, Choi KC, Cao X, et al. Association between household solid fuel use for cooking and sleep disturbance in rural China: findings from the China Kadoorie Biobank data. Sleep Med. 2021;83:13–20. https:// doi. org/ 10. 1016/J. SLEEP. 2021. 04. 029. 17. Po JY, FitzGerald JM, Carlsten C. Respiratory disease associated with solid biomass fuel exposure in rural women and children: systematic review and meta-analysis. Thorax. 2011;66:232–9. https:// doi. org/ 10. 1136/ THX. 2010. 147884. 18. Yan Z, Liu Y, Yin Q, et al. Impact of household solid fuel use on blood pres- sure and hypertension among adults in China. Air Qual Atmos Health. 2016;9:931–40. https:// doi. org/ 10. 1007/ S11869- 016- 0395-2/ TABLES/5. 19. Hystad P, Duong M, Brauer M, et al. Health effects of household solid fuel use: Findings from 11 Countries within the prospective urban and rural epidemiology study. Environ Health Perspect. 2019;127(5):57003. https:// doi. org/ 10. 1289/ EHP39 15. 20. Yu K, Qiu G, Chan KH, et al. Association of solid fuel use with risk of cardio- vascular and all-cause mortality in rural China. JAMA. 2018;319(13):1351– 61. https:// doi. org/ 10. 1001/ jama. 2018. 2151. 21. Pun VC, Manjourides J, Suh H. Association of ambient air pollution with depressive and anxiety symptoms in older adults: Results from the NSHAP study. Environ Health Perspect. 2017;125(3):342–8. https:// doi. org/ 10. 1289/ EHP494. 22. Li C, Zhou Y, Ding L. Effects of long-term household air pollution expo- sure from solid fuel use on depression: evidence from national longitudi- nal surveys from 2011 to 2018. Environ Pollut. 2021;283:117350. https:// doi. org/ 10. 1016/j. envpol. 2021. 117350. 23. Shao J, Ge T, Liu Y, Zhao Z, Xia Y. Longitudinal associations between household solid fuel use and depression in middle-aged and older Chi- nese population: a cohort study. Ecotoxicol Environ Saf. 2021;209:111833. https:// doi. org/ 10. 1016/j. ecoenv. 2020. 111833. 24. Lemmens HJ, Brodsky JB, Bernstein DP. Estimating ideal body weight–a new formula. Obes Surg. 2005;15(7):1082–3. 25. Xu W, Zhang H, Paillard-Borg S, Zhu H, Qi X, Rizzuto D. Prevalence of over- weight and obesity among Chinese adults: role of adiposity indicators and age. Obes Facts. 2016;9(1):17–28. https:// doi. org/ 10. 1159/ 00044 3003. 26. Zhou B, Coorperative Meta-Analysis Group of China Obesity Task Force. Predictive values of body mass index and waist circumference to risk fac- tors of related diseases in Chinese adult population. Zhonghua Liu Xing Bing Xue Za Zhi. 2002;23(1):5–10. 27. Chen Z, Lee L, Chen J, et al. Cohort profile: the Kadoorie Study of Chronic Disease in China (KSCDC). Int J Epidemiol. 2005;34(6):1243–9. https:// doi. org/ 10. 1093/ ije/ dyi174. 28 Chen Z, Chen J, Collins R, et al. China Kadoorie Biobank of 0.5 million peo- ple: survey methods, baseline characteristics and long-term follow-up. Int J Epidemiol. 2011;40(6):1652–66. https:// doi. org/ 10. 1093/ ije/ dyr120. Chair et al. BMC Public Health (2023) 23:1081 Page 9 of 9 Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in pub- lished maps and institutional affiliations. 29. Li LM, Lv J, Guo Y, et al. The China Kadoorie Biobank: Related methodol- ogy and baseline characteristics of the participants. Zhonghua Liu Xing Bing Xue Za Zhi. 2012;33(3):249–55. 30. Wang S, Luo K. Life expectancy impacts due to heating energy utilization in China: Distribution, relations, and policy implications. Sci Total Environ. 2018;610–611:1047–56. https:// doi. org/ 10. 1016/j. scito tenv. 2017. 08. 195. 31. Kessler RC, Andrews G, Mroczek D, et al. The World Health Organization Composite International Diagnostic Interview short-form (CIDI-SF). Int J of Methods Psychiatr Res. 1998;7:171–85. https:// doi. org/ 10. 1002/ MPR. 47. 32. Sun X, Zheng B, Lv J, Guo Y, Bian Z, Yang L, Chen Y, Fu Z, Guo H, Liang P, Chen Z, Chen J, Li L, Yu C, China Kadoorie Biobank (CKB) Collaborative Group. Sleep behavior and depression: findings from the China Kadoorie Biobank of 0.5 million Chinese adults. J Affect Disord. 2018;229:120–4. https:// doi. org/ 10. 1016/j. jad. 2017. 12. 058. 33 Meng R, Yu C, Liu N, et al. Association of depression with all-cause and cardiovascular disease mortality among adults in China. JAMA Netw Open. 2020;3(2):e1921043. https:// doi. org/ 10. 1001/ jaman etwor kopen. 2019. 21043. Published 2020 Feb 5. 34. Shen G, Chen Y, Du W, et al. Exposure and size distribution of nitrated and oxygenated polycyclic aromatic hydrocarbons among the population using different household fuels. Environ Pollut. 2016;216:935–42. https:// doi. org/ 10. 1016/j. envpol. 2016. 07. 002. 35. Zhang H, Zhu T, Wang S, et al. Indoor emissions of carbonaceous aerosol and other air pollutants from household fuel burning in Southwest China. Aerosol Air Qual Res. 2014;14:1779–88. https:// doi. org/ 10. 4209/ aaqr. 2013. 10. 0305. 36. Bhatt S, Nagappa AN, Patil CR. Role of oxidative stress in depression. Drug Discov Today. 2020;25(7):1270–6. https:// doi. org/ 10. 1016/j. drudis. 2020. 05. 001. 37. Calderón-Garcidueñas L, Calderón-Garcidueñas A, Torres-Jardón R, Avila- Ramírez J, Kulesza RJ, Angiulli AD. Air pollution and your brain: what do you need to know right now. Prim Health Care Res Dev. 2015;16(4):329– 45. https:// doi. org/ 10. 1017/ S1463 42361 40003 6X. 38. Hurley LL, Tizabi Y. Neuroinflammation, neurodegeneration, and depression. Neurotox Res. 2013;23(2):131–44. https:// doi. org/ 10. 1007/ s12640- 012- 9348-1. 39. Ji X, Liu R, Guo J, et al. Olfactory bulb microglia activation mediated neuronal death in real-ambient particulate matter exposure mice with depression-like behaviors. Sci Total Environ. 2022;821:153456. https:// doi. org/ 10. 1016/j. scito tenv. 2022. 153456. 40 Li H, Cai J, Chen R, et al. Particulate matter exposure and stress hormone levels: a randomized, double-blind, crossover trial of air purification [published correction appears in Circulation. 2017 Sep 12;136(11):e199]. Circulation. 2017;136(7):618–27. https:// doi. org/ 10. 1161/ CIRCU LATIO NAHA. 116. 026796. 41. Stetler C, Miller GE. Depression and hypothalamic-pituitary-adrenal acti- vation: a quantitative summary of four decades of research. Psychosom Med. 2011;73(2):114–26. https:// doi. org/ 10. 1097/ PSY. 0b013 e3182 0ad12b. 42. Balmes JR. Household air pollution from domestic combustion of solid fuels and health. J Allergy Clin Immunol. 2019;143(6):1979–87. https:// doi. org/ 10. 1016/j. jaci. 2019. 04. 016. 43. Carnevali L, Montano N, Statello R, Sgoifo A. Rodent models of depression-cardiovascular comorbidity: bridging the known to the new. Neurosci Biobehav Rev. 2017;76(Pt A):144–53. https:// doi. org/ 10. 1016/j. neubi orev. 2016. 11. 006. 44. Carvalho AF, Hyphantis T, Sales PM, et al. Major depressive disorder in breast cancer: a critical systematic review of pharmacological and psy- chotherapeutic clinical trials. Cancer Treat Rev. 2014;40(3):349–55. https:// doi. org/ 10. 1016/j. ctrv. 2013. 09. 009. 45. Hou B, Liao H, Huang J. Household cooking fuel choice and economic poverty: Evidence from a nationwide survey in China. Energy and Buil. 2018;166:319–29. https:// doi. org/ 10. 1016/j. enbui ld. 2018. 02. 012. 46. Duan X, Jiang Y, Wang B, et al. Household fuel use for cooking and heating in China: results from the first Chinese Environmental Exposure- Related Human Activity Patterns Survey (CEERHAPS). Appl Energy. 2014;136:692–703. https:// doi. org/ 10. 1016/j. apene rgy. 2014. 09. 066. 47. Liao W, Liu X, Kang N, et al. Effect modification of kitchen ventilation on the associations of solid fuel use and long-duration cooking with the increased prevalence of depressive and anxiety symptoms: the Henan Rural Cohort Study. Indoor Air. 2022;32(3):e13016. https:// doi. org/ 10. 1111/ ina. 13016. • fast, convenient online submission • thorough peer review by experienced researchers in your field• rapid publication on acceptance• support for research data, including large and complex data types• gold Open Access which fosters wider collaboration and increased citations maximum visibility for your research: over 100M website views per year • At BMC, research is always in progress.Learn more biomedcentral.com/submissionsReady to submit your researchReady to submit your research ? Choose BMC and benefit from: ? Choose BMC and benefit from:
null
10.1371_journal.pone.0257370.pdf
Data Availability Statement: This study was registered in the Open Science Framework Registry (https://osf.io/rqve6). The review protocol can be accessed at https://bookdown.org/MathiasHarrer/ Doing_Meta_Analysis_in_R/. Data are available from the Dryad Data Repository (https://datadryad. org/stash/dataset/doi:10.6078/D10T42).
This study was registered in the Open Science Framework Registry ( https://osf.io/rqve6 ). The review protocol can be accessed at https://bookdown.org/MathiasHarrer/ Doing_Meta_Analysis_in_R/ . Data are available from the Dryad Data Repository ( https://datadryad. org/stash/dataset/doi:10.6078/D10T42 ).
RESEARCH ARTICLE Short-term elevations in glucocorticoids do not alter telomere lengths: A systematic review and meta-analysis of non-primate vertebrate studies Lauren ZaneID 1*, David C. EnsmingerID 1 1,2, Jose´ Pablo Va´ zquez-MedinaID a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Zane L, Ensminger DC, Va´zquez-Medina JP (2021) Short-term elevations in glucocorticoids do not alter telomere lengths: A systematic review and meta-analysis of non-primate vertebrate studies. PLoS ONE 16(10): e0257370. https://doi. org/10.1371/journal.pone.0257370 Editor: Gabriele Saretzki, University of Newcastle, UNITED KINGDOM Received: April 28, 2021 Accepted: August 29, 2021 Published: October 1, 2021 Copyright: © 2021 Zane et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: This study was registered in the Open Science Framework Registry (https://osf.io/rqve6). The review protocol can be accessed at https://bookdown.org/MathiasHarrer/ Doing_Meta_Analysis_in_R/. Data are available from the Dryad Data Repository (https://datadryad. org/stash/dataset/doi:10.6078/D10T42). Funding: The authors received no specific funding for this work. Competing interests: The authors have declared that no competing interests exist. 1 Department of Integrative Biology, University of California, Berkeley, CA, United States of America, 2 Department of Biological Sciences, San Jose State University, San Jose, CA, United States of America * laurenzane@berkeley.edu Abstract Background The neuroendocrine stress response allows vertebrates to cope with stressors via the acti- vation of the Hypothalamic-Pituitary-Adrenal (HPA) axis, which ultimately results in the secretion of glucocorticoids (GCs). Glucocorticoids have pleiotropic effects on behavior and physiology, and might influence telomere length dynamics. During a stress event, GCs mobilize energy towards survival mechanisms rather than to telomere maintenance. Addi- tionally, reactive oxygen species produced in response to increased GC levels can damage telomeres, also leading to telomere shortening. In our systematic review and meta-analysis, we tested whether GC levels impact telomere length and if this relationship differs among time frame, life history stage, or stressor type. We hypothesized that elevated GC levels are linked to a decrease in telomere length. Methods We conducted a literature search for studies investigating the relationship between telomere length and GCs in non-human vertebrates using four search engines: Web of Science, Goo- gle Scholar, Pubmed and Scopus, last searched on September 27th, 2020. This review identified 31 studies examining the relationship between GCs and telomere length. We pooled the data using Fisher’s Z for 15 of these studies. All quantitative studies underwent a risk of bias assessment. This systematic review study was registered in the Open Science Framework Registry (https://osf.io/rqve6). Results The pooled effect size from fifteen studies and 1066 study organisms shows no relationship between GCs and telomere length (Fisher’s Z = 0.1042, 95% CI = 0.0235; 0.1836). Our meta-analysis synthesizes results from 15 different taxa from the mammalian, avian, amphibian groups. While these results support some previous findings, other studies have found a direct relationship between GCs and telomere dynamics, suggesting underlying PLOS ONE | https://doi.org/10.1371/journal.pone.0257370 October 1, 2021 1 / 17 PLOS ONE Elevated glucocorticoids and telomere attrition mechanisms or concepts that were not taken into account in our analysis. The risk of bias assessment revealed an overall low risk of bias with occasional instances of bias from miss- ing outcome data or bias in the reported result. Conclusion We highlight the need for more targeted experiments to understand how conditions, such as experimental timeframes, stressor(s), and stressor magnitudes can drive a relationship between the neuroendocrine stress response and telomere length. Introduction The vertebrate neuroendocrine stress response integrates external stimuli into a broad range of physiological adjustments through the activation of the Hypothalamic-Pituitary-Adrenal axis (HPA axis) and the concomitant secretion of glucocorticoids (GCs) [1, 2]. While the pri- mary GC produced varies by taxa (e.g., cortisol in humans and corticosterone in birds and other mammals [3]), the impacts of GCs on organismal physiology are remarkably similar. Across species, an increase in GC secretion can typically be detected in 3–5 minutes following interaction with a stressor [4]. Additionally, GCs are relatively easy to quantify because they are present in all vertebrates and can be measured noninvasively in multiple matrices includ- ing hair and feces using a variety of assays [5, 6]. Therefore, wildlife stress physiology studies often rely on GC measurements as an indicator of the neuroendocrine stress response [7]. Fol- lowing their secretion, GCs induce a myriad of acute behavioral and physiological effects to prioritize immediate survival [8, 9]. In addition to allowing animals to cope with immediate stressors, GCs can influence other cellular processes such as telomere length dynamics. Telomeres are evolutionarily conserved caps that protect chromosomes against the loss of coding nucleotides during cell replication and against chromosomal fusion [10]. Telomere shortening is associated with aging, the neu- roendocrine stress response, and survival, and is thus of interest to several fields of biology [1, 11]. In humans, increased telomere loss predicts the onset of age-related diseases, cardiovascu- lar complications, cellular senescence, and other aging phenotypes [12, 13]. Telomere attrition can be attributed to several causes including the “end replication problem” in which the termi- nal end of linear DNA cannot be completely replicated by the lagging strand [14]. Since the end replication problem occurs at every cell division, telomeres continuously shorten with age progression [15]. Other stressors such as inflammatory challenges erode telomeres regardless of age [16]. In non-human vertebrates including birds, mammals, fish, amphibians and reptiles, expo- sure to challenging environmental conditions correlates with shorter telomeres [17, 18]. Reproductive stressors such as an artificially increased brood size can also shorten telomeres in zebra finch parents compared to controls and parents with a reduced brood size [19]. Early telomere length is positively correlated with survival and lifetime breeding success in both wild purple-crowned fairy wrens and zebra finches. Thus, individuals with longer telomeres are more likely to survive and produce more offspring that survive to maturity [20, 21]. Therefore, telomere dynamics—the change in telomere length attributed to processes of elongation and shortening—is related to organismal fitness [22]. In addition to impacting telomere length, stressors that lead to energy limitation such as psychological stress, disease, accelerated growth, nutrient shortage and work load activate the HPA axis causing the release of GCs [23]. PLOS ONE | https://doi.org/10.1371/journal.pone.0257370 October 1, 2021 2 / 17 PLOS ONE Elevated glucocorticoids and telomere attrition Thus, several hypothesized connections between GCs and telomere length exist. Firstly, GCs are an essential part of the vertebrate stress response, and their primary function is to mobilize energy [5]. Accordingly, the “metabolic telomere attrition hypothesis” proposes that during events that require an increased amount of energy and metabolic rates, telomeres are shortened as collateral [20]. As a result of the high energy expenditure, the energetically expen- sive maintenance of telomeres cannot take place as an emergency survival mechanism due to a shift in energy allocation [23]. In addition, GCs stimulate the generation of reactive oxygen species (ROS) and subsequent oxidative damage to telomeres, which are particularly suscepti- ble to oxidation due to a high guanine content [11, 24–26]. Finally, cortisol reduces telomerase —the enzyme responsible for telomere maintenance—activity in human T lymphocytes [27]. This reduction in telomerase activity can result in excessive telomere attrition [28]. Since wild- life face an array of stressors throughout their lifetime and these stressors can erode telomeres, GCs may play a mechanistic role in telomere loss [1]. External stressors cause pleiotropic effects that can potentially influence telomere dynamics, however the evidence for a causal relationship between GCs and telomere length is sparse. Two recent literature reviews on the topic by Angelier et.al 2018 and Casagrande and Hau 2019 [11, 23] summarize the potential relationship between GCs and telomere length. How- ever, it is essential to build a quantitative understanding of the relationships between the neu- roendocrine stress response and its downstream effects. In this study, we review the existing literature for empirical evidence of the relationship between GC secretion and telomere length to better understand the underlying mechanism of telomere shortening as well as potential consequences of the neuroendocrine stress response in non-primate vertebrates. Using a meta-analytical framework, we tested whether GC levels impact telomere length and if this relationship can differ among time frame, life history stage, or stressor type. We hypothesized that elevated GC levels are linked to a decrease in telomere length. Methods Literature search and study selection We conducted a literature search for studies investigating the relationship between telomere length and GC levels in non-human vertebrates using four search engines: Web of Science, Google Scholar, Pubmed and Scopus. Five subsets of the following keywords ‘reactive oxygen species,’ ‘antioxidant,’ ‘glucocorticoid,’ ‘cortisol,’ ‘corticosterone,’ ‘telomere length,’ ‘chronic stress,’ ‘oxidative stress,’ ‘acute stress,’ ‘chronic stress,’ ‘telomeres,’ and ‘HPA axis,’ were con- ducted in each search engine. We did not specify a time frame in our literature search. Addi- tional records were obtained from the reference section of studies included in the meta- analysis. Our study includes a qualitative synthesis of 31 full-text, peer-reviewed studies, and we report effect sizes for 15 of these studies. Studies were excluded if (1) GCs were administered, but physiological measurements such as feather or plasma GC levels were not taken. Such studies were excluded because it would not be possible to calculate the appropriate effect size (Fisher’s Z) for correlation data. For homogeneity in effect size calculation and statistical analysis, we did not include studies in which (2) GCs and telomere length were not specifically measured at two different time points (before or after treatment) (3) raw data was not accessible to use for the effect size calculation, or (4) telomere length measurements or GC measurements were log transformed. Statistical data analyses Meta-analysis. We conducted statistical analyses exclusively on studies with raw data available. When data was not publicly accessible, we contacted authors via email for consensual PLOS ONE | https://doi.org/10.1371/journal.pone.0257370 October 1, 2021 3 / 17 PLOS ONE Elevated glucocorticoids and telomere attrition access. For each study, the correlation coefficient (R2) was calculated by fitting a linear mixed model using the “lme4” R package (version 3.6.1, R Development Core Team, Boston, MA). When possible, random effects such as multiple blood draws from a single individual were incorporated in the linear mixed model (LMER) to account for variability not captured by explanatory parameters. For studies where a random effect could not be determined, a linear model (LM) was fitted. From the LMs and LMERs, R2 values were obtained from the model and converted into Fisher’s Z, then adjusted for sample size and combined into a pooled effect size (Fisher’s Z; Z) using the R package “meta”. The random-effects model meta-analysis was implemented in our study as this model accounts for the assumption that studies come from different populations, rather than the same population. These pooled effect sizes were then visualized in a forest plot. The “meta” package was also used to assess the statistical difference between observed and fixed effect model estimate of effect size (Cochrane’s Q) and the percent of variability in effect sizes that is not caused by sampling error (I2). After estimating heterogeneity, we identified potential outliers. Studies were classified as outliers if the study had an effect size with a confi- dence interval that did not intersect with the confidence interval of the pooled effect size. Since some studies can have a larger influence on the pooled effect size than others due to its sample size or individual effect size, we conducted an influence analysis. The analysis was conducted by omitting each study one at a time and simulating the pooled effect size, with a confidence interval had the study not been included. This influence analysis was represented in a Baujat plot, which shows the contribution of each study to heterogeneity as Cochrane’s Q, and compares this to the study’s influence on the pooled effect size. Subgroup analysis. Since experimental design can affect the outcome of a study, differ- ences in effect size may be attributed to these variables. As such, further sources of between- study heterogeneity were investigated through subgroup analysis and meta-regression. In the subgroup analysis, studies were grouped based on different categorical experimental parame- ters. We completed eight different subgroup analyses for the following parameters—duration of stressor, type of GC assay, telomere assay, species, taxa, study type, life history stage, and stressor type. For each subgroup analysis, a pooled effect size (Fisher’s Z) was calculated. We then compared pooled effect sizes and tested for between-study subgroup differences. The meta-regression was analogous to the subgroup analysis, except the parameter of investigation is continuous rather than categorical. We conducted one meta-regression for publication year and subsequently tested for between-study subgroup differences. For all analyses the signifi- cance threshold was set at p<0.05. In the subgroup analysis, studies included in the meta-analysis were clustered based on cat- egorical grouping and represented as a pooled effect size with a 95% confidence interval. The between study difference was indicated by Cochrane’s Q and the subsequent p-value for this statistical measure. The first subgroup analysis “stressor duration” organized studies based on the timeframe of the experiment—less than one week (n = 1), one to two weeks (n = 2), two to three weeks (n = 7), three to four weeks (n = 1), or longer than four weeks (n = 4)—. The sec- ond subgroup analysis, “type of stress” compared anthropogenic (n = 5) to naturally occurring stress (n = 7), or if stress was simulated by GC administration (n = 3). The subsequent sub- group analysis “life history stage, “differentiates studies based on pre-maturate study organ- isms (n = 12), or post-maturate study organisms (n = 3). Next, the subgroup “GC assay,” separates studies into those that quantified plasma GCs (n = 13) or non-plasma GCs (n = 2). Similarly, by performing the subgroup analysis for the variable “telomere assay” we hoped to parse out potential differences between the three methods of telomere quantification: qPCR (n = 7), TeloTAGGG (n = 1), and Telomerase Restriction Fragment (TRF; n = 7). The fifth subgroup analysis contrasts avian (n = 12) and non-avian (n = 3) studies. To explore the PLOS ONE | https://doi.org/10.1371/journal.pone.0257370 October 1, 2021 4 / 17 PLOS ONE Elevated glucocorticoids and telomere attrition relationship between individual species, we performed an additional subgroup analysis for each species included in the study. Finally, the subgroup analysis “study type” distinguished studies based on study design: cross-sectional (n = 5), repeated measures (n = 2), or within individual (n = 8) design. Publication bias. Published studies may not accurately represent the total studies investi- gating an area of research due to selective outcome reporting, missing studies and a higher likelihood of publication of studies reporting a significant (p<0.05) result. While proving selective outcome reporting and other forms of publication biases is challenging, missing stud- ies can be visually represented using a funnel plot. Commonly, studies with small effect sizes and small sample studies are likely to be missing, which can be depicted with funnel asymme- try or holes in the funnel plot. We created a funnel plot by graphing effect size against study precision, defined as the standard error of the effect size to visualize potential publication bias. We also report an Egger’s test, which is represented by the intercept, it’s confidence interval, and the associated p-value to determine if publication bias was statistically significant. Risk of bias in included studies. We assessed studies for missing outcome-level data, measurement of the outcomes, and outcome reporting in each included study. For the missing outcome-level data domain, we considered studies that could not report values for telomere length or GCs in less than 10% of total study organisms as low risk. We designated studies that did not report these values for 10–50% of study organisms as moderate risk and studies that did not report values for over 50% of GCs or telomere length, as high risk. Secondly, we based risk of bias in the measurement of outcome on the type of GC and telomere measurement. Low risk studies utilized plasma GCs or salivary GCs because these quantifications capture ele- vations related to a short-term stress event within minutes. Studies that measured GCs in fecal matter received a ranking of some concern because fecal GCs typically encapsulate cumulative stress over the day rather than GCs related to a particular environmental stressor. Fecal GCs also received a ranking of some concern due to potential variations related to storage and col- lection times, which can affect the concentration of fecal GC metabolites in a sample [29]. We considered studies that measured GCs in feathers as high risk because feathers incorporate GCs in over a month. Additionally, we considered feather GC quantification as high risk because feather preparation and GC extraction can vary greatly [30]. Finally, for the risk of bias due to outcome reporting we denoted studies that based results off a subset of time points or measurements high risk. We denoted studies that report results based on all time points with low risk. We took these three domains into consideration when assessing overall risk of bias. Results Literature search and study selection We electronically screened 789 records for relevance from the following databases: Google Scholar (n = 512), Web of Science (n = 105), PubMed (n = 72), and Scopus (n = 100). 2113 additional records were hand screened from the reference section of the 31 studies used in qualitative analysis. Of the total 2902 records that were screened for relevance, 78 were removed as duplicates and 2,489 did not fit criteria for our study. For example, some excluded studies include human trials, cell culture work, or studies that only assessed research questions pertaining to either telomere length or GC levels, but not both (Fig 1; S1 Table). Of the 183 assessed full-text articles, we removed 152 studies that did not fulfill our inclusion criteria. We statistically analyzed 15 of the remaining 31 studies, the ones that provided raw data for analy- sis either within the manuscript or after contacting the corresponding author [16, 22, 29–42]. The other 16 studies appeared to fit criteria but did not provide raw data for analysis [26, 30, PLOS ONE | https://doi.org/10.1371/journal.pone.0257370 October 1, 2021 5 / 17 PLOS ONE Elevated glucocorticoids and telomere attrition Fig 1. PRISMA diagram. PRISMA diagram showing the selection process for references included in the meta-analysis of the effects of GCs on telomere length. https://doi.org/10.1371/journal.pone.0257370.g001 43–55]. The literature and study selection process is illustrated using a PRISMA diagram (Fig 1). Meta-analysis The random-effects model meta-analysis is represented as a pooled effect size (Fisher’s Z) with 95% confidence intervals (Fig 2). No studies were removed as outliers. The model found no relationship between GC levels and telomere length (Fisher’s Z = 0.1042, CI = 0.0235; 0.1836). Both heterogeneity measures, Cochrane’s Q (Q = 11.31, p = 0.6615) and I2 with 95% confi- dence intervals (I2 = 0.0%; CI = 0.0%; 42.6%) yielded similar results. The influence analysis indicated that theoretically removing one study at a time did not yield pooled effect sizes (Fisher’s Z = 0.09–0.11) that differed from the original pooled effect size (Fisher’s Z = 0.11, S1 Fig). Additionally, the influence analysis demonstrated that certain studies unevenly impacted the pooled effect size and/or overall heterogeneity (S2 Fig), but no studies were removed as outliers. Subgroup analysis The subgroup analysis for “stressor duration” found no differences between any of the tested time frames (Table 1). The difference between-studies was not statistically significant Fig 2. Forest plot. Distribution of effect sizes of GCs on telomere length and 95% CI of effect size. Dashed lines represent pooled effect sizes using a random and fixed effect model. Heterogeneity (I2), the percent of variability in effect sizes that is not caused by sampling error indicates very little variability in effect size. Weight indicates the influence the study has on the overall pooled effect. https://doi.org/10.1371/journal.pone.0257370.g002 PLOS ONE | https://doi.org/10.1371/journal.pone.0257370 October 1, 2021 6 / 17 PLOS ONE Table 1. Pooled effect sizes with 95% CI of experimental parameters investigated during the five subgroup analyses for stressor duration, stressor type, life history stage, GC assay and taxa group. Experimental Parameter Number of Studies Effect Size (Fisher’s Z) 95% CI Elevated glucocorticoids and telomere attrition Stressor Duration Type of Stress Life History Stage GC Assay Taxa Group Species Capreolus capreolus Coturnix japonica Fregata magnificens Hydrobates pelagicus Parus major Phalacrocorax aristotelis Rana temporaria Rissa tridactyla Sterna hirundo Sturnus unicolor Tachycineta bicolor Turdus merula Welsh pony Telomere Assay Study Type < 1 week n = 1 1–2 weeks n = 2 2–3 weeks n = 7 3–4 weeks n = 1 > 4 weeks n = 4 Anthropogenic n = 5 Naturally occurring n = 7 GC administration n = 3 Pre-maturation n = 12 Post-maturation n = 3 Plasma GCs n = 12 Non-Plasma GCs n = 3 Avian n = 12 Non-Avian n = 3 n = 1 n = 1 n = 1 n = 1 n = 2 n = 1 n = 1 n = 1 n = 1 n = 1 n = 2 n = 1 n = 1 n = 7 qPCR TeloTAGGG n = 1 TRF n = 7 n = 5 Cross sectional Repeated measure n = 2 Within individual n = 8 0.1902 0.1425 0.1111 0.0843 0.1012 0.1161 0.1012 0.1183 0.0135 0.0959 0.0741 0.0957 0.2181 0.0451 0.0993 0.0596 0.1306 0.4651 0.0707 0.1852 0.0067 0.0826 0.0957 0.0088 0.1134 0.0539 0.2693 0.1186 0.1306 0.0909 0.1687 0.0271 0.0984 -0.2717; 0.2965 -0.1663; 0.3454 -0.0157; 0.1628 -0.2950; 0.4591 0.0101; 0.4080 0.0059; 0.3621 -0.0388; 0.1284 -0.0320; 0.3085 0.0185; 0.2019 -0.1183; 0.2800 0.0061; 0.1945 -0.0437; 0.2701 0.0088; 0.1919 -0.0600; 0.0936 [-0.2072; 0.3881] [-0.1973; 0.3089] [-0.1232; 0.3684] [0.0857; 0.7267] [-0.1322; 0.2678] [-0.0600; 0.2893] [-0.1962; 0.2090] [-0.1686; 0.3238] [-0.2950; 0.4591] [-0.1182; 0.1356] [-0.2110; 0.4153] [-0.3004; 0.3952] [0.0252; 0.4831] [-0.0089; 0.2424] [-0.1232; 0.3684] [-0.0409; 0.2197] [-0.0219; 0.3474] [-0.1336; 0.1864] [0.0040; 0.1910] The meta-regression was performed for the continuous variable publication year and represented as Cochrane’s Q and the associated p = value. Publication dates ranged from 2014–2021. Publication date was not a significant predictor of effect size (Q = 1.252, p = 0.2632). https://doi.org/10.1371/journal.pone.0257370.t001 PLOS ONE | https://doi.org/10.1371/journal.pone.0257370 October 1, 2021 7 / 17 PLOS ONE Elevated glucocorticoids and telomere attrition (Q = 1.86, p = 0.7594). Similarly, the subgroup analysis for “stressor type,” did not reveal a dif- ference between types of stressors (Table 1). The between study difference was not significantly different (Q = 2.56, p = 0.2783). Likewise, our subgroup “life history stage,” did not show dif- ferences between effect sizes for pre- and post-maturation organisms (Table 1), and did not indicate a difference between groups (Q = 0.06, p = 0.8119). The fourth subgroup analysis, “GC assay” did not find a difference between plasma GCs and other GC measurements, yield- ing a non-significant difference between studies (Q = 0.03, p = 0.8742) (Table 1). Additionally, the between study difference for the telomere assay subgroup did not find a significant differ- ence between the three telomere quantification methods (Q = 0.12, p = 0.9401; Table 1). Our sixth subgroup analysis examined potential differences in effect size due to taxa, which could be divided into the binary categories avian and non-avian (Table 1). There was no difference between-studies (Q = 0.03, p = 0.8666). Our analysis further explored species-specific differ- ences and accordingly did not find a significant difference between species (Q = 9.27, p = 0.6797). Similarly, the final analysis investigated potential differences between study designs and yielded a non-significant difference between cross-sectional, repeated measures, or within individual designs (Q = 1.27, p = 0.5289). Publication bias We found publication bias against studies with small sample size and small effect size (S3 Fig; Egger’s test for small sample bias: intercept = 1.420616, CI = 0.3753223; 2.465909, p = 0.02064949). Risk of bias in included studies We represent the results of the risk of bias analysis in Table 2. Four of fifteen studies received a risk of bias ranking of moderate concern. These studies had some missing values for GCs or telomere length or selectively reported one time point in the results. The other eleven studies received a ranking of low risk and accordingly reported nearly all values for physiological parameters, measured GCs in plasma or saliva, and did not selectively report results. Table 2. Overall risk of bias assessed based on missing outcome data, measure of outcome and in the selection of reported results. Author Year Bias due to missing outcome data Bias in measure of outcome Bias in the selection of reported result Overall Risk of Bias Bauch et. al Burraco et. al Casagrande et. al Gil et. al Grunst et. al Hau et. al Herborn et. al Injaian et. al Lansade et. al Lemaitre et. al Pegan et. al Sebastiano et. al Stier et. al Watson et. al Young et. al 2016 2019 2020 2019 2020 2015 2014 2019 2018 2021 2019 2017 2020 2016 2017 high low high low low low low high low low high low low low moderate https://doi.org/10.1371/journal.pone.0257370.t002 low low low low high low low low low moderate low low low low low high low low low low low low high low low low low low low high some concern low risk some concern low risk low risk low risk low risk high risk low risk low risk some concern low risk low risk low risk some concern PLOS ONE | https://doi.org/10.1371/journal.pone.0257370 October 1, 2021 8 / 17 PLOS ONE Elevated glucocorticoids and telomere attrition Discussion External and internal stimuli can activate the neuroendocrine stress response in vertebrates, resulting in the secretion of GCs, which induces multiple downstream physiological and behavioral effects [8, 9]. GCs might directly or indirectly cause telomere erosion [1, 11, 32]. Therefore, our goal was to investigate the relationship between GCs and telomere length via meta-analysis using data from empirical studies. Though our sample size was limited (n = 15), our data do not support the hypothesis that elevated GC levels result in telomere shortening. The empirical evidence for a relationship between GCs and telomere length is mixed, with some studies showing that telomere shortening is directly related to GC levels, and other stud- ies finding no relationship. For example, GCs influence telomere dynamics in wild roe deer and great tits [32, 39], but not in red squirrels or magellanic penguins [46, 47]. These results suggest that the relationship between GCs and telomere length is species-specific. Alterna- tively, a potential relationship may be obscured by the methods used to measure GCs and telo- mere length or by differences in experimental design including time frame. A differential sensitivity of the HPA axis can also obscure conclusions made from GC measurements espe- cially in free-ranging vertebrates that can potentially encounter a variety of external stimuli [1]. For example, since GC levels in plasma remain elevated for several minutes after a stressor subsides, it can be challenging to assess whether a measured GC increase results from the stressor in question, the stress involved in obtaining a sample from the experimental subject, or an unrelated event triggering HPA axis activation [6, 56]. As baseline plasma GC samples must be collected quickly in many species, it can be logistically difficult to attain a true baseline GC value in the field [57–60]. GCs can also be incorporated into other matrixes such as saliva, feathers, and hair [4, 58]. The multitude of non-invasive GC sampling sources is advantageous to conservation physiology as their quantification does not require capture [6]. However, across tissues and fluids, the time required for GC incorporation varies. For example, eleva- tions in plasma GCs can be detected within minutes of stressor exposure, whereas GCs inte- grate into hair a week or more after stressor exposure [4]. Hence, there are caveats in the interpretation of each measurement such as incongruencies between GC levels in plasma and other tissues, hair and saliva [60]. Therefore, GC measurements in feces may be more repre- sentative of accumulated stress, rather than the event in question [6]. GC quantification in tissues and feces can also present specific uncertainty and imprecision during sampling, storage, and extraction. In fecal samples, GC metabolites can increase up to 92% in 120 days and provide an inaccurate assessment of GC levels [61, 62]. Excrement not collected immediately or across different time scales can obscure potential differences since exposure to abiotic factors like rainfall or extreme temperature can alter the concentration of fecal glucocorticoid metabolites [63]. Moreover, diet can affect GC metabolites in fecal sam- ples, since an increased amount of cellulose depresses fecal glucocorticoid metabolite concen- trations [61]. Similarly, feather preparation and extraction can also affect GC levels [64]. Further, different parts of the feather yield different concentrations of GCs. Saliva based GC extraction and quantification hosts similar shortcomings, though salivary GCs increase on a similar timeline (5–10 minutes) to circulating plasma GCs and thus prove a close proxy for plasma GC quantification [65]. Other factors such as time since last meal and recent activity also impact salivary GC measurement [66]. Similar considerations must be taken into account when assessing telomere length. Since telomere length can be influenced by environmental, maternal, and epigenetic effects, there is a large inter-individual variability in telomere dynamics [11, 67]. Several factors may contrib- ute to this variability including discrepancies between the repeatability of different telomere measurement assays. Seven studies included in our meta-analysis utilized the telomere PLOS ONE | https://doi.org/10.1371/journal.pone.0257370 October 1, 2021 9 / 17 PLOS ONE Elevated glucocorticoids and telomere attrition restriction fragment (TRF) assay, which depends on the distribution of the terminal restriction fragments to average the length of telomeres in a given cell population [68]. The other eight studies used the quantitative polymerase chain reaction (qPCR, n = 7), which relies on the quantification of the highly conserved (TTAGGG)n sequence for a Southern blot variation (TeloTAGGG for telomere quantification (n = 1) [69]. TRF-based studies are highly repeatable within individuals, whereas qPCR based studies are less repeatable and more variable than TRF because they are more prone to measurement errors [70]. qPCR can also bias measure- ments of telomere length because some species that exhibit interstitial telomeric repeats will artificially enlarge telomere length [71, 72]. In addition to methodological differences, there is large individual variability in telomere length based on tissue type [73]. In adult zebra finches, telomere length in red blood cells is correlated with telomere length in the spleen, liver and brain, but not muscle or heart [31]. While avian studies in our meta-analysis used red blood cells for telomere measurement, telomere length was measured in tail muscle and liver in mammals and amphibians, which could lead to discrepancies when comparing among studies [31, 46, 57]. A variety of biological factors also contribute to the diversity of telomere dynamics observed within a study and the large amount of observed inter-individual variability. The rate of telo- mere shortening can be influenced by the life histories and environmental conditions [22]. In accordance with the metabolic telomere attrition hypothesis, shortening is exacerbated by life history stages requiring more energy, such as reproduction [32]. Within an energy intensive process like reproduction, there can be a large inter-individual variability related to reproduc- tive effort, which can be attributed to brood size and food availability [74]. Differences in reproductive roles during the breeding season account for sex-specific telomere dynamics which can contribute to differences in the variability of telomere dynamics within a study [75]. Finally, individuals respond differently to environmental challenges which can act synergisti- cally with rapid growth or energy intensive life stages to magnify the rate of telomere shorten- ing in non-model vertebrates [71]. Telomere dynamics can be complicated by the presence of telomerase which in some cases can elongate telomeres [22, 76]. Typically, telomerase exhibits higher activity in developing organisms as compared to adults [77]. Ectotherms such as amphibians and reptiles have telo- merase that is active throughout adulthood while endotherms reduce telomerase expression almost to non-detectable levels as they reach maturity [11, 70]. However, there is conflicting evidence on these observations, as telomerase activity has been detected in adult common terns and European Storm Petrels among other species [78, 79]. Nonetheless, adult telomere shortening is observed in chickens, which have active telomerase in the adult life stage [26]. While there is an absence of empirical evidence on the long-term activity of telomerase in many avian species, even adults exhibit general shortening trends [76]. Many factors influence GC and telomere measurements. During the subgroup analysis, we attempted to disentangle the underlying causes of the variation in effect size. Ultimately, we found no impact of stressor, taxa, type of GC assay, or life history stage on the heterogeneity of the effect size. While no subgroup was identified as a predictor of heterogeneity in effect size, pooled effect sizes in certain categories with the subgroup indicate a higher pooled effect size than the overall pooled effect size. The small sample size for some parameters precluded fur- ther statistical analysis, however, we found variables of interest that may play a large role in the relationship between GCs and telomere length. For example, within “experimental time- frame,” (n = 4) the group of studies with a timeframe above four weeks had a pooled effect size of 0.2181, while all other groups’ pooled effect size was less than that of the overall pooled effect size. Since most studies took place in less than four weeks, this suggests that while almost immediate changes in GCs can be observed, the impact of GCs on telomere length cannot be PLOS ONE | https://doi.org/10.1371/journal.pone.0257370 October 1, 2021 10 / 17 PLOS ONE Elevated glucocorticoids and telomere attrition seen on short time scales. This idea is consistent with typical responses of telomere shortening observed in studies that take place for more than a year [29, 54, 79–81]. More work is needed to explore if long-term rather than short-term studies can be used to tease apart parameters that underlie the connection between GCs and telomere length such as stressor type or duration. While GC secretion is often viewed as the endpoint of HPA axis activation in response to external stimuli, GC manipulation is an oversimplification of the stress response which involves a multitude of physiological mechanisms that can each impact energy allocation and promote telomere erosion [8]. This highlights the problematic nature of the category “GC stress” which was investigated as a category during the subgroup analysis, in which studies subjected organisms to GC manipulation via an implant or oral administration. Since previous research found that organismal stress can result in adverse physiological responses without the involvement of the HPA axis, these results underscore the issue of using only GCs as a proxy for stress [82, 83]. Overall, we found no relationship between GCs and telomere length across studies. Cur- rently, the existing literature shows both a direct relationship and a lack of a relationship between GCs and telomere dynamics, suggesting that the underlying mechanisms driving this relationship are species-specific or altered by differences in experimental design. However, due to limited sample size, we are unable to investigate the underlying variables that play a role in this relationship. Here, we highlight the need for more studies with targeted experimental parameters to understand how conditions, such as experimental timeframes, stressor(s), and stressor magnitudes can drive a potential relationship between the neuroendocrine stress response and cellular aging. Thus, we recommend the following research priorities to groups studying similar questions. 1. Experimental timeframes and stressor magnitudes should be long enough to observe telo- mere erosion in relation to stressors when studying GCs. 2. When possible, studies should use a repeated measures design to measure cortisol levels and telomere lengths before and after stress exposure to account for individual variation. 3. While the avian taxa are well represented in this research topic, there is a dearth of informa- tion on other taxa. It will be important to investigate the neuroendocrine stress response in other vertebrates including mammals and reptiles to understand if similar principles hold true in these taxa or if telomere dynamics differ across taxa. 4. If possible, future research should assess the functionality of the study organisms’ HPA axis by ACTH/dexamethasone challenge prior to exposure to a stressor and completion of the study. Certainty of evidence We utilized the applicable Cochrane/GRADE categories “risk of bias,” “inconsistency,” and “publication bias,” for the determination of the certainty of evidence. Overall, we have a mod- erate confidence in the certainty of evidence. While most studies received a low risk of bias assessment, and had low heterogeneity, we report a considerable amount of publication bias as evidenced by Egger’s test and an asymmetrical funnel plot. Supporting information S1 Checklist. PRISMA 2020 checklist. (PDF) PLOS ONE | https://doi.org/10.1371/journal.pone.0257370 October 1, 2021 11 / 17 PLOS ONE Elevated glucocorticoids and telomere attrition S1 Fig. Influence analysis plot. The leave one out recalculation reveals a similar effect size across studies and indicates that studies evenly contribute to the pooled effect size. (TIF) S2 Fig. Baujat plot. Studies can have an unequal influence on the pooled effect size and con- tribute to the heterogeneity of effect sizes. The horizontal axis represents Cochrane’s Q and influence on the pooled effect size on the vertical axis. (TIF) S3 Fig. Funnel plot. The lack of studies in the bottom left of the “funnel” demonstrates publi- cation bias against studies with small sample sizes and small effect sizes. (TIF) S1 Table. Search strategy table. Details search term combinations used to search online data- bases and websites. (XLSX) Acknowledgments We would like to thank all the researchers who made their data freely available for this study. In particular, we thank Christine Bauch (University of Groningen), Stefania Casagrande (Max Planck Institute for Ornithology), Britt Heidinger (North Dakota State University), Marie- Pierre Moisan (French National Institute for Agriculture, Food, and Environment), Teresa Pegan (University of Michigan), Manrico Sebastiano (French National Centre for Scientific Research), Mathilde Tissier (Bishop’s University), and Hannah Watson (Lund University). Author Contributions Conceptualization: Lauren Zane, David C. Ensminger, Jose´ Pablo Va´zquez-Medina. Data curation: Lauren Zane. Formal analysis: Lauren Zane. Funding acquisition: Jose´ Pablo Va´zquez-Medina. Investigation: Lauren Zane. Methodology: Lauren Zane, David C. Ensminger. Project administration: David C. Ensminger. Resources: David C. Ensminger. Software: Lauren Zane. Supervision: David C. Ensminger, Jose´ Pablo Va´zquez-Medina. Validation: David C. Ensminger, Jose´ Pablo Va´zquez-Medina. Visualization: Lauren Zane. Writing – original draft: Lauren Zane. Writing – review & editing: David C. Ensminger, Jose´ Pablo Va´zquez-Medina. References 1. Haussmann MF, Marchetto NM. Telomeres: Linking stress and survival, ecology and evolution. Current Zoology. 2010; 56(6):714–27. PLOS ONE | https://doi.org/10.1371/journal.pone.0257370 October 1, 2021 12 / 17 PLOS ONE Elevated glucocorticoids and telomere attrition 2. Wingfield JC, Sapolsky RM. Reproduction and Resistance to Stress: When and How: Reproduction and resistance to stress. Journal of Neuroendocrinology. 2003; 1; 15(8):711–24. https://doi.org/10. 1046/j.1365-2826.2003.01033.x PMID: 12834431 3. Koren L, Whiteside D, Fahlman Å, Ruckstuhl K, Kutz S, Checkley S, et al. Cortisol and corticosterone independence in cortisol-dominant wildlife. General and Comparative Endocrinology. 2012 May; 177 (1):113–9. https://doi.org/10.1016/j.ygcen.2012.02.020 PMID: 22449618 4. Gormally BMG, Romero LM. What are you actually measuring? A review of techniques that integrate the stress response on distinct time-scales. Functional Ecology. 2020; 34(10):2030–44. 5. MacDougall-Shackleton SA, Bonier F, Romero LM, Moore IT. Glucocorticoids and “Stress” Are Not Synonymous. Integrative Organismal Biology. 2019; 1(1). 6. Sheriff MJ, Dantzer B, Delehanty B, Palme R, Boonstra R. Measuring stress in wildlife: techniques for quantifying glucocorticoids. Oecologia. 2011; 166(4):869–87. https://doi.org/10.1007/s00442-011- 1943-y PMID: 21344254 7. Arnold JM, Oswald SA, Voigt CC, Palme R, Braasch A, Bauch C, et al. Taking the stress out of blood collection: comparison of field blood-sampling techniques for analysis of baseline corticosterone. Jour- nal of Avian Biology. 2008; 39(5):588–92. 8. Angelier F, Wingfield JC. Importance of the glucocorticoid stress response in a changing world: Theory, hypotheses and perspectives. General and Comparative Endocrinology. 2013; 190:118–28. https://doi. org/10.1016/j.ygcen.2013.05.022 PMID: 23770214 9. Ricklefs RE, Wikelski M. The physiology/life-history nexus. Trends in Ecology & Evolution. 2002; 17 (10):462–8. 10. Young AJ. The role of telomeres in the mechanisms and evolution of life-history trade-offs and ageing. Philosophical Transactions of the Royal Society B. 2018; 373(1741):20160452. https://doi.org/10.1098/ rstb.2016.0452 PMID: 29335379 11. Angelier F, Costantini D, Ble´ vin P, Chastel O. Do glucocorticoids mediate the link between environmen- tal conditions and telomere dynamics in wild vertebrates? A review. General and Comparative Endocri- nology. 2018; 256:99–111. https://doi.org/10.1016/j.ygcen.2017.07.007 PMID: 28705731 12. Campisi J, Andersen JK, Kapahi P, Melov S. Cellular senescence: A link between cancer and age- related degenerative disease? Seminars in Cancer Biology. 2011; https://doi.org/10.1016/j.semcancer. 2011.09.001 PMID: 21925603 13. Donate LE, Blasco MA. Telomeres in cancer and ageing. Philosophical Transactions of the Royal Soci- ety B. 2011; 366(1561):76–84. https://doi.org/10.1098/rstb.2010.0291 PMID: 21115533 14. Levy MZ, Allsopp RC, Futcher AB, Greider CW, Harley CB. Telomere end-replication problem and cell aging. Journal of Molecular Biology. 1992; 225(4):951–60. https://doi.org/10.1016/0022-2836(92) 90096-3 PMID: 1613801 15. Monaghan P. Organismal stress, telomeres and life histories. Journal of Experimental Biology. 2014; 217(1):57–66. 16. Sebastiano M, Eens M, Angelier F, Pineau K, Chastel O, Costantini D. Corticosterone, inflammation, immune status and telomere length in frigatebird nestlings facing a severe herpesvirus infection. Con- servation Physiology. 2017; 5(1):cow073. https://doi.org/10.1093/conphys/cow073 PMID: 28070333 17. Chatelain M, Drobniak SM, Szulkin M. The association between stressors and telomeres in non-human vertebrates: a meta-analysis. Ecology Letters. 2020; 23(2):381–98. https://doi.org/10.1111/ele.13426 PMID: 31773847 18. Belmaker A, Hallinger KK, Glynn RA, Winkler DW, Haussmann MF. The environmental and genetic determinants of chick telomere length in Tree Swallows (Tachycineta bicolor). Ecology and Evolution. 2019; 9(14):8175–86. https://doi.org/10.1002/ece3.5386 PMID: 31380080 19. Reichert S, Stier A, Zahn S, Arrive´ M, Bize P, Massemin S, et al. Increased brood size leads to persis- tent eroded telomeres. Frontiers in Ecology and Evolution [Internet]. 2014; 2:9. 20. Eastwood JR, Hall ML, Teunissen N, Kingma SA, Hidalgo Aranzamendi N, Fan M, et al. Early-life telo- mere length predicts lifespan and lifetime reproductive success in a wild bird. Molecular Ecology. 2019; 28(5):1127–37. https://doi.org/10.1111/mec.15002 PMID: 30592345 21. Heidinger BJ, Blount JD, Boner W, Griffiths K, Metcalfe NB, Monaghan P. Telomere length in early life predicts lifespan. Proceedings of the National = Academy of Sciences. 2012; 109(5):1743–8. https:// doi.org/10.1073/pnas.1113306109 PMID: 22232671 22. Stier A, Metcalfe NB, Monaghan P. Pace and stability of embryonic development affect telomere dynamics: an experimental study in a precocial bird model. Proceedings of the Royal Society B. 2020; 287(1933):20201378. https://doi.org/10.1098/rspb.2020.1378 PMID: 32842933 23. Casagrande S, Hau M. Telomere attrition: metabolic regulation and signalling function? Biol Lett. 2019; 15(3):20180885. https://doi.org/10.1098/rsbl.2018.0885 PMID: 30890069 PLOS ONE | https://doi.org/10.1371/journal.pone.0257370 October 1, 2021 13 / 17 PLOS ONE Elevated glucocorticoids and telomere attrition 24. Reichert S, Stier A. Does oxidative stress shorten telomeres in vivo? A review. Biol Lett. 2017; 13 (12):20170463. https://doi.org/10.1098/rsbl.2017.0463 PMID: 29212750 25. Costantini D, Marasco V, Møller AP. A meta-analysis of glucocorticoids as modulators of oxidative stress in vertebrates. Journal of Comparative Physiology B. 2011; 181, 447–456. https://doi.org/10. 1007/s00360-011-0566-2 PMID: 21416253 26. Haussmann MF, Longenecker AS, Marchetto NM, Juliano SA, Bowden RM. Embryonic exposure to corticosterone modifies the juvenile stress response, oxidative stress and telomere length. Proceedings of the Royal Society B. 2012; 279(1732):1447–56. https://doi.org/10.1098/rspb.2011.1913 PMID: 22072607 27. Choi J, Fauce SR, Effros RB. Reduced telomerase activity in human T lymphocytes exposed to cortisol. Brain, Behavior, and Immunity. 2008; 22(4):600–5. https://doi.org/10.1016/j.bbi.2007.12.004 PMID: 18222063 28. Wei YB, Backlund L, Wegener G, Mathe AA, Lavebratt C. Telomerase Dysregulation in the Hippocam- pus of a Rat Model of Depression: Normalization by Lithium. International Journal of Neuropsychophar- macology. 2015; 18(7):pyv002–pyv002. https://doi.org/10.1093/ijnp/pyv002 PMID: 25618407 29. Bauch C, Riechert J, Verhulst S, Becker PH. Telomere length reflects reproductive effort indicated by corticosterone levels in a long-lived seabird. Molecular Ecology. 2016; 25(22):5785–94. https://doi.org/ 10.1111/mec.13874 PMID: 27696588 30. Burraco P, Valde´ s AE, Orizaola G. Metabolic costs of altered growth trajectories across life transitions in amphibians. Journal of Animal Ecology. 2020; 89(3):855–66. https://doi.org/10.1111/1365-2656. 13138 PMID: 31693168 31. Casagrande S, Stier A, Monaghan P, Loveland JL, Boner W, Lupi S, et al. Increased glucocorticoid con- centrations in early life cause mitochondrial inefficiency and short telomeres. Journal of Experimental Biology. 2020; 223(15):jeb222513. https://doi.org/10.1242/jeb.222513 PMID: 32532864 32. Gil D, Alfonso-Iñiguez S, Pe´ rez-Rodrı´guez L, Muriel J, Monclu´ s R. Harsh conditions during early devel- opment influence telomere length in an altricial passerine: Links with oxidative stress and corticoste- roids. Journal of Evolutionary Biology. 2019; 32(1):111–25. https://doi.org/10.1111/jeb.13396 PMID: 30387533 33. Grunst ML, Raap T, Grunst AS, Pinxten R, Parenteau C, Angelier F, et al. Early-life exposure to artificial light at night elevates physiological stress in free-living songbirds. Environmental Pollution. 2020; 259:113895. https://doi.org/10.1016/j.envpol.2019.113895 PMID: 31926393 34. Hau M, Haussmann MF, Greives TJ, Matlack C, Costantini D, Quetting M, et al. Repeated stressors in adulthood increase the rate of biological ageing. Frontiers in Zoology. 2015; 12(1):4. https://doi.org/10. 1186/s12983-015-0095-z PMID: 25705242 35. Herborn KA, Heidinger BJ, Boner W, Noguera JC, Adam A, Daunt F, et al. Stress exposure in early post-natal life reduces telomere length: an experimental demonstration in a long-lived seabird. Proceed- ings of the Royal Society B. 2014; 281(1782):20133151. https://doi.org/10.1098/rspb.2013.3151 PMID: 24648221 36. 37. 38. Injaian AS, Gonzalez-Gomez PL, Taff CC, Bird AK, Ziur AD, Patricelli GL, et al. Traffic noise exposure alters nestling physiology and telomere attrition through direct, but not maternal, effects in a free-living bird. General and Comparative Endocrinology. 2019; 276:14–21. https://doi.org/10.1016/j.ygcen.2019. 02.017 PMID: 30796896 Lansade L, Foury A, Reigner F, Vidament M, Guettier E, Bouvet G, et al. Progressive habituation to sep- aration alleviates the negative effects of weaning in the mother and foal. Psychoneuroendocrinology. 2018; 97:59–68. https://doi.org/10.1016/j.psyneuen.2018.07.005 PMID: 30005282 Lemaıˆtre J-F, Carbillet J, Rey B, Palme R, Froy H, Wilbourn RV, et al. Short-term telomere dynamics is associated with glucocorticoid levels in wild populations of roe deer. Comparative Biochemistry and Physiology Part A: Molecular & Integrative Physiology. 2021; 252:110836. https://doi.org/10.1016/j. cbpa.2020.110836 PMID: 33144154 39. Pegan TM, Winkler DW, Haussmann MF, Vitousek MN. Brief Increases in Corticosterone Affect Mor- phology, Stress Responses, and Telomere Length but Not Postfledging Movements in a Wild Songbird. Physiological and Biochemical Zoology. 2019; 92(3):274–85. https://doi.org/10.1086/702827 PMID: 30840539 40. Tissier ML, Williams TD, Criscuolo F. Maternal Effects Underlie Ageing Costs of Growth in the Zebra Finch (Taeniopygia guttata). White SA, editor. PLoS ONE. 2014; 9(5):e97705. https://doi.org/10.1371/ journal.pone.0097705 PMID: 24828412 41. Watson H, Bolton M, Heidinger BJ, Boner W, Monaghan P. Assessing the effects of repeated handling on the physiology and condition of semi-precocial nestlings. IBIS International Journal of Avian Science. 2016; 158(4):834–43. https://doi.org/10.1111/ibi.12402 PMID: 27708454 PLOS ONE | https://doi.org/10.1371/journal.pone.0257370 October 1, 2021 14 / 17 PLOS ONE Elevated glucocorticoids and telomere attrition 42. Young R, Orben R, Will A, Kitaysky A. Relationship between telomere dynamics and movement and behavior during winter in the thick-billed murre. Marine Ecological Progress Series. 2017; 578:253–61. 43. Cerchiara JA, Risques RA, Prunkard D, Smith JR, Kane OJ, Boersma PD. Magellanic penguin telo- meres do not shorten with age with increased reproductive effort, investment, and basal corticosterone. Ecological Evolution. 2017; 7(15):5682–91. https://doi.org/10.1002/ece3.3128 PMID: 28811878 44. Dantzer B, van Kesteren F, Westrick SE, Boutin S, McAdam AG, Lane JE, et al. Maternal glucocorti- coids promote offspring growth without inducing oxidative stress or shortening telomeres in wild red squirrels. Journal of Experimental Biology. 2020; 223(1):jeb212373. https://doi.org/10.1242/jeb.212373 PMID: 31796605 45. Drury SS, Howell BR, Jones C, Esteves K, Morin E, Schlesinger R, et al. Shaping long-term primate development: Telomere length trajectory as an indicator of early maternal maltreatment and predictor of future physiologic regulation. Developmental Psychopathology. 2017; 29(5):1539–51. https://doi.org/ 10.1017/S0954579417001225 PMID: 29162166 46. Gangoso L, Lambertucci SA, Cabezas S, Alarco´ n PAE, Wiemeyer GM, Sanchez-Zapata JA, et al. Sex- dependent spatial structure of telomere length in a wild long-lived scavenger. Ecosphere. 2016; 7(10): e01544. 47. Noguera JC, Velando A. Reduced telomere length in embryos exposed to predator cues. Journal of Experimental Biology. 2019; 222(24):jeb216176. https://doi.org/10.1242/jeb.216176 PMID: 31796604 48. Noreikiene K, O¨ st M, Seltmann MW, Boner W, Monaghan P, Jaatinen K. Nest cover and faecal gluco- corticoid metabolites are linked to hatching success and telomere length in breeding Common Eiders (Somateria mollissima). Canadian Journal of Zoology. 2017; 95(9):695–703. 49. Ouyang JQ, Lendvai A´ Z, Moore IT, Bonier F, Haussmann MF. Do Hormones, Telomere Lengths, and Oxidative Stress form an Integrated Phenotype? A Case Study in Free-Living Tree Swallows. Integra- tive and Comparative Biology. 2016; 56(2):138–45. https://doi.org/10.1093/icb/icw044 PMID: 27252220 50. Quirici V, Guerrero CJ, Krause JS, Wingfield JC, Va´ squez RA. The relationship of telomere length to baseline corticosterone levels in nestlings of an altricial passerine bird in natural populations. Frontiers in Zoology 2016; 13(1):1. 22. 51. Schultner J, Moe B, Chastel O, Bech C, Kitaysky AS. Migration and stress during reproduction govern telomere dynamics in a seabird. Biology Letters. 2014; 10(1):20130889. https://doi.org/10.1098/rsbl. 2013.0889 PMID: 24429681 52. Shen Q, Wu J, Ni Y, Xie X, Yu C, Xiao Q, et al. Exposure to jet lag aggravates depression-like behaviors and age-related phenotypes in rats subject to chronic corticosterone. Acta Biochimica et Biophysica Sinica. 2019; 51(8):834–44. https://doi.org/10.1093/abbs/gmz070 PMID: 31314053 53. Stevenson JR, McMahon EK, Boner W, Haussmann MF. Oxytocin administration prevents cellular aging caused by social isolation. Psychoneuroendocrinology. 2019; 103:52–60. https://doi.org/10.1016/ j.psyneuen.2019.01.006 PMID: 30640038 54. Xie X, Shen Q, Ma L, Chen Y, Zhao B, Fu Z. Chronic corticosterone-induced depression mediates pre- mature aging in rats. Journal of Affective Disorders; 229:254–61. https://doi.org/10.1016/j.jad.2017.12. 073 PMID: 29329057 55. Young RC, Welcker J, Barger CP, Hatch SA, Merkling T, Kitaiskaia EV, et al. Effects of developmental conditions on growth, stress and telomeres in black-legged kittiwake chicks. Molecular Ecology. 2017; 26(13):3572–84. https://doi.org/10.1111/mec.14121 PMID: 28370751 56. Romero LM, Meister CJ, Cyr NE, Kenagy GJ, Wingfield JC. Seasonal glucocorticoid responses to cap- ture in wild free-living mammals. American Journal of Physiology-Regulatory, Integrative and Compara- tive Physiology. 2008; 294(2):R614–22. https://doi.org/10.1152/ajpregu.00752.2007 PMID: 18094060 57. Tylan C, Camacho K, French S, Graham SP, Herr MW, Jones J, et al. Obtaining plasma to measure baseline corticosterone concentrations in reptiles: How quick is quick enough? General and Compara- tive Endocrinology. 2020; 287:113324. https://doi.org/10.1016/j.ygcen.2019.113324 PMID: 31733208 58. Dantzer B, Fletcher QE, Boonstra R, Sheriff MJ. Measures of physiological stress: a transparent or opaque window into the status, management and conservation of species? Conservation Physiology. 2014; 2(1). https://doi.org/10.1093/conphys/cou023 PMID: 27293644 59. Reeder DM, Kramer KM. STRESS IN FREE-RANGING MAMMALS: INTEGRATING PHYSIOLOGY, ECOLOGY, AND NATURAL HISTORY. Journal of Mammalogy. 2005; 86(2):225–35. 60. Keckeis K, Lepschy M, Scho¨ pper H, Moser L, Troxler J, Palme R. Hair cortisol: a parameter of chronic stress? Insights from a radiometabolism study in guinea pigs. J Comp Physiol B. 2012; 182(7):985–96. https://doi.org/10.1007/s00360-012-0674-7 PMID: 22592890 61. Goymann W. "Noninvasive Monitoring of Hormones in Bird Droppings: Physiological Validation, Sam- pling, Extraction, Sex Differences, and the Influence of Diet on Hormone Metabolite Levels." Annals of the New York Academy of Sciences (2005); 1046(1): 35–53. PLOS ONE | https://doi.org/10.1371/journal.pone.0257370 October 1, 2021 15 / 17 PLOS ONE Elevated glucocorticoids and telomere attrition 62. Khan M. Z., Altmann J., Isani S. S. and Yu J. "A matter of time: evaluating the storage of fecal samples for steroid analysis." General and Comparative Endocrinology (2002); 128(1): 57–64. https://doi.org/ 10.1016/s0016-6480(02)00063-1 PMID: 12270788 63. MILLSPAUGH J.J. & WASHBURN B.E. Use of fecal glucocorticoid metabolite measures in conserva- tion biology research: considerations for application and interpretation. General Comparative Endocri- nology (2004); 138: 189–199. https://doi.org/10.1016/j.ygcen.2004.07.002 PMID: 15364201 64. Freeman N. E. and Newman A. E. M. "Quantifying corticosterone in feathers: validations for an emerg- ing technique." Conservation Physiology (2018); 6: 9. 65. Hammond TT, Au ZA, Hartman AC, Richards-Zawacki CL. Assay validation and interspecific compari- son of salivary glucocorticoids in three amphibian species. Conservation Physiology. (2018); 6(1). https://doi.org/10.1093/conphys/coy055 PMID: 30279992 66. Garde AH, Persson R, Hansen Å. M, O¨ sterberg K, Ørbæk P, Eek F, et al. Effects of lifestyle factors on concentrations of salivary cortisol in healthy individuals. Scandinavian Journal of Clinical Lab Investiga- tions (2009); 69(2): 242–250. https://doi.org/10.1080/00365510802483708 PMID: 18985537 67. Shalev I, Entringer S, Wadhwa PD, Wolkowitz OM, Puterman E, Lin J, et al. Stress and telomere biol- ogy: A lifespan perspective. Psychoneuroendocrinology. 2013; 38(9):1835–42. https://doi.org/10.1016/ j.psyneuen.2013.03.010 PMID: 23639252 68. Mender I, Shay J. Telomere Restriction Fragment (TRF) Analysis. BIO-PROTOCOL. 2015; 5(22): e1658. https://doi.org/10.21769/bioprotoc.1658 PMID: 27500189 69. Gomes NMV, Shay JW, Wright WE. Telomere biology in Metazoa. FEBS Letters. 2010; 584(17):3741– 51. https://doi.org/10.1016/j.febslet.2010.07.031 PMID: 20655915 70. Ka¨ rkka¨inen T, Briga M, Laaksonen T, Stier A. Within-individual repeatability in telomere length: a meta- analysis in non-mammalian vertebrates. Preprints; 2021. https://doi.org/10.1111/mec.16155 PMID: 34455645 71. Rovatsos M, Kratochvı´l L, Altmanova´ M, Johnson Pokorna´ M. Interstitial Telomeric Motifs in Squamate Reptiles: When the Exceptions Outnumber the Rule. Stanyon R, editor. PLoS ONE. 2015; 10(8): e0134985. https://doi.org/10.1371/journal.pone.0134985 PMID: 26252002 72. Haussmann MF, Mauck RA. TECHNICAL ADVANCES: New strategies for telomere-based age estima- tion: TECHNICAL ADVANCES. Molecular Ecology Resources. 2008; 8(2):264–74. https://doi.org/10. 1111/j.1471-8286.2007.01973.x PMID: 21585768 73. Reichert S, Criscuolo F, Verinaud E, Zahn S, Massemin S. Telomere Length Correlations among Somatic Tissues in Adult Zebra Finches. Moreno E, editor. PLoS ONE. 2013; 8(12):e81496. https://doi. org/10.1371/journal.pone.0081496 PMID: 24349076 74. Sudyka J, Arct A, Drobniak S, Dubiec A, Gustafsson L, Cichoń M. Experimentally increased reproduc- tive effort alters telomere length in the blue tit (Cyanistes caeruleus). Journal of Evolutionary Biology. 2014; 27(10):2258–64. https://doi.org/10.1111/jeb.12479 PMID: 25228433 75. Parolini M, Romano A, Khoriauli L, Nergadze SG, Caprioli M, Rubolini D, et al. Early-Life Telomere Dynamics Differ between the Sexes and Predict Growth in the Barn Swallow (Hirundo rustica). Lustig AJ, editor. PLoS ONE. 2015; 10(11):e0142530. https://doi.org/10.1371/journal.pone.0142530 PMID: 26565632 76. Taylor HA, Delany ME. Ontogeny of telomerase in chicken: Impact of downregulation on pre- and post- natal telomere length in vivo. Development, Growth, and Differentiation. 2000; 42(6):613–21. https:// doi.org/10.1046/j.1440-169x.2000.00540.x PMID: 11142683 77. Gomes NMV, Ryder OA, Houck ML, Charter SJ, Walker W, Forsyth NR, et al. Comparative biology of mammalian telomeres: hypotheses on ancestral states and the roles of telomeres in longevity determi- nation: The comparative biology of mammalian telomeres. Aging Cell. 2011; 10(5):761–8. https://doi. org/10.1111/j.1474-9726.2011.00718.x PMID: 21518243 78. Haussmann MF, Winkler DW, Huntington CE, Nisbet ICT, Vleck CM. Telomerase activity is maintained throughout the lifespan of long-lived birds. Experimental Gerontology. 2007; 42(7):610–8. https://doi. org/10.1016/j.exger.2007.03.004 PMID: 17470387 79. Angelier F, Vleck CM, Holberton RL, Marra PP. Telomere length, non-breeding habitat and return rate in male American redstarts. Funct Ecol.2013; 27(2):342–50. 80. Asghar M, Hasselquist D, Hansson B, Zehtindjiev P, Westerdahl H, Bensch S. Hidden costs of infection: Chronic malaria accelerates telomere degradation and senescence in wild birds. Science. 2015; 347 (6220):436–8. https://doi.org/10.1126/science.1261121 PMID: 25613889 81. Barrett ELB, Burke TA, Hammers M, Komdeur J, Richardson DS. Telomere length and dynamics pre- dict mortality in a wild longitudinal study. Molecular Ecology. 2013; 22(1):249–59. https://doi.org/10. 1111/mec.12110 PMID: 23167566 PLOS ONE | https://doi.org/10.1371/journal.pone.0257370 October 1, 2021 16 / 17 PLOS ONE Elevated glucocorticoids and telomere attrition 82. Breuner CW, Delehanty B, Boonstra R. Evaluating stress in natural populations of vertebrates: total CORT is not good enough. Functional Ecology. 2013; 27(1):24–36. 83. Gabor CR, Knutie SA, Roznik EA, Rohr JR. Are the adverse effects of stressors on amphibians medi- ated by their effects on stress hormones? Oecologia. 2018; 186(2):393–404. https://doi.org/10.1007/ s00442-017-4020-3 PMID: 29222721 PLOS ONE | https://doi.org/10.1371/journal.pone.0257370 October 1, 2021 17 / 17 PLOS ONE
null
10.1098_rsos.220808.pdf
Data accessibility. This article has no additional data.
Data accessibility. This article has no additional data.
royalsocietypublishing.org/journal/rsos Review Cite this article: Dienes Z. 2023 The credibility crisis and democratic governance: how to reform university governance to be compatible with the nature of science. R. Soc. Open Sci. 10: 220808. https://doi.org/10.1098/rsos.220808 Received: 17 June 2022 Accepted: 6 January 2023 Subject Category: Science, society and policy Subject Areas: psychology Keywords: university governance, democracy, open democracy, open science Author for correspondence: Zoltan Dienes e-mail: z.dienes@sussex.ac.uk The credibility crisis and democratic governance: how to reform university governance to be compatible with the nature of science Zoltan Dienes School of Psychology, University of Sussex, Brighton, UK ZD, 0000-0001-7454-3161 To address the credibility crisis facing many disciplines, change is needed at the institutional level. Science will only function optimally if the culture by which it is governed becomes aligned with the way of thinking required in science itself. The paper suggests a series of graduated reforms to university governance, to radically reform the operation of universities. The reforms are based on existing established open democratic practices. The aim is for universities to become consistent with the flourishing of science and research more generally. 1. Introduction Many areas of science have been facing difficulties in credibility with a sense that the scientific process is not as healthy as it could be. There is low replicability of studies ([1], chapter 2), possibly associated with a failure of a field to self-correct [2]; and at the same time, there is a hyper-competitive culture aligned with perverse incentives that may reward substandard science [3]. The solutions to this credibility crisis will surely involve multiple levels of reform [4]. Specifically, cultural change is needed at the institutional level, which is the level that this paper addresses. Initially, a simple account of how knowledge grows is presented. Then a brief historical overview is provided of the growth of knowledge and its relation to how decisions are made in the broader social context. Classical Athens is taken as an example of matching between the governance of the society as a whole and the growth of knowledge. Next, the state of governance of UK universities is considered. Finally, open democratic practices that are Athenian-like and that have been tested in politics are considered for how a university might be governed in an open democratic way in order for the university to align itself with the way knowledge grows. © 2023 The Authors. Published by the Royal Society under the terms of the Creative Commons Attribution License http://creativecommons.org/licenses/by/4.0/, which permits unrestricted use, provided the original author and source are credited. 2 r o y a l s o c i e t y p u b i . l i s h n g o r g / j o u r n a l / r s o s R . S o c . O p e n S c i . 1 0 : 2 2 0 8 0 8 2. How does knowledge grow? Popper [5] asked ‘How does knowledge grow?’ His answer in general, is by trial and error; try ideas out and see what works; reject those that do not work. How could that process be enhanced? According to Popper, traditionally schools of thought were passed down from master to disciple with an aim to impart a doctrine pure and unchanged. There is in such a tradition a hierarchy with roles filled by people by virtue of their characteristics: master and disciple. This is a good way of conserving knowledge as it is, but not for promoting the growth of knowledge. But consider an alternative, which Popper calls the critical tradition: the master says ‘Here is my idea to solve a problem; can you improve on it?’ It can be difficult to encourage others to do better than oneself. Thus, the critical tradition as a second-order tradition that is passed on from mentor to pupil has to be constantly fought for: businesses, religious leaders, politicians and even academics will regularly try to stamp out criticism of their ideas. The critical tradition occurs when there is a culture of considering arguments for their own sake, with small regard for the authority of the person stating them. That is, in a critical tradition roles are fluid, and what is important is the quality of ideas. Taking part in a critical tradition may be psychologically easier when people see ideas, theories and data as having an objective existence apart from themselves, with properties that must be discovered; this is what it means to consider arguments for their own sake. Then people can refute a theory without thinking they themselves have been harmed [6]; cf. also [7]. Let an open society be a society in which such a critical tradition is encouraged (cf. [8]). Let democracy be an open society in which there are institutions that encourage a critical tradition independent of any individual. Thus an autocratic ruler may promote an open society if that was the sort of thing that ruler liked; but the society would not as such be a democracy, because the existence of the open society would depend on the whims of a particular ruler. 3. Lessons from history There is intriguing evidence of a historical relationship between the growth of knowledge and the existence of an open society, especially democracy. Popper [5] suggests how Thales, around 600 BCE, proposed a natural principle for how the world works, only for his apparent student, Anaximander to come up with something logically better, starting a critical tradition. For the next several hundred years in Athens, there was an astonishing flourishing of knowledge, in mathematics, astronomy, history, psychology and medicine. Knowledge and open critical discussion continued into the extended Greek and Roman world for some time AD, but the critical tradition gradually withered. For example, the Epidemics of the Hippocratic corpus (fifth century BCE) mainly indicated how their treatments failed ([9], chapter 5) in contrast to case histories from later centuries (consider the numerous triumphs Galen, fl. second century AD, described in outwitting other doctors ([10], e.g. chapter 7); after Galen, there were no students of his who tried to produce better solutions, at least not the for many centuries). Almost exactly contemporaneously with the rise and decline of flourishing of knowledge, there was a rise and decline in Athenian-style democracy. In Athens itself, the initial reforms of Solon (600 BCE) were strengthened by Cleisthenes (coming up to 500 BCE), then after further gradual refinements, Athenian democracy was formally ended in 332 BC by the conquest of Alexander the Great. However, as seen as part of an ecology of about 1000 Greek states, democracy robustly continued well into the second century BCE, with the number of democracies actually increasing for some time [11]. Over several hundred years from before until after the classic period, these Greek states showed shifting mixtures of democracies and elite control. Based on archaeological and historical evidence, Ober argued that cultural flourishing (and wealth) followed not the rise of conservative institutions but tracked the development of democratic rule-egalitarian institutions. There was both an explosion of knowledge and the implementation of a robust long-lasting democracy during the time of classic Athens, followed after a delay by a slow stagnation in knowledge growth. At other times and places, when there was an open society, knowledge also flourished—even without democracy. Saliently, from 800–1400 Arabic science drew from Greeks and Indians, and made major progress [12]. There was not democracy, but absolute rule by Abbasid caliphs. Any given caliph might support an open society (e.g. initially especially Al-Ma’mun), in the sense of encouraging critical discussion of ideas (e.g. Al-Khalili [12], pp. 67–68). When a caliph, or others in positions of authority, supported the free exploration of ideas, knowledge grew; when a subsequent caliph was more conservative, learning shrank (e.g. [12], p. 230, cf. also p. 194). The relation of general openness during this time and scientific progress bears further investigation. 3 r o y a l s o c i e t y p u b i . l i s h n g o r g / j o u r n a l / r s o s R . S o c . O p e n S c i . 1 0 : 2 2 0 8 0 8 The critical tradition sustained in the Arab world eventually found its way to the medieval Italian states. From about 1100, these states were already exploring democratic governance, using mixtures of lot and election (e.g. [13,14]). The assembly politics often characterizing decision making in those states entailed considering arguments for their own sake even if the person voicing them may be of low rank. Thus, the ground was laid for exploring new ideas; and later there is indeed the outpouring of new forms in art, invention and later science. Ferris [15] picks up the story of the intertwining of democratic values and the growth in science in the 1600s onwards, noting how science developed in the most liberal countries, and how conversely scientists were key people pushing for democratic change. A further explosive growth in knowledge occurred alongside the progressive rejection of authoritarian political values and the development of liberal values, from the enlightenment onwards. Yet the growth of knowledge has also occurred when society as a whole was not especially open, and conversely, science often did not occur where there were even democratic institutions. An example of the first point is the steady growth of knowledge throughout Chinese history. Up until about 1400, China was hundreds, and sometimes thousands of years, ahead of Europe in technological development. Needham [16] spent decades documenting how many innovations came from China, for example, China pioneered inoculations ( possibly tenth century and certainly by 16th); China developed mechanical clocks six centuries before Europe. True, China throughout its history has valued scholars very highly and had a well-organized civil service based on exams rather than (explicitly) on pedigree. But this time the emperor in principle had the final say on any matter (including an edict that held from 653 forbidding the private possession of astronomical instruments ([17], p. 228)), and there is little evidence of democratic processes (except briefly for a period in the Zhou dynasty, 1050–221 BCE ([14], p. 150)). There was the constant development of technology—but not the explosive development of science. Needham asked why did modern science not develop in China and only in Europe? Why did knowledge of the physical world grow steadily in China, and yet not explode like it did in Europe? Needham suggested that ‘There was no modern science in China because there was no democracy’ [16, p. 152]. Needham pointed out that science is indifferent to who makes the argument; thus ‘these civilizations which have developed an … exaggerated respect for teachers, will have to modify it’ [16, p. 140]. In sum, ‘there is a real kinship between the scientific ‘ namely, skepticism, anti-authoritarianism, not letting others mind and the democratic mentality, decide on aims or assessment of evidence, ‘a give and take, a live and let live attitude’ [16, p. 143]. (For a review of other hypotheses by Needham and others to address this question, see [17].) throughout Conversely, Stasavage [14] and Graeber & Wengrow [18] present historical, archaeological and anthropological evidence for democracy, in the sense of decision by assemblies, being a common solution to the problem of political governance, and often in large-scale societies. If that is true, why did science not emerge multiple times? Debates in assemblies, to the extent criticism of any individual’s views are welcome and not just tolerated, promote a critical tradition. And a critical tradition allows knowledge to grow, but it need not be specifically scientific knowledge. Graeber & Wengrow argue for the political sophistication of the indigenous Americans, whose skills were honed in assembly politics, and who could hold their own if not best European intellectuals of the 1600s in political and social arguments. Indeed, according to Graeber and Wengrow, native American arguments may have, unacknowledged, transformed the course of European intellectual history. Consider also the democratic politics in India at the time of the start of Buddhism (Mahaparinibbana sutta in [19]), which occurred contemporaneous with the development of ideas about the mind which continue to influence modern thinking (e.g. [20]). Clearly there is no deterministic relation between science and democracy; but there is synergy. The lesson to draw from history is that when science and democracy occur together, science is facilitated. Science might happen without democracy, and democracy without science—but put them together and see what happens. 4. Athenian-style democracy Athenian democracy emphasized the selection of arguments that could in principle be provided by anyone, rather than primarily the selection of people according to their traits as such [21]. Let us consider some details, because they will be useful for considering modern reform (for a readable overview with the specific aim of implementing the principles in modern corporate institutional governance, see [22]; see also [23,24]. Athenians were divided into 10 tribes pooling different types of citizens (urban, rural, coastal) into in-groups to unite pre-existing cultural divisions (consider the use of houses or colleges in some universities). Day-to-day business was organized by the Council of 500 classical Athens council of 500 (Boule) agenda sets agenda and implements decisions general administration and supervision of the system decisions law courts LOT Tribe 1 Tribe 2 popular assembly (Ecclesia) – all citizens (30k) discuss and approve motions nomothetai – checks new laws for consistency with old LOT Tribe 10 . . . (the citizens) 4 r o y a l s o c i e t y p u b i . l i s h n g o r g / j o u r n a l / r s o s R . S o c . O p e n S c i . 1 0 : 2 2 0 8 0 8 Figure 1. Sketch of classic Athenian democracy. In Athens circa fourth century BC, eligible citizens were assigned by lot, i.e. randomly, to the main governing bodies and the law courts. Of about 1000 posts, 90% were determined by lot and the rest by election. The majority of citizens would have spent at least a year at some point in their life in the main governing body, the boule. Some form of such governance lasted for hundreds of years in Athens and other Greek states. (the boule), consisting of 50 eligible citizens selected by random lot from each tribe for a period of a year. Each month the 50 from one tribe would set the agenda for the business of the day, and prescribe the meetings of the Assembly (which we will come to). Other duties include overseeing the work of all officials and financial oversight. Each day officials from the other tribes in the Council make decisions concerning the agenda. The leader of the Council was selected daily from the tribe in charge of agenda-setting that month: one person was the nominal head of state for one day alone! In sum, central decision-making was deliberately integrated over many people, chosen by lot from the citizenship. Roughly weekly, the citizens, or the subset who turned up (maybe a fifth of all citizens on any one occasion), formed the Assembly (ecclesia). The agenda was set by Council, and in principle any citizen could speak on any motion, before a vote to decide each proposal. One further institution is worth mentioning: the nomothetai. This was a panel formed by lot from eligible citizens to reflectively consider arguments for and against proposed general laws before a final decision was actually made to accept them. We will draw on this important institution later. Figure 1 for a sketch of the overall structure of Athenian democracy. 5. The state of UK universities Coming into this millennium, many UK universities were democracies at the school (i.e. faculty) level. School meetings were decision-making bodies, with decisions made usually by majority vote. Decisions fed up to the central university level. Senior management at the central level then had to do their best to render coherent at the institutional level the way these parts fitted together. Subsequently, in about the first decade of the millennium, many UK universities moved to more or less complete non-democratic top-down control, where senior management made decisions, and the Deans of schools were to work out how to implement those decisions to management’s satisfaction. The Dean rather than faculty had final say at school level. The Vice-Chancellor (VC) was appointed by Council with no involvement by faculty at large (with the members of Council being appointed by Council). These reforms occurred in the tradition of ‘New Public Management’, a philosophy of public sector management which started to be implemented in the Thatcher years, and has become dominant in the UK in its higher education policy [25–27]; see [28], for a history of UK education sympathetic to this approach). In many universities in the UK, top-down control is exerted by management, who are perceived to be a class separate from academics with no real accountability [29]. These managers may apply strong pressure on academics to achieve key performance indicators. The result is managerialism: the worth of an academic (for getting jobs, promotion, respect) is determined by set performance goals, typically including getting grants and publishing in high-impact-factor journals. The first duty of a researcher is not high-quality research by their own judgement, but to fulfil the agenda of an increasingly large class of managers (whose agenda is to boost e.g. world rankings and other metrics). It is not obvious 5 r o y a l s o c i e t y p u b i . l i s h n g o r g / j o u r n a l / r s o s R . S o c . O p e n S c i . 1 0 : 2 2 0 8 0 8 in project management the failure of complete top-down control that top-down control by senior managers is the best way of dealing with a rapidly changing and unpredictable environment that is necessarily what the interface of knowledge and ignorance consists in. For in unpredictable environments for aid agencies, see Honig [30]; and its failure in the secondary education sector, Honig [31]. Honig [31] presents evidence that top-down management with close monitoring and control backfires when those managed are already there because they want to be: such management produces both selection effects, the loss of good people, and motivational changes in those that remain (for the latter see also [32], chapter 5). Martin, reviewing the history of UK university governance with an eye on the management literature, asks, ‘Why, when the management literature of the last two decades has stressed the benefits of flatter organizational structures, of decentralization and local initiative … have many universities been intent on moving in precisely the opposite direction of greater centralization with a more hierarchical, organizational structure, top-down management … and ever more cumbersome and intrusive procedures?’ [27 p. 7].1 And of specific relevance to science, Xu et al. [38] found that scientific teams with a flat rather than hierarchical structure produced more novel ideas and a higher long-term citation impact. farms [40]. When an expert needs to exercise judgement Attempting to incentivize ‘performance’ when what really counts as performance cannot be easily measured—as is the case in attempting to understand the unknown—will generally backfire [30,39]. Simple-minded performance targets famously backfire even for simple problems. Consider the rat tails of Hanoi. At the beginning of the twentieth century the French colonial rulers of Hanoi wanted the city rid of rats. To bring on board the local population, money was offered for each rat tail delivered as proof of a killed rat. Yet the rats only increased. It turned out the locals, being resourceful, set up in an unpredictable environment, rat sustained incentives to maximize something just because it is measurable, typically distort best practice. The best person to judge strategy and tactics for dealing with a research problem is the researcher themselves. Incentivizing them to, for example, apply for grants, will distort how time is best spent. If a researcher needs a grant to further their research, they have no need of a manager to tell them to get a grant. Conversely, if their time would be better spent writing up that file drawer of papers, incentivizing them to apply for grants only promotes inefficiency. But the problem may be far worse than this. Key performance indicators filtering down from senior management as pressure on individual researchers may not just waste time—it may damage the integrity of science. It may produce, in effect, rat farms. In simulation studies, Smaldino & McElreath [41] consider a population of different laboratories who differ in the degree of p-hacking they engage in while competing for grants. Given reasonable assumptions, those laboratories that p-hack the most are most successful in obtaining grant money— and thereby produce more progeny laboratories (via the PhD students and post docs they train) which carry on the same culture as the parent laboratory. The current environment of intense competition for grant money plausibly promotes low-integrity science (while of course being consistent with some successful laboratories that do value rigour, as shown, for example, by their commitment to open science). Smaldino et al. [42] consider solutions. In their simulations, the way to break the effect of the selection of p-hackers, was to borrow an idea from classical Athens, selection by lot: award grants by random lottery for those submissions that passed a minimal standard of methodological rigour. (Such a procedure not only can restore integrity, it also ensures a lack of discrimination based on gender, race, or institution in grant allocation.2) While the concept of p-hacking is not relevant to all disciplines, the argument plausibly generalizes for any similar process where cutting corners to the detriment of the quality of the research nonetheless allows outcomes that look convincing. Satisfying key performance indicators typically involves publishing in journals with high impact factors. Often high-impact-factor journals are run by for-profit companies charging high publishing fees. A principle of how science should function is that anyone should be able to contribute according to only the quality of their contribution. High publishing fees mean that only those able to join a rich 1Leaf [33] describes widely varying levels of democratic governance across different US Universities, yet indicates the direction of change is toward less democracy. For the rise of managerialism in US Universities, see Ginsberg [34]. For the variability in extent of democratic governance across Europe, but the same direction of change as the UK, see de Boer & File [35]. And for the headlong embrace of managerialism by Australian Universities, see Biggs [36] and Hil [37]. 2For the greater efficiencies that would be achieved by grant lotteries, see Gross & Bergstrom [43]. For a different solution, see Brette [44]. 6 r o y a l s o c i e t y p u b i . l i s h n g o r g / j o u r n a l / r s o s R . S o c . O p e n S c i . 1 0 : 2 2 0 8 0 8 person’s club, at an intuitional or individual functioning of science [45]. And it may be worse than that. level, can contribute. This undermines the proper Does pressure to publish in high-impact journals produce p-hacked and less reproducible research? Direct evidence for pressure to publish producing poor quality publications as a general relationship is not yet in (consider the attempt by [46]). But there are some specific connections. Cash bonuses for publishing in high-impact-factor journals, as has been practised in Australia and China, is associated with retractions of papers [46]. Further, Fang & Casadevall [47] found a proportional relation between retractions and journal impact factor; most of these retractions were due to fraud [48]. While the retraction–impact factor relation may in part be because articles in high- rather than low-impact-factor journals are subject in his investigation of research quality across a range of scientific disciplines. He found a negative relation between journal impact factor and various measures of methodological rigour3 (also recently found by [51], in management science; and [52] found weakly negative relations in behavioural science and neuroscience). So if the papers are less good methodologically on average, why do they get published in higher impact-factor journals? Presumably because the authors were good at selling their results. In sum, managerialism at university level, situated in a dysfunctional ecosystem, selects for p-hacking/ corner-cutting salespeople. this explanation is ruled out by Brembs [49] to greater scrutiny, How is the rise of managerialism experienced by university staff? Shattock & Horvath [53] conducted extensive onsite interviews with staff at all levels of each university, at UK universities that spanned a ‘The wide range of rankings, to explore staff experiences. They found widespread dissatisfaction. sense that conditions for the pursuit of high quality academic work have worsened and are continuing to worsen is widespread, even in institutions that are most obviously successful. Criticisms that universities have become too top-down in their governance, and are insufficiently bottom-up, that good academic work is stifled by over-regulation and bureaucracy, and that too much academic business is handled by non-academic professionals, are commonplace’ [53, p. 104]. Similarly, Erickson et al. [29] in a survey of 5888 academic staff in the UK higher education sector, found only 10% of university staff were satisfied with senior management. The question is, in terms of university governance, could we be doing better? In the next section we consider democratic solutions. 6. Open democracy Hierarchical university governance may contribute to damage to scientific integrity. So what mode of governance might actually promote the growth of knowledge? That is, what way of governing universities would be consistent with a culture of the critical tradition, the tradition of carefully considering and selecting arguments rather than people? Arguments concerning the running of an institution can only be considered as such, by anyone with a stake in them, if there is transparency of information. Relevant information must be readily accessible ( just as is required for science to function). What organizational structure produces transparency in a way that is useful? Consider the following assumptions: (1) Each academic has information about how the university is working. (2) Different people have different relevant information. (3) Information will be expressed when a person has to use it to make decisions. (4) Those decisions will make best use of the information if people making decisions have to live by them. (5) To select the best ideas (that integrate information well), ideas should be selected for, not people. Ways of governing that satisfy these assumptions include those of Athenian-like democracy that we considered earlier. Athenian-like democracy has inspired a number of open democratic practices that have been explored in the political context in the last few decades [54]. We will review some of these practices now. 3Again, this finding is consistent with some papers in high-impact-factor journal being rigorous. For example, plausibly if the papers are Registered Reports, then they may have good methodology (e.g. [50]). But then consider the quality of Registered Reports in PCI RR, which is free to authors and readers, and whose rigour the reader can assess for themselves because the whole process is open: https://rr.peercommunityin.org/. 7 r o y a l s o c i e t y p u b i . l i s h n g o r g / j o u r n a l / r s o s R . S o c . O p e n S c i . 1 0 : 2 2 0 8 0 8 Before describing concrete ideas, consider an objection. Given the complex environment in which universities now operate, don’t we need decisions made not democratically by the uninformed but rather by experts (cf. [55]), with a top-down governance structure that thus allows nimble and agile decisions? As against this, if the above five assumptions are accepted, some form of democracy may foster decision-making where the most information is maximally integrated in the time available4: structures embedding such practices should produce networks between people closer to small-world networks than a strict top-down hierarchy could; and such networks allow more global integrated information [57]. Open democracy can ensure decisions are made by the well informed, as we discuss. And it is rare to hear modern universities described as nimble and agile [29,53]. Open democracy also presents a smorgasbord of ways of being democratic with different time scales of operation. Indeed, in Athens, important decisions were sometimes made or reversed very quickly, even in the space of a day ([22], e.g. pp. 133–134). Sometimes decisions should be fast, sometimes slow and reflective. With this in mind, let us see what is on offer. 6.1. The deliberative poll immigration or, The deliberative poll was developed by Fishkin [58,59], and will be used to illustrate the more general class of mini-publics, procedures by which people are selected from the population of citizens to deliberate an issue—such as citizen’s assemblies, citizen’s juries or citizens’ initiative reviews (which we will consider below) (see [60], for a review). Consider a difficult issue that concerns a community, and for which a reflective and informed decision is needed that takes into account diverse viewpoints—such as Brexit, in a university setting, the principles for allocating resources to schools in a context of shifting student demand. Randomly select 200–300 people from the total population (one may decide to over-represent certain groups of people particularly affected by the issue). Given the selection is approximately random, everyone has a chance to contribute and no one is selected simply because they have a vested interest. The total sample is allocated into groups of 15 to discuss the chosen issue. The discussants are given information packs, each pack prepared by experts or protagonists of different views. The discussions are moderated to encourage everyone to contribute more or less equally, and for debate to focus on arguments per se. There is an opportunity for discussants to ask a panel of experts any questions that remain unresolved. After several meetings, discussants anonymously vote: as Fishkin puts it, the vote is then what the people would think if they had reflected. The deliberative poll has been used in many countries over the last few decades, for example in the UK, Europe, Australia, China, the USA, Canada and Mongolia (for critical discussion see [61], chapter 3; [60], chapter 3). To take an example, in Ze Guo Township in China, the issue was how to spend the council’s money [58]. Thirty projects were listed by the council. Two hundred and seventy-five citizens were randomly selected; of these 235 completed the poll, so the final sample was close to random. Fishkin provides evidence that participants became more informed as a result of the discussions, that there was little to no domination by privilege, that there was little to no group polarization, and that priorities shifted toward projects that would benefit the whole town. Of course, the devil would be in the details to get such good procedural outcomes (e.g. well-trained moderators). People chose a sewage treatment plant, park and a main road; not, for example, a fancy town square. These choices surprised officials but they acted on the results of the poll. A minipublic can be used to either set the agenda of a committee, or to make decisions on issues defined by a committee, and both roles may be useful in a university. To adapt the minipublic to a university, one might adjust the number of people selected, or the length and number of meetings, according to the issue considered. Note the similarity of the minipublic in finalizing decisions to the nomethetai in the Athenian model. 6.2. Participatory budgeting Engaging the community as a whole in planning budgets was pioneered in Porto Alegre in Brazil in 1989 onwards in a procedure called participatory budgeting (see [60], for an introduction). In this case, neighbourhood assemblies voted locals to represent the neighbourhood for a year in a local committee 4Information integration in different organizational structures could be tested by injecting information into an organization at different places and then at a later time measure whether that information has been used to make decisions at different points in the organizational structure; compare measuring information integration in the human brain [56]. 8 r o y a l s o c i e t y p u b i . l i s h n g o r g / j o u r n a l / r s o s R . S o c . O p e n S c i . 1 0 : 2 2 0 8 0 8 to plan projects and their priorities. People were also elected to one of several thematic committees (education, transport, etc.) to plan projects and their priorities for each theme. Locals from each neighbourhood and thematic assembly were also voted to an overall coordinating central committee. The central committee decided projects and allocated a substantial proportion of the council’s budget. The committees also involved people with relevant technical and financial expertise. A person could be elected for no more than two terms. The choice of local people, the shortness of the term and the limited number of terms per person is what roughly corresponds to random sampling in deliberative polls in the sense of being the mechanism that limits entrenchment of certain people in decision making. Participatory budgeting was regarded as so successful it has been taken up in different forms in more than 2700 governments (though in 2017, participatory budgeting was suspended in Porto Alegre itself [62]. For a university following the Porto Alegre model, schools could, as local neighbourhoods, select faculty on a rotating basis to a school budgeting committee for a year to decide school projects. Staff from relevant groups could be selected on a rotating basis to committees devoted to certain themes at institutional level, such as IT, catering, grounds, etc. Similarly, each school could select on a rotating basis two representatives for a central budgeting committee to determine institution-wide spending and to finalize decisions of the other committees. In the participatory budget model, selection is by local election; but people could be selected (semi) randomly. 6.3. The citizens’ initiative review The citizens’ initiative review [63] is a one-group minipublic where the group jointly summarizes the best arguments pro and contra a proposal, and also summarizes how they voted. The aim is to provide an information leaflet for a referendum on the proposal that reflects the range of views ordinary people would have if they reflected on the issues and made themselves informed. Crucially, therefore the information leaflet is not provided by vested interests. The citizens review initiative has been used extensively in the state of Oregon, USA, where studies indicate that voters appreciated the information and were better informed as a result of it. 6.4. Allocation of citizens to roles Democracy is often associated with voting, as that is how our current representative democracies work. But when academics vote for people in the university to be on senate, council or some other position, often the vote is based on limited information, such as what school they work in, or what other committees they have been on. The facts given may be of marginal relevance to how that person would contribute to that role. And if the information is enough to be seen as relevant, voting is a mechanism for selecting people who stand out from other people, in other words, for selecting elites [13]. Having the resources or motivation to promote oneself in an election is not the same thing as having the qualities that would make one good at the job the election is for. The medieval Italian city states realized this. But they also thought selection strictly by lot may not select the best people for the job, even if it ensures the top jobs are not all held by people of a certain class. So these city states devised various combinations of election and lot to try to obtain the benefits of both procedures (see [64], chapter 7, for a systematic analysis of the possibilities). For example, one could select by lot a group of candidates, who are then chosen by election. An institution could decide what combination of processes will assign people to posts. One counter-defence of selection by lot alone, i.e. without election, is that the repeated use of such a procedure educates the citizenship; further if each person is merely a constraint in a global process of selecting ideas (as in science) the notion of selecting the right person may lose some of its relevance. Note that once people have been selected for the executive committee by an open democratic process, such a committee can in principle act in real time as fast as any committee now, depending on details. 7. Reforming the university Figure 2 shows a schematic summary of the flow of control from the top down in a hypothetical modern university. The simplest way of seeing how radical reform could be made is keeping the same structure but assigning people to roles by an open democratic process (figure 3a). Just as in Athens, assignment by a schematic possible structure of a UK university now top-down control senior management council senate Dean 1 Dean 2 Dean 3 committees e.g. teaching and learning School 1 School 2 School 3 Figure 2. Schematic diagram of top-down governance as might exist in a current UK university. Control runs from top to bottom. Senate allows some pushback on academic matters but senate may only in practice have the power for suggesting that senior management or council reconsider. (a) new democratic governance model? (b) new democratic governance model? council executive committee; constantly selected from each of the schools minipublics: agenda decisions selection (sortition?) council proposals decisions executive committee; constantly selected from each of the schools selection (sortition?) selection (sortition?) committees e.g. teaching and learning decisions selection (sortition ?) committees e.g. teaching and learning decisions School 1 School 2 School 3 School 1 School 2 School 3 9 r o y a l s o c i e t y p u b i . l i s h n g o r g / j o u r n a l / r s o s R . S o c . O p e n S c i . 1 0 : 2 2 0 8 0 8 (c) new democratic governance model? (d) new democratic governance model? council proposals decisions executive committee; constantly selected from each of the schools minipublics: agenda decisions selection (sortition?) enough signatures triggers proposal selection (sortition ?) committees e.g. teaching and learning decisions council proposals decisions executive committee; constantly selected from each of the schools minipublics: agenda decisions selection (sortition?) enough signatures triggers proposal proposals decisions assembly of citizens selection (sortition?) committees e.g. teaching and learning decisions citizen initiative review? School 1 School 2 School 3 School 1 School 2 School 3 Figure 3. A series of democratic changes to the governance structure shown in figure 2. (a) The same command flow through committees could be kept as in figure 2, but people assigned to committees by sortition (lot). (b) The executive committee could use minipublics to set certain agendas or make decisions. (c) To this could be added a standing right for decisions to be reviewed by the executive committee if a petition with enough signatures is submitted. (d) There could also be a general assembly of citizens to which the executive committee could submit some decisions. So that the assembly can make an informed choice, such referenda could be supported by citizen review initiatives. lot does not mean there are no restrictions. Some jobs may be open only to senior lecturers or above for example. Or some committees may require experience of other committees. While this one change is structurally simple, it is of course a radical first move. It involves the abolishment of senior management. Some people, notably senior management, might think this a possibly catastrophic first move. This possible first move is presented to illustrate how one can keep other things the same, yet radically change the democratic nature of governance to be similar to the style of governance that once thrived in a complex nation for over a hundred years. In practice, open democracy should be explored in small steps. One could first of all set up a minipublic, with commitment from senior management to abide by its decisions, on an issue of importance that needs 10 r o y a l s o c i e t y p u b i . l i s h n g o r g / j o u r n a l / r s o s R . S o c . O p e n S c i . 1 0 : 2 2 0 8 0 8 time to consider, for example, the university’s response to proposed pension reforms. The issue needs reflection and requires becoming immersed in relevant information, and should not be decided solely by people with one type of axe to grind. Figure 3b illustrates the addition of minipublics. Once their use has been explored and finessed, more changes could be explored. Figure 3c shows a further step that could be taken as an initial small step. To increase recurrent information flow through the system, so that it functions as close as possible to a self-organizing dynamic system that maximizes global constraint satisfaction, if enough signatures are obtained for a petition, the executive committee could be obliged to reconsider a decision, then provide information for why they kept it or changed the decision. Finally figure 3d shows the addition of an assembly of citizens. On some occasions the executive committee may wish decisions be decided by referenda. In this case, citizens’ initiative reviews should be used to provide unbiased information about what is at stake. Whether an institution makes any one of the changes suggested in this sequence should be a slow process of exploration. 8. Conclusion This paper outlined some broad principles for making decision making in a university more closely match decision making in science, arguing both that open democracy allows good decisions (else why do we use it in science, one of our most successful endeavours?); and that good science will be promoted when embedded in a broader culture that operates in the same way as itself. But much was left unaddressed. Who counts as a citizen? That will need to be addressed by an institution (bearing in mind we should be accountable to students, in a way we are not in the current system). What about professional services? Just as politicians need a civil service that has expertise and will offer alternative proposals, so academics need professional services to allow the university to run. The organization of professional services has not been addressed, but presumably some of the same principles could apply to their governance. What about Council, the ultimate governing body of a university? Shattock and Horvath ([53], p. 100) are scathing about how Councils are currently formed in terms of what is expected of them. An institution will need to decide how best to make Council better informed and more accountable through open democratic practices. The argument is not at all that the credibility crisis in science, and the current dissatisfaction with university governance, is the fault of any particular individuals, most of whom are simply people trying to make the best decisions. The problem is the structure in which senior management operate. No amount of focus groups to determine key catch phrases to repeat as slogans will change that. The current governance structure is almost designed to be divisive and demoralizing. Just like an Abbasid caliph, a current senior manager or VC of a modern university may somehow benignly run a happy and smooth operation that promotes the flourishing of knowledge despite the system. Until the next VC comes along. Let’s make the system itself work. One concern is whether open democracy will increase the admin load on people. If committees maintained the same numbers as currently, the average committee load ceteris paribus remains the same. There is an incentive difference, however, that may reduce average committee burden: while professional bureaucrats have an incentive to maximize number of committee meetings (are there not more data on key performance indicators and ‘academic drivers’ to be drilled down into?), people who are committee- averse have an incentive to reduce them. On the other hand, deliberative polls formed to consider an issue will increase admin time. The trade-off is that this increase in time is time spent being informed about how things work, being a part of its workings, and making a difference to how things work. A related concern is random lot means people may be selected who consider they are already being overworked. One could be allowed to refuse such an assignment a certain number of times in a certain period. But in the end the system will only function if people are committed to being citizens. The current pay of senior management could be split among citizens, and people could opt out of being citizens. People who opt out would not be taken seriously when they complained bitterly about decisions that were made. But is democracy weak under pressure, inherently irrational, maybe susceptible to mob rule? Tangian ([65], p. 294) argued that Athenian democracy would make unstable decisions for ‘situations close to controversy [i.e. 50/50 for and against in the population] because a negligible disturbance results in a significant change. At the controversial point, the decisive majority opinion, whether positive or negative, depends on the positions of just a few individuals.’ Similarly, there are long-standing theorems, such as Arrow’s theorem, that seem to show a group of people making decisions by voting are necessarily irrational in one way or another. Deutsch ([66], chapter 13) points out that these 11 r o y a l s o c i e t y p u b i . l i s h n g o r g / j o u r n a l / r s o s R . S o c . O p e n S c i . 1 0 : 2 2 0 8 0 8 arguments, that might be used against democracy, including Tangian’s just mentioned, presume there are fixed options regarded by people with fixed preferences. But the point of critical discussion is to change one’s preferences and allow new options to unfold. That is the real substance of decision making. Open democracy allows preferences and lists of options to change in a rational way. And to the extent that having people form small minipublics with moderated discussion promotes sensitivity to reason, democracy need not amount to mob rule. How could these reforms be implemented in practice? Popper [67] argued for piecemeal social engineering, that is taking small steps at any one time, as the politician is aware that the perfect state, if attainable, is far distant, and each change along the way will have unintended consequences, which are best dealt with one by one. Could a university VC be persuaded there is at least one decision that is important for people, yet one they could be willing to give up to a minipublic? It is sometimes in a leader’s interest to have handed over difficult decisions to a public body, such the Citizens’ Assemblies to consider abortion or marriage equality in Ireland [68]. Or if a VC genuinely considers a pension deal is the best one for the university, and they trust fully informed rational discussion will lead to the same conclusion, why not, by having a minipublic decide, save themselves from losing goodwill for the rest of their term? Or why not start with a decision about which they have no axe to grind, but staff do care about—might that not help staff morale? Might not a university who started along this path become a beacon for other universities, be the one that stood out from the other sheep by not copying their nearest neighbour sheep? Start with a single decision. Then finesse the procedure to make it better. Gradually finessing the procedure will take effort; but as indicated there are reasons to think it is worth the effort. In fact, open democracy may uniquely solve a key problem. Fukuyama [69] argued that all political systems face the problem of elite entrenchment bringing about inevitable decay of the system. Analogous arguments may apply to universities, not just nation states, when there is a limited pool of managers who, schooled by the same system, approach problems in similar ways. Despite what Fukuyama claimed, there is a way of hindering elite entrenchment: by constantly randomly selecting citizens to act as decision makers, there is a constant input of new viewpoints. Injections of randomness are necessary for creativity. Of course, people with many good ideas may still be especially influential. The goal is to make sure that in taking on board ideas, what matters is the quality of the ideas. Thus, good ideas should be selected, whatever their source. The benefits of open democracy may also be motivated in the light of McGregor’s [70] influential distinction between two theories that management might have about the psychology of people. According to Theory X, ‘the average human being has an inherent dislike of work … [and so] must be coerced, controlled, directed, threatened with punishment to get them to put forth adequate effort … ’ (p. 43). By contrast, according to Theory Y, ‘The expenditure of physical and mental effort in work is as natural as play or rest … [a person] will exercise self-direction and self-control to the service of objectives to which [they are] committed … the capacity to exercise a high degree of imagination, ingenuity and creativity in the solution of organizational problems is widely, not narrowly, distributed … ’ (pp. 59–60). A relevant principle may be that when management treats people in ways that express certain expectations of them (e.g. Theory X or Y), management tends to get what it expects (for a business case study, see [71]). Likewise, requiring academics to fulfil narrow key performance indicators, may tend to produce academics who do as they are directed (and no more). Managers who subscribe to Theory X may regard this as proof of their position—even as science and organizational creativity suffers. Yet trusting faculty to solve important organizational and research problems, as open democracy requires, may yield superior outcomes, by promoting those qualities presumed by trust (see [31], and [39] for a review of relevant evidence). Current problems with the university environment that were identified earlier included the overuse of key performance indicators and the use of a top-down governance structure. Yet democracy as such is orthogonal to both specific decisions concerning working conditions (e.g. whether staff are incentivized by particular metrics), and also by whether the flow of control runs from top down or bottom up in terms of committee structure. A democratic university may (or may not) decide to incentivize, for example, grant income in promotion criteria. If they do, this will be a decision whose details will be finessed by those who daily confront what the real trade-offs are. And given general dissatisfaction with the way senior management attempts to incentivize academics now, a democratic university may in at least some institutions come to different decisions in detail. Importantly, the decision will be made by faculty knowing they are trusted to make decisions, because open democracy embodies Theory Y thinking. When metrics are decided locally and with light touch by the experts, and used with judgement as a means and not an end, they can be useful and need not be demotivating [39]. Similarly, the flow of control can still be democratically top down as in figure 3, allowing greater coherence across the university (democratic decisions can apply to the whole university), or bottom 12 r o y a l s o c i e t y p u b i . l i s h n g o r g / j o u r n a l / r s o s R . S o c . O p e n S c i . 1 0 : 2 2 0 8 0 8 up, allowing different schools more independence. The two directions of flow can both be democratic because the people at the bottom and the top are, over time, the same when there is assortment by lot, or the use of deliberative polls. What the direction of the flow of control is for a given institution can be worked out according to the needs of a particular institution (say, by a deliberative poll). And in whichever direction control goes, once again a crucial difference remains compared with the status quo: An open democratic system embodies Theory Y thinking. Once a democratic organizational structure is in place, the decisions that result can be jointly owned. It will be us who made the decisions. We can say: this is our house, we built it. Our state. We as citizens may make a mess of it, as we invariably must, as any decision-making procedure will. But it will be our mess, our problems to solve—together. Data accessibility. This article has no additional data. Authors’ contributions. Z.D.: conceptualization, writing—original draft. Conflict of interest declaration. I declare I have no competing interests. Funding. I received no funding for this study. Acknowledgements. Thanks to Jason Chin and David Mellor for valuable comments. Thanks also to Josh Ober who gave talks at Sussex in 2015 which helped inspire me, in conversation with Jorg Huber. References 3. 2. 4. 5. 1. 6. Ritchie S. 2020 Science fictions: exposing fraud, bias. Negligence and hype in science. London, UK: The Bodley Head. Ioannidis JPA. 2012 Why science is not necessarily self-correcting. Pers. Psychol. Sci. 7, 645–654. (doi:10.1177/1745691612464056) Edwards MA, Roy S. 2017 Academic research in the 21st century: maintaining scientific integrity in a climate of perverse incentives and hypercompetition. Environ. Eng. Sci. 34, 51–61. (doi:10.1089/ees.2016.0223) Stewart S, Pennington CR, Silva G, Ballou N, Butler J, Dienes Z, Jay C, Rossit S, Samara A. 2022 Reforms to improve reproducibility and quality must be coordinated across the research ecosystem: the view from the UKRN Local Network Leads. BMC Research Notes. See: https://psyarxiv.com/xzfa2 Popper KR. 1963 Conjectures and refutations. London, UK: Routledge. Popper KR. 1979 Objective knowledge: an evolutionary approach. Oxford, UK: Oxford University Press. Flourishing Science Think Tank. 2022 Beyond kindness: a proposal for the flourishing of science and scientists. (doi:10.31231/osf.io/ 4zrmd) Nottourno MA. 2000 Science and the open society: the future of Karl Popper’s philosophy. Budapest, Hungary: Central University Press. Lloyd CER. 1970 Early Greek science: Thales to Aristotle. London, UK: Ghatto & Windus. 10. Mattern SP. 2013 The prince of medicine: Galen in the Roman empire. Oxford, UK: Oxford University Press. Ober J. 2015 The rise and fall of classical Greece (The Princeton history of the ancient world). Princeton, NJ: Princeton University Press. Al-Khalili J. 2012 The house of wisdom: how Arabic science saved ancient knowledge and gave us the Renaissance. London, UK: Penguin. 13. Manin B. 1997 The principles of representative government. Cambridge, UK: Cambridge University Press. 12. 11. 8. 7. 9. 14. 15. 16. 17. 18. Stasavage D. 2020 The decline and rise of democracy: a global history from antiquity to today. Princeton, NJ: Princeton University Press. Ferris T. 2011 The science of liberty: democracy, reason, and the laws of nature. New York, NY: Harper Collins. Needham J. 1969 The grand titration: science and society in east and west (China: history, philosopy, economic). London, UK: George Allen & Unwin. Lu J. 2011 Reassessing the Needham question: what forces impeded the development of modern science in China after the 15th century? Concord Rev. 21, 209–255. Graeber D, Wengrow D. 2021 The Dawn of everything: a new history of humanity. Dublin, Ireland: Allen Lane. 20. 19. Walshe M. 1995 The long discourses of the Buddha: a translation of the Digha Nikaya. Boston, MA: Wisdom Publications. Lovell M, Dienes Z. 2022 Minimal mindfulness of the world as an active control for a full mindfulness of mental states intervention: a registered report and pilot study, in principle acceptance of version 4 by Peer Community in Registered Reports. https://osf.io/tx54k Recommendation See: https://rr. peercommunityin.org/articles/rec?id=45 Saxonhouse AH. 2006 Free speech and democracy in ancient Athens. Cambridge, UK: Cambridge University Press. 21. 23. 22. Manville B, Ober J. 2003 A company of citizens: what the world’s first democracy teaches leaders about creating great organizations. Harvard, MA: Harvard Business Review Press. Cartledge P. 2016 Democracy: a life. Oxford, UK: Oxford University Press. Rhodes PJ. (translator). 2002 The Athenian constitution. London, UK: Penguin. Collini S. 2017 Speaking of universities. London, UK: Verso. Ferlie E, Musselin C, Andresani G. 2008 The steering of higher education systems: a 24. 25. 26. public management perspective. High. Educ. 56, 325. (doi:10.1007/s10734- 008-9125-5) 27. Martin BR. 2017 What’s happening to our universities? Prometheus 34, 7–24. 28. Willets D. 2017 A university education. Oxford, 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. UK: Oxford University Press. Erickson M. 2021 The UK higher education senior management survey: a statactivist response to managerialist governance. Stud. High. Educ. 46, 2134–2151. (doi:10.1080/ 03075079.2020.1712693) Honig D. 2020 Navigation by judgment: why and when top down management of foreign aid doesn’t work. Oxford, UK: Oxford University Press. Honig D. 2022 Managing for Motivation as Public Performance Improvement Strategy in Education & Far Beyond. See: https://bsc.cid. harvard.edu/publications/managing-motivation- public-performance-improvement-strategy- education-far-beyond. Brown R, Pehrson S. 2019 Group processes: dynamics within and between groups, 3rd edn. New Jersey, NJ: Wiley. Leaf MJ. 2018 An anthropology of academic governance and institutional democracy: the community of scholars in America. Cham, Switzerland: Palgrave Macmillan. Ginsberg B. 2011 The fall of the faculty: the rise of the all-administrative university and why It matters. New York, NY: Oxford University Press. De Boer H, File J. 2009 Higher Education Governance Reforms across Europe. See: https:// ris.utwente.nl/ws/portalfiles/portal/5147233/ c9hdb101+modern+project+report.pdf. Biggs JB. 2013 Changing universities: a memoir about academe in different places and times. Scarborough, Australia: Strictly Literary. Hil R. 2012 Whackademia: an insider’s account of the troubled university. Sydney, Australia: UNSW Press. Xu F, Wu L, Evans J. 2022 Flat teams drive scientific innovation. Proc. Natl Acad. Sci. USA 119, e2200927119. (doi:10.1073/pnas. 2200927119) 42. 43. 41. 40. 44. 39. Muller JZ. 2018 The tyranny of metrics. Princeton, NJ: Princeton University Press. Vann MG. 2003 Of rats, rice, and race: the great Hanoi rat massacre, an episode in French colonial history. Fr. Colon. Hist. 4, 191–203. (doi:10.1353/fch.2003.0027). S2CID 143028274. Smaldino PE, McElreath R. 2016 The natural selection of bad science. R. Soc. Open Sci. 3, 3160384160384. (doi:10.1098/rsos.160384) Smaldino PE, Turner MA, Contreras Kallens PA. 2019 Correction to ‘Open science and modified funding lotteries can impede the natural selection of bad science’. R. Soc. Open Sci. 6, 6191249191249. (doi:10.1098/rsos.191249) Gross K, Bergstrom CT. 2019 Contest models highlight inherent inefficiencies of scientific funding competitions. PLoS Biol. 17, e3000065. (doi:10.1371/journal.pbio.3000065) Brette R. 2022 Le modèle managérial de la recherche - Critique et alternative [A critique of the managerial model of research]. Med. Sci. (Paris) 2022, 84–88. French. (doi:10.1051/medsci/ 2021247). English: See: http://romainbrette.fr/ WordPress3/wp-content/uploads/2022/01/A- critique-of-the-managerial-model-of-research.pdf Racimo F, Galtier N, De Herde V, Aubert Bonn N, Phillips B, Guillemaud T, Bourguet D. 2022 Ethical publishing: how do we get there? Zenodo 14, 15. (doi:10.5281/zenodo.6289488) Fanelli D. 2020 Pressure to publish: what effects do we see? In Gaming the metrics: misconduct and manipulation in academic research (eds M Biagioli, A Lippman). Cambridge, MA: MIT Press. Fang FC, Casadevall A. 2011 Retracted science and the retraction index. Infect. Immun. 79, 3855–3859. (doi:10.1128/IAI.05661-11) Fang FC, Steen RG, Casadevall A. 2012 Misconduct accounts for the majority of retracted scientific publications. Proc. Natl Acad. 48. 47. 45. 46. 49. 50. 51. 52. 53. 54. 55. Sci. USA 109, 17– 028–17 033. (doi:10.1073/ pnas.1212247109) Brembs B. 2018 Prestigious science journals struggle to reach even average reliability. Front. Hum. Neurosci. 12, 37. (doi:10.3389/fnhum. 2018.00037) Dienes Z. 2020 The inner workings of Registered Reports. PsyArXiv preprint. (doi:10.31234/osf.io/ yhp2a) Kepes S, Keener SK, McDaniel MA, Hartman NS. 2022 Questionable research practices among researchers in the most research-productive management programs. J. Org. Behav. 43, 1190–1208. (doi:10.1002/job.2623) Dougherty MR, Horne Z. 2022 Citation counts and journal impact factors do not capture some indicators of research quality in the behavioural and brain sciences. R. Soc. Open Sci. 9, 9220334220334. (doi:10.1098/rsos.220334) Shattock M, Horvath A. 2021 The governance of British higher education: the impact of governmental, financial and market pressures. Bloomsbury Higher Education Research. London, UK: Bloomsbury Academic. Landemore H. 2020 Open democracy: reinventing popular rule for the 21st century. Princeton, NJ: Princeton University Press. Jones G. 2020 10% less democracy: why you should trust elites a little more and the masses a little less. Stanford, CA: Stanford University Press. 56. Massimini M, Ferrarelli F, Huber R, Esser SK, Singh H, Tononi G. 2005 Breakdown of cortical effective connectivity during sleep. Science 309, 2228–2232. (doi:10.1126/science.1117256) Latora V, Marchiori M. 2001 Efficient behavior of small-world networks. Phys. Rev. Lett. 87, 198701. (doi:10.1103/PhysRevLett.87.198701) Fishkin JS. 2011 When the people speak: deliberative democracy and public consultation. Oxford, UK: Oxford University Press. 57. 58. 59. 60. 61. 63. 62. Fishkin JS. 2018 Democracy when the people are thinking: revitalizing our politics through public deliberation. Oxford, UK: Oxford University Press. Smith G. 2009 Democratic innovations: designing institutions for citizen participation. Cambridge, UK: Cambridge University Press. O’Flynn I. 2022 Deliberative democracy. Cambridge, UK: Polity Press. Abers R, King R, Votto D, Brandao I. 2018 Porto Alegre: participatory budgeting and the challenge of sustaining transformative change. Case study, World Resources Institute. See: https://www.wri.org/research/porto-alegre- participatory-budgeting-and-challenge- sustaining-transformative-change. Gastil J, Knobloch KR. 2020 Hope for democracy: how citizens can bring reason back into politics. Oxford, UK: Oxford University Press. 64. McCormick JP. 2011 Machiavellian democracy. Cambridge, UK: Cambridge University Press. Tangian A. 2020 Analytic theory of democracy: history, mathematics, and applications. Cham, Switzerland: Springer. Deutsch D. 2011 The beginning of infinity: explanations that transform the world. London, UK: Allen Lane. Popper KR. 1945/1974 The open society and its enemies: volume 1, Plato. London, UK: Routledge. Farrell DM, Suiter J. 2019 Reimagining democracy: lessons in deliberative democracy from the Irish frontline. Ithaca, NY: Cornell University Press. Fukuyama F. 2014 Political order and political decay: from the industrial revolution to the globalization of democracy. London, UK: Profile Books. 66. 65. 68. 67. 69. 70. McGregor D. 1960 The human side of enterprise, 71. annotated edn. New York, NY: McGraw Hill. Semler R. 1993 Maverick! The success story behind the world’s most unusual workplace. London, UK: Random House Business Books. 13 r o y a l s o c i e t y p u b i . l i s h n g o r g / j o u r n a l / r s o s R . S o c . O p e n S c i . 1 0 : 2 2 0 8 0 8
null
10.1038_s41556-023-01184-y.pdf
Data availability All data that support the findings of this study are available within the paper and its supplementary files. Sequencing data that support the findings of this study have been deposited in the Gene Expression Omni- bus under accession code GSE208072. Previously published RNA-seq data from BCC, SCC and normal EpdSCs that were re-analysed here are available under accession code GSE152487. Source data are provided with this paper. All other data supporting the findings of this study are available from the corresponding author on reasonable request. Code availability All bioinformatic analysis tools and pipelines used in this study are documented in the method section. Codes are available from the cor- responding author upon reasonable request.
Data availability All data that support the findings of this study are available within the paper and its supplementary files. Sequencing data that support the findings of this study have been deposited in the Gene Expression Omnibus under accession code GSE208072 . Previously published RNA-seq data from BCC, SCC and normal EpdSCs that were re-analysed here are available under accession code GSE152487 . Source data are provided with this paper. All other data supporting the findings of this study are available from the corresponding author on reasonable request. Code availability All bioinformatic analysis tools and pipelines used in this study are documented in the method section. Codes are available from the corresponding author upon reasonable request. Article https://doi.org/10.1038/s41556-023-01184-y Extended Data Fig. 1 | See next page for caption. Nature Cell Biology Article https://doi.org/10.1038/s41556-023-01184-y Materials availability. Will be provided upon request and available upon publication.
The pioneer factor SOX9 competes for epigenetic factors to switch stem cell fates https://doi.org/10.1038/s41556-023-01184-y Received: 8 January 2023 Accepted: 8 June 2023 Published online: 24 July 2023 Check for updates Yihao Yang1,7, Nicholas Gomez1,2,7, Nicole Infarinato1,3, Rene C. Adam1,4, Megan Sribour1, Inwha Baek1,5, Mélanie Laurin1,6 & Elaine Fuchs  1 During development, progenitors simultaneously activate one lineage while silencing another, a feature highly regulated in adult stem cells but derailed in cancers. Equipped to bind cognate motifs in closed chromatin, pioneer factors operate at these crossroads, but how they perform fate switching remains elusive. Here we tackle this question with SOX9, a master regulator that diverts embryonic epidermal stem cells (EpdSCs) into becoming hair follicle stem cells. By engineering mice to re-activate SOX9 in adult EpdSCs, we trigger fate switching. Combining epigenetic, proteomic and functional analyses, we interrogate the ensuing chromatin and transcriptional dynamics, slowed temporally by the mature EpdSC niche microenvironment. We show that as SOX9 binds and opens key hair follicle enhancers de novo in EpdSCs, it simultaneously recruits co-factors away from epidermal enhancers, which are silenced. Unhinged from its normal regulation, sustained SOX9 subsequently activates oncogenic transcriptional regulators that chart the path to cancers typified by constitutive SOX9 expression. From development to malignancy, cells face decisions of fate determina- tion. Governing the reprogramming from one fate to another, pioneer factors are transcription factors that can recognize and access their cognate binding motifs in compacted and repressed chromatin1. In vitro studies have shown that when a pioneer factor binds, it displaces the nucleosome, permitting the opening and remodelling of the chromatin landscape to change gene expression2,3. Recent studies have begun to uncover interactions of various pioneer factors with histone-modifying enzymes and members of the SWI/SNF chromatin remodelling complex2. However, the order of events in chromatin remodelling has remained elusive due to the rapid time frame of reprogramming in vitro where cells are outside local restraints of their tissue microenvironments. Even less clear is the role of pioneer factors in accomplishing the other side of fate switching, namely the silencing of a cell’s previous identity2. In this Article, seeking the answers to these enigmas, we focused on the SOX superfamily of context-specific pioneer factors, whose members are at the nexus of critical cell fate choices in embryonic development, tissue homeostasis and transition to malignancy4–7. In skin, SOX9 is first expressed when multipotent embryonic epidermal progenitors bifurcate to become SOX9+ hair follicle stem cells (HFSCs) and SOX9neg epidermal stem cells (EpdSCs)8–10. In the next step of hair follicle morphogenesis, SOX9+ HFSCs bifurcate again to form SOX9neg transit amplifying hair shaft progenitors. Basal cell carcinoma (BCC) formation from EpdSCs resembles the initial steps of embryonic hair follicle morphogenesis, but once re-activated, SOX9 is sustained, leading to invaginating follicle-like tumour masses that lack hair line- ages11–14. Here we recapitulated these reprogramming events by gene- rating mice in which we could inducibly re-activate and sustain SOX9 expression in adult EpdSCs. Not encountered in vitro or in embryogenesis, the mature tis- sue stem cell niche imposed physiological constraints that slowed SOX-mediated chromatin reprogramming. This enabled the unravelling 1Howard Hughes Medical Institute, Robin Chemers Neustein Laboratory of Mammalian Cell Biology and Development, The Rockefeller University, New York, NY, USA. 2Present address: Allen Institute for Cell Sciences, Seattle, WA, USA. 3Present address: PRECISIONscientia, Yardley, PA, USA. 4Present address: Regeneron Pharmaceuticals, Tarrytown, NY, USA. 5Present address: Kyung Hee University, Seoul, South Korea. 6Present address: CHU de Québec-Université Laval Research Center, Quebec City, Quebec, Canada. 7These authors contributed equally: Yihao Yang, Nicholas Gomez.  e-mail: fuchslb@rockefeller.edu Nature Cell Biology | Volume 25 | August 2023 | 1185–1195 1185 nature cell biologyArticle of sequential events that happen as SOX9 achieves a cell fate switch that when dysregulated later progresses to a tumourigenic state. By dissect- ing the temporal steps of epigenetic reprogramming, we show that SOX9 binds to closed chromatin at HFSC enhancers, where it recruits histone and chromatin modifiers to remodel and subsequently open chromatin for transcription. In doing so, SOX9 redistributes co-factors away from EpdSC enhancers, thereby silencing these genes indirectly but efficiently. Moreover, when the ability of SOX9 to bind DNA is abro- gated, it still silences, but when it cannot bind chromatin remodellers, the switch fails altogether. Together, our findings illuminate how fate switching can be achieved through the direct activating functions of a pioneer factor, which then unleashes transcriptional repression through indirect competition for epigenetic co-factors. We further show that SOX9 regulates downstream transcription factors to drive tumourigen- esis, which explains the delay in subsequent reprogramming events. Results SOX9 launches a transcriptional cascade towards BCC in EpdSC To interrogate SOX9 reprogramming in adult tissue stem cells, we engi- neered mice harbouring a MYC-epitope-tagged Sox9 transgene con- trolled by a tetracycline responsive enhancer and a minimal promoter (TRE-Sox9) (Extended Data Fig. 1a). After validating the specificity of transgene induction (Extended Data Fig. 1b), we bred selected mice to lines expressing the requisite tetracycline-inducible transcriptional activator (rtTA) driven by an epidermal (Krt14) promoter (Krt14-rtTA)15, and selected mice that induced MYC–SOX9 in EpdSCs at levels compa- rable to SOX9 in adult HFSCs (Extended Data Fig. 1c,d). Upon doxycycline (DOX) administration (D0), adult mice were monitored weekly thereafter (Fig. 1a). Within the first 2 weeks, mor- phology and differentiation seemed unaffected (Extended Data Fig. 1e). However, by week (W)1, nuclear SOX9 was detected in the EpdSCs of the innermost (basal) epidermal layer (Fig. 1b). By W2, a rise in proliferation was detected, similar to that seen when SOX9 is naturally induced in embryonic epidermis10 (Extended Data Fig. 1f). Between W2 and W6, de novo invaginations began to grow between native HFs (Fig. 1b and Extended Data Fig. 1e). As differentiation defects necessitated killing mice by W6, we monitored later events in SOX9 reprogramming by engrafting neonatal Krt14-rtTA;TRE-Sox9 skin patches onto immunocompromised mice. Once normal skin patho- logy was restored (21 days after grafting), we induced SOX9 (Fig. 1a). By W12 post-induction, invaginations were dysplastic, resembling morphological and molecular (SOX9, EpCAM and KRT6) features of human BCCs (Extended Data Fig. 1g). To gain further insights, we profiled the transcriptomic changes occurring in EpdSCs during SOX9-driven reprogramming. At each timepoint, two biological replicates of RNA sequencing (RNA-seq) were performed on fluorescence-activated cell sorting (FACS)-purified EpdSCs from Krt14-rtTA;TRE-Sox9 skins (Extended Data Fig. 2a,b). By comparing transcriptomes across time, we identified the significantly variable genes (P < 0.05) along the reprogramming cascade (Fig. 1c and Supplementary Table 1). As expected, the D0 population displayed the hallmark signature of EpdSCs, replete with mRNAs encoding epidermal master regulator transcription factors, TRP63 and GATA3, key signal- ling pathways (NOTCH and EGFR), and epidermal structural proteins. Despite few morphological changes within 2 weeks after induc- tion, SOX9+ EpdSCs displayed dramatic transcriptional changes, mimicking transcriptional changes that occur when embryonic skin progenitors naturally induce SOX9 and divert from an epidermal to hair follicle fate10. Thus, epidermal genes were markedly suppressed, while classical markers of the embryonic hair bud and adult hair follicle outer root sheath (ORS) were upregulated, as supported by gene set enrichment analysis (GSEA) (Fig. 1c,d). The kinetics of these reprogram- ming events in adult EpdSCs, however, was markedly slower in the adult than in embryonic skin or in cultured cells, suggestive of the need to override the constraints of the mature epidermal niche. As in BCC development, progression to mature HFs did not hap- pen, in agreement with the need for Sox9 downregulation for HFSCs to generate the hair and its channel10,16. However, with sustained SOX9 expression, the transcriptional changes continued, and by W6–12, cancer-associated features appeared. At W12, GSEA revealed a strong correlation, both up and down, with the molecular signature of BCC compared with normal skin14,17 (Fig. 1c,d). Although the similarities in gene expression were strongest at late stages, they surfaced as early as W2, that is, before overt phenotypic changes, and clearly favoured a BCC versus squamous cell carcinoma (SCC) signature (Extended Data Fig. 2c,d). SOX9 is a bona fide pioneer factor To understand how SOX9 acts as a master regulator to induce these transcriptional dynamics, we began by performing CUT&RUN (cleav- age under targets and release using nuclease; hereafter termed CNR) sequencing18,19 to temporally assay the binding of SOX9 to chromatin, and assay for transposase-accessible chromatin with high-throughput sequencing (ATAC-seq)20,21 to interrogate chromatin accessibility dur- ing reprogramming (Fig. 2a,b). Biological replicates were concord- ant, and the SOX motif was most enriched in our SOX9 CNR peak sets (Extended Data Fig. 3a–c). SOX9 binding to chromatin occurred rapidly within W1 and before the rise in proliferation. In contrast, the increase in accessibil- ity at SOX9-binding sites occurred between W1 and W2, indicating that SOX9 can bind to closed chromatin (Fig. 2a). In fact, of all the SOX9 CNR peaks pooled from W1 to W12, nearly 30% were situated within closed chromatin at D0 (Fig. 2c). Moreover, by W2, nucleo- some occupancy was lost at these sites as measured by histone H3 (Fig. 2c). Additionally, these SOX9-bound opening peaks displayed a time-dependent decrease in CNR fragment length. These features are hallmarks of nucleosome displacement and pioneer factor activ- ity22, providing compelling evidence that SOX9 in skin EpdSCs binds to its cognate motifs within closed chromatin, and subsequently perturbs nucleosomes. SOX9 induces global chromatin changes at distal enhancers Since SOX9 bound to closed chromatin at W1, and presumptive nucleo- some loss occurred soon thereafter, these events seemed unlikely to account fully for the tumourigenic transcriptional dynamics (Fig. 1c). Probing deeper, we examined how ATAC peaks and their associated genes changed over time. Principal component analysis (PCA) of chromatin accessibility showed clustering according to times post-SOX9 induction (Fig. 2d). D0 and W1 samples clustered closely, W2 constituted an intermediary, and later timepoints (W6 and W12) made a second cluster. Comparative analyses across all timepoints revealed that many ATAC peaks were shared across samples, reflective of housekeeping genes and/or genes common to both EpdSCs and HFSCs (for example, Krt5) (Fig. 2e). By contrast, other ATAC peaks exhibited strikingly dynamic behaviour (for example, WNT-target Ctnnb1), indicative of SOX9-induced temporal chromatin remodelling. These dynamic changes were particularly striking at W2 after SOX9 induction (Fig. 2e). Of the peaks that opened by W2, many persisted thereafter. Upon binning peaks as either static (present at all timepoints, n = 38,079) or dynamic (absent in at least one timepoint, n = 62,626), it was clear that dynamic peaks were substantially more enriched in distal intergenic regions than static peaks (Fig. 2f and Extended Data Fig. 3e), suggesting a special role for SOX9 in eliciting chromatin changes at enhancers. Direct and indirect chromatin remodelling induced by SOX9 K-means clustering of the dynamic ATAC peaks resolved the temporal changes in chromatin accessibility following SOX9 induction. Although more than 10,000 peaks (cluster C4) opened at later timepoints (Fig. 3a), Nature Cell Biology | Volume 25 | August 2023 | 1185–1195 1186 Articlehttps://doi.org/10.1038/s41556-023-01184-y a Temporal Sox9 induction in EdpSCs Krt14-rtTA;TRE-Sox9 Age to P21 Graft P0 skin to Nude mice then age to P21 P0 c D0 W1 W2 W6 W12 Samples are collected at indicated timepoints D0 W1 W2 W6 Mouse dies Inject DOX Skin graft DOX feed b SOX9 DAPI D0 W1 W2 W6 Epidermis Dermis W12 Grafts are collected only at W12 W12 d GSEA confirms SOX9-induced EpdSC fate switching Hair placode Differential gene expression (P < 0.05) SOX9+ placode versus SOX9neg placode W2 (early) versus D0 expression W2 (early) versus D0 expression e r o c s t n e m h c i r n E 0.4 0.3 0.2 0.1 0 –0.1 P < 0. 001 NES 1.32 e r o c s t n e m h c i r n E 0.1 0 –0.1 –0.2 –0.3 –0.4 P < 0.001 NES –1.39 Bach2 Lrp6 Dkk3 Gli2 Irx3 Macf1 Foxp4 Trpv4 Ccnd2 SOX9+ placode genes SOX9neg placode genes BCC lesion Differential gene expression (P < 0.05) BCC cells versus Normal EpdSCs W12 (late) versus D0 expression W12 (late) versus D0 expression e r o c s t n e m h c i r n E 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 P < 0.001 NES 1.92 e r o c s t n e m h c i r n E 0.1 0 –0.1 –0.2 –0.3 –0.4 –0.5 –0.6 P < 0.001 NES –0.61 Gata3 Trp63 Cdkn1a Klf5 Jun Wnt7a Egr1 Tgfbr2 Bmp2 Jag2 Vdr Tcf7l2 Nfix Nfib Nfatc1 Lgr4 Sox4 Foxc1 Krt17 Runx3 Runx2 Runx1 Lgr5 Zeb1 Wnt5a Vim Tnc Sox18 Snai3 Pthlh Krt16 Fstl1 Elk3 Cd200 –2.00 0 2.00 BCC upregulated genes BCC downregulated genes Fig. 1 | In EpdSCs of adult skin, sustained SOX9 re-activation silences the epidermal while activating the hair follicle and subsequently BCC fates. a, Schematic of SOX9 induction in EpdSCs of adult and post-engrafted skin. b, Immunofluorescence reveals the appearance of nuclear SOX9 in EpdSCs after DOX induction. Dotted lines denote epidermal–dermal borders. Scale bars, 50 μm. c, Heat map of temporal RNA-seq data shows significantly variable genes (DESeq2 Wald test, adjusted P value <0.05) across combined independent replicates (r > 0.94) of each of five indicated timepoints. Hierarchical clustering revealed five distinct patterns, shown as coloured bars on the right with representative transcripts indicative of specific fates. d, GSEA of W2 versus D0 SOX9-induced expression changes in adult EpdSCs compared with SOX9+ and SOX9neg wild-type placode gene signatures from embryonic day E15.5 (top) (Kolmogorov–Smirnov test, P < 0.001 for both gene sets), and W12 versus D0 SOX9-induced EpdSC expression changes compared to BCC upregulated and downregulated signatures (bottom) (Kolmogorov–Smirnov test, P < 0.001 for both gene sets). NES, normalized enrichment score. the most substantial changes occurred between W1 and W2. Using Genomic Regions Enrichment of Annotations Tool (GREAT), we assessed the biological pathways associated with each of the six clusters. C1 and C6 closed within the first 2 weeks and were enriched for pathways with direct importance to EpdSCs (Fig. 3b and Supplemen- tary Table 2). By contrast, C2 and C5 markedly increased their chro- matin accessibility during this time and were enriched for hair follicle development and SHH signalling, key not only in stimulating ORS/ HFSC lineage proliferation23–25 but also in driving BCCs11,12,14 (Fig. 3b). Also notable was a downregulation of AP1, EGFR and TGFβ signalling pathways, which are known to be elevated in BCCs that develop resist- ance to SHH inhibitors26. Many of the pathways enriched in the C2/C5 clusters were also implicated in other cancers previously associated with SOX9 expression27–29. The role of SOX9 in activating the ORS/HFSC fate appeared to be direct, as chromatin that opened by W2 was associated with hair fol- licle development and displayed both SOX motifs and SOX9 binding (Fig. 3b,c). These peaks also persisted in an accessible state and were still prominent from W6 to W12 (Fig. 3a). By contrast, late-opening C4-associated genes were not related to HFSC fate. Many of their peaks only entered a more accessible state sometime after W2, coincident with the late BCC gene induction per our transcriptome analysis (Fig. 1c,d). Intriguingly, these peaks were not enriched for SOX but rather RUNX, AP1 and NF-κB motifs (Fig. 3c). Moreover, whereas the Sox9 transgene was induced by W1, Runx1–Runx3 in particular were high- est at W2-W6 (Fig. 1b,c). Given that RUNX1 suppresses basosquamous features in therapeutic-resistant human BCCs30, the sustained Runx expression underscored a BCC-like rather than SCC-like phenotype. Nature Cell Biology | Volume 25 | August 2023 | 1185–1195 1187 Articlehttps://doi.org/10.1038/s41556-023-01184-y a b s k a e p R N C 9 X O S l l A d Schematic for in vivo temporal chromatin profiling Homeostasis Upregulated HF fate BCC-like downgrowths c EpdSC Time on DOX D0 W1 W2 W6 W12 Isolate EpdSC via FACS SOX9 binds to closed chromatin before acquiring accessibility SOX9 CNR IgG D0 W1 W2 W6 W12 ATAC D0 W1 W2 W6 W12 0 300 600 ±2 kb from SOX9 peak centre 0 1 2 PCA plot at ATAC-seq samples l a n g i s 3 H 0.020 0.015 0.010 0.005 10 5 0 –5 –10 e c n a i r a v % 6 : 2 C P Early Late Mid –40 –20 0 20 PC1: 88% variance D0 W1 W2 W6 W12 f Dynamic peaks are more distal to TSS than static peaks Static Dynamic y t i s n e d e v i t a l u m u C 1.0 0.8 0.6 0.4 0.2 0 0 5 10 15 log2(distance to TSS) 20 SOX9-mediated nucleosome displacement and chromatin relaxation D0 ATAC peaks W1–W12 SOX9 peaks 29,563 27,771 9,815 Opening SOX9 peaks Opening SOX9 peaks Shuffled regions Transcription factor pA-MNase Primary Ab Compact nucleosome Longer CNR fragments Relaxed nucleosome ) p b ( h t g n e l t n e m g a r f R N C 9 X O S 150 100 50 0 D0 W2 Shorter CNR fragments W1 W2 W6 W12 –2 0 2 –2 0 2 Distance from SOX9 peak centre (kb) e ) 3 0 1 × ( s k a e p n i t c e s r e t n i f o . o N 40 38,079 7,954 8,093 6,843 5,272 4,600 30 20 10 0 D0 W1 W2 W6 W12 D0 W1 W2 W6 W12 D0 W1 W2 W6 W12 2 kb 1 kb Ctnnb1 Krt5 3,321 2,531 1,516 1,349 859 660 537 320 200 Static Dynamic Fig. 2 | Upon induction, SOX9 opens chromatin at enhancers by evicting the nucleosome at its binding site and remodelling the flanking chromatin. a, Schematic of morphological changes that occur temporally after DOX. Back skins were collected at indicated timepoints and subjected to EpdSC FACS purification followed by SOX9 CNR landscaping and ATAC-seq. b, Heat map of IgG control CNR (blue), SOX9 CNR (blue) and ATAC-seq (orange) signals at all SOX9- bound peaks across indicated timepoints. Peaks are arranged along the vertical axis on the basis of their accessibility at W1. Note that, within the cohort of peaks in the bottom half along this axis, SOX9 binding occurred by W1, while their ATAC- seq landscape did not change until the following week. c, Top left: Venn diagrams show that, although SOX9 binds to many pre-existing accessible chromatin peaks, 9,815 SOX9 peaks open de novo. MINT-ChIP for histone H3 (H3) coupled with SOX9 CNR reveals that, by W2 post-induction, SOX9 binds to these previously closed peaks, concomitant with displacement of histone H3 at the SOX9-bound site. Right: schematic and data showing that CNR fragment length shortens between W1 and W2, correlating with SOX9 binding and nucleosome eviction. n = 2 biological replicates. Box plot is centred at median and bound by first and the third quartile, and whiskers extend to 1.5 times interquartile range (IQR) on both ends. d, PCA of ATAC-seq duplicate samples over the five timepoints. e, Upset plot of ATAC-seq peaks for each timepoint following SOX9 induction. Shared regions between timepoints are indicated by the dots and connecting lines. Grey inset shows a genome browser track of the generic skin stem cell Krt5 locus as an example of a static gene region. Blue inset depicts a dynamic region of the WNT-target HFSC gene Ctnnb1 that is open by W2 and remains open across W6 and W12 timepoints. f, Empirical cumulative distribution plot of dynamic (blue) and static peaks (grey) and their density relative to the transcription start site (TSS) of the nearest gene. Note that dynamic peaks are primarily associated with distal regions, typically encompassing enhancers. We also performed temporal motif analysis with ChromVAR31, which considers both enrichment and chromatin accessibility vari- ability at each motif. In addition to SOX, AP1(FOS/JUN), GATA and RUNX were top variable motifs. To learn how motif accessibility varied over time, we plotted accessibility deviation scores for each timepoint and compared them with a motif (TBX) that showed no temporal variability. Agreeing with motif enrichments in C4, the RUNX motifs continued to gain accessibility from W2 to W6 (Fig. 3c and Extended Data Fig. 3f). Delving deeper, the Runx1 gene locus was closed at D0, but within W1 after induction, the locus revealed SOX9 binding at multiple sites (Fig. 3d). Since the dynamic peaks were enriched at distal intergenic regions (Fig. 2f), we performed multiplexed T7-indexed chromatin Nature Cell Biology | Volume 25 | August 2023 | 1185–1195 1188 Articlehttps://doi.org/10.1038/s41556-023-01184-y a Dynamic peaks K-means b Pathway enrichment in clusters c Motif enrichment in clusters Gene hits 30 60 90 120 −log(Padj) 90 60 30 C1: 9,298 C2: 7,398 C4: 10,740 Skin development Epidermis development Hair follicle development Cell junction assembly Glioma TGFβ pathway SMAD2/3 signalling Small-cell lung cancer Pancreatic cancer C5: 8,428 Chronic myeloid leukaemia C3: 18,447 C6: 8,345 Colorectal cancer Hedgehog signalling Integrin signalling VEGFR signalling mTOR signalling Protein translation 0.05 P value 0 0.1 STAT (STAT) NFKB (RHR) SOX (HMG) RUNX (Runt) AP1/FOS/JUN (bZIP) CTCF (Zf) REST (Zf) HNF/POU (homeodomain) GATA (Zf) MEF (MADS) C1 C2 C3 C4 C5 C6 Clusters C1 C2 C3 C4 C5 C6 Cluster e RUNX1 DAPI D0 W6 W12 f y t i l i b i s s e c c a n i t a m o r h C g Footprinting at C4 late-opening peaks RUNX 0.26 0.01 n = 420 W12 W2 y t i l i b i s s e c c a n i t a m o r h C 0.15 SOX 0.02 n = 939 W12 W2 –100 bp 0 100 bp –100 bp 0 100 bp Distance from motif footprint centre Distance from motif footprint centre GO terms of late opening RUNX footprints clustering D0 W1 W2 W6 W12 ATAC signal 0.5 1.5 2.5 d C A T A R N C 9 X O S 1 e m 4 K 3 H D0 W1 W2 W6 W12 D0 W1 W2 W6 W12 D0 W1 W2 W6 W12 Runx1 0 Blood vessel development Vasculature development Stem cell proliferation Blood vessel morphogenesis Regulation of stem cell proliferation –log10(binomial P value) 8 12 4 16 Fig. 3 | SOX9 triggers an activated HFSC fate, while its target gene transcription factors trigger BCC transformation. a, Heat map of K-means clustering of ATAC peaks based on signal across time. Number of peaks in each cluster are indicated on the right. b, GO-term analysis of the genes associated with each peak of a cluster. Size of circle reflects the number of gene hits for each pathway, while the shade of red indicates adjusted P value (Padj) from binomial test. Note that cluster C1, whose chromatin is suppressed by W2, is enriched for epidermal genes, while C2, whose chromatin is accessible by W2, is enriched for hair follicle genes. C4, whose chromatin is accessible at later times, is the cluster most enriched for BCC genes. c, Motif enrichment analysis of each cluster with HOMER. Shade of red indicates P value with white representing ≥0.01 (binomial is used to calculate the significance). Note enrichment for: GATA motif in epidermal genes whose chromatin closes after SOX9 induction; SOX motif in hair follicle genes whose chromatin opens early; and the SOX9 target RUNX1 motif in genes whose chromatin opens weeks after SOX9 has bound. d, Chromatin landscape of the Runx1 gene locus, showing that pioneer factor SOX9 binds to this gene (blue), concomitant with early increases in H3K4me1 modifications across the locus (purple), while chromatin accessibility (orange) comes afterwards. Red boxes indicate regions that are bound by SOX9, primed at W1 and opened at W2. e, RUNX1 immunofluorescence reveals its absence in epidermal homeostasis, but its presence at late timepoints (W6 and W12) following SOX9 induction. Scale bars, 50 μm. f, RUNX and SOX ATAC footprint analyses at W12 and W2, showing a late increase in chromatin accessibility at the RUNX footprint at a stage when phenotypic BCC-like invaginations are prevalent. By comparison, SOX footprints appear by W2 and are retained thereafter. g, GO terms of genes whose putative enhancers have RUNX footprints and open at W6 and W12 (binomial is used to calculate the significance). immunoprecipitation (MINT-ChIP)32 on enhancer histone modifica- tion, H3K4me1, which also showed binding to this locus within W1. By contrast, accessibility did not occur until a week later (Fig. 3d). Immunofluorescence corroborated the delay in activating the Runx1 locus, and underscored its prominence at later stages of reprogram- ming (Fig. 3e). Finally, footprint analyses exposed an increase in chro- matin accessibility at RUNX footprints over the late-opening C4 ATAC peaks (Fig. 3f and Extended Data Fig. 3g). This contrasted with SOX footprints, which appeared early and then remained constant from W2 to W12 (Fig. 3f). Gene Ontology (GO)-term analyses of the genes associated with these late-opening RUNX footprints reflected stem cell proliferation, and angiogenesis, hallmarks of cancers (Fig. 3g). Together, these data imply that later changes involved not only SOX9 but also transcrip- tion factors that were directly targeted by SOX9, notably of the RUNX family, whose motifs were also enriched as noted above. Although we did not address whether RUNX factors operate as pioneer factors, the enrichment of RUNX motifs coincident with the rise in proliferation during BCC-like downgrowth raised the possibility that proliferation may enhance if not allow accessibility of these factors to chromatin. Nature Cell Biology | Volume 25 | August 2023 | 1185–1195 1189 Articlehttps://doi.org/10.1038/s41556-023-01184-y92,560 92,600 92,640 92,680 165 kbchr16:92,533,035–92,698,884(0–9.35)(0–9.35)(0–9.35)(0–9.35)(0–9.35)(0–12,147)(0–12,147)(0–12,147)(0–12,147)(0–12,147)(0–2.04)(0–2.04)(0–2.04)(0–2.04)(0–2.04) a H3K4me1 increases at SOX9-bound peaks before their opening b ) 1 + l a n g i s C A T A 2 g o ( l 1.0 0.5 0 ATAC SOX9 CNR ATAC H3K4me1 1.0 1.0 l 0.5 Identification of SOX9 co-factors by BioID 0.5 o g 2 ( S O X 9 C N R S g n a l i + 1 ) ) 1 + l a n g i s C A T A 2 g o ( l 0.5 0.25 l ( o g 2 H 3 K 4 m e 1 s i g n a l + 1 ) Primary EpdSC +DOX Mass spectrometry Collect cells +Biotin Krt14-rtTA;TRE-Sox9-BioID2 or Krt14-rtTA;TRE-NLS-GFP-BioID2 D0 D5 D6 D7 D0 W1 W2 W6 W12 D0 W1 W2 W6 W12 0 0 0 c SOX9-specific chromatin activation interactors d MLL methyltransferases are recruited to SOX9-bound peaks e MLL3/4 are recruited to newly occupied SOX9 binding sites Arid1a Arid1b Smarcd2 Ruvbl1 Taf9 Chromatin remodellers and transcription initiation factors Kmt2a (Mll1) Kmt2c (Mll3) Kmt2d (Mll4) Ep300 JunB Fosl2 Histone modifiers AP1 TFs Counts/106 200 400 800 ) 1 + l a n g i s R N C 4 / 3 L L M ( 2 g o l 8 6 4 2 MLL3/4 CNR D0 MLL3/4 peaks W1 MLL3/4 peaks Fraction of peaks bound by SOX9 17,963 19,109 12,518 D0 W1 W2 Name Motif SOX(HMG) CTCF(Zf) P value 1 × 10−27 1 × 10−2 0.4 0.3 0.2 0.1 0 D0 only W1 only MLL3/4 peaks Fig. 4 | SOX9 recruits co-factors to epigenetically prime and then remodel chromatin to an open, transcriptionally accessible state. a, Left: box plot comparing ATAC (orange) and SOX9 CNR (blue) signals at SOX9 peaks that transition from closed to open chromatin over time. Note that SOX9 binding increases markedly by W1, preceding chromatin accessibility at these sites by nearly a week. Right: box plot comparing ATAC (orange) and MINT-ChIP H3K4me1 (purple) signals at SOX9 peaks that open over time. Note that H3K4me1 follows the time course of SOX9 binding, again preceding chromatin accessibility. n = 2 biological replicates. Box plots are centred at median and bound by first and the third quartile, and whiskers extend to 1.5 times interquartile range (IQR) on both ends. b, Schematic of BioID2 proximity labelling of proteins that interact with SOX9 induced in cultured EpdSCs. c, Selected SOX9-interacting proteins detected with mass spectrometry and that fall into the top GO terms of chromatin remodellers of the SWI/SNF family; transcription initiation factors (TAF9), AP1 transcription factors; and enzymes that modify histones (MLL1, MLL3, MLL4 and p300). For full list, see Supplementary Table 3. Circle size corresponds to the strength of the hit as delineated at right. d, Box plot of MLL3/4 CNR signals reveals the appearance of MLL3/4 at SOX9-bound peaks following DOX. n = 2 biological replicates. Box plot is centred at median and bound by the first and the third quartile, and whiskers extend to 1.5 times IQR on both ends. e, Venn diagrams providing further evidence that the de novo MLL3/4 peaks that appear between D0 and W1 are highly enriched for bound SOX9 (binomial P = 1 × 10−27), whereas the 17963 MLL3/4 peaks lost at W1 do not show any significant motif enrichment over all D0 MLL3/4 peaks and are not associated with SOX9-bound sites, inconsistent with a repressor role for SOX9. SOX9 induces epigenetic remodelling before opening chromatin The substantial delay between H3K4me1 and SOX9 versus chromatin accessibility and transcription of the Runx1 gene led us to wonder whether this might be a general phenomenon of SOX9 reprogram- ming. To address this, we compared ATAC and histone modification signals over time at all opening SOX9 peaks (Fig. 2c). Correlating with SOX9 binding, H3K4me1 deposition occurred within W1 and levelled off thereafter (Fig. 4a), preceding chromatin accessibility changes at W2. By contrast, H3K27ac changes, while appearing by W1, were less robust and, in further contrast, continued to rise over time relative to SOX9 and H3K4me1 (Extended Data Fig. 4b). Notably, although the nucleosomes directly over SOX9-binding sites appear to have been evicted, H3K4me1 was strongly enhanced on nucleosomes flanking SOX9 (Extended Data Fig. 4c). Moreover, the domain size of H3K4me1 gradually increased from D0 to W2. This did not occur at static peaks, but rather specifically at SOX9-bound opening peaks (Extended Data Fig. 4d). Activating HFSC enhancers To understand how SOX9 directly activates HFSC enhancers, we began by identifying SOX9-interacting co-factors. To this end, we transduced Krt14-rtTA primary EpdSCs in vitro with TRE-Sox9-BioID2 and control TRE-GFP-NLS-BioID2 and then induced expression of each transgene using DOX (Extended Data Fig. 5a). One week later, biotinylated SOX9-interacting proteins were purified and analysed by mass spec- trometry (Fig. 4b). Biological replicates correlated highly and formed distinct clusters by PCA (Extended Data Fig. 5b–d). Fifty-eight proteins interacted with SOX9 relative to NLS-GFP EpdSCs (Supplementary Table 3). On the basis of protein function and GO-term analysis, SOX9-interacting proteins were mainly DNA and chromatin binders enriched in chromatin modi- fications and nuclear activity. Among the strongest SOX9 interactions were with core members of the SWI/SNF chromatin remodelling com- plex (ARID1a/b and SMARCD2), TATA box binding protein TAF9 (TFIID) required for RNA polymerase II-mediated induction of transcrip- tion, and AP1 (FOSL2 and JUNB) (Fig. 4c and Extended Data Fig. 5e,f). Histone modifiers typifying key active enhancers in developmental contexts were also featured. As SOX9-induced opening peaks were more enriched at enhancers over promoters (Extended Data Fig. 3e), we were intrigued to find modifiers of two histone marks enriched at active enhancers: Ep300, the acetyltransferase for H3K27ac, and MLL3/ MLL4, histone methyltransferases that not only can deposit H3K4me1 but possibly play additional emerging roles in enhancer activation33–35. Since we observed an increase in H3K4me1 at SOX9 targeted enhancers before H3K27ac or chromatin opening, we first focused on whether, as predicted, MLL3/4 are recruited by SOX9 to closed Nature Cell Biology | Volume 25 | August 2023 | 1185–1195 1190 Articlehttps://doi.org/10.1038/s41556-023-01184-y a Closing chromatin is not bound by SOX9 Opening ATAC peaks (C2 + C4 + C5) W1–W12 SOX9 peaks 18,273 8,293 (31%) 19,691 Closing ATAC peaks (C1 + C6) W1–W12 SOX9 peaks 17,113 530 (3%) 27,454 d TRE PGK MycTag-WT-SOX9 PuroR OR MycTag-∆TA-SOX9 Lack trans-activation (TA) domain OR MycTag-∆HMG-SOX9 Lack DNA-binding domain D0 D3 D6 Primary Krt14-rtTA EpdSC Puromycin selection + DOX Collect cells f ATAC peaks closed by WT SOX9 ATAC peaks closed by ∆HMG SOX9 D0 W1 W2 D0 W1 W2 D0 W1 W2 Truncated versions of SOX9 fail to open chromatin MYC-tag CNR WT-SOX9 ∆TA No DOX ATAC ∆HMG No DOX WT-SOX9 ∆TA ∆HMG b C A T A 9 X O S 4 / 3 L L M e s k a e p 9 X O S - T W l l A c Closing chromatin loses H3K4me1 and MLL3/4 ) 1 + l a n g i s 1 e m 4 K 3 H 2 g o ( l 0.5 0.4 0.3 0.2 0.1 0 1 , 2 0 7 p e a k s b o u n d b y b o t h W T a n d ∆ T A S O X 9 H3K4me1 8 MLL3/4 CNR 7 6 5 4 3 ) 1 + l a n g i s R N C 4 / 3 L L M ( 2 g o l D0 W1 W2 D0 W1 W2 ∆TA SOX9 fails to recruit MLL3/4 or deposit H3K4me1 ) 1 + l a n g i s R N C 4 / 3 L L M ( 10 2 g o l 5 MLL3/4 CNR H3K4me1 ) 1 + l a n g i s 1 e m 4 K 3 H 2 g o ( l 12 10 8 6 500 1,250 ± 2 kb from WT-SOX9 peak centre 0 1 32 4 ∆TA SOX9 No DOX WT SOX9 No DOX ∆TA SOX9 WT SOX9 8,644 6,296 (64%) 3,592 GO of peaks closed upon both WT or ∆HMG SOX9 induction –log10(binomial P value) 16 12 0 4 8 Negative regulation of haemopoiesis Regulation of apoptotic signalling pathway Epithelial cell differentiation Keratinocyte differentiation Regulation of epidermis development Adherens junction organization Fig. 5 | SOX9 achieves EpdSC fate silencing independent from DNA binding. a, Top: Venn diagram shows robust overlap between opening peak clusters (C2 + C4 + C5) and SOX9 peaks. Bottom: Venn diagram shows only 3% overlap between closing peaks (C1 + C6) and SOX9 peaks. b, ATAC, SOX9 CNR and MLL3/4 CNR tracks at the Gata3 locus, showing that, by W1 after SOX9 induction, MLL3/4 CNR peaks were diminished, and by W2, ATAC peaks closed, even though CNR showed no SOX9 binding in this region (red box). c, Box plots showing loss of MLL3/4 and H3K4me1 signal beginning at W1 post SOX9-induction and specifically at ATAC peaks that close by W2 (C1, C6). n = 2 biological replicates. d, Schematic of inducing of MYC-tagged wild-type (WT) or mutant versions of SOX9 in transduced Krt14-rtTA cultured EpdSCs. e, Profile plot and heat map showing MYC-tagged wild-type or variant SOX9 binding (blue) and accessibility (orange) before and after DOX. Peaks are sorted the same way across samples. Note that ΔHMG-SOX9 fails to bind DNA, and ΔTA-SOX9 binds only to the subset Vim P = 0.0041 ** P = 0.0042 ** g n o i s s e r p x e e v i t a l e R 300 200 100 0 Gata3 Trp63 P = 0.032 P = 0.027 P = 0.039 P = 0.042 * * * * n o i s s e r p x e e v i t a l e R 1.5 1.0 0.5 0 n o i s s e r p x e e v i t a l e R 1.5 1.0 0.5 0 WT ∆TA ∆HMG WT ∆TA ∆HMG WT ∆TA ∆HMG of SOX9 peaks that were already accessible before DOX. Both mutants of SOX9 failed to open chromatin de novo. Right: box plot comparing MLL3/4 CNR and H3K4me1 signals at the peaks that are bound by wild-type SOX9 and ΔTA-SOX9. Note that only wild-type SOX9 brought additional MLL3/4 and deposited more H3K4me1 to these peaks. n = 2 biological replicates. f, Venn diagram shows overlap between the ATAC peaks that are closed by wild-type SOX9 or ΔHMG- SOX9 induction. GO terms reveal that epidermal enhancers close upon wild-type or ΔHMG-SOX9 induction (binomial is used to calculate the significance). g, Quantitative PCR analysis of genes that are directly induced (Vim) or indirectly repressed (Gata3 and Trp63) by SOX9 in EpdSC cells. All the error bars are mean ± s.d. *P < 0.05 and **P < 0.01, two-tailed t-test. n = 3 biological replicates. All box plots are centred at median and bound by the first and third quartile, and whiskers extend to 1.5 times interquartile range (IQR) on both ends. chromatin in vivo. To validate the physical interaction, we exploited the MYC tag of SOX9 and performed co-immunoprecipitations on cultured EpdSC lysates with or without SOX9 induction, and then probed for MLL4. Given the large size of MLL4 (>500 kDa) and the likelihood of degradation, we used CRISPR–Cas9 to ablate Mll4 in these EpdSCs to ensure correct band identification (Extended Data Fig. 5g). After validation, we performed MLL3/4 CNR, reasoning that, if MLL3/4 recruitment to chromatin is regulated by SOX9, de novo MLL3/4 targeted sites should be enriched with SOX9 binding. A marked increase in MLL3/4 association with chromatin occurred between D0 and W2 at opening SOX9-bound enhancers (Fig. 4d). Moreover, upon analysing de novo MLL3/4 recruitment sites on chromatin at W1, we found that SOX motifs were significantly enriched (Fig. 4e). These data began to provide a clearer picture of how SOX9 functions as a pioneer factor, as it not only binds to closed chromatin but also recruits co-factors to epigenetically modify flanking histones. The data from Fig. 4c further hinted that SOX9 recruits the SWI/SNF complex to make the chromatin accessible for transcription. Nature Cell Biology | Volume 25 | August 2023 | 1185–1195 1191 Articlehttps://doi.org/10.1038/s41556-023-01184-y9,860 kb9,870 kb25 kb(0–4.48)(0–4.48)(0–4.48)(0–10,000)(0–10,000)(0–10,000)(0–633)(0–633)(0–633)Gata3Chromosome 2 a 0.8 0.6 0.4 0.2 0 300 200 100 0 C A T A R N C 4 / 3 L L M ∆HMG SOX9 redistributes MLL3/4 and closes chromatin No DOX WT-SOX9 ∆HMG b y t i l i b i s s e c c a n i t a m o r h C AP1 footprint dynamics after SOX9 induction 0.87 n = 1,866 0.19 D0 W1 y t i l i b i s s e c c a n i t a m o r h C 0.74 n = 8,137 0.19 W2 W1 ± 2 kb from centres of 6,296 closing peaks Distance from AP1 motif footprint centre –100 bp 0 100 bp –100 bp 0 100 bp AP1 in C1/C6 closing peaks AP1 in W1–W12 SOX9 CNR peaks d R N C ) 1 P A N U J ( R N C ) 1 P A N U J ( No DOX WT-SOX9 ∆HMG ± 2 kb from centres of closing peaks 800 600 400 200 0 4,000 3,000 2,000 1,000 0 ± 2 kb from centres of WT-SOX9 bound peaks g e 1 G R B a 1 D R A I R N C ) F N S / I W S ( R N C ) F N S / I W S ( 4,000 3,000 2,000 1,000 0 4,000 3,000 2,000 1,000 0 No DOX WT-SOX9 ∆HMG No DOX WT-SOX9 ∆HMG R N C ) F N S / I W S ( R N C ) F N S / I W S ( 1 G R B a 1 D R A I 800 600 400 200 0 600 400 200 0 No DOX WT-SOX9 AFOS AFOS + WT-SOX9 c 3 2 1 l a n g i s C A T A ± 2 kb from centres of WT-SOX9 bound peaks No DOX WT-SOX9 AFOS AFOS + WT-SOX9 ± 2 kb from centres of closing peaks No DOX WT-SOX9 ∆HMG ARID1a + WT-SOX9 0.8 0.6 0.4 0.2 l a n g i s C A T A f 0.8 0.6 0.4 0.2 3 2 1 0 0.8 0.6 0.4 0.2 0 ± 2 kb from centres of WT SOX9 bound peaks ± 2 kb from centres of closing peaks ± 2 kb from centres of closing peaks SOX9 re-activation SOX motif AP1 motif HFSC enhancers SOX9 H3K4me1 AP1 SOX9 AP1 H3K27ac EpdSCs MLL3/4 SWI/SNF AP1 TF Enhancer activation complex SWI/SNF MLL3/4 Co-factor hijacking MLL3/4 SWI/SNF AP1 TF W2 SWI/SNF MLL3/4 HFSC genes ↑ and SOX9-induced BCC TFs ↑ (for example, Runx1) BCC genes ↑ W2–12 EpdSC enhancers EpdSC genes ↓ EpdSC genes ↓↓ Homeostasis (D0) Primed (W1) Fate switched (W2–12) Fig. 6 | SOX9 redistributes chromatin remodelling co-factors to activate HFSC enhancers and silence EpdSC enhancers. a, Profile plots showing that ATAC and MLL3/4 CNR signals wane at EpdSC enhancers upon wild-type SOX9 and ΔHMG-SOX9 induction. Note that, when SOX9 cannot bind to DNA (ΔHMG), it still diminishes MLL3/4 at endogenous EpdSC enhancers and closes their chromatin. b, AP1 footprint analyses in closing ATAC peaks and SOX9 CNR peaks. Note that chromatin accessibility at AP1 footprints decreases over closing epidermal peaks by W1 and increases over opening SOX9-bound peaks by W2 (see also Fig. 5a). c, Top: profile plots and heat maps comparing ATAC signals at wild-type SOX9 (WT-SOX9) bound peaks in vitro. While WT-SOX9 can open chromatin, dominant negative FOS (AFOS) inhibits the opening as shown in the last column. Bottom: profile plots and heat maps comparing ATAC signals at closing peaks in vitro. AFOS phenocopies the indirect closing effect of SOX9 on AP1-associated epidermal enhancers. d, Top: profile plots showing that JUN CNR signals are reduced at EpdSC enhancers upon wild-type SOX9 and ΔHMG-SOX9 induction. Bottom: supportive of competition, wild-type SOX9 recruits AP1 to HFSC enhancers, while ΔHMG-SOX9, lacking the DNA binding domain, fails to do so. e, Profile plots showing that both BRG1 and ARID1a of the SWI/SNF complex behave similarly to AP1 and MLL3/4: they are recruited to the opening HFSC enhancers by only wild-type SOX9 (left), while both wild-type SOX9 and ΔHMG-SOX9 reduce SWI/SNF association with epidermal enhancers (right). The red dotted line denotes CNR levels of indicated target at closing peaks before DOX. f, Profile plots comparing ATAC signals at closing epidermal enhancers with wild-type SOX9, ΔHMG-SOX9 or ARID1a overexpression together with wild-type SOX9. Note that, when ARID1a is overexpressed, SOX9 fails to close epidermal enhancers. g, Working model for how SOX9 achieves cell fate switching: pioneer factor SOX9 indirectly silences the epidermal fate by competing away co-factors and other transcription factor (TFs) including AP1 from active EpdSC enhancers. Concomitantly, SOX9 binds directly to key hair follicle enhancers, bringing with it the hijacked chromatin remodelling machinery and activating the hair follicle fate. Also among the SOX9 target genes are transcription factors such as RUNX1, whose footprints appear to participate in the delayed activation of BCC cancer genes, leading to further fate switching to BCC, downstream of the EpdSC-to- HFSC fate transition. Nature Cell Biology | Volume 25 | August 2023 | 1185–1195 1192 Articlehttps://doi.org/10.1038/s41556-023-01184-y Silencing the epidermal fate Interestingly, while SOX9 binding was highly enriched within peaks that opened over time, it accounted for only 3% of peaks that closed over time (Fig. 5a). The differences were even more striking when we restricted our analysis to ATAC peaks changing over the first two weeks (Extended Data Fig. 6a). These findings were consistent with our SOX9 interactome, which was dominated by chromatin-activating remodellers. Moreover, in contrast to the hair-follicle-associated GO terms prominent in W2 SOX9-bound opening peaks, Epd-associated GO-terms were featured among closing peaks that were not bound by SOX9 (Extended Data Fig. 6b). We therefore hypothesized that SOX9 silences epidermal fate indirectly. To further understand how epidermal fate is silenced, we were intrigued by GATA factors, whose motif was markedly enriched in ATAC peaks (C1, C6) that closed within the first 2 weeks after SOX9 induc- tion and whose transcription factor footprint declined upon SOX9 induction (Extended Data Fig. 6c). GATAs surfaced upon analysing the transcription factors expressed by EpdSCs and whose motifs are highly enriched in closing chromatin (Extended Data Fig. 6d). GATA3 transcript and protein expression also declined concomitantly with the closure of GATA motifs (Extended Data Fig. 6e). The Gata3 gene locus also lost chromatin accessibility by W2, but the decline happened only at non-SOX9-bound peaks. The near- est SOX9-bound enhancer was >30 kb from the Gata3 gene body, and like several other weaker peaks, this site was already open and MLL3/4-bound before SOX9 was induced. Notably, subsequent SOX9 binding had little or no effect on its status (Extended Data Fig. 6f). These findings suggest that the role of SOX9 in silencing epidermal fate is at least in part indirect. Moreover, the result appeared to be physiologically relevant as genes downregulated in SOX9+ embryonic skin progenitors were also silenced when SOX9 was induced in adult EpdSCs (Fig. 1d)36. MLL3/4 presence over opening SOX9-dependent enhancer peaks was robust by W2, as was H3K4me1 modification (Fig. 4a,d). By contrast, the >6,000 SOX9-independent enhancer peaks that closed during this time displayed plummeting MLL3/4 association and a more gradual loss of H3K4me1 (Extended Data Fig. 7a). These findings raised the tanta- lizing possibility that, in binding to nucleosomes at HFSC-enhancers, SOX9 might be recruiting co-factors including MLL3/4 away from active EpdSC enhancers. To test this hypothesis, we engineered DOX-inducible MYC-tagged wild-type and mutant forms of SOX9 that lacked either the transacti- vation (ΔTA) domain or the DNA binding (ΔHMG) domain (Fig. 5d and Extended Data Fig. 7b). In the transduced primary EpdSCs, immuno- fluorescence levels of three versions of SOX9 were comparable, and the ectopically expressed proteins were of the expected size (Extended Data Fig. 7c,d). Additionally, as judged by co-immunoprecipitation, only wild-type SOX9 and ΔHMG-SOX9, but not ΔTA-SOX9, associated with MLL4, consistent with the inability of ΔTA to interact with chro- matin remodellers (Extended Data Fig. 7e). By using CNR with a MYC-tag antibody recognizing all three SOX9 variants equivalently, we verified that wild-type SOX9 and the ΔTA-SOX9 mutant, but not ΔHMG-SOX9, bound to DNA (Fig. 5e). Inter- estingly, without the TA domain to interact with co-factors, ΔTA-SOX9 only bound to chromatin that was already accessible in EpdSCs. Con- sistent with this result, the 1,207 peaks that were open before DOX and bound by ΔTA-SOX9 did not show MLL3/4 recruitment nor did they show H3K4me1 modification (shown at right). Additionally, and in contrast to wild-type SOX9, ΔTA-SOX9 failed to stably bind to closed chromatin of HFSC enhancers, indicating that, without binding to co-factors, SOX9 lost the defining feature of pioneer factors. Although ΔHMG-SOX9 did not bind DNA, it had a striking effect on chromatin accessibility. Nearly 10,000 ATAC peaks closed and >8,000 peaks opened upon induction (Extended Data Fig. 7f). As this mutant was unable to bind DNA, it was not surprising to see that the GO-term profile of the opening peaks was dramatically different than that of wild-type SOX9 (Extended Data Fig. 7g). Rather than HFSC features, the changes were more reflective of a stressed state. By contrast, in the ATAC peaks that closed in response to ΔHMG-SOX9, 64% of them were also closed by wild-type SOX9, and the GO terms corresponded to the same EpdSC genes indirectly silenced by wild-type SOX9 (Fig. 5f,g). Competition for SOX9-interacting chromatin remodellers Consistent with the hypothesis that SOX9 closes chromatin by compet- ing for and redistributing co-factors, MLL3/4 CNR signal diminished over EpdSC enhancers upon ΔHMG-SOX9 induction (Fig. 6a). Prob- ing deeper, we turned to AP1 transcription factors, which surfaced in our SOX9 interactome. In agreement with the dynamics observed for MLL3/4, footprint analysis in vivo revealed that AP1 binding decreased in closing non-SOX9-bound epidermal enhancers and increased in SOX9-bound chromatin (Fig. 6b). Moreover, in these SOX9-bound opening peaks, SOX and AP1 motifs were mostly found within one-nucleosome distance, supporting a role for SOX9 in targeting AP1 transcription factors to their canonical binding sites upon opening hair follicle enhancers (Extended Data Fig. 7h). Notably, motif analyses revealed the presence of AP1-binding sites in both closing and opening enhancers (Extended Data Fig. 3f), suggesting that the interaction between SOX9 and AP1 may be function- ally important for both opening SOX9+ HFSC enhancers and closing SOX9neg EpdSC enhancers. To test the possibility that enhancers might be competing for AP1 binding, we used the strategy delineated in Fig. 5d to induce AFOS, a dominant negative version of c-FOS that can heter- odimerize with AP1 transcription factors and block their binding to DNA37,38. We performed these experiments in the presence and absence of wild-type SOX9, and then carried out ATAC-seq. In the peaks that were bound by wild-type SOX9, AFOS clearly interfered with the opening of the HFSC enhancers, while also phenocopying the closing effects of SOX9 at EpdSC enhancers when expressed alone (Fig. 6c). Our data thus far suggested that, like MLL3/4, AP1 transcription factors function on both sides of the fate coin. To test whether other members of the interactome are targets for this putative competition for SOX9-binding partners, we focused on AP1 transcription factors and the SWI/SNF complex. After first validating their association with SOX9 (Extended Data Fig. 7i,j), we performed CNR analysis. We found that induction of wild-type SOX9 resulted in increased JUN(AP1) binding at SOX9-bound peaks, and decreased JUN binding at closing epidermal peaks. Notably, while ΔHMG-SOX9 failed to recruit JUN and open hair follicle enhancers, it still diminished JUN binding at closing epidermal peaks (Fig. 6d). Similarly, when we performed CNR on both structural (ARID1a) and enzymatic (BRG1) members of the SWI/SNF complex, we observed a decline in their association with epidermal enhancers when either wild-type SOX9 or ΔHMG-SOX9 were induced, but an increased association with SOX9-bound peaks only after wild-type SOX9 and not ΔHMG-SOX9 induction (Fig. 6e). Together, these data suggest that SOX9 orchestrates the redistribu- tion of transcription factors and epigenetic co-factors that are shared by the enhancers of both cell fates. Moreover, this competition appeared to be predicated in part on limiting levels of chromatin remodelling factors, as when we overexpressed ARID1a in the presence of SOX9, epidermal enhancers were rescued from closing (Fig. 6f and Extended Data Fig. 7i,j). Discussion Elegant studies by the Zaret lab launched the field of pioneer factors, now examined in various fate-switching scenarios and distinguished by their ability to bind their sequence motifs within closed chromatin1,2. However, the precise sequence of nucleosome eviction, opening of surrounding chromatin, and reprogramming fate choices has been dif- ficult to unravel, particularly in in vitro settings, where fate choices lack constraints imposed by native tissue microenvironments. By exploiting the slowed kinetics of our in vivo reprogramming system, we discovered Nature Cell Biology | Volume 25 | August 2023 | 1185–1195 1193 Articlehttps://doi.org/10.1038/s41556-023-01184-y that SOX9 not only perturbs its target nucleosome but also recruits enzymes that modify the flanking enhancer nucleosomes. Like SOX9 binding itself, these features precede the subsequent chromatin open- ing to the transcriptional machinery. As our ΔTA-SOX9 studies imply, these dynamics appear to be achieved by SOX9 recruiting of chromatin remodelling factors such as AP1 TFs and the SWI/SNF complex39,40. It has generally been viewed that a pioneer factor can act either as a transcriptional activator or as a repressor through recruiting dif- ferent cohorts of co-activators or co-repressors1,2. At first glance, this notion seems well suited to nodes of lineage switching, where one fate is silenced while another is chosen. However, increasing evidence sug- gests that pioneer factors may bind and directly regulate the enhancers of only one lineage at the crossroads, leaving a conundrum as to how the other lineage becomes silenced to achieve the switch. Our findings showed clearly that EpdSC gene silencing occurs shortly after SOX9 induction, a timing that is at odds with the notion that SOX9 might induce transcriptional repressors that then sub- sequently silence epidermal genes. Moreover, in contrast to HFSC enhancers, many of which bind SOX9 and are opened de novo, EpdSC enhancers show a paucity of SOX9 binding and yet close rapidly upon SOX9 induction. Rather to prevailing notions, our findings favour a dual function model whereby a pioneer factor actively hijacks and redistributes shared co-factors to achieve cost-effective and coordinated fate switch- ing from one lineage to another (Fig. 6g). Thus, following SOX9 induc- tion in EpdSCs, MLL3/4 binding increased at SOX9-bound opening HFSC enhancers, while diminished at closing non-SOX9-bound EpdSC enhancers. Our studies with wild-type SOX9 and ΔHMG-SOX9 revealed that not only does SOX9 interact with MLL3/4, but also with a compen- dium of co-factors essential to activate enhancers, which include not only MLL3/4 but also AP1 and SWI/SNF complex. In closing, although direct repressive mechanisms independ- ent from chromatin accessibility are still formally possible, our data suggest that at least some chromatin remodellers that are generally required for enhancer activity are in short supply, thereby setting up the competition to achieve fate switching once a pioneer factor such as SOX9 is activated. By utilizing such a mechanism, cellular fate plasticity is minimized, while simultaneously expediting the shift in density of shared transcriptional regulators to genomic loci of new fate determinants. Finally, it is noteworthy that, to make tissue, stem cells must undergo a fate choice, which for SOX9+HFSCs, is achieved by down- regulating SOX9 (ref. 16). In our model as in BCC, SOX9 was constitutive and hence the choice to make hair was never made. Moreover, when left outside the instructive microenvironment of the quiescent hair follicle bulge niche, the proliferating cells with sustained SOX9 activated SOX9 downstream target transcription factor genes, such as those encoding the RUNX family, that secondarily drove further dynamic changes in the chromatin landscape. These findings begin to explain how and why in adult tissue stem cells, sustained re-activation of a pioneer factor involved in embryonic fate decisions frequently leads to cancer5,6,41. Online content Any methods, additional references, Nature Portfolio reporting sum- maries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author con- tributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41556-023-01184-y. References 1. Zaret, K. S. Pioneer transcription factors initiating gene network changes. Annu. Rev. Genet 54, 367–385 (2020). 2. Balsalobre, A. & Drouin, J. Pioneer factors as master regulators of the epigenome and cell fate. Nat. Rev. Mol. Cell Biol. 23, 449–464 (2022). 3. Soufi, A. et al. Pioneer transcription factors target partial DNA motifs on nucleosomes to initiate reprogramming. Cell 161, 555–568 (2015). 4. Fuglerud, B. M. et al. SOX9 reprograms endothelial cells by altering the chromatin landscape. Nucleic Acids Res. 50, 8547–8565 (2022). 5. Grimm, D. et al. The role of SOX family members in solid tumours 6. and metastasis. Semin. Cancer Biol. 67, 122–153 (2020). Julian, L. M., McDonald, A. C. & Stanford, W. L. Direct reprogramming with SOX factors: masters of cell fate. Curr. Opin. Genet Dev. 46, 24–36 (2017). 7. Kamachi, Y. & Kondoh, H. Sox proteins: regulators of cell fate specification and differentiation. Development 140, 4129–4144 (2013). 8. Nowak, J. A., Polak, L., Pasolli, H. A. & Fuchs, E. Hair follicle stem cells are specified and function in early skin morphogenesis. Cell Stem Cell 3, 33–43 (2008). 9. Kadaja, M. et al. SOX9: a stem cell transcriptional regulator of secreted niche signaling factors. Genes Dev. 28, 328–341 (2014). 10. Ouspenskaia, T., Matos, I., Mertz, A. F., Fiore, V. F. & Fuchs, E. WNT-SHH antagonism specifies and expands stem cells prior to niche formation. Cell 164, 156–169 (2016). 11. Oro, A. E. et al. Basal cell carcinomas in mice overexpressing sonic hedgehog. Science 276, 817–821 (1997). 12. Vidal, V. P. et al. Sox9 is essential for outer root sheath differentiation and the formation of the hair stem cell compartment. Curr. Biol. 15, 1340–1351 (2005). 13. Youssef, K. K. et al. Identification of the cell lineage at the origin of basal cell carcinoma. Nat. Cell Biol. 12, 299–305 (2010). 14. Larsimont, J. C. et al. Sox9 controls self-renewal of oncogene targeted cells and links tumor initiation and invasion. Cell Stem Cell 17, 60–73 (2015). 15. Nguyen, H., Rendl, M. & Fuchs, E. Tcf3 governs stem cell features and represses cell fate determination in skin. Cell 127, 171–183 (2006). 16. Yang, H., Adam, R. C., Ge, Y., Hua, Z. L. & Fuchs, E. Epithelial– mesenchymal micro-niches govern stem cell lineage choices. Cell 169, 483–496 e413 (2017). 17. Fiore, V. F. et al. Mechanics of a multilayer epithelium instruct tumour architecture and function. Nature 585, 433–439 (2020). 18. Skene, P. J., Henikoff, J. G. & Henikoff, S. Targeted in situ genome-wide profiling with high efficiency for low cell numbers. Nat. Protoc. 13, 1006–1019 (2018). 19. Skene, P. J. & Henikoff, S. An efficient targeted nuclease strategy for high-resolution mapping of DNA binding sites. eLife 6, e21856 (2017). 20. Buenrostro, J. D., Giresi, P. G., Zaba, L. C., Chang, H. Y. & Greenleaf, W. J. Transposition of native chromatin for fast and sensitive epigenomic profiling of open chromatin, DNA-binding proteins and nucleosome position. Nat. Methods 10, 1213–1218 (2013). 21. Corces, M. R. et al. The chromatin accessibility landscape of primary human cancers. Science https://doi.org/10.1126/science. aav1898 (2018). 22. Meers, M. P., Janssens, D. H. & Henikoff, S. Pioneer factor- nucleosome binding events during differentiation are motif encoded. Mol. Cell 75, 562–575 e565 (2019). 23. Oro, A. E. & Scott, M. P. Splitting hairs: dissecting roles of signaling systems in epidermal development. Cell 95, 575–578 (1998). 24. Hsu, Y. C., Li, L. & Fuchs, E. Transit-amplifying cells orchestrate stem cell activity and tissue regeneration. Cell 157, 935–949 (2014). 25. Oshimori, N. & Fuchs, E. Paracrine TGF-β signaling counterbalances BMP-mediated repression in hair follicle stem cell activation. Cell Stem Cell 10, 63–75 (2012). Nature Cell Biology | Volume 25 | August 2023 | 1185–1195 1194 Articlehttps://doi.org/10.1038/s41556-023-01184-y 26. Kuonen, F. et al. c-FOS drives reversible basal to squamous cell carcinoma transition. Cell Rep. 37, 109774 (2021). 27. Wang, L. et al. Oncogenic role of SOX9 expression in human malignant glioma. Med. Oncol. 29, 3484–3490 (2012). 28. Zhou, C. H. et al. Clinical significance of SOX9 in human non-small cell lung cancer progression and overall patient survival. J. Exp. Clin. Cancer Res. 31, 18 (2012). 29. Matheu, A. et al. Oncogenicity of the developmental transcription factor Sox9. Cancer Res. 72, 1301–1315 (2012). 30. Haensel, D. et al. LY6D marks pre-existing resistant basosquamous tumor subpopulations. Nat. Commun. 13, 7520 (2022). 31. Schep, A. N., Wu, B., Buenrostro, J. D. & Greenleaf, W. J. chromVAR: inferring transcription-factor-associated accessibility from single-cell epigenomic data. Nat. Methods 14, 975–978 (2017). 32. van Galen, P. et al. A multiplexed system for quantitative comparisons of chromatin landscapes. Mol. Cell 61, 170–180 (2016). 33. Sze, C. C. & Shilatifard, A. MLL3/MLL4/COMPASS family on epigenetic regulation of enhancer function and cancer. Cold Spring Harb. Perspect. Med https://doi.org/10.1101/cshperspect. a026427 (2016). 34. Dorighi, K. M. et al. Mll3 and Mll4 facilitate enhancer RNA synthesis and transcription from promoters independently of H3K4 monomethylation. Mol. Cell 66, 568–576 e564 (2017). 35. Rao, R. C. & Dou, Y. L. Hijacked in cancer: the KMT2 (MLL) family of methyltransferases. Nat. Rev. Cancer 15, 334–346 (2015). 36. Nowak, J. A., Polak, L., Pasolli, H. A. & Fuchs, E. Hair follicle stem cells are specified and function in early skin morphogenesis. Cell Stem Cell 3, 33–43 (2008). 37. Ahn, S. et al. A dominant-negative inhibitor of CREB reveals that it is a general mediator of stimulus-dependent transcription of c-fos. Mol. Cell. Biol. 18, 967–977 (1998). 38. Olive, M. et al. A dominant negative to activation protein-1 (AP1) that abolishes DNA binding and inhibits oncogenesis. J. Biol. Chem. 272, 18586–18594 (1997). 39. Vierbuchen, T. et al. AP-1 transcription factors and the BAF complex mediate signal-dependent enhancer selection. Mol. Cell 68, 1067–1082 e1012 (2017). 40. Wolf, B. K. et al. Cooperation of chromatin remodeling SWI/ SNF complex and pioneer factor AP-1 shapes 3D enhancer landscapes. Nat. Struct. Mol. Biol. https://doi.org/10.1038/ s41594-022-00880-x (2022). 41. Kumar, P. & Mistri, T. K. Transcription factors in SOX family: potent regulators for cancer initiation and development in the human body. Semin. Cancer Biol. 67, 105–113 (2020). Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons. org/licenses/by/4.0/. © The Author(s) 2023 Nature Cell Biology | Volume 25 | August 2023 | 1185–1195 1195 Articlehttps://doi.org/10.1038/s41556-023-01184-y Methods Ethical regulation compliance All animals used in this study were maintained and bred under specific-pathogen-free conditions at the Comparative Bioscience Center at The Rockefeller University, which is an Association for Assess- ment and Accreditation of Laboratory Animal Care-accredited facility. All procedures were performed with the Institutional Animal Care and Use Committee-approved protocols (20012-H and 20066-H). Generating and handling TRE-Sox9 mice To generate the conditional SOX9 transgenic mice, the Sox9 coding sequence was cloned into the pTRE2 vector harbouring a DOX-inducible, minimal CMV2 promoter. A MYC tag was added to the N terminus of SOX9. Transgenic mice were generated as described previously42. The resulting TRE-Sox9 mice were then genotyped and crossed to Krt14-rtTA transgenic mice15 to allow for DOX-inducible expression of MYC–SOX9 specifically in skin epithelium. Primary cell isolation Primary Krt14-rtTA;TRE-Sox9 EpdSCs were isolated from newborn male pups (postnatal day 0, or P0) as described previously16,43. Briefly, mouse back skin was collected from P0 pups and treated with dispase (Gibco) overnight at 4 °C. Epidermis was manually separated from dermis and disassociated into a single-cell suspension. Epidermal cells were pas- saged and maintained in E-low calcium medium44 (0.05 mM CaCl2) at 37 °C with 7.5% CO2. DOX treatment A total of 0.1 mg of DOX (Sigma) in 100 μl phosphate-buffered saline (PBS) was administered by intraperitoneal injection to Krt14-rtTA;TRE-Sox9 and Krt14-rtTA-only mice at postnatal day P21, and the mice were thereafter were maintained on mouse chow containing 2 mg g−1 DOX throughout the experimental time course. Phenotypic mice were housed with at least one control littermate for adequate grooming. To maintain proper body fluid, 100 μl PBS was administered through intraperitoneal injection every other day after 4 weeks of SOX9 induction. For W12 samples, epi- dermis from the back skin of P0 Krt14-rtTA;TRE-Sox9 or Krt14-rtTA-only pups were grafted onto 6–8-week-old immunocompromised (Nude) female mice. Grafts were allowed to heal for 21 days, and DOX was admin- istered as above. For induction of SOX9 and its variants, AFOS and ARID1a in cultured cells, DOX was added to a final concentration of 1 μg ml−1 in E-low medium for BioID or SOX9 variant experiments. Images were collected and analysed with Fiji (ImageJ v.2.3.0). For the Human Atlas immunostaining, the following antibodies were used: SOX9 (CAB068240), EpCAM (CAB030012) and KRT6A (HPA061168). For cultured cells, cells were plated onto chamber slides (Thermo Fisher). At collection, cells were fixed with 4% paraformaldehyde for 10 min, and then washed three times with PBS at room temperature. After washing, the cells were blocked and stained with primary anti- bodies the same way as described above for sections with the follow- ing primary antibodies: HA-tag (rabbit, 1:1,000, Cell Signaling), GFP (chicken, 1:2,000, Fuchs Lab), RFP (rat, 1:1,000, ChromoTek), and MYC-tag (rabbit, 1:1,000, Cell Signaling). For 5′-ethynyl-2′ deoxyuridine (EdU) experiments, mice were injected IP with EdU (50 μg g−1 body weight) 2 h before analysis. Quan- tifications were performed by counting the number of EdU+ EpdSCs within the basal layer. For quantifying the SOX9 signal in the native ORS and the SOX9-induced epidermis, sections were stained with same SOX9 antibody concentration (1:5,000), and same laser intensity and exposure time were used to acquire images. From each sample, 100 cells were quantified with the multi-point tool in Fiji. Flow cytometry and cell sorting Krt14-rtTA;TRE-Sox9 and Krt14-rtTA-only male mice were used for FACS experiments to obtain maximal cell numbers and to control for varia- tion due to sex. Briefly, the whole back skins were first dissected from the mouse. After scraping off the fat tissues from the dermal side, the tissues were incubated in 0.25% trypsin/ethylenediaminetetraacetic acid (EDTA) (Gibco) for 45–60 min at 37 °C. After quenching the trypsin with cold FACS buffer (5% foetal bovine serum, 10 mM EDTA and 1 mM HEPES in PBS), the epidermal layer and HFs were scraped off the epi- dermal side of the skin. The tissues were mechanically separated and filtered through a 70 μm cell strainer (BD) into a single-cell suspension for immunolabelling. Single-cell suspensions were immunolabelled with antibodies: Ly6A/E-APCCy7 (1:500, BioLegend), CD49f-PECy7 1:1,000, BioLegend), CD34-Alexa660 (1:50 Invitrogen), CD45-biotin (1:200, BioLegend), CD31-biotin (1:200, BioLegend), CD140a-biotin (1:200, BioLegend), CD117-biotin (1:200, BioLegend), TruStain FcX for blocking (1:1,000, BioLegend) and streptavidin-FITC (1:1,000, BioLegend) in 300 μl of FACS buffer. Stained cells were washed and resuspended with FACS buffer with 100 ng ml−1 DAPI before analysis or sorting. EpdSCs were collected using an Aria Cell Sorters (BD Biosciences) with BD FACSDiva (v. 8.0) into either FACS buffer for genomic experi- ments or TRIzol LS (Invitrogen) for RNA extraction. Immunofluorescence Mouse back skin was fixed in 4% paraformaldehyde at room tempera- ture for 15 min, and then washed three times with PBS for 15 min at 4 °C. Following PBS washes, samples were dehydrated in 30% sucrose in PBS 4 °C overnight. The dehydrated samples were then embedded in optimal cutting temperature (OCT) medium (VWR) and frozen on dry ice. Cryo- sections (16 μm) were blocked in immunofluorescence buffer containing 0.3% Triton X-100, 2.5% normal donkey serum, 2.5% normal goat serum, 1% bovine serum albumin and 1% gelatin in PBS for 1 h at room tempera- ture. After blocking, the sections were stained with primary antibodies in immunofluorescence buffer at 4 °C overnight: MYC-tag (rabbit, 1:1,000, Cell Signaling), SOX9 (rabbit, 1:5,000, Millipore), ITGA6 (rat, 1:1,000, BD), KRT14 (chicken, 1:1,000, BioLegend), KRT10 (rabbit, 1:250, Fuchs Lab), EpCAM (rabbit, 1:100, Abcam), KRT6 (guinea pig, 1:1,000, Fuchs Lab), RUNX1 (rabbit, 1:100, Abcam), and GATA3 (rat, 1:100, Invit- rogen). After primary antibody staining, all sections were washed three times with immunofluorescence buffer containing 0.1% Triton X-100 in PBS for 5 min at room temperature. Sections were then stained with Alexa 488, 546 or 647 conjugated secondary donkey antibodies (1:500, Thermo Fisher), mounted with Prolong Diamond anti-fade mounting medium with 4′,6-diamidino-2-phenylindole (DAPI, Thermo Fisher) and imaged with Zeiss Axio Observer Z1 with Apotome 2 microscope. RNA-seq and raw file processing EpdSCs were collected by FACS as described above directly into TRIzol LS (Invitrogen). RNA libraries were generated using SMARTer RNA kit for low-input RNA-seq. Libraries were sequenced on Illumina NovaSeq SP. Raw FASTQ files were trimmed of barcodes using Skewer (v.0.2.2) and transcript abundance quantified using Salmon (v.1.4.0) with a modified GENCODE transcript index (version GRCm38 release M24) to include TRE-Sox9. Gene level counts and transcripts per million (TPM) were calculated using the Tximport (v.1.12.3) package in R (v.3.6.1). For hair placode RNA-seq data, after generating the raw counts, differen- tially expressed (DEG) gene list was generated with DESeq2 (v.1.16.1). ATAC-seq and raw file processing ATAC-seq20 was performed on FACS-purified EpdSCs (two to four male mice per replicate) at indicated timepoints (D0, W1, W2, W6 and W12) and cultured keratinocytes. Briefly, cells were lysed in ATAC lysis buffer for 5 min and then transposed with Tn5 transposase (Illumina) for 30 min. Samples were barcoded and sequencing libraries were prepared according to the manufacturer’s guidelines (Illumina) and sequenced on an Illumina NextSeq. For sequencing analysis, 50 bp paired-end FASTQs were aligned to the mouse genome (GRCm38/mm10) using the PEPATAC (v0.10.3) pipeline45. Replicate BAM files were merged, and peak calling Nature Cell Biology Articlehttps://doi.org/10.1038/s41556-023-01184-y was performed using Model-based Analysis of ChIP-Seq 2 (MACS2) with the option of ‘–keep-dup all’ to keep duplicates generated during the combining of experimental replicates. Because peak calling is greatly influenced by number of reads and sequencing depth, we normalized peak calling as performed as described21 with a threshold of 3, and we quantified reads in filtered peaks (RIP) for generating normalized bigwig files. To do so, 1,000,000/RIP was used as input for Deeptools ‘bam- coverage’ with the ‘–scaleFactor’ option. Shared peaks were defined as regions that had ≥1 base pair overlap between two timepoints as shown in Fig. 2e. Dynamic peaks were defined as those accessible chromatin regions that were absent from at least one timepoint. For PCA analysis, peaks called from combined replicates were merged to create a union set of peaks across the samples. Read counts under the union peaks were summed for each individual replicate and used as input for PCA analysis or generating K-means clusters in R. CNR and raw file processing EpdSCs were FACS purified, and the CNR sequencing was performed as previously described19,46 with minor modifications indicated below. Briefly, 500,000–1,000,000 EpdSCs were washed with ice-cold PBS, resuspended in crosslinking buffer (10 mM HEPES–NaOH pH 7.5, 100 mM NaCl, 1 mM egtazic acid (EGTA), 1 mM EDTA and 1% formalde- hyde) and rotated at room temperature for 10 min. Crosslinked cells were quenched with glycine at a final concentration of 0.125 M for 5 min at room temperature. Cells were washed with cold 1× PBS and resuspended in NE1 buffer (20 mM HEPES–KOH pH 7.9, 10 mM KCl, 1 mM MgCl2, 1 mM dithiothreitol, 0.1% Triton X-100 supplemented with Roche complete protease inhibitor EDTA-free) and rotated for 10 min at 4 °C. Nuclei were washed twice with CNR wash buffer (20 mM HEPES pH 7.5, 150 mM NaCl, 0.5% bovine serum albumin and 0.5 mM spermidine supplemented with protease inhibitor) and incubated with concanavalin-A (ConA) beads washed with CNR binding buffer (20 mM HEPES–KOH pH 7.9, 10 mM KCl, 1 mM CaCl2 and 1 mM MnCl2) for 10 min at 4 °C. ConA-bead-bound nuclei were incubated overnight at 4 °C in CNR antibody buffer (CNR wash buffer supplemented with 0.1% Triton X-100 and 2 mM EDTA) and antibody. After antibody incu- bation, ConA-bead-bound nuclei were washed once with CNR Triton wash buffer (CUT&RUN wash buffer supplemented with 0.1% Triton X-100) then resuspended and incubated at 4 °C for 1 h in CUT&RUN antibody buffer and 2.5 μl pAG-MNase (EpiCypher). ConA-bound-nuclei were then washed twice with CUT&RUN Triton wash buffer and resus- pended in 100 μl of Triton wash buffer and incubated on ice for 5 min. Then, 2 μl 100 mM CaCl2 was added and mixed gently to each 100 μl ConA-bound-nuclei. The reaction was then incubated at 0 °C for 30 min. The reaction was stopped by addition of 100 μl 2× stop buffer (340 mM NaCl, 20 mM EDTA, 4 mM egtazic acid, 0.1% Triton X-100 and 50 μg ml−1 RNaseA) and incubated at 37 °C for 10 min. All buffers mentioned above were filtered with 0.22 μm filter before use. After incubation, ConA-bound-nuclei were captured using a magnet and supernatant containing CNR DNA fragments were collected. Superna- tant was incubated at 70 °C for 4 h with 2 μl 10% sodium dodecyl sulfate and 2.5 μL 20 mg ml−1 proteinase K. DNA was purified using PCI reagent (phenol:chloroform:isoamyl alcohol, Millipore) and overnight ethanol precipitation with glycogen at −20 °C. DNA was resuspended in elution buffer (1 mM Tris–HCl pH 8.0 and 0.1 mM EDTA). CNR sequencing libraries were generated using NEBNext Ultra II DNA Library Prep Kit for Illumina and NEBNext Multiplex Oligos for Illu- mina. PCR-amplified libraries were purified using 1× ratio of SPRI beads (Beckman) and eluted in 15 μl EB buffer (Qiagen). All CNR libraries were sequenced on Illumina NextSeq using 40 bp paired-end reads. Reads were trimmed with Skewer and aligned to reference genome (mm10) using Bowtie2 (v.2.2.9) and deduplicated with Java (v.2.3.0) Picard tools (http://broadinstitute.github.io/picard). Reads were filtered to ≤120 bp using Samtools (v.1.3.1). BAM files for each replicate were combined using Samtools. Bigwig files were generated using Deeptools (v.3.1.2) Nature Cell Biology with reads per kilobase of transcript per million mapped reads (RPKM) normalization and presented with Integrative Genomics Viewer software. CNR peaks were called using SEACR47 from bedGraph files generated from RPKM-normalized Bigwig files (bigWigToBedGraph, UCSC Tools) using stringent setting and a numeric threshold of 0.01. Peaks were further filtered to have peaks scores >1,800 for a set of high-confidence peaks. MINT-ChIP–seq and raw file processing EpdSCs were FACS purified and subjected to histone ChIP–seq (MINT-ChIP) with antibodies recognizing H3K4me1 (rabbit, Cell Signaling), H3K27ac (rabbit, Active Motif) and Total H3 (mouse, Active Motif). Pooled sam- ples were then sequenced using 50 bp paired-end Illumina NextSeq. Resulting FASTQ files were demultiplexed for specific histone antibod- ies by using the unique barcode present in sequenced read2. Resulting paired reads were then trimmed for adapters using Skewer and aligned to mouse genome (GRCm38/mm10) using Bowtie2. Duplicated reads were marked and removed using Picard, and replicates were merged with Samtools. Peak calling for H3K27ac was performed using MACS2, while broad domains of H3K4me1 were called using epic2 (ref. 48). Samples were independently normalized to the number of RIP. For visualization, Bigwig files were generated on the combined BAM files using Deeptools ‘bamcoverage’ with (1,000,000/RIP) as input for the ‘–scalefactor’ option. For total H3, RPKM was used for normalization. BioID and mass spectrometry For identification of SOX9-interacting partners we transduced pri- mary Krt14-rtTA EpdSCs with LV-TRE-MYC-BioID2-GFP-NLS-H2B-RFP or LV-TRE-MYC-BioID2-SOX9-H2B-RFP. RFP+ transduced cells were then isolated using FACS, and stable EpdSC lines were established. We induced expression of recombinant proteins using 1 μg ml−1 DOX. Cells were allowed to expand for 5 days and were pulsed with 50 μM biotin (Sigma) for 16 h before reaching confluence. Cells were purified and proteins isolated as previously described49 with minor modifica- tions mentioned below. Immediately after sonication, lysates were washed using Zeba desalting columns (7K molecular weight cut-off, ThermoFisher cat. no. 89894) with 50 mM Tris pH 7.4 to remove excess biotin. Beads were also washed three times with 2 M urea and a final two times with PBS before being resuspended with 500 μl 50 mM Tris, pH 8.0. All washes were performed using a magnetic stand. New tubes were used in between each urea and PBS washes. Wash buffer was removed from suspension of magnetic beads and replaced with 100 μl 8 M urea, 50 mM ammonium bicarbonate and 10 mM dithiothreitol for 1 h and replaced with 100 μl 40 mM iodoacetamide and incubated in the dark for 30 min. Alkylation solution was replaced with 1 μg trypsin (Promega) dissolved in 100 μl 50 mM ammonium bicarbonate and incu- bated for 4 h. Supernatant was then removed and re-digested overnight using 0.5 μg trypsin and 0.5 μg Endopeptidase Lys-C (Wako). Peptides were desalted and concentrated using C18-based Stage tips50 and sepa- rated by nanoLC (gradient: 2% B/98% A to 38% B/62% A in 70 min, A: 0.1% formic acid, B: 90% acetonitrile/0.1% formic acid) coupled to a Fusion Lumos (Thermo Scientific) operated in high/high mode. Data were queried with UniProts Complete Proteome mouse database and concatenated with known common contaminants. Proteome Discover and Mascot was used to analyse the result- ing data produced. Data were further filtered using a percolator51 to calculate peptide false discovery rates and set a threshold of 1%. Proteins were specific to SOX9’s proximity if they were identified in two of the three MYC-BioID2-SOX9 replicates and absent from all the MYC-BioID2-GFP-NLS samples. For the full list of SOX9-specific interac- tors and raw counts, see Supplementary Table 3. Generation of EpdSC lines expressing SOX9 and variants, AFOS or ARID1a Three versions of MYC-tagged SOX9 (WT, ΔTA and ΔHMG as indicated in Extended Data Fig. 7b) were cloned into plKO vectors with a TRE promoter Articlehttps://doi.org/10.1038/s41556-023-01184-y and a puromycin-resistance gene (puroR) under the control of a constitu- tive promoter (PGK). Three lentiviruses were produced as described52. Krt14-rtTA EpdSCs were cultured and transduced with 1 μl concentrated lentivirus in 10 ml E-low medium with 8 μg ml−1 polybrene (hexadime- thrine bromide, Sigma 107689-100MG) overnight. Transduced cells were then selected with 2 μg ml−1 puromycin for 5 days before DOX treatment. For AFOS and ARID1a experiments, Flag-tagged AFOS or Arid1a CDS were cloned into the described plKO vector for lentiviral production. Krt14-rtTA or Krt14-rtTA;TRE-mycSOX9 EpdSCs were cultured and transduced with 1 μl concentrated lentivirus as described above. Transduced cells were also selected with puromycin for 5 days before DOX treatment. CRISPR-mediated Mll4 knockout To generate Mll4 (also known as Kmt2d) null lines, we cultured keratino- cytes from the EpdSCs of our Krt14-rtTA, TRE-Sox9 mice. Lines were generated with the Alt-R CRISPR–Cas9 system (Integrated DNA Tech- nologies). Briefly, a recombinant Cas9 protein, a validated single guide RNA (TGCTCGGCAACAGACGTGAC) targeting Mll4 or a negative control single guide RNA (Integrated DNA Technologies), and an ATTO-550 conjugated tracer RNA were used to form a ribonucleoprotein were mixed with RNAiMax reagent (Thermo Fisher). Then, keratinocytes were transfected with the mixture overnight, and FACS purified into 96-well plates to produce clonal cell lines. The knockout cell lines were validated through sequencing of the target region for indel efficiency via MiSeq and used for the immunoblot of MLL4. Immunoblotting and co-immunoprecipitation Cultured EpdSCs were washed on the plate in cold 1× PBS, lysed in RIPA buffer (Millipore) supplemented with protease and phosphatase inhibitors (Roche), and collected by scraping. Cells were lysed for 15 min on ice and then centrifuged to collect the supernatant. Co-immunoprecipitation was performed as previously described53 with the modification where protein-A/G-conjugated magnetic beads (Pierce) were used to bind antibodies instead, and proteins were eluted from beads with 1× NuPAGE LDS Sample Buffer (Invitrogen) with 2.5% 2-mercaptoethanol at 70 °C for 10 min. Protein concentration was deter- mined by BCA Assay (Pierce) against a bovine serum albumin standard curve. Then 15 μg protein of each sample was run on NuPAGE 4–12% Bis-Tris Gels (Invitrogen) for 2 h at 110 V in NuPAGE MOPS SDS Running Buffer (Invitrogen). Protein was transferred onto nitrocellulose mem- brane (Cytiva) in NuPAGE Transfer Buffer (Invitrogen) at 15 V overnight at 4 °C. Given the marked differences in expected sizes of some of the pro- teins, overlapping host species of the antibodies raised, and the paucity of primary cell lysates for immunoprecipitates, we often cut the blots on the basis of size and performed immunoblotting on each piece with different antibodies. Membranes were then treated with blocking buffer with 5% non-fat dry milk and 0.1% Tween-20 in TBS for 1 h at room temperature before incubating with primary antibodies. The following primary antibodies were diluted in blocking buffer: MYC-tag (mouse, 1:1,000, Cell Signaling), MLL4 (mouse, 1:200, Santa Cruz Biotechnol- ogy), cJUN (rabbit, 1:1,000, Cell Signaling), ARID1a (rabbit, 1:1,000, Abcam) and β-actin (mouse, 1:10,000, Cell Signaling). The membranes were incubated in primary antibodies overnight at 4 °C. Membranes were then washed three times in 0.1% Tween-20 in TBS before incubating with HRP secondary (1:10,000) antibody for 1 h at room temperature. After secondary antibody incubation, membranes were then washed four times in 0.1% Tween-20 in TBS and incubated in ECL Prime reagents (Cytiva) for 5 min before chemiluminescence detection. Membranes were imaged with an GE Amsham AI600 Imager. For clarity, we show the bands of the correct sizes. However, all full blots (cut before processing as delineated above) are shown in corresponding source data. Quantitative PCR Equal amounts of RNA extracted from cultured cells were collected with AllPrep DNA/RNA Kits (Qiagen) and reverse transcribed using the Nature Cell Biology superscript VILO cDNA synthesis kit (Invitrogen). For quantitative PCR, biological replicates represent the average of three technical replicates per individual sample. Complementary DNAs from each sample were normalized using primers against Rps16. All primers used are provided in Supplementary Table 4. Bioinformatic analyses GSEA. For comparing with both hair placodes and BCC, TPM matrices in D0, W2 and W12 were used as GSEA (v. 4.1.0) input. The DEG lists as illustrated in Fig. 1d were used as gene set inputs. For the BCC sample, DEG list of genes with P < 0.05 was curated from GSE152487 in the Gene Expression Omnibus depository17. GSEA was run with default settings, without collapsing, and with the gene set as the permutation type. The leading-edge analysis function was used to determine the significance of gene set enrichment. Heat maps and box plots. All heat maps showing sequencing sig- nals over binding sites are generated with Deeptools from RIP- or RPKM-normalized bigwig files. Profileplyr (v. 1.4.3) was used to gen- erate ATAC, H3K4me1, H3K27ac and MLL3/4 CNR box plots in R with matrix output from Deeptools compute-matrix as input. The histone H3 profile plot was also generated with Profileplyr in R. GO analysis. We performed GO analysis of each ATAC-seq cluster by associating each region with genes and performing enrichment analysis using Genomic Regions Enrichment of Annotation Tool (GREAT, ver- sion 3)54 with default gene association settings and the whole mouse genome (GRCm38/mm10) as the background. Transcription factor motif and footprint analyses. For motif enrich- ment analysis on peak sets, HOMER55 (v. 4.10) findMotifGenome.pl was used with a customized motif database from JASPAR2018 (ref. 56). The motif input for HOMER was generated from the 79 clusters of JASPAR2018 vertebrates CORE central transcription factor motifs using 80% of the maximum log-odds expectation for each motif as the detection threshold for HOMER. To identify cluster-specific motif enrichment in our ATAC-seq clusters we ran HOMER for each cluster using the union set of dynamic peaks as our background (-bg) set with the options -size given –h. The resulting heat map was generated by combining the significant (P < 0.05) motifs for each cluster and plotting the associated P value. For motif distance measuring, we overlapped SOX9-bound opening peaks with known AP1 and SOX motifs curated by HOMER (mm10-191020) and measured the distance from SOX motifs to the closest AP1 motifs with Bedtools. For footprint analysis, we used HINT-ATAC57 with our 79 motif clusters as the input as well. For transcription factor motif variability score analysis, we ran ChromVAR31 (1.18.0) on the dynamic peaks for differential chromatin accessibility across our 79 motif clusters to find the top variable motifs in dynamic peaks. We further used ChromVAR to calculate the motif deviation scores over time at the top variable motifs. Illustrations. Schematics were prepared using BioRender and Adobe Illustrator (v. 26.0.1). Statistics and reproducibility. No statistical methods were used to pre-determine sample sizes, but our sample sizes are similar to those reported in previous publications14,16,46. No data points were excluded. Upon collection, mice with the same genetic background were randomly allocated to genomic or immunofluorescence experi- ments. Data collection and analysis were not performed blind to the conditions of the experiments as the mice appears phenotypical after SOX9 induction. All immunofluorescence experiments were repeated three times with samples collected from different mice. All co-immunoprecipitation and immunoblot experiments were repeated twice with samples collected on different days. The statistics in Fig. 5g Articlehttps://doi.org/10.1038/s41556-023-01184-y and Extended Data Fig. 1d were analysed with two-tailed t-test on the GraphPad Prism (9.0). Data distribution was assumed to be normal, but this was not formally tested. All the error bars are mean ± s.d. *P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001. 53. Guzzi, N. et al. Pseudouridine-modified tRNA fragments repress aberrant protein synthesis and predict leukaemic progression in myelodysplastic syndrome. Nat. Cell Biol. 24, 299–306 (2022). Resource availability Lead contact. Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, E.F. (fuchslb@rockefeller.edu). Materials availability. Will be provided upon request and available upon publication. Reporting summary Further information on research design is available in the Nature Port- folio Reporting Summary linked to this article. Data availability All data that support the findings of this study are available within the paper and its supplementary files. Sequencing data that support the findings of this study have been deposited in the Gene Expression Omni- bus under accession code GSE208072. Previously published RNA-seq data from BCC, SCC and normal EpdSCs that were re-analysed here are available under accession code GSE152487. Source data are provided with this paper. All other data supporting the findings of this study are available from the corresponding author on reasonable request. Code availability All bioinformatic analysis tools and pipelines used in this study are documented in the method section. Codes are available from the cor- responding author upon reasonable request. References 42. Vasioukhin, V., Degenstein, L., Wise, B. & Fuchs, E. The magical touch: genome targeting in epidermal stem cells induced by tamoxifen application to mouse skin. Proc. Natl Acad. Sci. USA 96, 8551–8556 (1999). 43. Blanpain, C., Lowry, W. E., Geoghegan, A., Polak, L. & Fuchs, E. Self-renewal, multipotency, and the existence of two cell populations within an epithelial stem cell niche. Cell 118, 635–648 (2004). 44. Rheinwald, J. G. & Green, H. Epidermal growth factor and the multiplication of cultured human epidermal keratinocytes. Nature 265, 421–424 (1977). 45. Smith, J. P. et al. PEPATAC: an optimized pipeline for ATAC-seq data analysis with serial alignments. NAR Genom. Bioinform. 3, lqab101 (2021). 46. Larsen, S. B. et al. Establishment, maintenance, and recall of inflammatory memory. Cell Stem Cell 28, 1758–1774 e1758 (2021). 47. Meers, M. P., Tenenbaum, D. & Henikoff, S. Peak calling by Sparse Enrichment Analysis for CUT&RUN chromatin profiling. Epigenetics Chromatin 12, 42 (2019). 48. Stovner, E. B. & Saetrom, P. epic2 efficiently finds diffuse domains in ChIP–seq data. Bioinformatics 35, 4392–4393 (2019). 49. Kim, D. I. & Roux, K. J. Filling the void: proximity-based labeling of proteins in living cells. Trends Cell Biol. 26, 804–817 (2016). 50. Rappsilber, J., Mann, M. & Ishihama, Y. Protocol for micro-purification, enrichment, pre-fractionation and storage of peptides for proteomics using StageTips. Nat. Protoc. 2, 1896–1906 (2007). 54. McLean, C. Y. et al. GREAT improves functional interpretation of cis-regulatory regions. Nat. Biotechnol. 28, 495–501 (2010). 55. Heinz, S. et al. Simple combinations of lineage-determining transcription factors prime cis-regulatory elements required for macrophage and B cell identities. Mol. Cell 38, 576–589 (2010). 56. Khan, A. et al. JASPAR 2018: update of the open-access database of transcription factor binding profiles and its web framework. Nucleic Acids Res. 46, D260–D266 (2018). 57. Li, Z. et al. Identification of transcription factor binding sites using ATAC-seq. Genome Biol. 20, 45 (2019). Acknowledgements We thank M. Nikolova, E. Wong, J. Racelis, T. Omenchenko and J. Levorse for experimental assistance; M. Parigi, N. Guzzi, M. D. Abdusselamoglu, S. Yuan, A. Gola, K. Gonzales and C. Cowley for discussions; S. Mazel, S. Semova, S. Han and S. Shalaby for conducting FACS sorting; C. Lai (for high-throughput sequencing and raw data analyses; H. Molina for conducting mass spectrometry. E.F. is a Howard Hughes Medical Investigator. N.G. was the recipient of Burroughs Welcome Diversity fellowship (1017355), and an F32 postdoctoral fellowship from the National Cancer Institute (5F32CA221353). N.I. was the recipient of an F31 from the National Institutes of Health (5F31AR073110). M.L. was the recipient CIHR postdoctoral fellowship. This study was supported by grants to E.F. from the National Institutes of Health (R01-AR31737 and R01-AR050452). Author contributions Y.Y., N.G. and E.F. conceptualized the study, designed the experiments, interpreted the data and wrote the manuscript. Y.Y. and N.G. performed and analysed in vivo high throughput data. M.L. assisted with proteomic experiments. R.C.A. generated the SOX9-inducible transgenic mice. Y.Y. performed in vitro studies with help from I.B., immunofluorescence microscopy and quantifications. M.S. participated in SOX9 mouse experiments and tumour cell engraftments. N.I. participated in high-throughput data generation, immunofluorescence microscopy and quantifications. All authors provided input on the final manuscript. Competing interests The authors declare no competing financial interests in this research, but E.F. was on the Scientific Advisory Boards of L’Oreal and Arsenal Biosciences during a period while these studies were ongoing. Additional information Extended data is available for this paper at https://doi.org/10.1038/s41556-023-01184-y. Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41556-023-01184-y. Correspondence and requests for materials should be addressed to Elaine Fuchs. 51. Kall, L., Canterbury, J. D., Weston, J., Noble, W. S. & MacCoss, M. J. Semi-supervised learning for peptide identification from shotgun proteomics datasets. Nat. Methods 4, 923–925 (2007). Peer review information Nature Cell Biology thanks Yali Dou, Anthony Oro and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. 52. Beronja, S., Livshits, G., Williams, S. & Fuchs, E. Rapid functional dissection of genetic fetworks via tissue specific transduction and RNAi in mouse embryos. Nat. Med. 16, 821–827 (2010). Reprints and permissions information is available at www.nature.com/reprints. Nature Cell Biology Articlehttps://doi.org/10.1038/s41556-023-01184-y Extended Data Fig. 1 | See next page for caption. Nature Cell Biology Articlehttps://doi.org/10.1038/s41556-023-01184-y Extended Data Fig. 1 | Ectopic reactivation of SOX9 in adult EpdSCs silences epidermal fate and activates hair follicle fate within 2W and then progresses to BCC-like lesions. a, Schematic of the TRE-Sox9 construct for generating SOX9-inducible transgenic mice. b, Immunofluorescence showing myc-tag staining at D0 and W12 SOX9-induction. This validates myc-tag expression is only induced and specifically induced in EpdSCs. All scale bars are 50μm. c, Immunofluorescence comparing SOX9 expression in the normal anagen ORS (containing activated HFSCs) and induced SOX9 expression in the adult EpdSCs. All scale bars are 50μm. d, Quantifications of SOX9 immunofluorescence intensity in the two states in (b) shows that SOX9 levels in ectopically induced adult EpdSCs are not higher than in native ORS HFSCs. n = 100 cells measured over 5 biological replicates. All the error bars are mean ± SD. Statistical significance from two-tailed t-test is denoted by ****(p < 0.0001). e, (left) KRT14 and KRT10 immunofluorescence of the skin after SOX9-induction in EpdSCs. Note that the KRT14 skin progenitor is markedly expanded over time. All scale bars are 50μm. (right) Quantification of the thickness of KRT14 and KRT10 layers over time. n = 5 biological replicates with 2 measurement per sample. Boxplots are centered at median and bound by 1st and the 3rd quartile, and whiskers extend to 1.5 times IQR on both ends. The solid dots are data points, and the empty circles are outliers beyond 1.5 times IQR. f, (top) cell proliferation as assessed by EdU immunofluorescence after SOX9-induction. All scale bars are 50μm. (bottom left) quantification of % proliferating cells in the basal layer of EpdSCs. g, (left) Immunostaining of human BCC samples from The Human Protein Atlas. SOX9 and EpCAM show strong staining in the body of the lesion, whereas KRT6A is restricted to the apical epidermis. (right) Immunofluorescence images show the lesions in 12-week samples have similar EpCAM and KRT6 staining pattern as human BCC. Dotted lines denote the dermo-epidermal border. All scale bars are 50μm. Nature Cell Biology Articlehttps://doi.org/10.1038/s41556-023-01184-y Extended Data Fig. 2 | Quality control for transcriptome analysis and temporal comparisons between SOX9-induced transcriptome analyses and cutaneous cancers. a, FACS gating strategies for isolating EpdSCs. DAPI-negative singlets were first gated on ITGA6 (epithelial progenitors) and immune cell, fibroblast/adipocyte, endothelial and melanocyte lineage markers (CD45, CD140a, CD31 and CD117). Lineage-negative and ITGA6+ cells were further gated on CD34 and Ly6A/E (SCA-1) to distinguish HFSCs and EpdSCs, respectively. b, Replicate correlation analyses of RNA-seq show strong correlation (r > 0.94) between samples across time points. c, At W2 following SOX9 induction, similarities to BCC were already apparent (K-S test, p < 0.001). By contrast, a negative correlation to SCC was observed (K-S test, p < 0.001). d, In comparing all BCC genes to our W6 and W2 data, the shift towards a BCC signature continued to rise, concomitant with the phenotypic changes described in Fig. 1 (K-S test, p = 0.014). Nature Cell Biology Articlehttps://doi.org/10.1038/s41556-023-01184-y Extended Data Fig. 3 | See next page for caption. Nature Cell Biology Articlehttps://doi.org/10.1038/s41556-023-01184-y Extended Data Fig. 3 | Analyses and dynamics of ATAC-seq and Cut-and-Run and peak distributions. a, Replicate correlation analyses of SOX9 Cut-and- Run show strong correlation between duplicate samples at each time point. b, HOMER motif analysis shows that SOX(HMG) motifs are strongly enriched in SOX9 CNR peaks in W1 to W12 samples. c, ATAC-seq samples show clear nucleosome patterning and TSS enrichment. d, Total histone H3 signals are mutually exclusive from ATAC signals at D0 and W2 time points. e, Distribution of dynamic and static peaks at different genomic features. Dynamic peaks are more enriched in intronic and intergenic regions, while static peaks are more enriched at promoters. f, ChromVAR analysis of motif deviation scores of SOX (blue), AP1/FOS/JUN (red), RUNX (purple), GATA (green), and TBX (brown) motifs at indicated time points. Note that SOX motif accessibility rises markedly in dynamic peaks that open within the first two weeks post SOX9 induction while GATA motif accessibility declines. AP1 and RUNX motif accessibility rise between W2 and W6. TBX motif is shown as a control which does not change accessibility overtime. g, HINT-ATAC footprint analysis shows how transcription factor (TF) motif footprints differ in activity score at indicated time points compared to D0. Motifs indicated with red dot show significant changes in activity score over D0. Nature Cell Biology Articlehttps://doi.org/10.1038/s41556-023-01184-y Extended Data Fig. 4 | MINT-ChIP experiments and analyses of H3K4me1 relative to SOX9 binding. a, Replicate correlation analyses of the H3K4me1 and H3K27ac MINT-ChIP show strong correlation between each duplicate sample across all time points. b, Boxplot of H3K27ac signals at opening SOX9 peaks overtime. Note that the signal increases gradually from W1 to W12. n = 2 biological replicates. Boxplot is centered at median and bound by 1st and the 3rd quartile, and whiskers extend to 1.5 times IQR on both ends. c, Profile plot (top) and heatmap (bottom) showing the H3K4me1 signals at all SOX9 Cut-and-Run peaks across time points. Note the strong flanking pattern of the H3K4me1 signals adjacent to each SOX9 binding site (center dip) from W1 to W12. d, Boxplot of H3K4me1 domain sizes at opening SOX9 peaks and SOX9 bound static peaks shows gradual increase of H3K4me1 domain size from W1 to W2 after SOX9 induction. n = 2 biological replicates. Boxplots are centered at median and bound by 1st and the 3rd quartile, and whiskers extend to 1.5 times IQR on both ends. Nature Cell Biology Articlehttps://doi.org/10.1038/s41556-023-01184-y Extended Data Fig. 5 | BioID experiments and identification of SOX9 co-factors. a, Immunofluorescence validation of Krt14rtTA primary keratinocytes transduced with TRE-HA-SOX9-BioID2; H2B-RFP (left) or TRE-GFP- NLS-BioID2; H2B-RFP (right). Transgenes were induced with doxycycline and immunolabeled for HA (left) or GFP (right). All scale bars are 50μm. b, Correlation plots of Label Free Quantification (LFQ) values identified by BioID experiments across replicates and samples. SOX9-BioID and GFP-NLS-BioID samples share stronger correlations between replicates than each other. Blue number represents r2 value. c, PCA analysis of GFP and SOX9 BioID protein interactors demonstrate sample-specific clusters. d, Histogram of LFQ intensity values for the GFP (left) and SOX (right) BioID replicates. e, Molecular function enrichment of proteins specifically interacting with SOX9. f, Gene ontology enrichment of proteins specific to SOX9 (binomial is used to calculate the significance). g, Two Mll4 null keratinocyte cell lines (Krt14rtTA; TRE-mycSox9) were generated with CRIPSR/Cas9 with 99.9% indel frequency. These lines validated the efficacy of the MLL4 antibody by immunoblot, which detected a ~500 kDa protein in the control but not the knockout (KO) cell lines. MW, molecular weight. h, Immunoblot showing that the ~500 kDa protein, identified in (g) as MLL4, is pulled down with mycSOX9 in an anti-myc tag antibody immunoprecipitation. See also in associated source data. Nature Cell Biology Articlehttps://doi.org/10.1038/s41556-023-01184-y Extended Data Fig. 6 | Epidermal TFs diminish rapidly upon SOX9 induction in EpdSCs. a, Upset plot shows % of peaks bound by SOX9 in the top peak sets from Fig. 2d. Note that peaks opened by W2 are more often bound by SOX9 than later opening or closing peaks. b, GO terms enriched in SOX9-bound opening peaks (C2, C4, C5) and all closing peaks (C1,C6). c, ATAC footprint analysis at D0 and W2 shows a decrease in chromatin accessibility at GATA footprint in dynamic ATAC peaks following SOX9 induction. d, EpdSCs expression of TFs that belong to TF families whose motifs are enriched in closing ATAC peaks (C1 and C6 in Fig. 3c) (binomial is used to calculate the significance). e, (top) Transcript levels of Gata3 over time following SOX9 induction. (bottom) GATA3 immunofluorescence of epidermis at D0, W6 and W12. Scale bars, 50μm. f, Integrative Genomics Viewer (IGV) snapshot of SOX9, ATAC and H3K4me1 tracks within the Gata3 locus. The red box indicates a peak >30kb downstream from the gene that is bound by SOX9 and MLL3/4, and stays open at W2 even though the gene body of Gata3 closes its chromatin (see Fig. 5b). Nature Cell Biology Articlehttps://doi.org/10.1038/s41556-023-01184-y Extended Data Fig. 7 | See next page for caption. Nature Cell Biology Articlehttps://doi.org/10.1038/s41556-023-01184-y Extended Data Fig. 7 | Truncated SOX9 displays impaired DNA binding or co-factor recruitment. a, (top) Venn diagram reveals almost no overlap between ATAC peaks (C2,C4,C5) that mostly open between W2-W12 and D0 MLL3/4 Cut-and-Run peaks. (bottom) Venn diagram shows substantial overlap between ATAC peaks (C1,C6) that mostly close between W1-W2 and D0 MLL3/4 Cut-and-Run peaks. b, Schematic illustrating constructs engineered to express WT SOX9 and two variants of SOX9. c, Immunofluorescence reveals similar intensities of WT and mutant SOX9 in induced EpdSCs. Scale bars, 100μm. d, Immunoblot validating the sizes of different versions of SOX9. MW, molecular weight. e, Immunoblot showing that the transactivating (TA) domain of SOX9 is sufficient in binding MLL4. f, Venn diagrams show that the peak sets closed by WT and ΔHMG-SOX9 are comparable in size. g, (top) Top 5 biological process gene ontologies of genes associated with SOX9-bound opening peaks upon WT SOX9 induction in vitro from GREAT. (bottom) Top 5 biological process gene ontologies of genes associated with peaks opened upon ΔHMG SOX9 induction in vitro from GREAT. Binomial is used to calculate the significance. h, Distribution of the distance between a SOX motif and its closest AP1 motif in the SOX9 bound opening peaks. Note that the x-axis is binned by multiplies of one nucleosome size (147bp). The cumulation plot of the distribution is shown in orange on the secondary y-axis. i, Immunoblot showing that both WT-SOX9 and ΔHMG-SOX9 are capable of binding c-JUN and ARID1a. j, Immunoblot validating ARID1a can be induced 3x higher in the SOX9 expressing keratinocytes. For full blots, see associated source data. Nature Cell Biology Articlehttps://doi.org/10.1038/s41556-023-01184-y
null
10.1371_journal.pcbi.1011795.pdf
null
Computational code is available at https://github.com/kieran12lamb/ SARS-CoV2_Mutational_Signatures GISAID data accessions are available at doi.org/10.55876/gis8 .
RESEARCH ARTICLE Mutational signature dynamics indicate SARS- CoV-2’s evolutionary capacity is driven by host antiviral molecules Kieran D. LambID T. Phan3,4, Matthew Cotten1,3,4,5, Ke YuanID 1,2☯, Martha M. Luka1,2☯, Megan Saathoff1, Richard J. Orton1, My V. 2,6,7*, David L. RobertsonID 1* a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 1 Medical Research Council - University of Glasgow Centre for Virus Research, School of Infection and Immunity, Glasgow, Scotland, United Kingdom, 2 School of Computing Science, University of Glasgow, Glasgow, Scotland, United Kingdom, 3 Medical Research Council/Uganda Virus Research Institute and London School of Hygiene & Tropical Medicine Uganda Research Unit, Entebbe, Uganda, 4 College of Health Solutions, Arizona State University, Phoenix, Arizona, United States of America, 5 Complex Adaptive Systems Initiative, Arizona State University, Scottsdale, Arizona, United States of America, 6 School of Cancer Sciences, University of Glasgow, Glasgow, Scotland, United Kingdom, 7 Cancer Research UK Scotland Institute, Glasgow, Scotland, United Kingdom ☯ These authors contributed equally to this work. * Ke.Yuan@glasgow.ac.uk (KY); David.L.Robertson@glasgow.ac.uk (DLR) OPEN ACCESS Citation: Lamb KD, Luka MM, Saathoff M, Orton RJ, Phan MVT, Cotten M, et al. (2024) Mutational signature dynamics indicate SARS-CoV-2’s evolutionary capacity is driven by host antiviral molecules. PLoS Comput Biol 20(1): e1011795. https://doi.org/10.1371/journal.pcbi.1011795 Editor: Roger Dimitri Kouyos, University of Zurich, SWITZERLAND Received: July 12, 2023 Accepted: January 3, 2024 Published: January 25, 2024 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pcbi.1011795 Copyright: © 2024 Lamb et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: Computational code is available at https://github.com/kieran12lamb/ SARS-CoV2_Mutational_Signatures GISAID data accessions are available at doi.org/10.55876/gis8. Abstract The COVID-19 pandemic has been characterised by sequential variant-specific waves shaped by viral, individual human and population factors. SARS-CoV-2 variants are defined by their unique combinations of mutations and there has been a clear adaptation to more efficient human infection since the emergence of this new human coronavirus in late 2019. Here, we use machine learning models to identify shared signatures, i.e., common underly- ing mutational processes and link these to the subset of mutations that define the variants of concern (VOCs). First, we examined the global SARS-CoV-2 genomes and associated metadata to determine how viral properties and public health measures have influenced the magnitude of waves, as measured by the number of infection cases, in different geographic locations using regression models. This analysis showed that, as expected, both public health measures and virus properties were associated with the waves of regional SARS- CoV-2 reported infection numbers and this impact varies geographically. We attribute this to intrinsic differences such as vaccine coverage, testing and sequencing capacity and the effectiveness of government stringency. To assess underlying evolutionary change, we used non-negative matrix factorisation and observed three distinct mutational signatures, unique in their substitution patterns and exposures from the SARS-CoV-2 genomes. Signa- tures 1, 2 and 3 were biased to C!T, T!C/A!G and G!T point mutations. We hypothe- sise assignments of these mutational signatures to the host antiviral molecules APOBEC, ADAR and ROS respectively. We observe a shift amidst the pandemic in relative mutational signature activity from predominantly Signature 1 changes to an increasingly high proportion of changes consistent with Signature 2. This could represent changes in how the virus and the host immune response interact and indicates how SARS-CoV-2 may continue to gener- ate variation in the future. Linkage of the detected mutational signatures to the VOC-defining amino acids substitutions indicates the majority of SARS-CoV-2’s evolutionary capacity is PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 1 / 26 PLOS COMPUTATIONAL BIOLOGY 221201qs, doi.org/10.55876/gis8.230406qg and doi.org/10.55876/gis8.230406fb. likely to be associated with the action of host antiviral molecules rather than virus replication errors. Mutational processes and SARS-CoV-2’s evolutionary capacity Funding: The authors acknowledge funding from the Medical Research Council (MRC, MC_UU_12014/12 to DLR, MC_UU_00034/5 to DLR and a Doctoral Training Programme in Precision Medicine studentship for KDL, MR/ N013166/1 to KY and DLR), the Wellcome Trust (220977/Z/20/Z to MC, KY, DLR), the UK Department for International Development (DFID) under the MRC/DFID Concordat agreement (MC_PC_20010 to MC), Engineering and Physical Sciences Research Council (EPSRC, EP/R018634/ 1 to KY), and the European Union’s Horizon 2020 research and innovation programme project PANCAIM (101016851 to KY). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Author summary We show that both public health measures and virus properties are associated with the rise and fall of regional SARS-CoV-2 reported infection numbers with regional differences attributable to the extent of vaccine usage and the effectiveness of public health measures. In our mutational signature analysis, using non-negative matrix factorisation, we detected three distinct mutational signatures that can be putatively attributed to the action of spe- cific host antiviral molecules. Interestingly, we observe a shift in mutational signature activity from predominantly Signature 1 changes to an increasingly high proportion of changes consistent with Signature 2. These mutation patterns influence SARS-CoV-2’s evolutionary capacity, the available genetic variation that selection can act on, and so can be linked to the mutations defining the variants of concern responsible for the distinct SARS-CoV-2 infection waves. The dominant types of nucleotide substitutions involved indicate that much of the mutation and hence variation come from the action of the host immune response rather than replication errors since the virus has an error correction system. Introduction The COVID-19 pandemic began in late 2019 following a zoonotic spillover event of a SARS- related coronavirus, subsequently named SARS-CoV-2, in Wuhan, China [1, 2]. The extensive and rapid global spread of this new human coronavirus and its detrimental impact on human health has rendered it among the most significant pandemics in recent history [3]. Different geographical regions of the world have reported varied infection patterns that are attributed to differences in population demographics and health care systems, diverse government responses [4, 5], the emergence of more transmissible variants [6, 7] and other viral, human and population factors. Since its emergence, SARS-CoV-2 has undergone significant genetic change such that numerous variants, i.e., distinct genotypes, have been identified [8], many with altered phenotypic properties [9]. The World Health Organization (WHO) and other public health bodies have broadly classi- fied variants that pose an increased risk to global public health (due to increased transmissibil- ity, increased virulence or decrease in the effectiveness of public health measures relative to 2019/early 2020 SARS-CoV-2 variants) as variants of concern (VOCs) and variants of interest (VOIs) [10]. The early SARS-CoV-2 variants to emerge in 2019 and the more transmissible +S:D614G variant followed by the VOCs (Alpha, Beta, Gamma, Delta and currently Omicron) have driven significant and sequential “waves” of SARS-CoV-2 infections internationally. The emergence of each variant showing a clear geographical link [11–13]. Viral mutations arise from a diverse set of processes (principally viral polymerase replica- tion errors and host anti-viral editing processes), which can be identified by the characteristic mutational signatures that they leave on the genome [14, 15]. Such characterisation of domi- nant mutational processes is routinely used in cancer genomics [16]. The catalogue of SARS- CoV-2 nucleotide changes show distinct mutational patterns suggestive of a role for host antiviral mutational processes in introducing changes in the viral RNA [17, 18]. These PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 2 / 26 PLOS COMPUTATIONAL BIOLOGY Mutational processes and SARS-CoV-2’s evolutionary capacity processes potentially dominate in SARS-CoV-2 evolution because point mutations introduced in replication are mostly corrected by the action of a proofreading enzyme. The generation of virus diversity, the key to virus persistence by generating novel variation and thus evolutionary capacity, is multi-faceted [19], yet our understanding of the relative importance of underlying mutational processes linked to the action of host anti-viral mole- cules is still very limited. Given that SARS-CoV-2 continues to develop new variants, many associated with sets of previously observed (convergent) and novel mutations [9], it is critical that we improve our understanding of the mechanisms and sources of evolutionary change. Along with routine surveillance of SARS-CoV-2 infections, there has been an unprece- dented global sequencing effort resulting in databases containing many millions of genome sequences, in particular GISAID [20]. Here we examined this data to describe the global molecular epidemiology and evolution of SARS-CoV-2. Using regression models we first examined how viral properties and public health measures have influenced the magnitude of infection waves in different geographic locations. Satisfied that SARS-CoV-2 variants have been an important driver of infections we then used non-negative matrix factorisation to char- acterise the mutational processes involved in the generation of variants and their changing pat- terns of activity over time. Results Characterising the SARS-CoV-2 waves regionally This first part of the study reports on global SARS-CoV-2 data from 24/12/2019 to 28/01/ 2022 only as limited public health measures were in place after this time. We observed 1,544 distinct SARS-CoV-2 lineages from 7,348,178 sequences. 88% of the infections in the global pandemic during this time frame were caused by a subset of 13 Pango and WHO variants (S1 Table). While there are geographical differences there is a clear dominance of a subset of variants and replacement of these through time (Fig 1). This “wave” infection pattern was evident in all geographic locations. Although biased by testing rates, Europe and the Ameri- cas had the highest infection rates, reporting up to 450 cases per million population per day (Fig 1). The emergence or introduction of VOCs coincided with a steep increase in infection rates globally. For example, cases in Asia showed a steep rise in February 2021, which peaked in May 2021 (Fig 1, panel Asia). During this period, Alpha and Delta comprised greater than 75% of the SARS-CoV-2 cases identified in the sequence data. Africa and Oceania on the other hand displayed overall sustained low case numbers. Despite this, Beta dominated the second wave in parts of Africa while Alpha dominated the third Oceanic wave. After its emergence in March 2021, Delta spread to become the predominant variant across all conti- nents. The Omicron variant of concern was first identified in South Africa in late November 2021 and, by January 2022, it had rapidly become the predominant cause of infections world- wide (Fig 1). Covariates of the waves We investigated the degree to which public health measures and viral properties explain conti- nent-specific reported cases of infection. Correlation analysis at the global level showed a sig- nificant correlation between infection rates and the predictor variables: government stringency, vaccination, previous infection burden, virus diversity and fitness (S2 Table). Regression analysis revealed that the impact of the predictor variables on the magnitude of reported cases were found across all continents. We classified significance levels as follows: no significance for p-values greater than 0.05, weak significance for p-values between 0.05 and 0.001, and high significance for p-values less than 0.001. Our findings indicated that PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 3 / 26 PLOS COMPUTATIONAL BIOLOGY Mutational processes and SARS-CoV-2’s evolutionary capacity Fig 1. Continent-level SARS-CoV-2 lineage dynamics and pandemic curves. Lines show a 14-day rolling average of reported SARS-CoV-2 cases. Bars show the biweekly proportions of common lineages and are coloured by lineage. The white space shows the proportion of sequences from other (non-majority) lineages. https://doi.org/10.1371/journal.pcbi.1011795.g001 government stringency had a weakly significant impact in Asia, Europe, and South America, but a strongly significant impact in Africa, Oceania, and North America. Virus fitness, previ- ous infection burden, and vaccination demonstrated a strongly significant impact across all continents. Virus diversity was strongly correlated with high infection numbers in Europe and PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 4 / 26 PLOS COMPUTATIONAL BIOLOGY Mutational processes and SARS-CoV-2’s evolutionary capacity North America, with a weaker association in Africa, Asia, Oceania, and South America. The R squared values, indicating the proportion of variance explained by our model, were greater than 0.5 for all continents, ranging from 0.66 in Oceania to 0.79 in Africa (S3 Table). Gener- ally, our predictions closely resembled the rise and fall of SARS-CoV-2 infection case numbers (Fig 2). For country-level analysis, we included 29 countries from six continents based on the com- pleteness of data (availability of sequence data in every 14 day bin). Pandemic plots were visu- alised using biweekly bins and multiple linear regression was fitted using the same approach. Different countries had varying lineage dynamics as illustrated in S1 Fig. The five predictor variables had varying impacts on infection rates across countries (S2 Fig). Despite some differ- ences related to the population level processes investigated here, there is a clear variant replace- ment process taking place. As the generation of novel variants is fundamentally a mutation dependent process we next investigated the underlying patterns of mutations being generated through time. The goodness of fit varied among countries, with the R squared varying from 0.28 (Japan) to 0.96 (Australia), with a median of 0.69 (S4 Table). Though our model success- fully captured the general infection wave patterns in many countries, it struggled to capture short-term data spikes in specific instances, such as in Belgium (November 2020), India (May 2021), Indonesia (August 2021) and Japan (September 2021) (S2 Fig). Fig 2. Association of SARS-CoV-2 infection rates and predictor variables globally. A. Pearson’s correlation matrix of infection rate and predictor variables. Positive correlations are denoted in orange and negative correlations in blue and colour intensity is directly proportional to coefficient value. B. Model fitting using multiple linear regression. Black solid lines show a 14-day rolling average of adjusted SARS-CoV-2 cases. Pink solid lines show fitted mean response values of infection rates with predictor values as input. https://doi.org/10.1371/journal.pcbi.1011795.g002 PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 5 / 26 PLOS COMPUTATIONAL BIOLOGY Mutational processes and SARS-CoV-2’s evolutionary capacity Identifying putative mutational processes contributing to changes in SARS-CoV-2 New variants of concern have displaced viral lineages that were previously dominant in the population in different geographical regions and in some cases globally (Fig 1). This behaviour has been observed with the original variants of concern (Alpha, Beta and Gamma) and then globally with the Delta and Omicron lineages. We investigated whether these variant wave events (periods of time where infections are dominated by a single variant) were linked to the activity of specific mutational processes. Each of the variants of interest/concern has evolved independently such that detecting the patterns of mutations in the SARS-CoV-2 sequence data allows us to observe which processes are most active and could be contributing to the emer- gence of variants. Mutations were called using inferred references for each of the Pango lineages, which we call tree-based referencing (S3 Fig). The SARS-CoV-2 alignment of 13,278,844 sequences up to 26/10/2022 was used. Of these 13 million sequences 2,195,182 sequences were selected as they contained 5,726,144 newly arisen mutations. Cytosine to thymine mutations (C!T) were the most common and were the primary substitution category for most weeks where sequences were recorded. Note, SARS-CoV-2 has an RNA genome but we refer to uracil as a thymine to match pre-existing DNA mutational signature notations. Three signatures were identified with distinct substitution patterns using non-negative matrix factorisation (NMF) (Fig 3 and S5 Fig). Signature 1 is heavily biased towards C!T mutations. Signature 1 had a high probability of ACA, ACT and TCT contexts (adjacent nucle- otides in the 5’ and 3’ direction of the mutated site), consistent with what was earlier reported by Simmonds et al. [17] as highly mutated contexts for C!T substitutions in SARS-CoV-2. Signature 2 is predominantly adenine to guanine (A!G), guanine to adenine (G!A) and thy- mine to cytosine (T!C) mutations. The proportion of A!G and T!C mutations is approxi- mately equal in this signature, which is indicative of a double-stranded mutational process. SARS-CoV-2 mutations at adenine positions on the negative strand will be counted as thymine mutations due to the negative strand being used to replicate positive sense RNA, with the mutated A!G now pairing with a cytosine on the +sense RNA and replacing the original thy- mine [21, 22]. Signature 3 is predominantly composed of guanine to thymine (G!T) substitutions. The dynamics of mutational processes through the pandemic By using the available SARS-CoV-2 sequences we can measure the mutational signature activ- ity across time as long as our samples are aggregated using time series annotations. Signature exposures (Fig 4) show that Signature 1 remained the most prominent signature throughout the pandemic, although following the emergence of Signature 2 its activity reduced propor- tionally. Absolute exposure values (Fig 4B) show that Signature 1 does not appear to reduce its exposure, rather Signature 2 increases its exposure. Signature 2 establishes itself as a substantial signature after December 2020. It continues to expand after October 2021, just prior to the emergence of the Delta VOC. Signature 3 is by far the least active of the three signatures but remains consistent until after January-February 2022 when it begins to drop towards zero. This is around the time Omicron began to emerge as the dominant VOC. Combined signature activity reached a peak between July and October 2021 (Fig 4B) coin- ciding with the peak number of unique mutations (Fig 5A and 5B). This is around the time the mutational signature dynamics appear to be shifting, with Signature 2 contributing more unique mutations. We can see that this also coincides with the Delta VOC wave, which, between May 2021 and January 2022, was the lineage group showing the greatest number of PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 6 / 26 PLOS COMPUTATIONAL BIOLOGY Mutational processes and SARS-CoV-2’s evolutionary capacity Fig 3. Mutational signatures extracted from the SARS-CoV-2 genome sequences by non-negative matrix factorisation. Signatures are patterns of probabilities for each category of substitution in a three nucleotide context. Each bar represents a context and is coloured by the substitution category of the mutation that occurs there. Each signature may represent a distinct mutational process. Signature 1 is heavily biased towards cytosine to thymine (C!T) mutations, particularly in 3’ CpG contexts TCG, CCG and ACG. Signature 2 from SARS-CoV-2 is predominantly adenine to guanine (A!G), guanine to adenine (G!A) and thymine to cytosine mutations (T!C). Signature 3 is strongly guanine to thymine (G!T), a pattern that is thought to be caused by the action of guanine oxidation by reactive oxygen species. Signatures are shown normalised against the tri-nucleotide composition of the SARS-CoV-2 genome. Non-normalised forms in the context of the SARS-CoV- 2 genome composition are shown in S5 Fig. https://doi.org/10.1371/journal.pcbi.1011795.g003 newly acquired mutations (Fig 5). Delta was the first VOC to dominate on a global scale, out- competing other VOCs like Alpha, Beta and Gamma in their regions of circulation. Omicron similarly repeated this phenomenon, almost entirely replacing Delta globally within weeks of its emergence (Fig 5B). We also see a marked decrease in the activity of Signature 3 following Omicron’s establishment as the dominant variant. A similar decrease in G!T mutations was PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 7 / 26 PLOS COMPUTATIONAL BIOLOGY Mutational processes and SARS-CoV-2’s evolutionary capacity Fig 4. Signature exposure plots showing the activities of the extracted mutation signatures over the duration of the COVID-19 pandemic. A. Shows the percentage activity of the signatures during a given week of the pandemic, with each colour representing a different signature. B. Shows the signature activities as their absolute values at each epidemic week. https://doi.org/10.1371/journal.pcbi.1011795.g004 also observed by Bloom et al. [23] and Ruis et al. [24]. This is different to Delta, where there was an increase in Signature 3 following its emergence. These Signature 3 changes become par- ticularly apparent when we begin to look at signature activities within variant-defined subsets of the data. Signature dynamics spatially and by variant After observing changes in signature activity during transitions between dominant variants, we next investigated the differences between signature activities in variant-defined subsets of the data as well as in continent-defined subsets. We used the globally extracted signatures to extract exposures from the subsets using a non-negative least squares regression to retain the non-negativity constraint. This allowed for the measurement of signature activity in each of the subsets of interest. PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 8 / 26 PLOS COMPUTATIONAL BIOLOGY Mutational processes and SARS-CoV-2’s evolutionary capacity Fig 5. A. Counts of unique SARS-CoV-2 mutations for each epidemic week, with colours representing which continent the mutations came from. B. Counts of unique mutations per week that are part of the mutational signature substitution-context features (i.e., no indel mutations included). Colours represent which lineage/group of lineages the mutations belong to. C. Ridgeline plot showing the exposure of mutational signatures in SARS-CoV-2 variant-defined subsets. Exposures are coloured by the signature they have been attributed to. D. Ridgeline plot showing the exposure of mutational signatures in SARS-CoV-2 continent-defined subsets. https://doi.org/10.1371/journal.pcbi.1011795.g005 Signature 1 was the most active in almost all the variant-defined subsets as was expected from the global activity. Signature 3 was most active in the Delta subset as well as during the Delta wave in the continent-defined subsets (Fig 5). The non-VOC, Beta and Omicron subsets appear to be the least impacted by Signature 3 with almost zero activity in Omicron. Signature 2 also shows low activity in the non-VOC subset but is very active in the other VOC subsets, in particular Alpha, where it appears to be the most active, overtaking the Signature 1 process. Continent-defined subsets of the data also consistently showed the high activity of Signa- ture 1. Signature 2 begins to consistently appear in all continents after 2020, with only small bursts of activity being detected before this (Fig 5D), again consistent with what we see in the global data. Signature 3 activity also follows the pattern of the global activity, appearing most prominently during the Delta wave. Bridging the gap between mutation signatures and amino acid substitutions Stratifying non-synonymous nucleotide substitutions by their association with mutational sig- natures should provide insights into how these mutational processes affect viral proteins. Exposures were calculated by stratifying nucleotide mutations by whether they were synony- mous or non-synonymous substitutions for each dataset (Fig 6A). The unattributed exposure PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 9 / 26 PLOS COMPUTATIONAL BIOLOGY Mutational processes and SARS-CoV-2’s evolutionary capacity Fig 6. A. Exposures for each of the SARS-CoV-2 mutational signatures for both synonymous and non-synonymous stratified datasets. Synonymous exposures are below 0 on the y-axis, while non-synonymous exposures are above 0. Each area represents signature exposures across epidemic weeks, with colours representing which signature the exposures are attributed to. B. Non-synonymous and synonymous mutations in the tree-based references of identified variants of concern. Signature 1 produces the majority of both synonymous and non-synonymous substitutions in all lineages. Signature 3 mutations are more often non-synonymous substitutions in the lineages of concern, with most lineages having few to no changes. Signature 2 non-synonymous mutations appear to have increased in the Omicron lineages (BA.1 and BA.2). C. Variant of concern associated non-synonymous mutations coloured by the mutational signature with the greatest likelihood of causing the change. D. Variant of concern synonymous mutations coloured by the putative mutational process that caused the change. https://doi.org/10.1371/journal.pcbi.1011795.g006 PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 10 / 26 PLOS COMPUTATIONAL BIOLOGY Mutational processes and SARS-CoV-2’s evolutionary capacity was calculated using the model error for mutational categories not contained within any of the extracted mutational signatures. The majority of non-synonymous substitutions can be described by the observed mutational signatures. Signature 1 likely produces most of the non- synonymous mutations, however, Signature 3 is an almost exclusively non-synonymous signa- ture, with particularly high activity during the Delta wave of infections. Signature 2 appears to produce predominantly synonymous mutations. Using the tree-based references, we can also look at individual lineage reference sequences to observe which mutational processes have probably produced their specific amino acid sub- stitution set. The tree-based references were used since they are equivalent to a high-quality representative sequence and because many of the early real sequences contain sequencing errors. For each variant of concern, mutations were assigned to a signature by calculating the maximum likelihood of the mutation and its context being produced by each of the three extracted signatures. Using the trinucleotide context C[C ! T]G as an example, the likelihood function is P(C[C ! T]G j Signature), which corresponds to the probability bars for CT-CCT in the extracted signatures. Mutations that contained substitution-context pairs not found within any of the mutational signatures were labeled as “unattributed”. The Alpha VOC tree-based reference sequence contains eleven Signature 1 changes, six Sig- nature 2 changes and a single Signature 3 change. Signature 1 changes account for 39% of all substitutions within the Alpha tree-reference sequence, with 75% of these mutations being non-synonymous substitutions. Signature 1 was frequently active prior to the Alpha VOC’s emergence. The activity plots (Fig 4) show that this was the case for much of the pandemic, particularly prior to the Alpha’s emergence around September 2020. It should be noted that while Signature 1 mutations are by far the most frequent, only one is found within the Spike protein (producing the S:T716I change). Signature 3 only had one change, which was non-syn- onymous appearing in ORF:8. Signature 2 mutations were non-synonymous substitutions 83% of the time, with three Spike mutations relating to the process including S:D614G, which is present within all known variants of concern. The Beta VOC emerged around the same time as Alpha (Autumn 2020) and is defined by a smaller set of mutations. A greater proportion of Signature 1 mutations are non-synonymous substitutions in Beta (66%). Signature 2 mutations resulted in S:D215G and S:E484K, the latter reported to help the virus evade neutralising antibodies [25]. Signature 3 mutations most likely produced S:K417N in spike, which is also reported to aid in antibody evasion [25, 26] similar to S:E484K. Gamma also emerged in Autumn 2020 and has 33 different defining substitutions. Signa- ture 1 mutations account for 11 of these with 54% being non-synonymous. Four are present in Spike including S:L18F, S:P26S, S:H655Y and S:T1027I. Signature 2 mutations resulted in six amino acid substitutions, with only 75% of changes being non-synonymous. Three of the five mutations in non-synonymous substitutions occurred in Spike. Signature 3 mutations in the Gamma lineage were all non-synonymous except for a single synonymous substitution in ORF1a/b. Delta was the first VOC to dominate worldwide and replace almost every other lineage in all regions. The initial Delta sequence (Pango lineage B.1.617.2) contains six Signature 1 muta- tions. 66% of these changes were non-synonymous and none occurred within Spike. Signature 2 mutations were all non-synonymous and displaced throughout the virus ORFs including ORF1a/b, S and M. Signature 3 mutations in Delta are found in non-coding regions and N, with the N mutations both being non-synonymous. Omicron is the most recent VOC to emerge, quickly replacing Delta globally. Omicron dif- fers from earlier VOCs with a much greater number of Spike mutations relative to the other ORFs. The first identified Omicron variant B.1.1.529 has 40 substitutions of which 32 are non- PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 11 / 26 PLOS COMPUTATIONAL BIOLOGY Mutational processes and SARS-CoV-2’s evolutionary capacity synonymous changes. This is almost double that of Delta, which only had 18. Seven of these substitutions were Signature 1 changes, two were Signature 3 and ten were Signature 2 changes. There are four non-synonymous ORF1a/b mutations despite this ORF being substan- tially longer than SARS-CoV-2’s other ORFs. Only one Spike substitution was synonymous out of the 21 total changes. This number is even greater when looking at the major Omicron variants BA.1 and BA.2. BA.1 had 31 non-synonymous substitutions in Spike alone while BA.2 had 28. Between these three Omicron variants, only two Spike substitutions are non-synony- mous out of a total of 40. Nine of the 40 changes are from Signature 1, 2 are from Signature 3 and 12 are from Signature 2. This means 23/40 of the changes appear to come from these three mutational processes. 20 of the 40 substitutions observed in these variants were present in the receptor-binding domain (RBD) of Omicron, with nine of these changes thought to help Omi- cron evade the immune response or increase its transmissibility [27]. Of these beneficial RBD changes, three are potentially the result of Signature 1 activity, 9 are Signature 2 and one is from Signature 3. The high density of Signature 2 RBD amino acid changes in a variant that has emerged as Signature 2 exposure increased suggests that the mutational process behind Signature 2 may have contributed to the emergence of the Omicron variant. Signature exposures and highly mutated sequences in wastewater data Similar trends over time in exposures are seen when the mutational signatures are applied to publicly available wastewater data. Although the trend is seen at a lower resolution than global data, Signature 1 and Signature 3 are gradually replaced by Signature 2 (Fig 7A). Although, Sig- nature 2 is not quite as strong as in the global data (Fig 4). This suggests trends in mutational processes can be monitored using wastewater, not only sequencing of the infected population. Additionally, at time periods where a high level of virus diversity is expected, there are highly mutated sequences present in the wastewater (Fig 7C). This suggests cryptic sequences in wastewater may be used to observe potential upcoming variants, similar to how known sequences have been back-traced to particular buildings using wastewater [28]. As chronic SARS-CoV-2 infections are implicated as a major contributor to VOC evolution [29, 30], it may be possible to parse highly-mutated cryptic sequences of interest from chronic infections out of wastewater data in the interest of detecting potential VOCs. Unfortunately, this is problematic to deconvolve as sequencing data for immunocompromised and chroni- cally infected individuals is sparse. When sequences from known chronic infections are exam- ined, the distribution of mutation types is consistent with global data, with Signature 1 mutations dominating as expected for samples from January 2022 (Fig 7B). Although, due to the low number of chronic infections for comparison this result is not very conclusive, it does demonstrate how mutational patterns can be potentially detected in this type of data. Studying these types of infections, and underlying mutational processes, will be important to under- stand better the origins of the sets of mutations that contribute to the generation of VOCs. Discussion In this study, we investigated SARS-CoV-2 lineage dynamics and identified temporal variables that are associated with increased numbers of infection cases. Both public health measures and virus properties were associated with the sequential waves of regional SARS-CoV-2 infections cases. These predictors have varying impact in different geographical locations. As more of the global population’s immune system becomes sensitised to existing SARS-CoV-2 variants, either through previous infection or vaccination, the virus has and will continue to undergo changes that enable reinfections. The continued emergence of new variants is thus expected. In some regions, government stringency had limited significant impact on patterns of PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 12 / 26 PLOS COMPUTATIONAL BIOLOGY Mutational processes and SARS-CoV-2’s evolutionary capacity Fig 7. A. Signature exposures per month from wastewater sequences show similar trends in mutational processes as the global data, although at a lower resolution and, interestingly, with a lower Signature 2 exposure. B. Substitutions in SARS-CoV-2 consensus sequences from infections of immunocompromised individuals contain mutation types corresponding with patterns observed in the distinct signatures. Of note, there are more synonymous mutations present in the chronic infection data than in the global sequences, although it is important to note the sample size for immunocompromised infections is low. C. Mutation counts in wastewater sequences for bi-yearly time periods. Highly mutated sequences cluster to the right especially during the 2021 July-December time period, as would be expected when Omicron was emerging. https://doi.org/10.1371/journal.pcbi.1011795.g007 infection. This could be due to differences in implementation strategies and support, other competing predictor variables, as well as behavioural changes in citizens as a response to the restrictions. Our analysis highlights the significant role of vaccination in influencing reported COVID- 19 case patterns across all continents, even in regions with lower vaccination coverage like Africa. Despite Africa’s lower vaccination rates, the continent has seen a relatively low-level of sustained transmission. This phenomenon might be attributed to factors such as the younger median age of the population, lower population density, immune priming due to prevalent PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 13 / 26 PLOS COMPUTATIONAL BIOLOGY Mutational processes and SARS-CoV-2’s evolutionary capacity infectious diseases, and limited testing capacity [31]. The weak impact of viral diversity on reported cases in Asia and South America may be explained by the emergence and dominance of variants such Delta and Gamma in the regions, respectively. For instance, the Delta variant, initially identified in Asia, quickly became the predominant strain, overshadowing other line- ages before spreading globally. Overall, the predictor variables significantly contributed to explaining the rise and fall of infection numbers across different continents, accounting for more than half of the variance in reported cases. The differences in the regression effectiveness can be attributed to intrinsic differences among continents, such as variations in vaccine cov- erage, testing and sequencing capabilities, and the effectiveness of government stringency measures. While our model effectively captured the general trends of infection waves, it struggled to accurately represent peaks within short time-frames in some countries. This discrepancy might be attributed to the omission of certain predictor variables, like mass gatherings, which are known to contribute to viral super-spreading events [32]. In utilizing the OWID and OxCGRT datasets, which are arguably among the most compre- hensive for addressing our research objectives, we note some limitations. First, there were dis- crepancies in parameter definitions, such as varying case classifications across regions. Second, positive tests are commonly labeled based on their reporting date rather than “date-of-event” [33]. Lastly, the cases reported in these datasets may not be fully representative of the actual disease burden. Although the Human Development Index (HDI) of a country can act as a proxy to bridge the gap between reported cases and the true disease burden, it does not fully capture the entire complexity. The extracted signatures from the global SARS-CoV-2 dataset show clear and distinct pat- terns describing mutational processes acting on the viral genome. The most prominent of these signatures, Signature 1 (Fig 3 and S5 Fig), shows a marked bias towards C!T mutations, a signal indicative of the APOBEC family of cytidine deaminases [17, 18]. APOBEC enzymes have been shown to cause extensive C!T editing of DNA and RNA in human and viral genomes. However, it is not yet clear whether they are the cause of this pronounced C!T bias in SARS-CoV-2 despite a number of other studies also observing other APOBEC-like muta- tional patterns [34–37]. Cytosines flanked by either an adenine or thymine in both the 3’ and 5’ direction appear to be the most pronounced targets of Signature 1. APOBEC editing was shown to have contexts outside of the traditional TpC when structural features of the nucleic acid such as hairpin loops are present [38]. Outside of structural features, APOBEC3A is thought to be the predominant cause for TpC changes and is found to be expressed in lung tis- sue [39]. ApC changes are considered to be caused by APOBEC1, which in cell models was shown to efficiently edit SARS-CoV-2 RNA [39]. APOBEC1 is found predominately in the liver and small intestine, tissues reported to be infected by SARS-CoV-2 [39, 40]. 3’ CpG nucle- otide contexts are the most targeted, in particular TCG, CCG and ACG. CpG suppression is a well-known dinucleotide bias. In RNA viruses, this appears to be a result of selective pressures exerted from the presence of host CpG sensing molecules such as Zinc-finger Antiviral Protein (ZAP). ZAP relies on host CpG suppression to allow it to specifically target non-host genomic material (such as viral RNA) with higher CpG content [41]. This allows viruses with lower CpG content to better evade restriction by ZAP since it more closely resembles the host CpG composition. While ZAP does not induce C!T changes, it may help explain why C!T sites in a CpG 3’ context are preferentially edited relative to other 3’ contexts. ZAP has been shown to restrict SARS-CoV-2 despite pre-existing CpG depletion [42]. ZAP isoforms have been shown to prevent necessary translational frame-shifting for SARS-CoV-2 ORF1b protein pro- duction. [43]. The non-normalised form of Signature 1 (S5 Fig) shows that when tri-nucleotide bias is not accounted for 3’ CpG’s are lower than the normalised signatures, yet 5’ TpC and PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 14 / 26 PLOS COMPUTATIONAL BIOLOGY Mutational processes and SARS-CoV-2’s evolutionary capacity ApC contexts remain the most prevalent(S5 Fig). The most targeted contexts do shift to ACA, ACT and TCT, likely reflecting their comparatively high abundance within the SARS-CoV-2 genome relative to 3’CpG contexts. These non-normalised contexts are consistent with what was earlier reported by Simmonds et al. [17]). Signature 2 (Fig 3 and S5 Fig) has a nearly identical proportion of A!G and T!C muta- tions. These are a known target of the ADAR family of adenine deaminases. ADAR enzymes typically operate on double-stranded RNA and convert adenine into inosine [21, 22]. Inosine forms base pairs with cytosine, which after another round of replication causes guanine to replace the inosine and complete the A!G change. As ADAR operates on both strands of dsRNA, the mutational signature resulting from the process is expected to contain an equal proportion of A!G and T!C mutations, which is the case for Signature 2 [21]. Signature 2 also contains a number of G!A mutations, which could be caused by low-level C!T activity on the negative sense RNA strand. Due to the cellular strand biases present between the posi- tive and negative sense RNA [36], C!T mutational processes acting on ssRNA are much less likely to produce a mutation on the negative strand (resulting in G!A substitutions) than C!T changes on the positive strand. The negative strand will only be present during the repli- cation phase of the virus while the positive strand will be present both on cell entry and on exit as the new viral particles are packaged to infect further cells. This could explain why the nega- tive sense Signature 1 changes are present in Signature 2, since it may be operating at a similar level to Signature 2 on the negative strand. The non-normalised form of Signature 2 (S5 Fig) does have different targeted contexts, just as with Signature 1. However, the main attribute of Signature 2 is its equal contributions of A!G and T!C substitutions, which still remain equal. Signature 3 (Fig 3 and S5 Fig) is dominated by G!T substitutions. A putative mechanism for this is Reactive Oxygen Species(ROS) in the cell. Increases in oxidative stress as part of a ROS ‘burst’ have been associated with viruses during the early stages of infection [34, 44]. Gua- nine nucleotides are known to be vulnerable to oxidation, with the product 7,8-dihydro-8-oxo- 2’-deoxyguanine (oxoguanine) pairing with adenine bases rather than cytosine [44, 45]. Similar to inosine causing A!G changes, this change to oxoguanine will result in a G!T mutation after a replication cycle. The lack of C!A changes in the signature also suggests that the mech- anism is most active on the positive single-stranded RNA rather than the negative single- stranded RNA. The initial positive single-stranded RNA is found in the cytoplasm, meaning it can be easily accessed by ROS and other mechanisms of mutation. Viral replication is thought to take place within membrane-bound environments that aim to protect the RNA. The pres- ence of double-stranded RNA within these environments strongly suggests that this is the case [46] and may explain the relative lack of negative strand mutations in SARS-CoV-2 signatures. The non-normalised G!T signature (S5 Fig) seems to display a context preference of TpG and ApG nucleotides, although this contextual bias is changed to CpG and ApG following normali- sation. These contextual biases mean that the signature could be some other as yet unknown editing mechanism on the viral RNA, although normalisation changing this context so heavily suggests that this bias perhaps has more to do with genome composition. The increased CpG context shift post-normalisation could also be another ZAP-induced effect, where CpG deple- tion is selected for to help the virus evade ZAP. Curiously, this G!T bias has been observed in other coronaviruses, but not widely among RNA viruses [47]. ROS has a verified cancer muta- tional signature [15, 48] although the context preferences do not match the signatures (normal- ised or non-normalised) observed here. However, there are a multitude of differences between viral RNA and human DNA that make these signatures difficult to compare. It is important to note that while SARS-CoV-2 does have an error correction mechanism resulting in fewer replicase-induced errors, this mechanism will not catch all changes. A PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 15 / 26 PLOS COMPUTATIONAL BIOLOGY Mutational processes and SARS-CoV-2’s evolutionary capacity number of the mutations picked up from the set of sequences (and included in our mutational signatures) will be derived from replication errors. However, the clear and repeatable extrac- tion of the signatures indicates that despite this potential contamination, the extracted signa- tures do appear to be predominantly other mutational processes. While a replication error- associated mutational signature may be identified in future, this signature is too diffuse to identify as a distinct process. Similarly, a high proportion of mutations are not accounted for by the extracted mutational signatures. These mutations were not present in large enough quantities to enable effective extraction from the data. Future methods may be able to tease out the more subtle mutational mechanisms that almost certainly exist to induce these less com- mon mutation types. Signature activities clearly change in both the global dataset and in the various subsets of the data for VOCs and continents. In the global data (Fig 4) Signature 1 is dominant through- out the pandemic. Signature 2 only begins to appear around November 2020, after which it appears consistently active for the remainder of the pandemic. This is approximately when variant of concern lineages began to emerge, as well as the beginning of the first vaccine roll- outs. This is particularly apparent in the Alpha subset where Signature 2 is the most highly active mutational process (Fig 5), with a large depletion of Signature 1 activity as well. Alpha was shown to increase sub-genomic RNA expression of several immune-antagonist viral proteins including nucleocapsid (N), ORF9b and ORF6 [49–52]. N is thought to shield dsRNA from detection by RNA sensors, which trigger downstream antiviral response path- ways [49, 52–54]. ORF9b antagonises TOM70, a protein required for the activation of mito- chondrial antiviral-signalling proteins (MAVS) [49] while ORF6 inhibits the transportation to the nucleus of inflammatory transcription factors [55]. Combined, the cumulative immune inhibition may have resulted in an observable change in the mutational processes that we observe within the Alpha lineage. Beta and Gamma (both VOCs that emerged around the same time as Alpha) gained amino acid substitutions that helped evade the immune system primarily via antigenic change. Alpha’s reliance on attenuating immune pathways rather than antibody binding may be why we see a different signature exposure pattern in this VOC rela- tive to the others. This could be due to the attenuated pathways being involved in signalling for the mutational processes behind Signatures 1 and 3, while not inhibiting Signature 2 as much. This Alpha pattern is not observed in the other VOC datasets, although Delta and Omicron have a high level of Signature 2 exposure as well, despite Signature 1 remaining the dominant process in those subsets. Signature 3 appears to be most prominently found in the Delta subset and remains consistently at low levels in the global data until January 2022 when it appears to disappear almost entirely. The Omicron subset has little to no exposure for Signature 3 and this happens to be the VOC almost exclusively circulating after January 2022. Why Omicron appears to have so little Signature 3 exposure is unclear, although unlike previous VOCs, Omi- cron differs in its preference of cell entry mechanism. Previous variants of the virus typically enter the cell using membrane fusion, where the viral membrane fuses with the cell membrane via the action of ACE-2 receptor binding and TMPRSS2 cleavage of the spike protein. Omi- cron instead favours an endosomal route of entry whereby the viral particle binds to the cell using ACE-2 and is enveloped by endocytosis into the cell. Cleavage of the spike protein then occurs via the action of Cathepsin L, which allows for the release of the viral RNA into the cytoplasm of the now-infected cell [56, 57]. Signature transitions from Signature 1 to Signature 2 changes occur from December 2020 onwards in the global dataset and appears consistently in the VOC and continent-defined sub- sets around this time point as well. Alpha underwent a major shift to Signature 2 mutations early in its time as a VOC, although Signature 1 returned as the predominant set of changes towards the end of its wave of infections. The non-VOC subset appears to be the least impacted PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 16 / 26 PLOS COMPUTATIONAL BIOLOGY Mutational processes and SARS-CoV-2’s evolutionary capacity by Signature 2 changes. However, this can mostly be explained by the number of non-VOC sequences quickly declining after the emergence of the VOC lineages. Delta underwent a dra- matic increase in Signature 2 and Signature 3 exposure from July 2021, with Signature 2 becoming the predominant signature towards the end of Deltas wave. Signature 2 changes continue into Omicrons introduction, although it does decrease after the initial BA.1 wave from December 2021 to March 2022. It seems clear that while Signature 1 mutations have dominated in contributing to the evolutionary capacity of SARS-CoV-2 throughout the pan- demic, this mutational environment is beginning to change. Such shifts in mutational pro- cesses are potentially evidence of changing interactions between the viruses and the immune systems of the hosts they circulate within. For example, changes in population-level immunity via vaccination or previous infections may influence the mutations that we observe in the data. Changing mutational process activity in consensus sequences from infections is unlikely to fully reflect the true activity of each process, but they are likely to show which processes are contributing mutations that eventually make it into circulating viruses. All variants of concern we assessed show predominantly non-synonymous mutations and all mutational signatures are associated with more non-synonymous than synonymous changes. More synonymous substitutions in the lineage references were found in ORF1a/b, which is expected due to it being the longest ORF. However, this pattern is not observed with non-synonymous mutations as these are mainly located in the spike protein (Fig 6C and 6D). This is consistent with spike being under intense immune pressure since it is the main glyco- protein for SARS-CoV-2. As such, spike must change in order to escape the host immune response, while maintaining its main function of binding and entry into host cells. Signature 1 changes are the predominant source of mutations in all SARS-CoV-2 VOCs that we analysed, followed by unattributed mutations, Signature 2 changes and Signature 3 changes. Signature 3 changes were unlikely to be synonymous mutations with only Beta, Gamma and Delta con- taining very few such changes (Fig 6D). This is also reflected in the global synonymous/non- synonymous exposures where Signature 3 appears completely inactive in the synonymous mutation subset (Fig 6A). Signature 2 exposure appears the most likely to be synonymous mutations (Fig 6A) but this does not seem to be observed in the VOC lineages where most Sig- nature 2 changes are non-synonymous mutations (Fig 6B). In conclusion, mutational signature analysis reveals important processes contributing to SARS-CoV-2 genetic variation and serves as a tool to track the dominant changes over time and to generate hypotheses about the main mechanistic processes in play. Specifically, host antiviral molecules as opposed to replication errors appear to be a the main generator of muta- tions (confirming earlier computational studies), a result that requires experimental confirma- tion. Despite limitations in potential biases, our findings contribute to a better understanding of the complex dynamics driving the evolution of SARS-CoV-2 and the emergence of VOCs. Methods Data The findings of this study are based on metadata associated with 13,281,213 sequences avail- able on GISAID up to October 26, 2022 and accessible at doi.org/10.55876/gis8.221201qs. Sequences were filtered to remove records from non-human hosts, with lengths less than 20,000 nucleotides, non-assigned lineages, with greater than 30% unknown bases, sequences reported to be collected before 24/12/2019 and those with excessive mutations/deletions. The cutoff for filtering out hypermutated sequences was 175 mutations in coding regions or more than 69 different deletions, the cutoffs were manually determined after evaluation of the PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 17 / 26 PLOS COMPUTATIONAL BIOLOGY Mutational processes and SARS-CoV-2’s evolutionary capacity mutation/deletion distribution and selecting the point where sequence counts were consis- tently observed in single digits, this resulted in 1,852 sequences being filtered out. Publicly available daily SARS-CoV-2 cases, tests performed and total vaccinations per capita were obtained from OWID [58] in September 2022. Prior to February 2023, the OWID data was piped from the Johns Hopkins University COVID-19 dashboard [33, 59]. Country-level government stringency indices were downloaded from OxCGRT [60]. Government stringency indices are composed of nine indicators: school closure, workplace closure, cancellation of public events, stay at home order, public information campaigns, restrictions on public gather- ings, public transport, internal movement and international travel. The index on a given day ranges from 0 to 100 and is calculated as the mean of the nine indicators, with higher indices indicating stricter regulations. If responses vary at sub-national levels, the index at the strictest level is used [60]. Wastewater findings are based on metadata associated with 1,343 sequences available on GISAID and accessible at doi.org/10.55876/gis8.230406qg. Wastewater sequences were down- loaded from the ‘wastewater data’ section of GISAID in December 2022. Sequences for immunocompromised individuals were downloaded from GISAID in November 2022. Analysis of this was based on the metadata associated with 34 sequences avail- able on GISAID and accessible at doi.org/10.55876/gis8.230406fb. Sequences were chosen based on the known list of sequences used in [30]. Sequences were aligned to the COVID refer- ence genome before use. Design Predictors of SARS-CoV-2 reported cases were explored using a linear model at both country and continent levels. We collected continuous dependent variables reported on a daily basis. These were classified into two groups: (i) public health measures (government stringency, test- ing capacity and vaccination), (ii) viral properties (diversity and fitness). We examined the data for completeness of predictive variables. In instances of missing vaccination data, we interpreted this as no vaccinations having been given. This was a reasonable assumption for periods prior to the vaccine rollouts in the respective countries. With the exception of vaccina- tions, variables with less than 70% of the countries reporting data were not included. The num- ber of SARS-CoV-2 diagnostic tests performed was excluded as a predictor due to missing data. We determined the previous burden by summing the adjusted new cases per capita over the past 90 days. Prior infection significantly reduces the risk of a subsequent infection, with a reduction in risk of up to 95% in the initial three months [61]. This was included as a predictor variable in the linear model. Amino acid substitutions were defined against the Wuhan-Hu-1 sequence. Building on findings from Obermeyer et al., we extracted a list of previously identified fitness-associated mutations [62]. Each fit mutation within a sequence was counted and the counts were normal- ized to the number of sequences per geographical location. Virus fitness was therefore defined as the sum of the frequencies of previously identified [62] amino acid substitutions that increase SARS-CoV-2 fitness divided by the sum of total genomes and the log of total muta- tions per location. Virus Fitness ¼ weekly sum of fit mutations total seqs per week þ logðtotal mutations per weekÞ Diversity was calculated by dividing distinct lineages by the total number of genomes in a given week. Sequences reported in GISAID were assumed to be representative of the diversity of infections for that continent/country. PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 18 / 26 PLOS COMPUTATIONAL BIOLOGY Mutational processes and SARS-CoV-2’s evolutionary capacity Linear model We employed a linear regression model, described by Heo et al. [63], to adjust reported cases per country using the Human Development Index (HDI), which encompasses not just eco- nomic growth but also reflects a country’s capacity for per capita testing. Countries with higher HDI levels, typically high-income nations, conducted more tests per million people, often lead- ing to more confirmed cases compared to nations with lower HDI levels. Adjusted daily cases were smoothed using a 14-days rolling average to limit possible noise and identify simplified changes over time. For continent-level analysis, data from all contributing countries was used to fit the linear model. To ensure that countries with a large number of cases didn’t artificially inflate the results, each country’s influence on the continent-level OxCGRT index was adjusted based on its percent contribution to the continent’s 14-day average daily case tally. Pearson’s correlation was used to test for correlation among the variables. Multiple linear regression was fitted to evaluate the relationship between infection rate (adjusted daily cases per capita) as the outcome and the public health measures and viral properties as predictors within the different continents. The regression models were fitted on data from 01 April 2020 onwards, as (sequence) data addition remained stable after this. The country-level analysis was carried out for countries with less than 50 days of missing genome data using a similar approach. Pandemic plots Case numbers and sequence data were aggregated by their respective continents, a 14-day roll- ing average was used to smooth out daily infection rates and categorical variables were sum- marised by counts. Proportions of lineages were calculated in 14-days bins and the most common lineages were visualised per continent. Tree-based referencing The rapid evolution of SARS-CoV-2 means that the majority of viral sequences are distinct from the early pandemic reference genome Wuhan-Hu-1 [64]. Continuing to count mutations against the early reference sequence can result in mutations being allocated the wrong substi- tution category (i.e., A!T instead of a C!T) where sites have mutated multiple times. Azgari et al. [35] tackled this issue by building a tree of clustered sequences to remove ancestral muta- tions. However, we utilise the available SARS-CoV-2 tree generated as part of the Pango [8] nomenclature to generate a reference sequence for each defined lineage. This means that sequences from the lineage B.1 are compared against a generated reference sequence for the B lineage rather than the Wuhan-1 sequence (See S3 Fig for diagrammatic description). One reference sequence was generated for each of the Pango lineages in the alignment. A nucleotide was included in the generated Pango reference if it exceeded a frequency threshold of greater than 75% of the samples from the lineage. If this threshold was not reached, the ref- erence nucleotide of the nearest parental lineage was used (i.e., if a mutation in B.1 is ambigu- ous, the nucleotide from the B lineage reference at that position is used). Building intermediate references also meant that counting inherited mutations could be avoided. Since mutations were identified relative to their nearest parental Pango lineage, inherited mutations are not counted because, relative to this sequence, there hasn’t been a mutation. Mutations are also only counted once per lineage set of sequences so that mutations that are observed many times due spread of the virus rather than acquisition by a mutational process are not over-counted. This means that convergent amino acid substitutions can be observed between lineage sets, although they may be undercounted within a lineage. However, this is necessary since it is very difficult to identify convergence within similar sequences (especially at a global scale). PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 19 / 26 PLOS COMPUTATIONAL BIOLOGY Mutational processes and SARS-CoV-2’s evolutionary capacity Overcounting of the mutations results in mutational signatures that reflect the circulating pre- dominant lineages rather than the mutational processes producing the mutations in those lineages. Pseudo-sampling Mutations were binned into categories composed of their substitution type (e.g., cytosine ! thymine = CT) and their mutation context. The mutation context is the mutated base and the nucleotides at the 5’ and 3’ positions of the mutated base. There are a total of 192 types of substitution-context matchings that can appear (12 possible single nucleotide changes x four possible nucleotide 5’ x four possible nucleotide 3’). Every sequence produces a single count vector of mutation category counts, with the total count matrix becoming the muta- tional catalogue of the virus. On average, a single SARS-CoV-2 genome sequence has very few new mutations. As extracting mutational signatures when mutation counts are low is unlikely to produce meaningful results, we define each sample as a time-point (all of the sequences collected in an epidemic week) and decompose signatures from the counts at each time-point rather than from each sequence. This shrinks the mutational catalogue of the virus from millions of samples down to less than 200 samples, one for each Epidemic Week. Non-negative matrix factorisation NMF (non-negative matrix factorisation) [65, 66] was used to split the mutational catalogue into two sub-matrices. One matrix represents the mutational signatures, the other matrix rep- resents the exposure of the signatures. These matrices were used to reconstruct the original mutational catalogue with some degree of error. To verify the validity of the identified signa- tures, NMF was performed 100 times for each value of N, with N representing the number of signatures to extract from the mutational catalogue. For this analysis, N was set to 2, . . ., 10. For each NMF run, a new mutational catalogue was generated using bootstrap re-sampling of the original matrix and removal of any mutational categories that did not account for more than 0.5% of mutations. Mutational categories are pseudo-sampled down into epidemic week matrices that NMF was run on. The signatures were then clustered together using K-means clustering, with the cluster means forming the new signatures. Clusters were then assessed using the silhouette score to determine the clustering quality. Clusters with high silhouette scores are well separated from other clusters and are dense and well-formed. Cosine similarity was used to determine if the signature was reliably extracted from the cluster. The cosine simi- larity was calculated between signatures extracted from the whole mutational catalogue and the cluster means of the signature clusters. A higher cosine similarity indicates that the cluster mean shows a similar pattern to the initial mutational signature. Following the best practices in Islam et al. [66], an N value of three was selected due to the reduction of the reconstruction error plateauing around three and the marked decrease in silhouette score for signatures greater than 3. The average cosine similarity between signatures and clusters was consistently above 0.95 for each cluster and had an average of 0.98 for all three clusters when clustering was repeated 100 times. Silhouette scores for each cluster were above 0.95, suggesting excellent sep- aration and density of clusters (S5 Table and S9 Fig). Signatures can therefore be reliably extracted from the bootstrapped catalogues, are robust and thus are unlikely to be artefacts. Counts of mutations were normalised by the tri-mer composition of the SARS-CoV-2 refer- ence sequence (dividing the counts by the number of contexts in the reference sequence). Composition biased versions of the signatures were then produced by rescaling the signatures using tri-mer composition. PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 20 / 26 PLOS COMPUTATIONAL BIOLOGY Mutational processes and SARS-CoV-2’s evolutionary capacity Non-negative least squares regression A non-negative least squares (NNLS) Regression was used to produce positive exposure weights for each of the signatures in each of the datasets. The non-negativity of the regression ensures that the weights of the signatures continue to represent an additive process. The NNLS weights can then represent the exposures of the signatures on each dataset. Consensus lineage and continent signatures Mutational catalogues were constructed for each continent and each of the Variant of Concern (VOC) lineages (Alpha, Beta, Gamma, Delta and Omicron). The global signatures were then used to extract exposures for each of the mutational catalogues to determine how processes varied between each mutational catalogue subset. VOC sequence sets were filtered so that weeks with fewer than 100 sequences were excluded. Supporting information S1 Fig. Country-level SARS-CoV-2 lineage dynamics. Solid bars show the biweekly propor- tions of the common lineages. Bars are coloured by lineage and white space shows the propor- tion of sequences from other lineages. The countries included in this analysis is based on temporal data completeness. (TIF) S2 Fig. Model-fitting of country-level SARS-CoV-2 reported cases. Black solid lines show a 14-day rolling average of adjusted SARS-CoV-2 cases. Pink solid lines show fitted mean response values of infection rates with predictor values as input and grey shaded areas high- light the confidence intervals. The countries included in this analysis is based on temporal data completeness. (TIF) S3 Fig. Diagrammatic depiction of how tree-based referencing works. Each Pango lineage has a reference generated for it. Arrows show which sequences use which reference sequence, with the arrow tip indicating the reference. For example, sequences from the B.1 lineage are compared against the reference for the B lineage so that B.1 lineage-defining mutations can be counted. (TIF) S4 Fig. Graphical description of the methods for NMF extraction of mutational signatures. For every value of N signatures, the mutational signatures are extracted 100 times for boot- straped and pseudo-sampled datasets. Once this has been completed, signatures are clustered into N clusters and the stability and density of those clusters are evaluated using the silhouette score. Signatures that have silhouette scores above 0.95 are evaluated as stable signatures. The cluster means become the extracted signatures. The best set of N signatures is selected by pick- ing the value of N that best minimises the reconstruction error and has the best silhouette score (with a minimum of 0.95). A further evaluation is the cosine similarity of the clustered signa- ture means with the signatures extracted by completing NMF on the original pseudo-sampled dataset. Again, signatures must have a cosine similarity of at least 0.95 to be considered. (TIF) S5 Fig. Non-normalised mutational signatures for SARS-CoV-2. Signatures were extracted using normalised counts calculated by dividing the mutation counts by the count of the tri- nucleotide context of the mutation context (Fig 4). These signatures were then multiplied post-analysis by the tri-nucleotide composition of the reference sequence to produce the non- PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 21 / 26 PLOS COMPUTATIONAL BIOLOGY Mutational processes and SARS-CoV-2’s evolutionary capacity normalised signatures shown here. (TIF) S6 Fig. Counts of unique substitutions per week of the pandemic. Areas are coloured by substitution category. (TIF) S7 Fig. Counts of unique substitutions per week of the pandemic for each VOC category. Areas are coloured by substitution category. (TIF) S8 Fig. Counts of unique substitutions per week of the pandemic for each continent cate- gory. Areas are coloured by substitution category. (TIF) S9 Fig. Signature evaluation metrics. The number of signatures was selected at N = 3 since this produced an “elbow” for the reconstruction error while having a suitable silhouette score greater than 0.95. (TIF) S1 Table. Proportion of common lineages/variants globally. (XLSX) S2 Table. Correlation between infection rate and predictor variables across different conti- nents. (XLSX) S3 Table. Effect of public health measures (government stringency and vaccination) and viral properties (diversity and fitness) on infection rates at continent level. (XLSX) S4 Table. Effect of public health measures (government stringency and vaccination) and viral properties (diversity and fitness) on infection rates at national levels. (XLSX) S5 Table. Evaluation Results for Signature with N = 3. (XLSX) Acknowledgments We gratefully acknowledge all data contributors, i.e., the authors and their originating labora- tories responsible for obtaining the specimens and their submitting laboratories for generating the genetic sequence and metadata and sharing via the GISAID Initiative, on which this research is based. For the purpose of open access, the author has applied a Creative Commons Attribution (CC BY) licence to any Author Accepted Manuscript version arising. We thank Spyros Lytras, Francesca Young, Sejal Modha, Andres Gomez and Procheta Sen for their help- ful comments throughout the process of writing and preparing this manuscript. Author Contributions Conceptualization: Kieran D. Lamb, Ke Yuan, David L. Robertson. Data curation: Richard J. Orton. Formal analysis: Kieran D. Lamb, Martha M. Luka, Megan Saathoff. PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 22 / 26 PLOS COMPUTATIONAL BIOLOGY Mutational processes and SARS-CoV-2’s evolutionary capacity Funding acquisition: Matthew Cotten, Ke Yuan, David L. Robertson. Investigation: Kieran D. Lamb, Martha M. Luka, Ke Yuan, David L. Robertson. Methodology: Kieran D. Lamb, Ke Yuan. Resources: Kieran D. Lamb. Software: Kieran D. Lamb. Supervision: My V. T. Phan, Matthew Cotten, Ke Yuan, David L. Robertson. Visualization: Kieran D. Lamb. Writing – original draft: Kieran D. Lamb, Martha M. Luka, Megan Saathoff. Writing – review & editing: Kieran D. Lamb, Martha M. Luka, My V. T. Phan, Matthew Cot- ten, Ke Yuan, David L. Robertson. References 1. Yang X, Yu Y, Xu J, Shu H, Xia J, Liu H, et al. Clinical course and outcomes of critically ill patients with SARS-CoV-2 pneumonia in Wuhan, China: a single-centered, retrospective, observational study. The Lancet Respiratory Medicine. 2020; 8:475–481. https://doi.org/10.1016/S2213-2600(20)30079-5 PMID: 32105632 2. Li Q, Guan X, Wu P, Wang X, Zhou L, Tong Y, et al. Early Transmission Dynamics in Wuhan, China, of Novel Coronavirus–Infected Pneumonia. New England Journal of Medicine. 2020; 382:1199–1207. https://doi.org/10.1056/NEJMoa2001316 PMID: 31995857 3. Petersen E, Koopmans M, Go U, Hamer DH, Petrosillo N, Castelli F, et al. Comparing SARS-CoV-2 with SARS-CoV and influenza pandemics. The Lancet Infectious Diseases. 2020; 20:e238–e244. https://doi.org/10.1016/S1473-3099(20)30484-9 PMID: 32628905 4. da Silva Filipe A, Shepherd JG, Williams T, Hughes J, Aranday-Cortes E, Asamaphan P, et al. Geno- mic epidemiology reveals multiple introductions of SARS-CoV-2 from mainland Europe into Scotland. Nature Microbiology 2020; 6:112–122. https://doi.org/10.1038/s41564-020-00838-z PMID: 33349681 5. Dewi A, Nurmandi A, Rochmawati E, Purnomo EP, Rizqi MD, Azzahra A, et al. Global policy responses to the COVID-19 pandemic: proportionate adaptation and policy experimentation: a study of country policy response variation to the COVID-19 pandemic. Health Promotion Perspectives. 2020; 10:359. https://doi.org/10.34172/hpp.2020.54 PMID: 33312931 6. Kirby T. New variant of SARS-CoV-2 in UK causes surge of COVID-19. The Lancet Respiratory medi- cine. 2021; 9:e20–e21. https://doi.org/10.1016/S2213-2600(21)00005-9 PMID: 33417829 7. Lauring AS, Hodcroft EB. Genetic Variants of SARS-CoV-2—What Do They Mean? JAMA. 2021; 325:529–531. PMID: 33404586 8. Rambaut A, Holmes EC, O’Toole A´ ine, Hill V, McCrone JT, Ruis C, et al. A dynamic nomenclature pro- posal for SARS-CoV-2 lineages to assist genomic epidemiology. Nature Microbiology 2020; 5:1403– 1407. https://doi.org/10.1038/s41564-020-0770-5 PMID: 32669681 9. Harvey WT, Carabelli AM, Jackson B, Gupta RK, Thomson EC, Harrison EM, et al. SARS-CoV-2 vari- ants, spike mutations and immune escape. Nature Reviews Microbiology 2021; 19:409–424. https:// doi.org/10.1038/s41579-021-00573-0 PMID: 34075212 10. WHO. Coronavirus Disease (COVID-19) Situation Reports; 2022. Available from: https://www.who.int/ emergencies/diseases/novel-coronavirus-2019/situation-reports. 11. Tegally H, Wilkinson E, Giovanetti M, Iranzadeh A, Fonseca V, Giandhari J, et al. Detection of a SARS- CoV-2 variant of concern in South Africa. Nature 2021; 592:438–443. https://doi.org/10.1038/s41586- 021-03402-9 PMID: 33690265 12. Bugembe DL, Phan MVT, Ssewanyana I, Semanda P, Nansumba H, Dhaala B, et al. Emergence and spread of a SARS-CoV-2 lineage A variant (A.23.1) with altered spike protein in Uganda. Nature Micro- biology 2021; 6:1094–1101. https://doi.org/10.1038/s41564-021-00933-9 PMID: 34163035 13. Mlcochova P, Kemp SA, Dhar MS, Papa G, Meng B, Ferreira IATM, et al. SARS-CoV-2 B.1.617.2 Delta variant replication and immune evasion. Nature. 2021; 599:7883. https://doi.org/10.1038/s41586-021- 03944-y PMID: 34488225 PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 23 / 26 PLOS COMPUTATIONAL BIOLOGY Mutational processes and SARS-CoV-2’s evolutionary capacity 14. Alexandrov LB, Stratton MR. Mutational signatures: the patterns of somatic mutations hidden in cancer genomes. Current opinion in genetics & development. 2014; 24:52–60. https://doi.org/10.1016/j.gde. 2013.11.014 PMID: 24657537 15. Alexandrov LB, Nik-Zainal S, Wedge DC, Aparicio SAJR, Behjati S, Biankin AV, et al. Signatures of mutational processes in human cancer. Nature 2013; 500:415–421. https://doi.org/10.1038/ nature12477 PMID: 23945592 16. Forbes SA, Beare D, Boutselakis H, Bamford S, Bindal N, Tate J, et al. COSMIC: Somatic cancer genetics at high-resolution. Nucleic Acids Research. 2017; 45:D777–D783. https://doi.org/10.1093/nar/ gkw1121 PMID: 27899578 17. Simmonds P. Rampant C!U Hypermutation in the Genomes of SARS-CoV-2 and Other Coronavi- ruses: Causes and Consequences for Their Short- and Long-Term Evolutionary Trajectories. mSphere. 2020; 5. https://doi.org/10.1128/mSphere.00408-20 PMID: 32581081 18. Ratcliff J, Simmonds P. Potential APOBEC-mediated RNA editing of the genomes of SARS-CoV-2 and other coronaviruses and its impact on their longer term evolution. Virology. 2021; 556:62. https://doi. org/10.1016/j.virol.2020.12.018 PMID: 33545556 19. Sanjua´ n R, Domingo-Calap P. Mechanisms of viral mutation. Cellular and molecular life sciences. 2016; 73:4433–4448. https://doi.org/10.1007/s00018-016-2299-6 PMID: 27392606 20. Shu Y, McCauley J. GISAID: Global initiative on sharing all influenza data—from vision to reality. Eurosurveillance. 2017; 22(13). https://doi.org/10.2807/1560-7917.ES.2017.22.13.30494 PMID: 28382917 21. Picardi E, Mansi L, Pesole G. Detection of A-to-I RNA Editing in SARS-COV-2. Genes 2021; 13:41. https://doi.org/10.3390/genes13010041 PMID: 35052382 22. Ringlander J, Fingal J, Kann H, Prakash K, Rydell G, Andersson M, et al. Impact of ADAR-induced edit- ing of minor viral RNA populations on replication and transmission of SARS-CoV-2. Proceedings of the National Academy of Sciences of the United States of America. 2022; 119:e2112663119. https://doi. org/10.1073/pnas.2112663119 PMID: 35064076 23. Bloom JD, Beichman AC, Neher RA, Harris K. Evolution of the SARS-CoV-2 Mutational Spectrum. Molecular Biology and Evolution. 2023; 40(4). https://doi.org/10.1093/molbev/msad085 PMID: 37039557 24. Ruis C, Peacock TP, Polo LM, Masone D, Alvarez MS, Hinrichs AS, et al. Mutational spectra distinguish SARS-COV-2 replication niches. 2022. https://doi.org/10.1101/2022.09.27.509649 25. Wang P, Nair MS, Liu L, Iketani S, Luo Y, Guo Y, et al. Antibody resistance of SARS-CoV-2 variants B.1.351 and B.1.1.7. Nature 2021; 593:130–135. https://doi.org/10.1038/s41586-021-03398-2 PMID: 33684923 26. Wang Z, Schmidt F, Weisblum Y, Muecksch F, Barnes CO, Finkin S, et al. mRNA vaccine-elicited anti- bodies to SARS-CoV-2 and circulating variants. Nature 2021; 592:616–622. https://doi.org/10.1038/ s41586-021-03324-6 PMID: 33567448 27. Ou J, Lan W, Wu X, Zhao T, Duan B, Yang P, et al. Tracking SARS-CoV-2 Omicron diverse spike gene mutations identifies multiple inter-variant recombination events. Signal Transduction and Targeted Therapy 2022; 7:1–9. https://doi.org/10.1038/s41392-022-00992-2 PMID: 35474215 28. Shafer MM, Gregory D, Bobholz MJ, Roguet A, Haddock Soto LA, Rushford C, et al. Tracing the origin of SARS-CoV-2 Omicron-like Spike sequences detected in wastewater. medRxiv. 2022; https://doi.org/ 10.1101/2022.10.28.22281553 29. Chaguza C, Hahn AM, Petrone ME, Zhou S, Ferguson D, Breban MI, et al. Accelerated SARS-CoV-2 intrahost evolution leading to distinct genotypes during chronic infection. Cell Reports Medicine. 2023; p. 100943. https://doi.org/10.1016/j.xcrm.2023.100943 PMID: 36791724 30. Harari S, Tahor M, Rutsinsky N, Meijer S, Miller D, Henig O, et al. Drivers of adaptive evolution during chronic SARS-CoV-2 infections. Nature Medicine. 2022; 28(7):1501–1508. https://doi.org/10.1038/ s41591-022-01882-4 PMID: 35725921 31. Bamgboye EL, Omiye JA, Afolaranmi OJ, Davids MR, Tannor EK, Wadee S, et al. COVID-19 Pan- demic: Is Africa Different? Journal of the National Medical Association. 2021; 113:324. https://doi.org/ 10.1016/j.jnma.2020.10.001 PMID: 33153755 32. Herng LC, Singh S, Sundram BM, Zamri ASSM, Vei TC, Aris T, et al. The effects of super spreading events and movement control measures on the COVID-19 pandemic in Malaysia. Scientific Reports. 2022; 12. https://doi.org/10.1038/s41598-022-06341-1 PMID: 35140319 33. Dong E, Ratcliff J, Goyea TD, Katz A, Lau R, Ng TK, et al. The Johns Hopkins University Center for Sys- tems Science and Engineering COVID-19 Dashboard: data collection process, challenges faced, and lessons learned; The Lancet Infectious Diseases, 22(12), e370–e376. https://doi.org/10.1016/S1473- 3099(22)00434-0 PMID: 36057267 PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 24 / 26 PLOS COMPUTATIONAL BIOLOGY Mutational processes and SARS-CoV-2’s evolutionary capacity 34. Graudenzi A, Maspero D, Angaroni F, Piazza R, Ramazzotti D. Mutational signatures and heteroge- neous host response revealed via large-scale characterization of SARS-CoV-2 genomic diversity. ISCIENCE. 2021; 24:102116. https://doi.org/10.1016/j.isci.2021.102116 PMID: 33532709 35. Azgari C, Kilinc Z, Turhan B, Circi D, Adebali O, Castilletti C, et al. The Mutation Profile of SARS-CoV-2 Is Primarily Shaped by the Host Antiviral Defense. Signal Transduction and Targeted Therapy 2022 7:1. 2021. https://doi.org/10.3390/v13030394 PMID: 33801257 36. Giorgio SD, Martignano F, Torcia MG, Mattiuz G, Conticello SG. Evidence for host-dependent RNA editing in the transcriptome of SARS-CoV-2. Science Advances. 2020; 6:5813–5830. https://doi.org/10. 1126/sciadv.abb5813 PMID: 32596474 37. Yi K, Kim SY, Bleazard T, Kim T, Youk J, Ju YS. Mutational spectrum of SARS-COV-2 during the global pandemic. Experimental & Molecular Medicine. 2021; 53(8):1229–1237. https://doi.org/10.1038/ s12276-021-00658-z PMID: 34453107 38. Langenbucher A, Bowen D, Sakhtemani R, Bournique E, Wise JF, Zou L, et al. An extended APO- BEC3A mutation signature in cancer. Nature Communications. 2021; 12. https://doi.org/10.1038/ s41467-021-21891-0 PMID: 33707442 39. Kim K, Calabrese P, Wang S, Qin C, Rao Y, Feng P, et al. The Roles of APOBEC-mediated RNA Edit- ing in SARS-CoV-2 Mutations, Replication and Fitness. bioRxiv. 2021.12.18.473309. https://doi.org/10. 1101/2021.12.18.473309 40. 41. Trypsteen W, Cleemput JV, van Snippenberg W, Gerlo S, Vandekerckhove L. On the whereabouts of SARS-CoV-2 in the human body: A systematic review. PLOS Pathogens. 2020; 16:e1009037. https:// doi.org/10.1371/journal.ppat.1009037 PMID: 33125439 Takata MA, Gonc¸alves-Carneiro D, Zang TM, Soll SJ, York A, Blanco-Melo D, et al. CG dinucleotide suppression enables antiviral defence targeting non-self RNA. Nature. 2017; 550(7674):124–127. https://doi.org/10.1038/nature24039 PMID: 28953888 42. Nchioua R, Kmiec D, Mu¨ller JA, Conzelmann C, Groß R, Swanson CM, et al. Sars-cov-2 is restricted by zinc finger antiviral protein despite preadaptation to the low-cpg environment in humans. mBio. 2020; 11(5):1–19. https://doi.org/10.1128/mBio.01930-20 43. Zimmer MM, Kibe A, Rand U, Pekarek L, Ye L, Buck S, et al. The short isoform of the host antiviral pro- tein ZAP acts as an inhibitor of SARS-CoV-2 programmed ribosomal frameshifting. Nature Communica- tions. 2021; 12(1):1–15. https://doi.org/10.1038/s41467-021-27431-0 PMID: 34893599 44. Mourier T, Sadykov M, Carr MJ, Gonzalez G, Hall WW, Pain A. Host-directed editing of the SARS-CoV- 2 genome. Biochemical and Biophysical Research Communications. 2021; 538:35–39. https://doi.org/ 10.1016/j.bbrc.2020.10.092 PMID: 33234239 45. Li Z, Wu J, DeLeo CJ. RNA damage and surveillance under oxidative stress. IUBMB Life. 2006; 58:581–588. https://doi.org/10.1080/15216540600946456 PMID: 17050375 46. V’kovski P, Kratzel A, Steiner S, Stalder H, Thiel V. Coronavirus biology and replication: implications for SARS-CoV-2. Nature Reviews Microbiology 2020; 19:155–170. https://doi.org/10.1038/s41579-020- 00468-6 PMID: 33116300 47. Simmonds P, Ansari MA. Extensive C->U transition biases in the genomes of a wide range of mamma- lian RNA viruses; potential associations with transcriptional mutations, damage- or host-mediated editing of viral RNA. PLoS Pathogens. 2021; 17. https://doi.org/10.1371/journal.ppat.1009596 PMID: 34061905 48. Kucab JE, Zou X, Morganella S, Joel M, Nanda AS, Nagy E, et al. A Compendium of Mutational Signa- tures of Environmental Agents. Cell. 2019; 177(4):821–836.e16. https://doi.org/10.1016/j.cell.2019.03. 001 PMID: 30982602 49. Thorne LG, Bouhaddou M, Reuschl AK, Zuliani-Alvarez L, Polacco B, Pelin A, et al. Evolution of enhanced innate immune evasion by SARS-CoV-2. Nature. 2022; 602:487–495. https://doi.org/10. 1038/s41586-021-04352-y PMID: 34942634 50. Carabelli AM, Peacock TP, Thorne LG, Harvey WT, Hughes J, de Silva TI, et al. SARS-CoV-2 variant biology: immune escape, transmission and fitness; Nat Rev Microbiol 21, 162–177 (2023). https://doi. org/10.1038/s41579-022-00841-7 PMID: 36653446 51. Markov PV, Ghafari M, Beer M, Lythgoe K, Simmonds P, Stilianakis NI, et al. The evolution of SARS-CoV- 2; Nat Rev Microbiol 21, 361–379 (2023). https://doi.org/10.1038/s41579-023-00878-2 PMID: 37020110 52. Liu G, Gack MU. SARS-CoV-2 learned the ‘Alpha’bet of immune evasion; Nature Immunology. 2022; 23:351–353. https://doi.org/10.1038/s41590-022-01148-8 PMID: 35194206 53. Chen K, Xiao F, Hu D, Ge W, Tian M, Wang W, et al. Sars-cov-2 nucleocapsid protein interacts with rig- i and represses rig-mediated ifn-β production. Viruses. 2021; 13. 54. Catanzaro M, Fagiani F, Racchi M, Corsini E, Govoni S, Lanni C. Immune response in COVID-19: addressing a pharmacological challenge by targeting pathways triggered by SARS-CoV-2; Sig Trans- duct Target Ther 5, 84 (2020). https://doi.org/10.1038/s41392-020-0191-1 PMID: 32467561 PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 25 / 26 PLOS COMPUTATIONAL BIOLOGY Mutational processes and SARS-CoV-2’s evolutionary capacity 55. Miorin L, Kehrer T, Sanchez-Aparicio MT, Zhang K, Cohen P, Patel RS, et al. SARS-CoV-2 Orf6 hijacks Nup98 to block STAT nuclear import and antagonize interferon signaling; Proceedings of the National Academy of Sciences, 117(45), 28344–28354. https://doi.org/10.1073/pnas.2016650117 PMID: 33097660 56. Willett BJ, Grove J, MacLean OA, Wilkie C, Lorenzo GD, Furnon W, et al. SARS-CoV-2 Omicron is an immune escape variant with an altered cell entry pathway. Nature Microbiology 2022; 7:1161–1179. https://doi.org/10.1038/s41564-022-01143-7 PMID: 35798890 57. Jackson CB, Farzan M, Chen B, Choe H. Mechanisms of SARS-CoV-2 entry into cells. Nature Reviews Molecular Cell Biology 2021; 23:3–20. https://doi.org/10.1038/s41580-021-00418-x PMID: 34611326 58. Ritchie H, Mathieu E, Rode´ s-Guirao L, Appel C, Giattino C, Ortiz-Ospina E, et al. Coronavirus Pan- demic (COVID-19). Our World in Data. 2020;. 59. Dong E, Du H, Gardner L. An interactive web-based dashboard to track COVID-19 in real time; The Lancet Infectious Diseases Volume 20, Issue 5, May 2020, Pages 533–534 https://doi.org/10.1016/ S1473-3099(20)30120-1 PMID: 32087114 60. Hale T, Angrist N, Goldszmidt R, Kira B, Petherick A, Phillips T, et al. A global panel database of pan- demic policies (Oxford COVID-19 Government Response Tracker). Nature Human Behaviour 2021; 5:529–538. https://doi.org/10.1038/s41562-021-01079-8 PMID: 33686204 61. Nordstro¨m P, Ballin M, Nordstro¨m A. Risk of SARS-CoV-2 reinfection and COVID-19 hospitalisation in individuals with natural and hybrid immunity: a retrospective, total population cohort study in Sweden. The Lancet Infectious Diseases. 2022; 22:781–790. https://doi.org/10.1016/S1473-3099(22)00143-8 PMID: 35366962 62. Obermeyer F, Jankowiak M, Barkas N, Schaffner SF, Pyle JD, Yurkovetskiy L, et al. Analysis of 6.4 mil- lion SARS-CoV-2 genomes identifies mutations associated with fitness. Science. 2022; 376:1327– 1332. https://doi.org/10.1126/science.abm1208 PMID: 35608456 63. Heo MH, Kwon YD, Cheon J, Kim KB, Noh JW. Association between the Human Development Index and Confirmed COVID-19 Cases by Country. Healthcare (Switzerland). 2022; 10. https://doi.org/10. 3390/healthcare10081417 PMID: 36011075 64. Wu F, Zhao S, Yu B, Chen YM, Wang W, Song ZG, et al. A new coronavirus associated with human respiratory disease in China. Nature 2020; 579:265–269. https://doi.org/10.1038/s41586-020-2008-3 PMID: 32015508 65. 66. Lee DD, Seung HS. Learning the parts of objects by non-negative matrix factorization. Nature 1999; 401:788–791. https://doi.org/10.1038/44565 PMID: 10548103 Islam SMA, Dı´az-Gay M, Wu Y, Barnes M, Vangara R, Bergstrom EN, et al. Uncovering novel muta- tional signatures by de novo extraction with SigProfilerExtractor. Cell Genomics. 2022; 2(11):100179. https://doi.org/10.1016/j.xgen.2022.100179 PMID: 36388765 PLOS Computational Biology | https://doi.org/10.1371/journal.pcbi.1011795 January 25, 2024 26 / 26 PLOS COMPUTATIONAL BIOLOGY
Could not heal snippet
10.1038_s41598-020-77562-5.pdf
Data availability All data generated and analyzed during the present study are available upon request from the corresponding author upon reasonable request.
Data availability All data generated and analyzed during the present study are available upon request from the corresponding author upon reasonable request. Received: 22 April 2020; Accepted: 9 November 2020
OPEN Vascular endothelial growth factor promotes atrial arrhythmias by inducing acute intercalated disk remodeling Louisa Mezache1, Heather L. Struckman1, Amara Greer‑Short2, Stephen Baine2,3, Sándor Györke2,3, Przemysław B. Radwański2,3,4, Thomas J. Hund1,2 & Rengasayee Veeraraghavan1,2,3* Atrial fibrillation (AF) is the most common arrhythmia and is associated with inflammation. AF patients have elevated levels of inflammatory cytokines known to promote vascular leak, such as vascular endothelial growth factor A (VEGF). However, the contribution of vascular leak and consequent cardiac edema to the genesis of atrial arrhythmias remains unknown. Previous work suggests that interstitial edema in the heart can acutely promote ventricular arrhythmias by disrupting ventricular myocyte intercalated disk (ID) nanodomains rich in cardiac sodium channels (NaV1.5) and slowing cardiac conduction. Interestingly, similar disruption of ID nanodomains has been identified in atrial samples from AF patients. Therefore, we tested the hypothesis that VEGF‑induced vascular leak can acutely increase atrial arrhythmia susceptibility by disrupting ID nanodomains and slowing atrial conduction. Treatment of murine hearts with VEGF (30–60 min, at clinically relevant levels) prolonged the electrocardiographic P wave and increased susceptibility to burst pacing‑ induced atrial arrhythmias. Optical voltage mapping revealed slower atrial conduction following VEGF treatment (10 ± 0.4 cm/s vs. 21 ± 1 cm/s at baseline, p < 0.05). Transmission electron microscopy revealed increased intermembrane spacing at ID sites adjacent to gap junctions (GJs; 64 ± 9 nm versus 17 ± 1 nm in controls, p < 0.05), as well as sites next to mechanical junctions (MJs; 63 ± 4 nm versus 27 ± 2 nm in controls, p < 0.05) in VEGF–treated hearts relative to controls. Importantly, super‑ resolution microscopy and quantitative image analysis revealed reorganization of NaV1.5 away from dense clusters localized near GJs and MJs to a more diffuse distribution throughout the ID. Taken together, these data suggest that VEGF can acutely predispose otherwise normal hearts to atrial arrhythmias by dynamically disrupting NaV1.5‑rich ID nanodomains and slowing atrial conduction. These data highlight inflammation‑induced vascular leak as a potential factor in the development and progression of AF. Atrial fibrillation (AF) is the most common cardiac arrhythmia, affecting 2–3% of the US population1. Inflam- mation, vascular leak, and associated tissue edema are common sequelae of pathologies associated with AF2–8, and are emerging as proarrhythmic factors. Inflammatory signaling involving cytokines compromises the vas- cular barrier function, and increase vascular leak9. Specifically, multiple studies in early stage AF patients (lone/ paroxysmal AF) report elevated levels of vascular endothelial growth factor A (VEGF; 89–560 pg/ml)3–6,8 and VEGF receptor 2, its primary receptor in the vascular endothelium7. Likewise, elevated levels of vascular leak- inducing cytokines predict AF recurrence following ablation10. Although vascular leak is known to promote adverse remodeling and cardiovascular disease in the chronic condition (days-weeks)11–13, its acute (< 4 h) con- tribution to arrhythmogenesis has yet to be explored. Myocardial edema, a direct consequence of vascular leak, is linked to arrhythmias in multiple pathologies, including AF14–18. Likewise, cardiac edema has been linked 1Department of Biomedical Engineering, College of Engineering, The Ohio State University, 460 Medical Center Dr., Rm 415A, IBMR, Columbus, OH 43210, USA. 2The Frick Center for Heart Failure and Arrhythmia, Dorothy M. Davis Heart and Lung Research Institute, College of Medicine, The Ohio State University Wexner Medical Center, Columbus, OH, USA. 3Department of Physiology and Cell Biology, College of Medicine, The Ohio State University, Columbus, OH, USA. 4Division of Pharmacy Practice and Sciences, College of Pharmacy, The Ohio State University, Columbus, OH, USA. *email: veeraraghavan.12@osu.edu Scientific Reports | (2020) 10:20463 | https://doi.org/10.1038/s41598-020-77562-5 1 Vol.:(0123456789)www.nature.com/scientificreports to AF recurrence following ablation19,20. Previous work by us and others, suggests that interstitial edema can acutely (within minutes) elevate arrhythmia susceptibility21–24. In these studies, the proarrhythmic impact of edema resulted from disruption of cardiac sodium channel (NaV1.5)–rich intercalated disk (ID) nanodomains and consequent slowing of action potential propagation22–25. Interestingly, similar disruption of ID nanodomains has been identified in AF patients26. Therefore, we hypothesized that VEGF (at clinically-relevant levels) may acutely promote atrial arrhythmias by disrupting ID nanodomains and slowing atrial conduction. We provide structural and functional evidence, from the nanoscale to the in vivo level, demonstrating that this mechanism can promote atrial arrhythmias. We also identify a novel form of tissue remodeling involving the dynamic reorganization of NaV1.5 within the ID occurring in the aftermath of acute exposure to VEGF, resulting in the dispersal of channels from dense clusters located within nanodomains. Methods All animal procedures were approved by Institutional Animal Care and Use Committee at The Ohio State Uni- versity and performed in accordance with the Guide for the Care and Use of Laboratory Animals published by the U.S. National Institutes of Health (NIH Publication No. 85-23, revised 2011). Langendorff preparation, tissue collection. Male C57/BL6 mice (30 g, 6–18 weeks) were anesthetized with 5% isoflurane mixed with 100% oxygen (1 l/min). After loss of consciousness, anesthesia was maintained with 3–5% isoflurane mixed with 100% oxygen (1 l/min). Once the animal was stably in a surgical plane of anes- thesia, the heart was excised, leading to euthanasia by exsanguination. The isolated hearts were prepared in one of the following three ways: i) Langendorff preparations: For optical mapping and ex vivo electrocardiography (ECG) studies, hearts were perfused (at 60–80 mm Hg) in a Langendorff configuration with oxygenated, modified Tyrode’s solution (containing, in mM: NaCl 140, KCl 5.4, MgCl2 0.5, CaCl2 1.2, dextrose 5.6, HEPES 10; pH adjusted to 7.4) at 37 °C as previously described22,25,27–29. ii) Cryopreservation: Hearts were embedded in optimal cutting temperature compound and frozen using liquid nitrogen for cryosectioning and fluorescent immunolabeling as in previous studies22,23,25,30. These samples were used for light microscopy experiments as described below. iii) Fixation for Transmission Electron Microscopy (TEM): Atria were dissected and fixed overnight in 2% glutaraldehyde at 4 °C for resin embedding and ultramicrotomy as previously described22,25. For both structural and functional studies, the left atrium was prioritized in order to avoid any influence from pacemaker tissue. FITC‑dextran extravasation. Langendorff-perfused mouse hearts were perfused for 60 min with Tyrode’s solution with or without VEGF (500 pg/ml) and FITC-dextran (10 mg/ml) was added to the final 10 ml of perfu- sate. Perfused hearts were then cryopreserved as described above and extravasated FITC-dextran levels assessed by confocal microscopy of cryosections. Optical mapping and volume‑conducted electrocardiography (ECG). Optical voltage mapping was performed using the voltage sensitive dye, di-4-ANEPPS (15 µM; ThermoFisher Scientific, Grand Island, NY), as previously described22,23,29, in order to quantify conduction velocity. Motion was suppressed by add- ing blebbistatin (10  µM) to the perfusate. Preparations were excited by 510  nm light and fluorescent signals passed through a 610 nm longpass filter (Newport, Irvine, CA) and recorded at 1000 frames/sec using a MiCAM Ultima-L CMOS camera (SciMedia, Costa Mesa, CA). Activation time was defined as the time of the maximum first derivative of the AP31, and activation times were fitted to a parabolic surface32. Gradient vectors evaluated along this surface were averaged along the fast axis of propagation (± 15°) to quantify CV. Hearts were paced epicardially from the left atrium at a cycle length of 100 ms with 1 ms current pulses at 1.5 times the pacing threshold for all CV measurements. A volume-conducted ECG was collected concurrently using silver chloride electrodes placed in the bath and digitized at 1 kHz. Atrial arrhythmia inducibility was assessed by 10 s of burst pacing at cycle lengths of 50, 40, and 30 ms as previously described33,34. In subsets of experiments, vascular endothelial growth factor A (VEGF; Sigma SRP4364) was added to the perfusate at 100 (low) and 500 pg/ml (high). These concentrations were selected based on VEGF levels observed in human AF patients (89–560 pg/ml)3–6,8. Measurements were made following 30 min of treatment. In vivo ECG. Continuous ECG recordings (PL3504 PowerLab 4/35, ADInstruments) were obtained from mice anesthetized with isoflurane (1–1.5%) as previously described35. Briefly, after baseline recording (5 min.), animals received either intraperitoneal VEGF (10 or 50  ng/kg; Sigma) or vehicle (PBS). After an additional 20 min, animals were injected intraperitoneally with epinephrine (1.5 mg/kg; Sigma) and caffeine (120 mg/kg; Sigma) challenge and ECG recording continued for 40 min. ECG recordings were analyzed using the LabChart 8 software (ADInstruments). Primary antibodies. The following primary antibodies were used for Western immunoblotting and fluo- rescence microscopy studies: • Connexin43 (Cx43; rabbit polyclonal; Sigma C6219) Scientific Reports | (2020) 10:20463 | https://doi.org/10.1038/s41598-020-77562-5 2 Vol:.(1234567890)www.nature.com/scientificreports/ • Connexin40 (Cx40; rabbit polyclonal; ThermoFisher Scientific 36–4900) • N-cadherin (N-cad; mouse monoclonal; BD Biosciences 610,920) • Cardiac isoform of the voltage-gated sodium channel (NaV1.5; rabbit polyclonal; custom antibody25) • The sodium channel β subunit (β1; rabbit polyclonal; custom antibody25) Western immunoblotting. Whole cell lysates of mouse hearts frozen using liquid nitrogen were prepared as previously described25,35,36. These were electrophoresed on 4–15% TGX Stain-free gels (BioRad, Hercules, CA) before being transferred onto a nitrocellulose membrane. The membranes were probed with primary antibodies against Cx43, Cx40, NaV1.5 and β1 as well as mouse monoclonal antibody against GAPDH (loading control; Fitzgerald Industries, Acton, MA), followed by goat anti-rabbit and goat anti-mouse HRP-conjugated second- ary antibodies (Promega, Madison, WI). Signals were detected by chemiluminescence using SuperSignal West Femto Extended Duration Substrate (ThermoFisher Scientific, Grand Island, NY), imaged using a Chemidoc MP imager (BioRad, Hercules, CA), and analyzed using Image Lab software (BioRad, Hercules, CA). Fluorescent immunolabeling. Immuno-fluorescent labeling of cryosections (5 µm thickness) of fresh- frozen myocardium was performed, as previously described 22,25,35,37. Briefly, cryosections were fixed with para- formaldehyde (2%, 5 min at room temperature), permeabilized with Triton X-100 (0.2% in PBS for 15 min at room temperature) and treated with blocking agent (1% BSA, 0.1% triton in PBS for 2 h at room temperature) prior to labeling with primary antibodies (overnight at 4 °C). Samples were then washed in PBS (3 × 5 min in PBS at room temperature) prior to labeling with secondary antibodies. For confocal microscopy, samples were then labeled with goat anti-mouse and goat anti-rabbit secondary antibodies conjugated to Alexa 405, Alexa 488, Alexa 568 and Alexa 647 were used (1:8000; ThermoFisher Scien- tific, Grand Island, NY). Simultaneous labeling with two rabbit or mouse primary antibodies was accomplished by direct fluorophore conjugation of primary antibodies (Zenon labeling kits, ThermoFisher Scientific, Grand Island, NY). Samples were then washed in PBS (3 × 5 min in PBS at room temperature) and mounted in ProLong Gold (Invitrogen, Rockford, IL). For STimulated Emission Depletion (STED) microscopy, samples were prepared similar to confocal microscopy but labeled with Alexa 594 and Atto 647 N fluorophores. For STochastic Optical Reconstruction Microscopy (STORM), samples were labeled with Alexa 647 and Biotium CF 568 fluorophores. STORM samples were then washed in PBS (3 × 5 min in PBS at room temperature) and optically cleared using Scale U2 buffer (48 h at 4 °C) prior to imaging23,25,30. Transmission electron microscopy (TEM). TEM images of the ID, particularly gap junctions (GJs) and mechanical junctions (MJs), were obtained at 60,000 × magnification on a FEI Tecnai G2 Spirit electron micro- scope. Intermembrane distance at various ID sites was quantified using ImageJ (NIH, http://rsbwe b.nih.gov/ij/), as previously described22,25. Sub‑diffraction confocal imaging (sDCI). Confocal imaging was performed using an A1R-HD laser scanning confocal microscope equipped with four solid-state lasers (405 nm, 488 nm, 560 nm, 640 nm, 30 mW each), a 63×/1.4 numerical aperture oil immersion objective, two GaAsP detectors, and two high sensitivity photomultiplier tube detectors (Nikon, Melville, NY). Individual fluorophores were imaged sequentially with the excitation wavelength switching at the end of each frame. Images were collected as z-stacks with fluorophores images sequentially (line-wise) to achieve optimal spectral separation. Sub-diffraction structural information (130 nm resolution) was recovered by imaging with a 12.8 µm pinhole (0.3 Airy units) with spatial oversampling (4 × Nyquist sampling) and applying 3D deconvolution, as previously described38. STimulated emission depletion (STED) microscopy. Samples were imaged using a time-gated STED 3X system (Leica, Buffalo Grove, IL) based on a TCS SP8 laser scanning confocal microscope and equipped with STED modules, a pulsed white-light laser (470–670 nm; 80 MHz pulse rate), a Plan Apochromat STED WHITE 100×/1.4 numerical aperture oil immersion objective, HyD hybrid detectors, and three STED depletion lasers (775 nm, 660 nm, 592 nm). Depletion beam was applied in the classical vortex donut configuration to achieve the best lateral resolution (25  nm) as well as in a z-donut configuration to achieve the best axial resolution (50 nm). Time gating of light collection (1.5–3.5 ns following each laser pulse) was also applied to aid in achiev- ing optimal resolution. Images were collected as z-stacks with fluorophores images sequentially (line-wise) and subjected to 3D deconvolution. These images were analyzed using object-based segmentation in 3D (OBS3D), as previously described22,23. Single molecule localization. STORM imaging was performed using a Vutara 352 microscope (Bruker Nano Surfaces, Middleton, WI) equipped with biplane 3D detection, and fast sCMOS imaging achieving 20 nm lateral and 50 nm axial resolution, as previously described 25,30,36,39. Individual fluorophore molecules were local- ized with a precision of 10 nm. The two color channels were precisely registered using localized positions of several TetraSpeck Fluorescent Microspheres (ThermoFisher Scientific, Carlsbad, CA) scattered throughout the field of view, with the procedure being repeated at the start of each imaging session. Protein clustering and spatial organization were quantitatively assessed from single molecule localization data using STORM-RLA, a machine learning-based cluster analysis approach, as previously described30. Statistical analysis. Treatments were applied in unblinded fashion for all studies. All data which passed the Shaprio-Wilk test for normality were treated as follows. The Wilcoxon signed rank test or a single factor ANOVA Scientific Reports | (2020) 10:20463 | https://doi.org/10.1038/s41598-020-77562-5 3 Vol.:(0123456789)www.nature.com/scientificreports/ Figure 1. Acute effects of VEGF on atrial conduction. (A) Representative volume-conducted ECGs. (B) Summary plots of P wave duration (n = 5/group; *p < 0.05 vs. control). (C) Representative isochrone maps of left atrial activation. (D) Summary plots of CV (n = 5/group; *p < 0.05 vs. control). was used for single comparisons. For multiple comparisons, the Šidák correction was applied. Fisher’s exact test was used to test differences in nominal data. For non-normal data, a Friedman rank sum test or Kruskal–Wal- lis 1-way analysis of variance for paired and unpaired data was applied. A p < 0.05 was considered statistically significant. All values are reported as mean ± standard error unless otherwise noted. To ensure unbiased results, all image analyses were conducted using automated batch processing algorithms. Results Multiple studies in early stage AF patients (lone/paroxysmal AF) report elevated levels of VEGF (89–560 pg/ ml)3–6,8 and VEGF receptor 27. In order to assess the acute impact of VEGF on AF susceptibility, we assessed the structural and electrophysiological impacts of treating Langendorff-perfused WT mouse hearts with clinically relevant levels of VEGF (low: 100 pg/ml and high: 500 pg/ml) for 30 min. VEGF-induced vascular leak was first confirmed by extravasation of FITC-dextran from cryosections of VEGF-treated (500 pg/ml) and vehicle control hearts. Levels of FITC-dextran extravasated into VEGF-treated (500 pg/ml) hearts was doubled relative to vehicle controls (201 ± 7% vs. 100 ± 9%, p < 0.05, n = 3 hearts/group; Supplementary Fig. 1). These data are consistent with acute enhancement of vascular leak by VEGF. Atrial conduction is slowed following acute VEGF treatment. To examine the functional impacts of VEGF-induced ID remodeling, volume-conducted electrocardiograms (ECG) were recorded from Langendorff- perfused mouse hearts (Fig. 1). Significant P-wave prolongation was observed following 30 min of VEGF perfu- sion compared to control (Fig. 1A,B). VEGF exerted similar effects on P-wave duration in vivo (Supplementary Fig. 2). These data point to possible slowing of atrial conduction following VEGF treatment. Next, we directly assessed atrial conduction velocity using optical voltage mapping. Representative optical isochrone maps of activation in Fig. 1C demonstrate increased conduction delay in VEGF treated hearts compared to untreated controls. Overall, VEGF significantly and dose-dependently decreased atrial conduction velocity (Fig. 1D). VEGF‑treated hearts are susceptible to atrial arrhythmias. Conduction slowing is a well-estab- lished substrate for cardiac arrhythmias in general40–42, and AF in particular43,44. Therefore, we assessed the acute effects of VEGF-induced conduction slowing on AF risk. A representative volume-conducted ECG trace in Fig. 2A (top) illustrates resumption of sinus rhythm following atrial burst pacing. In contrast, an atrial arrhyth- Scientific Reports | (2020) 10:20463 | https://doi.org/10.1038/s41598-020-77562-5 4 Vol:.(1234567890)www.nature.com/scientificreports/ Figure 2. Acute impact of VEGF on atrial arrhythmia susceptibility. (A) Representative volume-conducted ECGs show response to burst pacing. (B) Incidence of atrial arrhythmias following burst pacing (n = 5/ group, * p < 0.05 vs. control). (C) Representative in vivo surface ECG illustrates atrial arrhythmia observed in a VEGF-treated mouse following caffeine + epinephrine challenge. (D) Total atrial arrhythmia burden under caffeine + epinephrine challenge quantified as seconds of arrhythmia per hour of observation (n = 10/group, *p < 0.05 vs. control). mia is apparent on the trace from a VEGF-treated heart (Fig. 2A, bottom). Overall, VEGF increased the inci- dence of burst pacing-induced atrial arrhythmias in dose-dependent fashion (Fig. 2A,B; Supplementary Fig. 3). Next, we assessed the acute impact of VEGF on atrial arrhythmia risk in vivo. Promotion of arrhythmic trig- gers via caffeine and epinephrine challenge elicited atrial arrhythmias in VEGF-treated mice but not in untreated controls (Fig. 2C,D; Supplementary Fig. 4). Taken together, these data suggest that conduction slowing increases the risk of atrial arrhythmias. VEGF does not acutely alter expression of key ID proteins. In order to determine the structural basis of VEGF-induced atrial arrhythmias, we assessed the expression of key ID proteins. Western immunoblot- ting revealed no significant difference in the levels of Na+ channel subunits (NaV1.5, β1), the gap junction protein Cx43, or the mechanical junction protein, N-cad between VEGF-treated (high dose) hearts and untreated con- trols (Supplementary Fig. 5). Expression of the gap junction protein Cx40 was slightly elevated in VEGF-treated hearts. Increased Cx40 expression could enhance GJ coupling, although the small change observed is unlikely to have appreciable functional impact. In any case, changes in ID protein expression cannot explain VEGF-induced conduction slowing and proarrhythmia. ID structural remodeling following acute VEGF insult. Previous studies link cardiac interstitial edema to ultrastructural remodeling within the ID, specifically, increased intermembrane distance near GJ. Sim- ilar changes have also been reported in AF patients26. Therefore, we performed transmission electron micros- copy (TEM) to assess the acute effects of VEGF on ID structure. Representative TEM images show narrow intermembrane spacing at GJ- and MJ-adjacent sites in untreated control hearts, and marked widening at these sites following VEGF treatment (Fig. 3A). Overall, both low and high doses of VEGF significantly increased intermembrane distances at GJ- and MJ-adjacent sites compared to untreated controls (Fig. 3B). The swelling occurred in dose-dependent fashion at GJ-adjacent perinexi but not near MJ. ID proteins undergo reorganization following acute VEGF treatment. Next, we performed super- resolution microscopy studies to assess the effects VEGF on ID molecular organization. As a first step, we used sDC imaging (130 nm resolution) to examine the overall layout of key proteins within the murine atrial ID. Although lacking the resolution of other super-resolution imaging methods such as STED and STORM, sDCI Scientific Reports | (2020) 10:20463 | https://doi.org/10.1038/s41598-020-77562-5 5 Vol.:(0123456789)www.nature.com/scientificreports/ Figure 3. VEGF effects on ID ultrastructure. (A) Representative TEM images of IDs. (B) Summary plots of intermembrane distance at GJ-adjacent perinexal sites (solid bars) and MJ-adjacent (striped bars) ID sites (> 100 measurements/group/location from n = 3 hearts/group, *p < 0.05 vs. control). offers greater capability for multicolor imaging. Therefore, we used sDCI to examine the organization of sodium channel α (NaV1.5) and β (β1) subunits relative to GJ (Cx40, Cx43) and MJ (N-cad) proteins (Fig. 4). Both connexin isoforms predominantly expressed in the atria, Cx40 and Cx43, displayed similar patterns of localization (Fig. 4A,B), suggesting that either isoform could be used as a marker for atrial GJs. N-cad immu- nosignal was localized to distinct ID regions compared to Cx40, Cx43, with very little co-localization. These results are consistent with the enrichment of GJ and MJ within interplicate and plicate ID regions respectively. Representative sDCI images (Fig. 4C,D) illustrate an ID in en face orientation from a murine atrial section labeled for NaV1.5, β1, Cx43 and N-cad. NaV1.5 and β1 were distributed extensively throughout the ID. Having established the overall layout of Na+ channel components within the atrial ID, we switched to higher resolution techniques to assess the effects of VEGF-induced vascular leak on their localization. Three dimen- sional en face views of IDs from control hearts obtained by STED microscopy (25 nm resolution) reveal extensive clustering of NaV1.5 throughout the ID, particularly in close proximity to Cx43 clusters and at N-cad-rich sites (Fig. 5A, top). In VEGF-treated hearts, NaV1.5 clusters appeared fragmented, were located further from Cx43 clusters, and co-distributed less with N-cad (Fig. 5A, bottom). Similar to NaV1.5, β1 was also organized into clus- ters in control hearts, and was found in close proximity to Cx43 clusters (Fig. 5B, top). However, unlike NaV1.5, β1 displayed very little co-distribution with N-cad. In VEGF-treated hearts, β1 clusters appeared more diffuse and were distributed farther away from Cx43 clusters (Fig. 5B, bottom). Quantitative analysis by object-based segmentation was used to calculate NaV1.5 and β1 signal enrichment ratio, defined as the ratio of NaV1.5 / β1 immunosignal cluster mass (volume x normalized intensity) at sites near (< 100 nm away) Cx43 and N-cad vs. the signal cluster mass at other ID sites. Overall, we observed significant enrichment of NaV1.5 immunosignal near (< 100 nm) Cx43 and N-cad, and β1 near Cx43 in control hearts (Fig. 6). VEGF-treatment significantly decreased NaV1.5 and β1 enrichment ratio near Cx43, while NaV1.5 also trended towards a decrease at N-cad- rich sites. These results suggest that VEGF-induced vascular leak induces acute nanoscale reorganization of NaV1.5 and β1 within the ID. Despite its high resolution, STED microscopy still has limited ability to assess protein density. In any fluores- cence image, intensity is determined by a combination of the density of fluorescently-labeled proteins and the number of photons emitted by each. In order to obtain orthogonal validation of the STED results and overcome this limitation, we turned to STORM single molecule localization microscopy and STORM-RLA machine learn- ing-based cluster analysis. By localizing individual molecules, STORM offers the unique ability to assess relative differences in protein density between different ID regions. Representative three-dimensional en face views of atrial IDs obtained by STORM show dense clusters of NaV1.5 occurring in close proximity to Cx43 and within N-cad-rich regions in control hearts (Fig. 7A,B). In VEGF-treated hearts, NaV1.5 clusters appeared more diffuse and were shifted away from Cx43 and N-cad clusters (Fig. 7C,D). In contrast, β1 was preferentially localized near Cx43 clusters and throughout N-cad-free ID regions in control hearts (Fig. 8A,B). In VEGF-treated hearts, β1 clusters appeared further from Cx43 clusters (Fig. 8). Close-up views of Cx43 clusters and associated NaV1.5 clusters supported these findings (Fig. 9A,B). STORM data were quantitatively analyzed using STORM-RLA to determine the percent of total NaV1.5/β1 signal at the ID, which was localized within Cx43-adjacent perinexal sites (≤ 100 nm from Cx43 clusters; Fig. 9C,E) and at N-cad-rich plicate ID sites (Fig. 9D,E). Additionally, signal enrichment ratio, defined as the ratio of NaV1.5/β1 molecular density at these sites vs. the density at other ID sites was also calculated. In control hearts, 59 ± 2% of NaV1.5 was localized within Cx43-adjacent perinexal sites (enrichment ratio: 10.5 ± 0.3) and 35 ± 2% within N-cad-rich plicate ID sites (enrichment ratio: 6.5 ± 0.4). In contrast, β1 displayed a marked preference for Cx43-adjacent perinexal sites (69 ± 4% of ID-localized β1, enrichment ratio: 10.7 ± 1.9) in comparison to N-cad-rich plicate ID sites (14 ± 3% of ID-localized β1). In VEGF treated hearts, NaV1.5 density was significantly reduced at both Cx43-adjacent perinexal sites (32 ± 3% of sig- nal, enrichment ratio: 6.9 ± 0.8) and N-cad-rich plicate ID sites (26 ± 3% of signal, enrichment ratio: 4.6 ± 0.4). Scientific Reports | (2020) 10:20463 | https://doi.org/10.1038/s41598-020-77562-5 6 Vol:.(1234567890)www.nature.com/scientificreports/ Figure 4. sDCI imaging of IDs. Representative 3D sDCI images of en face IDs from murine atria immunolabeled for (A, B) NaV1.5, Cx40, Cx43, and N-cad, and (C, D) NaV1.5, β1, Cx43, and N-cad. Likewise, β1 density was also reduced at Cx43-adjacent perinexal sites (49 ± 3% of signal, enrichment ratio: 5.4 ± 0.7) without significant changes at N-cad-rich plicate ID sites. Overall, the STORM-RLA results indicated dynamic reorganization of ID-localized NaV1.5 and β1 following VEGF treatment. Scientific Reports | (2020) 10:20463 | https://doi.org/10.1038/s41598-020-77562-5 7 Vol.:(0123456789)www.nature.com/scientificreports/ Figure 5. STED imaging of atrial IDs. Representative 3D STED images of en face IDs from VEGF-treated and control murine atria immunolabeled for (A) NaV1.5 and (B) β1 along with Cx43 and N-cad. Figure 6. OBS3D analysis of STED images. (A) Bivariate histograms of NaV1.5 cluster mass (normalized intensity summed over the cluster) as a function of distance from Cx43 clusters. These provide representative examples of intermediate steps in image analysis involved in assessing enrichment ratios, calculated as the ratio of NaV1.5/β1 immunosignal cluster mass (volume x normalized intensity) at sites near (< 100 nm away) Cx43 (GJ) and N-cad (MJ) clusters vs. the signal cluster mass at other ID sites. (B) Summary plots of enrichment ratio (n = 3 hearts/group, 3 images/heart; *p < 0.05 vs. control). Scientific Reports | (2020) 10:20463 | https://doi.org/10.1038/s41598-020-77562-5 8 Vol:.(1234567890)www.nature.com/scientificreports/ Figure 7. STORM imaging of atrial IDs—NaV1.5. Representative 3D STORM images of en face IDs immunolabeled for NaV1.5 along with Cx43 and N-cad from (A, B) control and (C, D) VEGF-treated murine atria. STORM data are rendered as point clouds with each localized molecule represented as a 50 nm sphere. Although 20 nm resolution was achieved, the 50 nm size was chosen for rendering to guarantee visibility in print. Figure 8. STORM imaging of atrial IDs—β1. Representative 3D STORM images of en face IDs immunolabeled for β1 along with Cx43 and N-cad from (A, B) control and (C, D) VEGF-treated murine atria. Discussion Patients with new-onset AF show elevated levels of VEGF3–6,45, a cytokine that promotes vascular leak. Indeed, inflammation, vascular leak, and associated tissue edema are common sequelae of AF2–8, and are emerging as proarrhythmic factors. In previous studies in the ventricles, myocardial edema acutely (within minutes) disrupted ID nanodomains, slowed conduction, and precipitated arrhythmias22–24. Interestingly, patients with AF also evidence swelling of ID nanodomains26 and conduction slowing has been linked to AF in human patients43,44. However, the mechanism by which tissue edema due to vascular leak precipitates AF is unknown. Therefore, we tested the hypothesis that VEGF may acutely promote atrial arrhythmias by disrupting ID nanodomains and compromising atrial conduction (Fig. 10). Here, we demonstrate that VEGF insult acutely induces ID nanodo- main swelling and translocation of sodium channel subunits from these sites, likely generating a substrate for slowed atrial conduction, and atrial arrhythmias. Cytokines such as VEGF, which induce vascular leak, have been shown to have a multitude of other impacts, including directly reducing the expression of Cx43 in cardiac myocytes46–51. In contrast, our Western blots indi- cated no change in the expression of Cx43 or Na+ channel subunits, and a slight increase in Cx40 expression fol- lowing acute VEGF insult. The apparent divergence of our results from the aforementioned studies may reflect the much longer time courses (> 4 h) involved in those compared to our study (< 1 h). Overall, our data suggest that reduced expression of ID proteins cannot explain the rapid proarrhythmic impact of VEGF in our experiments. In previous studies, acute interstitial edema induced swelling of the perinexus, a GJ-adjacent ID nanodomain, and brought about conduction slowing and spontaneous arrhythmias within 10 min22–24. Likewise, George et al. demonstrated elevated extracellular volume, ID nanodomain swelling, and conduction slowing during acute Scientific Reports | (2020) 10:20463 | https://doi.org/10.1038/s41598-020-77562-5 9 Vol.:(0123456789)www.nature.com/scientificreports/ Figure 9. STORM-RLA analysis of NaV1.5, β1 localization. Representative 3D STORM images of a Cx43 cluster and associated NaV1.5 clusters from (A) control and (B) VEGF-treated murine atria. (C, D) Bivariate histograms of NaV1.5 cluster density as a function of distance from Cx43 clusters. Dashed circles highlight the decrease in NaV1.5 clusters located near Cx43. (E) Summary plots of STORM-RLA results. Left: % of ID-localized NaV1.5 and β1 located within 100 nm of Cx43 (GJ) and N-cad (MJ) clusters. Right: Enrichment ratio, calculated as the ratio of NaV1.5/β1 cluster density within 100 nm of Cx43 (GJ) and N-cad (MJ) clusters to NaV1.5/β1 cluster density at other ID sites (n = 3 hearts/group, 10 images/heart; *p < 0.05 vs. control). Figure 10. Proposed mechanism for the genesis and progression of AF. Elevated VEGF levels in AF patients increase vascular leak, in turn promoting cardiac edema. The resulting disruption of NaV1.5-rich ID nanodomains slows atrial conduction, thereby providing a substrate for further atrial arrhythmias. inflammatory response (90 min of exposure to pathophysiological levels of TNFα)21. Consistent with these, our TEM studies identified significant swelling of ID nanodomains (near both GJs and MJs) following VEGF treat- ment. Taken together, these results suggest that ID nanodomain swelling may contribute to atrial arrhythmias following acute VEGF insult. Notably, the ultrastructural impact of VEGF in our experiments closely corresponds with observations from human AF patients26. A concomitant impact during acute swelling of ID nanodomains, suggested by previous work, is the transloca- tion of sodium channels from these sites25. Perinexal swelling was found to decrease local INa density near GJs, albeit without any change in whole-cell INa and was sufficient to induce proarrhythmic conduction slowing. These results suggest that the precise localization of sodium channels within the ID may be an important determinant of Scientific Reports | (2020) 10:20463 | https://doi.org/10.1038/s41598-020-77562-5 10 Vol:.(1234567890)www.nature.com/scientificreports/ cardiac electrical propagation. Therefore, we used super-resolution microscopy to test whether VEGF-induced ID remodeling included any reorganization of sodium channel proteins. Overall, STED and STORM both identified NaV1.5 enrichment near Cx43 clusters as well as at N-cad-rich sites, consistent with previous reports22,23,25,30,52. In contrast, β1 was preferentially localized near Cx43 and predominantly within N-cad-free ID sites, again in keeping with previous data25. These data suggest that NaV1.5 at N-cad-rich sites may associate with a different β subunit, an idea which merits future investigation. Importantly, both STED and STORM images revealed changes consistent with decreased NaV1.5 near GJs and MJs in VEGF-treated hearts relative to controls. Quantitative analysis of STED and STORM data revealed a substantial depletion of NaV1.5 from GJ-adjacent perinexal sites, and to a somewhat lesser degree, also from MJ-adjacent sites. Likewise, VEGF treatment also decreased β1 density at GJ-adjacent sites. Overall, these data, along with previously published results25, suggest that local INa density at GJ- and MJ- adjacent sites might be decreased following acute VEGF insult. Taken in the context of our TEM results, these data suggest that intermembrane adhesion within ID nanodomains may play a role in retaining sodium channels at these sites. Inhibition of adhesive interactions may enhance lateral diffusion of ion channels within the membrane, resulting in their dispersal from dense clusters. While further research will be required to uncover the precise mechanism by which nanodomain swelling induces sodium channel translocation, we provide here the first direct demonstration of this dynamic remodeling phenomenon. Taken together, our light and electron microscopy results identify two forms of dynamic ID remodeling fol- lowing acute exposure to VEGF: (1) swelling of the extracellular cleft near GJs and MJs, and (2) translocation of NaV1.5, wherein dense NaV1.5 clusters located near GJs and MJs are redistributed more diffusely. These changes could impair atrial conduction via two, non-mutually exclusive mechanisms: (1) Direct effects on membrane excitability via cooperative activation. The earliest activating NaV1.5 channels promote positive feedback acti- vation of further NaV1.5 channels, when these channels are tightly clustered, and face a restricted extracellular cleft53,54. NaV1.5 translocation away from dense clusters into a more diffuse pattern would weaken this effect, and could thereby compromise excitability. (2) Indirect effects on intercellular coupling via ephaptic coupling: When dense NaV1.5 clusters from adjacent cells face each other across a narrow (< 30 nm) extracellular cleft, channel activation on one side prompts transient depletion of sodium (positive charge) from the cleft, and subsequent depolarization of the apposed cell’s membrane, activating its NaV1.5 channels55–58. Both nanodomain swelling and the more diffuse reorganization of NaV1.5 would weaken local electrochemical transients within ID nano- domains, and could thereby impair atrial conduction22,23,25,59–61. Notably, based on their structural properties, both perinexi and plicate nanodomains would support cooperative activation but only perinexi are predicted to support ephaptic coupling59,62. However, since VEGF impacted both locations simultaneously, our results do not delineate the relative contributions of the two mechanisms, or indeed of the two different ID nanodomains. While future work will be required to answer these mechanistic questions, the totality of structural and func- tional results indicate that VEGF can acutely induce proarrhythmic conduction slowing, and likely does so by disrupting ID nanodomains (Fig. 10). Our results, identifying acute remodeling of ID nanodomains as an arrhythmia mechanism, have important implications for our broader understanding of arrhythmia substrates. Classically, structural arrhythmia substrates are viewed as being permanent (e.g. an infarct), while functional substrates are thought to be dynamic (e.g. a line of block resulting from repolarization heterogeneities). However, vascular leak-induced edema and consequent nanodomain remodeling, as demonstrated here, may represent a dynamic and transient structural arrhythmic substrate. This may contribute to the intermittent nature of arrhythmias in pathologies such as AF in the early stages. The results presented here also have important implications for the treatment of AF. First, they suggest that therapies which mitigate cytokine-induced vascular leak may be effective in preventing atrial arrhythmias. Second, they suggest that direct targeting of ID nanodomains to prevent swelling and sodium channel transloca- tion could also be an effective antiarrhythmic strategy. Limitations. VEGF’s impact on the heart is multi-factorial in nature, involving direct effects on cardiac myocytes as well as effects on non-myocyte cells. These include effects on GJs, which could contribute to con- duction slowing46–51. Although our Western blot analysis did not identify any decrease in Cx40 or Cx43 expres- sion, functional GJ coupling may have been impacted without altering overall protein expression. However, VEGF’s effects on GJs have been demonstrate to occur over much longer time courses (> 4 h) than those involved in the present study (< 4  h). Intermembrane spacing measured by TEM may have been impacted the effects of glutaraldehyde fixation on tissue63. However, such effects would uniformly impact all samples and do not detract from the observation that VEGF increases intermembrane spacing near GJs and MJs. Super-resolution microscopy revealed translocation of NaV1.5 from ID nanodomains, occurring in conjunction with increase in intermembrane spacing. While our data link these effects to arrhythmogenic conduction slowing, the inability to separate these two effects experimentally precludes delineation of their relative impacts on conduction. While this merits future investigation using experimental and modeling approaches, our data indicate that remodeling of ID nanodomains secondary to VEGF-induced vascular leak is acutely proarrhythmic. Conclusion In summary, we demonstrate that VEGF, at levels occurring in AF patients, can acutely increase susceptibility to atrial arrhythmias. We provide, to our knowledge, the first evidence that sodium channel clusters at the ID can undergo dynamic reorganization. Importantly, we identify a novel mechanism for atrial arrhythmias, wherein dynamic disruption of ID nanodomains, secondary to VEGF-induced vascular leak, induces proarrhythmic slow- ing of atrial conduction. This mechanism may contribute to the genesis and progression of AF in the early stages and help explain the link between inflammation and AF. Our work identifies vascular leak and ID nanodomains are potential therapeutic targets for the treatment and prevention of AF in the early stages. Scientific Reports | (2020) 10:20463 | https://doi.org/10.1038/s41598-020-77562-5 11 Vol.:(0123456789)www.nature.com/scientificreports/ Data availability All data generated and analyzed during the present study are available upon request from the corresponding author upon reasonable request. Received: 22 April 2020; Accepted: 9 November 2020 References 1. Zoni-Berisso, M., Lercari, F., Carazza, T. & Domenicucci, S. Epidemiology of atrial fibrillation: European perspective. Clin. Epi- demiol. 6, 213–220. https ://doi.org/10.2147/CLEP.S4738 5 (2014). 2. Weis, S. M. Vascular permeability in cardiovascular disease and cancer. Curr. Opin. Hematol. 15, 243–249. https ://doi.org/10.1097/ MOH.0b013 e3282 f97d8 6 (2008). 3. Li, J. et al. Role of inflammation and oxidative stress in atrial fibrillation. Heart Rhythm Off. J. Heart Rhythm Soc. 7, 438–444. https ://doi.org/10.1016/j.hrthm .2009.12.009 (2010). 4. Ogi, H. et al. Is structural remodeling of fibrillated atria the consequence of tissue hypoxia?. Circul. J. Off. J. Jpn. Circul. Soc. 74, 1815–1821 (2010). 5. Scridon, A. et al. Increased intracardiac vascular endothelial growth factor levels in patients with paroxysmal, but not persistent atrial fibrillation. Europace 14, 948–953. https ://doi.org/10.1093/europ ace/eur41 8 (2012). 6. Seko, Y., Nishimura, H., Takahashi, N., Ashida, T. & Nagai, R. Serum levels of vascular endothelial growth factor and transforming growth factor-beta1 in patients with atrial fibrillation undergoing defibrillation therapy. Jpn. Heart J. 41, 27–32 (2000). 7. Gramley, F. et al. Atrial fibrillation is associated with cardiac hypoxia. Cardiovasc. Pathol. 19, 102–111. https ://doi.org/10.1016/j. carpa th.2008.11.001 (2010). 8. Chung, N. A. et al. Is the hypercoagulable state in atrial fibrillation mediated by vascular endothelial growth factor?. Stroke 33, 2187–2191 (2002). 9. Sukriti, S., Tauseef, M., Yazbeck, P. & Mehta, D. Mechanisms regulating endothelial permeability. Pulm. Circ. 4, 535–551. https :// doi.org/10.1086/67735 6 (2014). 10. Kimura, T. et al. Serum inflammation markers predicting successful initial catheter ablation for atrial fibrillation. Heart Lung Circ. 23, 636–643. https ://doi.org/10.1016/j.hlc.2014.02.003 (2014). 11. Bertoluci, M. C. et al. Endothelial dysfunction as a predictor of cardiovascular disease in type 1 diabetes. World J. Diabetes 6, 679–692. https ://doi.org/10.4239/wjd.v6.i5.679 (2015). 12. de Zeeuw, D., Parving, H. H. & Henning, R. H. Microalbuminuria as an early marker for cardiovascular disease. J. Am. Soc. Nephrol. 17, 2100–2105. https ://doi.org/10.1681/ASN.20060 50517 (2006). 13. Montezano, A. C. et al. Oxidative stress and human hypertension: vascular mechanisms, biomarkers, and novel therapies. Can. J. Cardiol. 31, 631–641. https ://doi.org/10.1016/j.cjca.2015.02.008 (2015). 14. Amano, Y. et al. T2-weighted cardiac magnetic resonance imaging of edema in myocardial diseases. Sci. World J. 2012, 194069. https ://doi.org/10.1100/2012/19406 9 (2012). 15. Boyle, A., Maurer, M. S. & Sobotka, P. A. Myocellular and interstitial edema and circulating volume expansion as a cause of mor- bidity and mortality in heart failure. J. Cardiac. Fail. 13, 133–136 (2007). 16. White, S. K. et al. Remote ischemic conditioning reduces myocardial infarct size and edema in patients with ST-segment elevation myocardial infarction. JACC Cardiovasc. Interv. 8, 178–188. https ://doi.org/10.1016/j.jcin.2014.05.015 (2015). 17. Zia, M. I. et al. Comparison of the frequencies of myocardial edema determined by cardiac magnetic resonance in diabetic versus nondiabetic patients having percutaneous coronary intervention for ST elevation myocardial infarction. Am. J. Cardiol. 113, 607–612. https ://doi.org/10.1016/j.amjca rd.2013.10.040 (2014). 18. Migliore, F., Zorzi, A., Perazzolo Marra, M., Iliceto, S. & Corrado, D. Myocardial edema as a substrate of electrocardiographic abnormalities and life-threatening arrhythmias in reversible ventricular dysfunction of takotsubo cardiomyopathy: imaging evi- dence, presumed mechanisms, and implications for therapy. Heart Rhythm Off. J. Heart Rhythm Soc. https ://doi.org/10.1016/j. hrthm .2015.04.041 (2015). 19. Neilan, T. G. et al. Myocardial extracellular volume expansion and the risk of recurrent atrial fibrillation after pulmonary vein isolation. JACC Cardiovasc. Imaging 7, 1–11. https ://doi.org/10.1016/j.jcmg.2013.08.013 (2014). 20. Arujuna, A. et al. Acute pulmonary vein isolation is achieved by a combination of reversible and irreversible atrial injury after cath- eter ablation: evidence from magnetic resonance imaging. Circul. Arrhythmia Electrophysiol. 5, 691–700. https ://doi.org/10.1161/ CIRCE P.111.96652 3 (2012). 21. George, S. A., Calhoun, P. J., Gourdie, R. G., Smyth, J. W. & Poelzing, S. TNFalpha modulates cardiac conduction by altering electrical coupling between myocytes. Front. Physiol. 8, 334. https ://doi.org/10.3389/fphys .2017.00334 (2017). 22. Veeraraghavan, R. et al. Sodium channels in the Cx43 gap junction perinexus may constitute a cardiac ephapse: an experimental and modeling study. Pflugers Arch. 467, 2093–2105. https ://doi.org/10.1007/s0042 4-014-1675-z (2015). 23. Veeraraghavan, R., Lin, J., Keener, J. P., Gourdie, R. & Poelzing, S. Potassium channels in the Cx43 gap junction perinexus modulate ephaptic coupling: an experimental and modeling study. Pflugers Arch. 468, 1651–1661. https ://doi.org/10.1007/s0042 4-016-1861-2 (2016). 24. Veeraraghavan, R., Salama, M. E. & Poelzing, S. Interstitial volume modulates the conduction velocity-gap junction relationship. Am. J. Physiol. Heart Circul. Physiol. 302, H278–H286 (2012). 25. Veeraraghavan, R. et al. The adhesion function of the sodium channel beta subunit (beta1) contributes to cardiac action potential propagation. Elife https ://doi.org/10.7554/eLife .37610 (2018). 26. Raisch, T. B. et al. Intercalated disc extracellular nanodomain expansion in patients with atrial fibrillation. Front. Physiol. 9, 398 (2018). 27. Radwanski, P. B. et al. Neuronal Na+ channel blockade suppresses arrhythmogenic diastolic Ca2+ release. Cardiovasc. Res. 106, 143–152. https ://doi.org/10.1093/cvr/cvu26 2 (2015). 28. Radwanski, P. B., Veeraraghavan, R. & Poelzing, S. Cytosolic calcium accumulation and delayed repolarization associated with ventricular arrhythmias in a guinea pig model of Andersen-Tawil syndrome. Heart Rhythm Off. J. Heart Rhythm Soc. 7, 1428–1435 (2010). 29. Veeraraghavan, R. & Poelzing, S. Mechanisms underlying increased right ventricular conduction sensitivity to flecainide challenge. Cardiovasc. Res. 77, 749–756 (2008). 30. Veeraraghavan, R. & Gourdie, R. Stochastic optical reconstruction microscopy-based relative localization analysis (STORM-RLA) for quantitative nanoscale assessment of spatial protein organization. Mol. Biol. Cell 27, 3583–3590. https ://doi.org/10.1091/mbc. E16-02-0125 (2016). 31. Girouard, S. D., Laurita, K. R. & Rosenbaum, D. S. Unique properties of cardiac action potentials recorded with voltage-sensitive dyes. J. Cardiovasc. Electrophysiol. 7, 1024–1038 (1996). 32. Bayly, P. V. et al. Estimation of conduction velocity vector fields from epicardial mapping data. IEEE Trans. Bio-med. Eng. 45, 563–571 (1998). Scientific Reports | (2020) 10:20463 | https://doi.org/10.1038/s41598-020-77562-5 12 Vol:.(1234567890)www.nature.com/scientificreports/ 33. Greer-Short, A. et al. Calmodulin kinase II regulates atrial myocyte late sodium current, calcium handling, and atrial arrhythmia. Heart Rhythm Off. J. Heart Rhythm Soc. 17, 503–511. https ://doi.org/10.1016/j.hrthm .2019.10.016 (2020). 34. Aschar-Sobbi, R. et al. Increased atrial arrhythmia susceptibility induced by intense endurance exercise in mice requires TNFalpha. Nat. Commun. 6, 6018. https ://doi.org/10.1038/ncomm s7018 (2015). 35. Koleske, M. et al. Tetrodotoxin-sensitive Navs contribute to early and delayed afterdepolarizations in long QT arrhythmia models. J. Gener. Physiol. https ://doi.org/10.1085/jgp.20171 1909 (2018). 36. Struckman, H. L. et al. Super-resolution imaging using a novel high-fidelity antibody reveals close association of the neuronal sodium channel NaV1.6 with ryanodine receptors in cardiac muscle. Microsc. Microanal. https ://doi.org/10.1017/S1431 92761 90152 89 (2020). 37. Radwański, P. B. et al. Neuronal Na+ channels are integral components of pro-arrhythmic Na+/Ca2+ signaling nanodomain that promotes cardiac arrhythmias during β-adrenergic stimulation. JACC Basic Transl. Sci. 1, 251–266. https ://doi.org/10.1016/j.jacbt s.2016.04.004 (2016). 38. Lam, F., Cladiere, D., Guillaume, C., Wassmann, K. & Bolte, S. Super-resolution for everybody: an image processing workflow to obtain high-resolution images with a standard confocal microscope. Methods 115, 17–27. https ://doi.org/10.1016/j.ymeth .2016.11.003 (2017). 39. Bonilla, I. M. et al. Enhancement of cardiac store operated calcium entry (SOCE) within novel intercalated disk microdomains in arrhythmic disease. Sci. Rep. 9, 10179. https ://doi.org/10.1038/s4159 8-019-46427 -x (2019). 40. Kleber, A. G. & Rudy, Y. Basic mechanisms of cardiac impulse propagation and associated arrhythmias. Physiol. Rev. 84, 431–488 (2004). 41. Kleber, A. G. Discontinuous propagation of the cardiac impulse and arrhythmogenesis. J. Cardiovasc. Electrophysiol. 10, 1025–1027 (1999). 42. Radwanski, P. B., Johnson, C. N., Gyorke, S. & Veeraraghavan, R. Cardiac arrhythmias as manifestations of nanopathies: an emerg- ing view. Front. Physiol. 9, 1228. https ://doi.org/10.3389/fphys .2018.01228 (2018). 43. Zheng, Y., Xia, Y., Carlson, J., Kongstad, O. & Yuan, S. Atrial average conduction velocity in patients with and without paroxysmal atrial fibrillation. Clin. Physiol. Funct. Imaging 37, 596–601. https ://doi.org/10.1111/cpf.12342 (2017). 44. Lalani, G. G. et al. Atrial conduction slows immediately before the onset of human atrial fibrillation: a bi-atrial contact mapping study of transitions to atrial fibrillation. J. Am. Coll. Cardiol. 59, 595–606. https ://doi.org/10.1016/j.jacc.2011.10.879 (2012). 45. Smorodinova, N. et al. Bioptic study of left and right atrial interstitium in cardiac patients with and without atrial fibrillation: interatrial but not rhythm-based differences. PLoS ONE 10, e0129124. https ://doi.org/10.1371/journ al.pone.01291 24 (2015). 46. Dhein, S., Polontchouk, L., Salameh, A. & Haefliger, J. A. Pharmacological modulation and differential regulation of the cardiac gap junction proteins connexin 43 and connexin 40. Biol. Cell 94, 409–422 (2002). 47. Pimentel, R. C., Yamada, K. A., Kleber, A. G. & Saffitz, J. E. Autocrine regulation of myocyte Cx43 expression by VEGF. Circ. Res. 90, 671–677 (2002). 48. Fernandez-Cobo, M., Gingalewski, C., Drujan, D. & De Maio, A. Downregulation of connexin 43 gene expression in rat heart during inflammation. The role of tumour necrosis factor. Cytokine 11, 216–224. https ://doi.org/10.1006/cyto.1998.0422 (1999). 49. Herve, J. C. & Dhein, S. Pharmacology of cardiovascular gap junctions. Adv. Cardiol. 42, 107–131. https ://doi.org/10.1159/00009 2565 (2006). 50. Salameh, A. et al. Chronic regulation of the expression of gap junction proteins connexin40, connexin43, and connexin45 in neonatal rat cardiomyocytes. Eur. J. Pharmacol. 503, 9–16. https ://doi.org/10.1016/j.ejpha r.2004.09.024 (2004). 51. Sawaya, S. E. et al. Downregulation of connexin40 and increased prevalence of atrial arrhythmias in transgenic mice with cardiac- restricted overexpression of tumor necrosis factor. Am. J. Physiol. Heart Circul. Physiol. 292, H1561-1567. https ://doi.org/10.1152/ ajphe art.00285 .2006 (2007). 52. Leo-Macias, A. et al. Nanoscale visualization of functional adhesion/excitability nodes at the intercalated disc. Nat. Commun. 7, 10342. https ://doi.org/10.1038/ncomm s1034 2 (2016). 53. Hichri, E., Abriel, H. & Kucera, J. P. Distribution of cardiac sodium channels in clusters potentiates ephaptic interactions in the intercalated disc. J. Physiol. https ://doi.org/10.1113/JP275 351 (2017). 54. Clatot, J. et al. Voltage-gated sodium channels assemble and gate as dimers. Nat. Commun. 8, 2077. https ://doi.org/10.1038/s4146 7-017-02262 -0 (2017). 55. Veeraraghavan, R., Gourdie, R. & Poelzing, S. Mechanisms of cardiac conduction: a history of revisions. Am. J. Physiol. Heart Circul. Physiol. https ://doi.org/10.1152/ajphe art.00760 .2013 (2014). 56. Veeraraghavan, R., Poelzing, S. & Gourdie, R. G. Old cogs, new tricks: a scaffolding role for connexin43 and a junctional role for sodium channels?. FEBS Lett. https ://doi.org/10.1016/j.febsl et.2014.01.026 (2014). 57. Veeraraghavan, R., Poelzing, S. & Gourdie, R. G. Intercellular electrical communication in the heart: a new, active role for the intercalated disk. Cell Commun. Adhes. https ://doi.org/10.3109/15419 061.2014.90593 2 (2014). 58. Veeraraghavan, R. & Radwanski, P. B. Sodium channel clusters: harmonizing the cardiac conduction orchestra. J. Physiol. 596, 549–550. https ://doi.org/10.1113/JP275 632 (2018). 59. Mori, Y., Fishman, G. I. & Peskin, C. S. Ephaptic conduction in a cardiac strand model with 3D electrodiffusion. Proc. Natl. Acad. Sci. U.S.A. 105, 6463–6468 (2008). 60. Kucera, J. P., Rohr, S. & Rudy, Y. Localization of sodium channels in intercalated disks modulates cardiac conduction. Circ. Res. 91, 1176–1182 (2002). 61. Lin, J. & Keener, J. P. Modeling electrical activity of myocardial cells incorporating the effects of ephaptic coupling. Proc. Natl. Acad. Sci. U.S.A. 107, 20935–20940 (2010). 62. Lin, J. & Keener, J. P. Ephaptic coupling in cardiac myocytes. IEEE Trans. Bio-med. Eng. 60, 576–582 (2013). 63. Raisch, T., Khan, M. & Poelzing, S. Quantifying intermembrane distances with serial image dilations. J. Vis. Exp. JoVE https ://doi. org/10.3791/58311 (2018). Author contributions L.M. conducted the majority of the experiments, performed data analysis, prepared figures, and wrote, revised and edited the manuscript. H.L.S. performed confocal microscopy experiments and assisted with optical map- ping studies. A.G.S. performed the optical mapping experiments and helped with data analysis. S.B. performed the immunoblotting studies. S.G. helped with hypothesis development and data interpretation, and helped with manuscript editing. P.B.R. helped with electrocardiography studies, hypothesis development and data interpreta- tion, and helped with manuscript editing. T.J.H. helped with optical mapping experiments, hypothesis develop- ment, data interpretation and manuscript preparation and editing. R.V. conceived the study, oversaw experiments, performed image analysis, manuscript writing, revision, and editing. All authors reviewed the manuscript. Scientific Reports | (2020) 10:20463 | https://doi.org/10.1038/s41598-020-77562-5 13 Vol.:(0123456789)www.nature.com/scientificreports/ Funding This work was supported by the National Institutes of Health [R01HL148736 awarded to RV, R01HL063043 and R01HL074045 awarded to S.G, and R01HL135096 and R01HL134824 to TJH] and the American Heart Association [16SDG29870007 awarded to RV, 19TPA34910191 to PBR, and Postdoctoral fellowship to AGS]. Competing interests The authors declare no competing interests. Additional information Supplementary information is available for this paper at https ://doi.org/10.1038/s4159 8-020-77562 -5. Correspondence and requests for materials should be addressed to R.V. Reprints and permissions information is available at www.nature.com/reprints. Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/. © The Author(s) 2020 Scientific Reports | (2020) 10:20463 | https://doi.org/10.1038/s41598-020-77562-5 14 Vol:.(1234567890)www.nature.com/scientificreports/
null
10.1073_pnas.2304611120.pdf
Data, Materials, and Software Availability. Structure factors and refined coordinates obtained from X- ray crystallography have been deposited into the Protein Data Bank (www.wwpdb.org) under PDB accession codes: 8SSP (72) (AurA- danusertib- Mb1), 8SSO (73)  (AurA- danusertib- Mb2), and 8SSN (74) (Abl64– 510-SKI- ascimi
null
RESEARCH ARTICLE | BIOCHEMISTRY OPEN ACCESS A biophysical framework for double- drugging kinases Chansik Kima,b,1 , Adelajda Hadzipasica,b,2, Steffen Kuttera,b,3, Vy Nguyena,b,4, and Dorothee Kerna,b,5 , Hannes Ludewiga,b Edited by Melanie Cobb, The University of Texas Southwestern Medical Center, Dallas, TX; received March 26, 2023; accepted July 6, 2023 Selective orthosteric inhibition of kinases has been challenging due to the conserved active site architecture of kinases and emergence of resistance mutants. Simultaneous inhibition of distant orthosteric and allosteric sites, which we refer to as “double- drugging”, has recently been shown to be effective in overcoming drug resistance. However, detailed biophysical characterization of the cooperative nature between orthosteric and allosteric modulators has not been undertaken. Here, we provide a quantitative framework for double- drugging of kinases employing isothermal titration calorimetry, Förster resonance energy transfer, coupled- enzyme assays, and X- ray crystallography. We discern positive and negative cooperativity for Aurora A kinase (AurA) and Abelson kinase (Abl) with different combinations of orthosteric and allosteric modulators. We find that a confor- mational equilibrium shift is the main principle governing cooperativity. Notably, for both kinases, we find a synergistic decrease of the required orthosteric and allosteric drug dosages when used in combination to inhibit kinase activities to clinically relevant inhibition levels. X- ray crystal structures of the double- drugged kinase complexes reveal the molecular principles underlying the cooperative nature of double- drugging AurA and Abl with orthosteric and allosteric inhibitors. Finally, we observe a fully closed confor- mation of Abl when bound to a pair of positively cooperative orthosteric and allosteric modulators, shedding light on the puzzling abnormality of previously solved closed Abl structures. Collectively, our data provide mechanistic and structural insights into rational design and evaluation of double- drugging strategies. kinase | conformational equilibrium | cooperativity | double- drugging Protein allostery is one of the fundamental regulatory mechanisms involved in various biological processes (1). Specifically, the allosteric regulation of protein kinases has been found essential for signaling cascades. Thus, dysregulation and overexpression of protein kinases are often related to many human diseases, including various cancers. However, due to the highly conserved catalytic site architecture of kinases, specific orthosteric inhi- bition is often unsuccessful, causing off- target effects (2). In addition, cancers often develop resistant mutations circumventing treatments with orthosteric drugs (3, 4). To overcome these problems, the field has been exploring allosteric sites of kinases for specific and efficacious inhibition (5, 6). A recently approved allosteric inhibitor of Abelson kinase (Abl), asciminib, has been highly effective in inhibiting Abl in vitro and in vivo (7–12). Remarkably, dual inhibition of Abl with this allosteric inhibitor combined with the orthosteric inhibitors (including imatinib, nilotinib, and ponatinib), which we refer to as “double- drugging”, has been impressively successful in abolishing the emergence of resistant mutants for Abl (12–14). Considering this clinical benefit, this approach has been applied to inhibit other targets such as EGFR kinase and SHP2 phosphatase (15, 16). However, the biophysical mechanisms underlying double- drugging of distant orthosteric and allosteric sites have not been well studied. Herein, we provide the quantitative framework for double- drugging using two targets: Aurora A kinase (AurA) and Abl. Both kinases participate in various cellular pathways, and their dysregulation results in a multitude of cancers, such as breast cancer and leukemia (17–19). Common obstacles faced by orthosteric inhibitors for AurA and Abl include cytotoxicity, off- target effects, and emergence of resistance mutants (3, 4, 20, 21). For both systems, we exploit a rational selection of ligands to probe positive and negative coopera- tivity between remote orthosteric and allosteric sites using isothermal titration calorimetry (ITC), Förster resonance energy transfer (FRET), and coupled- enzyme assays. We find that both orthosteric and allosteric ligands exhibit preferred binding to the active or inactive states and that cooperativity occurs by shifting this active–inactive conformational equi- librium through long- range allosteric networks that are encoded for natural regulation of those kinases. X- ray crystal structures of the double- drugged complexes shed light on the atomistic mechanisms of cooperativity. After we determine negative cooperativity for the double- drug combination used by Novartis in their clinical trials, we rationally chose a different orthosteric inhibitor, Src inhibitor 1 (SKI), for positive cooperativity with Significance While immensely successful, drugging kinases by active site inhibitors has faced major challenges. Selectivity issues leading to side effects and emergence of resistance mutations rendered treatments targeting active sites ineffective. Double- drugging via active and allosteric sites is a recently developed approach to overcome these obstacles. Using Aurora A and Abelson kinase, we provide a quantitative biophysical evaluation of double- drugging by rationally selecting inhibitor combinations with positive cooperativity. The results shed light on the interplay of kinase conformational equilibria and inhibitor- dose requirements for effective inhibition. Due to our rational selection of a positively cooperative drug combination for Abl, we deliver a fully closed, inactive Abl structure, including regulatory SH3 and SH2 domains. Collectively, this biophysical framework aids future rational double- drug designs. Preprint: This manuscript has been submitted to bioRxiv under a CC- BY 4.0 International license. This article is a PNAS Direct Submission. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1Present address: NoveltyNobility, Gyeonggi- do, Seongnam- si 13477, Republic of Korea. 2Present address: Novartis Institutes for Biomedical Research, Inc., Oncology Drug Discovery, Cambridge, MA 02139. 3Present address: Schrödinger, Inc., Natick, MA 01760. 4Present address: Relay Therapeutics, Cambridge, MA 02139. 5To whom correspondence may be addressed. Email: dkern@brandeis.edu. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2304611120/- /DCSupplemental. Published August 17, 2023. PNAS  2023  Vol. 120  No. 34  e2304611120 https://doi.org/10.1073/pnas.2304611120   1 of 11 asciminib. This double- drug combination forms a unique ternary complex, revealing a fully closed Abl structure. Results Cooperative Binding between Orthosteric and Allosteric Modulators of AurA. In solution, AurA exists in a conformational equilibrium between active and inactive states (22–24). We previously designed monobodies (Mbs) that are fully selective allosteric modulators, which bind to the natural allosteric regulatory site of AurA on the N- terminal lobe (N- lobe), the binding site for the natural coactivator protein TPX2 (25, 26). Different monobodies either act as activators or inhibitors depending on how they shift the active/inactive conformational equilibrium of AurA (25). To achieve double- drugging on AurA, we combined these Mbs with the orthosteric inhibitor danusertib (PHA739358) that tightly binds to AurA [IC50 = 13 nM, Ki = 0.87 ± 1.44 nM] (27). Since it has been shown that danusertib preferentially binds to the inactive conformation of AurA (22, 28), we hypothesized that inhibiting Mbs would bind tighter to AurA when in complex with the orthosteric inhibitor danusertib. Conversely, binding of activating Mbs to AurA should be weakened in the presence of danusertib (Fig. 1A). Aligning with our hypothesis, we find that the binding affinity of activating monobody (Mb1) to AurA weakens 16- fold when AurA is presaturated with danusertib (Fig. 1B). To test whether the binding of Mb1 and danusertib is mutually exclusive, we repeated this experiment by preincubating AurA with a higher concentration of danusertib (SI Appendix, Fig. S1E). Identical Mb1 binding affin- ities, independent of saturating danusertib concentrations, reveal that the simultaneous binding of Mb1 and danusertib to AurA is possible. Thus, we reason that this 16- fold negative cooperativity for Mb1 binding arises from a conformational equilibrium shift of AurA to the inactive state induced by danusertib. To achieve desired positive cooperativity between allosteric and orthosteric binders to AurA, we chose the inhibiting monobodies Mb2 and Mb3 because i) Mb2 is an inhibiting monobody for which we had obtained an X- ray crystal structure in complex with AurA, ii) Mb3 exhibits larger inhibition than Mb2, and iii) AurA- Mb3 complex exists in a monomeric form unlike the dimeric AurA- Mb2 complex (25). We indeed measure a twofold tighter binding of Mb2 to the AurA- danusertib complex compared to apo AurA (Fig. 1B). Using the equilibrium constant for active/inactive states of AurA previously determined [Keq= 0.67 (22)] and assuming identical affin- ities of Mb2 to the inactive states of apo- or danusertib- bound AurA, we fit our apparent affinities to a reversible two- state allosteric model. We find that the twofold positive cooperativity can be explained solely by the shift in the conformational equilibrium (SI Appendix, Fig. S2). Thus, a further increase in positive cooperativity would only be possible if the Mb affinity was tighter to the inactive state of the AurA- danusertib complex than to the inactive state of apo AurA. We indeed observed a threefold positive cooperativity for Mb3 with danusertib (Fig. 1B). We speculate that this increased affinity of Mb3 to the AurA- danusertib complex compared to apo AurA could result from favorable interactions with a closed activation loop, since danu- sertib binding shifts the equilibrium of the activation loop toward such conformation (23, 28). To confirm whether the mechanism of cooperativity between Mbs and danusertib follows a classic allosteric model, we tested binding of Mb6 to the AurA- danusertib complex. Despite high affinity, Mb6 binding does not change AurA’s activity, implying that Mb6 binding does not shift the active/inactive conformational equilibrium of AurA (25). Indeed, the binding affinity of Mb6 to AurA is not changed in the presence of danusertib (Fig. 1B and SI Appendix, Fig. S3). Fig.  1. Double- drugging of AurA kinase with orthosteric drug danusertib and different allosteric modulators. (A) Schematic representation of active/ inactive equilibrium of AurA [green, PDB- ID: 5G15, and orange, PDB- ID: 6C83 (25)]. Arrows indicate binding of danusertib and Mbs to their preferred AurA conformations. The table represents the rationale of positive and negative cooperativity for double- drugging of AurA. (B) Effect of preincubation with danusertib on observed dissociation constants (apparent Kd) of different monobodies measured by ITC. Activating monobody Mb1 shows 16- fold negative cooperativity, while inhibiting monobodies, Mb2 and Mb3, show twofold and threefold positive cooperativity, respectively. Mb6 binding is not affected by the presence of danusertib (SI Appendix, Fig. S3). (C) Reversal of preincubation order during affinity measurements shows identical cooperativities for orthosteric/allosteric ligand combinations (SI  Appendix, Fig. S1). Errors in (B and C) ITC data bar graph represent 68.3% CI (±1 SD) of the fit of the data. (D) Kinase inhibition curves of AurA and AurA in the presence of saturating concentrations of Mb1, Mb2 and Mb3 as a function of danusertib concentration. Enzyme assays were conducted (n = 2, mean ± SDM) under kcat/Km condition with 3 mM Lats2 peptide, measuring observed activity (kobs). With inhibiting Mb2 and Mb3, fourfold and 20- fold lower concentrations of danusertib, respectively, are required to inhibit to 10% residual AurA activity [(Danusertib)10% act., and dashed line]. Errors in this bar graph were determined by jackknifing the inhibition curve data. For a reversible two- state allosteric model, the same fold- change of cooperativity must be observed when reversing the order of binding. To measure changes in the affinity of danusertib upon Mb binding, we had to employ competitive replacement ITC with adenosine 5′- (α, β- methylene) diphosphate (AMPCP), since 2 of 11   https://doi.org/10.1073/pnas.2304611120 pnas.org danusertib binds too tightly to AurA for direct measurement. Indeed, the measured cooperativities are matching quantitatively regardless of the binding order (Fig. 1C). Double- Drugging of AurA Lowers Inhibitor Concentration Needed for Efficacious Inhibition. Next, we probed biological relevance of these observed cooperativities by measuring the inhibition of AurA kinase activity using Lats2 peptide as a substrate with cellular ATP concentrations. Preincubation of inhibiting Mbs resulted in a vastly decreased amount of danusertib required to cause 90% inhibition of AurA activity (Fig. 1D). This combined inhibition effect is the direct consequence of positive cooperativity. For instance, Mb3, which displays a larger degree of positive cooperativity than Mb2, causes a larger reduction in required amount of danusertib for effective inhibition. X- Ray Crystal Structures of Ternary Complexes: AurA- danusertib- Mb1 and AurA- danusertib- Mb2. We solved X- ray crystal structures of double- drugged AurA complexes to further understand the structural features responsible for the positive and negative cooperativity between Mbs and danusertib (SI Appendix, Table S1). The complex of AurA- danusertib- Mb1 [active, DFGin, BLAminus (29, 30)] displays hallmarks of an active kinase, such as an intact regulatory spine, the α- C helix in the “in” position, and the “DFG- in” state (Fig. 2A). However, we note that in contrast to the AurA- AMPPCP- Mb1 structure (PDB- ID: 5G15), D274 is rotated away from danusertib to avoid a steric clash with the terminal phenyl ring of danusertib (SI Appendix, Fig. S4A). This crystal structure corroborates the capability of danusertib to bind to the active conformation of AurA as we had tested biochemically (SI Appendix, Fig. S1B). The most interesting structure for “double- drugging” with maxi- mal inhibition is of course the ternary complex of AurA- danusertib- Mb2 [inactive, DFGinter (29, 30)]. Like AurA- AMPPCP- Mb2 (PDB- ID: 6C83), this ternary complex displays features of an inactive kinase: α- C helix “out”, “DFG- out,” as well as both a broken regu- latory spine and a broken canonical salt bridge (K162- E181) (Fig. 2B). This is expected due to the conformational equilibrium shift caused by Mb2 binding and the preferential binding of danus- ertib to inactive AurA. Furthermore, the activation loop is fully shifted Fig.  2. Proposed molecular mechanism for negative and positive cooperativity for double- drugging AurA with danusertib in combination with Mb1 and Mb2, respectively. (A and B) Zoom- in of X- ray crystal structures of AurA (gray) complexed with danusertib and either Mb1 (A, green) or Mb2 (B, gold). An intact regulatory spine, DFG- in conformation and extended activation loop in A is contrasted to a broken regulatory spine, DFG- out, and closed activation loop in (B). This closed activation loop provides additional hydrophobic interaction to the terminal ring of danusertib. (C–H) Orthosteric binding sites for six different AurA states reveal why danusertib has higher affinity for inactive AurA (22, 27, 31). K162 and E181, which form the canonical salt bridge in active AurA, and D274 and F275 (DFG- motif), are shown in stick representation. (D–E) While K162- E181 salt bridge is established in an active AurA conformations, (C) this salt bridge is broken in AurA- danusertib- Mb1 structure as K162 interacts with danusertib. (F–H) In the inactive AurA conformations, DFG- out F275 is positioned between K162 and E181, physically blocking the salt- bridge interaction, thereby prepositioning K162 for danusertib binding. (A–H) Oxygen, nitrogen, and phosphorous atoms are colored in red, blue, and orange, respectively. Carbon atoms are colored according to their respective protein cartoon. PNAS  2023  Vol. 120  No. 34  e2304611120 https://doi.org/10.1073/pnas.2304611120   3 of 11 toward the active site, providing additional hydrophobic interactions to the terminal ring of danusertib (Fig. 2B). This shifted activation loop is a major structural feature of an inactive AurA (23, 28), as observed in AurA bound to the orthosteric inhibitor MLN8054 (PDB- ID: 2WTV) (32). The structure of AurA- AMPPCP- Mb2 (PDB- ID: 6C83) displays a similar activation loop, however, with an extended portion being disordered (residues 276 to 290) to circum- vent clashing with the β- and γ- phosphate groups of AMPPCP (SI Appendix, Fig. S4B). The binary complex between AurA and danusertib (PDB- ID: 2J50) did not exhibit such a shift in the acti- vation loop. However, it is unclear whether the activation loop conformation in the AurA- danusertib structure reflects the solution state since the activation loop is directly involved in crystal contacts (SI Appendix, Fig. S5). Our crystal structures and ITC experiments showed that danu- sertib can bind to both the AurA- Mb2 and AurA- Mb1 complexes. To reveal why danusertib, however, binds with much higher affin- ity to AurA- Mb2 than AurA- Mb1 (there is no steric hindrance), we scrutinized the thermodynamic parameters of our ITC studies on danusertib binding to different AurA- Mb complexes (SI Appendix, Fig. S1 A–D). We find that the enthalpy for danu- sertib binding to AurA- Mb1 is reduced by 22.8 kJ/mol compared to AurA- Mb2, which approximates the equivalence of one salt bridge [12.6 to 20.9 kJ/mol (33)]. The canonical salt bridge between K162 and E181 is a feature of an active AurA, in both its apo form and bound to AMPPNP (PDB: 6CPE and 2DWB, respectively) (22) (Fig. 2 D and E). In contrast, the ternary com- plex of AurA- danusertib- Mb1 displays a broken salt bridge, as K162 interacts now with danusertib, while maintaining the α- C helix in the “in” position (Fig. 2C). Thus, we propose that the K162- E181 salt bridge in AurA- Mb1 must be broken for danus- ertib binding, as reflected by the lowered binding enthalpy. To confirm that apo AurA- Mb1 complex establishes the K162- E181 salt bridge, we deleted danusertib from the structure of AurA- danusertib- Mb1 and carried out molecular dynamics simulations in triplicate. We observed that K162- E181 indeed forms this salt bridge on average 80.8% in a 10 ns simulation (SI Appendix, Fig. S6A). However, in the presence of danusertib, we observe K162 to rather form a hydrogen bond with O- 27 of danusertib’s methoxy moiety than with E181 in the MD simulation, which is the state sampled in our crystal structure as well (Fig. 2C and SI Appendix, Fig. S6 B and C). In the inactive conformations of AurA, the broken K162- E181 salt bridge stems from the α- C helix and DFG- motif being posi- tioned in the “out” conformation such that F275 positions between K162 and E181 (PDB: 4C3R and 2J50) (27, 31) (Fig. 2 G and H). Thus, we propose that K162 in inactive AurA conforma- tions, such as AurA- Mb2 complex, is prepositioned for danusertib binding (Fig. 2F), which results in the tighter binding of danus- ertib to the inactive state of AurA. Cooperative Effect of Imatinib and Asciminib Binding on Abl. Intrigued by our mechanistic insights into double- drugging of AurA, we turned to Abl, the only target currently in clinical trials for double- drugging. It has been shown that the combination of the orthosteric inhibitor imatinib and the allosteric inhibitor asciminib abolishes the emergence of resistance mutations (7–12), an impressive breakthrough. Therefore, Abl embodies a powerful target to delineate the biophysical constraints, or “framework”, for successful double- drugging. Since the quantitative biophysical parameters for this drug combination are not known, we set out to biophysically investigate the cooperativity and modulation of Abl’s open/closed conformational equilibrium first using this exact combination of orthosteric and allosteric inhibitors. Note that we use the well- established relevant construct of SH3- SH2- KD Abl (Abl64– 510) (SI Appendix, Fig. S7). Abl exists in a conformational equilibrium between open, active, and closed, inactive conformations (34–36) (Fig.  3A). In the open conformation, the regulatory domains are elongated so that the SH2 domain moves onto the N- lobe of the kinase domain, forming a “top- hat” conformation (35). In the closed conformation, the regulatory domains tightly interact with the kinase domain, SH3:N- lobe and SH2:C- lobe, the latter facilitated by the bent C- terminal α- I helix (12, 35, 37) (Fig. 3A). This conformational equilibrium is susceptible to modulation by single agents such as imatinib and asciminib (12, 34, 38). In ITC experiments, we find that imatinib binds fivefold tighter to AblKD, which exists exclusively in the open conformation, than to Abl64– 510 (Fig. 3B). This confirms imatinib’s preferential binding to the open state of Abl (34). In full agreement with this model, imatinib binds to Abl64– 510 with a fourfold decreased affin- ity in the presence of asciminib, since asciminib shifts the equi- librium to the closed state (12) (Fig. 3B). We conclude that this fourfold negative cooperativity between imatinib and asciminib stems from a shift in the conformational equilibrium of Abl, where both drugs preferentially bind to the open and closed conforma- tion, respectively. Akin to AurA, preincubation of Abl64– 510 with increased concentration of asciminib did not result in a weakened imatinib affinity, confirming the simultaneous binding of the two inhibitors (SI Appendix, Fig. S8B). Surprisingly, we found that imatinib and asciminib display a twofold negative cooperativity for AblKD (Fig. 3B). This implies the presence of an additional conformational equilibrium within the kinase domain itself (39) and that asciminib and imatinib shift this equilibrium in opposite directions. We refer herein to the asciminib- favoring conformation as the “closing- competent” conformation of AblKD. Importantly, we measure identical negative cooperativities between imatinib and asciminib on both AblKD and Abl64– 510, regardless of binding order, within the range of errors (Fig. 3C and SI Appendix, Fig. S8E). Due to the tight binding of asciminib, its affinity was measured via competitive replacement ITC using N- Myr peptide as a weak- binding ligand (12, 40). Collectively, we conclude that the binding of imatinib to the orthosteric site and asciminib to the allosteric site in Abl64– 510 follow a two- state allosteric model, in which the two drugs favor the closed and open conformation, respectively. Positive Cooperativity between SKI and Asciminib on Abl. Considering the negative cooperativity between imatinib and asciminib described by our ITC experiments, we wanted to rationally select an orthosteric inhibitor that exhibits positive cooperativity with asciminib. We chose Src inhibitor 1 (SKI), an orthosteric inhibitor that tightly binds to Src kinase (IC50 = 44 nM) (41, 42), because Bannister et al. recently measured that SKI preferentially binds to the α- C helix out, and thus closed- inactive conformation of Src kinase, despite the DFG- motif being in the “in” position (SKI was therefore traditionally classified as type I inhibitor) (Unpublished data, Bannister et  al.). Due to Abl and Src kinases’ close structural homology, we hypothesized that SKI would bind to Abl in a similar fashion, thus exhibiting positive cooperativity with asciminib by preferentially binding to the closed state of Abl. Since SKI binding to Abl did not result in a detectable heat change in ITC, we turned to FRET experiments to quantify this interaction (Fig.  3D and SI  Appendix, Fig.  S9 and Fig.  S10). SKI indeed binds preferentially to the closed conformation of Abl64– 510, as seen by the fivefold tighter binding of SKI to Abl64– 510 than to AblKD. Furthermore, we observe a modest positive cooperativity between SKI and asciminib binding in AblKD, indicating that SKI binds to the “closing- competent” 4 of 11   https://doi.org/10.1073/pnas.2304611120 pnas.org Fig. 3. Double- drugging of Abl kinase. (A) Schematic representation of conformational equilibrium in Abl kinase. Arrows indicate binding of orthosteric inhibitors imatinib and SKI, and allosteric inhibitor asciminib to preferred Abl conformations. X- ray crystal structures from PDB- ID: 1OPL (green) and PDB- ID: 5MO4 (red) were used for open and closed Abl structures, respectively (12, 35). Note that the SH3 domain in the open structure is missing due to lacking electron density. (B) With ITC experiments, we observe twofold and fourfold negative cooperativity for open- conformation binder imatinib when AblKD and Abl64– 510, respectively, are preincubated with closed- conformation binder asciminib (SI Appendix, Fig. S8). (C) Matching fold- change of negative cooperativity is observed reversing the order of modulators used in preincubation and titration, using ITC (SI Appendix, Fig. S8). (B and C) Errors in the ITC data bar graphs represent 68.3% CI (±1 SD) of the fit of the data. (D) FRET experiments to detect SKI binding (10 nM of enzyme in all experiments. Data (n = 2 to 5, mean ± SDM) have been fitted to quadratic binding equation. Unlike imatinib, SKI binds tighter to AblKD- asciminib and Abl64– 510 than to AblKD. In AblKD, asciminib exhibits small positive cooperativity with the binding of SKI. Kd errors are SE of the fit. conformation induced by asciminib. Unexpectedly, we did not find a difference between the binding affinities of SKI to Abl64– 510 and Abl64– 510- asciminib. We interpret this result as evidence that the conformational equilibrium of apo Abl is already far shifted to the closed conformation. Hence, the binding of asciminib had no effect on this equilibrium. This is, in fact, in agreement with a NMR study by Grzesiek and colleagues reporting overlapping chemical shifts between apo and GNF- 5 (a predecessor of asciminib) bound Abl for open/closed equilibrium markers (34). Effect of Orthosteric and Allosteric Modulators on Abl Activity. Interestingly, it had been reported that allosteric inhibitors of Abl other than asciminib (such as GNF- 2, GNF- 5, myristate, and myristoyl- peptide) actually do not inhibit the catalytic activity despite binding to AblKD (40, 43, 44). However, with ITC, we observed that asciminib shifts the conformation of AblKD to the “closing- competent” conformation (Fig.  3 B and C). Is this “closing- competent” conformation of the kinase domain a catalytically inactive state of Abl? Inhibition curves of AblKD generated using a coupled- enzyme assay with Srctide as substrate and asciminib as an inhibitor reveal 30% inhibition at saturating asciminib concentration (Fig.  4A). We conclude that the closing- competent conformation of the kinase domain is indeed catalytically inactive and that asciminib shifts the conformational equilibrium of AblKD to be 30% in this conformation by binding to the C- lobe and allosteric propagation to the orthosteric site. This model also reconciles the moderate synergistic effect of SKI and asciminib binding to AblKD (Fig. 3D). We note that this unique allosteric propagation by asciminib could contribute to its increased potency relative to other myristate pocket binders. Most importantly, and stressing the importance of studying full- length kinases in drug development, asciminib causes a 93% inhibition of Abl64– 510 at saturating concentration (Fig.  4A). This vastly increased inhibition is caused by the closing of the regulatory domains leading to an inactive kinase. Next, we quantified the inhibition of AblKD and Abl64– 510 by the two orthosteric inhibitors imatinib and SKI. In agreement with our affinity measurements (Fig. 3 B–D), imatinib exhibited a lower IC50 for AblKD than for Abl64– 510, while SKI exhibited a higher IC50 for AblKD than for Abl64– 510. Second, preincubation of AblKD with asciminib increased the IC50 for imatinib. This negative cooperativity arises from binding preferences of imatinib and asciminib to opposite conformations. In contrast, double- drugging of AblKD with SKI and asciminib resulted in a reduced IC50 since both have a binding preference to the same, “closing competent” conformation, highlighting their positive cooperativity (Fig. 4A). We note that SKI’s IC50 is higher than imatinib’s IC50 with respect to AblKD, Abl64– 510, and AblKD- asciminib (Fig. 4A), whereas this trend is reversed in our binding experiments (Fig. 3 B–D). We ascribe this discrepancy to the presence of ATP in the coupled- enzyme assay: AMPPCP binds twofold tighter to AblKD than Abl64– 510 (SI Appendix, Fig. S11). Thus, under our assay condition, we reason that the ATP shifts the conformational equilibrium of PNAS  2023  Vol. 120  No. 34  e2304611120 https://doi.org/10.1073/pnas.2304611120   5 of 11 Fig. 4. Catalytic activities of Abl kinase under double- drugging conditions. (A) Inhibition curves of AblKD and Abl64– 510 with asciminib, imatinib, and SKI. IC50 values shift according to the favored binding conformations of the corresponding inhibitor. (B and C) In vitro synergy studies using (B) imatinib and asciminib (C) or SKI to examine cooperativity between both inhibitors for Abl64– 510 activity. (C, Left) Numbers in the grid represent kobs with respective concentrations of inhibitor combinations. (C, Middle and Right) graphic representation of the data to illustrate inhibitor concentration needed to achieve 10% residual kinase activity [(Imatinib)10% act. and (SKI)10% act, dashed line]. Note that SKI required for 10% residual kinase activity decreases with increasing asciminib, especially with higher fold- change than that observed for imatinib. All assays were measured (n = 4 for 0 nM orthosteric inhibitor, n = 2 for all other assays, mean ± SDM) under kcat/Km condition with 2 mM Srctide. Errors in IC50 are SE of the fit. Errors in the bar graphs were determined by jackknifing the inhibition curve data. Abl64– 510 to the open state, which is favored by imatinib over SKI binding. Inhibition of Abl Kinase Activity under Double- Drugging Condition. The key question for clinical application is: What is the effect of different dosing concentration combinations of the two inhibitors on Abl’s kinase activity? Therefore, we performed synergy studies on Abl64– 510 kinase activity varying the concentration of both orthosteric and allosteric inhibitors (Fig. 4 B and C). These experiments underscore the negative cooperativity between imatinib and asciminib and corroborate the positive cooperativity between SKI and asciminib. First, we find a more pronounced inhibition of Abl64– 510 by SKI than by imatinib in the presence of asciminib. On the other hand, when used as a single agent, imatinib inhibits Abl64– 510 stronger than SKI, highlighting the difference in cooperativity. Second, we observe that in the presence of asciminib, less SKI is required for 90% inhibition of Abl activity compared to imatinib due to the positive cooperativity between SKI and asciminib (Fig.  4 B and C). X- Ray Crystal Structure of the Ternary Complex of Abl64– 510- SKI- Asciminib. Intrigued by the synergistic effect of SKI and asciminib on Abl activity, we structurally characterized this ternary complex by cocrystallization, resulting in a 2.86 Å crystal structure of Abl64– 510- SKI- asciminib [inactive, DFGin, BLBplus (29, 30)] (Fig. 5A and SI Appendix, Fig. S12 and Table S1). Surprisingly, this Abl structure adopts a closed conformation with striking differences to previously reported closed structures; Abl in complex with nilotinib and asciminib (PDB- ID: 5MO4), as well as in complex with PD166326 and myristic acid, a groundbreaking structure of full- length Abl in the inhibited state (PDB- ID: 1OPK) (Fig. 5B and SI Appendix, Fig. S13) (12, 35). First, we note that the entire N- terminal lobe is ~30° twisted only for Abl64– 510- SKI- asciminib, when aligned by the 6 of 11   https://doi.org/10.1073/pnas.2304611120 pnas.org regulatory domains (Fig. 5B and SI Appendix, Fig. S13). Second, the α- C helix is adopting the “out” position resulting from this N- lobe twist, since an α- C helix “in position” would clash with strands β4 and β5 (Fig. 5 B and C). In consequence, the canonical salt bridge between K290 and E305, a hallmark of an active kinase, is broken in our structure, whereas D400 (DFG- motif) is positioned in the “in” position. Paradoxically, the two other closed ternary complexes of Abl possess an α- C helix located in the “in” position and an established canonical salt bridge (K290- E305), both reminiscent of an active kinase conformation (SI Appendix, Fig. S15), while their DFG- motif is in the “out” position. This highlights i) the importance of the α- C helix conformation and the canonical salt bridge, and not only the DFG- motif, in determining open/closed conformation which is directly correlated to active/inactive states in full- length kinases, and ii) that through binding of SKI and asciminib, we were able to capture the strictly closed and inactive conformation of Abl with regulatory domains. We note, that the orthosteric site is fully occupied by SKI and the twisted N- lobe aids in forming this tightly packed binding pocket (Fig. 5C and SI Appendix, Fig. S13). In fact, K290 located on β3 strand is wrapping over SKI burying the inhibitor in Abl’s orthosteric site. Besides extensive van der Waals interactions between SKI and Abl, the quinazoline ring of SKI shares two hydrogen bonds with Abl, one between the side chain hydroxyl of T334 on β- strand 5 and N- 2 of SKI as well as between the amide of M337 and N- 0 of SKI (Fig. 5C). When compared to other closed Abl structures, we find an extended domain interface between SH3, linker, and N- lobe of the kinase domain, which explains the positive cooperativity Fig. 5. X- ray structure of ternary Abl64– 510- SKI- asciminib complex reveals a fully closed conformation compared to previous “energetically frustrated” ternary closed Abl structures. (A) Abl64– 510 bound to SKI and asciminib. (B) Superposition of Abl64– 510- SKI- asciminib (blue) and Abl- nilotinib- asciminib (pink, PDB- ID: 5MO4) (12). When superimposed by the regulatory SH2 and SH3 domains, the N- lobe of Abl64– 510- SKI- asciminib twists and exhibits α- C helix “out” position. (C) Zoom into the SKI and nilotinib binding sites. Van der Waals radii for the interacting Abl residues (spheres) with SKI (orange) and nilotinib (green) show more confined binding pocket for SKI than nilotinib. (D) Comparison of interface residues between N- SH3 domain (S94, D96, T98), linker (V247, S248), and N- lobe of kinase domain (W280, K282, Y283, S284, and L285) for Abl64– 510- SKI- asciminib (blue), Abl- nilotinib- asciminib (pink, PDB- ID: 5MO4), and Abl- PD166326- myristate (yellow, PDB- ID: 1OPK) (12, 35). Due to the twist in the N- lobe for Abl64– 510- SKI- asciminib, residues in the domain/domain interface exhibit better packing. For Abl64– 510- SKI- asciminib, an additional hydrogen bond is established between S248 and D96 which contributes to this extended interface. Oxygen and nitrogen atoms are colored in red and blue, respectively. Carbon atoms are colored according to their respective protein cartoon. PNAS  2023  Vol. 120  No. 34  e2304611120 https://doi.org/10.1073/pnas.2304611120   7 of 11 between SKI and asciminib. This improved interface is a direct result of the twisted N- lobe. The repositioned β2 and β3 strands cause Y283 to be completely buried within this interface. In other ternary complexes of Abl, this interface is only partially formed (Fig. 5D). Moreover, S248 (located on the linker) forms a hydro- gen bond with D96 in the SH3 domain, which is only present in the ternary complex of Abl64– 510- SKI- asciminib. Strikingly, S248P was identified as a resistant mutation for GNF- 2 and asciminib in cell culture- based screening (7, 45) with no mechanistic under- standing, given that this mutation is far away from the allosteric inhibitor binding site. Our structure now reveals the important role of S248 in allosteric closing of Abl, and hence asciminib inhibition! We conclude that our complex of Abl64– 510- SKI- asciminib represents the only example of a fully closed and inactive Abl structure. Discussion To combat on- target cancer drug resistance, double- drugging holds promise to be a powerful strategy. The rationale behind is multiplication of individual resistance mutational probabilities for each drug. Impressively, combinations of asciminib and various orthosteric inhibitors, including imatinib, indeed abolish the emergence of Abl resistance mutants and this double- drugging of Abl is currently in clinical trials (12). Given these groundbreaking clinical results, we used Abl kinase to interrogate the biophysical mechanism underlying this drug combination to learn a quanti- tative biophysical framework for successful double- drugging. In a second step, we used our knowledge of conformational equilibria in kinases to rationally select alternative orthosteric drugs exhib- iting improved synergy with the allosteric drug. Our results have major implications: i) Knowledge of conformational equilibria in drug targets indeed enables rational selection of inhibitor combi- nations with positive cooperativity and therefore better synergy. ii) Our Abl structure solves the apparent mystery of all previous closed Abl structures with the α- C helix in the active “in position” that contradicted the common features of inactive- closed kinases with the α- C helix in the canonical “out position”. Structural investigation of our double- drugged ternary complex with SKI and asciminib reveals a true α- C helix “out” state observed in Abl structures (SI Appendix, Fig. S14). This originates from an SKI- induced twist in the N- lobe causing a fully closed conformation, thus, releasing an energetically frustrated conformation observed in other double- drugged Abl complexes, since SKI and asciminib both preferentially bind to the closed state of Abl to cause positive cooperativity. In contrast, double- drugging with an open conforma- tion binder (nilotinib) and a closed conformation binder ( asciminib) results in an energetically frustrated Abl structure (PDB- ID: 5MO4). This structural study highlights how understanding of conforma- tional equilibria crucially aids the discovery of further inhibited states. Our finding of the energetically frustrated conformation agrees with previous NMR experiments reporting an opposing bind- ing preference of imatinib and GNF- 5 for Abl (34). In contrast, Johnson et al. claimed that such an antagonism arises from mutually exclusive binding of orthosteric and allosteric inhibitors (38). This conclusion contradicts previous studies characterizing Abl- imatinib- GNF- 5 by NMR as well as crystallographic studies on the ternary complex of Abl bound to both nilotinib and asciminib (12, 34). Our ITC studies resolve this controversy by ruling out mutual exclusivity for binding of imatinib and asciminib. iii) Our Abl data solve a heated debate: Recently, Kalodimos and colleagues argued that imatinib opens Abl via binding to its allosteric site (reported Kd >10 µM), and not via binding to its active site (46). This is in disagreement with NMR and cellular studies by Grzesiek et al. (34, 47, 48). Tighter binding of imatinib to open AblKD (Kd = 15 nM) compared to closed Abl64– 510 (Kd = 72.4 nM) and negative cooperativity between imatinib and asciminib buttress Grzesiek’s model where imatinib’s preferential binding to the open conformation of Abl arises from its orthosteric site binding with nanomolar affinity. Double- drugging has been applied to two additional targets, SHP2 phosphatase (16) and EGFR kinase (15, 49). Fodor et al. used a combination of two allosteric binders, SHP099 and SHP504, to inhibit the phosphatase SHP2 (16).The authors demonstrate that the combination reduces the dosage require- ments of these allosteric inhibitors to achieve effective inhibition of SHP2; however, SHP504 is a very weak binder with an IC50 of 21 μM (16). For EGFR kinase, a combination of the inhibitor JBJ- 04- 125- 02 binding right next to the irreversible orthosteric inhibitor osimertinib has been found to be more efficacious, than single agents, for inhibiting tumor growth in a mouse model. Furthermore, Jänne and colleagues demonstrated that this double- drugging resulted in the reduced emergence of resistance mutants in cellular assays (49). Here, the allosteric inhibitor bind- ing site is in immediate proximity to the orthosteric site, resulting in direct interactions between the two inhibitors potentially driv- ing positive cooperativity (15, 49). In contrast, we investigated the mechanism of dual inhibition in AurA and Abl kinase targeting a distant allosteric site that is involved in natural regulation, in combination with active site drugs. Rationally targeting those natural allosteric sites has the advantage that it assures allosteric coupling to activity. We demon- strate with our amateur attempts on both kinases that rational selection of double- drug combinations with positive cooperativity, and hence increased synergy, is possible based on knowledge of involved conformational equilibria. Furthermore, we note that such kinase activity- based synergy studies could easily be per- formed in a high- throughput manner to test orthosteric and allosteric inhibitor combinations. In summary, this work proposes a biophysical framework for designing and evaluating double- drugging synergy utilizing ortho- steric and allosteric modulators. As highlighted here, positive coop- erativity is desirable for double- drugging approaches improving selectivity and dosage requirements. However, while extreme negative cooperativity is undesirable, the clinical success of Novartis’ drug combination for Abl (12) with fourfold negative cooperativity as measured here suggests a clinical efficacy window ranging from small negative to strong positive cooperativity, given single- drug efficacy. Single drug efficacy is crucial, as otherwise a single resistance muta- tion abolishing binding of one drug would render the dual treatment to combat drug resistance essentially ineffective. Methods Cloning and Purification of Aurora A and Monobodies. AurA (residues 122 to 403, TEV- cleavable, N- terminal His6- tagged, kanamycin- resistance) in pET28a and LPP (#79748) from Addgene were cotransformed in BL21(DE3) cells and plated on Kan/Spec LB plate. Expression cultures were grown in TB to OD = 0.6–0.8 and induced with 0.6 mM IPTG for 16 h at 21 °C. Harvested cells were resuspended in 50 mM Tris–HCl, 300 mM NaCl, 20 mM MgCl2, and 10% glycerol, pH 8.0, and sonicated in the presence of EDTA- free protease inhibitor cocktail, lysozyme and DNAse. Clarified lysate was purified via Ni- NTA columns. AurA was eluted in 100% of 50 mM Tris–HCl, 300 mM NaCl, 500 mM imidazole, 20 mM MgCl2, and 10% glycerol, pH 8.0, which was combined with TEV and GST- LPP, and then dialyzed overnight against 50 mM Tris–HCl, 300 mM NaCl, 1 mM MnCl2, 5 mM TCEP, and 10% glycerol, pH 7.5 at 4 °C. Cleaved Aurora A was purified with Ni- NTA and GST columns and subsequently polished with a 26/600 S200 pg gel filtration column equilibrated in 20 mM Tris–HCl, 200 mM NaCl, 20 mM MgCl2, 5 mM TCEP, and 8 of 11   https://doi.org/10.1073/pnas.2304611120 pnas.org 10% glycerol, pH 7.5. Pure fractions were pooled and concentrated to around 40  μM, and stored in −80 °C. Monobodies (TEV- cleavable, N- terminal His6- tagged) were purified with on- column refolding as described in Zorba et al. (25). Cloning and Purification of AblKD and Abl64– 510. AblKD (residues 229 to 510, TEV- cleavable, N- terminal MBP- His6- tagged) and Abl64– 510 (residues 64 to 510, TEV- cleavable, N- terminal MBP- His6- tagged) were cloned into pETm41 (GenScript) (SI Appendix, Fig.  S7). Residue numbering follows Abl1b isoform that naturally consists of N- myristoylation. All Abl constructs were cotransformed with phosphatase YOPH (streptomycin- resistance) in BL21(DE3) cells and plated on Kan/Strep LB plate. Expression was performed in TB media and induced at OD = 0.6–0.8 with 0.1 mM (for AblKD) or 0.2 mM IPTG (for Abl64– 510) for 16 to 20 h at 18 °C. Harvested cells were resuspended in 50 mM Tris–HCl, 500 mM NaCl, and 1 mM TCEP, pH 8.0 (buffer A). Cells were sonicated in the presence of EDTA- free protease inhibitor cocktail, lysozyme, and DNAse. Clarified lysate was with Ni- NTA columns. The protein was eluted with 100% 50 mM Tris–HCl, 500 mM NaCl, 500 mM imidazole, and 1 mM TCEP, pH 8.0 and combined with TEV and CIP (#M0525, NEB) and dialyzed overnight against buffer A at 4 °C. Cleaved Abl was further purified with Ni- NTA and a Q column (gradient elution with 50 mM Tris–HCl, 1 M NaCl, and 1 mM TCEP, pH 8.0). Prior to anion exchange chroma- tography Abl was dialyzed into 50 mM Tris–HCl, 1 mM TCEP, and 10% glycerol, pH 8.0. Dephosphorylated Abl fractions were polished with 26/600 S75 pg (for AblKD) or 26/600 S200 pg (for Abl64– 510) gel filtration column with buffer A. Pure fractions were aliquoted to around 40 μM and stored in −80 °C. ITC. All titrations were carried out using Nano ITC (TA Instruments) and analyzed via the NanoAnalyze software either using the independent fit model or compet- itive replacement model. The first injection of each experiment was discarded according to the software manual. For AurA, danusertib (Selleckchem #S1107) was reconstituted to 100 mM in 100% DMSO and was diluted to appropriate concentration to match final 5% DMSO (vol/vol) for each experiment. An ADP- analogue, AMPCP, was used for competitive replacement experiments to measure and fitting of the binding of danusertib to AurA. All proteins were dialyzed in 20 mM Tris–HCl, 200 mM NaCl, 10% (vol/vol) glycerol, and 5 mM TCEP, pH 7.5. AMPCP was resuspended with the same buffer and was matched to pH 7.5. DMSO was added prior to each experiment to match 5% between titrant and titrand. Each injection was added in 2 µL increments with 180 s interval at a constant stirring speed of 300 rpm and at 25 °C. Concentrations used for the experiments are noted in SI Appendix. For Abl, N- Myr peptide (Myr- GQQPGKVLGDQR), ordered from GenScript, was used for competitive replacement experiments to measure and fitting of the binding of asciminib to Abl. All proteins were dialyzed in 50 mM Tris–HCl, 500 mM NaCl, and 1 mM TCEP, pH 8.0. Imatinib- mesylate (Sigma #SML- 1027) and asci- minib (MedKoo #206490) were reconstituted to 10 mM in 100% DMSO and were diluted to appropriate concentration for each experiment. N- Myr peptide was resuspended with the same buffer and was matched to pH 8.0. DMSO was added prior to each experiment to match 5% between titrant and titrand. Each injection was added in 1 to 1.5 µL increments with 180 s interval at a constant stirring speed of 300 rpm and at 25 °C. Concentrations used for the experiments are noted in SI Appendix. In Vitro Kinase Assay. To measure the IC50 of danusertib to AurA, ADP- GloTM Max assay (Promega #V7001) was used. 20 nM AurA in the absence or presence of either saturating concentration of Mb1 or Mb2 or Mb3 was incubated with 3 mM Lats2 (ATLARRDSLQKPGLE), 0.6 mg/mL BSA, and varying concentrations of danusertib with final 5% (vol/vol) of DMSO at 25 °C in 20 mM Tris–HCl, 200 mM NaCl, 10% (vol/vol) glycerol, and 5 mM TCEP, pH 7.50. The bolded and underlined residue indicates site of phosphorylation. The reaction was initiated by adding 5 mM ATP, and the final samples were collected after 2 h for AurA- Mb1 complex, 10 h for apo AurA, and 20 h for AurA- Mb2 and AurA- Mb3 complexes. The amount of ADP in the samples was measured by following the manufacturer’s protocol and used to calculate the observed rate. Assays for Abl were performed at 25 °C with half- well 96- well plate (Corning #3994) in 50 mM Tris–HCl, 500 mM NaCl, and 1 mM TCEP, pH 8.0, supplemented with 20 nM Abl kinase (AblKD or Abl64– 510), 2 mM Srctide (EIYGEFKK), 0.6 mg/mL BSA, 20 mM MgCl2, 750 µM NADH, 6 mM PEP, and 2.5 units of PK/LDH (Sigma #P0294). The bolded and underlined residue indicates site of phosphorylation. Oxidation of NADH at A340 was monitored using SpectraMAX by starting the assay with 1 mM ATP. The final volume of the assay was 100 µL. The observed rate (kobs) was calculated following Zorba et al. (25). All data were processed using GraphPad Prism and fitted to a four- parameter dose- response model. Molecular Dynamics Simulation. All- atom molecular dynamics simulations were conducted using OpenMM 7.6 (50) and “Making it rain” cloud- based notebook environment (51). The structure of AurA- danusertib- Mb1 was used as an initial model. To mimic danusertib binding to AurA- Mb1 under ITC con- ditions, we created such structure via removal of danusertib from our ternary complex AurA- danusertib- Mb1 [since the published AurA- Mb1 structure (PDB- ID: 5G15) has AMPPCP bound to active site (25). Parameterization for all MD runs was conducted using LEaP (52) with Amber ff14SB force field (53), GAFF2 (54) for ligand, and TIP3P (55, 56) water model. The systems were neutralized with NaCl at 0.2 mM, following the ITC conditions, and box size was set at 20 Å. AurA- Mb1 and AurA- danusertib- Mb1 structures were equilibrated to 298 K via Langevin dynamics (57) and 1 bar via Monte Carlo barostat (58) with 2 fs integration time. We set 10,000 steps of energy minimization with 1,000 kJ/mol of harmonic position restraints. The systems were equilibrated for 0.2 ns and 1 ns for AurA- Mb1 and AurA- danusertib- Mb1, respectively, in the NVT ensemble. Then, with accordingly equilibrated systems, triplicates of 10 ns production runs were done in the NPT ensemble. Trajectories were analyzed using VMD 1.9.4a53 (59). FRET Measurements. FluoroMax- 4 (Horiba Scientific) with temperature con- troller (water bath) was used to measure FRET between intrinsic tryptophan fluorescence and SKI. Either 10 nM Abl or 10 nM Abl + 200 nM asciminib was preincubated with varying concentrations of SKI for 40 min at 25 °C before meas- urements. An increase in the fluorescence was measured when the complex, specifically tryptophan, was excited at 295 nm to emit at 340 nm, which then excites SKI to emit at 460 nm (SI Appendix, Fig. S9). Both 5 nm of excitation and emission slit width were used. Control experiments (buffer- only, protein- only, and inhibitor- only) were confirmed that the increase of fluorescence is caused by the fluorescence energy transfer. The fluorescence intensity at 460 nm versus SKI concentration was fitted to the quadratic equation below in GraphPad Prism to obtain apparent Kd. + ( I [ ] Et] [ + K d) − F = F0 + A √( I + + K d)2 − 4 [Et] [I] Et] [ [ ] 2[Et] We simulated curves with tighter Kd for comparison to ensure that the fitted curves are not step functions due to the high enzyme concentration (SI Appendix, Fig. S10). Crystallographic Methods. Crystals of AurA in complex with Mb1 and danus- ertib were obtained by combining 2 µL of 300 µM (10 mg/mL) AurA + 315 µM (4 mg/mL) Mb1 + 2 mM AMPPCP + 4 mM MgCl2 with 2 µL reservoir of 0.1 M MES pH 6.5 + 0.2 M ammonium sulfate + 4% (v/v) 1,3- propanediol + 15 to 18% PEG 8,000. Streak seeding was used to obtain bigger crystals. Crystals were grown at 18 °C by hanging drop. The crystals were transferred to a drop of fresh reservoir for 30 s to remove excess nucleotides from the crystal sur- face. Then, the crystals were transferred to a drop with reservoir with 1 mM danusertib for 16 h of soaking. For cryoprotection, the crystals were transferred into 17.5% PEG 400, 17.5% ethylene glycol, 15% reservoir, and 50% water for a few seconds. Crystals of AurA in complex with Mb2 and danusertib were obtained by com- bining 0.5 µL of 300 µM (10 mg/mL) AurA + 315 µM (4 mg/mL) Mb2 + 1 mM danusertib with 0.5 µL of 0.1 M BIS–TRIS pH 5.5 + 0.2 M Ammonium acetate + 25% PEG3350. Crystals were grown at 18 °C by sitting drop. Crystals were harvested and subsequently flash frozen. Diffraction data for AurA- danusertib- Mb1 and AurA- danusertib- Mb2 were collected at 100 K Advanced Light Source (Lawrence Berkeley National Laboratory) at beamlines BL821 and BL501, respectively, and were integrated with XIA2 (60) or XDS (61). Data were scaled and merged with AIMLESS (62). Initial phases were obtained with molec- ular replacement programs MOLREP (63) and PHASER (64) by using AurA + Mb1 PNAS  2023  Vol. 120  No. 34  e2304611120 https://doi.org/10.1073/pnas.2304611120   9 of 11 + AMPPCP (PDB- ID: 5G15) for AurA- danusertib- Mb1 structure and AurA + AMPPCP (PDB- ID: 4C3R) and HA4Mb (PDB- ID: 3K2M) for AurA- danusertib- Mb2 structure using two molecules each in the asymmetric unit. The structures were iteratively refined using refmac and phenix.refine (Version1.19.1) (65) followed by manual model building in COOT (66). Models were validated with MolProbity (67). Molecular structures were represented and rendered with ChimeraX (68, 69). Crystals of Abl64– 510 in complex with SKI and asciminib were obtained by combin- ing 0.3 µL of 600 µM Abl64– 510 + 700 µM SKI + 700 µM asciminib (~32 mg/mL) in 5% DMSO with 0.4 µL reservoir of 0.1 M Tris–HCl pH 8 + 1.75 M Ammonium sulfate + 2% (v/v) polypropylene glycol 400 (PPG 400). The final stock of complex was con- centrated from 1 µM Abl64– 510 with ~1.2 µM SKI/asciminib after incubation at 4 °C for 6 h. Screening around this condition yielded crystals in a transparent diamond- shaped or plate- shaped crystals. Crystals were grown at 18 °C by sitting drop for a few days. The crystals were transferred to a drop of fresh reservoir containing 20% xylitol with matching concentration of inhibitors in 5% DMSO for few seconds for cryoprotection. Single crystal X- ray diffraction data were collected at 100 K at Advanced Light Source Berkeley (BL201). Data were integrated with XDS (61) as well as scaled and merged with AIMLESS (62). Analysis of processed data with phenix.xtriage (70) found outliers in the dataset and further revealed substantial translational noncrystallographic symmetry with a Patterson peak of 56.63% height relative to origin, complicating refinement. Initial phases were obtained by molecular replacement (PHASER) (64) using Abl- nilotinib- asciminib (PDB- ID: 5MO4) as a search model with two molecules in the asymmetric unit. The kinase domain, SH3, and SH2 (regulatory domains) were individually placed during molecular replacement. Refinement and manual model building were performed by phenix.refine (version 1.19.1) and Coot, respectively (65, 66). Models were validated with MolProbity (67). Molecular structures were represented and rendered with ChimeraX (68, 69) and PyMol (71). Data, Materials, and Software Availability. Structure factors and refined coordinates obtained from X- ray crystallography have been deposited into the Protein Data Bank (www.wwpdb.org) under PDB accession codes: 8SSP (72) (AurA- danusertib- Mb1), 8SSO (73)  (AurA- danusertib- Mb2), and 8SSN (74) (Abl64– 510-SKI- asciminib). ACKNOWLEDGMENTS. D.K. is supported by the Howard Hughes Medical Institute (HHMI). The Berkeley Center for Structural Biology is supported by the HHMI, Participating Research Team members, and the NIH, National Institute of General Medical Sciences, ALS- ENABLE grant P30 GM124169. The Advanced Light Source is a Department of Energy Office of Science User Facility under Contract No. DE- AC02- 05CH11231. The Pilatus detector on beamline 2.0.1 was funded under NIH grant S10OD021832. The Pilatus detector on beamline 5.0.1 was funded under NIH grant S10OD026941. Author affiliations: aDepartment of Biochemistry, Brandeis University, Waltham, MA 02454; and bHHMI, Brandeis University, Waltham, MA 02454 Author contributions: C.K., A.H., V.N., and D.K. designed research; C.K., H.L., A.H., S.K., and V.N. performed research; C.K., H.L., A.H., S.K., V.N., and D.K. analyzed data; and C.K., H.L., and D.K. wrote the paper. Competing interest statement: D.K. is co- founder of Relay Therapeutics and MOMA Therapeutics. The remaining authors declare no competing interests. 1. 2. 3. 4. 5. J.- P. Changeux, Allostery and the monod- wyman- changeux model after 50 years. Annu. Rev. Biophys. 41, 103–133 (2012). S. K. Hanks, A. M. Quinn, T. Hunter, The protein kinase family: Conserved features and deduced phylogeny of the catalytic domains. Science 241, 42–52 (1988). S. A. Rosenzweig, Acquired resistance to drugs targeting tyrosine kinases. Adv. Cancer Res. 138, 71–98 (2018). S. Lu et al., Emergence of allosteric drug- resistance mutations: New challenges for allosteric drug discovery. Drug Discov. Today 25, 177–184 (2020). X. Lu, J. B. Smaill, K. Ding, New promise and opportunities for allosteric kinase inhibitors. Angew. Chem. Int. Ed. Engl. 59, 13764–13776 (2020). 25. A. Zorba et al., Allosteric modulation of a human protein kinase with monobodies. Proc. Natl. Acad. Sci. U.S.A. 116, 13937–13942 (2019). 26. R. Bayliss, T. Sardon, I. Vernos, E. Conti, Structural basis of Aurora- A activation by TPX2 at the mitotic spindle. Mol. Cell 12, 851–862 (2003). 27. D. Fancelli et al., 1, 4, 5, 6- tetrahydropyrrolo [3, 4- c] pyrazoles: Identification of a potent Aurora kinase inhibitor with a favorable antitumor kinase inhibition profile. J. Med. Chem. 49, 7247–7251 (2006). 28. E. W. Lake et al., Quantitative conformational profiling of kinase inhibitors reveals origins of selectivity for Aurora kinase activation states. Proc. Natl. Acad. Sci. U.S.A. 115, E11894–E11903 (2018). 6. H. Mobitz, W. Jahnke, S. W. Cowan- Jacob, Expanding the opportunities for modulating kinase 29. V. Modi, R. L. Dunbrack Jr., Defining a new nomenclature for the structures of active and inactive 7. targets with allosteric approaches. Curr. Topics Med. Chem. 17, 59–70 (2017). J. Zhang et al., Targeting Bcr- Abl by combining allosteric with ATP- binding- site inhibitors. Nature 463, 501–506 (2010). kinases. Proc. Natl. Acad. Sci. U.S.A. 116, 6818–6827 (2019). 30. V. Modi, R. L. Dunbrack Jr., Kincore: A web resource for structural classification of protein kinases and their inhibitors. Nucleic Acids Res. 50, D654–D664 (2022). 8. D. Rea et al., A phase 3, open- label, randomized study of asciminib, a STAMP inhibitor, vs bosutinib 31. A. Zorba et al., Molecular mechanism of Aurora A kinase autophosphorylation and its allosteric 9. in CML after 2 or more prior TKIs. Blood 138, 2031–2041 (2021). T. P. Hughes et al., Asciminib in chronic myeloid leukemia after ABL kinase inhibitor failure. New Engl. J. Med. 381, 2315–2323 (2019). 10. J. Schoepfer et al., Discovery of Asciminib (ABL001), an Allosteric Inhibitor of the Tyrosine Kinase Activity of BCR- ABL1 (ACS Publications, 2018). 11. P. W. Manley, L. Barys, S. W. Cowan- Jacob, The specificity of asciminib, a potential treatment for chronic myeloid leukemia, as a myristate- pocket binding ABL inhibitor and analysis of its interactions with mutant forms of BCR- ABL1 kinase. Leukemia Res. 98, 106458 (2020). 12. A. A. Wylie et al., The allosteric inhibitor ABL001 enables dual targeting of BCR–ABL1. Nature 543, 733–737 (2017). 13. C. A. Eide et al., Combining the allosteric inhibitor asciminib with ponatinib suppresses emergence of and restores efficacy against highly resistant BCR- ABL1 mutants. Cancer Cell 36, 431–443.e435 (2019). activation by TPX2. Elife 3, e02667 (2014). 32. C. A. Dodson et al., Crystal structure of an Aurora- A mutant that mimics Aurora- B bound to MLN8054: Insights into selectivity and drug design. Biochem. J. 427, 19–28 (2010). 33. D. E. Anderson, W. J. Becktel, F. W. Dahlquist, pH- induced denaturation of proteins: A single salt bridge contributes 3–5 kcal/mol to the free energy of folding of T4 lysozyme. Biochemistry 29, 2403–2408 (1990). 34. L. Skora, J. Mestan, D. Fabbro, W. Jahnke, S. Grzesiek, NMR reveals the allosteric opening and closing of Abelson tyrosine kinase by ATP- site and myristoyl pocket inhibitors. Proc. Natl. Acad. Sci. U.S.A. 110, E4437–E4445 (2013). 35. B. Nagar et al., Structural basis for the autoinhibition of c- Abl tyrosine kinase. Cell 112, 859–871 (2003). 36. T. Saleh, P. Rossi, C. G. Kalodimos, Atomic view of the energy landscape in the allosteric regulation of Abl kinase. Nat. Struct. Mol. Biol. 24, 893–901 (2017). 14. K. V. Gleixner et al., Asciminib and ponatinib exert synergistic anti- neoplastic effects on CML cells 37. W. Jahnke et al., Binding or bending: Distinction of allosteric Abl kinase agonists from antagonists expressing BCR- ABL1T315I- compound mutations. Am. J. Cancer Res. 11, 4470 (2021). by an NMR- based conformational assay. J. Am. Chem. Soc. 132, 7043–7048 (2010). 15. T. S. Beyett et al., Molecular basis for cooperative binding and synergy of ATP- site and allosteric EGFR 38. T. K. Johnson et al., Synergy and antagonism between allosteric and active- site inhibitors of abl inhibitors. Nat. Commun. 13, 1–11 (2022). tyrosine kinase. Angew. Chem. 133, 20358–20361 (2021). 16. M. Fodor et al., Dual allosteric inhibition of SHP2 phosphatase. ACS Chem. Biol. 13, 647–656 39. T. Xie, T. Saleh, P. Rossi, C. G. Kalodimos, Conformational states dynamically populated by a kinase (2018). 17. J. Colicelli, ABL tyrosine kinases: Evolution of function, regulation, and specificity. Sci. Signaling 3, re6 (2010). 18. N. Ertych et al., Increased microtubule assembly rates influence chromosomal instability in colorectal cancer cells. Nat. Cell Biol. 16, 779–791 (2014). determine its function. Science 370, eabc2754 (2020). 40. O. Hantschel et al., A myristoyl/phosphotyrosine switch regulates c- Abl. Cell 112, 845–857 (2003). 41. G. Tian, M. Cory, A. A. Smith, W. B. Knight, Structural determinants for potent, selective dual site inhibition of human pp60 c- src by 4- anilinoquinazolines. Biochemistry 40, 7084–7091 (2001). 42. J. Bain et al., The selectivity of protein kinase inhibitors: A further update. Biochem. J. 408, 297–315 19. H. Zhou et al., Tumour amplified kinase STK15/BTAK induces centrosome amplification, aneuploidy (2007). and transformation. Nat. Genet. 20, 189–193 (1998). 43. Y. Choi et al., N- myristoylated c- Abl tyrosine kinase localizes to the endoplasmic reticulum upon 20. R. Du, C. Huang, K. Liu, X. Li, Z. Dong, Targeting AURKA in cancer: Molecular mechanisms and binding to an allosteric inhibitor. J. Biol. Chem. 284, 29005–29014 (2009). opportunities for cancer therapy. Mol. Cancer 20, 1–27 (2021). 44. D. Fabbro et al., Inhibitors of the Abl kinase directed at either the ATP- or myristate- binding site. 21. C. O. De Groot et al., A cell biologist’s field guide to aurora kinase inhibitors. Front. Oncol. 5, 285 Biochim. Biophys. Acta 1804, 454–462 (2010). (2015). 45. W. Qiang et al., Mechanisms of resistance to the BCR- ABL1 allosteric inhibitor asciminib. Leukemia 22. W. Pitsawong et al., Dynamics of human protein kinase Aurora A linked to drug selectivity. Elife 7, 31, 2844–2847 (2017). e36656 (2018). 46. T. Xie, T. Saleh, P. Rossi, D. Miller, C. G. Kalodimos, Imatinib can act as an Allosteric Activator of Abl 23. J. A. Gilburt et al., Dynamic equilibrium of the aurora A kinase activation loop revealed by single- Kinase. J. Mol. Biol. 434, 167349 (2022). molecule spectroscopy. Angew. Chem. 129, 11567–11572 (2017). 24. S. Cyphers, E. F. Ruff, J. M. Behr, J. D. Chodera, N. M. Levinson, A water- mediated allosteric network governs activation of Aurora kinase A. Nat. Chem. Biol. 13, 402 (2017). 47. R. Sonti, I. Hertel- Hering, A. J. Lamontanara, O. Hantschel, S. Grzesiek, ATP site ligands determine the assembly state of the Abelson kinase regulatory core via the activation loop conformation. J. Am. Chem. Soc. 140, 1863–1869 (2018). 10 of 11   https://doi.org/10.1073/pnas.2304611120 pnas.org 48. S. Grzesiek, J. Paladini, J. Habazettl, R. Sonti, Imatinib disassembles the regulatory core of Abelson kinase by binding to its ATP site and not by binding to its myristoyl pocket. Magn. Res. 3, 91–99 (2022). 49. C. To et al., Single and dual targeting of mutant EGFR with an allosteric inhibitor. Cancer Discov. 9, 926–943 (2019). 61. W. Kabsch, XDS. Acta Crystallogr. D Biol. Crystallogr. 66, 125–132 (2010). 62. P. R. Evans, G. N. Murshudov, How good are my data and what is the resolution? Acta Crystallogr. D Biol. Crystallogr. 69, 1204–1214 (2013). 63. A. Vagin, A. Teplyakov, MOLREP: An automated program for molecular replacement. J. Appl. 50. P. Eastman et al., OpenMM 7: Rapid development of high performance algorithms for molecular Crystallogr. 30, 1022–1025 (1997). dynamics. PLoS Comput. Biol. 13, e1005659 (2017). 51. P. R. Arantes, M. D. Polêto, C. Pedebos, R. Ligabue- Braun, Making it rain: Cloud- based molecular simulations for everyone. J. Chem. Inf. Model. 61, 4852–4856 (2021). 52. D. A. Case et al., Amber 2021 (University of California, San Francisco, 2021). 53. J. A. Maier et al., ff14SB: Improving the accuracy of protein side chain and backbone parameters 64. A. J. McCoy et al., Phaser crystallographic software. J. Appl. Crystallogr. 40, 658–674 (2007). 65. P. V. Afonine et al., Towards automated crystallographic structure refinement with phenix. refine. Acta Crystallogr. D Biol. Crystallogr. 68, 352–367 (2012). 66. P. Emsley, K. Cowtan, Coot: Model- building tools for molecular graphics. Acta Crystallogr. D Biol. Crystallogr. 60, 2126–2132 (2004). from ff99SB. J. Chem. Theory Comput. 11, 3696–3713 (2015). 67. C. J. Williams et al., MolProbity: More and better reference data for improved all- atom structure 54. X. He, V. H. Man, W. Yang, T.- S. Lee, J. Wang, A fast and high- quality charge model for the next validation. Protein Sci. 27, 293–315 (2018). generation general AMBER force field. J. Chem. Phys. 153, 114502 (2020). 68. E. F. Pettersen et al., UCSF ChimeraX: Structure visualization for researchers, educators, and 55. H. J. Berendsen, J. P. Postma, W. F. van Gunsteren, J. Hermans, “Interaction models for water in developers. Protein Sci. 30, 70–82 (2021). relation to protein hydration” in Intermolecular Forces, B. Pullman, Eds. (Springer, Dordrecht, 1981), https://doi.org/10.1007/978- 94- 015- 7658- 1_21. pp. 331–342. 69. T. D. Goddard et al., UCSF ChimeraX: Meeting modern challenges in visualization and analysis. Protein Sci. 27, 14–25 (2018). 56. W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, M. L. Klein, Comparison of simple 70. P. D. Adams et al., PHENIX: A comprehensive Python- based system for macromolecular structure potential functions for simulating liquid water. J. Chem. Phys. 79, 926–935 (1983). solution. Acta Crystallogr. D Biol. Crystallogr. 66, 213–221 (2010). 57. P. Turq, F. Lantelme, H. L. Friedman, Brownian dynamics: Its application to ionic solutions. J. Chem. 71. W. L. DeLano, Pymol: An open- source molecular graphics tool. CCP4 Newsl. Protein Crystallogr. 40, Phys. 66, 3039–3044 (1977). 82–92 (2002). 58. J. Åqvist, P. Wennerström, M. Nervall, S. Bjelic, B. O. Brandsdal, Molecular dynamics simulations of 72. C. Kim et al., AurA bound to danusertib and activating monobody Mb1. Protein Data Bank. https:// water and biomolecules with a Monte Carlo constant pressure algorithm. Chem. Phys. Lett. 384, 288–294 (2004). 59. W. Humphrey, A. Dalke, K. Schulten, VMD: Visual molecular dynamics. J. Mol. Graphics 14, 33–38 (1996). 60. G. Winter, xia2: An expert system for macromolecular crystallography data reduction. J. Appl. www.rcsb.org/structure/8SSP. Accessed 8 May 2023. 73. C. Kim et al., AurA bound to danusertib and inhibiting monobody Mb2. Protein Data Bank. https:// www.rcsb.org/structure/8SSO. Accessed 8 May 2023. 74. C. Kim et al., Abl kinase in complex with SKI and asciminib. Protein Data Bank. https://www.rcsb.org/ Crystallogr. 43, 186–190 (2010). structure/8SSN. Accessed 8 May 2023. PNAS  2023  Vol. 120  No. 34  e2304611120 https://doi.org/10.1073/pnas.2304611120   11 of 11
null
10.1001_jamanetworkopen.2019.12416.pdf
null
null
Original Investigation | Oncology Common Secondary Genomic Variants Associated With Advanced Epithelioid Hemangioendothelioma Nathan D. Seligson, PharmD; Achal Awasthi, MS; Sherri Z. Millis, PhD; Brian K. Turpin, DO; Christian F. Meyer, MD, PhD; Anne Grand'Maison, MD; David A. Liebner, MD; John L. Hays, MD, PhD; James L. Chen, MD Abstract IMPORTANCE Epithelioid hemangioendothelioma (EHE) is a rare, malignant vascular sarcoma characterized in most cases by a WWTR1-CAMTA1 fusion. The clinical course of EHE exhibits a dual nature. The condition is often indolent but can rapidly grow and metastasize unpredictably. No biomarkers to date are available to predict this phenotype. The hypothesis of this study was that better defining the genomic landscape of EHE using next-generation sequencing could offer additional therapies and insight into clinical outcomes. OBJECTIVE To characterize secondary EHE genomic alterations and their association with clinical outcomes. Key Points Question Can next-generation sequencing reveal rationale for the dichotomous biological activity of epithelioid hemangioendothelioma (EHE) while illuminating potentially actionable alterations? Findings In a cross-sectional study of next-generation sequencing results collected from 49 participants diagnosed with EHE, more than half of DESIGN, SETTING, AND PARTICIPANTS Multicenter, cross-sectional, retrospective study of next- patients with EHE profiled exhibited generation sequencing results collected from participants diagnosed with EHE. Data were abstracted pathogenic genomic variants in addition between May 1, 2013, and May 31, 2019. This analysis was conducted from January through June 2019. Summary genomic data were provided by commercial genomic testing companies. MAIN OUTCOMES AND MEASURES Presence or absence of secondary pathogenic genomic variants and their association with disease stage and clinical features. RESULTS A total of 49 participants with EHE were assessed for the presence or absence of secondary genomic variants. Of these, 32 (65.3%) were female; the mean (SD) age at diagnosis was 49.9 (18.3) years (range, 11-81 years). In all, 46 participants (93.9%) had confirmed WWTR1-CAMTA1 fusion; 26 participants (57.1%) exhibited a pathogenic genomic variant secondary to the WWTR1- CAMTA1 fusion; and 9 participants (18.4%) exhibited potentially targetable genomic variants. Commonly altered genes included CDKN2A/B, RB1, APC, and FANCA. Participants older than 45 years at diagnosis had an increased prevalence of secondary genomic variants that was not statistically significant (65.6% vs 38.5%; difference, 27.1%; 95% CI, −3.5% to 58.0%; P = .16) and were more likely to have a clinically targetable variant (28.1% vs 0%; difference, 28.1%; 95% CI, 11.2%-40.2%; P = .03). In 14 participants with clinical data available, those with stage III/IV EHE were more likely to exhibit a secondary pathogenic genomic variant (80% vs 0%; difference, 80%; 95% CI, 55.2%-100%; P = .006). Participants with stage III/IV EHE were diagnosed at an older age (mean [SD] age, 54.6 [14.1] years vs 31.7 [16.0] years; P = .05) and had elevated WWTR1-CAMTA1 fusion expression that was not statistically significant (mean [SD] expression, 677 [706] copies vs 231 [213] copies; P = .20). CONCLUSIONS AND RELEVANCE Although EHE exhibits few secondary genomic variants, presence of key secondary variants may be prognostic for aggressive EHE. Further research is (continued) to the WWTR1-CAMTA1 fusion, with 18.4% of participants exhibiting a potentially targetable variant. Participants with stage III/IV EHE were more likely to exhibit a secondary pathogenic variant. Meaning Next-generation sequencing may identify secondary genomic variants that are associated with EHE aggressiveness; additionally, these variants may represent potential therapeutic targets. + Supplemental content Author affiliations and article information are listed at the end of this article. Open Access. This is an open access article distributed under the terms of the CC-BY License. JAMA Network Open. 2019;2(10):e1912416. doi:10.1001/jamanetworkopen.2019.12416 (Reprinted) October 2, 2019 1/8 Downloaded from jamanetwork.com by guest on 12/03/2024 JAMA Network Open | Oncology Secondary Genomic Variants Associated With Epithelioid Hemangioendothelioma Abstract (continued) needed to confirm this finding and determine whether more intensive upfront treatment is necessary for these patients. JAMA Network Open. 2019;2(10):e1912416. doi:10.1001/jamanetworkopen.2019.12416 Introduction Epithelioid hemangioendothelioma (EHE) is a rare vascular sarcoma with a prevalence of approximately 1 per 1 000 000 persons.1 A hallmark molecular characteristic of EHE is the fusion of the WWTR1 and CAMTA1 genes, present in 90% of EHE cases and pathognomonic for disease.2-4 The clinical course of EHE may be either indolent (often locally limited) or aggressive (characterized by local invasiveness or metastasis); however, indolent disease can unpredictably become aggressive. Key molecular biomarkers indicative of EHE course have yet to be established5; however, mitotic count and tumor size have been associated with prognosis.6 While limited disease can be amenable to observation or local therapy, metastatic EHE is typically resistant to chemotherapy and carries a poor prognosis. Treatment for advanced-stage EHE is not well established.7 Pathway-specific targeted therapies hold some promise, but improved systemic therapies are still needed.8 Few reports describe the genomic landscape of EHE outside of the driver fusions with their clinical correlates and have described a mostly quiet genome.9 In this article, we present the largest assessment, to our knowledge, of the clinicogenomic landscape of WWTR1-CAMTA1 (WC) fusion–associated EHE. Methods Data were abstracted between May 1, 2013, and May 31, 2019. This analysis was conducted from January through June 2019. Summary genomic data was provided by commercial genomic testing companies. This study is reported in accordance with the Strengthening the Reporting of Genetic Association Studies (STREGA) reporting guideline.10 Retrospective Analysis Approval for the retrospective collection of genomic data from Foundation Medicine, including a waiver of informed consent and HIPAA waiver of authorization, was obtained from the Western Institutional Review Board. Participants diagnosed with EHE were identified from retrospective sarcoma studies at The Ohio State University James Comprehensive Cancer Center, Roswell Park Cancer Institute, Johns Hopkins Medical Center, and Cincinnati Children's Hospital Medical Center. Waiver of informed consent for the original studies was approved by local institutional review boards. Participant characteristics, tumor stage at time of biopsy, and genomic data were extracted for this study. All participants identified were included. Sample size was based on data available, and no sample size calculations were performed. Genomic Analysis Genomic profiling data were collected from 46 patients with EHE who underwent genomic sequencing by Foundation Medicine (FMI)11 and 3 who underwent genomic sequencing by OmniSeq.12 Participants’ WC fusion status was only confirmed for those profiled by FMI. The FMI FoundationOne Heme panel includes coverage of 426 fully sequenced genes, rearrangement of 32 genes, and fusions of 282 genes. The OmniSeq panel includes coverage of 26 fully sequenced genes, hot spots in 73 genes, copy number variants in 52 genes, and fusions of 23 genes. Full genomic coverage of both targeted next-generation platforms is outlined in eTable 1 in the Supplement. Pathogenicity of genomic variants for participants sequenced by OmniSeq was determined via the COSMIC database.13 JAMA Network Open. 2019;2(10):e1912416. doi:10.1001/jamanetworkopen.2019.12416 (Reprinted) October 2, 2019 2/8 Downloaded from jamanetwork.com by guest on 12/03/2024 JAMA Network Open | Oncology Secondary Genomic Variants Associated With Epithelioid Hemangioendothelioma Pathogenic variants and variants of unknown significance (VUS) were included in our analysis. Genomic variants identified apart from the WC fusion were considered secondary variants. Gene enrichment was performed using Superpaths14 (eTable 2 in the Supplement). Targetable variants were defined using OncoKB classification as previously described.15 Statistical Analysis All data were analyzed in R statistical software version 3.4.3 (R Project for Statistical Computing) or Prism analysis and graphing software version 8.0.0 (GraphPad). For continuous variables, t tests were used. For categorical variables, χ2 tests were used to generate P values and a test of proportions was used to generate 95% confidence intervals of proportion difference. Continuous data are presented as mean (SD) unless otherwise stated and 2-tailed P values (cid:2).05 were considered statistically significant. Results Patient Characteristics Of 49 participants with EHE analyzed (32 [65.3%] female; mean [SD] age at diagnosis, 49.9 [18.3] years [range, 11-81 years]), 46 (93.9%) had WC fusion confirmation. These participants were primarily female (29 patients [63.0%]), and the mean (SD) age at diagnosis was 50.2 (18.5) years (range, 11-81 years). Full demographic characteristics are available in the Table. Participants had a low tumor mutation burden (mean [SD] mutations per megabase, 1.1 [1.5]). Quantification of WC expression from available participants demonstrated a right-skewed, log-normal distribution (eFigure 1A and B in the Supplement). EHE and Secondary Alteration in Established Oncogenic Pathways In all, 21 participants with EHE (42.9%) exhibited a WC fusion as a sole pathogenic genomic variant. A single additional pathogenic variant was identified in 14 participants (28.6%), while 2 or more pathogenic variants were identified in an additional 14 participants. The most commonly identified secondary variants were seen in CDKN2A (6 pathogenic, 1 VUS), CDKN2B (4 pathogenic, 0 VUS), RB1 (2 pathogenic, 1 VUS), ATRX (2 pathogenic, 1 VUS), APC (2 pathogenic, 1 VUS), and FANCA (2 pathogenic, 0 VUS) (eTable 3 in the Supplement). Pathways identified as altered in EHE included cell Table. Demographic Characteristics Characteristic No. (%) WWTR1-CAMTA1 Fusion (n = 46) No WWTR1-CAMTA1 Fusion (n = 3) Total (N = 49) Age at diagnosis, mean (SD) [range], y 50.2 (18.5) [11-81] 45.3 (18.7) [25-62] 49.9 (18.3) [11-81] Sex Male Female Microsatellite status Microsatellite stable Not tested Tumor mutation burden, mean (SD), mutations per megabase Pathogenic genomic variants Data not available WWTR1-CAMTA1 only Additional variants, No. 1 2 ≥3 17 (37.0) 29 (63.0) 32 (70.0) 14 (30.0) 1.1 (1.5) 0 20 (43.5) 14 (30.4) 9 (19.5) 3 (6.6) 0 3 (100) 0 3 (100) ND 1 (33.3) 0 0 1 (33.3) 1 (33.3) 17 (34.7) 32 (65.3) 32 (65.3) 17 (34.7) 1.1 (1.5) 1 (2.0) 20 (40.8) 14 (28.6) 10 (20.4) 4 (8.2) Abbreviation: ND, no data. JAMA Network Open. 2019;2(10):e1912416. doi:10.1001/jamanetworkopen.2019.12416 (Reprinted) October 2, 2019 3/8 Downloaded from jamanetwork.com by guest on 12/03/2024 JAMA Network Open | Oncology Secondary Genomic Variants Associated With Epithelioid Hemangioendothelioma cycle regulation, growth signaling, epigenetic modulators, and DNA damage repair (Figure 1A). Twenty-six participants (57.1%) exhibited a pathogenic genomic variant secondary to the WC fusion, and 9 participants (18.4%) exhibited genomic variants in multiple pathways. Commonly altered genes included CDKN2A/B, RB1, APC, and FANCA. Sex did not segregate secondary genomic variant frequency (47% male vs 62% female; difference, 15%; 95% CI, 13.6%-44.5%; P = .32) (Figure 1B). Older Age and Increased Genomic Complexity Age at diagnosis demonstrated a bimodal distribution with a division at 45 years (log-likelihood 1-component model, −207.2 vs 2-component model, −197.6; difference, 9.6; 95% CI, 0.0-23.8; P = .02) (eFigure 2 and eMethods in the Supplement). Participants aged 45 years or older at diagnosis had a higher prevalence of pathogenic secondary genomic variants that was not statistically significant (65.6% vs 38.5%; difference, 27.1%; 95% CI, −3.5% to 58.0%; P = .16) (Figure 1C). Notably, variants in the most commonly altered gene in this data set, CDKN2A, were exclusively seen in participants aged 45 years or older at diagnosis (eTable 4 in the Supplement). A total of 19 targetable variants were identified by OncoKB (Figure 2A; eTable 5 in the Supplement). Pathogenic genomic variants identified to be targetable were seen in 9 participants (18.4%), with 5 participants (10.2%) harboring variants associated with US Food and Drug Administration–approved therapies. Participants aged 45 years or older at diagnosis were more likely to have a targetable pathogenic genomic variant (28.1% vs 0%; difference, 28.1%; 95% CI, 11.2%-40.2%; P = .03) (Figure 2B). Presence of Secondary Alterations and Advanced-Stage Disease To assess the clinicogenomic landscape of EHE, 14 participants with clinical data available were identified (4 with stage I/II, 10 with stage III/IV). Participants with stage III/IV EHE were significantly more likely to exhibit a pathogenic secondary genomic variant (80% vs 0%; difference, 80%; 95% CI, 55.2%-100%; P = .006) (Figure 3A and C). Additionally, those with stage III/IV EHE were older at diagnosis (mean [SD] age, 54.6 [14.1] years vs 31.7 [16.0] years; P = .05) (Figure 3B) and had greater WC fusion expression that was not statistically significant (mean [SD], 677 [706] vs 231 [213] copies; P = .20) (eFigure 1C in the Supplement). Figure 1. Genomic Landscape of Epithelioid Hemangioendothelioma A Genetic variants and pathways 57.1% Cell Cycle Regulation Growth Signaling Epigenetic Modulators DNA Damage Repair Other B Variants by sex y r a d n o c e S c i n e g o h t a P % , t n a i r a V c i m o n e G 100 80 60 40 20 0 P =.32 Participants C Variants by age y r a d n o c e S c i n e g o h t a P % , t n a i r a V c i m o n e G 100 80 60 40 20 0 P =.16 Female Male Sex <45 ≥45 Age at Diagnosis, y A, Heatmap of the presence (shaded) or absence (white) of known genomic variants based on their shared pathways. While a majority of epithelioid hemangioendothelioma tumors demonstrated a secondary genomic variant (57.1%), few tumors exhibited genomic variants in multiple pathway categories. B, Pathogenic secondary genomic variants were more common in female participants, but the difference was not statistically significant. C, Pathogenic secondary genomic variants were also more common in participants aged 45 years or older at diagnosis, but the difference was not statistically significant. JAMA Network Open. 2019;2(10):e1912416. doi:10.1001/jamanetworkopen.2019.12416 (Reprinted) October 2, 2019 4/8 Downloaded from jamanetwork.com by guest on 12/03/2024 JAMA Network Open | Oncology Secondary Genomic Variants Associated With Epithelioid Hemangioendothelioma Discussion Epithelioid hemangioendothelioma is characterized molecularly by its primary gene fusions, but owing to the rarity of this disease, little is known regarding the clinical significance of secondary genomic alterations. Here, we present the largest assessment, to our knowledge, of the genomic landscape of WC fusion–positive EHE. Of the 49 participants included in this study, 46 were positively identified to have the WC fusion with no additional fusions detected. An additional 3 participants with EHE were included after histopathological review. The less common EHE fusion, YAP1-TFE3,16 was not explicitly tested for by the OmniSeq panel. No participants tested with the FMI panel were identified to have concurrent WC and YAP1-TFE3 fusions. We have included this OmniSeq data as well, given the high likelihood of a WC fusion. Although half of all EHE tumors included in this study exhibited a pathogenic secondary genomic variant, it was rare for a tumor to have 2 or more secondary variants present. The identified variants are linked to well-studied oncogenic pathways. The most prevalent gene alteration identified in this study was deletion of the CDKN2A/B locus, corresponding to well-studied tumor suppressor genes responsible for regulation of the cell cycle and p53-mediated apoptosis. The data available here are unable to test the importance of CDKN2A/B loss in the natural history or development of EHE. Further study is necessary to identify the role of CDKN2A/B loss in EHE. In other sarcomas, including gastrointestinal stromal tumors, loss of CDKN2A expression is associated with poor prognosis and a greater potential for metastatic disease.17-19 The biological meaning of CDKN2A/B loss in EHE requires further elucidation. Approximately 20% of EHEs studied exhibited a clinically actionable secondary genomic alteration. Further assessment identified an enriched prevalence in participants aged 45 years or older at diagnosis. In our clinically enriched subset, stage III/IV EHE was strongly associated with the presence of pathogenic secondary genomic variants and older age. Importantly, this was true when either including or excluding participants lacking confirmation of the WC fusion. Taken together, these data suggest that the fusion event may represent the first step in the development of EHE with a secondary genomic change required for tumor aggressiveness. This has several potential practice implications: for one, participants with newly diagnosed EHE could be considered for genomic profiling to evaluate the presence or absence of secondary alterations; in addition, participants with EHE with secondary alterations may potentially be considered for more aggressive treatment. Prospective clinical trials will need to confirm this guidance. Figure 2. Targetable Genomic Variants in Epithelioid Hemangioendothelioma A Targetable genomic variants B Targetable variants by age CDKN2A ROS1 ERBB2 IDH1 TSC2 TSC1 BRCA1 BRCA2 NTRK1 PTCH1 Participants VUS Pathogenic variant % , t n a i r a V c i m o n e G e l b a t e g r a T 30 25 20 15 10 5 0 P =.03 Age at diagnosis <45 y ≥45 y P =.63 Pathogenic VUS Variant Type A, Heatmap of the presence of targetable pathogenic genomic variants (dark blue), targetable variants of unknown significance (VUS) (light blue), or the absence of targetable genomic variants (white) as defined by OncoKB. B, Participants aged 45 years or older at diagnosis were more likely to exhibit a targetable pathogenic genomic variant, but equally likely to demonstrate a targetable VUS compared with participants younger than 45 years at diagnosis. JAMA Network Open. 2019;2(10):e1912416. doi:10.1001/jamanetworkopen.2019.12416 (Reprinted) October 2, 2019 5/8 Downloaded from jamanetwork.com by guest on 12/03/2024 JAMA Network Open | Oncology Secondary Genomic Variants Associated With Epithelioid Hemangioendothelioma Limitations To our knowledge, this is the largest genomic assessment of EHE to date; however, limitations inherent to studies of extremely rare diseases apply here. As tumor sequencing is not a standard recommendation in the treatment of EHE, the data available may suffer from selection bias toward more aggressive EHE. Additionally, the 2 next-generation sequencing platforms used to test for genomic variants are targeted to a specific set of genes and vary significantly in their coverage. Whole-genome sequencing approaches may provide a more comprehensive assessment; however, the targeted panels used here provide strong, validated assessment of genes known to have biological and clinical associations with cancer. It is important to note that participants who underwent sequencing by FMI and OmniSeq exhibited similar trends in association between genomic variants and tumor stage. Secondary genomic variants are certainly important in EHE; however, the prevalence of these alterations may be lower in a prospectively curated data set. Additionally, this data set is limited temporally and is unable to differentiate between passenger and active genomic variants. Further longitudinal research is necessary to define the genomic progression of EHE. Figure 3. Prevalence of Pathogenic Secondary Genomic Variants A Variants by stage B Age at diagnosis and stage P =.006 I/II III/IV Stage y r a d n o c e S c i n e g o h t a P % , t n a i r a V c i m o n e G 100 80 60 40 20 0 Specific variants by stage C I I / I e g a t S V I / I I I e g a t S P =.05 Any variant Growth signaling Epigenetics Cell cycle DNA damage Other y , s i s o n g a i D t a e g A 80 60 40 20 0 I/II III/IV Stage VUS Pathogenic variant a a 0 a 5 10 15 Secondary Variants, No. 4 A C R A M S 1 T O P 3 L L M 8 9 P U N 2 K R R L 1 D R A B 4 M D M K S A P 2 T E T 1 C S T K C I 2 B B R E A C N A F B 2 N K D C 1 H C T O N 6 3 D C 0 0 3 P E 1 H L M N L E R P B B E R C 1 S O R 2 L L M M L B 1 H D I A I B K F N A 2 N K D C C P A 1 P I R B X 3 X D D M T A 1 H C T P 1 A N N T C 2 K 2 P A M 1 M R B P 3 0 7 F N Z C P A X R T A 1 H C T O N M T A 7 W X B F A 2 N K D C T R E T 3 A T A G 2 A C R B 0 9 P S H 2 M D M 1 B H P E 6 K 3 P A M 4 K 2 P A M 1 K 3 P A M 3 F C T 2 K E H C A T I I C N C L F 4 2 1 R P G 2 H C T O N 1 M D R P B 2 F E M C 6 4 M A F 1 H C W Y L F E 1 H 1 T S I H Variant A, Stage III/IV tumors were more likely to harbor a pathogenic secondary genomic variant. B, Stage III/IV tumors were associated with older age at diagnosis. Horizontal lines indicate group median. C, Heatmap of secondary genomic variants with total secondary variants noted on the rightmost y-axis. VUS indicates variant of unknown significance. a Three participants clinically diagnosed with stage III/IV epithelioid hemangioendothelioma with genomic profiling data available but lacking confirmation of a WWTR1-CAMTA1 fusion were included in a secondary clinical assessment. These participants exhibited similar characteristics to other participants previously described. JAMA Network Open. 2019;2(10):e1912416. doi:10.1001/jamanetworkopen.2019.12416 (Reprinted) October 2, 2019 6/8 Downloaded from jamanetwork.com by guest on 12/03/2024 JAMA Network Open | Oncology Secondary Genomic Variants Associated With Epithelioid Hemangioendothelioma Conclusions In this study, more than half of participants with EHE and WWTR1-CAMTA1 fusion exhibited a secondary genomic variant, with up to 20% that are potentially clinically actionable. Participants with advanced-stage EHE were significantly more likely to have secondary genomic variants. Prospective, multigroup clinical trials are necessary to confirm these findings and their clinical utility. ARTICLE INFORMATION Accepted for Publication: August 12, 2019. Published: October 2, 2019. doi:10.1001/jamanetworkopen.2019.12416 Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2019 Seligson ND et al. JAMA Network Open. Corresponding Author: James L. Chen, MD, Division of Medical Oncology, Departments of Internal Medicine and Biomedical Informatics, The Ohio State University, A445A, 320 W 10th Ave, Columbus, OH 43210 (james.chen@ osumc.edu). Author Affiliations: The Ohio State University Wexner Medical Center and Comprehensive Cancer Center, The Ohio State University, Columbus (Seligson); Department of Biomedical Informatics, The Ohio State University, Columbus (Awasthi, Liebner, Chen); Foundation Medicine Inc, Cambridge, Massachusetts (Millis); Division of Pediatric Hematology/Oncology, Cincinnati Children's Hospital Medical Center, Cincinnati, Ohio (Turpin); Division of Medical Oncology, Johns Hopkins Medical Center, Baltimore, Maryland (Meyer); Department of Medical Oncology, Roswell Park Cancer Center, Buffalo, New York (Grand'Maison); Division of Medical Oncology, Department of Internal Medicine, The Ohio State University, Columbus (Liebner, Hays, Chen); Division of Gynecologic Oncology, Department of Obstetrics and Gynecology, The Ohio State University, Columbus (Hays). Author Contributions: Drs Seligson and Chen had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Concept and design: Seligson, Awasthi, Millis, Hays, Chen. Acquisition, analysis, or interpretation of data: All authors. Drafting of the manuscript: Seligson, Awasthi, Chen. Critical revision of the manuscript for important intellectual content: Seligson, Millis, Turpin, Meyer, Grand'Maison, Liebner, Hays, Chen. Statistical analysis: Seligson, Awasthi, Millis, Chen. Administrative, technical, or material support: Seligson, Millis, Meyer, Hays, Chen. Supervision: Seligson, Grand'Maison, Liebner, Hays, Chen. Conflict of Interest Disclosures: Dr Millis reported employment by Foundation Medicine outside the submitted work. Dr Meyer reported serving on the advisory board of Bayer Pharmaceuticals and serving as a speaker for Novartis regarding pazopanib outside the submitted work. Dr Liebner reported personal fees from Foundation Medicine outside the submitted work. Dr Chen reported receiving personal fees from Foundation Medicine outside the submitted work. No other disclosures were reported. REFERENCES 1. Sardaro A, Bardoscia L, Petruzzelli MF, Portaluri M. Epithelioid hemangioendothelioma: an overview and update on a rare vascular tumor. Oncol Rev. 2014;8(2):259. doi:10.4081/oncol.2014.259 2. Errani C, Zhang L, Sung YS, et al. A novel WWTR1-CAMTA1 gene fusion is a consistent abnormality in epithelioid hemangioendothelioma of different anatomic sites. Genes Chromosomes Cancer. 2011;50(8):644-653. doi:10. 1002/gcc.20886 3. Tanas MR, Sboner A, Oliveira AM, et al. Identification of a disease-defining gene fusion in epithelioid hemangioendothelioma. Sci Transl Med. 2011;3(98):98ra82. doi:10.1126/scitranslmed.3002409 4. Doyle LA, Fletcher CD, Hornick JL. Nuclear expression of CAMTA1 distinguishes epithelioid hemangioendothelioma from histologic mimics. Am J Surg Pathol. 2016;40(1):94-102. doi:10.1097/PAS. 0000000000000511 5. Rosenberg A, Agulnik M. Epithelioid hemangioendothelioma: update on diagnosis and treatment. Curr Treat Options Oncol. 2018;19(4):19. doi:10.1007/s11864-018-0536-y JAMA Network Open. 2019;2(10):e1912416. doi:10.1001/jamanetworkopen.2019.12416 (Reprinted) October 2, 2019 7/8 Downloaded from jamanetwork.com by guest on 12/03/2024 JAMA Network Open | Oncology Secondary Genomic Variants Associated With Epithelioid Hemangioendothelioma 6. Deyrup AT, Tighiouart M, Montag AG, Weiss SW. Epithelioid hemangioendothelioma of soft tissue: a proposal for risk stratification based on 49 cases. Am J Surg Pathol. 2008;32(6):924-927. doi:10.1097/PAS. 0b013e31815bf8e6 7. Amin RM, Hiroshima K, Kokubo T, et al. Risk factors and independent predictors of survival in patients with pulmonary epithelioid haemangioendothelioma: review of the literature and a case report. Respirology. 2006;11 (6):818-825. doi:10.1111/j.1440-1843.2006.00923.x 8. Trametinib in treating patients with epithelioid hemangioendothelioma that is metastatic, locally advanced, or cannot be removed by surgery. https://clinicaltrials.gov/ct2/show/NCT03148275. Accessed March 15, 2019. 9. Rubin B, Ali S, Subbiah B. Cell cycle dysregulation in epithelioid hemangioendothelioma. Paper presented at: Annual Meeting of The Connective Tissue Oncology Society; November 9, 2017; Maui, HI. 10. Little J, Higgins JP, Ioannidis JP, et al. STrengthening the REporting of Genetic Association studies (STREGA): an extension of the STROBE Statement. Ann Intern Med. 2009;150(3):206-215. doi:10.7326/0003-4819-150-3- 200902030-00011 11. Frampton GM, Fichtenholtz A, Otto GA, et al. Development and validation of a clinical cancer genomic profiling test based on massively parallel DNA sequencing. Nat Biotechnol. 2013;31(11):1023-1031. doi:10.1038/nbt.2696 12. OmniSeq website. https://www.omniseq.com/comprehensive/. Accessed May 21, 2018. 13. Forbes SA, Beare D, Gunasekaran P, et al. COSMIC: exploring the world’s knowledge of somatic mutations in human cancer. Nucleic Acids Res. 2015;43(Database issue):D805-D811. doi:10.1093/nar/gku1075 14. Superpaths. http://www.genecards.org. Accessed May 22, 2019. 15. Chakravarty D, Gao J, Phillips SM, et al OncoKB: a precision oncology knowledge base [published online May 16, 2017]. JCO Precis Oncol. doi:10.1200/PO.17.00011 16. Antonescu CR, Le Loarer F, Mosquera JM, et al. Novel YAP1-TFE3 fusion defines a distinct subset of epithelioid hemangioendothelioma. Genes Chromosomes Cancer. 2013;52(8):775-784. doi:10.1002/gcc.22073 17. Schneider-Stock R, Boltze C, Lasota J, et al. Loss of p16 protein defines high-risk patients with gastrointestinal stromal tumors: a tissue microarray study. Clin Cancer Res. 2005;11(2, pt 1):638-645. 18. Schmieder M, Wolf S, Danner B, et al. p16 expression differentiates high-risk gastrointestinal stromal tumor and predicts poor outcome. Neoplasia. 2008;10(10):1154-1162. doi:10.1593/neo.08646 19. Lagarde P, Pérot G, Kauffmann A, et al. Mitotic checkpoints and chromosome instability are strong predictors of clinical outcome in gastrointestinal stromal tumors. Clin Cancer Res. 2012;18(3):826-838. doi:10.1158/1078- 0432.CCR-11-1610 SUPPLEMENT. eTable 1. Genomic Coverage of Next-Generation Sequencing Platform eTable 2. Description of Pathways eFigure 1. Distribution of Genomic WWTR1-CAMTA1 Fusion Expression eTable 3. Gene Variants in EHE eFigure 2. Age at Diagnosis eMethods. Models eTable 4. Clinicogenomic Features of CDKN2A/B Variant Subjects eTable 5. Targetable Genomic Variants JAMA Network Open. 2019;2(10):e1912416. doi:10.1001/jamanetworkopen.2019.12416 (Reprinted) October 2, 2019 8/8 Downloaded from jamanetwork.com by guest on 12/03/2024
null
10.1103_physrevx.13.021007.pdf
null
null
PHYSICAL REVIEW X 13, 021007 (2023) Entanglement Phase Transition Induced by the Non-Hermitian Skin Effect Kohei Kawabata,1,* Tokiro Numasawa ,2 and Shinsei Ryu1 1Department of Physics, Princeton University, Princeton, New Jersey 08544, USA 2Institute for Solid State Physics, University of Tokyo, Kashiwa 277-8581, Japan (Received 10 June 2022; revised 2 March 2023; accepted 13 March 2023; published 12 April 2023) Recent years have seen remarkable development in open quantum systems effectively described by non- Hermitian Hamiltonians. A unique feature of non-Hermitian topological systems is the skin effect, anomalous localization of an extensive number of eigenstates driven by nonreciprocal dissipation. Despite its significance for non-Hermitian topological phases, the relevance of the skin effect to quantum entanglement and critical phenomena has remained unclear. Here, we find that the skin effect induces a nonequilibrium quantum phase transition in the entanglement dynamics. We show that the skin effect gives rise to a macroscopic flow of particles and suppresses the entanglement propagation and thermalization, leading to the area law of the entanglement entropy in the nonequilibrium steady state. Moreover, we reveal an entanglement phase transition induced by the competition between the unitary dynamics and the skin effect even without disorder or interactions. This entanglement phase transition accompanies non- equilibrium quantum criticality characterized by a nonunitary conformal field theory whose effective central charge is extremely sensitive to the boundary conditions. We also demonstrate that it originates from an exceptional point of the non-Hermitian Hamiltonian and the concomitant scale invariance of the skin modes localized according to the power law. Furthermore, we show that the skin effect leads to the purification and the reduction of von Neumann entropy even in Markovian open quantum systems described by the Lindblad master equation. Our work opens a way to control the entanglement growth and establishes a fundamental understanding of phase transitions and critical phenomena in open quantum systems far from thermal equilibrium. DOI: 10.1103/PhysRevX.13.021007 Subject Areas: Condensed Matter Physics, Quantum Physics, Statistical Physics I. INTRODUCTION Nonequilibrium quantum dynamics provides a profound understanding about quantum many-body systems. Closed quantum systems driven out of equilibrium eventually reach thermal equilibrium, which validates the foundations of quantum statistical mechanics [1–4]. Thanks to the recent advances in quantum simulations and technologies, such thermalization dynamics was experimentally obser- ved in ultracold atoms [5–7] and trapped ions [8]. Thermalization arises from the propagation of quantum correlations and entanglement throughout the whole system and the consequent entanglement entropy proportional to the subsystem [9–11]. Beyond closed the volume of quantum systems, the nonequilibrium dynamics of open quantum systems has recently been studied extensively. *kohei.kawabata@princeton.edu Published by the American Physical Society under the terms of license. the Creative Commons Attribution 4.0 International Further distribution of this work must maintain attribution to the author(s) and the published article’s title, journal citation, and DOI. Researchers have found entanglement phase transitions induced by quantum measurements [12–25]. There, suffi- ciently strong quantum measurements prevent thermal- ization and drive the system into a steady state far from equilibrium for which the entanglement entropy is only proportional to the boundary of the subsystem (i.e., the area law [26]). Such measurement-induced phase transitions also accompany nonequilibrium critical phenomena unique to open quantum systems. As another platform of open systems, the physics effec- tively described by non-Hermitian Hamiltonians has recently attracted growing interest [27,28]. In the classical regime, non-Hermiticity is implemented by controlling gain and loss, and leads to unique phenomena and functionalities without Hermitian counterparts, such as power oscillations [29–31], unidirectional invisibility [32–35], high-performance lasers [36–40], and enhanced sensitivity [41–43]. In the quantum regime, effective non-Hermitian Hamiltonians are justi- fied as conditional dynamics subject to continuous moni- toring and postselection of the null measurement outcome [44–48], as well as the Feshbach projection formalism [49–52]. Non-Hermitian systems have been realized in several open quantum systems, including atoms [53–55], 2160-3308=23=13(2)=021007(26) 021007-1 Published by the American Physical Society KAWABATA, NUMASAWA, and RYU PHYS. REV. X 13, 021007 (2023) photons [56–59], exciton polaritons [60], electronic spins [61,62], and superconducting qubits [63]. On the theoretical side, researchers have studied open quantum dynamics of non-Hermitian systems [64–72]. Notably, non-Hermitian systems at critical points support anomalous singularities called exceptional points [73–75], at which the non- Hermitian Hamiltonians are no longer diagonalizable. Phase transitions and critical phenomena due to exceptional points date back to the Yang-Lee edge singularity [76–79]. Exceptional points are also the key to the real-complex spectral transition protected by parity-time symmetry [80,81] and induce new universality classes of phase transitions in non-Hermitian quantum systems [82–92]. Another unique feature of non-Hermitian systems is the skin effect [93–95]. This is anomalous localization of an extensive number of eigenstates driven by reciprocity- breaking non-Hermiticity, which has no analogs in Hermi- tian systems. The skin effect plays a central role in the topological phases of non-Hermitian systems [96–112]. Since the skin effect leads to extreme sensitivity of the bulk to the boundary conditions, it changes the nature of the bulk-boundary correspondence [93–95,113–120]. More- over, the skin effect originates from the topological invari- ants intrinsic to non-Hermitian systems [111,121,122]. The skin effect has recently been observed in classical experi- ments of mechanical metamaterials [123], electrical circuits [124,125], photonic lattices [126], and active particles [127], as well as quantum experiments of single photons [128] and ultracold atoms [129]. In these experiments, reciprocity-breaking dissipation is introduced by the asym- metry of the hopping amplitudes. It is also relevant to Liouvillians for a quantum master equation [130–134]. The skin effect may open up a way to actively control the phases of matter. Despite the significance of the skin effect for non- Hermitian topological phases, its impact on the genuine quantum nature has remained unclear. While several recent works studied the entanglement dynamics in non- Hermitian quantum systems [66–72], they focused only on non-Hermitian systems that are subject to reciprocal dis- sipation and free from the skin effect. On the basis of the important role of the skin effect in non-Hermitian physics, it may crucially change the entanglement dynamics in open quantum systems. Furthermore, the relevance of the skin effect on quantum phase transitions has also been unclear. The previous works focused on the Yang-Lee edge singu- larity [76–79] and its variants [57,67,85–88,91,92], which do not accompany the skin effect. Although the skin effect may lead to new universality classes of phase transitions and critical phenomena far from thermal equilibrium, no research has hitherto addressed this problem. In this work, we study the impact of the skin effect on the entanglement dynamics and nonequilibrium phase transi- tions in open quantum systems. First, we show that the skin effect gives rise to a macroscopic flow of particles and suppresses the entanglement propagation, leading to a nonequilibrium steady state characterized by the area law of entanglement entropy. This is contrasted with the thermal equilibrium states, which exhibit the volume law of entanglement entropy. Second, we reveal a new type of entanglement phase transition induced by the skin effect. It arises from the competition between coherent coupling and nonreciprocal dissipation; the nonequilibrium steady state exhibits the volume law for small dissipation but the area law for large dissipation, between which the entan- glement entropy grows subextensively (i.e., logarithmically with respect to the subsystem size). Anomalously, this nonequilibrium quantum criticality is characterized by a nonunitary conformal field theory whose effective central charge is extremely sensitive to the boundary conditions. We also demonstrate that it originates from an excep- tional point in the non-Hermitian Hamiltonian and the concomitant scale invariance of the skin modes localized according to the power law. In addition to the conditio- nal dynamics effectively described by non-Hermitian Hamiltonians, we show that the skin effect leads to the purification and the reduction of von Neumann entropy even in Markovian open quantum systems described by the Lindblad master equation. From these results, we show that the skin effect is a new mechanism that triggers entanglement phase transitions and nonequilibrium critical phenomena in open quantum systems. The measurement-induced phase transitions typ- ically rely on spatial or temporal randomness [12–25] while they can occur in some models with no randomness except in measurement outcomes [14]. The entanglement phase transition in this work relies not on any randomness but on the skin effect. While the Yang-Lee edge singularity [76–79] originates from an exceptional point, it does not accompany the skin effect. Furthermore, the boundary- sensitive effective central charge, which implies a new universality class, has never been reported in conformal field theory. Since the skin effect is a universal phenome- non arising solely from non-Hermitian topology, our entanglement phase transition can generally appear in a wide variety of open quantum systems. We hope that these results will deepen our understanding of quantum phases far from thermal equilibrium. The rest of this work is organized as follows. In Sec. II, we describe general behavior of the entanglement dynamics in closed and open quantum systems. In Sec. III, we show the entanglement suppression induced by the skin effect for a non-Hermitian spinless-fermionic model. In Sec. IV, we demonstrate the entanglement phase transition and discuss its nonequilibrium quantum criticality for a non-Hermitian spinful-fermionic model. In Sec. V, we show that the skin effect leads to the purification and reduction of von Neumann entropy in a Liouvillian of the Lindblad master equation. In Sec. VI, we conclude this work with several outlooks. In Appendix A, we describe the implementation 021007-2 ENTANGLEMENT PHASE TRANSITION INDUCED … PHYS. REV. X 13, 021007 (2023) of effective non-Hermitian Hamiltonians in the quantum trajectory approach. In Appendix B, we describe the numerical method to effectively simulate the dynamics of non-Hermitian free fermions. In Appendix C, we provide additional numerical results for different initial conditions. In Appendix D, we describe details of the Liouvillian skin effect. II. ENTANGLEMENT DYNAMICS AND NON-HERMITIAN SKIN EFFECT Before the detailed calculations, we discuss the general behavior of nonequilibrium dynamics in closed and open quantum systems. For simplicity, we assume the quasipar- ticle picture, which is applicable to integrable systems discussed in this work. Under the time evolution of closed quantum systems, the quasiparticles coherently move in all the directions and diffuse throughout the entire system [Fig. 1(a)]. Such a bidirectional propagation of quasipar- ticles arises from the conservation of the particle number and energy. Consequently, quantum correlations develop throughout the system, leading to extensive entanglement for the steady state. This means the entanglement entropy proportional to the subsystem size, i.e., volume law (S ∝ ld with the subsystem length l and spatial dimensions d) [9]. The volume law of the entanglement entropy lies at the heart of thermalization and validates quantum statistical mechanics [1–4]. In open quantum systems, the particle number or energy is not necessarily conserved because of the coupling to the external environment. As a direct result of the violation of the conservation laws, quasiparticles can be amplified or attenuated. As long as such an external coupling is reciprocal, quantum correlations propagate uniformly (a) (b) FIG. 1. Quasiparticle propagation in closed and open quantum systems. (a) Closed quantum systems. Quasiparticles propagate in both directions and diffuse throughout the system, leading to the volume law of entanglement entropy. (b) Open quantum to the skin effect. Nonreciprocal dissipation systems subject makes quasiparticles move toward only one direction, sup- pressing the entanglement propagation and leading to the area law of entanglement entropy. throughout the system in a manner similar to closed quantum systems. However, when the external coupling is nonreciprocal, quasiparticles can be amplified toward one direction and attenuated toward the other direction [Fig. 1(b)]. In such a case, the quasiparticles move only in one direction and accumulate at a boundary for a suffi- ciently long time, i.e., non-Hermitian skin effect [93–95]. Since the quasiparticles are present only at a boundary, the quantum correlations extend not over the entire system but only at the boundary. The entanglement is greatly suppressed and carried only by the skin modes at the boundary, leading to the area law of the entanglement entropy (i.e., S ∝ ld−1). This is a unique consequence of nonreciprocal dissipation for quantum entanglement dynamics. We confirm such a suppression of entanglement for a non-Hermitian spinless-fermionic model (i.e., Hatano- Nelson model [135]) in Sec. III. Notably, an extensive number of localized modes are the entanglement suppression. A possible needed for known mechanism that gives rise to it is disorder. In the presence of sufficiently strong disorder, the system is to the Anderson [136,137] or many-body [3] subject localization, in which thermalization is prohibited. We emphasize that the skin effect is a different mechanism that suppresses the entanglement growth. In fact, the skin effect does not rely on disorder, and occurs only in open quantum systems. The skin effect originates solely from non-Hermitian topology [111,121,122] and hence appears in a wide variety of open quantum systems. Even if the skin effect suppresses the quasiparticle diffusion and the entanglement propagation, it is unclear whether the skin effect can compete with the unitary dynamics and give rise to a continuous phase transition. In fact, in the Hatano-Nelson model and many other non- Hermitian models, even infinitesimal non-Hermiticity causes the skin effect and results in no continuous phase transition. Nevertheless, we show that the skin effect indeed induces new nonequilibrium phase transitions and critical phenomena intrinsic to open quantum systems. There, an entanglement phase transition arises from the competition between the coherent coupling and the nonreciprocal dissipation: the system reaches a thermal equilibrium state exhibiting the volume law for small dissipation while it reaches a nonequilibrium steady state exhibiting only the area law for large dissipation, between which the entan- glement entropy grows subextensively (i.e., S ∝ log l) with an unconventional nonequilibrium quantum critica- lity described by a nonunitary conformal field theory. We demonstrate such an entanglement phase transition induced by the skin effect by explicitly constructing and investigating an illustrative example of non-Hermitian spinful-fermionic models (i.e., symplectic Hatano-Nelson model [122,138]) in Sec. IV. As well as the conditional dynamics effectively described by non-Hermitian Hamiltonians, the skin effect 021007-3 KAWABATA, NUMASAWA, and RYU PHYS. REV. X 13, 021007 (2023) has a considerable impact also on the open quantum dynamics described by a master equation. While a Markovian open quantum system typically exhibits the thermal equilibrium state with infinite temperature as the steady state, the skin effect dramatically changes the properties of the steady state toward far from equilibrium. We show the purification and suppression of von Neumann entropy for Markovian open quantum systems described by the Lindblad master equation in Sec. V. Entanglement phase transitions can also occur as a the competition between the unitary consequence of dynamics and quantum measurements [12–25]. However, the entanglement phase transition in this work exhibits properties distinct from the measurement-induced phase transitions. First, the boundary-sensitive critical behaviors have never been found in the previous works on the measurement-induced phase transitions. Additionally, the measurement-induced phase transitions typically rely on spatial or temporal randomness and many-body inter- actions aside from some exceptions [25]. By contrast, the skin effect induces the entanglement phase transition even without randomness and interactions, which enables a deep understanding of the phase transition and critical the behavior measurement-induced phase transitions manifest them- selves only in a conditional quantum trajectory postselected by measurements and disappear in the open quantum dynamics averaged over multiple quantum trajectories. On the other hand, the skin effect occurs and yields purification even in the averaged open quantum dynamics described by the Markovian master equation. in open quantum systems. Furthermore, III. ENTANGLEMENT SUPPRESSION INDUCED BY THE NON-HERMITIAN SKIN EFFECT We study the nonequilibrium quantum dynamics induced by the non-Hermitian skin effect. To this end, we investigate the Hatano-Nelson model [135] as a prototypical example that exhibits the skin effect: ˆH ¼ − 1 2 X l ½ðJ þ γÞˆc† lþ1 ˆcl þ ðJ − γÞˆc† l ˆclþ1(cid:2); ð1Þ where ˆcl (ˆc† l ) annihilates (creates) a spinless fermion at site l, J > 0 denotes the Hermitian hopping amplitude, and γ ∈ R denotes the asymmetric hopping amplitude as a source of non-Hermiticity. Here, we assume jγj < J for simplicity. The asymmetric hopping can be implemented in the quantum trajectory approach (see Appendix A for details) [44–48] and has been realized in the recent experi- ments of single photons [128] and ultracold atoms [129]. the Bloch Under the periodic boundary conditions, Hamiltonian for the Hatano-Nelson model reads Thus, the complex-valued spectrum of HðkÞ winds around the origin in the complex-energy plane when the momen- tum k goes around the Brillouin zone ½0; 2πÞ. From this complex-spectral winding, we introduce a topological invariant [104,105]: I W ≔ − 0 2π dk 2πi d dk log det HðkÞ: ð3Þ Since such complex-spectral winding is ill defined in Hermitian systems, the winding number W is intrinsic the to non-Hermitian systems. As a consequence of intrinsic non-Hermitian topology, an extensive number of boundary modes appear under the open boundary con- ditions [121,122], i.e., non-Hermitian skin effect [93–95]. While we here focus on the Hatano-Nelson model in Eq. (1) as a prototypical example, the skin effect generally occurs and leads to the entanglement suppression as long as the intrinsic non-Hermitian topology is non- trivial W ≠ 0. In the following, we impose the open boundary conditions and prepare the initial state as the charge density wave state, (cid:2)YL=2 (cid:3) jψ 0i ¼ ˆc† 2l jvaci; ð4Þ l¼1 where jvaci is the fermionic vacuum state, and the system length L is assumed to be even. The many-particle wave function evolves by the non-Hermitian Hamiltonian ˆH in Eq. (1) as jψðtÞi ¼ e−i ˆHtjψ 0i ke−i ˆHtjψ 0ik : ð5Þ Despite non-Hermiticity of the Hamiltonian, the particle number N ¼ L=2 is conserved under dynamics. Thanks to the free (i.e., quadratic) nature of the model, its dynamics can be efficiently calculated (see Appendix B for details). We show that the skin effect leads to a nonequilibrium steady state whose entanglement is suppressed, which is to be contrasted with the thermal equilibrium states in closed quantum systems. While we here consider Eq. (4) as an initial state, the entanglement suppression depends only on the skin effect, and the specific details of the initial state should be irrelevant. A. Skin effect We begin with investigating the time evolution of the local particle number: nlðtÞ ≔ hψðtÞj ˆnljψðtÞi: ð6Þ HðkÞ ¼ −J cos k þ iγ sin k: ð2Þ In Hermitian systems, particles are distributed uniformly [Fig. 2(a)]. In the presence of non-Hermiticity, by contrast, 021007-4 ENTANGLEMENT PHASE TRANSITION INDUCED … PHYS. REV. X 13, 021007 (2023) (a) (b) 1.0 0.8 0.6 0.4 0.2 0 FIG. 2. Time evolution of the local particle number nlðtÞ ≔ hψðtÞj ˆnljψðtÞi in the Hatano-Nelson model with open boundaries (L ¼ 100, J ¼ 1.0) for (a) γ ¼ 0.0 and (b) γ ¼ 0.8. The initial state is prepared as the charge density wave state in Eq. (4). In the presence of non-Hermiticity, particles accumulate at the boundary, which is a clear signature of the non-Hermitian skin effect. particles accumulate at the right (left) edge of the system for γ > 0 (γ < 0) [Fig. 2(b)]. Such localization of an extensive number of particles is impossible in closed quantum systems and is a clear signature of the non-Hermitian skin effect. transformation [GL(1) gauge The skin effect can be understood by the imaginary transformation; gauge GLðnÞ is the general linear group of n × n invertible matrices] [94,119,135]. Let us introduce the new fermionic operators by l ≔ elθ ˆc† ˆp† l ; ˆql ≔ e−lθ ˆcl; ð7Þ where θ ∈ C plays a role of the complex-valued gauge. The Hamiltonian in Eq. (1) is rewritten as ˆH ¼ − 1 2 h XL−1 l¼1 e−θðJ þ γÞ ˆp† lþ1 ˆql þ eθðJ − γÞ ˆp† l ˆqlþ1 i : ð8Þ In particular, when we choose θ so that it will satisfy e−θðJ þ γÞ ¼ eθðJ − γÞ, i.e., (cid:2) 1 2 log J þ γ J − γ (cid:3) ; θ ¼ ð9Þ the Hamiltonian reduces to ˆH ¼ − p ffiffiffiffiffiffiffiffiffiffiffiffiffiffi J2 − γ2 2 XL−1 (cid:5) l¼1 lþ1 ˆql þ ˆp† ˆp† l ˆqlþ1 (cid:6) : ð10Þ Now that the asymmetric hopping formally disappears, the Hamiltonian is diagonalized to q ffiffiffiffiffiffiffiffiffiffiffiffiffiffi J2 − γ2 X ˆH ¼ − ðcos kÞ ˆp† k ˆqk k ð11Þ by the Fourier transforms, r r ffiffiffiffiffiffiffiffiffiffiffiffi 2 L þ 1 ffiffiffiffiffiffiffiffiffiffiffiffi 2 L þ 1 ˆpk ≔ ˆqk ≔ XL l¼1 XL l¼1 ˆpl sinðklÞ; ˆql sinðklÞ; ð12Þ ð13Þ with momentum k ¼ nπ=ðL þ 1Þ (n ¼ 1; 2; …; L). Thus, the spectrum of ˆH is entirely real. Non-Hermiticity of ˆH originates solely from the nonorthogonality of the quasi- particles (i.e., ˆpk ≠ ˆqk). In the presence of the skin effect, while the spectrum of an infinite non-Hermitian system coincides with the infinite-size limit of the spectrum of the corresponding finite system with periodic boundaries, it does not coincide with the spectrum of the infinite-size limit of the corresponding finite system with open boun- daries [122,139]. This extreme sensitivity yields unique open quantum phenomena, as we show below. Because of the GL(1) transformation in Eq. (7), the quasiparticle ˆpk is exponentially localized at the right (left) edge while the quasiparticle ˆqk is exponentially localized at the left (right) edge for Reθ > 0 (Reθ < 0). All the the edges, which is the quasiparticles are localized at hallmark of the skin effect unique to non-Hermitian systems. Thus, the Hamiltonian ˆH annihilates the quasi- particles around one edge and creates the quasiparticles around the other edge under its time evolution. Here, θ−1 characterizes the localization length of the quasiparticles. It should be noted that the above transformation is possible only for the open boundary conditions and is unfeasible so that the periodic boundary conditions can be satisfied. The quasiparticles form Bloch waves delocalized through- out the system under the periodic boundary conditions, where no length scale appears as a consequence of non- Hermiticity. The emergent length scale θ−1 is unique to the open boundary conditions. We also investigate the time evolution of the correlation matrix, CijðtÞ ≔ hψðtÞjˆc† i ˆcjjψðtÞi; ð14Þ for i; j ¼ 1; 2; …; L. In the absence of non-Hermiticity, the quasiparticles propagate in both directions, leading to the diffusion of particles and quantum information [Fig. 3(a)]. In the presence of non-Hermiticity, on the other hand, the quasiparticles cease to move, and the correlation propaga- tion is frozen [Fig. 3(b)]. This is another consequence of the skin effect. Because of the localization of the quasiparticles, they move toward the right (left) edge for γ > 0 (γ < 0) at the beginning of the dynamics. However, once the quasi- particles accumulate at the edge, they are no longer mobile because of the Pauli exclusion principle. Under the skin effect, the system soon reaches a nonequilibrium steady state in which an extensive number of particles are the frozen localized at an edge. It is noteworthy that 021007-5 KAWABATA, NUMASAWA, and RYU PHYS. REV. X 13, 021007 (2023) (a) (b) 0.10 0.08 0.06 0.04 0.02 0 FIG. 3. Correlation propagation in the Hatano-Nelson model with open boundaries (L ¼ 100, J ¼ 1.0) for (a) γ ¼ 0.0 and (b) γ ¼ 0.8. The absolute values jCl;l0 j of the correlation matrix are shown as a function of site l and time t with l0 ¼ L=2 ¼ 50. The initial state is prepared as the charge density wave state in Eq. (4). In the presence of non-Hermiticity, the correlation propagation is frozen as a consequence of the non-Hermitian skin effect. correlation propagation due to the skin effect is different from the supersonic correlation propagation in non- Hermitian quantum systems with reciprocal dissipation [66,67]. This difference also shows a unique role of the skin effect in open quantum systems. B. Current Next, we investigate the charge current, IlðtÞ ≔ hψðtÞjˆIljψðtÞi; ð15Þ is the local current operator between sites l where ˆIl and l þ 1: ˆIl ≔ iJ 2 ðˆc† l ˆclþ1 − ˆc† lþ1 ˆclÞ: ð16Þ While no current flows in closed quantum systems at thermal equilibrium, the skin effect gives rise to a current in open quantum systems. Figure 4 shows the behavior of the P total charge current IðtÞ ≔ l¼1 IlðtÞ induced by the skin effect. In the presence of non-Hermiticity, the current takes a nonzero steady value for sufficiently long time [Fig. 4(a)]. This means that the system reaches a nonequilibrium steady L−1 state accompanying a nonzero current in contrast with the thermal equilibrium states, where the current should vanish [i.e., I ¼ oðLÞ] [140]. The current for the steady state monotonically increases as a function of non-Hermiticity [Fig. 4(b)]. Furthermore, it grows linearly with respect to the system length L [Fig. 4(c)] and hence is indeed a macroscopic quantity. The macroscopic current induced by the skin effect may be characterized by topological field theory [111]. To understand how the skin effect gives rise to a current in more detail, we also study the local distribution of the current (Fig. 5). Notably, in the presence of non- Hermiticity, the current arises only in the bulk and vanishes around the edges [Fig. 5(b)]. On the basis of the local particle distribution in Fig. 2(b), the current arises only in the region where the particles are neither dense nor sparse. This is because particles cannot enter such dense or sparse regions from the environment because of the Pauli exclu- sion principle. is also compatible with the frozen correlation propagation in Fig. 3(b). Moreover, the con- tinuity equation of our non-Hermitian system reads It ∂ ∂t nl þ ðIl − Il−1Þ ¼ σl; ð17Þ where σl is the local inflow of particles from the external environment at site l. In Hermitian systems, σl vanishes for arbitrary l and t owing to the conservation of the particle number [Fig. 5(c)]. Under the skin effect, a pair of a source and sink appears, between which the current flows [Fig. 5(d)]. It is also notable that the current does not arise for small non-Hermiticity or a short system length (Fig. 4). In such a case, the localization length of the many-body skin modes is comparable with the system length, and consequently particles cannot enter the system from the environment. C. Entanglement dynamics The non-Hermitian skin effect gives rise to a nonequili- brium flow not only of particles but also of quantum information. To show this, we investigate the time evolu- tion of the entanglement entropy in the Hatano-Nelson (a) (b) (c) P l¼1 hψðtÞjˆIljψðtÞi in the Hatano-Nelson model with open boundaries (J ¼ 1.0). The initial state FIG. 4. Total charge current IðtÞ ≔ is prepared as the charge density wave state in Eq. (4). (a) Time evolution of the current (L ¼ 100) for γ ¼ 0.0 (black dashed curve), 0.2 (blue curve), 0.4 (green curve), 0.6 (light green curve), 0.8 (orange curve), and 1.0 (red curve). (b) Charge current for the steady state as a function of non-Hermiticity γ for L ¼ 100. (c) Charge current for the steady state as a function of the system length for γ ¼ 0.8. L−1 021007-6 ENTANGLEMENT PHASE TRANSITION INDUCED … PHYS. REV. X 13, 021007 (2023) (a) (c) (b) (d) (a) (c) (b) (d) FIG. 5. Local current distribution in the Hatano-Nelson model with open boundaries (J ¼ 1.0). The initial state is prepared as the charge density wave state in Eq. (4). (a),(b) Time evolution of the local current IlðtÞ ≔ hψðtÞjˆIljψðtÞi for (a) γ ¼ 0.0 and (b) γ ¼ 0.8. (c),(d) Time evolution of the local particle inflow for (c) γ ¼ 0.0 and (d) γ ¼ 0.8. model. We focus on the von Neumann entanglement entropy SðL; lÞ between the subsystem in ½1; l(cid:2) and the rest of the system. Here, we calculate the entanglement entropy from a single wave function jψðtÞi instead of the biorthogonal density operator [87,89,91]. In Hermitian systems, SðL; lÞ grows linearly in time until it saturates to the extensive entanglement entropy S ∝ l [9,141], which is consistent with our numerical calculations for γ ¼ 0 [Fig. 6(a)]. In the presence of non-Hermiticity, however, the growth of the entanglement entropy is greatly suppressed. The entanglement entropy for the steady state is much smaller than that for the Hermitian case and monotonically decreases as a function of non-Hermiticity [Fig. 6(b)]. In the Hermitian case γ ¼ 0, the steady-state entanglement entropy grows linearly with the system length, i.e., volume law; the steady-state entanglement entropy is independent of the system length, i.e., area law [Figs. 6(c) and 6(d)]. in the non-Hermitian case γ ≠ 0, The suppression of the entanglement entropy originates the from the skin effect. In closed quantum systems, quasiparticles diffuse throughout the system and let the system be a thermal equilibrium state exhibiting the extensive entanglement entropy. On the other hand, a macroscopic current from the external environment pushes the quasiparticles only in one direction and forbids quan- tum diffusion throughout the system. Consequently, the quasiparticles are localized only at one edge (i.e., skin effect) and cannot develop a global quantum correlation, leading to the area law of the entanglement entropy for the nonequilibrium steady state. FIG. 6. Entanglement entropy (EE) of the Hatano-Nelson model with open boundaries (J ¼ 1.0). The initial state is prepared as the charge density wave state in Eq. (4). (a) Time evolution of the entanglement entropy SðL; L=2Þ (L ¼ 100) for γ ¼ 0.0 (black dashed curve), 0.1 (violet curve), 0.2 (blue curve), 0.4 (green curve), 0.6 (light green curve), 0.8 (orange curve), and 1.0 (red curve). (b) Entanglement entropy SðL; L=2Þ for the steady state as a function of non-Hermiticity γ (L ¼ 100). (c) Entanglement entropy SðL; L=2Þ for the steady state as a function of the system length L. (d) Entanglement entropy SðL; lÞ (L ¼ 100) for the steady state as a function of the subsystem length l. It should be noted that the area law of the entanglement entropy can also occur in non-Hermitian systems with broken parity-time symmetry [70]. In such systems, the suppression of the entanglement is due to the relaxation toward a pure state with the largest imaginary part of the complex-valued energy. By contrast, our non-Hermitian system hosts the entirely real spectrum under the open boundary conditions and hence does not rely on parity- time-symmetry breaking. The non-Hermitian skin effect is a new mechanism of open quantum systems that hinders the growth of the quantum correlation and entanglement. IV. ENTANGLEMENT PHASE TRANSITION INDUCED BY THE NON-HERMITIAN SKIN EFFECT In the Hatano-Nelson model, even infinitesimal non- Hermiticity induces the skin effect and makes the system relax to far from equilibrium. To understand the non- equilibrium quantum criticality induced by the skin effect, we consider the symplectic generalization of the Hatano- Nelson model [122,138]: XL ˆH ¼ − 1 2 ½ˆc† lþ1ðJ þ γσz − iΔσxÞˆcl l¼1 l ðJ − γσz þ iΔσxÞˆclþ1(cid:2); þ ˆc† ð18Þ 021007-7 KAWABATA, NUMASAWA, and RYU PHYS. REV. X 13, 021007 (2023) with Pauli matrices σi’s (i ¼ x, y, z). The fermionic ˆcl ¼ ðˆcl;↑ ˆcl;↓ÞT [creation operator annihilation operator l;↑ ˆc† l ¼ ðˆc† ˆc† l;↓Þ] now includes the spin degree of freedom. Because of non-Hermiticity γ > 0 (γ < 0), the up-spin fermions are pushed toward the right (left) while the down- spin fermions are pushed toward the left (right). In addition, Δ ∈ R controls the spin-orbit coupling between the up-spin fermions and down-spin fermions. Owing to the spin-orbit coupling Δ, the model is free from the skin effect even in the presence of non-Hermiticity γ as long as jγj < jΔj is satisfied. Similarly to the original Hatano-Nelson model, the symplectic Hatano-Nelson model in Eq. (18) can be implemented in the quantum trajectory approach (see Appendix A for details). It is notable that non-Hermitian spin-orbit-coupled fermions have been realized in recent experiments of ultracold atoms [55], and our model can also be realized in a similar experiment. Under the periodic boundary conditions, the Bloch Hamiltonian of the symplectic Hatano-Nelson model reads HðkÞ ¼ −J cos k þ ðiγσz þ ΔσxÞ sin k; ð19Þ whose complex spectrum is obtained as ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi γ2 − Δ2 EðkÞ ¼ −J cos k (cid:3) i q FIG. 7. Phase diagram of the symplectic Hatano-Nelson model. For jγj < jΔj (blue region), no skin effect occurs, and the entanglement entropy for the steady state obeys the volume law. For jγj > jΔj (red region), the reciprocal skin effect occurs, and the entanglement entropy for the steady state obeys the area law. The phase boundary jγj ¼ jΔj ≠ 0 (black line) marks critical points, at which the skin modes exhibit the scale invariance, and the entanglement entropy for the steady state grows subexten- sively (i.e., logarithmically with respect to the subsystem length). In fact, the non-Hermitian Hamiltonian in Eq. (18) respects reciprocity, sin k: ð20Þ ˆT ˆH† ˆT−1 ¼ ˆH; ð22Þ Therefore, for small non-Hermiticity jγj < jΔj, the spectrum is entirely real, and no skin effect occurs. For large non- Hermiticity jγj > jΔj, on the other hand, each band is characterized by the complex-spectral winding and subject to the skin effect [122]. There, up-spin fermions and down- spin fermions are localized at opposite boundaries. This is ensured by the Z2 topological reciprocal skin effect invariant ν ∈ f0; 1g unique to non-Hermitian systems [105]: (cid:7) ð−1Þν ≔ sgn Pf½Hðk ¼ πÞT(cid:2) Pf½Hðk ¼ 0ÞT(cid:2) (cid:8) Z 1 2 k¼π k¼0 × exp − d log det ½HðkÞT(cid:2) (cid:9)(cid:10) ; ð21Þ with the unitary operator T ≔ σy for the symplectic Hatano- Nelson model. The presence or absence of the skin effect is controlled by the competition between non-Hermiticity γ and spin-orbit coupling Δ, and jγj ¼ jΔj marks a phase transition point, between which the skin effect occurs or not (Fig. 7). The reciprocal skin effect generally occurs as long as the Z2 topological invariant in Eq. (21) is nontrivial. Thus, while we here consider the symplectic Hatano-Nelson model in Eq. (18) for illustrative purposes, the Z2 skin effect and the concomitant entanglement phase transition should appear in a wide variety of open quantum systems. It is also notable that the symplectic Hatano-Nelson model respects reciprocity, which is one of the fundamental [105]. symmetry for non-Hermitian systems internal where ˆT is an antiunitary operator satisfying ˆT ˆcl and ˆTz ˆT−1 ¼ z(cid:4) for z ∈ C. In terms of Hamiltonian in Eq. (19), reciprocity is written as ˆT−1 ¼ σy ˆcl the Bloch THTðkÞT−1 ¼ Hð−kÞ; TT(cid:4) ¼ −1; ð23Þ with the unitary operator T ≔ σy. The Kramers pair structure between up-spin and down-spin fermions, as well as the concomitant skin effect, is protected by reciprocity. Below, we study the nonequilibrium quantum dynamics of the symplectic Hatano-Nelson model. We choose the initial state as jψ 0i ¼ (cid:2)YL=2 l¼1 (cid:3) 2l−1;↑ ˆc† ˆc† 2l;↓ jvaci; ð24Þ where the system length L is assumed to be even. We confirm that the system reaches a many-body steady state subject to the reciprocal skin effect in Sec. IVA. This nonequilibrium steady state is characterized by a spin current in contrast to the thermal equilibrium states, as we show in Sec. IV B. Furthermore, in Sec. IV C, we demonstrate that the phase boundary jγj ¼ jΔj marks an entanglement phase transition, between which the steady state exhibits the volume law or the area law (Fig. 7). The critical point characterized by a conformal field theory that is anomalously sensitive to the boundary conditions. In Sec. IV D, we also show that jγj ¼ jΔj is 021007-8 ENTANGLEMENT PHASE TRANSITION INDUCED … PHYS. REV. X 13, 021007 (2023) this nonequilibrium quantum criticality originates from the scale-invariant skin modes decaying according to the power law. While we here choose Eq. (24) as an initial state, the universal properties of the entanglement phase transition— the critical behaviors in Eqs. (31), (38), and (46)—arise solely from the scale invariance of the skin modes and should not depend on the specific details of the initial state (see Appendix C for details). the skin effect freezes the correlation propagation in a to the original Hatano-Nelson model. similar manner the quasiparticles cease to propagate even at Notably, the critical point (i.e., jγj ¼ jΔj). The frozen correlation propagation implies the skin effect even at the critical point. In Sec. IV D, we indeed demonstrate the skin effect at the critical point while the critical skin modes are localized algebraically instead of exponentially. A. Reciprocal skin effect B. Spin current We begin with investigating the time evolution of local particle numbers for each spin (Fig. 8). Below the critical point (i.e., jγj < jΔj), the particles are distributed almost uniformly throughout the system. Above the critical point (i.e., jγj > jΔj), on the other hand, the skin effect indeed occurs, and the particles are localized at the edges. In to the original Hatano-Nelson model, up-spin contrast fermions are localized at (left) edge while the right down-spin fermions are localized at the left (right) edge for γ > 0 (γ < 0) [Fig. 8(d)]. Consequently, particles are uniformly distributed on average. This is a unique feature of the reciprocity-protected skin effect in the symplectic Hatano-Nelson model. We also investigate the correlation propagation in the symplectic Hatano-Nelson model (Fig. 9). The correlation matrix now includes the spin degree of freedom: Cis;js0ðtÞ ≔ hψðtÞjˆc† i;s ˆcj;s0jψðtÞi: ð25Þ Below the critical point (i.e., jγj < jΔj), the correlation bidirectionally propagates throughout the system even in the presence of non-Hermiticity, which is a signature of the quantum diffusion. Above the critical point (i.e., jγj > jΔj), We next investigate the time evolution of the current. Owing to the spin degree of freedom, we consider both the total charge current, ˆIc ≔ ˆI↑ þ ˆI↓; and the total spin current, ˆIs ≔ ˆI↑ − ˆI↓; ð26Þ ð27Þ with ˆIs ≔ iJ 2 XL−1 l¼1 ðˆc† l;s ˆclþ1;s − ˆc† lþ1;s ˆcl;sÞ ðs ¼ ↑; ↓Þ: ð28Þ While ˆIs is not conserved in the presence of the spin-orbit coupling Δ, it gives an intuitive measure for the spin current. Even in the presence of non-Hermiticity γ, the charge current IcðtÞ always vanishes as a consequence of reciprocity [Fig. 10(a)]. On the other hand, the spin current IsðtÞ exhibits characteristic behavior unique to the sym- plectic Hatano-Nelson model. Below the critical point FIG. 8. Time evolution of the local particle number nl;sðtÞ ≔ hψðtÞj ˆnl;sjψðtÞi for s ¼ ↑ (top panels) and s ¼ ↓ (bottom panels) in the symplectic Hatano-Nelson model with open boundaries (L ¼ 100, J ¼ 1.0, Δ ¼ 0.5). The initial state is prepared as Eq. (24). Non- Hermiticity is chosen to be (a) γ ¼ 0.0, (b) γ ¼ 0.4, (c) γ ¼ 0.5, and (d) γ ¼ 0.8. While no skin effect occurs for jγj < jΔj, the reciprocal skin effect occurs for jγj > jΔj. 021007-9 KAWABATA, NUMASAWA, and RYU PHYS. REV. X 13, 021007 (2023) FIG. 9. Correlation propagation in the symplectic Hatano-Nelson model with open boundaries (L ¼ 100, J ¼ 1.0; Δ ¼ 0.5). The absolute values jCl↑;l0↑j ¼ jCl↓;l0↓j (top panels) and jCl↑;l0↓j ¼ jCl↓;l0↑j (bottom panels) of the correlation matrix are shown as a function of site l and time t with l0 ¼ L=2 ¼ 50. The initial state is prepared as Eq. (24). Non-Hermiticity is chosen to be (a) γ ¼ 0.0, (b) γ ¼ 0.4, (c) γ ¼ 0.5, and (d) γ ¼ 0.8. (i.e., jγj < jΔj), the spin current just oscillates and vanishes after averaging over time; above the critical point (i.e., jγj > jΔj), the skin effect occurs and induces a nonzero spin current. Similarly to the steady-state charge current in the original Hatano-Nelson model, the steady-state spin current grows as we increase non-Hermiticity or the system length [Figs. 10(c) and 10(d)]. Thus, the system reaches a non- equilibrium steady state with a nonzero spin current. The spin current characterizes the nonequilibrium quantum phases of the symplectic Hatano-Nelson model as an order parameter. This is contrasted with the thermal equilibrium states and the nonequilibrium steady states in the original Hatano-Nelson model, which are respectively characterized by zero current and nonzero charge currents. C. Entanglement phase transition Now, we investigate the entanglement dynamics of the symplectic Hatano-Nelson model (Fig. 11). In the Hermitian case γ ¼ 0, the system reaches the thermal equilibrium state (or the generalized Gibbs state) under the dynamics, and the entanglement entropy for the steady state grows linearly with the system length, i.e., volume law. Even in the presence of non-Hermiticity, the volume law of the entanglement entropy persists for jγj < jΔj. This contrasts with the original Hatano-Nelson model, in which the volume law is violated by infinitesimal non-Hermiticity (Sec. III C). The robust volume law is consistent with the quantum diffusion of quasiparticles shown in Fig. 9. As non-Hermiticity increases, the entanglement entropy for the steady state gradually decreases and sharply vanishes at jγj ¼ jΔj. For the larger non-Hermiticity jγj > jΔj, the entanglement entropy is greatly suppressed and no longer grows even if we increase the system length L, i.e., the area law. Similarly to the original Hatano-Nelson model, the area law of the steady-state entanglement entropy arises from the skin effect. Here, jγj ¼ jΔj marks a nonequili- brium phase transition across which the steady-state entan- glement entropy exhibits the volume law or the area law (a) (c) (b) (d) FIG. 10. Current in the symplectic Hatano-Nelson model with open boundaries (J ¼ 1.0, Δ ¼ 0.5). The initial state is prepared as Eq. (24). Time evolution of the (a) charge current IcðtÞ ≔ hψðtÞjˆIcjψðtÞi and (b) spin current IsðtÞ ≔ hψðtÞjˆIsjψðtÞi for L ¼ 100 and γ ¼ 0.0, 0.2, 0.4, 0.5, 0.6, 0.8, 1.0. (c) Spin current for the steady state as a function of non-Hermiticity γ for L ¼ 100. (d) Spin current for the steady state as a function of the system length L for γ ¼ 0.8. 021007-10 ENTANGLEMENT PHASE TRANSITION INDUCED … PHYS. REV. X 13, 021007 (2023) (a) (c) (b) (d) (a) (b) FIG. 12. Entanglement entropy of the symplectic Hatano- Nelson model with open boundaries (J ¼ 1.0) at the critical (γ ¼ Δ). The initial state is prepared as Eq. point (24). (a) Entanglement entropy SðL; lÞ (L ¼ 100) for the steady state as a function of the subsystem length l for γ ¼ 0.0, 0.01, 0.02, 0.05, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, and 1.0. (b) Effective central charge c as a function of γ ¼ Δ (L ¼ 100). FIG. 11. Entanglement entropy of the symplectic Hatano- Nelson model with open boundaries (J ¼ 1.0, Δ ¼ 0.5). The initial state is prepared as Eq. (24). (a) Time evolution of the entanglement entropy SðL; L=2Þ (L ¼ 100) for γ ¼ 0.0 (black dashed curve), 0.2 (blue curve), 0.4, 0.48, 0.5 (green curves), 0.6 (light green curve), and 0.8 (orange curve). (b) Entanglement entropy density SðL; L=2Þ=L (L ¼ 100) for the steady state as a function of non-Hermiticity γ. The black dashed curve is the fitting result SðL; L=2Þ=L ¼ 0.94ðΔ=J − γ=JÞ0.44 around the critical point γ ¼ Δ. (c) Entanglement entropy SðL; L=2Þ for the steady state as a function of the system length L for γ ¼ 0.0, 0.2, 0.4, 0.45, 0.48, 0.49, 0.495, 0.5, 0.6, and 0.8. (d) Entangle- ment entropy SðL; lÞ (L ¼ 100) for the steady state as a function of the subsystem length l. (Fig. 7). Around this transition point jγj ¼ jΔj, the density the steady-state entanglement entropy exhibits the of critical behavior, (cid:2) (cid:3) 0.44(cid:3)0.06 SsðL; L=2Þ L ∝ jΔj − jγj J ; ð29Þ for jγj ≤ jΔj [Fig. 11(b)]. Notably, the entanglement phase transition induced by the skin effect occurs even without randomness. This contrasts with the phase transitions induced by quantum measurements, which typically rely on spatial or temporal randomness [12–25], although some models can exhibit the phase transitions even without randomness [14]. The skin effect provides a new mechanism for the entanglement phase transition and gives rise to a new universality class of nonequilibrium quantum phase transitions. To unveil the nonequilibrium quantum criticality, we further study the entanglement entropy at the transition point jγj ¼ jΔj. We numerically calculate the steady-state entanglement entropy as a function of the system parameter jγ=Jj ¼ jΔ=Jj. According to the conformal field theory the entanglement entropy SsðL; lÞ description [9,142], of a one-dimensional quantum critical system with open boundaries grows logarithmically with respect subsystem length l: to the SsðL; lÞ ¼ c 6 log (cid:2) sin (cid:3) πl L þ S0; ð30Þ where c is the central charge that characterizes the relevant conformal field theory, and S0 is a nonuniversal constant. the steady-state entanglement Despite non-Hermiticity, entropy of the symplectic Hatano-Nelson model at the critical point jγj ¼ jΔj is well fitted by this subextensive scaling [Fig. 12(a)]. Remarkably, the effective central charge c is sensitive to the system parameter γ=J ¼ Δ=J in contrast to unitary conformal field theory for closed quantum systems [Fig. 12(b)]. It can take large values for small non-Hermiticity jγ=Jj, in which a crossover between the unitary and nonunitary critical points should occur. For larger jγ=Jj, on the other hand, the effective central charge c exhibits the power-law behavior: c ∝ jγ=Jj−ð0.66(cid:3)0.03Þ; ð31Þ whose critical exponent is close to 2=3. Here, we identify the effective central charge from the logarithmic scaling of the entanglement entropy. We note that this is apparently different from the effective central charge in the context of nonunitary conformal field theory, which is defined by subtracting the dimension of the lowest-dimensional oper- ator from the central charge. Still, the parameter-dependent effective central charge c implies nonunitary or irrational conformal field theory that underlies the nonequilibrium quantum criticality induced by the skin effect. It merits further study to identify this anomalous type of conformal field theory. It should also be noted that a couple of recent works on random nonunitary quantum dynamics have reported a similar subextensive growth of the steady-state entangle- ment entropy with the parameter-dependent effective cen- in the nonunitary tral charge [21,23,68]. For example, random dynamics of free fermions in Ref. [68], the effective central charge obeys c ∝ β−1, where β is the degree of non-Hermiticity. The different exponents, 2=3 of 021007-11 KAWABATA, NUMASAWA, and RYU PHYS. REV. X 13, 021007 (2023) [21,23,68]. By contrast, our symplectic Hatano-Nelson model and 1 of the nonuni- tary random dynamics in Ref. [68], signal the different universality classes of the entanglement phase transition. Furthermore, as also discussed above, temporal randomness plays a crucial role in the entanglement phase transitions in Refs. the entanglement phase transition in this work is based not on the randomness but on the skin effect. As shown below, it arises from the scale invariance of skin modes, and consequently, the underlying nonunitary conformal field theory is also anoma- lously sensitive to the boundary conditions. Our model provides a new type of nonequilibrium quantum phase transitions that belongs to a different universality class. D. Criticality of skin modes We demonstrate that the nonequilibrium quantum criti- cality at the phase transition point jγj ¼ jΔj originates from the scale invariance of the skin modes due to an exceptional point. To understand this, we first perform an imaginary gauge transformation in a manner similar to the original Hatano-Nelson model (Sec. III A). Here, because of the spin degree of freedom, we consider the following SL(2) gauge transformation rather than the GL(1) one [138]: l ≔ ˆc† ˆp† l V (cid:2) elθ 0 (cid:3) ; 0 e−lθ (cid:2) e−lθ 0 0 elθ (cid:3) V−1 ˆcl; ˆql ≔ ð32Þ for θ ∈ C and V ∈ SLð2Þ [SLðnÞ is the special linear group of n × n matrices with determinant 1]. This transformation retains reciprocity in Eqs. (22) and (23). With these new fermion operators ˆp† l and ˆql, the symplectic Hatano-Nelson model reads ˆH ¼ − 1 2 (cid:8) XL (cid:2) e−ðlþ1Þθ ˆp† lþ1 (cid:3) V−1ðJ þ γσz − iΔσxÞV 0 eðlþ1Þθ (cid:2) elθ (cid:3) ˆql 0 e−lθ (cid:9) 0 (cid:3) 0 (cid:3) l¼1 (cid:2) þ ˆp† l e−lθ 0 0 elθ V−1ðJ − γσz þ iΔσxÞV ˆqlþ1 : ð33Þ (cid:2) eðlþ1Þθ 0 0 e−ðlþ1Þθ Away from the critical point jγj ¼ jΔj, the non-Hermitian matrix J − γσz þ iΔσx can be diagonalized by appropri- ately choosing V: V−1ðJ − γσz þ iΔσxÞV ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi γ2 − Δ2 J þ p ¼ 0 ! : 0 p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi γ2 − Δ2 J − ð34Þ Furthermore, e−θðJ þ p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi γ2 − Δ2 let us choose θ such that Þ, i.e., ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi γ2 − Δ2 Þ ¼ eθðJ − p it satisfies exceptional point under the periodic boundary conditions (see Sec. IV E for details). If the skin effect occurs, the localization properties of the skin modes are captured by the quasiparticles ˆpl and ˆql. For Reθ > 0, the up-spin (down-spin) component of ˆpl is exponentially localized at the right (left) edge while the up- spin (down-spin) component of ˆql is exponentially local- ized at the left (right) edge. Here, all the quasiparticles are subject to the skin effect, and no delocalized modes are present in the bulk. The localization length ξ of the single- particle skin modes is obtained from Eq. (35) as (cid:7) θ ¼ 1 2 log (cid:2) J þ J − p p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi γ2 − Δ2 ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi γ2 − Δ2 (cid:3) : ð35Þ ξ ¼ 1 Reθ ¼ ∞ 1=jθj ðjγj < jΔjÞ ðjγj > jΔjÞ: ð37Þ With these choices of V and θ, the Hamiltonian reduces to ˆH ¼ − p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi J2 − γ2 þ Δ2 2 XL−1 l¼1 ð ˆp† lþ1 ˆql þ ˆp† l ˆqlþ1Þ; ð36Þ in which the asymmetric hopping vanishes formally. It can be further diagonalized similarly to Eq. (11). The imaginary gauge transformation is feasible only under the open boundary conditions in such a manner that the boundary conditions are respected. The spectrum does not show any singular behavior even across the critical point jγj ¼ jΔj, which contrasts with the emergence of an Thus, no skin effect occurs for jγj < jΔj while the recip- rocal skin effect occurs for jγj > jΔj, which is consistent with our numerical calculations in Fig. 8. Notably, around the critical point jγj ¼ jΔj, the localization length ξ exhibits the critical behavior: ξ ≃ p J ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi γ2 − Δ2 ∝ ðjγj − jΔjÞ−1=2: ð38Þ At the critical point jγj ¼ jΔj, the localization length ξ of the skin modes diverges, which signals the scale invariance. Consequently, we find that there emerge skin modes decaying according to the power law due to an exceptional 021007-12 ENTANGLEMENT PHASE TRANSITION INDUCED … PHYS. REV. X 13, 021007 (2023) point. At the critical point jγj ¼ jΔj, the above imaginary gauge transformation is no longer applicable. In fact, the non-Hermitian matrix J − γσz þ iΔσx is nondiagonalizable for jγj ¼ jΔj and supports an exceptional point. Instead of the diagonalization in Eq. (34), the matrix is only trans- formed into the Jordan normal form: V−1ðJ − γσz þ iΔσxÞVjjγj¼jΔj ¼ (cid:2) (cid:3) : J −γ 0 J ð39Þ As a result, the Hamiltonian reduces to ˆH ¼ − (cid:2) (cid:8) ˆp† lþ1 1 0 J 2 XL−1 l¼1 (cid:3) (cid:2) ˆql þ ˆp† l γ=J 1 1 −γ=J 0 1 (cid:3) ˆqlþ1 (cid:9) : ð40Þ Because of this defective the nondiagonalizability, Hamiltonian supports scale-invariant skin modes linearly localized at the boundary. To see this, we study the spatial distribution of the single-particle wave functions in a transfer-matrix method (see, for example, Ref. [143]). Let E ∈ C be a single-particle eigenenergy and jϕi ¼ P l;s ϕl;sjlijsi be the corresponding eigenstate, where l and s denote the sites and spins, respectively. The single- particle Schrödinger equation in real space reads (cid:2) (cid:2) (cid:3) − J 2 1 γ=J 0 1 ⃗ϕl−1 − J 2 1 −γ=J 0 1 (cid:3) ⃗ϕlþ1 ¼ E ⃗ϕl; ð41Þ with ⃗ϕl ¼ ðϕl;↑ϕl;↓ÞT. For simplicity, we consider a zero- energy eigenstate (i.e., E ¼ 0). Then, we have leading to (cid:2) ⃗ϕlþ1 ¼ − 1 γ=J 0 1 (cid:3) 2 ⃗ϕl−1; ⃗ϕ2lþ1 ¼ ð−1Þl ⃗ϕ2lþ2 ¼ ð−1Þl (cid:2) (cid:2) 1 0 1 0 (cid:3) 2l γ=J 1 (cid:3) 2l γ=J 1 ⃗ϕ1; ⃗ϕ2: ð42Þ ð43Þ ð44Þ As an important property of the Jordan normal form, it is nilpotent with index 2, i.e., (cid:8)(cid:2) 1 γ=J 0 1 (cid:3) (cid:9) n − 1 ¼ 0; ð45Þ for n ≥ 2. Consequently, we have (cid:3) (cid:2) k ⃗ϕ2lþ1k ¼ (cid:11) (cid:11) (cid:11) (cid:11) 1 2lγ=J 0 1 (cid:11) (cid:11) (cid:11) (cid:11) ∝ 2ljγj J k ⃗ϕ1k; ⃗ϕ1 for sufficiently large l, meaning the linear growth of the norm kϕlk of the wave function with respect to the site l. Thus, the skin modes at the critical point are localized linearly in contrast to the exponentially localized skin modes off the critical point. The linear localization of the critical skin modes gets stronger for larger non- Hermiticity jγj, which is compatible with the decrease of entanglement entropy as a function of jγj (Fig. 12). We note that similar power-law decay arises even for E ≠ 0 since the lth power of the Jordan normal form still appears. It is also notable that the lth power of a diagonalizable matrix gives λl with the eigenvalue λ of the matrix. The emergence of the power law in terms of l, rather than the exponential, is a unique feature of nondiagonalizable matrices. In general, the (n − 1)th power-law localization k ⃗ϕlk ∝ l−ðn−1Þ (l ≫ 1) appears if an n × n Jordan matrix is concerned while only the linear localization appears in the symplectic Hatano- Nelson model. The criticality of skin modes is understood also by a continuum model. To have such a continuum model, let us focus on a gapless point k ¼ π=2, around which the Bloch Hamiltonian HðkÞ in Eq. (19) reads HðkÞ ≃ Jk þ iγσz þ Δσx: ð47Þ Now, we consider a semi-infinite system with a domain wall at x ¼ 0. The system is prepared as a vacuum for x < 0 while the Hamiltonian for x > 0 is HðxÞ ¼ −iJ∂x þ iγσz þ Δσx: ð48Þ Let E ∈ R be an eigenenergy and ⃗ϕðxÞ ∈ C2 be the corresponding right eigenstate. For x > 0, the Schrödinger equation reads ð−iJ∂x þ iγσz þ ΔσxÞ ⃗ϕðxÞ ¼ E ⃗ϕðxÞ; ð49Þ which is solved as ⃗ϕðxÞ ¼ eiðE−iγσzþΔσxÞx=J ⃗ϕð0Þ (cid:8) (cid:3) ¼ eiEx=J cosh (cid:2) x ξ ðγσz þ iΔσxÞξ J þ (cid:3)(cid:9) (cid:2) x ξ ⃗ϕð0Þ; sinh ð50Þ p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi γ2 − Δ2 with ξ ≔ J= [i.e., Eq. (38)]. Thus, away from the critical point (i.e., jγj ≠ jΔj), the wave function for large x behaves as ( k ⃗ϕðxÞk ≃ k ⃗ϕð0Þk ðjγj < jΔjÞ ex=ξk ⃗ϕð0Þk ðjγj > jΔjÞ; ð51Þ ð46Þ which is consistent with the results for the corresponding lattice model. At the critical point jγj ≠ jΔj, by contrast, 021007-13 KAWABATA, NUMASAWA, and RYU PHYS. REV. X 13, 021007 (2023) the relevant length scale ξ diverges, and the wave function behaves as (a) (b) k ⃗ϕðxÞk ≃ jγjx J k ⃗ϕð0Þk; ð52Þ which also reproduces the linear localization of the skin modes [i.e., Eq. (46)]. in equilibrium, The scale invariance at the critical point appears also for thermal phase transitions [144,145] and quantum phase transitions [146] in equilibrium. At such a critical the correlation length diverges point and the power-law correlation arises. By contrast, the scale invariance of our non-Hermitian system originates from the exceptional point and the concomitant scale- invariant skin modes, which are intrinsic to open quantum systems. Our results provide a new type of nonequili- brium quantum criticality that has no analogs in closed quantum systems. We note in passing that the phase transition in the symplectic Hatano-Nelson model is distinct from a dis- continuous phase transition in Refs. [147,148], which studied the finite-size scaling of skin modes in the presence of a symmetry-breaking perturbation. In these previous works, skin modes are localized exponentially even at the phase transition point. By contrast, the symplectic Hatano-Nelson model exhibits a continuous phase transi- tion that hosts skin modes localized according to the power law, for which the universal critical exponents such as Eqs. (31), (38), and (46) are well defined. E. Criticality for the periodic boundary conditions To understand the significance of the skin effect, we also study the entanglement dynamics of the symplectic Hatano-Nelson model with periodic boundaries. Under the periodic boundary conditions, the model exhibits a phase transition also at jγj ¼ jΔj. However, the phase transition is not characterized by the skin effect but the reality of the spectrum. In fact, eigenstates are always delocalized throughout the system because of translation invariance. Meanwhile, the spectrum EðkÞ in Eq. (20) is entirely real for jγj ≤ jΔj but no longer real for jγj > jΔj. At the critical point jγj ¼ jΔj, the Bloch Hamiltonian HðkÞ in Eq. (19) is not diagonalizable and forms an exceptional point. Similarly to the open boundary conditions, the time- averaged spin current vanishes below the critical point (Fig. 13). Above the critical point, the spectrum is complex, and the system relaxes to the many-body eigenstate that possesses the largest the complex spectrum. This nonequilibrium steady state is characterized by the nonzero spin current, which is qualitatively similar to the spin current induced by the skin effect (Fig. 10). It should be noted that the spin current for the open boundary conditions is carried by a superposition of many-body skin modes instead of a single eigenstate. Around the critical imaginary part of FIG. 13. Spin current in the symplectic Hatano-Nelson model with periodic boundaries (L ¼ 100, J ¼ 1.0, Δ ¼ 0.5). The initial state is prepared as Eq. (24). (a) Time evolution of the spin current for γ ¼ 0.0 (black dashed curve), 0.2 (blue curve), 0.4, 0.5, 0.6 (green curves), 0.8 (orange curve), and 1.0 (red curve). (b) Spin current for the steady state as a function of non-Hermiticity γ. The black dashed curve is the fitting result Is ¼ 123Jðγ=J − Δ=JÞ0.50 around the critical point γ ¼ Δ. point, the steady-state spin current exhibits the power-law behavior, (cid:3) 0.50(cid:3)0.02 (cid:2) jγj − jΔj J Is ∝ J ðjγj ≥ jΔjÞ; ð53Þ where the critical exponent 0.50 (cid:3) 0.02 is close to 1=2. This critical exponent may be related to the point-gap closing and the concomitant emergence of an exceptional point, where the complex spectrum in Eq. (20) exhibits the similar critical behavior ImEðkÞ ∝ ðjγj − jΔjÞ1=2 for jγj ≥ jΔj. We also study the entanglement dynamics for the periodic boundary conditions. Qualitatively, it is similar to the entanglement dynamics for the open boundary conditions: the entanglement entropy of the nonequilibrium steady state is extensive below the critical point while it is suppressed above the critical point [Fig. 14(a)]. However, the steady state exhibits a distinct critical behavior around the phase transition point jγj ¼ jΔj. Below the transition the steady-state point entanglement critical behavior [Fig. 14(b)], entropy exhibits the density of jγj ≤ jΔj), (i.e., the (cid:2) (cid:3) 0.33(cid:3)0.02 SsðL; L=2Þ L ∝ jΔj − jγj J ; ð54Þ jγj ¼ jΔj, whose critical exponent 0.33 (cid:3) 0.02 deviates from that under the open boundary conditions in Eq. (29). At the critical point the steady-state entanglement entropy under the periodic boundary conditions is much smaller than that under the open boundary conditions. According to conformal field theory [9,142], the entangle- ment entropy SsðL; lÞ of a one-dimensional quantum critical system with periodic boundaries behaves by SsðL; lÞ ¼ c 3 log (cid:2) sin (cid:3) πl L þ S0: ð55Þ 021007-14 ENTANGLEMENT PHASE TRANSITION INDUCED … PHYS. REV. X 13, 021007 (2023) (a) (b) (c) (d) (e) FIG. 14. Entanglement entropy of the symplectic Hatano- Nelson model with periodic boundaries (L ¼ 100, J ¼ 1.0). The initial state is prepared as Eq. (24). (a) Time evolution of the entanglement entropy SðL; L=2Þ (Δ ¼ 0.5) for γ ¼ 0.0 (black dashed curve), 0.2 (blue curve), 0.4, 0.5, 0.6 (green curves), 0.8 (orange curve), and 1.0 (red curve). (b) Entanglement entropy density SðL; L=2Þ=L for the steady state as a function of non- Hermiticity γ (Δ ¼ 0.5). The black dashed curve is the fitting result SðL; L=2Þ=L ¼ 0.56ðΔ=J − γ=JÞ0.33 around the critical point γ ¼ Δ. (c) Entanglement entropy SðL; lÞ for the steady state at the critical point (γ ¼ Δ) as a function of the subsystem length l for γ ¼ 0.0, 0.1, 0.2, 0.4, 0.6, 0.8, and 1.0. (d) Effective central charge c as a function of γ at the critical point (γ ¼ Δ). The black dashed line shows c ¼ 2. (e) cn obtained from the R´enyi entanglement entropy SðnÞðL; lÞ for the steady state at the critical point (γ ¼ Δ ¼ 1.0) as a function of the R´enyi index n. The black dashed curve shows the conformal field theory result cn ¼ cð1 þ 1=nÞ=2 with c ¼ 2. We confirm that our numerical results for the steady states are consistent with this subextensive behavior [Fig. 14(c)]. Remarkably, in contrast to the parameter-dependent central charge for the open boundary conditions, the effective central charge does not depend on the system parameter jγ=Jj ¼ jΔ=Jj and is obtained as the following constant [Fig. 14(d)]: c ¼ 2.04 (cid:3) 0.08; ð56Þ which is compatible with the effective central charge c ¼ 2 of non-Hermitian free fermions [91]. The different behavior of the effective central charge c means the different universality classes of the entanglement phase transition. Moreover, we investigate the R´enyi entanglement en- tropy for the steady state, which is defined as SðnÞ s ≔ ðtr log ˆρnÞ=ð1 − nÞ for the reduced density operator ˆρ and coincides with the von Neumann entanglement entropy Ss for n → 1. According to conformal field theory, the R´enyi entanglement entropy also follows the scaling in Eq. (55), where the central charge c is replaced by cn ≔ cð1 þ 1=nÞ=2 [9,142]. We also confirm this conformal field theory scaling with respect to the R´enyi index n [Fig. 14(e)]. We note that the parameter dependence of the effective central charge for small non-Hermiticity γ is due to a finite-size effect that interpolates between the unitary and nonunitary critical points. Importantly, the mechanism of the entanglement phase transition is different depending on the boundary conditions. Under the periodic boundary conditions, the entanglement phase transition originates from the real-complex spectral transition. At the critical point, the Bloch Hamiltonian is defective and exhibits an exceptional point. This is similar to the entanglement phase transition due to parity-time-sym- metry breaking [70]. In such a case, the effective central charge is the constant in Eq. (56). Under the open boundary conditions, on the other hand, the model exhibits no spectral transitions. While non-Hermiticity is irrelevant to the spec- trum, it gives rise to a length scale of the skin modes. Then, the nonequilibrium quantum criticality is induced by the scale invariance of the skin modes, as discussed in Sec. IV D. The effective central charge depends on the system parameter [i.e., Eq. (31)] in contrast to unitary conformal field theory. Despite these differences, the critical behavior of the bulk modes and that of the boundary (i.e., skin) modes may have a hidden connection with each other. In fact, the skin effect under the open boundary conditions originates from the non-Hermitian topological invariant under the periodic boundary conditions [111,121,122], which can be consid- ered as the bulk-boundary correspondence of non- Hermitian topological systems. In this respect, it is of importance to consider the different critical behaviors of the bulk and boundary modes in terms of nonunitary conformal field theory. It is also notable that while the bulk and boundary modes are clearly separated in the symplectic Hatano-Nelson model, they can appear simultaneously in more generic non-Hermitian models. V. PURIFICATION INDUCED BY THE LIOUVILLIAN SKIN EFFECT We have so far considered the conditional dynamics effectively described by non-Hermitian Hamiltonians. Notably, the skin effect occurs also in the open quantum dynamics described by the master equation [149,150], d dt ˆρ ¼ Lˆρ; ð57Þ 021007-15 KAWABATA, NUMASAWA, and RYU PHYS. REV. X 13, 021007 (2023) where L is a Liouvillian that acts on the density operator ˆρ (see Appendix A for a relationship between non-Hermitian Hamiltonians and Liouvillians in the quantum trajectory approach). Although the Liouvillian L is not an operator but a superoperator, it is still non-Hermitian. Consequently, L can exhibit the skin effect in a similar manner to non- Hermitian Hamiltonians [130–134]. Here, we demonstrate that the Liouvillian skin effect has a significant influence on the open quantum dynamics described by the master equation. In particular, we show that the Liouvillian skin effect leads to the purification and the reduction of von Neumann entropy for the steady state. We consider the following prototypical model that exhibits the Liouvillian skin effect [131]: Lˆρ ≔ XL X l¼1 n¼R;L (cid:2) ˆLln ˆρ ˆL† ln − 1 2 f ˆL† ln ˆLln; ˆρg (cid:3) ; ð58Þ where the jump operators are ˆLlR ≔ r ffiffiffiffiffiffiffiffiffiffiffi J þ γ 2 ˆc† lþ1 ˆcl; ˆLlL ≔ r ffiffiffiffiffiffiffiffiffiffiffi J − γ 2 ˆc† l ˆclþ1; ð59Þ ð60Þ to the right and from the right with J > 0 and jγj ≤ J. Similarly to the Hatano-Nelson model, ˆLnR and ˆLnL describe the dissipative hopping from the left to the left, respectively. Consequently, in the presence of the asym- metry of the hopping (i.e., γ ≠ 0), the spectrum and eigenstates of the Liouvillian dramatically change accord- ing to the boundary conditions. In particular, the steady state ˆρs greatly depends on the boundary conditions. In this Liouvillian, the total particle number is conserved. This in Refs. [130,132,134], in which the total particle number decreases with time. l¼1 ˆc† l ˆcl contrasts with the Liouvillians ˆN ¼ P L Z ≔ XL l¼1 rL ¼ rðrL − 1Þ r − 1 : ð63Þ P We note in passing that the steady state in Eq. (62) is formally equivalent to the Gibbs state Z−1 l¼1 e−βEljlihlj with the linear potential βEl ≔ −l log r. While the effective temperature is infinite in the absence of the asymmetric hopping (i.e., γ ¼ 0), it decreases as the asymmetric hopping jγj increases and reaches zero for the completely asymmetric hopping γ ¼ (cid:3)J. L We demonstrate that the skin effect has a considerable influence on the open quantum dynamics even in the Markovian regime. In particular, the skin effect can purify mixed states. Let us prepare an initial state as the com- pletely mixed state ˆρ0 ∝ 1 and consider the dynamics described by the Liouvillian in Eq. (58). As shown in Fig. 15(a), the initially low purity monotonically increases with time. The purity for the steady state increases with the larger asymmetry jγj, leading to a pure state for the completely asymmetric hopping γ ¼ (cid:3)J [Fig. 15(b)]. The steady-state purity is analytically obtained from Eq. (62) as Ps ≔ tr ˆρ2 s ¼ r − 1 r þ 1 rL þ 1 rL − 1 ≃ γ J ; ð64Þ (a) (b) (c) (d) For the single-particle case, the steady state for the periodic boundary conditions is the completely mixed state (see Appendix D for details), ˆρs ¼ 1 L ; ð61Þ while the steady state for the open boundary conditions is the skin modes, ˆρs ¼ 1 Z XL l¼1 rljlihlj; ð62Þ with r ≔ ðJ þ γÞ=ðJ − γÞ, jli ≔ ˆc† zation constant, l jvaci, and the normali- FIG. 15. Purification induced by the Liouvillian skin effect (L ¼ 50, J ¼ 1.0). The initial state is prepared as the completely mixed state ˆρ0 ¼ 1=L with the purity P0 ¼ 1=L and the von Neumann entropy S0 ¼ log L. (a) Time evolution of the purity for γ ¼ 0.0 (black dashed curve), γ ¼ 0.2 (blue curve), γ ¼ 0.4 (green curve), γ ¼ 0.6 (light green curve), γ ¼ 0.8 (orange curve), and γ ¼ 1.0 (red curve). (b) Steady-state purity as a function of γ (red curve), consistent with the analytical result Ps ≃ γ=J (black dashed curve). (c) Time evolution of the von Neumann entropy. (d) Steady-state von Neumann entropy as a function of γ (red curve), consistent with the analytical result Ss ≃ ðJ þ γ=2γÞ logðJ þ γ=2γÞ − ðJ − γ=2γÞ logðJ − γ=2γÞ (black dashed curve). 021007-16 ENTANGLEMENT PHASE TRANSITION INDUCED … PHYS. REV. X 13, 021007 (2023) for γ > 0 and L → ∞. This analytical formula is consistent with the numerical results. We also calculate the time evolution of the von Neumann entropy S ≔ −tr ˆρs log ˆρs, as shown in Fig. 15(c). While the reciprocal dynamics realizes the maximal entropy, the asymmetry of the dissipative hopping lowers the entropy. The entropy Ss for the steady state monotonically decreases as a function of jγj, reaching zero for the completely asymmetric case γ ¼ (cid:3)J [Fig. 15(d)]. Here, Ss is also analytically obtained as Ss ≔ −tr ˆρs log ˆρs ¼ log Z − log r r − 1 (cid:2) (cid:2) (cid:3) J þ γ 2γ log ≃ LrLþ1 Z þ (cid:3) J þ γ 2γ log r r − 1 (cid:2) − J − γ 2γ (cid:3) (cid:2) J − γ 2γ (cid:3) ; log ð65Þ for γ > 0 and L → ∞. Notably, while the steady-state entropy Ss subextensively increases with respect to the system length L (i.e., Ss ¼ log L) in the absence of the skin effect, Ss is independent of L (i.e., area law) in the presence of the skin effect. This is similar to the entanglement suppression of the open quantum dynamics effectively described by a non-Hermitian Hamiltonian that is discussed in the previous sections. The purification and suppression of the von Neumann entropy are induced by the Liouvillian skin effect. Under the periodic boundary conditions, no skin effect occurs, and the steady state is the completely mixed state in Eq. (61). Consequently, no purification occurs, and the steady state is characterized by the maximal entropy. [12–25]. However, It is also notable that purification can arise from quantum measurements such measurement- induced purification occurs only in the conditional dynam- ics of a particular quantum trajectory. This conditional nature of the open quantum dynamics is a key to the measurement-induced phase transitions. By contrast, we here demonstrate that the skin effect leads to the purifica- tion even in the Markovian master equation characterized by a Liouvillian, which describes the open quantum dynamics averaged over multiple quantum trajectories. This also shows a significant role of the skin effect in the open quantum dynamics. VI. DISCUSSION The entanglement dynamics provides the foundations of quantum statistical physics. However, the nature of entan- glement in open quantum systems has remained elusive in contrast to closed quantum systems. In this work, we show that the skin effect, a universal feature intrinsic to non- Hermitian systems, has a significant impact on the entan- glement dynamics in open quantum systems. We show that the skin effect suppresses the entanglement growth and even induces an entanglement phase transition. This is triggers the different from the known mechanism that entanglement phase transition such as quantum measure- ments [12–25]. While we consider the prototypical models for illustrative purposes, the skin effect originates solely from non-Hermitian topology, and hence our entanglement phase transition should generally arise in a wide range of open quantum systems. On the basis of the recent exper- imental observations of the skin effect in open quantum systems [128,129], as well as the realization of non- Hermitian spin-orbit-coupled fermions [55], our results should be observed in a similar experimental setup. We show that our entanglement phase transition accom- panies anomalous nonequilibrium quantum criticality that is described by the boundary-sensitive effective central charges [cf. the difference between Figs. 12(b) and 14(d)]. These anomalous critical behaviors imply a new univer- sality class of phase transitions in open quantum systems. It merits further study to derive the nonunitary conformal field theory that describes the nonequilibrium quantum criticality induced by the skin effect. The different critical behaviors in the bulk and boundaries may be unified into the same field theory. In this respect, it is worth noting that the skin effect can be considered as a quantum anomaly of a topological field theory intrinsic to non- Hermitian systems [111]. Furthermore, we demonstrate that our entanglement phase transition is induced by the criticality of skin modes that decay according to the power law. Notably, while the conventional Bloch band theory cannot describe the skin modes, recent works developed a non-Bloch band theory that correctly characterizes the skin modes [94,119,122]. However, the non-Bloch band theory only predicts the exponentially localized skin modes and cannot describe the critical skin modes discovered in this work. It is significant to generally develop a modified band theory that captures the phase transitions and critical phenomena induced by the non-Hermitian skin effect. Additionally, the skin effect leads to the slowdown of relaxation processes [131]. The critical skin effect should yield the logarithmic correction of the relaxation time. We also show that the skin effect plays an important role in the open quantum dynamics described by the master equation. In particular, the skin effect changes the proper- ties of the nonequilibrium steady state and increases the purity and decreases the von Neumann entropy. These findings may lead to potential applications of the skin effect in quantum information science. They also imply that the skin effect has a considerable impact in a wide range of open classical and quantum dynamics. In this research direction, it is worth studying the role of the skin effect, for example, in quantum circuits. We note in passing that recent works have found signatures of non-Hermitian [151,152]. topology in monitored quantum circuits 021007-17 KAWABATA, NUMASAWA, and RYU PHYS. REV. X 13, 021007 (2023) Moreover, it is meaningful to explore the relevance of the skin effect in classical stochastic processes such as the asymmetric simple exclusion process [153]. Another remarkable mechanism that prohibits the quan- tum diffusion is disorder. In closed quantum systems, sufficiently strong disorder drives the systems into the Anderson [136,137] or many-body [3] localization, result- ing in the absence of thermalization. While the skin effect also accompanies an extensive number of localized eigenmodes similarly to the disorder-induced localization, we emphasize that it does not rely on disorder and thus gives a different mechanism that hinders the entanglement propagation and thermalization. Meanwhile, it is intriguing to consider the open quantum dynamics in the presence of disorder. In fact, non-Hermiticity changes the universality [135,138,154–161]. classes of The interplay of disorder and dissipation should further enrich phase transitions and critical phenomena in open quantum systems. localization transitions While we focus on one-dimensional systems in this work, it is also worthwhile to study non-Hermitian systems in higher dimensions. Different types of skin effects appear in higher dimensions, such as the chiral magnetic skin effect [111,162–164], higher-order skin effect [165,166], and defect-induced skin effect [167–169]. These higher- dimensional skin effects may give rise to further different universality classes of phase transitions and critical phe- nomena in open quantum systems. It is also of interest to study the entanglement dynamics of non-Hermitian inter- acting systems. Several recent works have shown that the interplay of non-Hermiticity and many-body inter- [170–177]. actions Similarly to the many-body localized phases due to dis- order [178–180], many-body skin modes may exhibit the logarithmic violation of the area law for the entanglement growth. to new quantum phases leads ACKNOWLEDGMENTS We thank Anish Kulkarni and Yuhan Liu for helpful discussions. K. K. is supported by the Japan Society for the Promotion of Science (JSPS) through the Overseas Research Fellowship. S. R. is supported by the National Science Foundation under Award No. DMR-2001181, and by a Simons Investigator Grant from the Simons Foundation (Award No. 566116). This work is supported by the Gordon and Betty Moore Foundation through Grant No. GBMF8685 toward the Princeton theory program. APPENDIX A: EFFECTIVE NON-HERMITIAN HAMILTONIANS The non-Hermitian Hamiltonians in Eqs. (1) and (18) can be realized in the quantum trajectory approach [44–48]. Let us consider a Markovian open quantum system, ðA2Þ ðA3Þ which is generally described by the Lindblad master equation [149,150]: d dt ˆρ ¼ −i½ ˆH; ˆρ(cid:2) þ (cid:2) X n ˆLn ˆρ ˆL† n − (cid:3) ˆLn; ˆρg ; 1 2 f ˆL† n ðA1Þ where ˆρ is the density operator, ˆH is the Hamiltonian that describes the coherent dynamics, and ˆLn’s are the jump operators that describe the coupling to the external envi- ronment. This master equation can be written as d dt ˆρ ¼ −ið ˆH eff ˆρ − ˆρ ˆH† effÞ þ X ˆLn ˆρ ˆL† n; n with the effective non-Hermitian Hamiltonian: ˆH eff ≔ ˆH − i 2 X ˆL† n ˆLn: n P ˆLn ˆρ ˆL† n p ˆLn ffiffiffiffiffi dt to stochastic loss events. Here, The last term n specifies each quantum trajectory subject can be considered to be a measurement operator for a signal n in the time interval ½t; t þ dt(cid:2), and 1 − i ˆH dt can be consid- ered to be a measurement operator for no signals. Under continuous monitoring and postselection of the null meas- urement outcome, the quantum jumps are no longer relevant, and the dissipative dynamics is described by the effective non-Hermitian Hamiltonian ˆH To obtain the Hatano-Nelson model eff. in Eq. (1), we choose the Hamiltonian ˆH and the jump operators ˆLl’s (l ¼ 1; 2; …; L) to be [104] eff J 2 p XL l¼1 ffiffiffiffiffi jγj ˆH ¼ − ðˆc† lþ1 ˆcl þ ˆc† l ˆclþ1Þ; ˆLl ¼ ½ˆcl þ i sgn ðγÞˆclþ1(cid:2): ðA4Þ ðA5Þ Although the effective Hamiltonian ˆH eff differs from Eq. (1) P by the background constant loss −ijγj ˆN ¼ −ijγj L l ˆcl, it only describes the total decay of the system and does not contribute to the dynamics of the wave function. Similarly, to obtain the symplectic Hatano-Nelson model in Eq. (18), we choose ˆH and ˆLl’s (l ¼ 1; 2; …; L) to be l¼1 ˆc† ˆH ¼ − 1 2 XL l¼1 lþ1ðJ − iΔσxÞˆcl þ ˆc† ½ˆc† l ðJ þ iΔσxÞˆclþ1(cid:2); ˆLl;↑ ¼ ˆLl;↓ ¼ p p ffiffiffiffiffi jγj ffiffiffiffiffi jγj ½ˆcl þ i sgn ðγÞˆclþ1(cid:2); ½ˆcl − i sgn ðγÞˆclþ1(cid:2): ðA6Þ ðA7Þ ðA8Þ As described above, the open quantum dynamics char- acterized by the non-Hermitian Hamiltonian is conditional, and the success probability of having the desirable non- Hermitian Hamiltonian can be low at long time. This is 021007-18 ENTANGLEMENT PHASE TRANSITION INDUCED … PHYS. REV. X 13, 021007 (2023) different from the quantum master equation, which describes the open quantum dynamics of the mixed states averaged over many quantum trajectories and hence is free from the postselection. However, in certain cases, this difficulty can be circumvented, and the effective non- Hermitian Hamiltonian is well realized with a reasonable probability (see, for example, Ref. [85]). In this respect, it is also notable that a similar experimental difficulty should arise also in the measurement-induced phase transitions. In fact, only a quantum trajectory conditioned on a set of measurement outcomes can exhibit an entanglement phase transition, while the mixed quantum state averaged over many quantum trajectories should not exhibit such a phase transition. Still, a different way to realize the measurement- induced phase transitions without the postselection of a certain set of measurement outcomes has recently been proposed [24]. Finally, while we here focus on the quantum trajectory approach, it should be noted that the effective non-Hermitian Hamiltonians can be justified also by the Feshbach projection formalism [49–52]. APPENDIX B: NUMERICAL METHOD FOR THE DYNAMICS OF NON-HERMITIAN FREE FERMIONS We describe a numerical method to investigate the dynamics of non-Hermitian free (i.e., quadratic) fermions. An initial state jψ 0i evolves by the non-Hermitian Hamiltonian ˆH as Eq. (5). The denominator ke−i ˆHtjψ 0ik describes the normalization of the evolved state due to the conditional nature of the non-Hermitian Hamiltonian. This time evolution is equivalently described by the nonlinear Schrödinger equation [65]: i d dt jψi ¼ ð ˆH − hψj ˆHjψiÞjψi: ðB1Þ Despite non-Hermiticity of the Hamiltonian, the total particle number is conserved under the dynamics when the initial state is an eigenstate of the particle number operator (i.e., ˆNjψ 0i ¼ Njψ 0i). This is a consequence of U(1) symmetry ½ ˆH; ˆN(cid:2) ¼ 0. We first consider a spinless-fermionic system such as the Hatano-Nelson model. We prepare an initial state as a Gaussian state with a fixed particle number N. As an advantage of the quadratic Hamiltonian, the evolved state remains to be a Gaussian state through the time evolution in Eq. (5). Thus, the state can always be represented as (cid:2)XL YN (cid:3) jψðtÞi ¼ UlnðtÞˆc† l jvaci; ðB2Þ n¼1 l¼1 where jvaci is the fermionic vacuum and U is the L × N isometry satisfying U†U ¼ 1: ðB3Þ In this representation, the matrix U ¼ UðtÞ contains all information about the quantum dynamics. In particular, the L × L correlation matrix, CijðtÞ ≔ hψðtÞjˆc† i ˆcjjψðtÞi; ðB4Þ is obtained as CðtÞ ¼ ½UðtÞU†ðtÞ(cid:2)T: ðB5Þ From the correlation matrix, the local particle number in Eq. (6) reads nlðtÞ ¼ CllðtÞ; and the local charge current in Eq. (15) reads IlðtÞ ¼ J Im½Clþ1;lðtÞ(cid:2): ðB6Þ ðB7Þ To calculate the entanglement entropy S between the subsystem ½x1; x2(cid:2) and the rest of the system, we diago- x2 nalize the ðx2 − x1 þ 1Þ × ðx2 − x1 þ 1Þ submatrix ½C(cid:2) i;j¼x1 and obtain its eigenvalues λn’s (n ¼ 1; 2; …; x2 − x1 þ 1). Then, the von Neumann entanglement entropy is given as S ¼ − Xx2−x1þ1 i¼1 ½λi log λi þ ð1 − λiÞ log ð1 − λiÞ(cid:2); ðB8Þ and the R´enyi entanglement entropy is SðnÞ ¼ 1 1 − n Xx2−x1þ1 i¼1 log ½λn i þ ð1 − λiÞn(cid:2); ðB9Þ with the R´enyi index n. Here, we calculate the entangle- ment entropy from a single wave function instead of the biorthogonal density operator constructed from both right and left eigenstates [87,91]. The time evolution of U ¼ UðtÞ is efficiently calculated as follows. After the time interval Δt, the state evolves as jψðt þ ΔtÞi ∝ e−i ˆHΔtjψðtÞi (cid:2)XL YN ¼ n¼1 l¼1 ½e−ihΔtU(cid:2)lnðtÞˆc† l (cid:3) jvaci; ðB10Þ P where h is the L × L single-particle Hamiltonian (i.e., ˆH ¼ i hij ˆcj). To restore the normalization condi- tion hψðtÞjψðtÞi ¼ 1, we perform the QR decomposition, i;j¼1 ˆc† L e−ihΔtU ¼ QR; ðB11Þ 021007-19 KAWABATA, NUMASAWA, and RYU PHYS. REV. X 13, 021007 (2023) where Q is an L × N matrix satisfying Q†Q ¼ 1 and R is an upper triangular matrix. The L × N matrix Uðt þ ΔtÞ is obtained as Uðt þ ΔtÞ ¼ Q: ðB12Þ In our numerical calculations, we choose Δt ¼ 0.05 for J ¼ 1.0. This numerical method is applicable even in the presence of spatial or temporal disorder. A similar numeri- cal method was used to investigate the open quantum dynamics of monitored free fermions [17,23]. The dynamics of a spinful system including the sym- plectic Hatano-Nelson model in Eq. (18) can also be calculated in a similar manner. In the spinful case, the state is represented as (cid:2)XL X YN (cid:3) jψðtÞi ¼ UlsnðtÞˆc† ls jvaci; ðB13Þ n¼1 l¼1 s¼↑;↓ where s describes the spin degree of freedom, and the isometry U is now the 2L × N matrix satisfying U†U ¼ 1. From U, the 2L × 2L correlation matrix C is obtained as Cis;js0ðtÞ ≔ hψðtÞjˆc† is ˆcjs0jψðtÞi ¼ ½UðtÞU†ðtÞ(cid:2)js0;is: ðB14Þ APPENDIX C: DIFFERENT INITIAL CONDITIONS We provide additional numerical results on the critical behavior for different initial conditions. We prepare the initial state as the fully polarized state, in the symplectic Hatano-Nelson model jψ 0i ¼ (cid:3) ˆc† l;↑ jvaci; (cid:2)YL l¼1 ðC1Þ and obtain the effective central charge from the logarithmic scaling of the steady-state entanglement entropy for both open and periodic boundary conditions (Fig. 16). The obtained effective central charges are consistent with those for the different initial state in Eq. (24). We also prepare the initial state as jψ 0i ¼ (cid:2)YL=4 l¼1 (cid:3) 4l−3;↑ ˆc† ˆc† 4l−3;↓ jvaci; ðC2Þ the which has the different particle number. Under open boundary conditions, the effective central charges behave differently for jγj ≪ J, in which the universal behavior should not be expected because of a significant (a) (b) FIG. 16. Effective central charge c of the symplectic Hatano- Nelson model (L ¼ 100, J ¼ 1.0) at the critical point (γ ¼ Δ) under the (a) open boundary conditions and (b) periodic boun- dary conditions. For each γ ¼ Δ, the effective central charge c is obtained from the logarithmic scaling of the steady-state entan- glement entropy for the initial states in Eq. (24) (red dots), Eq. (C1) (blue dots), and Eq. (C2) (green dots). The black dashed lines are (a) c ∝ γ−2=3 and (b) c ¼ 2. crossover between the unitary and nonunitary critical points. For jγj ≃ J, on the other hand, the power-law scaling c ∝ γ−2=3 in Eq. (31) appears. Under the periodic boundary conditions, the effective central charge is obtained as c ≃ 2 and hence consistent with those for the different initial conditions. APPENDIX D: DIAGONALIZATION OF LIOUVILLIANS We exactly solve the Liouvillian described by Eqs. (58), (59), and (60) in the single-particle Hilbert space [131]. First, for the periodic boundary conditions, we have Ljlihlj ¼ 1 2 ½ðJ þ γÞjl þ 1ihl þ 1j þ ðJ − γÞjl − 1ihl − 1j(cid:2) − Jjlihlj; ðD1Þ for l ¼ 1; 2; …; L. Here, jli ≔ ˆc† l jvaci is the single-particle state at site l, and we have j0i ¼ jLi and jL þ 1i ¼ j1i owing to the periodic boundary conditions. Notably, Eq. (D1) is formulated in the subspace spanned solely by the diagonal states fj1ih1j; j2ih2j; …; jLihLjg. The matrix representation of L in this subspace coincides with the single-particle matrix of the Hatano-Nelson model in Eq. (1) with periodic boundaries. Therefore, the eigenval- ues of L are 1 2 ½ðJ þ γÞe−ik þ ðJ − γÞeik(cid:2) − J ¼ Jðcos k − 1Þ − iγ sin k; ðD2Þ and the corresponding eigenstates are the plane waves, 1 L XL l¼1 eikljlihlj; ðD3Þ 021007-20 ENTANGLEMENT PHASE TRANSITION INDUCED … PHYS. REV. X 13, 021007 (2023) with k ∈ f0; 2π=L; 4π=L; …; 2πðL − 1Þ=Lg. Thus, the steady state, which is the zero mode of L, is given as the plane wave with zero momentum k ¼ 0: ˆρs ¼ 1 L XL l¼1 jlihlj: ðD4Þ The other eigenstates superposed by off-diagonal states do not contribute to the steady state [131]. For the open boundary conditions, on the other hand, the Liouvillian exhibits the skin effect in a similar manner to the Hatano-Nelson model. We still have Eq. (D1) for the bulk l ¼ 2; 3; …; L − 1. At the boundaries, we have Lj1ih1j ¼ J þ γ 2 j2ih2j − J þ γ 2 j1ih1j; ðD5Þ LjLihLj ¼ J − γ 2 jL − 1ihL − 1j − J − γ 2 jLihLj: ðD6Þ Because of the different boundary conditions, the steady state of L is now given as ˆρs ∝ XL l¼1 rljlihlj; ðD7Þ with r ≔ ðJ þ γÞ=ðJ − γÞ. It is also notable that the relaxation process speeds up for the larger asymmetry γ. The relaxation time τ subject to the Liouvillian skin effect is given as τ ≃ L ξΔ ; ðD8Þ where ξ is the localization length of the skin mode, and Δ is the Liouvillian gap [131]. From ξ ¼ 1=log r and Δ ¼ J − for Eq. (58), we have ffiffiffiffiffiffiffiffiffiffiffiffiffiffi J2 − γ2 p τ ≃ L ffiffiffiffiffiffiffiffiffiffiffiffiffiffi p J2 − γ2 J − log J þ γ J − γ : ðD9Þ This is a decreasing function of 0 ≤ γ ≤ J and consistent with the numerical results in Fig. 15. [1] A. Polkovnikov, K. Sengupta, A. Silva, and M. Vengalattore, Colloquium: Nonequilibrium Dynamics of Closed Interacting Quantum Systems, Rev. Mod. Phys. 83, 863 (2011). [2] J. Eisert, M. Friesdorf, and C. Gogolin, Quantum Many- Body Systems Out of Equilibrium, Nat. Phys. 11, 124 (2015). [3] R. Nandkishore and D. A. Huse, Many-Body Localization and Thermalization in Quantum Statistical Mechanics, Annu. Rev. Condens. Matter Phys. 6, 15 (2015). [4] L. D’Alessio, Y. Kafri, A. Polkovnikov, and M. Rigol, From Quantum Chaos and Eigenstate Thermalization to Statistical Mechanics and Thermodynamics, Adv. Phys. 65, 239 (2016). [5] S. Trotzky, Y.-A. Chen, A. Flesch, I. P. McCulloch, U. Schollwöck, J. Eisert, and I. Bloch, Probing the Relaxation towards Equilibrium in an Isolated Strongly Correlated One-Dimensional Bose Gas, Nat. Phys. 8, 325 (2012). [6] M. Gring, M. Kuhnert, T. Langen, T. Kitagawa, B. Rauer, M. Schreitl, I. Mazets, D. A. Smith, E. Demler, and J. Schmiedmayer, Relaxation and Prethermalization in an Isolated Quantum System, Science 337, 1318 (2012). [7] A. M. Kaufman, M. E. Tai, A. Lukin, M. Rispoli, R. Schittko, P. M. Preiss, and M. Greiner, Quantum Thermal- ization through Entanglement in an Isolated Many-Body System, Science 353, 794 (2016). [8] J. Smith, A. Lee, P. Richerme, B. Beyenhuis, P. W. Heiss, P. Hauke, M. Heyl, D. A. Huse, and C. Monroe, Many- Body Localization in a Quantum Simulator with Program- mable Random Disorder, Nat. Phys. 12, 907 (2016). [9] P. Calabrese and J. Cardy, Evolution of Entanglement Entropy in One-Dimensional Systems, J. Stat. Phys. (2005) P04010; Time Dependence of Correlation Functions Following a Quantum Quench, Phys. Rev. Lett. 96, 136801 (2006). [10] H. Kim and D. A. Huse, Ballistic Spreading of Entangle- ment in a Diffusive Nonintegrable System, Phys. Rev. Lett. 111, 127205 (2013). [11] A. Nahum, J. Ruhman, S. Vijay, and J. Haah, Quantum Entanglement Growth under Random Unitary Dynamics, Phys. Rev. X 7, 031016 (2017). [12] A. Chan, R. M. Nandkishore, M. Pretko, and G. Smith, Unitary-Projective Entanglement Dynamics, Phys. Rev. B 99, 224307 (2019). [13] B. Skinner, J. Ruhman, and A. Nahum, Measurement- Induced Phase Transitions in the Dynamics of Entangle- ment, Phys. Rev. X 9, 031009 (2019). [14] Y. Li, X. Chen, and M. P. A. Fisher, Quantum Zeno Effect and the Many-Body Entanglement Transition, Phys. Rev. B 98, 205136 (2018); Measurement-Driven Entanglement Transition in Hybrid Quantum Circuits, Phys. Rev. B 100, 134306 (2019). [15] S. Choi, Y. Bao, X.-L. Qi, and E. Altman, Quantum Error Correction in Scrambling Dynamics and Measurement- Induced Phase Transition, Phys. Rev. Lett. 125, 030505 (2020); Y. Bao, S. Choi, and E. Altman, Theory of the Phase Transition in Random Unitary Circuits with Mea- surements, Phys. Rev. B 101, 104301 (2020). [16] M. J. Gullans and D. A. Huse, Dynamical Purification Phase Transition Induced by Quantum Measurements, Phys. Rev. X 10, 041020 (2020). [17] X. Cao, A. Tilloy, and A. D. Luca, Entanglement in a Fermion Chain under Continuous Monitoring, SciPost Phys. 7, 24 (2019). [18] C.-M. Jian, Y.-Z. You, R. Vasseur, and A. W. W. Ludwig, Measurement-Induced Criticality in Random Quantum Circuits, Phys. Rev. B 101, 104302 (2020). [19] A. Lavasani, Y. Alavirad, and M. Barkeshli, Measurement- in Induced Topological Entanglement Transitions 021007-21 KAWABATA, NUMASAWA, and RYU PHYS. REV. X 13, 021007 (2023) Symmetric Random Quantum Circuits, Nat. Phys. 17, 342 (2021). [20] S. Sang and T. H. Hsieh, Measurement-Protected Quan- tum Phases, Phys. Rev. Res. 3, 023200 (2021). [21] M. Ippoliti, M. J. Gullans, S. Gopalakrishnan, D. A. Huse, and V. Khemani, Entanglement Phase Transitions in Measurement-Only Dynamics, Phys. Rev. X 11, 011030 (2021). [22] Y. Fuji and Y. Ashida, Measurement-Induced Quantum Criticality under Continuous Monitoring, Phys. Rev. B 102, 054302 (2020). [23] O. Alberton, M. Buchhold, and S. Diehl, Entanglement Transition in a Monitored Free-Fermion Chain: From Extended Criticality to Area Law, Phys. Rev. Lett. 126, 170602 (2021). [24] M. Ippoliti and V. Khemani, Postselection-Free Entangle- ment Dynamics via Spacetime Duality, Phys. Rev. Lett. 126, 060501 (2021); M. Ippoliti, T. Rakovszky, and V. Khemani, Fractal, Logarithmic, and Volume-Law En- tangled Nonthermal Steady States via Spacetime Duality, Phys. Rev. X 12, 011045 (2022). [25] T.-C. Lu and T. Grover, Spacetime Duality between Localization Transitions and Measurement-Induced Tran- sitions, PRX Quantum 2, 040319 (2021). [26] J. Eisert, M. Cramer, and M. B. Plenio, Colloquium: Area Laws for the Entanglement Entropy, Rev. Mod. Phys. 82, 277 (2010). [27] V. V. Konotop, J. Yang, and D. A. Zezyulin, Nonlinear Waves in PT -Symmetric Systems, Rev. Mod. Phys. 88, 035002 (2016). [28] R. El-Ganainy, K. G. Makris, M. Khajavikhan, Z. H. Musslimani, S. Rotter, and D. N. Christodoulides, Non- Hermitian Physics and PT Symmetry, Nat. Phys. 14, 11 (2018). [29] K. G. Makris, R. El-Ganainy, D. N. Christodoulides, and Z. H. Musslimani, Beam Dynamics in PT Symmetric Optical Lattices, Phys. Rev. Lett. 100, 103904 (2008). [30] C. E. Rüter, K. G. Makris, R. El-Ganainy, D. N. Christodoulides, M. Segev, and D. Kip, Observation of Parity-Time Symmetry in Optics, Nat. Phys. 6, 192 (2010). [31] A. Regensburger, C. Bersch, M.-A. Miri, G. Onishchukov, D. N. Christodoulides, and U. Peschel, Parity-Time Syn- thetic Photonic Lattices, Nature (London) 488, 167 (2012). [32] A. Mostafazadeh, Spectral Singularities of Complex Scat- tering Potentials and Infinite Reflection and Transmission Coefficients at Real Energies, Phys. Rev. Lett. 102, 220402 (2009). [33] Z. Lin, H. Ramezani, T. Eichelkraut, T. Kottos, H. Cao, and D. N. Christodoulides, Unidirectional Invisibility Induced by PT -Symmetric Periodic Structures, Phys. Rev. Lett. 106, 213901 (2011). [34] L. Feng, Y.-L. Xu, W. S. Fegadolli, M.-H. Lu, J. E. B. Oliveira, V. R. Almeida, Y.-F. Chen, and A. Scherer, Experimental Demonstration of a Unidirectional Reflec- tionless Parity-Time Metamaterial at Optical Frequencies, Nat. Mater. 12, 108 (2013). [35] B. Peng, S. K. Özdemir, F. Lei, F. Monifi, M. Gianfreda, G. L. Long, S. Fan, F. Nori, C. M. Bender, and L. Yang, Parity-Time-Symmetric Whispering-Gallery Microcavities, Nat. Phys. 10, 394 (2014). [36] S. Longhi, PT -Symmetric Laser Absorber, Phys. Rev. A 82, 031801(R) (2010). [37] Y. D. Chong, L. Ge, and A. D. Stone, PT -Symmetry Breaking and Laser-Absorber Modes in Optical Scattering Systems, Phys. Rev. Lett. 106, 093902 (2011). [38] B. Peng, S. K. Özdemir, S. Rotter, H. Yilmaz, M. Liertzer, F. Monifi, C. M. Bender, F. Nori, and L. Yang, Loss- Induced Suppression and Revival of Lasing, Science 346, 328 (2014). [39] L. Feng, Z. J. Wong, R.-M. Ma, Y. Wang, and X. Zhang, Single-Mode Laser by Parity-Time Symmetry Breaking, Science 346, 972 (2014). [40] H. Hodaei, M.-A. Miri, M. Heinrich, D. N. and M. Khajavikhan, Parity-Time- Christodoulides, Symmetric Microring Lasers, Science 346, 975 (2014). [41] J. Wiersig, Enhancing the Sensitivity of Frequency and Energy Splitting Detection by Using Exceptional Points: Application to Microcavity Sensors for Single-Particle Detection, Phys. Rev. Lett. 112, 203901 (2014). [42] H. Hodaei, A. U. Hassan, S. Wittek, H. Garcia-Gracia, R. El-Ganainy, D. N. Christodoulides, and M. Khajavikhan, Enhanced Sensitivity at Higher-Order Exceptional Points, Nature (London) 548, 187 (2017). [43] W. Chen, S. K. Özdemir, G. Zhao, J. Wiersig, and L. Yang, Exceptional Points Enhance Sensing in an Optical Micro- cavity, Nature (London) 548, 192 (2017). [44] J. Dalibard, Y. Castin, and K. Mølmer, Wave-Function Approach to Dissipative Processes in Quantum Optics, Phys. Rev. Lett. 68, 580 (1992); K. Mølmer, Y. Castin, and J. Dalibard, Monte Carlo Wave-Function Method in Quantum Optics, J. Opt. Soc. Am. B 10, 524 (1993). [45] R. Dum, P. Zoller, and H. Ritsch, Monte Carlo Simulation of the Atomic Master Equation for Spontaneous Emission, Phys. Rev. A 45, 4879 (1992). [46] H. Carmichael, An Open Systems Approach to Quantum Optics (Springer, Berlin, 1993). [47] M. B. Plenio and P. L. Knight, The Quantum-Jump Approach to Dissipative Dynamics in Quantum Optics, Rev. Mod. Phys. 70, 101 (1998). [48] A. J. Daley, Quantum Trajectories and Open Many-Body Quantum Systems, Adv. Phys. 63, 77 (2014). [49] G. Gamow, Zur Quantentheorie des Atomkernes, Z. Phys. 51, 204 (1928). [50] H. Feshbach, C. E. Porter, and V. F. Weisskopf, Model for Nuclear Reactions with Neutrons, Phys. Rev. 96, 448 (1954); H. Feshbach, Unified Theory of Nuclear Reac- tions, Ann. Phys. (N.Y.) 5, 357 (1958); A Unified Theory of Nuclear Reactions. II, Ann. Phys. (N.Y.) 19, 287 (1962). [51] I. Rotter, A Non-Hermitian Hamilton Operator and the Physics of Open Quantum Systems, J. Phys. A 42, 153001 (2009). [52] N. Moiseyev, Non-Hermitian Quantum Mechanics (Cambridge University Press, Cambridge, England, 2011). [53] P. Peng, W. Cao, C. Shen, W. Qu, J. Wen, L. Jiang, and Y. Xiao, Anti-Parity-Time Symmetry with Flying Atoms, Nat. Phys. 12, 1139 (2016). [54] J. Li, A. K. Harter, J. Liu, L. de Melo, Y. N. Joglekar, and L. Luo, Observation of Parity-Time Symmetry Breaking Transitions in a Dissipative Floquet System of Ultracold Atoms, Nat. Commun. 10, 855 (2019). 021007-22 ENTANGLEMENT PHASE TRANSITION INDUCED … PHYS. REV. X 13, 021007 (2023) [55] Z. Ren, D. Liu, E. Zhao, C. He, K. K. Pak, J. Li, and G.-B. Jo, Chiral Control of Quantum States in Non-Hermitian Spin-Orbit-Coupled Fermions, Nat. Phys. 18, 385 (2022). [56] L. Xiao, X. Zhan, Z. H. Bian, K. K. Wang, X. Zhang, X. P. Wang, J. Li, K. Mochizuki, D. Kim, N. Kawakami, W. Yi, H. Obuse, B. C. Sanders, and P. Xue, Observation of Topological Edge States in Parity-Time-Symmetric Quan- tum Walks, Nat. Phys. 13, 1117 (2017). [57] K. Kawabata, Y. Ashida, and M. Ueda, Information Retrieval and Criticality in Parity-Time-Symmetric Sys- tems, Phys. Rev. Lett. 119, 190401 (2017); L. Xiao, K. Wang, X. Zhan, Z. Bian, K. Kawabata, M. Ueda, W. Yi, and P. Xue, Observation of Critical Phenomena in Parity- Time-Symmetric Quantum Dynamics, Phys. Rev. Lett. 123, 230401 (2019). [58] B. Dóra, M. Heyl, and R. Moessner, The Kibble-Zurek Mechanism at Exceptional Points, Nat. Commun. 10, 2254 (2019); L. Xiao, D. Qu, K. Wang, H.-W. Li, J.-Y. Dai, B. Dóra, M. Heyl, R. Moessner, W. Yi, and P. Xue, Non- Hermitian Kibble-Zurek Mechanism with Tunable Com- plexity in Single-Photon Interferometry, PRX Quantum 2, 020313 (2021). [59] F. E. Öztürk, T. Lappe, G. Hellmann, J. Schmitt, J. Klaers, F. Vewinger, J. Kroha, and M. Weitz, Observation of a Non-Hermitian Phase Transition in an Optical Quantum Gas, Science 372, 88 (2021). [60] T. Gao, E. Estrecho, K. Y. Bliokh, T. C. H. Liew, M. D. Fraser, S. Brodbeck, M. Kamp, C. Schneider, S. Höfling, Y. Yamamoto, F. Nori, Y. S. Kivshar, A. G. Truscott, R. G. Dall, and E. A. Ostrovskaya, Observation of Non- Hermitian Degeneracies in a Chaotic Exciton-Polariton Billiard, Nature (London) 526, 554 (2015). [61] Y. Wu, W. Liu, J. Geng, X. Song, X. Ye, C.-K. Duan, X. Rong, and J. Du, Observation of Parity-Time Symmetry Breaking in a Single-Spin System, Science 364, 878 (2019). [62] W. Zhang, X. Ouyang, X. Huang, X. Wang, H. Zhang, Y. Yu, X. Chang, Y. Liu, D.-L. Deng, and L.-M. Duan, Observation of Non-Hermitian Topology with Nonunitary Dynamics of Solid-State Spins, Phys. Rev. Lett. 127, 090501 (2021). [63] M. Naghiloo, N. Abbasi, Y. N. Joglekar, and K. W. Murch, the Exceptional Quantum State Tomography across Point in a Single Dissipative Qubit, Nat. Phys. 15, 1232 (2019). [64] C. M. Bender, D. C. Brody, H. F. Jones, and B. K. Meister, Faster than Hermitian Quantum Mechanics, Phys. Rev. Lett. 98, 040403 (2007). [65] D. C. Brody and E.-M. Graefe, Mixed-State Evolution in the Presence of Gain and Loss, Phys. Rev. Lett. 109, 230405 (2012). [66] Y. Ashida and M. Ueda, Full-Counting Many-Particle Dynamics: Nonlocal and Chiral Propagation of Correla- tions, Phys. Rev. Lett. 120, 185301 (2018). [67] B. Dóra and C. P. Moca, Quantum Quench in PT - Symmetric Luttinger Liquid, Phys. Rev. Lett. 124, 136802 (2020); A. Bácsi and B. Dóra, Dynamics of Entanglement after Exceptional Quantum Quench, Phys. Rev. B 103, 085137 (2021). [68] X. Chen, Y. Li, M. P. A. Fisher, and A. Lucas, Emergent Conformal Symmetry in Nonunitary Random Dynamics of Free Fermions, Phys. Rev. Res. 2, 033017 (2020). [69] A. Biella and M. Schiró, Many-Body Quantum Zeno Effect and Measurement-Induced Subradiance Transition, Quantum 5, 528 (2021); X. Turkeshi and M. Schiró, Entanglement and Correlation Spreading in Non- Hermitian Spin Chains, Phys. Rev. B 107, L020403 (2023). [70] S. Gopalakrishnan and M. J. Gullans, Entanglement and Purification Transitions in Non-Hermitian Quantum Me- chanics, Phys. Rev. Lett. 126, 170503 (2021). [71] S.-K. Jian, Z.-C. Yang, Z. Bi, and X. Chen, Yang-Lee Edge Singularity Triggered Entanglement Transition, Phys. Rev. B 104, L161107 (2021). [72] T. Orito and K.-I. Imura, Unusual Wave-Packet Spreading and Entanglement Dynamics in Non-Hermitian Disor- dered Many-Body Systems, Phys. Rev. B 105, 024303 (2022). [73] T. Kato, Perturbation Theory for Linear Operators (Springer, Berlin, 1966). [74] M. V. Berry, Physics of Nonhermitian Degeneracies, Czech. J. Phys. 54, 1039 (2004). [75] W. D. Heiss, The Physics of Exceptional Points, J. Phys. A 45, 444016 (2012). [76] C. N. Yang and T. D. Lee, Statistical Theory of Equations of State and Phase Transitions. I. Theory of Condensation, Phys. Rev. 87, 404 (1952); T. D. Lee and C. N. Yang, Statistical Theory of Equations of State and Phase Tran- sitions. II. Lattice Gas and Ising Model, Phys. Rev. 87, 410 (1952). [77] M. E. Fisher, Yang-Lee Edge Singularity and ϕ3 Field Theory, Phys. Rev. Lett. 40, 1610 (1978). [78] J. L. Cardy, Conformal Invariance and the Yang-Lee Edge Singularity in Two Dimensions, Phys. Rev. Lett. 54, 1354 (1985). [79] C. Itzykson and J.-M. Drouffe, Statistical Field Theory, Vol. I (Cambridge University Press, Cambridge, England, 1989). [80] C. M. Bender and S. Boettcher, Real Spectra in Non- Hermitian Hamiltonians Having PT Symmetry, Phys. Rev. Lett. 80, 5243 (1998). [81] C. M. Bender, Making Sense of Non-Hermitian Hamilto- nians, Rep. Prog. Phys. 70, 947 (2007). [82] T. Fukui and N. Kawakami, Breakdown of the Mott Insulator: Exact Solution of an Asymmetric Hubbard Model, Phys. Rev. B 58, 16051 (1998). [83] Y. Nakamura and N. Hatano, A Non-Hermitian Critical Point and the Correlation Length of Strongly Corre- lated Quantum Systems, J. Phys. Soc. Jpn. 75, 104001 (2006). [84] T. Oka and H. Aoki, Dielectric Breakdown in a Mott Insulator: Many-Body Schwinger-Landau-Zener Mecha- nism Studied with a Generalized Bethe Ansatz, Phys. Rev. B 81, 033103 (2010). [85] T. E. Lee and C.-K. Chan, Heralded Magnetism in Non- Hermitian Atomic Systems, Phys. Rev. X 4, 041001 (2014); T. E. Lee, F. Reiter, and N. Moiseyev, Entangle- ment and Spin Squeezing in Non-Hermitian Phase Tran- sitions, Phys. Rev. Lett. 113, 250401 (2014). 021007-23 KAWABATA, NUMASAWA, and RYU PHYS. REV. X 13, 021007 (2023) [86] Y. Ashida, S. Furukawa, and M. Ueda, Parity-Time- Symmetric Quantum Critical Phenomena, Nat. Commun. 8, 15791 (2017). [87] R. Couvreur, J. L. Jacobsen, and H. Saleur, Entanglement in Nonunitary Quantum Critical Spin Chains, Phys. Rev. Lett. 119, 040601 (2017). [88] M. Nakagawa, N. Kawakami, and M. Ueda, Non-Hermi- in Ultracold Alkaline-Earth Atoms, tian Kondo Effect Phys. Rev. Lett. 121, 203001 (2018). [89] L. Herviou, J. H. Bardarson, and N. Regnault, Defining a Bulk-Edge Correspondence for Non-Hermitian Hamilto- nians via Singular-Value Decomposition, Phys. Rev. A 99, 052118 (2019); L. Herviou, N. Regnault, and J. H. Bardarson, Entanglement Spectrum and Symmetries in Non-Hermitian Fermionic Non-Interacting Models, Sci- Post Phys. 7, 069 (2019). [90] K. Yamamoto, M. Nakagawa, K. Adachi, K. Takasan, M. Ueda, and N. Kawakami, Theory of Non-Hermitian Fer- mionic Superfluidity with a Complex-Valued Interaction, Phys. Rev. Lett. 123, 123601 (2019). [91] P.-Y. Chang, J.-S. You, X. Wen, and S. Ryu, Entanglement Spectrum and Entropy in Topological Non-Hermitian Systems and Nonunitary Conformal Field Theory, Phys. Rev. Res. 2, 033069 (2020); Y.-T. Tu, Y.-C. Tzeng, and P.-Y. Chang, R´enyi Entropies and Negative Central Charges in Non-Hermitian Quantum Systems, SciPost Phys. 12, 194 (2022). [92] C. H. Lee, Exceptional Bound States and Negative Entanglement Entropy, Phys. Rev. Lett. 128, 010402 (2022). [93] T. E. Lee, Anomalous Edge State in a Non-Hermitian Lattice, Phys. Rev. Lett. 116, 133903 (2016). [94] S. Yao and Z. Wang, Edge States and Topological Invariants of Non-Hermitian Systems, Phys. Rev. Lett. 121, 086803 (2018); S. Yao, F. Song, and Z. Wang, Non- Hermitian Chern Bands, Phys. Rev. Lett. 121, 136802 (2018). [95] F. K. Kunst, E. Edvardsson, J. C. Budich, and E. J. Bergholtz, Biorthogonal Bulk-Boundary Correspondence in Non-Hermitian Systems, Phys. Rev. Lett. 121, 026808 (2018). [96] M. S. Rudner and L. S. Levitov, Topological Transition in a Non-Hermitian Quantum Walk, Phys. Rev. Lett. 102, 065703 (2009). [97] M. Sato, K. Hasebe, K. Esaki, and M. Kohmoto, Time- Reversal Symmetry in Non-Hermitian Systems, Prog. Theor. Phys. 127, 937 (2012); K. Esaki, M. Sato, K. Hasebe, and M. Kohmoto, Edge States and Topological Phases in Non-Hermitian Systems, Phys. Rev. B 84, 205128 (2011). [98] Y. C. Hu and T. L. Hughes, Absence of Topological Insulator Phases in Non-Hermitian PT-Symmetric Ham- iltonians, Phys. Rev. B 84, 153101 (2011). [99] H. Schomerus, Topologically Protected Midgap States in Complex Photonic Lattices, Opt. Lett. 38, 1912 (2013). [100] S. Longhi, D. Gatti, and G. D. Valle, Robust Light Trans- port in Non-Hermitian Photonic Lattices, Sci. Rep. 5, 13376 (2015). [101] D. Leykam, K. Y. Bliokh, C. Huang, Y. D. Chong, and F. Nori, Edge Modes, Degeneracies, and Topological Numbers in Non-Hermitian Systems, Phys. Rev. Lett. 118, 040401 (2017). [102] Y. Xu, S.-T. Wang, and L.-M. Duan, Weyl Exceptional Rings in a Three-Dimensional Dissipative Cold Atomic Gas, Phys. Rev. Lett. 118, 045701 (2017). [103] H. Shen, B. Zhen, and L. Fu, Topological Band Theory for Non-Hermitian Hamiltonians, Phys. Rev. Lett. 120, 146402 (2018); V. Kozii and L. Fu, Non-Hermitian Topological Theory of Finite-Lifetime Quasiparticles: Prediction of Bulk Fermi Arc due to Exceptional Point, arXiv:1708.05841. [104] Z. Gong, Y. Ashida, K. Kawabata, K. Takasan, S. Higashikawa, and M. Ueda, Topological Phases of Non- Hermitian Systems, Phys. Rev. X 8, 031079 (2018); K. Kawabata, S. Higashikawa, Z. Gong, Y. Ashida, and M. Ueda, Topological Unification of Time-Reversal and Particle-Hole Symmetries in Non-Hermitian Physics, Nat. Commun. 10, 297 (2019). [105] K. Kawabata, K. Shiozaki, M. Ueda, and M. Sato, Symmetry and Topology in Non-Hermitian Physics, Phys. Rev. X 9, 041015 (2019). [106] H. Zhou and J. Y. Lee, Periodic Table for Topological Bands with Non-Hermitian Symmetries, Phys. Rev. B 99, 235112 (2019). [107] H.-G. Zirnstein, G. Refael, and B. Rosenow, Bulk- Boundary Correspondence for Non-Hermitian Hamilto- nians via Green Functions, Phys. Rev. Lett. 126, 216407 (2021). [108] D. S. Borgnia, A. J. Kruchkov, and R.-J. Slager, Non- Hermitian Boundary Modes and Topology, Phys. Rev. Lett. 124, 056802 (2020). [109] K. Kawabata, T. Bessho, and M. Sato, Classification of Exceptional Points and Non-Hermitian Topological Semi- metals, Phys. Rev. Lett. 123, 066405 (2019). [110] J. Y. Lee, J. Ahn, H. Zhou, and A. Vishwanath, Topologi- cal Correspondence between Hermitian and Non- Hermitian Systems: Anomalous Dynamics, Phys. Rev. Lett. 123, 206404 (2019). [111] K. Kawabata, K. Shiozaki, and S. Ryu, Topological Field Theory of Non-Hermitian Systems, Phys. Rev. Lett. 126, 216405 (2021). [112] E. J. Bergholtz, J. C. Budich, and F. K. Kunst, Exceptional Topology of Non-Hermitian Systems, Rev. Mod. Phys. 93, 015005 (2021). [113] Y. Xiong, Why Does Bulk Boundary Correspondence Fail in Some Non-Hermitian Topological Models?, J. Phys. Commun. 2, 035043 (2018). [114] V. M. Martinez Alvarez, J. E. Barrios Vargas, and L. E. F. Foa Torres, Non-Hermitian Robust Edge States in One Dimension: Anomalous Localization and Eigenspace Condensation at Exceptional Points, Phys. Rev. B 97, 121401(R) (2018). [115] A. McDonald, T. Pereg-Barnea, and A. A. Clerk, Phase- Dependent Chiral Transport and Effective Non-Hermitian Dynamics in a Bosonic Kitaev-Majorana Chain, Phys. Rev. X 8, 041031 (2018). [116] C. H. Lee and R. Thomale, Anatomy of Skin Modes and Topology in Non-Hermitian Systems, Phys. Rev. B 99, 201103(R) (2019). [117] T. Liu, Y.-R. Zhang, Q. Ai, Z. Gong, K. Kawabata, M. Ueda, and F. Nori, Second-Order Topological Phases in 021007-24 ENTANGLEMENT PHASE TRANSITION INDUCED … PHYS. REV. X 13, 021007 (2023) Non-Hermitian Systems, Phys. Rev. Lett. 122, 076801 (2019). [118] C. H. Lee, L. Li, and J. Gong, Hybrid Higher-Order Skin- Topological Modes in Nonreciprocal Systems, Phys. Rev. Lett. 123, 016805 (2019). [119] K. Yokomizo and S. Murakami, Non-Bloch Band Theory of Non-Hermitian Systems, Phys. Rev. Lett. 123, 066404 (2019). [120] N. Okuma and M. Sato, Quantum Anomaly, Non- Hermitian Skin Effects, and Entanglement Entropy in Open Systems, Phys. Rev. B 103, 085428 (2021). [121] K. Zhang, Z. Yang, and C. Fang, Correspondence between Winding Numbers and Skin Modes in Non-Hermitian Systems, Phys. Rev. Lett. 125, 126402 (2020). [122] N. Okuma, K. Kawabata, K. Shiozaki, and M. Sato, Topological Origin of Non-Hermitian Skin Effects, Phys. Rev. Lett. 124, 086801 (2020); K. Kawabata, N. Okuma, and M. Sato, Non-Bloch Band Theory of Non-Hermitian Hamiltonians in the Symplectic Class, Phys. Rev. B 101, 195147 (2020). [123] M. Brandenbourger, X. Locsin, E. Lerner, and C. Coulais, Non-Reciprocal Robotic Metamaterials, Nat. Commun. 10, 4608 (2019); A. Ghatak, M. Brandenbourger, J. van Wezel, and C. Coulais, Observation of Non-Hermitian Topology and Its Bulk-Edge Correspondence in an Active Mechanical Metamaterial, Proc. Natl. Acad. Sci. U.S.A. 117, 29561 (2020). [124] T. Helbig, T. Hofmann, S. Imhof, M. Abdelghany, T. Kiessling, L. W. Molenkamp, C. H. Lee, A. Szameit, and R. Thomale, Generalized Bulk- M. Greiter, Boundary Correspondence in Non-Hermitian Topolec- trical Circuits, Nat. Phys. 16, 747 (2020); T. Hofmann, T. Helbig, F. Schindler, N. Salgo, M. Brzezińska, M. Greiter, T. Kiessling, D. Wolf, A. Vollhardt, A. Kabaši, C. H. Lee, A. Bilušić, R. Thomale, and T. Neupert, Reciprocal Skin Effect and Its Realization in a Topolectrical Circuit, Phys. Rev. Res. 2, 023265 (2020). [125] X. Zhang, Y. Tian, J.-H. Jiang, M.-H. Lu, and Y.-F. Chen, Observation of Higher-Order Non-Hermitian Skin Effect, Nat. Commun. 12, 5377 (2021). [126] S. Weidemann, M. Kremer, T. Helbig, T. Hofmann, A. Stegmaier, M. Greiter, R. Thomale, and A. Szameit, Topological Funneling of Light, Science 368, 311 (2020). [127] L. S. Palacios, S. Tchoumakov, M. Guix, I. P. S. Sánchez, and A. G. Grushin, Guided Accumulation of Active Par- ticles by Topological Design of a Second-Order Skin Effect, Nat. Commun. 12, 4691 (2021). [128] L. Xiao, T. Deng, K. Wang, G. Zhu, Z. Wang, W. Yi, and P. Xue, Non-Hermitian Bulk-Boundary Correspondence in Quantum Dynamics, Nat. Phys. 16, 761 (2020). [129] W. Gou, T. Chen, D. Xie, T. Xiao, T.-S. Deng, B. Gadway, W. Yi, and B. Yan, Tunable Nonreciprocal Quantum Transport through a Dissipative Aharonov-Bohm Ring in Ultracold Atoms, Phys. Rev. Lett. 124, 070402 (2020); Q. Liang, D. Xie, Z. Dong, H. Li, H. Li, B. Gadway, W. Yi, and B. Yan, Dynamic Signatures of Non-Hermitian Skin Effect and Topology in Ultracold Atoms, Phys. Rev. Lett. 129, 070401 (2022). [130] F. Song, S. Yao, and Z. Wang, Non-Hermitian Skin Effect and Chiral Damping in Open Quantum Systems, Phys. Rev. Lett. 123, 170401 (2019). [131] T. Haga, M. Nakagawa, R. Hamazaki, and M. Ueda, Liouvillian Skin Effect: Slowing Down of Relaxation Processes without Gap Closing, Phys. Rev. Lett. 127, 070402 (2021). [132] C.-H. Liu, K. Zhang, Z. Yang, and S. Chen, Helical Damping and Dynamical Critical Skin Effect in Open Quantum Systems, Phys. Rev. Res. 2, 043167 (2020). [133] T. Mori and T. Shirai, Resolving a Discrepancy between Liouvillian Gap and Relaxation Time in Boundary- Dissipated Quantum Many-Body Systems, Phys. Rev. Lett. 125, 230604 (2020). [134] F. Yang, Q.-D. Jiang, and E. J. Bergholtz, Liouvillian Skin Effect in an Exactly Solvable Model, Phys. Rev. Res. 4, 023160 (2022). [135] N. Hatano and D. R. Nelson, Localization Transitions in Non-Hermitian Quantum Mechanics, Phys. Rev. Lett. 77, 570 (1996); Vortex Pinning and Non-Hermitian Quantum Mechanics, Phys. Rev. B 56, 8651 (1997). [136] P. A. Lee and T. V. Ramakrishnan, Disordered Electronic Systems, Rev. Mod. Phys. 57, 287 (1985). [137] F. Evers and A. D. Mirlin, Anderson Transitions, Rev. Mod. Phys. 80, 1355 (2008). [138] K. Kawabata and S. Ryu, Nonunitary Scaling Theory of Non-Hermitian Localization, Phys. Rev. Lett. 126, 166801 (2021). [139] L. N. Trefethen and M. Embree, Spectra and Pseudospec- tra (Princeton University Press, Princeton, NJ, 2005). [140] H. Watanabe, A Proof of the Bloch Theorem for Lattice Models, J. Stat. Phys. 177, 717 (2019). [141] M. Fagotti and P. Calabrese, Evolution of Entanglement Entropy Following a Quantum Quench: Analytic Results for the XY Chain in a Transverse Magnetic Field, Phys. Rev. A 78, 010306(R) (2008). [142] P. Calabrese and J. Cardy, Entanglement Entropy and Quantum Field Theory, J. Stat. Phys. (2004) P06002. [143] B. Kramer, A. MacKinnon, T. Ohtsuki, and K. Slevin, Finite Size Scaling Analysis of the Anderson Transition, Int. J. Mod. Phys. B 24, 1841 (2010). [144] N. Goldenfeld, Lectures on Phase Transitions and the Renormalization Group (Westview Press, Boulder, CO, 1992). [145] J. Cardy, Scaling and Renormalization in Statistical Physics (Cambridge University Press, Cambridge, England, 1996). [146] S. Sachdev, Quantum Phase Transitions (Cambridge University Press, Cambridge, England, 1999). [147] L. Li, C. H. Lee, S. Mu, and J. Gong, Critical Non- Hermitian Skin Effect, Nat. Commun. 11, 5491 (2020). [148] K. Yokomizo and S. Murakami, Scaling Rule for the Critical Non-Hermitian Skin Effect, Phys. Rev. B 104, 165117 (2021). [149] H.-P. Breuer and F. Petruccione, The Theory of Open Quantum Systems (Oxford University Press, Oxford, 2007). [150] A. Rivas and S. F. Huelga, Open Quantum Systems (Springer, Berlin, 2012). [151] G. Kells, D. Meidan, and A. Romito, Topological Tran- sitions in Weakly Monitored Free Fermions, SciPost Phys. 14, 031 (2023). 021007-25 KAWABATA, NUMASAWA, and RYU PHYS. REV. X 13, 021007 (2023) [152] C. Fleckenstein, A. Zorzato, D. Varjas, E. J. Bergholtz, J. H. Bardarson, and A. Tiwari, Non-Hermitian Topology in Monitored Quantum Circuits, Phys. Rev. Res. 4, L032026 (2022). [153] B. Derrida, An Exactly Soluble Non-Equilibrium System: The Asymmetric Simple Exclusion Process, Phys. Rep. 301, 65 (1998). [154] K. B. Efetov, Directed Quantum Chaos, Phys. Rev. Lett. 79, 491 (1997). [155] J. Feinberg and A. Zee, Non-Hermitian Random Matrix Theory: Method of Hermitian Reduction, Nucl. Phys. B504, 579 (1997). [156] P. W. Brouwer, P. G. Silvestrov, and C. W. J. Beenakker, Theory of Directed Localization in One Dimension, Phys. Rev. B 56, R4333 (1997). [157] R. Hamazaki, K. Kawabata, and M. Ueda, Non-Hermitian Many-Body Localization, Phys. Rev. Lett. 123, 090603 (2019). [158] S. Longhi, Topological Phase Transition in Non- Hermitian Quasicrystals, Phys. Rev. Lett. 122, 237601 (2019). [159] A. F. Tzortzakakis, K. G. Makris, and E. N. Economou, Non-Hermitian Disorder in Two-Dimensional Optical Lattices, Phys. Rev. B 101, 014202 (2020). [160] Y. Huang and B. I. Shklovskii, Anderson Transition in Three-Dimensional Systems with Non-Hermitian Dis- order, Phys. Rev. B 101, 014204 (2020). [161] X. Luo, T. Ohtsuki, and R. Shindou, Universality Classes of the Anderson Transitions Driven by Non-Hermitian Disorder, Phys. Rev. Lett. 126, 090402 (2021); Transfer Matrix Study of the Anderson Transition in Non-Hermitian Systems, Phys. Rev. B 104, 104203 (2021); X. Luo, Z. Xiao, K. Kawabata, T. Ohtsuki, and R. Shindou, Unifying the Anderson Transitions in Hermitian and Non-Hermitian Systems, Phys. Rev. Res. 4, L022035 (2022). [162] F. Terrier and F. K. Kunst, Dissipative Analog of Four- Dimensional Quantum Hall Physics, Phys. Rev. Res. 2, 023364 (2020). [163] T. Bessho and M. Sato, Nielsen-Ninomiya Theorem with Bulk Topology: Duality in Floquet and Non-Hermitian Systems, Phys. Rev. Lett. 127, 196404 (2021). [164] M. M. Denner, A. Skurativska, F. Schindler, M. H. Fischer, R. Thomale, T. Bzdušek, and T. Neupert, Exceptional Topological Insulators, Nat. Commun. 12, 5681 (2021). [165] R. Okugawa, R. Takahashi, and K. Yokomizo, Second- Order Topological Non-Hermitian Skin Effects, Phys. Rev. B 102, 241202(R) (2020). [166] K. Kawabata, M. Sato, and K. Shiozaki, Higher-Order Non- Hermitian Skin Effect, Phys. Rev. B 102, 205118 (2020). [167] X.-Q. Sun, P. Zhu, and T. L. Hughes, Geometric Response and Disclination-Induced Skin Effects in Non-Hermitian Systems, Phys. Rev. Lett. 127, 066401 (2021). [168] F. Schindler and A. Prem, Dislocation Non-Hermitian Skin Effect, Phys. Rev. B 104, L161106 (2021). [169] B. A. Bhargava, I. C. Fulga, J. van den Brink, and A. G. Moghaddam, Non-Hermitian Skin Effect of Dislocations and Its Topological Origin, Phys. Rev. B 104, L241402 (2021). [170] C.-X. Guo, X.-R. Wang, C. Wang, and S.-P. Kou, Non- Hermitian Dynamic Strings and Anomalous Topological Degeneracy on a Non-Hermitian Toric-Code Model with Parity-Time Symmetry, Phys. Rev. B 101, 144439 (2020). [171] T. Yoshida, K. Kudo, and Y. Hatsugai, Non-Hermitian Fractional Quantum Hall States, Sci. Rep. 9, 16895 (2019). [172] S. Mu, C. H. Lee, L. Li, and J. Gong, Emergent Fermi Surface in a Many-Body Non-Hermitian Fermionic Chain, Phys. Rev. B 102, 081115(R) (2020). [173] W. Xi, Z.-H. Zhang, Z.-C. Gu, and W.-Q. Chen, Classi- fication of Topological Phases in One Dimensional Inter- acting Non-Hermitian Systems and Emergent Unitarity, Sci. Bull. 66, 1731 (2021). [174] E. Lee, H. Lee, and B.-J. Yang, Many-Body Approach to Non-Hermitian Physics in Fermionic Systems, Phys. Rev. B 101, 121109(R) (2020). [175] N. Matsumoto, K. Kawabata, Y. Ashida, S. Furukawa, and M. Ueda, Continuous Phase Transition without Gap Closing in Non-Hermitian Quantum Many-Body Systems, Phys. Rev. Lett. 125, 260601 (2020). [176] T. Liu, J. J. He, T. Yoshida, Z.-L. Xiang, and F. Nori, Non-Hermitian Topological Mott Insulators in One- Dimensional Fermionic Superlattices, Phys. Rev. B 102, 235151 (2020). [177] K. Yang, S. C. Morampudi, and E. J. Bergholtz, Excep- tional Spin Liquids from Couplings to the Environment, Phys. Rev. Lett. 126, 077201 (2021). [178] M. Žnidarič, T. Prosen, and P. Prelovšek, Many-Body Localization in the Heisenberg XXZ Magnet in a Random Field, Phys. Rev. B 77, 064426 (2008). [179] J. H. Bardarson, F. Pollmann, and J. E. Moore, Unbounded Growth of Entanglement in Models of Many-Body Locali- zation, Phys. Rev. Lett. 109, 017202 (2012). [180] M. Serbyn, Z. Papić, and D. A. Abanin, Universal Slow Growth of Entanglement in Interacting Strongly Disor- dered Systems, Phys. Rev. Lett. 110, 260601 (2013). 021007-26
null
10.3390_biom13060952.pdf
Data Availability Statement: The data supporting this study are available from the corresponding authors upon reasonable request.
Data Availability Statement: The data supporting this study are available from the corresponding authors upon reasonable request.
Article Ca2+ Influx through TRPC Channels Is Regulated by Homocysteine–Copper Complexes Gui-Lan Chen 1, Bo Zeng 1 and Shang-Zhong Xu 1,2,* , Hongni Jiang 1, Nikoleta Daskoulidou 1 , Rahul Saurabh 1, Rumbidzai J. Chitando 1 1 Centre for Atherothrombosis and Metabolic Disease, Hull York Medical School, University of Hull, Hull HU6 7RX, UK; chenguilan@swmu.edu.cn (G.-L.C.); zengbo@swmu.edu.cn (B.Z.) 2 Diabetes, Endocrinology and Metabolism, Hull York Medical School, University of Hull, Hull HU6 7RX, UK * Correspondence: sam.xu@hyms.ac.uk; Tel.: +44-1482-465372 Abstract: An elevated level of circulating homocysteine (Hcy) has been regarded as an independent risk factor for cardiovascular disease; however, the clinical benefit of Hcy lowering-therapy is not satisfying. To explore potential unrevealed mechanisms, we investigated the roles of Ca2+ influx through TRPC channels and regulation by Hcy–copper complexes. Using primary cultured human aortic endothelial cells and HEK-293 T-REx cells with inducible TRPC gene expression, we found that Hcy increased the Ca2+ influx in vascular endothelial cells through the activation of TRPC4 and TRPC5. The activity of TRPC4 and TRPC5 was regulated by extracellular divalent copper (Cu2+) and Hcy. Hcy prevented channel activation by divalent copper, but monovalent copper (Cu+) had no effect on the TRPC channels. The glutamic acids (E542/E543) and the cysteine residue (C554) in the extracellular pore region of the TRPC4 channel mediated the effect of Hcy–copper complexes. The interaction of Hcy–copper significantly regulated endothelial proliferation, migration, and angiogenesis. Our results suggest that Hcy–copper complexes function as a new pair of endogenous regulators for TRPC channel activity. This finding gives a new understanding of the pathogenesis of hyperhomocysteinemia and may explain the unsatisfying clinical outcome of Hcy-lowering therapy and the potential benefit of copper-chelating therapy. Keywords: homocysteine; calcium channel; TRPC; TRPM2; copper; endothelial cells; angiogenesis; 2-aminoethoxydiphenyl borate 1. Introduction Cardiovascular disease (CVD) is the leading cause of death in developed nations and is increasing rapidly in developing countries. The well-described risk factors include high blood pressure, dyslipidemia, smoking, diabetes mellitus, obesity, and new independent risk factors, such as C-reactive protein, lipoprotein (a), fibrinogen, and homocysteine (Hcy). The association between elevated Hcy levels and atherosclerosis was first demonstrated in patients with hyperhomocysteinemia in 1969 [1]; however, the importance of Hcy as a risk factor has been especially acknowledged during the last two decades in that even a mild or moderate increase in Hcy level (>15 µmol/L) in serum or plasma is closely associated with the morbidity and mortality of coronary heart diseases [2–6], stroke [7,8], peripheral vascular disease [9], venous thrombosis [10], dementia or Alzheimer’s disease [11], nerve degeneration [12], diabetes [13], osteoporotic fractures [14], end-stage renal disease [15], and other conditions, such as adverse pregnancy outcome (early abortion, placental vas- culopathy, and birth defects) [16] and liver fibrosis [17]. In patients with genetic enzyme defects including cystathionine β-synthase (CBS), methylenetetrahydrofolate reductase (MTHFR), and methionine synthase (MS) in the Hcy metabolic pathway, the concentration of Hcy is much higher and accompanied with more severe cardiovascular damage [8,18]. The MTHFR (T677C point mutation) variant is the most common enzyme defect associated Citation: Chen, G.-L.; Zeng, B.; Jiang, H.; Daskoulidou, N.; Saurabh, R.; Chitando, R.J.; Xu, S.-Z. Ca2+ Influx through TRPC Channels Is Regulated by Homocysteine–Copper Complexes. Biomolecules 2023, 13, 952. https://doi.org/10.3390/ biom13060952 Academic Editor: Fabrice Antigny Received: 25 April 2023 Revised: 15 May 2023 Accepted: 17 May 2023 Published: 6 June 2023 Copyright: © 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/). Biomolecules 2023, 13, 952. https://doi.org/10.3390/biom13060952 https://www.mdpi.com/journal/biomolecules biomolecules Biomolecules 2023, 13, 952 2 of 16 with high Hcy and its prevalence is 5~15% in Caucasian and Asian populations. The mech- anisms of how Hcy causes diseases or becomes a risk for diseases are still unknown [19,20]; in particular, the intervention for lowering plasma Hcy levels in patients did not show any preventive effects against cardiovascular diseases [21,22], suggesting unrecognised mechanisms or interactions with Hcy may exist in vivo. Since Hcy is involved in the patho- genesis of many diseases and is associated with all-cause mortality [23], it is reasonable to hypothesise that Hcy may target some ubiquitously expressed proteins or key signalling molecules in the body. Calcium is a key signalling messenger in the cell and several studies have suggested that Hcy may interfere with Ca2+ signalling pathways. For example, Ca2+ influx and intracellular Ca2+ release were enhanced by Hcy [24], and the ligand-gated Ca2+ channel NMDA receptor was stimulated by Hcy [25]. Interestingly, it has been shown that the up-regulation of Ca2+ permeable channels, such as TRPC1 and TRPC5, is related to vascular neointimal growth and cell mobility [26,27], while neointimal growth was also observed in the blood vessels from patients with hyperhomocysteinemia [1]. TRPC channels are ubiqui- tously expressed in the cardiovascular system and mediate the common pathway of Ca2+ entry via G-protein coupled receptor activation and/or the depletion of the endoplasmic reticulum (ER) Ca2+ store [28,29]. Therefore, we hypothesised that TRPC channels could be involved in the pathophysiology of hyperhomocysteinemia. On the other hand, the correlation between Hcy and copper in cardiovascular disease has been demonstrated in clinical surveys [30–33], and copper-lowering therapy with a chelator could be beneficial for cardiac hypertrophy [34]. We, therefore, aimed to investigate the effects of Hcy on TRPC channels and its regulatory mechanisms with copper ions in causing endothelial dysfunction and subsequent atherogenicity. 2. Materials and Methods 2.1. Cell Culture and Transfection Human TRPC4α (NM_016179), TRPC4β1 (NM_001135955, but the β1 isoform was cloned from the endothelial cell with one glutamic acid deletion at E785), and TRPC5 (AF054568) in the tetracycline-regulatory vector pcDNA4/TO (Invitrogen, Paisley, UK) were transfected into HEK-293 T-REx cells using the LipofectamineTM 2000 transfection reagent (Invitrogen, Paisley, UK). TRPC4 was tagged with an enhanced yellow fluorescent protein (EYFP) at the N-terminus. Expression was induced by 1 µg·mL−1 tetracycline for 48–72 h before recording. The non-induced cells without the addition of tetracycline were used as a control. Cells were grown in DMEM-F12 medium (Invitrogen, Paisley, UK) containing 10% foetal calf serum (FCS), 100 units·mL−1 penicillin, and 100 µg·mL−1 streptomycin at 37 ◦C under 95% air and 5% CO2. Cells were seeded on coverslips prior to experiments. Human aortic endothelial cells (HAECs) were purchased from PromoCell (Heidel- berg, Germany) and cultured in an endothelial cell growth medium as we described previously [35,36]. The medium was supplemented with 2% foetal calf serum, 5.0 µg·L−1 epidermal growth factor, 0.5 µg·L−1 vascular endothelial growth factor, 10 µg·L−1 basic fibroblast factor, 20 µg·L−1 R3 IGF-1, and 22.5 mg·L−1 heparin. Cells in passages 2 to 4 were used in the experiment to avoid age-dependent variations. 2.2. Electrophysiological Recordings and Ca2+ Measurements A whole-cell clamp was performed at room temperature (23–26 ◦C) as described before [37,38]. Briefly, the signal was amplified with an Axopatch B200 amplifier and controlled with pClamp software 10. A 1 s ramp voltage protocol from −100 mV to +100 mV was applied at a frequency of 0.2 Hz from a holding potential of 0 mV. Signals were sampled at 3 kHz and filtered at 1 kHz. A glass microelectrode with a resistance of 3–5 MΩ was used. The 200 nM Ca2+ buffered pipette solution contained 115 CsCl, 10 EGTA, 2 MgCl2, 10 HEPES, and 5.7 CaCl2 in mM. The pH was adjusted to 7.2 with CsOH and the osmolarity was adjusted to ~290 mOsm with mannitol. The calculated free Ca2+ was Biomolecules 2023, 13, 952 3 of 16 200 nM using EQCAL (Biosoft, Cambridge, UK). The standard bath solution contained (mM): 130 NaCl, 5 KCl, 8 D-glucose, 10 HEPES, 1.2 MgCl2, and 1.5 CaCl2. The pH was adjusted to 7.4 with NaOH. For excised patch recordings, the procedures were similar to our previous reports [39,40]. Intracellular Ca2+ was measured using a cuvette-based system as we described previously [35,41]. Briefly, HAECs were loaded with Fluo3-AM (5 µM) in a Ca2+ free standard bath solution (130 NaCl, 5 KCl, 8 D-glucose, 10 HEPES, and 1.2 MgCl2 in mM), then washed and resuspended in the standard bath solution. A total volume of 2 mL of standard bath solution with suspended cells was pipetted into a cuvette and the fluores- cence was measured using a Perkin–Elmer LS50B fluorimeter. All electrophysiological recordings and Ca2+ measurements were performed at room temperature (25 ◦C). 2.3. RT-PCR Total RNA was extracted from the cultured endothelial cells using TRI Reagent (Sigma- Aldrich, Poole, UK) and reverse transcribed with the Moloney murine leukaemia virus (M-MLV) reverse transcriptase using random primers (Promega, Southampton, UK). The PCR primer sequences used in this study and the detailed procedures were described in our previous report [42]. PCR products were confirmed by 2% agarose gel electrophoresis or direct sequencing. 2.4. Cell Proliferation, Migration, and Angiogenesis Assays Endothelial cells were grown to confluence in 24-well plates in an endothelial cell medium. Cell proliferation was assayed by a WST-1 kit (Roche) as we reported [42,43]. For the cell migration assay, a linear scrape of ~0.3 mm width was made through a pipette tip [26]. The cells were cultured in an endothelial cell medium with or without Hcy. After 24 h of culture, the cells were fixed with 4% paraformaldehyde, and cells across the edge of the wound were counted. For the angiogenesis experiment, bovine skin collagen (Sigma, Hertfordshire, UK) was diluted to 1.5 mg/mL with extracellular matrix (ECM) (Sigma) at 2–8 ◦C as a working solution. The pH and osmolarity were adjusted by 1 M NaOH and 10× phosphate-buffered saline, respectively. Human vascular endothelial growth factor (Sigma, UK) was added to a final concentration of 20 ng/mL. Collagen working solution at a volume of 120 µL was added to each well of a 48-well plate and allowed to gelatinise for 30 min at 37 ◦C. EA.hy926 cells were resuspended in the ECM solution and added to each well at a volume of 300 µL (~3 × 104 cells/well) and incubated at 37 ◦C for 30 min under 95% air and 5% CO2. After 24 h of culture with Hcy or the vehicle, cells were fixed with 4% paraformaldehyde, stained with 0.025% crystal violet, and photographed. The angiogenesis score was calculated by a semi-quantitative method as reported previously [44]. The BD MatrigelTM (BD Bioscience, Chester, UK) was also used to see the effects of Hcy and Cu2+ on endothelial cell tube formation. The angiogenesis was analysed with Wim Tube software (Wimasis, Munich, Germany). 2.5. Reagents and Drugs All general salts and reagents were purchased from Sigma-Aldrich (Poole, UK). L- homocysteine, lanthanum chloride (La3+), CuSO4 (Cu2+), gadolinium chloride (Gd3+), 2-aminoethoxydiphenyl borate (2-APB), trypsin, thapsigargin (TG), D-(−)-2-amino-5- phosphonopentanoic acid (D-AP5), verapamil, A23187, (1,10-phenanthroline)bis (triphenylphosphine)copper(I) nitrate dichloromethane adduct, and foetal calf serum were purchased from Sigma-Aldrich. Matrigel was purchased from BD Biosciences (UK) and Fluo-3 AM from Invitrogen (Paisley, UK). Fluo-3 AM (5 mM), TG (1 mM), and 2-APB (100 mM) were made up as stock solutions in 100% dimethyl sulphoxide (DMSO). 2.6. Statistics Data are expressed as mean ± s.e.m. where n is the cell number for electrophysiological recordings and Ca2+ imaging. Data sets were compared using a paired t-test for the results Biomolecules 2023, 13, 952 4 of 16 before and after treatment, or the ANOVA Bonferroni’s post-hoc analysis for comparing more than two groups with significance indicated if p < 0.05. 3. Results 3.1. Ca2+ Influx Induced by Hcy in HAECs The effect of Hcy on Ca2+ influx was measured in the primary cultured HAECs using Fluo-3 AM Ca2+ dye. Hcy at 1–100 µM increased the intracellular [Ca2+]i, which accounted for 33.1 ± 1.1% of the amplitude of the Ca2+ signal induced by calcium ionophore A231872 (Figure 1A,B). Blocking the voltage-gated Ca2+ channels with verapamil or using 100 mM K+ in the bath solution (equal molar substitution of Na+) to clamp the membrane potential did not prevent the effect of Hcy (Figure 1C,D), suggesting that Hcy-induced Ca2+ increase is mediated by non-voltage gated Ca2+-permeable channels. We also examined the Ca2+ release using the sarco/endoplasmic reticulum Ca2+-ATPase (SERCA) inhibitor thapsigargin (TG). Depletion of the ER Ca2+ store showed no significant blocking effect on Hcy-induced intracellular Ca2+ increase (Figure 1E). Hcy has been reported to induce Ca2+ transient through NMDA receptor activation in cultured neurons [24], therefore, we tested the effect of Hcy in cells treated with the NMDA antagonist D-(−)-2-amino-5- phosphonopentanoic acid (D-AP5). D-AP5 at 50 µM was unable to prevent the Hcy-induced Ca2+ influx (Figure 1F), suggesting that other Ca2+ entry pathways exist in endothelial cells. These results suggest that Hcy increases Ca2+ influx mainly through non-voltage gated channels, rather than the Ca2+ release or NMDA receptors in vascular endothelial cells. Figure 1. Effect of Hcy on Ca2+ influx in HAECs. Ca2+ influx was measured using Fluo-3 AM. (A) Example of Hcy on Ca2+ influx. Hcy was added accumulatedly and followed by calcium ionophore A23187 (2 µM). (B) The mean ± s.e.m. for the effect of Hcy. (C) Effect of Hcy under the bath solution with 100 mM K+. (D) Response to Hcy after blocking the voltage-gated Ca2+ channel with 10 µM verapamil. (E) Thapsigargin (2 µM) was added to block the SERCA. (F) NMDA antagonist 5-AP (50 µM) added. The ANOVA test was used and n = 6–8 for each experiment. *** p < 0.001. Biomolecules 2023, 13, x FOR PEER REVIEW 4 of 17 2.6. Statistics Data are expressed as mean ± s.e.m. where n is the cell number for electrophysiolog-ical recordings and Ca2+ imaging. Data sets were compared using a paired t-test for the results before and after treatment, or the ANOVA Bonferroni’s post-hoc analysis for com-paring more than two groups with significance indicated if p < 0.05. 3. Results 3.1. Ca2+ Influx Induced by Hcy in HAECs The effect of Hcy on Ca2+ influx was measured in the primary cultured HAECs using Fluo-3 AM Ca2+ dye. Hcy at 1–100 µM increased the intracellular [Ca2+]i, which accounted for 33.1 ± 1.1% of the amplitude of the Ca2+ signal induced by calcium ionophore A231872 (Figure 1A,B). Blocking the voltage-gated Ca2+ channels with verapamil or using 100 mM K+ in the bath solution (equal molar substitution of Na+) to clamp the membrane potential did not prevent the effect of Hcy (Figure 1C,D), suggesting that Hcy-induced Ca2+ increase is mediated by non-voltage gated Ca2+-permeable channels. We also examined the Ca2+ release using the sarco/endoplasmic reticulum Ca2⁺-ATPase (SERCA) inhibitor thapsigar-gin (TG). Depletion of the ER Ca2+ store showed no significant blocking effect on Hcy-induced intracellular Ca2+ increase (Figure 1E). Hcy has been reported to induce Ca2+ tran-sient through NMDA receptor activation in cultured neurons [24], therefore, we tested the effect of Hcy in cells treated with the NMDA antagonist D-(−)-2-amino-5-phosphono-pentanoic acid (D-AP5). D-AP5 at 50 µM was unable to prevent the Hcy-induced Ca2+ influx (Figure 1F), suggesting that other Ca2+ entry pathways exist in endothelial cells. These results suggest that Hcy increases Ca2+ influx mainly through non-voltage gated channels, rather than the Ca2+ release or NMDA receptors in vascular endothelial cells. Figure 1. Effect of Hcy on Ca2+ influx in HAECs. Ca2+ influx was measured using Fluo-3 AM. (A) Example of Hcy on Ca2+ influx. Hcy was added accumulatedly and followed by calcium ionophore Biomolecules 2023, 13, 952 5 of 16 3.2. Hcy-Induced Ca2+ Influx through TRPC4 and TRPC5 Channels To explore which pathway is involved in Hcy-induced Ca2+ entry, we examined the expression and function of TRPC channels in endothelial cells. The mRNAs of TRPC1, 3, 4, and 6 were detected in the HAECs using RT-PCR. TRPC1 and TRPC4 were more abundant in HUVEC, but TRPC5 was low and TRPC3, TRPC6, and TRPC7 seemed to be absent in HUVEC (Figure 2A). The spliced isoforms of TRPC1E9del, TRPC4β1, and TRPC4ε1 were also identified in the HAECs using the primer sets we reported previously [42] (Figure 2B). Figure 2. Hcy-induced Ca2+ influx through TRPC4 and TRPC5 channels in endothelial cells. (A) mRNAs of TRPCs in vascular endothelial cells (HAECs and HUVECs). The plasmid cDNAs for TRPC3, 6, and 7 were used as positive controls. (B) Detection of TRPC1 and TRPC4 spliced variants in HAECs. The PCR primers and the corresponding size of amplicons were given in our previous reports [42]. (C) TRPC4 current recorded in HEK293 T-REx cells inducibly overexpressing TRPC4α channels and the effect of Hcy (100 µM). (D) Current for induced TRPC5 cells. (E) Non-induced T-REx cell as control. (F) The mean ± s.e.m. measured at ±80 mV after exposure to each compound. n = 5–6 for each group. *** p < 0.001 compared with La3+ treatment measured at ±80 mV. Using whole-cell patch recordings, the effects of Hcy on TRPC4 and TRPC5 currents were examined in the HEK293 T-REx cells inducibly expressing TRPC channels [38]. Lan- thanides (La3+ or Gd3+) were used as channel activators in our experiment as we used before [41,45]. After perfusion with Hcy, the currents of TRPC4 and TRPC5 were signifi- cantly stimulated (Figure 2C,D) while no effects were observed on the non-induced cells (Figure 2E,F), suggesting that Hcy induced Ca2+ influx via the activation of TRPC4 and TRPC5 channels. Biomolecules 2023, 13, x FOR PEER REVIEW 5 of 17 A23187 (2 µM). (B) The mean ± s.e.m. for the effect of Hcy. (C) Effect of Hcy under the bath solution with 100 mM K+. (D) Response to Hcy after blocking the voltage-gated Ca2+ channel with 10 µM verapamil. (E) Thapsigargin (2 µM) was added to block the SERCA. (F) NMDA antagonist 5-AP (50 µM) added. The ANOVA test was used and n = 6–8 for each experiment. *** p < 0.001. 3.2. Hcy-Induced Ca2+ Influx through TRPC4 and TRPC5 Channels To explore which pathway is involved in Hcy-induced Ca2+ entry, we examined the expression and function of TRPC channels in endothelial cells. The mRNAs of TRPC1, 3, 4, and 6 were detected in the HAECs using RT-PCR. TRPC1 and TRPC4 were more abun-dant in HUVEC, but TRPC5 was low and TRPC3, TRPC6, and TRPC7 seemed to be absent in HUVEC (Figure 2A). The spliced isoforms of TRPC1E9del, TRPC4β1, and TRPC4Ɛ1 were also identified in the HAECs using the primer sets we reported previously [42] (Figure 2B). Using whole-cell patch recordings, the effects of Hcy on TRPC4 and TRPC5 currents were examined in the HEK293 T-REx cells inducibly expressing TRPC channels [38]. Lan-thanides (La3+ or Gd3+) were used as channel activators in our experiment as we used be-fore [41,45]. After perfusion with Hcy, the currents of TRPC4 and TRPC5 were signifi-cantly stimulated (Figure 2C,D) while no effects were observed on the non-induced cells (Figure 2E,F), suggesting that Hcy induced Ca2+ influx via the activation of TRPC4 and TRPC5 channels. Figure 2. Hcy-induced Ca2+ influx through TRPC4 and TRPC5 channels in endothelial cells. (A) mRNAs of TRPCs in vascular endothelial cells (HAECs and HUVECs). The plasmid cDNAs for TRPC3, 6, and 7 were used as positive controls. (B) Detection of TRPC1 and TRPC4 spliced variants Biomolecules 2023, 13, 952 6 of 16 3.3. Activation of TRPC4 and TRPC5 by Divalent Cu2+ and the Interference by Hcy Hcy and copper are two important regulators of cellular oxidative stress and both are involved in atherogenicity, however, their mechanisms are unclear [30]. We found that divalent Cu2+ showed an initial transient inhibition and then a gradual activation of TRPC4α and TRPC5 currents after perfusion with 10 µM Cu2+ (Figure 3A,B). The current of TRPC4β1 was also activated by Cu2+ (Figure S1A). The EC50 of Cu2+ for TRPC4α channel activation was 6.8 µM (Figure S1B). The Cu2+-induced currents were also sensitive to the non-selective TRPC blocker 2-APB as the currents of TRPC4 and TRPC5 induced by lanthanides [41,45]. Interestingly, perfusion with Hcy (100 µM) completely prevented the TRPC4 and TRPC5 channel activation by Cu2+ (Figure 3C,D), suggesting that the interaction of Hcy and copper is critical for regulating TRPC channel activity. We also examined the interaction on TRPM2 channels, since the channel is expressed in endothelial cells and inhibited by Cu2+ [35,46]. Hcy had no significant effect on TRPM2, but it prevented the inhibitory effect of Cu2+ (Figure S2). These data indicate that the complexes of Hcy–copper or the charge of copper ions may be the determinant for their effects on ion channels. Figure 3. TRPC channel activated by Cu2+ and counteracted by Hcy. (A,B) Representative time course and IV curve for TRPC4 and TRPC5 activated by Cu2+. 2-APB (100 µM) as a control channel blocker. (C,D) TRPC4 and TRPC5 currents after perfusion with 100 µM Hcy, the addition of 10 µM Cu2+, and the washout of Hcy. (E) The mean ± s.e.m. data for the effect of Cu2+ (n = 6–8. *** p < 0.01). (F) The mean ± s.e.m. data for Hcy plus Cu2+ (n = 5–6. *** p < 0.001). 3.4. No Effect of Monovalent Cu+ on TRPC Channel To test the role of copper ion charges, we examined the effects of monovalent copper (I) compounds. As shown in Figure 4, the copper (I), (1,10-phenanthroline)bis(triphenylphosphine) copper (I) nitrate dichloromethane adduct, had no effect on TRPC4α and TRPC5 chan- nel activity, but the divalent Cu2+ activated them (Figure 4A–C). Similarly, no effects of the monovalent copper, copper (I) 1-butanethiolate), and copper (I) tetrakis(acetonitrile) copper(I) tetrafluoroborate) were observed on TRPC4α channels (Figure S3). These data suggest that the divalent copper ions are essential for TRPC channel activation, but there Biomolecules 2023, 13, x FOR PEER REVIEW 6 of 17 in HAECs. The PCR primers and the corresponding size of amplicons were given in our previous reports [42]. (C) TRPC4 current recorded in HEK293 T-REx cells inducibly overexpressing TRPC4α channels and the effect of Hcy (100 µM). (D) Current for induced TRPC5 cells. (E) Non-induced T-REx cell as control. (F) The mean ± s.e.m. measured at ±80 mV after exposure to each compound. n = 5–6 for each group. *** p < 0.001 compared with La3+ treatment measured at ±80 mV. 3.3. Activation of TRPC4 and TRPC5 by Divalent Cu2+ and the Interference by Hcy Hcy and copper are two important regulators of cellular oxidative stress and both are involved in atherogenicity, however, their mechanisms are unclear [30]. We found that divalent Cu2+ showed an initial transient inhibition and then a gradual activation of TRPC4α and TRPC5 currents after perfusion with 10 µM Cu2+ (Figure 3A,B). The current of TRPC4β1 was also activated by Cu2+ (Figure S1A). The EC50 of Cu2+ for TRPC4α channel activation was 6.8 µM (Figure S1B). The Cu2+-induced currents were also sensitive to the non-selective TRPC blocker 2-APB as the currents of TRPC4 and TRPC5 induced by lan-thanides [41,45]. Interestingly, perfusion with Hcy (100 µM) completely prevented the TRPC4 and TRPC5 channel activation by Cu2+ (Figure 3C,D), suggesting that the interac-tion of Hcy and copper is critical for regulating TRPC channel activity. We also examined the interaction on TRPM2 channels, since the channel is expressed in endothelial cells and inhibited by Cu2+[35,46]. Hcy had no significant effect on TRPM2, but it prevented the inhibitory effect of Cu2+ (Figure S2). These data indicate that the complexes of Hcy–copper or the charge of copper ions may be the determinant for their effects on ion channels. Figure 3. TRPC channel activated by Cu2+ and counteracted by Hcy. (A,B) Representative time course and IV curve for TRPC4 and TRPC5 activated by Cu2+. 2-APB (100 µM) as a control channel blocker. (C,D) TRPC4 and TRPC5 currents after perfusion with 100 µM Hcy, the addition of 10 µM Biomolecules 2023, 13, 952 7 of 16 are no effects for monovalent Cu+ ions. In addition, Se2+ with antioxidant properties had no effect on TRPC4α channels (Figure 4D–F), suggesting that the TRPC channel has metal ion specificity. The conversion from divalent to monovalent copper ions under oxidative stress conditions could be an important part of endogenous regulators for TRPC4 and TRPC5 channel activity. Figure 4. Monovalent copper (Cu+) had no effect on TRPC channels. (A) TRPC4 cells were perfused with 10 µM monovalent copper ((1, 10-phenanthroline), bis (triphenylphosphine) copper (I) nitrate dichloromethane adduct), and then 10 µM divalent Cu2+. (B) Similar to (A) but TRPC5 cells were used. (C) The mean ± s.e.m. data measured at ± 80 mV after perfusion with Cu+ and Cu2+. n = 5–7 for each group, ** p < 0.01 and *** p < 0.001. (D) Effect of sodium selenite on TRPC4 current. (E) IV curves for (D). (F) The mean ± s.e.m. data for the effect of Se2+ and Cu2+ on TRPC4 current. 3.5. Extracellular Activation of Cu2+ on TRPC4 and 5 Channels Whole-cell patch recordings were performed using a pipette solution containing 10 µM Cu2+. The activation of the TRPC4 current by the intracellular Cu2+ application did not happen after the whole-cell configuration was formed for more than 5 min; however, bath perfusion with 10 µM Cu2+ significantly activated the current of TRPC4α with typical IV curves (Figure 5A). A similar effect on TRPC5 was observed (Figure 5B). We also performed outside-out excised membrane patches and the stimulating effects on TRPC4 and TRPC5 currents by Cu2+ were significant after the external surface exposure to Cu2+ by bath perfu- sion (Figure 5C,D). These data suggest that the action site for Cu2+ is extracellularly located. Biomolecules 2023, 13, x FOR PEER REVIEW 7 of 17 Cu2+, and the washout of Hcy. (E) The mean ± s.e.m. data for the effect of Cu2+ (n = 6–8. *** p < 0.01). (F) The mean ± s.e.m. data for Hcy plus Cu2+ (n = 5–6. *** p < 0.001). 3.4. No Effect of Monovalent Cu+ on TRPC Channel To test the role of copper ion charges, we examined the effects of monovalent copper (I) compounds. As shown in Figure 4, the copper (I), (1,10-phenanthroline)bis(tri-phenylphosphine) copper (I) nitrate dichloromethane adduct, had no effect on TRPC4α and TRPC5 channel activity, but the divalent Cu2+ activated them (Figure 4A–C). Similarly, no effects of the monovalent copper, copper (I) 1-butanethiolate), and copper (I) tetrakis(acetonitrile) copper(I) tetrafluoroborate) were observed on TRPC4α channels (Figure S3). These data suggest that the divalent copper ions are essential for TRPC chan-nel activation, but there are no effects for monovalent Cu+ ions. In addition, Se2+ with an-tioxidant properties had no effect on TRPC4α channels (Figure 4D–F), suggesting that the TRPC channel has metal ion specificity. The conversion from divalent to monovalent cop-per ions under oxidative stress conditions could be an important part of endogenous reg-ulators for TRPC4 and TRPC5 channel activity. Figure 4. Monovalent copper (Cu+) had no effect on TRPC channels. (A) TRPC4 cells were perfused with 10 µM monovalent copper ((1, 10-phenanthroline), bis (triphenylphosphine) copper (I) nitrate dichloromethane adduct), and then 10 µM divalent Cu2+. (B) Similar to (A) but TRPC5 cells were used. (C) The mean ± s.e.m. data measured at ± 80 mV after perfusion with Cu+ and Cu2+. n = 5–7 for each group, ** p < 0.01 and *** p < 0.001. (D) Effect of sodium selenite on TRPC4 current. (E) IV curves for (D). (F) The mean ± s.e.m. data for the effect of Se2+ and Cu2+ on TRPC4 current. Biomolecules 2023, 13, 952 8 of 16 Figure 5. Extracellular effect of Cu2+ on TRPC4 and TRPC5 channels. (A) A whole-cell patch was recorded in the HEK293 T-REx cells overexpressing TRPC4α with a pipette solution containing 10 µM Cu2+ (n = 4 for each group). (B) Same as (A) but cells overexpressing TRPC5 cells were used. (C) Example of outside-out patches showing the effect of Cu2+ on TRPC4α. (D) Outside-out patches for TRPC5 channels. (E) The mean ± s.e.m. for (A) and (B) (n = 4). (F) The mean ± s.e.m. for (C,D) (n = 4). * p < 0.05, ** p < 0.01, and *** p < 0.001. 3.6. Amino acid Residues of TRPC4 Involved in Copper Activation To identify the action site of channel activation by Cu2+, we substituted the negatively charged glutamic acids (E) at the position of E542, E543, and E555 with the uncharged amino acid glutamine (Q); the cysteine (C554) with tryptophan (W); and the positively charged lysine (K) with the negatively charged glutamic acid (E) in the putative extracellular loops between the S5 and S6 domain of TRPC4α (Figure 6). The mutants of E542Q/E543Q, E555Q, C554W, and K556E did not affect the membrane trafficking of the channel proteins; however, the mutants of E542Q/E543Q and C554W caused resistance to Cu2+, but these mutants did not alter the sensitivity to trypsin, since trypsin is assumed to be an intracellular signalling process through GPCR activation (Figure 6). The mutants E555Q and K556E did not significantly change the effect of copper activation. These data indicate that negatively charged glutamic acids and the cysteine residue in the third extracellular loop are functional targets for divalent copper. Biomolecules 2023, 13, x FOR PEER REVIEW 8 of 17 3.5. Extracellular Activation of Cu2+ on TRPC4 and 5 Channels Whole-cell patch recordings were performed using a pipette solution containing 10 µM Cu2+. The activation of the TRPC4 current by the intracellular Cu2+ application did not happen after the whole-cell configuration was formed for more than 5 min; however, bath perfusion with 10 µM Cu2+ significantly activated the current of TRPC4α with typical IV curves (Figure 5A). A similar effect on TRPC5 was observed (Figure 5B). We also per-formed outside-out excised membrane patches and the stimulating effects on TRPC4 and TRPC5 currents by Cu2+ were significant after the external surface exposure to Cu2+ by bath perfusion (Figure 5C,D). These data suggest that the action site for Cu2+ is extracellu-larly located. Figure 5. Extracellular effect of Cu2+ on TRPC4 and TRPC5 channels. (A) A whole-cell patch was recorded in the HEK293 T-REx cells overexpressing TRPC4α with a pipette solution containing 10 µM Cu2+ (n = 4 for each group). (B) Same as (A) but cells overexpressing TRPC5 cells were used. (C) Example of outside-out patches showing the effect of Cu2+ on TRPC4α. (D) Outside-out patches for TRPC5 channels. (E) The mean ± s.e.m. for (A) and (B) (n = 4). (F) The mean ± s.e.m. for (C,D) (n = 4). * p < 0.05, ** p < 0.01, and *** p < 0.001. Biomolecules 2023, 13, 952 9 of 16 Figure 6. Identification of amino acids involved in channel activation by Cu2+. The mutants of TRPC4α tagged with EYFP were made by site-mutagenesis and membrane localisation was examined using a fluorescent microscope. (A) The double glutamic acid mutants (TRPC4-E542Q/E543Q) showed the loss of channel activation by Cu2+, but the robust current through the mutant channel can also be activated by trypsin (2 nM). (B) The TRPC4-E555Q mutant was activated by Cu2+. (C) Less sensitivity to Cu2+ for the cysteine mutant (TRPC4-C555W). (D) Glysine at the position of 556 substituted with glutamic acid (TRPC4-K556E). (E) Amino acid alignment of the transmembrane region (S5-S6) for TRPCs (red asterisks indicate residues subject to mutagenesis) and the mean ± s.e.m. data showing the amplitude of currents corresponding to the mutants and the wild-type control after perfusion with Cu2+ (n = 8). *** p < 0.001. 3.7. TRPC and Homocysteine-Copper Complexes in the Regulation of Endothelial Cell Proliferation The blocking of TRPC channels has been shown to inhibit cell proliferation by us and others [27,42,47]. Here we further demonstrated the roles of TRPCs in the endothelial cells from macrovasculature. The proliferation of HAECs was significantly inhibited by specific pore-blocking TRPC antibodies (Figure 7A), which was consistent with the nonselective blocker 2-APB (Figure 7B). The over-expression of TRPC1 or TRPC4 promoted proliferation (Figure 7C), suggesting the significant contribution of TRPC channel activity to endothelial cell proliferation. However, Hcy inhibited the proliferation of HAECs but increased the proliferation of HUVECs. The pro-proliferative effect was more pronounced in the culture medium omitting cysteine and methionine (Figure 7D), or in the T-REx cells overexpressing Hcy-sensitive TRPC5 channels (Figure S4). On the other hand, divalent Biomolecules 2023, 13, x FOR PEER REVIEW 9 of 17 3.6. Amino acid Residues of TRPC4 Involved in Copper Activation To identify the action site of channel activation by Cu2+, we substituted the negatively charged glutamic acids (E) at the position of E542, E543, and E555 with the uncharged amino acid glutamine (Q); the cysteine (C554) with tryptophan (W); and the positively charged lysine (K) with the negatively charged glutamic acid (E) in the putative extracel-lular loops between the S5 and S6 domain of TRPC4α (Figure 6). The mutants of E542Q/E543Q, E555Q, C554W, and K556E did not affect the membrane trafficking of the channel proteins; however, the mutants of E542Q/E543Q and C554W caused resistance to Cu2+, but these mutants did not alter the sensitivity to trypsin, since trypsin is assumed to be an intracellular signalling process through GPCR activation (Figure 6). The mutants E555Q and K556E did not significantly change the effect of copper activation. These data indicate that negatively charged glutamic acids and the cysteine residue in the third ex-tracellular loop are functional targets for divalent copper. Figure 6. Identification of amino acids involved in channel activation by Cu2+. The mutants of TRPC4α tagged with EYFP were made by site-mutagenesis and membrane localisation was exam-ined using a fluorescent microscope. (A) The double glutamic acid mutants (TRPC4-E542Q/E543Q) Biomolecules 2023, 13, 952 10 of 16 copper had no significant effect on the proliferation of HAECs but significantly reduced the proliferation of HUVECs and the HUVEC-derived cell line EA.hy926 (Figure 7E). Combined incubation with Hcy and Cu2+ showed inhibitory effects at low concentrations of copper but stimulatory effects at a high concentration (100 µM Cu2+) (Figure 7F), which exhibited significant differences from the groups treated with Hcy alone. These data suggest that the sensitivity to Hcy and Cu2+ may rely on the types of vascular endothelial cells and the ratio of Hcy and copper complexes. Figure 7. Endothelial cell proliferation regulated by TRPC channels and the effects of Hcy and copper. Cell proliferation was assayed by a WST-1 kit and absorbance was measured at a wavelength of 450 nm. (A) Endothelial cells were incubated with the pore-blocking TRPC antibodies [28,42,48] for 24 h. The TRPC5 antibody targeting the C-terminal (T5C3) and preimmune serum (Preimmune) were used as controls. (B) 2-APB. (C) HAEC cells transfected with plasmid cDNAs for TRPC1 and TRPC4 using the electroporation method [49]. (D) Effect of Hcy on HAECs and HUVECs. (E) Effect of Cu2+ on HAEC, HUVEC, and the HUVEC-derived cell line Eahy926. (F) Combined effect of Hcy (10 µM) and Cu2+. n = 8 for each group, * p < 0.05, ** p < 0.01, and *** p < 0.001, ##, not significant. Biomolecules 2023, 13, x FOR PEER REVIEW 11 of 17 Figure 7. Endothelial cell proliferation regulated by TRPC channels and the effects of Hcy and cop-per. Cell proliferation was assayed by a WST-1 kit and absorbance was measured at a wavelength of 450 nm. (A) Endothelial cells were incubated with the pore-blocking TRPC antibodies [28,42,48] for 24 h. The TRPC5 antibody targeting the C-terminal (T5C3) and preimmune serum (Preimmune) were used as controls. (B) 2-APB. (C) HAEC cells transfected with plasmid cDNAs for TRPC1 and TRPC4 using the electroporation method [49]. (D) Effect of Hcy on HAECs and HUVECs. (E) Effect of Cu2+ on HAEC, HUVEC, and the HUVEC-derived cell line Eahy926. (F) Combined effect of Hcy (10 µM) and Cu2+. n = 8 for each group, * p < 0.05, ** p < 0.01, and *** p < 0.001, ##, not significant. 3.8. Hcy–Copper Complexes in the Regulation of Cell Migration and Angiogenesis TRPC channels are involved in cell migration and angiogenesis [26,50,51], so we ob-served the effects of Hcy and copper on endothelial cell migration and angiogenesis. Us-ing a linear wound assay, the number of migrated cells was seen to be significantly re-duced after treatment with Hcy (Figure 8A,B). Angiogenesis was examined using the ex-tracellular matrix (ECM) gel and Matrigel assays. The score of angiogenesis in the ECM gel and the tube formation in the Matrigel were significantly inhibited by Hcy (Figure 8C–G). However, the addition of Cu2+ in the culture medium alleviated the inhibitory effects of Hcy on endothelial cell tube formation and angiogenesis, suggesting that endothelial Biomolecules 2023, 13, 952 11 of 16 3.8. Hcy–Copper Complexes in the Regulation of Cell Migration and Angiogenesis TRPC channels are involved in cell migration and angiogenesis [26,50,51], so we ob- served the effects of Hcy and copper on endothelial cell migration and angiogenesis. Using a linear wound assay, the number of migrated cells was seen to be significantly reduced after treatment with Hcy (Figure 8A,B). Angiogenesis was examined using the extracel- lular matrix (ECM) gel and Matrigel assays. The score of angiogenesis in the ECM gel and the tube formation in the Matrigel were significantly inhibited by Hcy (Figure 8C–G). However, the addition of Cu2+ in the culture medium alleviated the inhibitory effects of Hcy on endothelial cell tube formation and angiogenesis, suggesting that endothelial cell mobility and angiogenesis are regulated by the complexes of homocysteine and copper. Taken together, regulation by Hcy and copper complexes via TRPC4/TRPC5 channels could be regarded as a new mechanism to control endothelial function. Figure 8. Endothelial cell migration and angiogenesis regulated by Hcy and Cu2+ complexes. (A) Example of endothelial cell migration using a linear wound assay. (B) Effect of Hcy on cell migration after 24 h of incubation. (C) Example of angiogenesis using ECM gel (i) and Matrigel for HUVEC (ii) and Eahy926 cells (iii). (D) The mean ± s.e.m. showing the effect of Hcy on angiogenesis (n = 40–60 imaging fields from six cell culture dishes for each group). (E–G) Effect of Hcy (10 µM) and Cu2+ (10 µM) on endothelial cell tube formation. n = 6 for each group. The number of loops, branching, and total length of tubes were analysed by software. *** p < 0.001. Biomolecules 2023, 13, x FOR PEER REVIEW 12 of 17 cell mobility and angiogenesis are regulated by the complexes of homocysteine and cop-per. Taken together, regulation by Hcy and copper complexes via TRPC4/TRPC5 channels could be regarded as a new mechanism to control endothelial function. Figure 8. Endothelial cell migration and angiogenesis regulated by Hcy and Cu2+ complexes. (A) Example of endothelial cell migration using a linear wound assay. (B) Effect of Hcy on cell migration after 24 h of incubation. (C) Example of angiogenesis using ECM gel (i) and Matrigel for HUVEC (ii) and Eahy926 cells (iii). (D) The mean ± s.e.m. showing the effect of Hcy on angiogenesis (n = 40–60 imaging fields from six cell culture dishes for each group). (E–G) Effect of Hcy (10 µM) and Cu2+ (10 µM) on endothelial cell tube formation. n = 6 for each group. The number of loops, branching, and total length of tubes were analysed by software. *** p < 0.001. Biomolecules 2023, 13, 952 12 of 16 4. Discussion Our data show that Hcy can increase Ca2+ influx in HAECs. The increase is mediated by the opening of TRPC4 and TRPC5 channels. Divalent copper acts as a non-selective activator of TRPC4/5 channels. The channel activation by divalent copper is regulated by Hcy. The charge of copper ions is critical for TRPC channel opening because monovalent copper (I) shows no significant effect on TRPC channel activity. We also explored the action site for divalent copper using excised membrane patches and site mutagenesis. The cysteine (C554) and glutamic acids (E542 and E543) in the third extracellular loop of TRPC4α are responsible for copper activation. Moreover, we showed that copper and Hcy are essential regulators for endothelial cell proliferation, migration, and angiogenesis. Divalent copper seems to counteract the effect of Hcy on proliferation and angiogenesis which suggests the importance of the Hcy–copper interaction in causing endothelium dysfunction and atherosclerosis. The regulation of TRPC channels is the sought-after underlying mechanism for the pathogenesis of patients with hyperhomocysteinemia. The effect of Hcy on intracellular [Ca2+]i is still unclear in endothelial cells, although there are several reports showing that Hcy increases the Ca2+ influx in human platelets [52], cultured vascular smooth muscle cells [23], podocytes [53], and neurons [24,54]. Here, we found that Hcy increased the Ca2+ influx in HAECs which is mediated by the activation of TRPC4 and TRPC5. The blocking of voltage-gated Ca2+ channels and NMDA receptors was unable to prevent the Hcy-induced Ca2+ influx, suggesting that the Hcy-induced Ca2+ entry pathway is not through the voltage-gated channel or the ligand-gated NMDA receptor channel in vascular endothelial cells. In addition, the Hcy-induced intracellular Ca2+ increase has been linked to ER calcium release via the homocysteine-inducible ER stress protein [55]; however, Hcy-induced Ca2+ influx also happened in the cells acutely treated with SERCA blocker TG which suggests that main pathways of Ca2+ influx are across the plasma membrane rather than the intracellular Ca2+ release from the ER. The effect of Hcy on store-operated channels or ORAI channels is unknown, but high concentrations (≥100 µM) of Hcy may inhibit the store-operated Ca2+ influx [56]. Hcy also inhibits BKCa and thus depolarises the membrane potential and increases the vascular tone [57]. This action may explain the diverse responses in vascular tone or [Ca2+]i observed in some cell types [58,59]. The N-methyl-D-aspartate (NMDA) receptor activation by Hcy could also be a mechanism for Ca2+ influx in the nervous system [24] but this mechanism may be less significant in vascular endothelial cells. Homocysteine contains sulphuric residues so its toxic effect has been attributed to redox homeostasis, such as the production of different reactive oxygen species (ROS), thus leading to the oxidation of low-density lipoprotein [20]. Cellular oxidative stress including ER stress has also been proposed for Hcy pathophysiology [19]; the increased ROS production activates ROS-sensitive Ca2+ channels. In addition, we demonstrated that the TRPC5 channel is a redox-sensitive channel that can be activated by thioredoxin and reducing agents [37] and mercury compounds [41]. Here, we found that TRPC4 and TRPC5 channel activities can be enhanced by Hcy, especially when the channels are opened by lanthanides. TRPM2 is also a redox-sensitive channel; however, Hcy itself had no acute effect on TRPM2 but significantly regulated the effect of Cu2+ on the TRPM2 channel. Chronic exposure to Hcy may change gene expressions, through Ca2+ channels and ROS signalling molecules [53,60], but we did not observe such gene expression in this study. The total Hcy level in the blood is determined by both genetic and environmental factors and is typically maintained at a normal range (2–14 µM). Vitamin deficiencies, in particular folate acid and vitamins B6 and B12, appear to be the most common causes of elevated Hcy [61]. A supplement of folic acid alone or with vitamin B12 or B6 can help to lower Hcy levels, but it is still uncertain how effective this will be in the prevention of cardiovascular disease or Hcy-related diseases. It has been demonstrated that both Hcy and copper are increased in diseased vessels and diabetic patients; however, the question of how Hcy interacts with copper and causes occlusive diseases remains unanswered. Here, we show for the first time that copper can interact with Hcy, controlling TRPC channel Biomolecules 2023, 13, 952 13 of 16 activity, thus changing intracellular Ca2+ signalling, and subsequently the endothelial function. This mechanism gives a new understanding of the two factors in the pathogenesis of cardiovascular diseases. Too low or too high concentrations of copper are detrimental, but we have demonstrated that the charge of copper ions could be more important than the copper concentration. Although treatment with a divalent-copper-selective chelator, triethylenetetramine (TETA), to lower the copper in the body may improve the cardiac structure and function in patients and rats with diabetic cardiomyopathy [34], a more pre- cise clinical trial is needed, especially regarding the charge of copper ions and consideration of the redox environment in the body. The inhibition of TRPCs shows anti-proliferative effects while the activation of TRPC channels shows proliferative effects in vascular endothelial cells, which is consistent with the observations in other cell types [26,42]. However, different types of endothelial cells may show differences, such as the HAECs showing inhibitory characteristics and the HUVECs showing pro-proliferative characteristics. This could be related to the predominance of Hcy-sensitive channels. In patients with hyperhomocysteinemia, neointimal hyperplasia in small vessels is evident [1]. In summary, we revealed a new mechanism of Hcy and copper and their interplay with TRPC channels in endothelial cells. This new concept could be extended to other cell types since many diseases are related to Hcy and copper and Hcy is associated with all-cause mortality. The findings suggest the importance of copper ion charges in the pathogenesis of vascular disorders, particularly in patients with increased homocysteine levels, and may also provide an alternative explanation for why Hcy-lowering therapy is not very significant in clinical trials and how Hcy-copper complexes could be the determinants. Supplementary Materials: The following supporting information can be downloaded at: https:// www.mdpi.com/article/10.3390/biom13060952/s1, Figure S1. Activation of TRPC4β1 by Cu2+ and the dose-response of Cu2+ on TRPC4α; Figure S2. Effect of Hcy and copper on TRPM2 current; Figure S3. Example of monovalent copper (I) 1-butanethiolate on TRPC4α current; and Figure S4. Hcy increased cell proliferation of T-Rex cells overexpressing TRPC5. Author Contributions: Conceptualisation, G.-L.C., B.Z. and S.-Z.X.; methodology, G.-L.C., B.Z., N.D., H.J. and R.J.C.; validation, G.-L.C., B.Z., N.D. and S.-Z.X.; formal analysis, G.-L.C., B.Z., N.D. and R.J.C.; investigation, G.-L.C., B.Z., H.J., N.D., R.S. and R.J.C.; writing—original draft preparation, G.- L.C., B.Z. and S.-Z.X.; writing—review and editing, S.-Z.X.; supervision, S.-Z.X.; funding acquisition, S.-Z.X. All authors have read and agreed to the published version of the manuscript. Funding: This research was funded by the British Heart Foundation (PG/08/071/25473) (to S.-Z.X.). B.Z. received a Scholarship from the China Scholarship Council. H.J. was supported by a Leverhulme Trust fellowship. Institutional Review Board Statement: Not applicable. Informed Consent Statement: Not applicable. Data Availability Statement: The data supporting this study are available from the corresponding authors upon reasonable request. Acknowledgments: We thank Neil Watson and Sahar Avazzadeh for their technical assistance. Conflicts of Interest: The authors declare no conflict of interest. References 1. McCully, K.S. Vascular pathology of homocysteinemia: Implications for the pathogenesis of arteriosclerosis. Am. J. Pathol. 1969, 2. 3. 56, 111–128. [PubMed] Jensen, M.K.; Bertoia, M.L.; Cahill, L.E.; Agarwal, I.; Rimm, E.B.; Mukamal, K.J. Novel metabolic biomarkers of cardiovascular disease. Nat. Rev. Endocrinol. 2014, 10, 659–672. [CrossRef] [PubMed] Chen, S.-C.; Su, H.-M.; Chang, J.-M.; Liu, W.-C.; Tsai, J.-C.; Tsai, Y.-C.; Lin, M.-Y.; Hwang, S.-J.; Chen, H.-C. Increasing prevalence of peripheral artery occlusive disease in hemodialysis patients: A 2-year follow-up. Am. J. Med. Sci. 2012, 343, 440–445. [CrossRef] [PubMed] Biomolecules 2023, 13, 952 14 of 16 4. 5. 6. 7. 8. 9. Schaffer, A.; Verdoia, M.; Cassetti, E.; Marino, P.; Suryapranata, H.; De Luca, G.; Novara Atherosclerosis Study Group (NAS). Relationship between homocysteine and coronary artery disease. Results from a large prospective cohort study. Thromb. Res. 2014, 134, 288–293. [CrossRef] Zylberstein, D.E.; Bengtsson, C.; Björkelund, C.; Landaas, S.; Sundh, V.; Thelle, D.; Lissner, L. Serum homocysteine in relation to mortality and morbidity from coronary heart disease: A 24-year follow-up of the population study of women in Gothenburg. Circulation 2004, 109, 601–606. [CrossRef] Homocysteine Studies Collaboration. Homocysteine and risk of ischemic heart disease and stroke: A meta-analysis. JAMA 2002, 288, 2015–2022. [CrossRef] Zylberstein, D.E.; Skoog, I.; Björkelund, C.; Guo, X.; Hultén, B.; Andreasson, L.-A.; Palmertz, B.; Thelle, D.S.; Lissner, L. Homocysteine levels and lacunar brain infarcts in elderly women: The prospective population study of women in Gothenburg. J. Am. Geriatr. Soc. 2008, 56, 1087–1091. [CrossRef] Casas, J.P.; Bautista, L.E.; Smeeth, L.; Sharma, P.; Hingorani, A.D. Homocysteine and stroke: Evidence on a causal link from mendelian randomisation. Lancet 2005, 365, 224–232. [CrossRef] Kuan, Y.M.; Dear, A.E.; Grigg, M.J. Homocysteine: An aetiological contributor to peripheral vascular arterial disease. ANZ J. Surg. 2002, 72, 668–671. [CrossRef] 10. Den Heijer, M.; Lewington, S.; Clarke, R. Homocysteine, MTHFR and risk of venous thrombosis: A meta-analysis of published epidemiological studies. J. Thromb. Haemost. 2005, 3, 292–299. [CrossRef] 11. Loscalzo, J. Homocysteine and dementias. N. Engl. J. Med. 2002, 346, 466–468. [CrossRef] 12. Rozycka, A.; Jagodzinski, P.P.; Kozubski, W.; Lianeri, M.; Dorszewska, J. Homocysteine Level and Mechanisms of Injury in Parkinson’s Disease as Related to MTHFR, MTR, and MTHFD1 Genes Polymorphisms and L-Dopa Treatment. Curr. Genom. 2013, 14, 534–542. [CrossRef] 13. Elias, A.N.; Eng, S. Homocysteine concentrations in patients with diabetes mellitus-relationship to microvascular and macrovas- cular disease. Diabetes Obes. Metab. 2005, 7, 117–121. [CrossRef] 14. Van Meurs, J.B.; Dhonukshe-Rutten, R.A.; Pluijm, S.M.; Van Der Klift, M.; De Jonge, R.; Lindemans, J.; De Groot, L.C.; Hofman, A.; Witteman, J.C.; Van Leeuwen, J.P.; et al. Homocysteine levels and the risk of osteoporotic fracture. N. Engl. J. Med. 2004, 350, 2033–2041. [CrossRef] 15. Yi, F.; Li, P.L. Mechanisms of homocysteine-induced glomerular injury and sclerosis. Am. J. Nephrol. 2008, 28, 254–264. [CrossRef] 16. Mills, J.L.; Lee, Y.J.; Conley, M.R.; Kirke, P.N.; McPartlin, J.M.; Weir, D.G.; Scott, J.M. Homocysteine metabolism in pregnancies complicated by neural-tube defects. Lancet 1995, 345, 149–151. [CrossRef] 17. Adinolfi, L.E.; Ingrosso, D.; Cesaro, G.; Cimmino, A.; D’Antò, M.; Capasso, R.; Zappia, V.; Ruggiero, G. Hyperhomocysteinemia and the MTHFR C677T polymorphism promote steatosis and fibrosis in chronic hepatitis C patients. Hepatology 2005, 41, 995–1003. [CrossRef] 18. Wang, L.; Chen, X.; Tang, B.; Hua, X.; Klein-Szanto, A.; Kruger, W.D. Expression of mutant human cystathionine {beta}-synthase rescues neonatal lethality but not homocystinuria in a mouse model. Hum. Mol. Genet. 2005, 14, 2201–2208. [CrossRef] 19. Austin, R.C.; Lentz, S.R.; Werstuck, G.H. Role of hyperhomocysteinemia in endothelial dysfunction and atherothrombotic disease. Cell Death Differ. 2004, 11 (Suppl. S1), S56–S64. [CrossRef] 20. Becker, J.S.; Adler, A.; Schneeberger, A.; Huang, H.; Wang, Z.; Walsh, E.; Koller, A.; Hintze, T.H. Hyperhomocysteinemia, a cardiac metabolic disease: Role of nitric oxide and the p22phox subunit of NADPH oxidase. Circulation 2005, 111, 2112–2118. [CrossRef] 21. Toole, J.; Malinow, M.; Chambless, L. Lowering homocysteine in patients with ischemic stroke to prevent recurrent stroke, myocardial infarction, and death: The Vitamin Intervention for Stroke Prevention (VISP) randomized controlled trial. JAMA 2004, 291, 565–575. [CrossRef] [PubMed] Spence, J.D.; Stampfer, M.J. Understanding the complexity of homocysteine lowering with vitamins: The potential role of subgroup analyses. JAMA 2011, 306, 2610–2611. [CrossRef] [PubMed] 22. 23. Mujumdar, V.S.; Hayden, M.R.; Tyagi, S.C. Homocyst(e)ine induces calcium second messenger in vascular smooth muscle cells. J. Cell Physiol. 2000, 183, 28–36. [CrossRef] 24. Abushik, P.A.; Niittykoski, M.; Giniatullina, R.; Shakirzyanova, A.; Bart, G.; Fayuk, D.; Sibarov, D.A.; Antonov, S.M.; Giniatullin, R. The role of NMDA and mGluR5 receptors in calcium mobilization and neurotoxicity of homocysteine in trigeminal and cortical neurons and glial cells. J. Neurochem. 2014, 129, 264–274. [CrossRef] 25. Ganapathy, P.S.; White, R.E.; Ha, Y.; Bozard, B.R.; McNeil, P.L.; Caldwell, R.W.; Kumar, S.; Black, S.M.; Smith, S.B. The role of N-methyl-D-aspartate receptor activation in homocysteine-induced death of retinal ganglion cells. Investig. Ophthalmol. Vis. Sci. 2011, 52, 5515–5524. [CrossRef] 26. Xu, S.-Z.; Muraki, K.; Zeng, F.; Li, J.; Sukumar, P.; Shah, S.; Dedman, A.M.; Flemming, P.K.; McHugh, D.; Naylor, J.; et al. A sphingosine-1-phosphate-activated calcium channel controlling vascular smooth muscle cell motility. Circ. Res. 2006, 98, 1381–1389. [CrossRef] 27. Kumar, B.; Dreja, K.; Shah, S.S.; Cheong, A.; Xu, S.Z.; Sukumar, P.; Naylor, J.; Forte, A.; Cipollaro, M.; McHugh, D.; et al. Upregulated TRPC1 channel in vascular injury in vivo and its role in human neointimal hyperplasia. Circ. Res. 2006, 98, 557–563. [CrossRef] 28. Xu, S.Z.; Beech, D.J. TrpC1 is a membrane-spanning subunit of store-operated Ca2+ channels in native vascular smooth muscle cells. Circ. Res. 2001, 88, 84–87. [CrossRef] Biomolecules 2023, 13, 952 15 of 16 29. Beech, D.J.; Muraki, K.; Flemming, R. Non-selective cationic channels of smooth muscle and the mammalian homologues of Drosophila TRP. J. Physiol. 2004, 559 Pt 3, 685–706. [CrossRef] 30. Kang, Y.J. Copper and homocysteine in cardiovascular diseases. Pharmacol. Ther. 2010, 129, 321–331. [CrossRef] 31. Mansoor, M.A.; Bergmark, C.; Haswell, S.J.; Savage, I.F.; Evans, P.H.; Berge, R.K.; Svardal, A.M.; Kristensen, O. Correlation between plasma total homocysteine and copper in patients with peripheral vascular disease. Clin. Chem. 2000, 46, 385–391. [CrossRef] 32. Dudman, N.P.; Wilcken, D.E. Increased plasma copper in patients with homocystinuria due to cystathionine beta-synthase deficiency. Clin. Chim. Acta 1983, 127, 105–113. [CrossRef] 33. Gromadzka, G.; Rudnicka, M.; Chabik, G.; Przybyłkowski, A.; Członkowska, A. Genetic variability in the methylenetetrahydrofo- late reductase gene (MTHFR) affects clinical expression of Wilson’s disease. J. Hepatol. 2011, 55, 913–919. [CrossRef] 34. Zhang, L.; Ward, M.-L.; Phillips, A.R.; Zhang, S.; Kennedy, J.; Barry, B.; Cannell, M.B.; Cooper, G.J. Protection of the heart by treatment with a divalent-copper-selective chelator reveals a novel mechanism underlying cardiomyopathy in diabetic rats. Cardiovasc. Diabetol. 2013, 12, 123. [CrossRef] 35. Xu, S.Z.; Zhong, W.; Watson, N.M.; Dickerson, E.; Wake, J.D.; Lindow, S.W.; Newton, C.J.; Atkin, S.L. Fluvastatin reduces oxidative damage in human vascular endothelial cells by upregulating Bcl-2. J. Thromb. Haemost. 2008, 6, 692–700. [CrossRef] 36. Daskoulidou, N.; Zeng, B.; Berglund, L.M.; Jiang, H.; Chen, G.-L.; Kotova, O.; Bhandari, S.; Ayoola, J.; Griffin, S.; Atkin, S.L.; et al. High glucose enhances store-operated calcium entry by upregulating ORAI/STIM via calcineurin-NFAT signalling. J. Mol. Med. 2015, 93, 511–521. [CrossRef] 37. Xu, S.-Z.; Sukumar, P.; Zeng, F.; Li, J.; Jairaman, A.; English, A.; Naylor, J.; Ciurtin, C.; Majeed, Y.; Milligan, C.J.; et al. TRPC 38. channel activation by extracellular thioredoxin. Nature 2008, 451, 69–72. [CrossRef] Jiang, H.; Zeng, B.; Chen, G.-L.; Bot, D.; Eastmond, S.; Elsenussi, S.E.; Atkin, S.L.; Boa, A.N.; Xu, S.-Z. Effect of non-steroidal anti-inflammatory drugs and new fenamate analogues on TRPC4 and TRPC5 channels. Biochem. Pharmacol. 2012, 83, 923–931. [CrossRef] 39. Chen, G.-L.; Zeng, B.; Eastmond, S.; Elsenussi, S.E.; Boa, A.; Xu, S.-Z. Pharmacological comparison of novel synthetic fenamate analogues with econazole and 2-APB on the inhibition of TRPM2 channels. Br. J. Pharmacol. 2012, 167, 1232–1243. [CrossRef] 40. Li, P.; Rubaiy, H.N.; Chen, G.L.; Hallett, T.; Zaibi, N.; Zeng, B.; Saurabh, R.; Xu, S.Z. Mibefradil, a T-type Ca(2+) channel blocker also blocks Orai channels by action at the extracellular surface. Br. J. Pharmacol. 2019, 176, 3845–3856. [CrossRef] 41. Xu, S.-Z.; Zeng, B.; Daskoulidou, N.; Chen, G.-L.; Atkin, S.L.; Lukhele, B. Activation of TRPC cationic channels by mercurial compounds confers the cytotoxicity of mercury exposure. Toxicol. Sci. 2012, 125, 56–68. [CrossRef] [PubMed] 42. Zeng, B.; Yuan, C.; Yang, X.; Atkin, S.L.; Xu, S.-Z. TRPC channels and their splice variants are essential for promoting human ovarian cancer cell proliferation and tumorigenesis. Curr. Cancer Drug Targets 2013, 13, 103–116. [CrossRef] [PubMed] 43. Zaibi, N.; Li, P.; Xu, S.Z. Protective effects of dapagliflozin against oxidative stress-induced cell injury in human proximal tubular cells. PLoS ONE 2021, 16, e0247234. [CrossRef] [PubMed] 44. Aranda, E.; Owen, G.I. A semi-quantitative assay to screen for angiogenic compounds and compounds with angiogenic potential using the EA.hy926 endothelial cell line. Biol. Res. 2009, 42, 377–389. [CrossRef] 45. Xu, S.-Z.; Zeng, F.; Boulay, G.; Grimm, C.; Harteneck, C.; Beech, D.J. Block of TRPC5 channels by 2-aminoethoxydiphenyl borate: A differential, extracellular and voltage-dependent effect. Br. J. Pharmacol. 2005, 145, 405–414. [CrossRef] 46. Zeng, B.; Chen, G.L.; Xu, S.Z. Divalent copper is a potent extracellular blocker for TRPM2 channel. Biochem. Biophys. Res. Commun. 2012, 424, 279–284. [CrossRef] 47. Kuang, C.-Y.; Yü, Y.; Wang, K.; Qian, D.-H.; Den, M.-Y.; Huang, L. Knockdown of transient receptor potential canonical-1 reduces the proliferation and migration of endothelial progenitor cells. Stem. Cells Dev. 2012, 21, 487–496. [CrossRef] 48. Xu, S.-Z.; Zeng, F.; Lei, M.; Li, J.; Gao, B.; Xiong, C.; Sivaprasadarao, A.; Beech, D. Generation of functional ion-channel tools by E3 targeting. Nat. Biotechnol. 2005, 23, 1289–1293. [CrossRef] 49. Zeng, B.; Chen, G.-L.; Garcia-Vaz, E.; Bhandari, S.; Daskoulidou, N.; Berglund, L.M.; Jiang, H.; Hallett, T.; Zhou, L.-P.; Huang, L.; et al. ORAI channels are critical for receptor-mediated endocytosis of albumin. Nat. Commun. 2017, 8, 1920. [CrossRef] 50. Yu, P.-C.; Gu, S.-Y.; Bu, J.-W.; Du, J.-L. TRPC1 is essential for in vivo angiogenesis in zebrafish. Circ. Res. 2010, 106, 1221–1232. [CrossRef] 51. Antigny, F.; Girardin, N.; Frieden, M. Transient receptor potential canonical channels are required for in vitro endothelial tube formation. J. Biol. Chem. 2012, 287, 5917–5927. [CrossRef] 52. Alexandru, N.; Jardín, I.; Popov, D.; Simionescu, M.; García-Estañ, J.; Salido, G.M.; Rosado, J.A. Effect of homocysteine on calcium mobilization and platelet function in type 2 diabetes mellitus. J. Cell Mol. Med. 2008, 12, 2586–2597. [CrossRef] 53. Han, H.; Wang, Y.; Li, X.; Wang, P.A.; Wei, X.; Liang, W.; Ding, G.; Yu, X.; Bao, C.; Zhang, Y.; et al. Novel role of NOD2 in mediating Ca2+ signaling: Evidence from NOD2-regulated podocyte TRPC6 channels in hyperhomocysteinemia. Hypertension 2013, 62, 506–511. [CrossRef] 54. Ovey, I.S.; Naziroglu, M. Homocysteine and cytosolic GSH depletion induce apoptosis and oxidative toxicity through cytosolic calcium overload in the hippocampus of aged mice: Involvement of TRPM2 and TRPV1 channels. Neuroscience 2015, 284, 225–233. [CrossRef] Biomolecules 2023, 13, 952 16 of 16 55. Chigurupati, S.; Wei, Z.; Belal, C.; Vandermey, M.; Kyriazis, G.; Arumugam, T.; Chan, S.L. The homocysteine-inducible endoplasmic reticulum stress protein counteracts calcium store depletion and induction of CCAAT enhancer-binding protein homologous protein in a neurotoxin model of Parkinson disease. J. Biol. Chem. 2009, 284, 18323–18333. [CrossRef] 56. Zhang, H.-S.; Xiao, J.-H.; Cao, E.-H.; Qin, J.-F. Homocysteine inhibits store-mediated calcium entry in human endothelial cells: Evidence for involvement of membrane potential and actin cytoskeleton. Mol. Cell Biochem. 2005, 269, 37–47. [CrossRef] 57. Cai, B.; Gong, D.; Pan, Z.; Liu, Y.; Qian, H.; Zhang, Y.; Jiao, J.; Lu, Y.; Yang, B. Large-conductance Ca2+-activated K+ currents blocked and impaired by homocysteine in human and rat mesenteric artery smooth muscle cells. Life Sci. 2007, 80, 2060–2066. [CrossRef] 58. Cortés, M.P.; Becerra, J.P.; Vinet, R.; Álvarez, R.; Quintana, I. Inhibition of ATP-induced calcium influx by homocysteine in human umbilical vein endothelial cells. Cell Biol. Int. 2013, 37, 600–607. [CrossRef] 59. Cai, B.; Gong, D.; Chen, N.; Li, J.; Wang, G.; Lu, Y.; Yang, B. The negative inotropic effects of homocysteine were prevented by matrine via the regulating intracellular calcium level. Int. J. Cardiol. 2011, 150, 113–115. [CrossRef] 60. Thilo, F.; Liu, Y.; Krueger, K.; Förste, N.; Wittstock, A.; Scholze, A.; Tepel, M. Do cysteine residues regulate transient receptor potential canonical type 6 channel protein expression? Antioxid. Redox Signal 2012, 16, 452–457. [CrossRef] 61. Lucock, M.; Yates, Z. Folic acid-vitamin and panacea or genetic time bomb? Nat. Rev. Genet. 2005, 6, 235–240. [CrossRef] [PubMed] Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.
null
10.1177_03010066231175014.pdf
Data Availability Data and scripts used to conduct this analysis can be viewed at Open Science Framework: Data and analysis for Effects of cortical distance on the Ebbinghaus and Delboeuf illusions. https://doi.org/10.17605/OSF.IO/ GUHSF.
Data Availability Data and scripts used to conduct this analysis can be viewed at Open Science Framework: Data and analysis for Effects of cortical distance on the Ebbinghaus and Delboeuf illusions. https://doi.org/10.17605/OSF.IO/ GUHSF .
Article Effects of cortical distance on the Ebbinghaus and Delboeuf illusions Perception 2023, Vol. 52(7) 459–483 © The Author(s) 2023 Article reuse guidelines: sagepub.com/journals-permissions DOI: 10.1177/03010066231175014 journals.sagepub.com/home/pec Poutasi W. B. Urale and Dietrich Samuel Schwarzkopf School of Optometry & Vision Science, The University of Auckland, New Zealand Abstract The Ebbinghaus and Delboeuf illusions affect the perceived size of a target circle depending on the size and proximity of circular inducers or a ring. Converging evidence suggests that these illusions are driven by interactions between contours mediated by their cortical distance in primary visual cortex. We tested the effect of cortical distance on these illusions using two methods: First, we manipulated retinal distance between target and inducers in a two-interval forced choice design, finding that targets appeared larger with a closer surround. Next, we predicted that targets pre- sented peripherally should appear larger due to cortical magnification. Hence, we tested the illu- sion strength when positioning the stimuli at various eccentricities, with results supporting this hypothesis. We calculated estimated cortical distances between illusion elements in each experi- ment and used these estimates to compare the relationship between cortical distance and illusion strength across our experiments. In a final experiment, we modified the Delboeuf illusion to test whether the influence of the inducers/annuli in this illusion is influenced by an inhibitory surround. We found evidence that an additional outer ring makes targets appear smaller compared to a sin- gle-ring condition, suggesting that near and distal contours have antagonistic effects on perceived target size. Keywords neural mechanisms, perception, crowding, eccentricity, Ebbinghaus illusion, Delboeuf illusion, size perception Date Received: 7 February 2022; accepted: 12 April 2023 Corresponding author: Dietrich Samuel Schwarzkopf, School of Optometry & Vision Science, The University of Auckland, 85 Park Road, Grafton, Auckland. Email: s.schwarzkopf@auckland.ac.nz 460 Perception 52(7) The Ebbinghaus illusion (see Figure 1) has bamboozled our visual systems for over a century (Ebbinghaus, 1902; Titchener, 1905). Yet, despite a mountain of research on this illusion, the neural mechanisms underlying it remain poorly understood. Filling this lacuna is crucial for under- standing how the brain determines visual object size, which is itself an unresolved question (Schwarzkopf, 2015). Theories of the Ebbinghaus Illusion Several theories have attempted to explain the Ebbinghaus illusion. Most illustrations of this illusion show an apparent “size-contrast” effect, where the center circle (target) surrounded by small inducers appears larger, while large inducers make the target appear smaller (e.g., Massaro & Anderson, 1970, 1971; Obonai, 1954). Many authors have thus described the illusion in terms of this size-contrast mechanism (Aglioti et al., 1995; Haffenden et al., 2001; Yamazaki et al., 2010). However, Todorović and Jovanović (2018) point out that size-contrast is descriptive rather than explanatory and offers an incomplete account for the illusion. They argue that size contrast is nebulously defined, and that there is no explanation for why there is a size-contrast effect instead of an assimilation effect, as found for other visual illusions such as the tilt illusion (Clifford, 2014). Part of this objection stems from the observation that geometrical features other than inducer size also modulate the strength of the Ebbinghaus illusion. One such factor is object-level similarity between targets and inducers, which make the illusion stronger (Coren & Miller, 1974; Massaro & Anderson, 1971; Rose & Bressan, 2002). The illusion also depends on the amount of empty space between inducers, with a more complete ring of inducers around the periphery of the target strengthening the illusion (Girgus et al., 1972; Massaro & Anderson, 1970; Roberts et al., 2005). Roberts et al. (2005) investigated the role of completeness by directly comparing the Ebbinghaus illusion to the Delboeuf illusion (see Figure 1), another size-perception illusion that uses a circular ring that surrounds the target stimulus instead of multiple circular inducers (Delboeuf, 1865; Evans, 1995). They showed that an Ebbinghaus configuration composed of Figure 1. Key stimuli. (A-B) The Ebbinghaus illusion. Most observers will perceive the white filled-in circle in B (small inducers) as larger than that in A (large inducers). (C) The test targets Experiments 1, 2, 3a, and 3b varies according to a staircase procedure. (D-E) The Delboeuf illusion. Most observers will perceive the white filled-in circle in E (close ring) as larger than that in D (far ring). (F) Novel two-ring Delboeuf stimulus used in Experiment 3a, featuring near and far annuli. Urale and Schwarzkopf 461 small inducers that formed a complete ring yield about the same illusory effect as a Delboeuf con- figuration, and that both configurations yielded stronger effects than Ebbinghaus configurations with less complete inducer annuli. Lastly, Ebbinghaus illusion strength changes with the distance between the target and inducers (target-inducer distance). Massaro and Anderson (1971) found that point of subjective equality (PSE) decreased with greater target-inducer distances, with the farthest distances failing to affect the perceived size of the target at all. Other work (Girgus et al., 1972; Jaeger, 1978; Roberts et al., 2005; Weintraub, 1979) also showed that target-inducer distance mod- ulates the illusion, but with an unexpected reversal of the effect of small inducers in some config- urations. That is, a sufficiently large distance between the target and small inducers causes the target to appear smaller, not larger, compared to control. Similarly, Roberts et al. (2005) found that even when controlling for completeness, increasing the target-inducer distance reversed the effect of small inducers, making the target appear smaller rather than larger. In contrast, large inducers always elicited a perceived shrinkage of the target, and this effect only became stronger with inducer-target distance. These findings demonstrate that to describe the Ebbinghaus illusion as an example of “size-contrast” is an oversimplification. Contour-based accounts (Jaeger, 1978; Jaeger & Long, 2007; Jaeger & Lorden, 1980; Sherman & Chouinard, 2016; Todorović & Jovanović, 2018; Weintraub, 1979; Weintraub & Schneck, 1986) are explanations based on interactions between the low-level contours that make up a stimulus. On the whole, these theories account better for the experimental evidence. Biphasic contour-interaction theory (BCIT) is one such account (Roberts et al., 2005; Sherman & Chouinard, 2016; Weintraub & Schneck, 1986). In contrast to the mid-level size comparison mechanism needed for a size-contrast account, the premise of BCIT explains the illusion in terms of low-level representations of contours: Nearby contours are attracted, whereas distant contours repel each other, hence the effect is “biphasic.” Roberts and colleagues’ (2005) findings are consistent with this theory, with their data showing a tendency for Ebbinghaus inducers to make the target appear smaller as target-inducer dis- tance is increased. They found a similar effect with the Delboeuf illusion. Relatedly, Sherman and Chouinard (2016) showed a correlation between the Delboeuf illusion and Ebbinghaus illusion. They argue this is incompatible with a size-contrast account because the ring in the Delboeuf illusion is always larger than the target. Todorović and Jovanović (2018) addressed the size-contrast theory more directly by using a novel stimulus. They found that increasing the number of small inducers in an Ebbinghaus configuration can counterintuitively eliminate the illusion if the target is embedded in a grid of inducers (also see Jaeger & Klahs, 2015). Their finding disputes a size-contrast account that predicts more inducers to amplify the size contrast between the inducers and the target, while supporting a BCIT-based explanation, which posits that attractive and repellent effects of contours located at varying distances from the target cancel out. Nevertheless, BCIT does not offer a complete explanation of the Ebbinghaus illusion. BCIT cannot explain the effect of similarity between inducers and targets when the distribution of near and far contours are controlled for (Coren & Miller, 1974; Deni & Brigner, 1997). As others have noted, there are often multiple contributing factors to visual illusions (Coren & Girgus, 1978), and it is possible that the Ebbinghaus illusion may represent the outcome of several distinct processes along the visual stream. Importantly, Rose and Bressan (2002) replicated the similarity effect and further showed that illusion magnitude was boosted only when both inducers and targets were circles or triangles, but not hexagons or irregular angular shapes. This disputes size contrast is based on the sum of Euclidean distances between contours. Considering this, contour-interaction seems to be a neces- sary but insufficient factor in the Ebbingaus illusion. As Rose and Bressan and others (Coren & Girgus, 1978; Schwarzkopf, 2015) have pointed out, the Ebbinghaus illusion may incorporate non-linear effects arising from top-down feedback, or even multiple contributing mechanisms. theories, as well as any contour-interaction account the complete explanation of that 462 Perception 52(7) Neural Correlates of the Ebbinghaus Illusion Converging evidence suggests that the effect of inducers on perceived size of the target is mediated by processes located in V1. Illusion magnitude was reduced—but not abolished—when inducers and target were shown to separate eyes (Song et al., 2011); indicative of a cortical mechanism in V1 where there are still many monocular neurons, although this cannot rule out a contribution from higher visual areas. Additionally, Schwarzkopf et al. (2011) used functional magnetic reson- ance imaging (fMRI) and retinotopic mapping to show that functional primary visual cortex (V1) surface area can predict Ebbinghaus PSE. V1’s selectivity for local contrast edges makes it a likely candidate site for mediating low-level interactions as posited by the contour-interaction account. They used the classical Ebbinghaus illusion, where observers judged the difference in target size between a large-inducer and small-inducer configuration. In a follow-up study, Schwarzkopf and Rees (2013) also found a correlation between V1 area and the PSEs for large and small inducers tested separately. Both Ebbinghaus configurations made the target appear relatively larger in indi- viduals with small V1s. The authors surmised this may indicate the effect of local circuits within V1, which are contingent on cortical distance. This could represent an attenuation of the effects of these circuits at greater distances or because of the time taken by those signals to propagate. Moreover, while large inducers reliably made the target appear smaller, small inducers made the target appear larger for some and smaller for other observers depending on their V1 surface area. Relatedly, Moutsiana et al. (2016) found that the expansive effect of the Delboeuf illusion is enhanced when it is encoded by larger population receptive fields (pRFs) in V1 (Harvey & Dumoulin, 2011). This was true both within observers with variation of pRF size across the visual field, and between individuals with different pRF sizes. While not able to explain the repul- sive effect between contours, these results and those of Schwarzkopf et al. (2011) can be concep- tualized as indicative of an antagonistic center-surround field of local interactions that defines a gradient of modulation based on cortical distance (Schwarzkopf, 2015). When target-inducer dis- tance is small, there is an attractive effect, and the target appears larger. Conversely, when the dis- tance is large the repulsive effect dominates, and the target appears smaller. In between is a point of equilibrium where inducers would have neither an attractive nor repulsive effect. The sign change of the illusion with small inducers across observers is consistent with this theory. In the Current Work While Schwarzkopf and Rees (2013) and prior work (Schwarzkopf et al., 2011) has shown evi- dence that the Ebbinghaus illusion depends on between-subject differences in cortical topography, the present work looks at the effect of varying cortical target-inducer distance within individuals. As such, we manipulate cortical distance in two ways: by varying the retinal target-inducer distance in visual space (Experiment 1), and by varying the eccentricity of stimuli when target-inducer distance is constant (Experiment 2). If proximity of contours affects the illusion as described by Schwarzkopf and Rees (2013) then reduced cortical distance will modulate the illusion so that the target appears larger. Furthermore, an account of the Ebbinghaus illusion based on cortical distance should also explain the difference in PSEs between large- and small-inducer Ebbinghaus configurations. Proponents of contour-based accounts (Sherman & Chouinard, 2016; Todorović & Jovanović , 2018; Weintraub, 1979) claim that large inducers cause a repulsive effect because they possess both near and far contours, while small inducers do not. In Experiments 3a and 3b we investigate this claim by using single- and double-ring configurations of the Delboeuf illusion. To draw further comparisons with the Ebbinghaus illusion, we also varied the retinal distance between targets and surround as in Experiment 1. If it is true that large inducers in the Ebbinghaus illusion cause Urale and Schwarzkopf 463 Figure 2. Trial procedure for Experiments 1, 2, 3a, and 3b. (A) Edge-to-edge inducer/ring distance(s) from reference target across experiments. (B) Trial sequence for Experiments 1 and 3a/3b. Observers maintained fixation on a cross before being shown reference and test intervals. The target in the test interval changed according to a staircase procedure. The order of these two intervals was counterbalanced across the experiment. Following these observers made a size comparison of targets in the two intervals. (C) Trial sequence for Experiment 2. This was like the other two experiments, except the first interval was preceded by an exogenous cue that indicated where the stimuli would appear, and gaze position was monitored using eye-tracking. Stimuli in the first and second intervals appeared above and below the horizontal meridian, respectively. repulsion because of the antagonistic effect of near and far contours, then we should observe a similar effect with the addition of a second ring in the Delboeuf illusion (Figure 2). Experiment 1 In Experiment 1 we varied the retinal distance between the target and the inducer. Varying the retinal distance entails changes in distances between representations of visual elements across the visual stream. This is evident in topographically organized areas such as V1 and V2, where we would expect an increase of cortical distance between representations with retinal distance. Our study here is a conceptual replication of the study by Roberts et al. (2005), who varied the dis- tance between inducers/annuli and the target for the Ebbinghaus and Delboeuf illusions, respect- ively. In that study, both inducers and annuli made the target appear smaller at farther distances compared to closer distances. More recently, Knol et al. (2015) also varied the Ebbinghaus illusion along various dimensions, including target size, inducer size, and target-inducer distance, finding enlargement in cases where small- or medium-sized targets (∼0.5° and ∼1°) were displayed with inducers at short distances. In our study we used a similar manipulation with the addition of some key differences. Firstly, we included shorter target-inducer distances compared to both studies. In Roberts and colleagues’ study, the closest target-inducer distance was 1.9° for small 464 Perception 52(7) inducers, and 2.53° for large inducers. Our study used a minimum distance of 0.14° for both inducer types. Furthermore, Roberts and colleagues limited the target-inducer distance since a closer dis- tance would require overlap between large inducers. In our study, we allowed inducers to overlap to achieve a short target-inducer distance. The inclusion of this smallest distance tests a key hypothesis posited by Schwarzkopf and Rees (2013) who proposed that large inducers could make the target appear larger if sufficiently close. Secondly, we showed stimuli close to fixation in two temporal intervals. Most other studies testing the Ebbinghaus illusion typically use a 2-alter- native forced-choice task (like ours, but where two stimuli are presented simultaneously side by side (in opposite hemifields) and are either flashed briefly (Schwarzkopf & Rees, 2013; Song et al., 2013) or remain on screen until the observer responds (Knol et al., 2015; Roberts et al., 2005; Todorović & Jovanović , 2018). Presenting stimuli in close proximity in separate intervals removes the need to split attention across the two stimulus locations and reduces the possibility of crowding effects of peripherally located stimuli. Materials and Methods Participants. We recruited 12 observers (8 females, age range 21–50), all with normal or corrected-to-normal visual acuity. Observers provided written and informed consent and all proce- dures were approved by the University of Auckland Human Participants Ethics Committee (UAHPEC). Experimental Setup. Stimuli were displayed on a 621 × 341 mm LCD monitor (Expt-1: Dell, S2817Q, USA; Expt-2: Samsung, U28D590D, South Korea), at a resolution of 3840 × 2160 × 8-bit resolution running 60 Hz. Monitors were linearized in software based on measurements made with a photometer (LS100, Konica Minolta, Japan). Stimuli were generated using program- ming environment MATLAB (version 2017B, MathWorks Inc.) and Psychtoolbox 3 (Brainard, 1997; Kleiner et al., 2007; Pelli, 1997) using customized scripts. Observers’ heads were stabilized with a chin rest. Stimuli. A single trial consisted of two stimuli: a reference stimulus, consisting of a target (always 0.56° diameter) and a surround, depending on the condition of the given trial, and a test stimulus, which consisted of only a target which varied in size according to an adaptive staircase procedure (see below). Stimuli were presented on a grey (175 cd/m2) background. Targets were always filled, white circles (341 cd/m2) inducers were white outlined circles with a ∼0.08° stroke and had dia- meters of 0.84° and 0.2° for large and small inducers, respectively. Example stimuli can be seen in Figure 3. The edge-to-edge distance between the target and inducers (target-inducer distance) could be one of seven possible distances: 0.14°, 0.42°, 0.7°, 1.13°, 1.55°, 2.25°, and 4.5°. In add- ition, there was a control condition without inducers. All Ebbinghaus configurations in both experi- ments had eight inducers, and large inducers were allowed to overlap in conditions with very short target-inducer distances. The centers of inducers were positioned at evenly spaced radial positions relative to the target ranging from 0° to 315° in steps of 45°. Targets in each interval were shown at a horizontally offset position relative to fixation (see “Procedure” section), so in the 0.42° condition some individual inducers above and below the target overlapped the vertical meridian. Procedure. Observers completed a single session lasting roughly 45 min seated in a darkened room seated at a distance of 82 cm from the screen. Observers were given a brief verbal description of the task prior to commencement of testing. They were told to maintain fixation on a cross (0.05° × 0.05°) in the center of the monitor through- out each block. Blocks consisted of 100 trials and were separated by a rest period of at least 30 s. All Urale and Schwarzkopf 465 Figure 3. Example stimuli at various target-inducer edge-to-edge distances. (A) Large inducer Ebbinghaus configuration at 0.14° and (B) 4.5°. (C) Small inducer configuration 0.14° and (D) 4.5°. (E) Delboeuf configuration with single ring at 0.14° and (F) 4.5°. (G) Two ring Delboeuf configuration, at 0.14°. stimuli were presented on a half-tone background. On each trial, stimuli were displayed sequen- tially with the target centered at 0.42° either left or right of fixation. The first interval always appeared just to the left of fixation, followed by the stimulus in the second interval which appeared to the right. The order of the presentation of the reference and test stimuli was decided pseudo- randomly on a per-trial basis. At the beginning of each trial, observers saw a blank (fixation only) screen for 500 milliseconds (ms) before seeing one stimulus for ∼100 ms, followed by a blank screen again for 500 ms before the final stimulus for ∼100 ms. They were told that they would be able to respond following pres- entation of all stimuli. Observers pressed the left or right keyboard button to indicate whether the left or right stimulus was larger or smaller. In alternating blocks, observers were instructed to either indicate the target that appeared larger or smaller. They were also told to ignore the inducers. In case of any prior knowledge of the Ebbinghaus illusion, observers were instructed to report on their prima facie experience instead of what they anticipated the correct answer to be. Pressing a button to indicate their response immediately began the next trial. The ratio of the test stimulus diameter relative to the reference diameter was varied using a 1-up-1-down staircase procedure. The procedure was used to determine the PSE for each condition. With two Ebbinghaus configura- tions, seven target-inducer distances, and a control condition, there were a total of 15 conditions. There were two staircases for each condition, progressing in steps on a binary logarithmic scale. We chose to use a binary logarithm because it linearizes stimulus size increments in line with Weber’s Law. Adjusting sizes in proportions, rather than a binary logarithmic scale as we do here, would be mathematically unsound as the non-linearity of the stimulus size ratios will theor- etically skew statistical and curve fitting analyses. As an example, a stimulus half the size of the reference will have a ratio of 0.5, while a stimulus of the equivalent larger size will have a ratio of 2. The arithmetic mean of these values would be 1.25 above a ratio of 1. However, these two sizes are linearly comparable when represented as binary logarithmic units, that is, −1 (2−1) and 1 (21), respectively. Moreover, in logarithmic units, 0 corresponds to the absence of an illusion (i.e., a size ratio of 1). Nevertheless, some readers might find it difficult to interpret logarithmic units; we therefore plot our results in linear units of degrees of visual angle but this is done purely for visualization. On a given trial, the size of the test stimulus in degrees of visual angle was 0.56 × 2g. The stair- case was varied by adjusting g. For each condition, one staircase began with a test diameter 0.2g 466 Perception 52(7) larger than the reference target (i.e., ∼115% of the reference target diameter), and the other 0.2g smaller (i.e., ∼87% of the reference target diameter). The step size of the staircase varied depending on the number of reversals: 0.1g for trials up until the 2nd reversal, then 0.075 until the 4th reversal, followed by 0.05 until the 8th reversal, and then 0.025 for the remaining reversals (25 in total). Trials from each of the 30 staircases were randomly interleaved and discontinued after the requisite number of reversals. The experiment ended when all staircases were complete. We calculated the PSE across conditions for each observer by fitting a cumulative Gaussian psy- chometric function to each condition using the weighted stimulus levels and responses from both staircases (R2 ≥ 0.98 for all fits for the present experiment and fits for psychometric functions in all subsequent experiments in this work). Assigned weighting to each data point was proportionate to the number of trials occurring at that stimulus level. All PSEs were taken as the 50% point of that function. To test the validity of our estimates we compared these values to PSEs calculated by taking the average size of the stimulus level during the last 8 reversals across both staircases for each condition, excluding values beyond twice the median absolute deviation in either direction. Using either method did not meaningfully change the pattern of results or conclusions of this manu- script. We chose a psychometric fit across all experiments as it is a more sensitive and theoretically grounded analysis. Results and Discussion Figure 4 shows the group-level average PSEs for Experiment 1. Prior to analysis, we subtracted the PSE in the control condition from the PSE for both inducer conditions at each distance. These base- lined PSEs were used in all subsequent analyses. For both large and small inducers, we fit a power function of the form axb + c, where a, b, and c are free parameters and x is target-inducer distance. We used a bootstrap technique to calculate the 95% confidence bands for this function by randomly selecting a sample of 12 (with replacement) from the pool of observers and then re-calculating the group means and re-fitting the power function to the new sample. This was repeated for a total of 10,000 times for each inducer type. We calculated goodness-of-fit measures for both small, R2 = .838, and large inducers, R2 = .732, as well as observed model parameters (Supplemental Table 1). A plot containing individual-observer model fits can be viewed in Supplemental Figure 1. In addition to the power function shown here, we also performed the same analysis with a two-term exponential function of the form aebx + cedx. We chose this as an alternative model because of the known exponential relationship between eccentricity and cortical magnifica- tion (Duncan & Boynton’s, 2003). This model performed well with small inducers but we chose a power model here because the exponential model performed poorly with large inducers (see Supplemental Table 2). Our results support our hypothesis that shorter target-inducer distances lead to an increase in per- ceived target size (larger, positive PSEs). For targets surrounded by small inducers, there was a clear uptick in PSEs for shorter distances. Moreover, with enough distance the sign of illusion inverted. The pattern for large inducers was more ambiguous. At all target-inducer distances, PSEs were negative, meaning the target was perceived as smaller. Importantly, our results also showed that the basic Ebbinghaus effect occurs with our novel presentation procedure where stimuli are pre- sented near the fovea in separate temporal intervals. Schwarzkopf and Rees (2013) hypothesized that at a short enough distance to the target, large inducers could make the target appear larger. We tested this by allowing large inducers to overlap and display at a distance much closer to the target compared to Roberts et al.’s (2005) study. Our results did not support this hypothesis, with a modest increase in PSE when large inducers were very close to the target. This may be due to the attractive effect of the nearer contours in large indu- cers being counteracted by contours on the far side of the inducers (Todorović & Jovanović , 2018), Urale and Schwarzkopf 467 Figure 4. Group mean PSEs across target-inducer retinal distances in Experiment 1. The horizontal dotted black line indicates the size of the reference stimulus, that is, the absence of any illusion. Solid and dashed lines are the fit to the data for the small- and large-inducer conditions, respectively. Shaded regions show the 95% bootstrapped bands for the power functions for each inducer type. Error bars indicate ±1 standard error of the mean across observers. “dva” = degrees of visual angle. but may also reflect an unanticipated effect of allowing large inducers to overlap at short distances from the target. Specifically, this would also reduce the figural similarity between the inducers and the target, which has been shown to affect the strength of the illusion (Choplin & Medin, 1999; Coren & Enns, 1993; Deni & Brigner, 1997; Jaeger & Guenzel, 2001; Rose & Bressan, 2002). Generally, our results bear important similarities and differences compared with the results of Roberts et al. (2005). Their study also found a similar pattern when increasing target-inducer dis- tance with large and small inducer conditions. However, they found a more reliable reduction in PSEs at greater target-inducer distances compared to our study. Unlike Roberts et al. we did not vary the numbers of inducers to always form a complete ring around the target, so this discrepancy may reflect lower stimulus energy due to the large distances between them. In both studies, the illu- sion for small inducers does invert at greater differences, although the crossover point for Roberts et al.’s study (∼3–3.25°) differs considerably to the crossover seen here (∼1.2°). This may be due to changes in illusion strength related to overall stimulus size, a factor shown to reliably effect illusion strength in other studies (Knol et al., 2015; Massaro & Anderson, 1971). Given that sequential presentation of the elements of the Ebbinghaus illusion can reduce the illu- sion magnitude (Jaeger & Pollack, 1977), a potential concern stems from our choice to present stimuli at nearby locations. Potentially, an afterimage from the target or inducers from the first interval could affect perception of the second interval; a persistent image of a target may enhance perceived simi- larity with a second target, and residual images of inducers may introduce an illusory effect on a lone 468 Perception 52(7) target in the second interval. However, we think these concerns are unlikely for the following reasons. Firstly, observers (including the two authors) did not report seeing afterimages. Secondly, we delib- erately offset each interval horizontally (and vertically in Experiment 2, see below), which should reduce the ability for any direct comparisons between stimuli. Thirdly, the temporal order of reference and test stimuli were counterbalanced, meaning any effect of inducers in the first interval would be counteracted by trials where the inducer condition was in the second interval. Experiment 1 supports the hypothesized relationship between distance in visual space and PSE in the Ebbinghaus illusion. Specifically, we predicted that for a given inducer type (i.e., small, large) as cortical distance between target and inducers decreases, perceived size of the target should increase. We observed this effect, albeit more clearly for small inducers. In Experiment 2 we test the relationship between the Ebbinghaus and cortical distance further by taking advantage of the change in cortical magnification across the visual field. Experiment 2 Cortical magnification in visual cortex falls off with eccentricity (Duncan & Boynton, 2003; Smith et al., 2001). Ebbinghaus stimuli at greater eccentricities therefore reduce cortical distance between representations. Chen et al. (2018) found the Ebbinghaus illusion was stronger when observers first viewed a low-spatial frequency prime compared to a high-spatial frequency prime. Sensitivity to low-spatial frequencies increases with eccentricity (Henriksson et al., 2008), and categorization of low-spatial frequency scenes elicits greater activity in brain areas associated with the peripheral visual field compared to high-spatial frequency scenes (Musel et al., 2013). We hypothesize that shorter cortical distances between the target and inducers produce an increase in perceived target size, irrespective of the inducer type. Thus, we should observe generally larger PSEs as stimuli are moved further into the periphery. The effect of eccentricity on the Ebbinghaus illusion has been investigated previously by Eymond et al. (2020). In one experiment, observers in their study compared a foveal test circle with a peripheral or foveal reference circle that was either an isolated control circle or an Ebbinghaus configuration with large inducers. They found the PSE for the Ebbinghaus condition did not differ depending on eccentricity while the control condition appeared smaller in the periphery. While this may initially seem inconsistent with our hypotheses, observers in their study compared a foveal test stimulus with a peripheral target, and the authors note there is a general reduction in perceived size when stimuli are placed into the periphery (Baldwin et al., 2016). Therefore, the lack of an effect of eccentricity on PSE in the Ebbinghaus illusion in their experiment may indicate that the effects of the inducers are counteracting a reduc- tion in perceived size in the periphery. Our experiment differs from these studies in two key ways: Firstly, we test perception of the Ebbinghaus illusion at multiple distances from fixation. This will allow us to observe graded effects of eccentricity. Secondly, targets in reference and test stimuli occurred at the same eccentric location. By doing this, account for stimuli varies in size in absolute terms across the visual field. Methods Participants. We recruited 12 observers (10 females, age range 22–53) all with normal or corrected-to-normal vision. Observers provided written and informed consent and procedures were approved by UAHPEC. Experimental Setup. We conducted Experiment 2 on the same experimental setup as Experiment 1 with the addition of an Eyelink 1000 Desktop System eye-tracker (operating at 1,000 Hz; SR Research). Urale and Schwarzkopf 469 Stimuli. The retinal dimensions of target and inducer stimuli were identical to Experiment 1. Unlike Experiment 1, in Experiment 2 retinal target-inducer distances were fixed while we manipulated the location of the Ebbinghaus stimuli along the visual field’s horizontal meridian. To avoid crowding, we applied Bouma’s law (Bouma, 1970; Pelli & Tillman, 2008), which states that the absence of visual crowding effect can be achieved if the retinal distance between the two visual elements is no less than 50% of the distance between these elements and fixation. Thus, the centers of target stimuli were positioned at a maximum distance of 4.5° from fixation, dictating a suitable target-inducer distance of 2.25°. This distance was used for both large and small-inducer configura- tions. To avoid influence from attentional capture on each trial, the presentation of the stimuli was preceded by a primer stimulus to alert the observer to the location of the forthcoming stimuli. Procedure. The procedure for Experiment 2 was mostly the same as Experiment 1, except that the retinal distance between target and inducers was kept constant while the distance between the loca- tion of the target and foveal vision was manipulated. Observers sat in a dimly lit room where they positioned their head on a headrest and chinrest apparatus located in front of a computer monitor where they performed a 9-point calibration routine for the eye-tracker. The experimenter verbally instructed observers to maintain fixation on the fixation cross located in the center of the screen, that the eye-tracker was tracking their eyes, and to try to avoid blinking during stimulus presentation time. Each trial began with presentation of a fixation cross. On a given trial the test and reference target stimuli could either occur at fixation or at an eccentric location close to the horizontal meridian. Eccentric locations could occur either to the left or right hemifield. Exogenous cueing can affect perceived size in the objects in the periphery (Kirsch et al., 2020), so to avoid any extraneous effects of attentional re-orienting, we ensured that attentional allocation was consistent across con- ditions. We did this with an exogenous cueing stimulus: If on the current trial the target and test targets were to appear at an eccentric location, they were preceded by “×” shaped cue (0.3° × 0.3°) at the location of the forthcoming reference and test stimuli. This cue appeared for 100 ms, followed by a 500 ms interval of only the fixation cross again, followed by the reference and test intervals. The order of test and reference intervals was pseudo-randomly determined on a trial-by-trial basis. Just as in Experiment 1, the reference target stimulus could be either the large- or small-inducer Ebbinghaus configurations or the control stimulus with no inducers, each with a 0.56° diameter target circle. The test stimulus varied according to the same staircase proced- ure described in Experiment 1. Each interval lasted 100 ms, separated by a 500 ms interval. Unlike Experiment 1, we offset the location of reference and test target in a vertical (rather than horizontal) orientation to avoid extraneous effects of one stimulus occurring at a more central loca- tion than the other. Thus, the center of the target circles in the first and second interval always appeared 0.2° above and below the horizontal meridian, respectively. We used the eye-tracker to ensure observers were always looking at fixation during presentation of the reference and test stimuli: A trial would be aborted if, during the reference and stimulus inter- vals, the observer blinked or if their gaze was tracked as deviating more than 1° from fixation. We performed a single-point drift-correction procedure between each 100-trial block. If the reference and test intervals ran to completion, observers were again shown a fixation cross while they responded by pressing a keyboard button to indicate whether they thought the target in the first (top) or second (bottom) interval was larger or smaller, depending on the instructions of the current block. This response period was untimed and giving a response would immediately initiate the next trial. Consequently, observers were asked to blink and orient their gaze to the fixation cross before giving their response. If the trial was aborted due to blinking or looking away from fixation, the screen would show the fixation cross for 500 ms before initiating the next trial. 470 Perception 52(7) There were 12 conditions in total with a 3 × 4 design: three types of inducer conditions (large, small, no inducers) and four eccentricities (0°, 1.69°, 2.8°, and 4.5°). There were two staircases for each of these conditions that operated as in Experiment 1. A given staircase ended after 25 reversals and the whole experiment ended when all staircases reached completion. Each block of trials ended after 100 trials or if all staircases were completed, ending the experiment. We calculated PSEs for each condition using the same procedure as Experiment 1. Results and Discussion As before, for a given eccentricity we subtracted the PSE for the control condition from the PSE for both inducer conditions. These baselined PSEs were used in all subsequent analyses. Figure 5 shows the group-level average PSEs for Experiment 2. Our analysis was to investigate whether PSEs increased or decreased with target eccentricity, and for this purpose we determined a linear function of function used in Experiment 1. We fit this to the small, R2 = .732, and large, R2 = .839, inducer conditions. Confidence bounds were generated using the same procedure as Experiment 1 and can be found in Supplemental Table 1. Generally, the target appeared larger as target-fixation distance increased, the form y = a + bx as appropriate, than the power rather Figure 5. Group mean PSEs (illusion magnitude) across target-fixation retinal distances (eccentricity) in Experiment 2 (units as in Figure 4). The horizontal dotted black line indicates the size of the reference stimulus, that is, the absence of any illusion. Shaded regions show the 95% bootstrapped bands for the linear fit to both types of inducers. Solid and dashed lines show fit to small- and large-inducer conditions, respectively. Error bars indicate ±1 standard error of the mean across observers. “dva” = degrees of visual angle. Urale and Schwarzkopf 471 although this effect was not observed to the same extent in large inducers. We see this in the con- fidence interval for the slope parameter for large inducers, b = 0.009 (95% CI [0.025, −0.01]), which overlapped zero. This indicates that there is no clear direction (either positive or negative) of the slope representing the relationship between eccentricity and PSE in the large-inducer condi- tion, and that the slope itself is close to zero. However, the interval for small inducers did not cross zero, b = 0.031 (95% CI [0.0445, 0.0166]), indicating a reliable positive relationship between eccentricity and PSE as determined by our bootstrap procedure. We also observed that for some observers the PSE for small inducers switched sign as target-fixation distance increased, in line with our findings while increasing target-inducer distance in Experiment 1. Cortical Distance and PSE. We looked at the effect of cortical distance on the Ebbinghaus illusion. Our approach takes inspiration from Mareschal et al.’s (2010) investigation of the effect of cortical distance on the tilt illusion. The tilt illusion (Gibson, 1937) is an illusion where the perceived tilt of a target line is influenced by the angle of surrounding lines. Mareschal and colleagues estimated cortical distance between target and surround across various retinal distances and concluded that the strength of the tilt illusion increases with cortical proximity. In a similar way, estimates of cor- tical distance allow us to investigate the relationship between cortical distance and PSE in Experiments 1 and 2. To do this, we chose to estimate linear cortical magnification factor (M ), which is the millimeters of cortex per degree of visual angle (Daniel & Whitteridge, 1961), using Duncan and Boynton’s (2003) formula: M = 9.81 × δ−.083, where δ denotes eccentricity in degrees of visual angle. By subtracting M between two different points (see Mareschal et al., 2010), we can estimate of the cortical distance between inducers and target across inducer condi- tions and experiments. For stimuli presented at fixation, all inducers in each configuration were equidistant to the target both in terms of visual space and cortical distance. However, when stimuli were presented at parafoveal locations in Experiment 2, the distances between individual inducers and the target were asymmetric; for example, cortical distance from the target is greater for the inducers positioned closer to fixation compared to the more peripherally located inducers. To capture these variations, we calculated an index of cortical distance for each condition based on the average edge-to-edge cortical distance between the nearest edge of all eight inducers and the target. We plot the estimates of cortical distance against PSE in Figure 6. We used a bootstrap method to plot confidence bands by taking 10,000 resamples (with replacement) of observers’ PSEs across both experiments. For the observed and each iteration of the bootstrapped data, we fit a linear func- tion of the form y = a + bx, where a is the intercept, and b is the slope coefficient. This was per- formed separately for both target-fixation distance and estimated cortical distance. We chose a linear function as a parsimonious way to characterize a simple relationship between two variables. Cortical distance predicted PSE for both small inducers, R2 = .876, and large inducers, R2 = .304. For the relationship between cortical distance and PSE, slope coefficients (b) for the large and small inducers were −0.01 (95% CI [−0.003, −0.017]) and −0.022 (95% CI [−0.016, −0.028]), respectively. We also ran the same procedure with a Difference-of-Gaussians (DoG) model (see Equation 1), in accordance with our theoretical expectations and to maintain consistency with the analysis in Experiments 3a and 3b (see below). The goodness-of-fit and parameter values (including confidence intervals derived from the bootstrap procedure) can be found in Supplemental Table 3. Upon visual inspection and comparison of goodness-of-fit estimates, we determined that the linear model was a better fit to the data from Experiments 1 and 2. This could be because the data points in the Ebbinghaus experiments fell within the steep portion of this function. Encouragingly, the two separate methods of manipulating cortical distance between target and inducers had comparable effects on the illusion strength. We observed agreement between PSEs 472 Perception 52(7) Figure 6. Group mean PSE as a function of estimated cortical distance. The horizontal dashed black line indicates the size of the reference stimulus, that is, the absence of any illusion. Small and large inducers are shown as circles and triangles, and PSEs from Experiment 1 and 2 are denoted by open and filled symbols, respectively. The size of the filled symbols denotes eccentricity in Experiment 2. Error bars indicate standard error (1±) of the mean across observers. Confidence bounds show the 95% upper and lower bounds of the line fit, produced from the bootstrap procedure. “dva” = degrees of visual angle, “mm” = millimeters. across the two experiments in conditions with similar cortical distance estimates, particularly for the small-inducer condition. Specifically, in Figure 6, markers at a similar position on the x axis, irre- spective of experiment, have similar PSEs. We observed a negative correlation between cortical distance and Ebbinghaus PSE in both large and small inducers, such that smaller cortical distance corresponded to larger perceived target size. These findings are consistent with the predictions based on previous neuroimaging work showing that smaller cortical extents associate with larger PSEs (Schwarzkopf et al., 2011) and especially perceptually larger stimuli (Schwarzkopf & Rees, 2013). Moreover, the shallower slope seen in the large inducer condition mirrors the results from Experiment 1. This may reflect non-linear interactions between the target and inducers due to antagonistic effects of the near and far contours in large inducers (Todorović & Jovanović , 2018). Mareschal et al.’s (2010) study also described opponent processes, which, in the context of the tilt illusion, were antagonistic “repulsive” and “assimilative” forces. These same mechanisms may account for repulsion and attraction in the Ebbinghaus illusion. Experiments 3a and 3b In the next experiment, we investigate why large and small inducers have contrasting effects on perceived target size. We saw support in Experiments 1 and 2 for the link between PSE and Urale and Schwarzkopf 473 estimated cortical distance, but they also replicated the different perceptual effects of large and small inducers. As these results and others (Roberts et al., 2005; Sherman & Chouinard, 2016; Todorović & Jovanović , 2018) have shown, this disparity is unlikely to be caused by a hypothetical size-contrast effect originating in mid or high-level vision. An alternative account is that these dif- ferences are driven by opponent processes which depend on the spatial (or cortical) extent of con- tours around the target. As others have stated (Rose & Bressan, 2002), contour-based accounts offer an incomplete account of the Ebbinghaus illusion, but it may be necessary. Proponents of accounts such as BCIT hold that this difference can be explained in terms of low-level contour interactions (Jaeger, 1978; Jaeger & Grasso, 1993; Sherman & Chouinard, 2016; Todorović & Jovanović , 2018; Weintraub, 1979; Weintraub et al., 1969; Weintraub & Schneck, 1986). The “biphasic” element of BCIT (Sherman & Chouinard, 2016) stipulates that contours nearer to the target have an attractive effect, while contours at more distance locations repel the target. Thus, the reason large inducers make the target look smaller is because large inducers have additional contours at farther distances from the target. An alternative account for these differences sees them as driven by higher level- categorization of inducers as whole objects beyond simple size contrast (Knol et al., 2015; Rose & Bressan, 2002). We test the effect of additional contours with the Delboeuf illusion (Delboeuf, 1865; Evans, 1995), an illusion in which perceived target size is affected by the proximity of a ring surrounding the target. The Delboeuf illusion is suitable for this purpose because it likely shares a common mechanism with the Ebbinghaus illusion. Supporting this, both Pressey (1977) and Sherman and Chouinard (2016) found that the two illusions share around a quarter of their variability. In another study, Roberts et al. (2005) found that a complete ring comprised of small Ebbinghaus inducers had the same illusory effect as a Delboeuf ring at a range of distances from the target. Accordingly, if negative PSEs associated with large inducers in the Ebbinghaus illusion are due to near and far contour placement, and if the Delboeuf and Ebbinghaus share a common mechan- ism, the simple addition of another ring in the Delboeuf illusion (Figure 3) should resemble the effect of large inducers in the Ebbinghaus illusion, such that we observe a downward shift in PSE. We test this hypothesis in Experiment 3a (henceforth “3a”). In Experiment 3a we observed that PSE did not trend towards zero with greater target-ring dis- tances (see “Results and discussion” section). The interaction between the surround and the target in the Ebbinghaus illusion has been conceptualized as sombrero-shaped center-surround of contextual interactions (Schwarzkopf, 2015), and in such a model we would expect the contextual effects to diminish to zero as it approaches the “brim” of the hat (i.e., the boundaries of any suppressive effect). To this end, in Experiment 3b (henceforth “3b”) we increased the ring-target distance further to observe if its effect on the target attenuates at even farther target-ring distances. Methods Participants. We recruited 12 volunteers (seven females, age range 22–49) for 3a and 14 volunteers (9 females, 21–52) for 3b, all with normal or corrected-to-normal vision. Observers provided written and informed consent, and procedures were approved by UAHPEC. Experimental Setup. The experimental setup for 3a was identical to Experiment 1. In order to cover an area of the visual field ∼40° in diameter, we reduced the viewing distance 3b from 82 to 42 cm. Stimuli. In both experiments, the reference-interval stimuli consisted of a central circle (0.56° in diameter) with either a single- or double-ring configuration in 3a (see Figure 3) and a single-ring only condition in 3b. Rings in both experiments had a thickness of ∼0.04°. 474 Perception 52(7) For 3a, on a given trial the inner-ring of the double-ring condition could be one of several dis- tances from the edge of the central circle: 0.14°, 0.44°, .7°, 1.13°, 1.55°, 2.25°, and 4.5°. The dis- tances were the same for the single-ring configuration, with the addition of a 5.34° condition (i.e., the distance of the outer ring in the double-ring condition at 4.5°). The distance between the borders of the inner and outer rings was always 0.84°. We chose this distance to match the diameter of the large inducers from Experiment 1, and in doing so emulate the antagonistic effects of near and far contours in those stimuli. Experiment 3b featured the single ring conditions at the following dis- tances: 0.14°, 11°, 15°, and 20°. Procedure. The procedures for both experiments were the same as Experiment 1 (see Figure 2). Observers typically completed the experiment in 45 min for Experiment 3a, and 20 min for 3b. Results and Discussion Prior to analysis, we again subtracted the PSE from the control condition from all other conditions and used those baselined PSEs for all subsequent analyses. Figure 7 shows the group-level average PSEs for 3a and 3b as a function of target-ring distance in degrees of visual angle and estimated cortical distance. We also plotted data from Experiments 3a and 3b separately, as PSE as a function of retinal distance and using the same power function used in Experiment 1 (see Supplemental Figure 5 for plots, and goodness-of-fit and parameter estimates in Supplemental Table 1). In 3a we used the Delboeuf illusion to determine if the difference between large and small indu- cers in the Ebbinghaus illusion are explainable as an interplay of near and far contours (Sherman & Chouinard, 2016; Todorović & Jovanović , 2018). On this basis our results support our hypothesis; compared to a single ring, the two-ring configuration had the effect of numerically reducing the PSE (Figure 7). Looking at Figure 7b, this downward shift is most pronounced for cortical distances between 2 and 8 mm. Biphasic-contour interaction theory explains this as a result of antagonistic effects of contours, with farther contours working to repulse the percept of the target. At the largest cortical distances, the two rings probably fall within the same large peripheral receptive fields. Thus, the two-ring condition may effectively be a single-ring condition at these distances, only with somewhat increased stimulus energy. Our results also agree with earlier research, albeit with updated psychophysical methods. For example, with a target-inducer distance compar- able to our own, Weintraub and Schneck (1986) observed that the target appeared larger when only the inner-arc of large inducers were visible, but that PSE decreased and eventually changed sign as the outer fragments of those inducers were filled in with successively more dots. This resembles the addition of the outside contour in 3a, which we observed as shifting the PSE for most target-ring distances. Similarities aside, the effects of the two ring conditions in this experiment and those of small and large inducers in Experiment 1 (Figure 4) are not an exact match. This may be explained in terms of a contour-based account, as these two illusions differ in terms of variations in stimulus energy (i.e., the amount of contour). Unlike Roberts et al. (2005), we did not modify the number of inducers to maintain an uninterrupted surround of inducers in our experiments, meaning that at all distances the Delbouef presented a more complete surround than in Experiment 1. By contrast, Roberts and col- leagues found that an uninterrupted surround of small inducers affected perceived target size almost identically to a Delboeuf ring at various distances from the target. The two conditions in 3a had the same effect at very close distances (0.14°), yet there was a total separation of PSEs between the large- and small-inducer conditions at that same distance in Experiment 1. This, too, may be attrib- utable to a difference in stimulus energy because of intermediary contours between the nearest and farthest edges in large inducers, which are absent in the Delboeuf illusion. Alternatively, the differ- ence we observe between these two illusions may indicate a contribution of a second mechanism, Urale and Schwarzkopf 475 Figure 7. Group mean PSEs across target-ring distances in Experiments 3a and 3b as a function of retinal distance (A) and estimated cortical distance (B). The horizontal dashed black line indicates the size of the reference stimulus, that is, the absence of any illusion. Solid and dashed colored lines show the Difference-of-Gaussians (DoG) function fitted to the data for the two-ring and single-ring conditions, respectively. Shaded regions show the 95% bootstrapped bands for DoG for each inducer type. Error bars indicate ±1 standard error of the mean across observers. “dva” = degrees of visual angle, “mm” = millimeters. possibly located higher in the visual stream. We discuss these possibilities further in the General discussion. DoG Function. We modelled the relationship between PSE and retinal and cortical distance, respect- ively, using a DoG function. DoG has been used to model inhibitory signals in extra-classical recep- tive fields (Cavanaugh et al., 2002), and it can account for the antagonistic center-surround proposed by Schwarzkopf (2015) and in BCIT (Sherman & Chouinard, 2016; Todorović & Jovanović , 2018; Weintraub & Schneck, 1986). To model a center-surround, we calculated the 476 Perception 52(7) difference of two Gaussian functions with peaks at zero distance. This took for form of Equation 1, where values σa, σb, a, and b, are left as free parameters. The DoG function is also used here to investigate whether the relationship between cortical distance and PSE resembles the hypothesized profile of an antagonistic center-surround mechanism. We fit this function using a least-squares pro- cedure and generated 95% confidence bounds using a bootstrap technique with 10,000 repetitions. A single function was fit to the combined PSEs from the single-ring condition in Experiments 3a and 3b, and another on the two-ring condition from 3a. For PSE as a function of retinal distance we fit DoG functions for small ring, R2 = .955, and large inducer conditions, R2 = .99. For PSE as a function of cortical distance we did the same for the small ring, R2 = .974, and large ring conditions, R2 = .997. f (x) = ae −x2 a − be 2σ2 −x2 2σ2 b (1) Equation 1: DoG function We observe something closely resembling the sagittal cross-section of a sombrero, as described by Schwarzkopf (2015), with an excitatory center and inhibitory surround (Cavanaugh et al., 2002). Due to a theoretical interest relating to the cortical point image (see General discussion), we also calculated zero-crossing for the two functions in Figure 7B (PSE as a function of cortical distance), with 95% confidence bounds generated from the bootstrapping procedure. That crossing was 5.57 mm (95% CI [6.53 4.25]) for the small ring and 3.65 mm (95% CI [4.76 2.98]) for large ring condition. An implication of a putative center-surround zone of interaction (Schwarzkopf, 2015) is a non- monotonic relationship between cortical distance and PSE. Theoretically, the effect should reduce to zero at sufficiently large cortical distances as the interaction between contours drops off. Mareschal et al. (2010) found such a “sombrero” pattern when varying cortical distance in the tilt illusion. In our study, even the farthest distances in 3b that did not return to zero. In aggregate, PSEs for distances between 11° and 20° (the three filled green squares on the right side of Figure 7) averaged slightly closer to zero in log units, that is, no illusion (M = −0.083, SE = 0.029), than the farthest single-ring condition in 3a (5.34°) (M = −0.123, SE = 0.024), although the difference between these was not significant, t(24) = 1.045, p = .307. Hence, the PSEs across 3a and 3b trend towards zero, but the cortical distance necessary to see this is evidently beyond the dimen- sions we measured. We note that the results from 3b are not inconsistent with a contour-based account. Specifically, because of cortical magnification (Duncan & Boynton, 2003), the large dis- tances used in 3b translate to only minor differences in cortical distance compared to 3a (see Figure 7). General Discussion Numerous studies and many theories have been put forward to explain the Ebbinghaus and Delboeuf illusions (as many as 10; for a review, see Robinson, 1998). Despite that, little is known about what lies behind the illusion, and where in the brain that mechanism occurs. The present research adds to a growing body of research pointing to striate cortex as a promising loca- tion for this substrate. Such work shows that V1 encodes perceived size, like in the cases of the hallway illusion (Murray et al., 2006), and retinal afterimages projected onto near and far surfaces (Sperandio et al., 2012). Additionally, there is partial interocular transfer of the Ebbinghaus illusion (Song et al., 2011), suggesting a cortical process and again implicating V1, as this region is partially monocular, although of course this could also involve monocular and binocular neurons across mul- tiple stages of the visual hierarchy (Dougherty et al., 2019). Finally, there is a correlation between PSE and between-subject variability in cortical magnification (Schwarzkopf et al., 2011; Urale and Schwarzkopf 477 Schwarzkopf & Rees, 2013). Based on these findings, Schwarzkopf and Rees (2013) and later Schwarzkopf (2015), raised the possibility that interaction between representations on the topog- raphy of V1 may be the substrate for the Ebbinghaus illusion, and that these interactions depend in part on the cortical distance between those interactions. This proposal offers a plausible neural basis for contour-based accounts of the Ebbinghaus illusions, in which the low-level inter- actions between contours underlies the effect. To query this theory further, in the first two experi- ments we tested the effect of cortical magnification on the Ebbinghaus illusion. Experiment 1 replicated previous work showing that shorter retinal distances between targets and Ebbinghaus inducers increases PSE (Knol et al., 2015; Roberts et al., 2005). Following this, Experiment 2 showed that for small inducers PSE increases when the target is positioned more per- ipherally. The prevailing trend in both experiments is consistent with the predictions that (1) in a general sense, the Ebbinghaus illusion depends in part on cortical magnification, and the specific prediction that (2) PSE correlates negatively with cortical distance. We were able to show conver- ging evidence to support this claim by manipulating cortical distances in two ways: firstly, by adjusting the retinal distance between target and inducer (Experiment 1), and secondly, by taking advantage of a reduction in cortical magnification across the visual field and displaying targets in the periphery (Experiment 2). This culminated in the combined results of both experi- ments and comparing PSE against the estimated cortical distance between targets and inducers (Figure 6). In Experiment 3a, we used the Delboeuf illusion to show that, compared to a single-ring condition, a two-ring condition produced a perceptual shrinkage of the target (PSEs shifted down), lending support to a biphasic-contour account where more distal contours cause a decrease in the perceived size of the target. The attractive or repulsive effect of Ebbinghaus-style inducers may depend on a cortical distance equivalent to the cortical point image. The point image is the cortical representation of a single point stimulus expressed in millimeters (Mcllwain, 1986), calculated as the product of the cortical mag- nification factor and receptive field size. Harvey and Dumoulin (2011) used pRF mapping to show that the cortical point image is near constant in V1, with only small decreases with eccentricity (similar findings are reported in non-human primates (Palmer et al., 2012). That is, in V1, there is a constancy in the ratio between receptive field size and cortical magnification. Looking at the Ebbinghaus illusion, Schwarzkopf and Rees (2013) found that perceived enlargement of the target in a small-inducer Ebbinghaus configuration occurred in observers with a relatively small V1. Using similar calculations to those used here (Duncan & Boynton, 2003), they estimated the inducer condition) to be cortical distance between target and inducer in their study (small ∼3.3 mm, falling within the 3–4 mm range of given for point images in V1 (Harvey & Dumoulin, 2011). Schwarzkopf and Rees surmised that the stability of the point image across the cortex indicates “constancy in spatial extent of cortical responses” (p. 11), and that the critical cortical distance between target and inducers in order for a perceived enlargement (i.e., attraction between contours) to occur might be equivalent to the cortical point image. Of interest here is whether a shift between attraction and repulsion occurs at a similar cortical distance in the present experiments. We estimated cortical distance at which a sign change (i.e., from perceived smaller to perceived larger) occurs as ∼7.2 mm for the small inducers in Ebbinghaus illusion (Experiment 1 & Experiment 2). These differ considerably to point image size in V1, which may reflect influences of surround completeness. Moreover, other studies have shown (Knol et al., 2015), target size influences PSE, so these differences may also reflect a difference in overall target size, which was 1.03° in Schwarzkopf and Rees’s study, compared to 0.56° in our study. Additionally, the arrangement of small inducers in their study were smaller (small inducer diameter was 26% of target diameter) compared to those used in the current study (35% of target diameter) and formed a more complete ring around the target compared to our study. The relative size of each inducer and completeness of the ring formed are known influences on PSE 478 Perception 52(7) (Roberts et al., 2005). Supporting this is that when we used the Delboeuf illusion, which consists of an uninterrupted ring, the sign change occurred between ∼5.5 and ∼3.7 mm for the single and double ring conditions, respectively (Experiment 3), which are closer to the estimates of Schwarzkopf and Rees. Thus, if the cortical point image is relevant to the magnitude with these illusions it likely interacts with overall stimulus energy. Despite our evidence for a link between cortical distance and the Ebbinghaus and Delboeuf illu- sions, conversion of retinal distances into cortical distance does not completely account for strength of illusions in other studies. As mentioned above, direct comparisons between the present experi- ments and those of Schwarzkopf and Rees (2013) and other comparable works (Knol et al., 2015; Roberts et al., 2005) are complicated by differences in stimulus dimensions and placement. For instance, Schwarzkopf and Rees (2013), Roberts et al. (2005), and Knol et al. (2015) presented ref- erence and target stimuli at a distance (distances from fixation to target center: 4.65°, ∼15.2°, and 13°, respectively) at either side of fixation, whereas we (with a few exceptions) presented stimuli at the same foveal locations at separate intervals. Compared to the experiments here, PSEs are not affected by cortical distance in the same way in those experiments. For instance, one of the condi- tions in Knol et al.’s (2015) study showed a repulsive effect of inducers (i.e., the target appeared larger) with a 0.48° target, a 1.9° target-inducer distance (measured as distance between target and inducer centers), and inducer radius of 0.09°. Considering the location of the target center in that condition (15.2° horizontal displacement from fixation) and using Duncan and Boynton’s (2003) formula, the estimated cortical distance between targets and the nearest edge of the farthest inducers along the horizontal meridian averages to 0.198 mm. This is well within the range where attraction occurred in our experiments, yet in their experiment this distance coincided with a shrink- age of perceived target size. There are several factors that may account for these differences. Firstly, the stimuli in Roberts et al. (2005), Schwarzkopf and Rees (2013), and Knol et al. (2015) may have been affected by crowding effects; all three studies chose horizontal eccentricities and target-inducer spacing was generally shorter in those experiments than what Bouma’s law (Bouma, 1970) would dictate as necessary to avoid crowding. Indeed, there is evidence that size perception (as opposed to only rec- ognition), is affected by crowding (van den Berg et al., 2007). Crowding thus likely interferes with discriminating peripheral target sizes presented in close proximity to the inducers. This would render measurements of PSEs more variable and potentially obscures other effects. Knol et al. and Schwarzkopf and Rees both failed to find substantial repulsive effects on perceived target size, yet Roberts et al., who used large horizontal displacements of their targets, reported an attract- ive effect at short target-inducer distances with small inducers, and effect resembling our findings from Experiment 1 when stimuli were presented at fixation. It is also a distinct possibility that attractive effects in the illusion and crowding share the same underlying neural mechanism. When cortical distance is sufficiently large, this would result in a perceptual overestimation of the target, but when cortical distance is too short it completely disrupts size discrimination. Finally, are potential differences originating in the task: In addition to greater horizontal displace- ment, Roberts et al., Schwarzkopf and Rees, and Knol et al. also presented stimuli simultaneously, while we presented stimuli in separate temporal intervals. We stress that a low-level contour interaction underlying two size illusions of the nature we describe here may fall short of providing a complete account of these illusions. Two decades ago, Rose and Bressan (2002) observed that research with inducers the same size or larger than the target does not have effects on perceived size consistent with a static zone of repulsion and attraction (Girgus et al., 1972; Jaeger & Grasso, 1993; Massaro & Anderson, 1971; Weintraub, 1979). This is reflected in the current study, with the large inducer condition being less responsive than small inducers to manipulations of retinal size and eccentricity in Experiments 1 and 2, respect- ively. A constant gradient of interaction between contours would not explain this discrepancy Urale and Schwarzkopf 479 between conditions, even when considering potential antagonistic effects between relatively near and far contours. This is even more apparent when comparing Experiment 1 (Ebbinghaus illusion) with Experiment 3 (Delboeuf illusion). We have suggested that these inconsistencies may be attrib- utable to less consistent spacing (and thus differences in stimulus energy) of Ebbinghaus inducers versus the continuous ring in the Delboeuf illusion, but mid- and high-level cognitive factors may also fill this explanatory gap. Last but not least, several studies suggest that cognitive factors modulate the strength of size illu- sions, including the Ebbinghaus illusion. For example, Gestalt grouping of a surround reduces PSE in the Ebbinghaus illusion (Rashal et al., 2020), and several studies have observed that size is affected by figural similarity between inducers and targets, even while controlling for the proximity and distribution of contours (Choplin & Medin, 1999; Coren & Enns, 1993; Deni & Brigner, 1997; Jaeger & Guenzel, 2001; Rose & Bressan, 2002). These figural effects have commonly been explained in terms of object-level categorization and attention. Indeed, attention in other contexts has been known to affect perceived size; Kirsch et al. (2020) found that a peripheral target appears small while attending to a central target. Fang et al. (2008) used fMRI to show that spatial distri- bution of activity in V1 reflected the perceived size of two attended targets embedded in the hallway illusion, and that this activity was significantly diminished when attention was narrowed by a demanding central task. There are also reports that semantic knowledge affects the Ebbinghaus illu- sion, with objects of a known size biasing their perceived size when surrounded by Ebbinghaus-style inducers (Hughes & Fernandez-Duque, 2010). This relates to findings that ventral temporal cortex (Konkle & Oliva, 2012) show selectivity for real-world size of objects, independent of image transformations. Consistent with known models of predictive processing across various stages of visual processing (Ballard & Jehee, 2012; Rao & Ballard, 1999), these size-encoded representations could drive feedback signals that boost predicted signals in earlier visual areas in V1 consistent with these predictions. Thus, while a mechanism that involves spa- tially contingent interactions between low-level representations shows promise in explaining some of the known characteristics of the Ebbinghaus and Delboeuf illusions, more work is needed to determine whether contour-interaction alone can explain these illusions or if there is a need for the addition of other (potentially cognitive) mechanisms. Conclusion In addition to showing a relationship between cortical distance and the Ebbinghaus illusion (Schwarzkopf & Rees, 2013), our results broadly support biphasic contour-based accounts of the Ebbinghaus and Delboeuf illusions. Specifically, shorter cortical distances between inducers and target in the Ebbinghaus illusion, whether due to retinal distance (Experiment 1) or cortical mag- nification (Experiment 2), associate with perceptual enlargement of the target. We did not confirm the prediction that large Ebbinghaus inducers produce perceptual enlargement at short dis- tances, a finding potentially due to repulsive effects of distal contours (on the far side of the inducer relative to the target) counteracting attractive effects of nearer contours. We noted that predictions based on estimated cortical distance did not align with select findings from other studies (Knol et al., 2015; Roberts et al., 2005; Schwarzkopf & Rees, 2013), possibly due to differences in stimu- lus dimensions, stimulus location, and task design. In the Delboeuf illusion, we showed that the addition of a second, more distant contour reliably decreases perceived target size compared to a single-ring, a finding aligned with an antagonistic surround described in contour-based accounts, such as BCIT (Roberts et al., 2005; Sherman & Chouinard, 2016; Todorović & Jovanović , 2018; Weintraub & Schneck, 1986). Lastly, we found that at large retinal distances (>11°), a single-ring Delboeuf ring still decreased perceived target size, a finding that may reflect low cortical magnification (the relatively minor changes in cortical distance) at points ranging into peripheral 480 Perception 52(7) vision. Future studies should continue to characterize effects of the surround in these illusions, including interactions between surround elements, potential influences of task design, and contri- butions of mid- and high-level vision. Author contribution(s) Poutasi W. B. Urale: Conceptualization; Formal analysis; Investigation; Methodology; Project administra- tion; Software; Visualization; Writing – original draft; Writing – review & editing. D. Samuel Schwarzkopf: Conceptualization; Formal analysis; Investigation; Methodology; Resources; Software; Supervision; Writing – review & editing. Declaration of Conflicting Interests The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publica- tion of this article. Funding The authors received no financial support for the research, authorship, and/or publication of this article. Data Availability Data and scripts used to conduct this analysis can be viewed at Open Science Framework: Data and analysis for Effects of cortical distance on the Ebbinghaus and Delboeuf illusions. https://doi.org/10.17605/OSF.IO/ GUHSF. ORCID iDs Poutasi W. B. Urale Dietrich Samuel Schwarzkopf https://orcid.org/0000-0001-6106-2297 https://orcid.org/0000-0003-3686-1622 Supplemental Material Supplemental material for this article is available online. References Aglioti, S., DeSouza, J. F. X., & Goodale, M. A. (1995). Size-contrast illusions deceive the eye but not the hand. Current Biology, 5, 679–685. https://doi.org/10.1016/S0960-9822(95)00133-3 Baldwin, J., Burleigh, A., Pepperell, R., & Ruta, N. (2016). The perceived size and shape of objects in periph- eral vision. I-Perception, 7, 2041669516661900. https://doi.org/10.1177/2041669516661900 Ballard, D. H., & Jehee, J. (2012). Dynamic coding of signed quantities in cortical feedback circuits. Frontiers in Psychology, 3, 254. https://doi.org/10.3389/FPSYG.2012.00254/BIBTEX Bouma, H. (1970). Interaction effects in parafoveal letter recognition. Nature, 226, 177–178. https://doi.org/10. 1038/226177a0 Brainard, D. H. (1997). The Psychophysics Toolbox. Spatial Vision, 10, 433–436. http://color.psych.ucsb.edu/ psychtoolbox Cavanaugh, J. R., Bair, W., & Movshon, A. J. (2002). Nature and interaction of signals from the receptive field center and surround in macaque V1 neurons. Journal of Neurophysiology, 88, 2530–2546. https://doi.org/ 10.1152/jn.00692.2001 Chen, L., Qiao, C., & Jiang, Y. (2018). Low-spatial-frequency bias in context-dependent visual size percep- tion. Journal of Vision, 18, 2–9. https://doi.org/10.1167/18.8.2 Choplin, J. M., & Medin, D. L. (1999). Similarity of the perimeters in the Ebbinghaus illusion. Perception & Psychophysics, 61, 3–12. https://doi.org/10.3758/BF03211944 Clifford, C. W. G. (2014). The tilt illusion: Phenomenology and functional implications. Vision Research, 104, 3–11. https://doi.org/10.1016/J.VISRES.2014.06.009 Urale and Schwarzkopf 481 Coren, S., & Enns, J. T. (1993). Size contrast as a function of conceptual similarity between test and inducers. Perception & Psychophysics, 54, 579–588. https://doi.org/10.3758/BF03211782 Coren, S., & Girgus, J. S. (1978). Seeing is deceiving: The psychology of visual illusions. Routledge. Coren, S., & Miller, J. (1974). Size contrast as a function of figural similarity. Perception & Psychophysics, 16, 355–357. https://doi.org/10.3758/BF03203955 Daniel, P. M., & Whitteridge, D. (1961). The representation of the visual field on the cerebral cortex in monkeys. The Journal of Physiology, 159, 203–221. https://doi.org/10.1113/JPHYSIOL.1961.SP006803 Delboeuf, F. J. (1865). Note sur certaines illusions d’optique: Essai d’une théorie psychophysique de la maniere dont l’oeil apprécie les distances et les angles. Bulletins de l’Académie Royale Des Sciences, Lettres et Beaux-Arts de Belgique, 19, 195–216. Deni, J. R., & Brigner, W. L. (1997). Ebbinghaus illusion: Effect of figural similarity upon magnitude of illu- sion when context elements are equal in perceived size. Perceptual and Motor Skills, 84, 1171–1175. https://doi.org/10.2466/pms.1997.84.3c.1171 Dougherty, K., Cox, M. A., Westerberg, J. A., & Maier, A. (2019). Binocular modulation of monocular V1 neurons. Current Biology, 29, 381–391.e4. https://doi.org/10.1016/J.CUB.2018.12.004 Duncan, R. O., & Boynton, G. M. (2003). Cortical magnification within human primary visual cortex corre- lates with acuity thresholds. Neuron, 38, 659–671. https://doi.org/10.1016/S0896-6273(03)00265-4 Ebbinghaus, H. (1902). The principles of psychology. Veit, Leipzig, 437. Evans, R. B. (1995). Joseph Delboeuf on visual illusions: A historical sketch. The American Journal of Psychology, 108, 563–574. https://doi.org/10.2307/1423073 Eymond, C., Seidel Malkinson, T., & Naccache, L. (2020). Learning to see the Ebbinghaus illusion in the per- iphery reveals a top-down stabilization of size perception across the visual field. Scientific Reports, 10, 1– 10. https://doi.org/10.1038/s41598-020-69329-9 Fang, F., Boyaci, H., Kersten, D., & Murray, S. O. (2008). Attention-dependent representation of a size illusion in human V1. Current Biology, 18, 1707–1712. https://doi.org/10.1016/j.cub.2008.09.025 Gibson, J. J. (1937). Adaptation, after-effect, and contrast in the perception of tilted lines. II. Simultaneous contrast and the areal restriction of the after-effect. Journal of Experimental Psychology, 20, 553–569. https://doi.org/10.1037/H0057585 Girgus, J. S., Coren, S., & Agdern, M. (1972). The interrelationship between the Ebbinghaus and Delboeuf illusions. Journal of Experimental Psychology, 95, 453–455. https://doi.org/10.1037/H0033606 Haffenden, A. M., Schiff, K. C., & Goodale, M. A. (2001). The dissociation between perception and action in the Ebbinghaus illusion: Nonillusory effects of pictorial cues on grasp. Current Biology, 11, 177–181. https://doi.org/10.1016/S0960-9822(01)00023-9 Harvey, B. M., & Dumoulin, S. O. (2011). The relationship between cortical magnification factor and popu- lation receptive field size in human visual Cortex: Constancies in cortical architecture. Journal of Neuroscience, 31, 13604–13612. https://doi.org/10.1523/JNEUROSCI.2572-11.2011 Henriksson, L., Nurminen, L., Hyvärinen, A., & Vanni, S. (2008). Spatial frequency tuning in human retino- topic visual areas. Journal of Vision, 8, 5–5. https://doi.org/10.1167/8.10.5 Hughes, M., & Fernandez-Duque, D. (2010). Knowledge influences perception: Evidence from the Ebbinghaus illusion. Journal of Vision, 10, 954–954. https://doi.org/10.1167/10.7.954 Jaeger, T. (1978). Ebbinghaus illusions: Size contrast or contour interaction phenomena? Perception & Psychophysics, 24, 337–342. https://doi.org/10.3758/BF03204250 Jaeger, T., & Grasso, K. (1993). Contour lightness and separation effects in the Ebbinghaus illusion. Perceptual and Motor Skills, 76, 255–258. https://doi.org/10.2466/pms.1993.76.1.255 Jaeger, T., & Guenzel, N. (2001). Similarity and lightness effects in Ebbinghaus illusion created by keyboard characters. Perceptual and Motor Skills, 92, 151–156. https://doi.org/10.2466/pms.2001.92.1.151 Jaeger, T., & Klahs, K. (2015). The Ebbinghaus illusion: New contextual effects and theoretical considera- tions. Perceptual and Motor Skills, 120, 177–182. https://doi.org/10.2466/24.27.PMS.120v13x4 Jaeger, T., & Long, S. (2007). Effects of contour proximity and lightness on Delboeuf illusions created by cir- cumscribed letters. Perceptual and Motor Skills, 105, 253–260. https://doi.org/10.2466/PMS.105.1.253- 260 Jaeger, T., & Lorden, R. (1980). Delboeuf illusions: Contour or size detector interactions? Perceptual and Motor Skills, 50, 376–378. https://doi.org/10.1177/003151258005000205 482 Perception 52(7) Jaeger, T., & Pollack, R. H. (1977). Effect of contrast level and temporal order on the Ebbinghaus circles illu- sion. Perception & Psychophysics, 21, 83–87. https://doi.org/10.3758/BF03199473 Kirsch, W., Pfister, R., & Kunde, W. (2020). On why objects appear smaller in the visual periphery. Psychological Science, 31, 88–96. https://doi.org/10.1177/0956797619892624 Kleiner, M., Brainard, D., & Pelli, D. G. (2007). What’s new in Psychtoolbox-3? Perception 36 ECVP Abstract Supplement. http://www.psychtoolbox.org. Knol, H., Huys, R., Sarrazin, J.-C., & Jirsa, V. K. (2015). Quantifying the Ebbinghaus figure effect: Target size, context size, and target-context distance determine the presence and direction of the illusion. Frontiers in Psychology, 6, 1679. https://doi.org/10.3389/fpsyg.2015.01679 Konkle, T., & Oliva, A. (2012). A real-world size organization of object responses in occipitotemporal cortex. Neuron, 74, 1114–1124. https://doi.org/10.1016/J.NEURON.2012.04.036 Mareschal, I., Morgan, M. J., & Solomon, J. A. (2010). Cortical distance determines whether flankers cause crowding or the tilt illusion. Journal of Vision, 10, 13–13. https://doi.org/10.1167/10.8.13 Massaro, D. W., & Anderson, N. H. (1970). A test of a perspective theory of geometrical illusions. Source: The American Journal of Psychology, 83, 567–575. Massaro, D. W., & Anderson, N. H. (1971). Judgmental model of the Ebbinghaus illusion. Journal of Experimental Psychology, 89, 147–151. Mcllwain, J. T. (1986). Point images in the visual system: New interest in an old idea. Trends in Neurosciences, 9, 354–358. https://doi.org/10.1016/0166-2236(86)90113-X Moutsiana, C., de Haas, B., Papageorgiou, A., van Dijk, J. A., Balraj, A., Greenwood, J. A., & Schwarzkopf, D. S. (2016). Cortical idiosyncrasies predict the perception of object size. Nature Communications, 7, 1–12. https://doi.org/10.1038/ncomms12110 Murray, S. O., Boyaci, H., & Kersten, D. (2006). The representation of perceived angular size in human primary visual cortex. Nature Neuroscience, 9, 429–434. https://doi.org/10.1038/nn1641 Musel, B., Bordier, C., Dojat, M., Pichat, C., Chokron, S., le Bas, J.-F., & Peyrin, C. (2013). Retinotopic and lateralized processing of spatial frequencies in human visual cortex during scene categorization. Journal of Cognitive Neuroscience, 25, 1315–1331. https://doi.org/10.1162/JOCN_A_00397 Obonai, T. (1954). Induction effects in estimates of extent. Journal of Experimental Psychology, 47, 57–60. https://doi.org/10.1037/H0057223 Palmer, C. R., Chen, Y., & Seidemann, E. (2012). Uniform spatial spread of population activity in primate parafoveal V1. Journal of Neurophysiology, 107, 1857–1867. https://doi.org/https://doi.org/https:// doi.org/ 10.1152/jn.00117.2011 Pelli, D. G. (1997). The VideoToolbox software for visual psychophysics: Transforming numbers into movies. Spatial Vision, 10, 437–442. Pelli, D. G., & Tillman, K. A. (2008). The uncrowded window of object recognition. Nature Neuroscience, 11, 1129–1135. https://doi.org/10.1038/nn.2187 Pressey, A. W. (1977). Measuring the Titchener circles and Delboeuf illusions with the method of adjustment. Bulletin of the Psychonomic Society, 10, 118–120. https://doi.org/10.3758/BF03329298 Rao, R. P. N., & Ballard, D. H. (1999). Predictive coding in the visual cortex: A functional interpretation of some extra-classical receptive-field effects. Nature Neuroscience, 2, 79–87. https://doi.org/10.1038/4580 Rashal, E., Cretenoud, A. F., & Herzog, M. H. (2020). Perceptual grouping leads to objecthood effects in the Ebbinghaus illusion. Journal of Vision, 20, 1–15. https://doi.org/10.1167/JOV.20.8.11 Roberts, B., Harris, M. G., & Yates, T. A. (2005). The roles of inducer size and distance in the Ebbinghaus illusion (Titchener circles). Perception, 34, 847–856. https://doi.org/10.1068/p5273 Robinson, J. O. (1998). The Psychology of Visual Illusion. Hutchinson and Co. Lt. Rose, D., & Bressan, P. (2002). Going round in circles: Shape effects in the Ebbinghaus illusion. Spatial Vision, 15, 191–203. Schwarzkopf, D. S. (2015). Where is size in the brain of the beholder? Multisensory Research, 28, 285–296. https://doi.org/10.1163/22134808-00002474 Schwarzkopf, D. S., & Rees, G. (2013). Subjective size perception depends on central visual cortical magni- fication in human V1. PLoS ONE, 8, e60550. https://doi.org/10.1371/journal.pone.0060550 Schwarzkopf, D. S., Song, C., & Rees, G. (2011). The surface area of human V1 predicts the subjective experi- ence of object size. Nature Neuroscience, 14, 28–30. https://doi.org/10.1038/nn.2706 Urale and Schwarzkopf 483 Sherman, J. A., & Chouinard, P. A. (2016). Attractive contours of the Ebbinghaus illusion. Perceptual and Motor Skills, 122, 88–95. https://doi.org/10.1177/0031512515626632 Smith, A. T., Singh, K. D., Williams, A. L., & Greenlee, M. W. (2001). Estimating receptive field size from fMRI data in human striate and extrastriate visual cortex. Cerebral Cortex, 11, 1182–1190. https://doi.org/ 10.1093/cercor/11.12.1182 Song, C., Samuel Schwarzkopf, D., Lutti, A., Li, B., Kanai, R., & Rees, G. (2013). Effective connectivity within human primary visual cortex predicts interindividual diversity in illusory perception. Journal of Neuroscience, 33, 18781–18791. https://doi.org/10.1523/JNEUROSCI.4201-12.2013 Song, C., Schwarzkopf, D. S., & Rees, G. (2011). Interocular induction of illusory size perception. BMC Neuroscience, 12, 27. https://doi.org/10.1186/1471-2202-12-27 Sperandio, I., Lak, A., & Goodale, M. A. (2012). Afterimage size is modulated by size-contrast illusions. Journal of Vision, 12, 18–18. https://doi.org/10.1167/12.2.18 Titchener, E. B. (1905). Experimental psychology: A manual of laboratory practice (Vol. 2). Johnson Reprint Company. Todorović , D., & Jovanović , L. (2018). Is the Ebbinghaus illusion a size contrast illusion? Acta Psychologica, 185, 180–187. https://doi.org/10.1016/J.ACTPSY.2018.02.011 van den Berg, R., Roerdink, J. B. T. M., & Cornelissen, F. W. (2007). On the generality of crowding: Visual crowding in size, saturation, and hue compared to orientation. Journal of Vision, 7, 14–14. https://doi.org/ 10.1167/7.2.14 Weintraub, D. J. (1979). Ebbinghaus illusion: Context, contour, and age influence the judged size of a circle amidst circles. Journal of Experimental Psychology: Human Perception and Performance, 5, 353–364. https://doi.org/10.1037/0096-1523.5.2.353 Weintraub, D. J., & Schneck, M. K. (1986). Fragments of Delboeuf and Ebbinghaus illusions: Contour/context explorations of misjudged circle size. Perception & Psychophysics, 40, 147–158. https://doi.org/10.3758/ BF03203010 Weintraub, D. J., Wilson, B. A., Greene, R. D., & Palmquist, M. J. (1969). Delboeuf illusion: Displacement versus diameter, arc deletions, and brightness contrast. Journal of Experimental Psychology, 80, 505–511. https://doi.org/10.1037/h0027424 Yamazaki, Y., Otsuka, Y., Kanazawa, S., Yamaguchi, M. K., & Kanazawa, S. O. (2010). Perception of the Ebbinghaus illusion in 5- to 8-month-old infants. Japanese Psychological Research, 52, 33–40. https:// doi.org/10.1111/j.1468-5884.2009.00420.x
null
10.1038_s42003-023-04955-3.pdf
Data availability The main data supporting the results in this study are available within the paper and its Supplementary Information. Source data for all figures can be found in Supplementary Data 1. The raw and analysed datasets generated during the study are too large to be publicly shared, yet they are available for research purposes from the corresponding authors on reasonable request.
Data availability The main data supporting the results in this study are available within the paper and its Supplementary Information. Source data for all figures can be found in Supplementary Data 1. The raw and analysed datasets generated during the study are too large to be publicly shared, yet they are available for research purposes from the corresponding authors on reasonable request.
ARTICLE https://doi.org/10.1038/s42003-023-04955-3 OPEN A spike-targeting bispecific T cell engager strategy provides dual layer protection against SARS-CoV-2 infection in vivo Fanlin Li1,2,6, Wei Xu3,6, Xiaoqing Zhang1,4,6, Wanting Wang1,2,6, Shan Su3, Ping Han1,2, Haiyong Wang1,2, Yanqin Xu1,2, Min Li1,2, Lilv Fan1,2, Huihui Zhang1,2, Qiang Dai1,2, Hao Lin1,2, Xinyue Qi1,2, Jie Liang1,2, Xin Wang5, & Xuanming Yang Shibo Jiang 3, Youhua Xie , Lu Lu 1,2✉ 3✉ 3✉ ; , : ) ( 0 9 8 7 6 5 4 3 2 1 Neutralizing antibodies exert a potent inhibitory effect on viral entry; however, they are less effective in therapeutic models than in prophylactic models, presumably because of their limited efficacy in eliminating virus-producing cells via Fc-mediated cytotoxicity. Herein, we present a SARS-CoV-2 spike-targeting bispecific T-cell engager (S-BiTE) strategy for con- trolling SARS-CoV-2 infection. This approach blocks the entry of free virus into permissive cells by competing with membrane receptors and eliminates virus-infected cells via powerful T cell-mediated cytotoxicity. S-BiTE is effective against both the original and Delta variant of SARS-CoV2 with similar efficacy, suggesting its potential application against immune- escaping variants. In addition, in humanized mouse model with live SARS-COV-2 infection, S-BiTE treated mice showed significantly less viral load than neutralization only treated group. The S-BiTE strategy may have broad applications in combating other coronavirus infections. 1 Sheng Yushou Center of Cell Biology and Immunology, School of Life Sciences and Biotechnology, Shanghai Jiao Tong University, Shanghai 200240, China. 2 Joint International Research Laboratory of Metabolic & Developmental Sciences, Shanghai Jiao Tong University, Shanghai 200240, China. 3 Key Laboratory of Medical Molecular Virology (MOE/NHC/CAMS), Shanghai Institute of Infectious Disease and Biosecurity, School of Basic Medical Sciences and Biosafety Level 3 Laboratory, Fudan University, Shanghai 200032, China. 4 Department of Physiology, Naval Medical University, Shanghai 200433, China. 5 Shanghai Longyao Biotechnology Limited, Shanghai 201203, China. 6These authors contributed equally: Fanlin Li, Wei Xu, Xiaoqing Zhang, Wanting Wang. ✉ email: yhxie@fudan.edu.cn; lul@fudan.edu.cn; xuanmingyang@sjtu.edu.cn COMMUNICATIONS BIOLOGY | (2023) 6:592 | https://doi.org/10.1038/s42003-023-04955-3 | www.nature.com/commsbio 1 ARTICLE COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-023-04955-3 The emergence of the novel human coronavirus SARS-CoV- 2 has caused a worldwide pandemic of coronavirus disease 2019 (COVID-19), hampering health and economic sys- tems globally1–3. Different vaccination formats have been demonstrated to confer protection against the original or mutated SARS-CoV-2 strains at different efficacies. The mRNA-based, adenoviral vector-based, and inactivated virus-based vaccines are reported to be 95%, 66.9%, and 65.9% effective, respectively, in preventing COVID-194–6. Despite the availability of these working vaccines, the emergence of immune-escaping variants significantly slowed the controlling of the pandemic7,8. the efficacy, bispecific Various small molecules targeting SARS-CoV-2 entry or repli- cation in cells are under evaluation9 and some of them have been authorized for clinic usage10. Among them, Paxlovid has shown clinical benefit of reducing the risk of progression to severe COVID-19 or death11–13. Neutralizing antibodies have been used respiratory syncytial virus and Ebolavirus clinically against disease14,15, and also quickly been developed against SARS-CoV-2 infection to exert potent neutralizing activity in preclinical models and clinic15–19. Angiotensin-converting enzyme 2 (ACE2) is a key entry receptor for SARS-CoV-21. Several studies have reported the antiviral effects of soluble ACE2-based therapeutics, functioning as a competitive inhibitor of membranous ACE220–23. To further enhance two epitopes24–27 and bispecific fusion proteins with antibody arm and soluble ACE2 arm28,29 have been developed to enhance neu- tralization to prevent escaping mutation of SARS-CoV-2. However, owing to the high mutation rate of SARS-CoV-2, these therapeutics will face challenges of treatment-escape or resistance eventually. Furthermore, spike mutations have been reported to be associated with the reduced neutralization ability of monoclonal antibodies and convalescent individuals30–38. Therefore, to achieve the optimal therapeutic potential, a novel approach, in addition to the utilization of neu- tralizing antibodies and small molecular inhibitors, needs to be developed. To address this need, in this paper, we present a SARS- CoV-2 spike-targeting bispecific T-cell engager (S-BiTE) strategy for controlling SARS-CoV-2 infection. serum antibodies vaccinated antibodies targeting and in Results Generation and characterization of S-BiTE. T cells are con- sidered to be the most effective immune cells in eliminating virus- infected cells or cancer cells39,40. BiTE, a potent T cell-activation strategy, has been widely used for treating various types of cancer41, but little is known about its effect on SARS-CoV-2 infection. Thus, we designed a fusion protein, S-BiTE, consisting of the ACE2 extracellular domain (amino acid 1–740) to block viral entry and the anti-CD3ε single-chain variable fragment (scFv) to activate T cells and eliminate viral-producing cells (Fig. 1a and supplementary Fig. 1a). SARS-CoV-2 enters permissive cells through ACE2 in a SARS-CoV-2 spike-dependent manner1,42. Owing to the specific and strong interaction between ACE2 and SARS- CoV-2 spike43, we used the extracellular domain of ACE2 as a specific ligand to identify spike-expressing cells, which mimic SARS-CoV-2-infected cells in vivo. This monovalent ACE2 extracellular domain has a relatively high affinity for the receptor binding domain (RBD) of spike-Fc fusion protein and spike-expressing 293-spike cells (Fig. 1b, c). The other portion of the fusion protein is the monovalent anti-CD3ε scFv portion that showed a significantly reduced affinity for CD3ε compared with the parental bivalent anti-CD3 antibody (Fig. 1d and Supplementary Fig. 1b), which disfavors the binding and the SARS-CoV-2 activation of T cells in the absence of spike44,45. The ACE2 portion of S-BiTE could function as a competitive receptor and entry blocker of SARS-CoV-2. To test this hypothesis, we performed a pseudotyped SARS-CoV-2 blockade assay in vitro and found that S-BiTE significantly blocked pseudotyped SARS-CoV-2 infection in permissive 293-ACE2 and A549-ACE2 cells (Fig. 1e–f). S-BiTE induced target-dependent T cell activation and cyto- toxicity in the presence of the SARS-CoV-2 spike. The ability of S-BiTE to activate T cells was investigated in an in vitro T cell- activation assay with co-cultured cells engineered to express the SARS-CoV-2 spike. S-BiTE specifically activated T cells to release IFN-γ and TNF in the presence of 293-spike cells in a dose- dependent manner (Fig. 2a, b and Supplementary Fig. 2a, b). In the same experimental setting, there was no T cell activation in response to the spike-negative 293 cells even with the highest tested concentration of S-BiTE, suggesting the high specificity of the ACE2-spike interaction. The same specific and sensitive T cell activation was observed in response to spike-expressing lung epithelial A549 cells (Fig. 2c, d and Supplementary Fig. 2c, d). To further investigate whether the increased T cell activation could lead to death of the target cells, we performed a flow cytometry- based cell-killing assay. S-BiTE induced strong cytotoxicity towards the spike-expressing 293 and A549 cells, rather than towards the corresponding spike-negative control cells (Fig. 2e, f, Supplementary Fig. 2e, f, and Supplementary Fig. 3). To further confirm its specifity, we established one CD20 targeting BiTE and performed similar in vitro T cell activation assay, S-BiTE induce more T cell activation than control CD20-BiTE dependent on spike expression (Supplementary Fig. 5). These results demon- strate that S-BiTE can induce CD3-mediated activation of human T cells and the killing of SARS-CoV-2 spike-expressing cells. on into entry antibodies Numerous studies have shown a strong inhibitory effect of neutralizing permissive viral cells14,15,46,47. However, only a few therapeutic neutralizing antibodies have been approved for clinical use48. The clinical efficacy of neutralizing antibodies may be limited by the emergence of viral mutations that enable escape from neutraliza- tion or the relatively weak ability of neutralizing antibodies to eliminate virus-infected cells. A recent publication showed that neutralization antibodies are less effective when administered 2 h after infection compared to their administration 24 h before infection49. Thus, we compared the cytotoxicity of S-BiTE and ACE2-human IgG1 Fc fusion proteins in vitro. Although IgG1 Fc is the most potent isotype in mediating antibody-dependent cell- mediated cytotoxicity (ADCC) and complement-dependent cytotoxicity (CDC)50, ACE2-Fc showed weak cytotoxic effects through ADCC and CDC in vitro (Fig. 2g, h). In the same setting, S-BiTE exhibited more than 2000 times higher killing effect on spike-expressing Raji cells (Fig. 2i). S-BiTE inhibited pseudotyped SARS-CoV-2 viral release. The elimination of virus-infected cells to prevent the production of more viruses is critical for viral control, especially in the early stages of infection. Thus, we assessed the ability of S-BiTE to prevent virus release in our pseudotyped SARS-CoV-2 produc- tion system. To mimic virus-producing cells, 293 cells were transfected with four viral component-encoding plasmids (Fig. 3a). Upon co-culture with the engineered 293 cells, S-BiTE significantly triggered the activation of T cells, killing of viral- producing cells, and reduction of virus release (Fig. 3b–d). Importantly, under these conditions, the presence of free viruses in the culture medium did not affect S-BiTE-mediated T cell activation and cytotoxicity towards the spike-expressing cells. the pseudotyped virus in the supernatant was Consistently, 2 COMMUNICATIONS BIOLOGY | (2023) 6:592 | https://doi.org/10.1038/s42003-023-04955-3 | www.nature.com/commsbio COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-023-04955-3 ARTICLE c I F M a d e S-BiTE T cell S-BiTE Permissive cell Neutraliza(cid:2)on of free virus Virus producing cell Elimina(cid:2)on of viral source ) M n 5 0 4 ( D O b f Fig. 1 Generation and characterization of the S-BiTE fusion protein. a A schematic diagram of the potential anti-SARS-CoV-2 mechanism of the spike- targeting bispecific T cell engager (S-BiTE) used in this study. b ELISA-binding curves of S-BiTE to immobilized RBD of the SARS-CoV-2 spike. c MFI of the binding of S-BiTE to 293-spike cells as determined by flow cytometry. d Median fluorescence intensity (MFI) of the binding of S-BiTE or parental anti-CD3 to primary human T cells. e, f Lentiviruses pseudotyped with the SARS-CoV-2 spike were incubated with 293-ACE2 cells (e) or A549-ACE2 cells (f) in the presence of indicated concentration of S-BiTE. Fluorescent IRFP-positive cells were measured by flow cytometry. Relative infection was calculated as ratio of the IRFP readout in the presence of S-BiTE to the IRFP readout in the absence of S-BiTE. One-way ANOVA with Dunnett’s multiple comparison and correction was performed and significance was shown. All data shown as mean ± SEM. Representative results from one of three repeated experiments are shown (n = 3/group) (b–f). significantly reduced after S-BiTE treatment (Fig. 3e, f). Collec- tively, these results indicate that S-BiTE is superior to the Fc- directed Ab therapeutic strategy in controlling viral infection at the cellular level. S-BiTE eliminated spike-expressing cells in vivo with good safety profile. Given the potent in vitro cytotoxicity of S-BiTE towards spike-expressing cells, its cytotoxicity was also assessed in vivo. Similar to that of other reported BiTE-like molecules, the half-life of S-BiTE in mice is approximately 1.5 h (Supplementary Fig. 4). Despite its short half-life, a single treatment of S-BiTE was demonstrated to kill spike-positive target cells to a significant extent in the in vivo cell-killing assay, suggesting the high potential of S-BiTE to kill virus-infected cells in vivo (Fig. 4a and Supplementary Fig. 6). To test whether S-BiTE could induce unwanted T cell activation or T cell depletion, which is critical for its safety profile. hACE2 and hCD3e humanized mice were adminstrated with S-BiTE. Our data showed that there is no significant difference in immune cell subtypes, T cell activation marker and major tissue immune cell infiltration (Supplementary Fig. 7). Furthermore, since mesenchymal stem cells (MSCs) have been treating various diseases and have been widely used for demonstrated to have good safety profiles in clinical settings51, in order to meet urgent clinical needs, we engineered MSCs to stably secrete S-BiTE (Fig. 4b). Serum expression of S-BiTE in mice lasted for more than 14 days after a single infusion (Fig. 4c). Importantly, when we checked the bio-distribution of MSCs, both engineered and unmodified MSCs were preferentially located in the lungs (Fig. 4d, e), suggesting their potential application in treating SARS-CoV-2-induced pneumonia. This MSC-based S- BiTE delivery approach provides a potential clinical ready option for the treatment of severe COVID-19 patients. COMMUNICATIONS BIOLOGY | (2023) 6:592 | https://doi.org/10.1038/s42003-023-04955-3 | www.nature.com/commsbio 3 ARTICLE COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-023-04955-3 a c g b d h e f i 0 1 10 100 1000 0 1 10 100 1000 Fig. 2 S-BiTE induced target-dependent T cell activation and cytotoxicity in the presence of spike. 293, 293-spike, A549, A549-spike, Raji, or Raji-spike cells were co-cultured with human primary T cells in the presence of the indicated concentration of S-BiTE. a–d After 24 h, levels of IFN-γ and TNF in the cell supernatant were analyzed by the CBA assay. e, f After 48 h, cytotoxicity was determined by measuring CD45- cells via flow cytometry. h, i Raji or Raji- spike cells were co-cultured with human primary NK cells and supplemented with human AB serum or human T cells in the presence of the indicated concentration of ACE2-Fc or S-BiTE. ADCC, CDC, or T cell-mediated cytotoxicity was analyzed by flow cytometry. Two-way ANOVA with Dunnett’s multiple comparison and correction was performed and significance was shown. All data shown as mean ± SEM. Representative results from one of three repeated experiments are shown (n = 3/group) (a–i). S-BiTE eliminated live virus-infected cells. Given the potential cytotoxicity of S-BiTE in eliminating spike-expressing cell lines, we further tested the efficacy of S-BiTE on live virus-infected cells. To establish infection, we infected A549-ACE2 cells with live SARS-CoV-2 for 2 h. Then, S-BiTE was added to the infected cells in the presence of T cells. Consistent with the results of the pseudovirus assay, S-BiTE significantly inhibited viral replication in permissive A549-ACE2 cells (Fig. 5a, b and Supplementary Fig. 5). S-BiTE is effective against the Delta variant of SARS-CoV-2. Owing to its high mutation rate, SARS-CoV-2 has evolved rapidly, leading to the emergence of many variants with immune- escape ability or enhanced proliferation and transmission cap- abilities worldwide. Among these variants, the Delta variant has been reported to increase the viral load in patients, contributing to the rapid global spread of this variant52. The efficacy of neu- tralization antibodies against the Delta variant has been reported to be 3 to 5 times lower than that against the original SARS-CoV- 2 strain30,38. Thus, we investigated whether S-BiTE is effective against the Delta variant spike. We first tested the binding of S-BiTE to Delta-spike-expressing cells. S-BiTE bound to the 293-Delta-spike at an EC50 of 3.45 nM, which is similar to that observed for the WT spike (Fig. 6a). We then investigated whether S-BiTE could cause the lysis of Delta-spike-expressing cells in the presence of T cells. S-BiTE could specifically activate T cells to release IFN-γ and TNF in the presence of 293-Delta-spike cells or A549-Delta-spike cells in a dose-dependent manner (Fig. 6b, c). Consistent with the increased T cell activation, S-BiTE also induced strong cytotoxi- city towards Delta-spike-expressing 293 and A549 cells, rather than towards the corresponding spike-negative control cells (Fig. 6d, e). These results demonstrate that S-BiTE can induce CD3-mediated activation of human T cells and kill cells expressing a mutated SARS-CoV-2 spike, which may be useful for the treatment of immune-escaping variants of SARS-CoV-2. S-BiTE shows better viral control ability in vivo than soluble ACE2 neutralizing agent. To explore the protection efficacy of S-BiTE against challenge with SARS-CoV2 Delta-variant virus in vivo, hACE2-hCD3ε transgenic mice were chosen as in vivo model. In hACE2-hCD3ε transgenic mice, hACE2 provides the entry receptor for SARS-CoV2 infection, and hCD3ε provides the 4 COMMUNICATIONS BIOLOGY | (2023) 6:592 | https://doi.org/10.1038/s42003-023-04955-3 | www.nature.com/commsbio COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-023-04955-3 ARTICLE a GFP-luc Gag/Pol Spike Rev b Pseudovirus S-BiTE T cell c e S-BiTE 0 100 1000 ng/ml S-BiTE 0 100 1000 ng/ml d f P < 0.0001 P < 0.0001 50 40 30 20 10 0 S-BiTE ng/ml 0 100 1000 T cells + + + S-B TE ng/ml i T cells + + + Fig. 3 S-BiTE inhibited viral release in a pseudotyped SARS-CoV-2 production assay. a Schematic illustrating the co-culture experiments to monitor the effect of S-BiTE on virus-producing cells. b–d Lenti-X 293 T cells were transfected with pseudotyped SARS-CoV-2 plasmids. After 24 h, human primary T cells were added to the culture in the presence of the indicated concentration of S-BiTE. Forty-eight hours post transfection, levels of TNF and IFN-γ in supernatant were analyzed by the CBA assay (n = 3/group) (b). The remaining virus-releasing cells were imaged by fluorescent microscopy (c) and analyzed by flow cytometry (d) (n = 6/group). e, f The released virus in the supernatant was used to infect 293-ACE2 cells, and virus-infected GFP- expressing cells were analyzed by fluorescent microscopy (e) and flow cytometry (f) (n = 5/group). One-way ANOVA with Dunnett’s multiple comparison and correction was performed and significance was shown. Scale bar: 120 μm. All data shown as mean ± SEM. Representative results from one of three repeated experiments are shown (b–f). T cell activating target by S-BiTE. To compare the protection efficacy of S-BiTE with neutralizing agent, a soluble ACE2-treated group was included as well. The body weight of ACE2 and S-BiTE group decreased significantly less compared with the PBS group (Fig. 7a). Since the lung and intestine are two major tissues- infected by SARS-CoV2 in this transgenic mouse, the number of viral RNA copies were measured and compared at 3 dpi. The RNA copies in the lung of both ACE2 group and S-BiTE group were significantly lower than the PBS group, reducing to about 10−2 and 10−3 of PBS group, respectively (Fig. 7b, c). Impor- tantly, the RNA copies of S-BiTE group were reduced to about 10−2 of ACE2 group, suggesting the powerful protection by S- BiTE-mediated T cell activation, which is consistent with our in vitro observation. These results demonstrate that S-BiTE can induce both neutralization and CD3-mediated T cell activation in vivo, which provide dual layer protection against immune- escaping variants of SARS-CoV-2 and may be a potential treat- ment for COVID-19. Discussion This study presents S-BiTE as an attractive therapeutic strategy for treating COVID-19 and other diseases caused by cor- onaviruses that use ACE2 as their receptor. S-BiTE could block viral entry by competing with the SARS-CoV-2 spike in binding to membrane ACE2. Furthermore, S-BiTE stimulated the pow- erful cytotoxic capabilities of T cells and increased sensitivity in eliminating virus-infected spike-expressing cells. This dual COMMUNICATIONS BIOLOGY | (2023) 6:592 | https://doi.org/10.1038/s42003-023-04955-3 | www.nature.com/commsbio 5 ARTICLE a mCD45-hCD45+ CD3-CD19+ COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-023-04955-3 Spike+ 62% Spike- 37% PBS S-BiTE Spike+ 37% Spike- 61% o i t a R - / e k i p S + e k i p S b c d e Fig. 4 S-BiTE eliminated spike-expressing cells in vivo. a Approximately 2 × 106 CFSE-labeled Raji and Raji-spike cells were mixed with 5 × 106 T cells and intraperitoneally injected into NSG mice (n = 5/group). The mice were treated with PBS or S-BiTE, and cells in peritoneal cavity were collected and analyzed by flow cytometry 6 h after treatment. b The continuous S-BiTE production by stably engineered S-BiTE-MSCs was measured by ELISA after 1-month culture (n = 2/group). c S-BiTE-MSCs was injected to NSG mice (n = 6/group) and serum levels of S-BiTE were determined by ELISA at the indicated time points. d, e S-BiTE-MSCs (n = 6/group) (d) or MSCs (n = 9/group) (e) were injected into NSG mice, and bio-distribution in the indicated tissue was analyzed by q-PCR. Unpaired Student’s T-tests (a) and one-way ANOVA with Dunnett’s multiple comparison and correction (b–e) was performed and significance was shown. All data shown as mean ± SEM. a–c representative results from one of four repeated experiments are shown. d, e pooled results from three independent experiments are shown. a b r e b m u n y p o c e v i t a l e R Fig. 5 S-BiTE inhibited live SARS-CoV-2 replication in permissive cells. a, b A549-ACE2 cells were infected with live SARS-CoV-2 at an MOI of 0.05. After 2 h, free viruses were removed by washing with PBS, and human primary T cells were added to the culture in the presence of the indicated concentration of S-BiTE. At 24 h (a) and 48 h (b) post infection, SARS-CoV-2 replication in cells was analyzed by quantitative real-time PCR (n = 9/group). One-way ANOVA with and Dunnett’s multiple comparison and correction was performed and significance was shown. All data shown as mean ± SEM. Representative results from one of three experiments are shown (a, b). functional design can simultaneously prevent viral spread and reduce virus production. Compared to the widely used neutralization-antibody strategy, the S-BiTE concept has two distinct advantages. The first advantage is the use of ACE2, the entry receptor for SARS-CoV-2, as a tar- geting moiety as theoretically, no SARS-CoV-2 mutant should be able to escape the ACE2-targeting S-BiTE treatment. However, owing to the different binding epitopes of neutralization antibodies, it is possible that the new mutants of SARS-CoV-2 are able to escape the predesigned neutralization-antibody treatment; anti- bodies from vaccinated and convalescent individuals have been reported to show reduced re-neutralization abilities against muta- ted SARS-CoV-2 variants30–38. Different approaches have been used to improve the ability of neutralization antibodies against escaping variants of SARS-CoV-2, such as antibody cocktails53,54 and bispecific antibodies25,55. Targeting multiple non-overlap 6 COMMUNICATIONS BIOLOGY | (2023) 6:592 | https://doi.org/10.1038/s42003-023-04955-3 | www.nature.com/commsbio COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-023-04955-3 ARTICLE a b c I F M l / m g p - N F I l / m g p F N T 0 1 10 100 1000 0 1 10 100 1000 d e 0 1 10 100 1000 0 1 10 100 1000 ) % ( r e b m u n l l e c e v i t a l e R 0 1 10 100 1000 0 1 10 100 1000 Fig. 6 S-BiTE induced T cell activation and cytotoxicity against SARS-CoV-2 Delta-variant spike. a MFI of the binding of S-BiTE to 293-Delta spike cells was determined by flow cytometry. b–e 293, 293-Delta spike, A549, and A549-Delta spike cells were co-cultured with human primary T cells in the presence of the indicated concentration of S-BiTE. After 24 h, levels of IFN-γ and TNF in supernatant were analyzed by the CBA assay (b, c). After 48 h, cytotoxicity was determined by measuring CD45- cells via flow cytometry (d, e). Representative results from one of three experiments are shown (n = 3/ group) (a–e). Two-way ANOVA with Dunnett’s multiple comparison and correction was performed and significance was shown. All data shown as mean ± SEM. epitopes including RBD can sufficiently improve affinity against spike protein and prevent escaping variants. One unique bispe- cific design, ACE-MAB (STI-4920)28, composes of a truncated extracellular ACE2 and an antibody against different epitope of SARS-CoV-2 spike. This design can simultaneously neutralize SARS-CoV-2 by competing membrane ACE2 and block CD147 binding to reduce lung inflammation and cytokine storm. The second advantage is the use of the anti-CD3 moiety to activate T cells, which makes the difference between our strategy and monoclonal, bispecific or pre-mixed neutralization antibodies. T cells are critical in eliminating virus-infected cells and tumor cells with high sensitivity. By activating T cells, S-BiTE is much more effective in eliminating virus-infected cells than antibody- mediated cytotoxicity. In the current design, we did not include an Fc portion in S-BiTE, which resulted in a short half-life in vivo. The function of S-BiTE in vivo may be enhanced by further engineering with Fc to improve its half-life. In addition, because they involve different design concepts, it is possible to develop combinational therapy including both neutralization antibodies and S-BiTE: neutralization antibodies would be focused on preventing viral spread and S-BiTE would be focused on eliminating virus-productive source cells. A similar strategy can be applied with a combination of viral-replication inhibitors and S-BiTE. Furthermore, although we used ACE2 as a targeting moiety, we believe that antibodies against spikes can also be used as targeting moieties. Binding to conserved epitopes and binding- induced T cell activation are key in engineering antibody-based BiTE against SARS-CoV-2 infection. A similar approach has demonstrated ACE2-anti-CD3 fusion protein can trigger effec- tive CD8 T cell activation in vitro against spike-expressing cell line56. Our work has demonstrated it’s sufficient in activating T cells and its efficacy in controlling live SARS-CoV2 infection both in vitro and in vivo. One limitation of our study is the using of ACE2 transgenic mice and lack of human clinical supporting data. The ACE2 expression is under the control of ubiquitous promoter, which may not reflect the physiological distribution of ACE2. An alternative approach would involve studies in human ACE2 knock-in mice. However, because we focused on spike-mediated virus neutralization and T cell acti- vation rather than the viral life cycle, we would predict the efficacy would be similar in human ACE2 knock-in mice. Nonetheless, knowledge on the infection rate of alveolar epithelial type II cells and other ACE2+ cells in ACE2 knock-in mice at different infection stages will provide valuable insights from the perspective of evalu- ating efficacy and safety57. In summary, S-BiTE can be used in a T cell-based strategy with the capability to neutralize viruses and to eliminate virus- producing cells. Further optimization of BiTE molecule stability, target moiety selection, neutralization capability, safety, and combination strategies with anti-inflammatory therapies are warranted to improve the clinical value of the S-BiTE approach in controlling SARS-CoV-2 infection. COMMUNICATIONS BIOLOGY | (2023) 6:592 | https://doi.org/10.1038/s42003-023-04955-3 | www.nature.com/commsbio 7 COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-023-04955-3 ARTICLE a % t h g i e w y d o b e v i t a l e R b c Lung p<0.0001 p<0.0001 p<0.0001 10 12 10 11 10 10 10 9 10 8 10 7 10 6 10 5 Intestines p<0.0001 p=0.0002 p=0.0016 10 9 10 8 10 7 10 6 10 5 PBS ACE2 S-BiTE PBS ACE2 S-BiTE Fig. 7 The protection efficiency of S-BiTE against SARS-CoV2 in hACE2-hCD3ε transgenic mice in vivo. a–c All hACE2-hCD3ε transgenic mice (n = 4/ group) were challenged intranasally with SARS-CoV2 Delta-variant, and 25 μg of S-BiTE were intranasally administrated -1, 24, and 48 h post-infection. Equal mole of soluble ACE2 or equal volume of PBS was used as controls. a The body weight of mice was recorded and normalized at indicated time points. b, c The virus titer in lungs and intestines of three groups were determined at 3 dpi by qRT-PCR. Unpaired Student’s T-tests (a–c) was performed and significance was shown (n = 4/group). All data shown as mean ± SEM. Methods Cell lines and reagents. Lenti-X 293 T cells were purchased from Clontech (Mountain View, CA, USA). The A549 cell line was obtained from the American Type Culture Collection (Manassas, VA, USA). Raji cells were provided by the Stem Cell Bank, Chinese Academy of Sciences (Shanghai, China). Human PBMCs from cord blood, NK cells, and MSCs were provided by Shanghai Longyao Bio- technology Co., Ltd. (Shanghai, China). Plasmids encoding SARS-CoV-2 spike and ACE2 were obtained from Molecular Cloud (Nanjing, China) and sub-cloned into pCDH-EF lentiviral vector plasmid (System Biosciences, Mountain View, CA, USA) with the puromycin-resistance marker. To establish the SARS-CoV-2 spike-, SARS-CoV-2 Delta spike-, or ACE2- expressing cell lines, Lenti-X 293 T, A549, or Raji cells were infected with the SARS-CoV-2 spike-, SARS-CoV-2 Delta spike-, or ACE2-expressing lentivirus. After selection with puromycin, the pooled resistant cells were identified by flow cytometry analysis. The cell culture medium was supplemented with 10% heat- inactivated fetal bovine serum (FBS), 2 mmol/L L-glutamine, 100 units/mL penicillin, and 100 μg/mL streptomycin. Lenti-X 293 T cells, A549 cells, and their derivatives were cultured in complete DMEM. Raji cells and their derivatives were cultured in complete RPMI medium. Production of S-BiTE, ACE2-His, ACE2-Fc, and RBD-Fc fusion proteins. For the production of S-BiTE, ACE2-Fc, and RBD-Fc fusion proteins, DNA sequences encoding the indicated proteins were cloned into the pCDH-EF vector (System Biosciences). Plasmids containing the indicated fusion protein were transfected into Lenti-X 293 T cells, and supernatants were collected and purified using a Diamond Protein A or Ni Bestarose FF column according to the manufacturer’s protocol (Bestchrom, Shanghai, China). Neutralization assay with pseudotyped SARS-CoV-2. Lenti-X 293 T cells were transfected with lentivirus package component plasmids, Gap/pol (#12251, Addgene), RSV-Rev (#12253, Addgene), pCDH-EF-IRFP-luc, and pcDNA3.1(+)-2019-nCoV-spike-P2A-eGFP (#MC_0101087, Molecular Cloud). Supernatants containing lentivirus particles were collected 48 and 72 h post transfection for direct usage or concentration by ultracentrifugation. The viral titer in TU/mL was determined by flow cytometric analysis of the transduced 293-ACE2 cells. In the virus neutralization assay, S-BiTE was serially diluted to the indicated concentrations in complete Dulbecco’s modified Eagle medium (DMEM). Pseudotyped lentiviral particles were inoculated on 293-ACE2 or A549-ACE2 monolayers in 96-well plates in the presence of 10 μg/mL of polybrene and indicated concentration of S-BiTE, and further incubated at 37 °C for 48 h. For infecting A549-ACE2 cells, a 60-min spin at 2500 rpm at 32 °C was used to improve infection efficiency. IRFP reporter activity was measured using CytoFLEX S (Beckman Coulter). The percentage of infectivity was calculated as the ratio of the IRFP readout in the presence of the fusion protein to the IRFP readout in the absence of the fusion protein. The half-maximal inhibitory concentrations (IC50) 8 COMMUNICATIONS BIOLOGY | (2023) 6:592 | https://doi.org/10.1038/s42003-023-04955-3 | www.nature.com/commsbio COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-023-04955-3 ARTICLE were determined using a 4-parameter logistic regression (GraphPad Prism, version 8). Flow cytometry. Single-cell suspensions were stained with conjugated antibodies. Samples from in vivo experiments were pre-incubated with anti-CD16/32 (anti- FcγIII/II receptor, clone 2.4G2) for 10 min before antibody staining. All fluores- cently labeled monoclonal antibodies were purchased from Biolegend (San Diego, CA, USA) or eBioscience (San Diego, CA, USA). All fluorescently labeled sec- ondary antibodies were purchased from Jackson ImmunoResearch Laboratories (West Grove, PA, USA). Samples were analyzed on a CytoFLEX S (Beckman Coulter), and data were analyzed using FlowJo software (TreeStar, Inc.). Enzyme-linked immunosorbent assay (ELISA) and flow cytometry analysis of S-BiTE. ELISA plates (Jet Biofil, Guangzhou, China) were coated with 2 μg/mL of RBD-Fc at 4 °C overnight. Plates were washed three times with phosphate buffered saline (PBS) containing 0.05% Tween-20 and blocked with 2% FBS in PBS at room temperature for 1 h. Diluted S-BiTE-containing samples were added, and plates were incubated for 1 h at room temperature. Then, plates were washed three times and incubated for 1 h at room temperature with alkaline phosphatase (AP)-con- jugated goat anti-mouse Fab secondary antibody (Jackson ImmunoResearch Laboratories) diluted to 1:2000 in blocking buffer. AP activity was measured at 405 nm using a SpectraMax 190 microplate reader (Molecular Devices, San Jose, CA) with p-nitrophenyl phosphate (Guangzhou Howei Pharmaceutical Technol- ogy Co. Ltd., Guangzhou, China) as substrate. The half-maximum effective con- centration (EC50) binding values were calculated using GraphPad Prism version 8. 293-CD3 and human primary T cells were incubated with the indicated S-BiTE- containing samples at 4 °C for 30 min, washed three times with 2% FBS in PBS, incubated with 2 μg/mL of RBD-hFc at 4 °C for 30 min, washed three times, and incubated with 1:200 diluted Alexa Fluor 647 conjugated goat anti-human IgG Fc antibodies (Jackson ImmunoResearch Laboratories). The cells were then subjected to flow cytometric analysis. 293-spike or 293-Delta-spike cells were incubated with the indicated S-BiTE- containing samples at 4 °C for 30 min, washed three times with 2% FBS in PBS, and incubated with 1:200 diluted Alexa Fluor 647 conjugated goat anti-mouse Fab antibodies (Jackson ImmunoResearch Laboratories). The cells were then subjected to flow cytometric analysis. Lentivirus production. Lentivirus was produced by transient transfection of Lenti- X 293 T cells with a four-plasmid system. Supernatants containing the lentiviral particles were collected at 48 and 72 h post-transfection and used to establish stable cell lines. In vitro T cell activation and killing assay. Human T cells were stimulated with anti-CD3 (0.25 μg/mL; Bio X Cell, Lebanon, NH, USA) and anti-CD28 (1 μg/mL; Bio X Cell) for 2 days, rested for 3 days, and used for the in vitro activation and killing assay in complete RPMI 1640 supplemented with IL-2 (50 IU/mL) (Beijing Four Rings Bio-Pharmaceutical Co., Beijing, China) and IL-21 (4 ng/mL; Biole- gend). Approximately 1 × 105 T cells were co-cultured with 2.5 × 104 Raji or Raji- spike cells; 3.75 × 104 293, 293-spike, or 293-Delta spike cells; or 1.25 × 104 A549, A549-spike, or A549-Delta spike cells in the presence of various concentrations of S-BiTE. After 1 day, levels of TNF and IFN-γ in the supernatant were analyzed by a cytometric bead array (CBA) assay according to the manufacturer’s protocol (BD Biosciences, San Jose, CA). After 2 days, the killing assay was conducted using flow cytometry. Anti-CD45 antibody was used to distinguish T cells from 293 or A549 cells. Anti-CD3 and anti-CD19 antibodies were used to distinguish T cells from Raji-derived cells. A549 and A549-spike cells were labeled with CFSE (MedChemExpress, Shanghai, China) and CellTraceTM violet (Life Technologies Corporation, Eugene, OR, USA), respectively, according to the manufacturer’s protocol. Then, 1.5 × 104 fluorescent dye-labeled A549 and A549-spike cells were plated on a 96-well plate. After 6 h, 1 × 105 T cells and various concentrations of S-BiTE were added to the culture. Killing efficiency was determined using the Operetta CLS (PerkinElmer, Waltham, MA, USA). Pseudotyped virus release assay. Lenti-X 293 T cells were transfected with Gag/ Pol (#12251, Addgene), RSV-Rev (#12253, Addgene), pcDNA3.1(+)-2019-nCoV- spike-P2A-eGFP (#MC_0101087, MolecularCloud), and pCDH-EF-GFP-luc in 24- well plates. After 24 h, 1 × 105 T cells were added to the culture with or without S-BiTE stimulation. Supernatants were collected at 48 h post-transfection. The viral titer was measured by infecting 293-ACE2 cells. Levels of TNF and IFN-γ in the supernatant were analyzed by the CBA assay. The remaining virus-releasing cells were imaged by REVOLVE fluorescent microscopy (ECHO, San Diego, CA, USA) and analyzed by flow cytometry. After 2 h, the free viruses were removed by washing with PBS. Human primary T cells were added to the culture in the presence of 1 or10 μg/mL of S-BiTE. At 24 h and 48 h post infection, cells were collected and mRNA was isolated. Reverse- transcription quantitative polymerase chain reaction (RT-qPCR) was used to test the SARS-CoV-2 mRNA viral titer using the One-Step PrimeScript RT-PCR Kit (Takara, Shiga, Japan) with the following primers: SARS-CoV-2-N-F: GGGGAACTTCTCCTGCTAGAAT; SARS-CoV-2-N-R: CAGACATTTTGCTCTCAAGCTG; SARS-CoV-2-N-probe: 5’-FAM- TTGCTGCTGCTTGACAGATT-TAMRA-3’. hACE2- hCD3ε transgenic mice were intranasally administrated with twenty- five μg of S-BiTE, equal mole of ACE2-His, or equal volume of PBS. One hour later, mice were intranasally infected with 10,000 PFU of SARS-CoV2 Delta-variant. Twenty-five μg of S-BiTE, equal mole of ACE2-His, or equal volume of PBS were administrated 24, and 48 h post-infection. Three days later, lungs and intestines of three groups collected for RNA isolation and RT-PCR. Mice. Six-eight weeks old female C57BL/6 J mice were purchased from Beijing Vital River Laboratory Animal Technology Co., Ltd. (Beijing, China). Eight-ten weeks old hACE2-hCD3ε transgenic mice, Six-eight weeks old female NOD- PrkdcscidIL2rγtm1 (NSG) mice were purchased from the Shanghai Model Organ- isms Center, Inc. (Shanghai, China). All mice were maintained under specific pathogen-free conditions. Animal care and use were in accordance with institu- tional and NIH protocols and guidelines, and all studies were approved by the Animal Care and Use Committee of Shanghai Jiao Tong University and the Institutional Laboratory Animal Care of Fudan University (20220609-001). In vivo T cell killing assay. Raji and Raji-spike cells were labeled with 5 or 50 μM of CFSE, respectively, according to the manufacturer’s protocol (MedChemEx- press). Approximately 2 × 106 CFSE-labeled Raji and Raji-spike cells were mixed with 5 × 106 T cells and injected intraperitoneally into NSG mice. The mice were treated with PBS or S-BiTE, and cells in the peritoneal cavity were collected and analyzed by flow cytometry 6 h after treatment. Statistics and reproducibility. The number of independent biological replicates (n) of each experiment was noted in the figure legends. All attempts at replication were successful. All statistical analyses were performed using GraphPad Prism 8. Error bars represent standard deviation (SD) or standard error of the mean (SEM). Statistical analyses were performed using the Student’s t-test and one-way or two- way analysis of variance (ANOVA) with Dunnett multiple comparisons correction. Reporting summary. Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Data availability The main data supporting the results in this study are available within the paper and its Supplementary Information. Source data for all figures can be found in Supplementary Data 1. The raw and analysed datasets generated during the study are too large to be publicly shared, yet they are available for research purposes from the corresponding authors on reasonable request. Received: 21 February 2023; Accepted: 18 May 2023; References 1. Zhou, P. et al. A pneumonia outbreak associated with a new coronavirus of probable bat origin. Nature 579, 270–273 (2020). 2. Guan, W. J. et al. Clinical characteristics of coronavirus disease 2019 in China. N. Engl. J. Med. 382, 1708–1720 (2020). 4. 3. Coronaviridae Study Group of the International Committee on Taxonomy of, V. The species Severe acute respiratory syndrome-related coronavirus: classifying 2019-nCoV and naming it SARS-CoV-2. Nat. Microbiol. 5, 536–544 (2020). Jara, A. et al. Effectiveness of an inactivated SARS-CoV-2 vaccine in Chile. N. Engl. J. Med. https://doi.org/10.1056/NEJMoa2107715 (2021). Polack, F. P. et al. Safety and efficacy of the BNT162b2 mRNA Covid-19 vaccine. N. Engl. J. Med. 383, 2603–2615 (2020). Sadoff, J. et al. Safety and efficacy of single-dose Ad26.COV2.S vaccine against Covid-19. N. Engl. J. Med. 384, 2187–2201 (2021). 5. 6. Live SARS-CoV-2 infection assay in vitro and in vivo. The inhibition assay for live SARS-CoV-2 was performed in a biosafety level 3 (BSL3) facility at Fudan University. A549-ACE2 cells were seeded in 24-well plates and infected with live SARS-CoV-2 (GenBank: MT121215.1) at a multiplicity of infection (MOI) of 0.05. 7. Davies, N. G. et al. Estimated transmissibility and impact of SARS-CoV-2 lineage B.1.1.7 in England. Science 372, eabg3055 (2021). 8. Wang, Z. et al. mRNA vaccine-elicited antibodies to SARS-CoV-2 and circulating variants. Nature 592, 616–622 (2021). COMMUNICATIONS BIOLOGY | (2023) 6:592 | https://doi.org/10.1038/s42003-023-04955-3 | www.nature.com/commsbio 9 ARTICLE COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-023-04955-3 9. Xiang, R. et al. Recent advances in developing small-molecule inhibitors against SARS-CoV-2. Acta Pharm Sin B https://doi.org/10.1016/j.apsb.2021. 06.016 (2021). 10. Nepali, K., Sharma, R., Sharma, S., Thakur, A. & Liou, J. P. Beyond the vaccines: a glance at the small molecule and peptide-based anti-COVID19 arsenal. J. Biomed. Sci. 29, 65 (2022). 11. Dryden-Peterson, S. et al. Nirmatrelvir plus ritonavir for early COVID-19 in a large U.S. health system: a population-based cohort study. Ann. Intern Med. 176, 77–84 (2023). 12. Hammond, J. et al. Oral nirmatrelvir for high-risk, nonhospitalized adults with Covid-19. N. Engl. J. Med. 386, 1397–1408 (2022). 43. Yan, R. et al. Structural basis for the recognition of SARS-CoV-2 by full-length human ACE2. Science 367, 1444–1448 (2020). 44. Staflin, K. et al. Target arm affinities determine preclinical efficacy and safety of anti-HER2/CD3 bispecific antibody. JCI Insight 5 https://doi.org/10.1172/ jci.insight.133757 (2020). 45. Bortoletto, N., Scotet, E., Myamoto, Y., D’Oro, U. & Lanzavecchia, A. Optimizing anti-CD3 affinity for effective T cell targeting against tumor cells. Eur. J. Immunol. 32, 3102–3107 (2002). 46. Wu, Y. et al. A noncompeting pair of human neutralizing antibodies block COVID-19 virus binding to its receptor ACE2. Science https://doi.org/10. 1126/science.abc2241 (2020). 13. Ganatra, S. et al. Oral nirmatrelvir and ritonavir in non-hospitalized vaccinated 47. Pinto, D. et al. Cross-neutralization of SARS-CoV-2 by a human monoclonal patients with Covid-19. Clin. Infect. Dis. https://doi.org/10.1093/cid/ciac673 (2022). SARS-CoV antibody. Nature https://doi.org/10.1038/s41586-020-2349-y (2020). 14. Chen, X. et al. Human monoclonal antibodies block the binding of SARS- 48. Salazar, G., Zhang, N., Fu, T. M. & An, Z. Antibody therapies for the CoV-2 spike protein to angiotensin converting enzyme 2 receptor. Cell. Mol. Immunol. https://doi.org/10.1038/s41423-020-0426-7 (2020). 15. Wang, C. et al. A human monoclonal antibody blocking SARS-CoV-2 infection. Nat. Commun. 11, 2251 (2020). 16. Shi, R. et al. A human neutralizing antibody targets the receptor-binding site of SARS-CoV-2. Nature 584, 120–124 (2020). 17. Cao, Y. et al. Potent neutralizing antibodies against SARS-CoV-2 identified by high-throughput single-cell sequencing of convalescent patients’ B cells. Cell 182, 73–84 e16 (2020). 18. Chi, X. et al. A neutralizing human antibody binds to the N-terminal domain of the Spike protein of SARS-CoV-2. Science 369, 650–655 (2020). 19. Weinreich, D. M. et al. REGN-COV2, a neutralizing antibody cocktail, in outpatients with Covid-19. N. Engl. J. Med. 384, 238–251 (2021). 20. Lei, C. et al. Neutralization of SARS-CoV-2 spike pseudotyped virus by recombinant ACE2-Ig. Nat. Commun. 11, 2070 (2020). 21. Glasgow, A. et al. Engineered ACE2 receptor traps potently neutralize SARS- CoV-2. Proc. Natl Acad. Sci. USA 117, 28046–28055 (2020). 22. Chan, K. K. et al. Engineering human ACE2 to optimize binding to the spike protein of SARS coronavirus 2. Science 369, 1261–1265 (2020). 23. Monteil, V. et al. Inhibition of SARS-CoV-2 infections in engineered human tissues using clinical-grade soluble human ACE2. Cell https://doi.org/10.1016/ j.cell.2020.04.004 (2020). 24. Cho, H. et al. Bispecific antibodies targeting distinct regions of the spike protein potently neutralize SARS-CoV-2 variants of concern. Sci. Transl. Med. 13, eabj5413 (2021). 25. De Gasparo, R. et al. Bispecific IgG neutralizes SARS-CoV-2 variants and prevents escape in mice. Nature 593, 424–428 (2021). 26. Li, Z. et al. An engineered bispecific human monoclonal antibody against SARS-CoV-2. Nat. Immunol. 23, 423–430 (2022). 27. Hanke, L. et al. A bispecific monomeric nanobody induces spike trimer dimers and neutralizes SARS-CoV-2 in vivo. Nat. Commun. 13, 155 (2022). 28. Qian, K. & Hu, S. Ig-like ACE2 protein therapeutics: A revival in development during the COVID-19 pandemic. MAbs 12, e1782600 (2020). 29. Miao, X. et al. A novel biparatopic hybrid antibody-ACE2 fusion that blocks SARS-CoV-2 infection: implications for therapy. MAbs 12, 1804241 (2020). 30. Edara, V. V. et al. Infection and Vaccine-Induced Neutralizing-Antibody Responses to the SARS-CoV-2 B.1.617 Variants. N. Engl. J. Med. 385, 664–666 (2021). 31. Lopez Bernal, J. et al. Effectiveness of Covid-19 Vaccines against the B.1.617.2 (Delta) Variant. N. Engl. J. Med. 385, 585–594 (2021). 32. Garcia-Beltran, W. F. et al. Multiple SARS-CoV-2 variants escape neutralization by vaccine-induced humoral immunity. Cell 184, 2372–2383 e2379 (2021). 33. Tada, T. et al. Convalescent-phase sera and vaccine-elicited antibodies largely maintain neutralizing titer against global SARS-CoV-2 variant spikes. mBio 12, e0069621 (2021). 34. Wang, P. et al. Antibody resistance of SARS-CoV-2 variants B.1.351 and B.1.1.7. Nature 593, 130–135 (2021). 35. Wu, K. et al. Serum neutralizing activity elicited by mRNA-1273 vaccine. N. Engl. J. Med. 384, 1468–1470 (2021). 36. Xie, X. et al. Neutralization of SARS-CoV-2 spike 69/70 deletion, E484K and N501Y variants by BNT162b2 vaccine-elicited sera. Nat. Med. 27, 620–621 (2021). 37. Liu, J. et al. BNT162b2-elicited neutralization of B.1.617 and other SARS- CoV-2 variants. Nature 596, 273–275 (2021). 38. Planas, D. et al. Reduced sensitivity of SARS-CoV-2 variant Delta to antibody neutralization. Nature 596, 276–280 (2021). 39. Leen, A. M., Rooney, C. M. & Foster, A. E. Improving T cell therapy for cancer. Annu Rev. Immunol. 25, 243–265 (2007). 40. Barry, M. & Bleackley, R. C. Cytotoxic T lymphocytes: all roads lead to death. Nat. Rev. Immunol. 2, 401–409 (2002). 41. Baeuerle, P. A. & Reinhardt, C. Bispecific T-cell engaging antibodies for cancer therapy. Cancer Res. 69, 4941–4944 (2009). 42. Hoffmann, M. et al. SARS-CoV-2 cell entry depends on ACE2 and TMPRSS2 and is blocked by a clinically proven protease inhibitor. Cell 181, 271–280. e278 (2020). prevention and treatment of viral infections. NPJ Vaccines 2, 19 (2017). 49. Cao, Y. et al. Potent neutralizing antibodies against SARS-CoV-2 identified by high-throughput single-cell sequencing of convalescent patients’ B cells. Cell https://doi.org/10.1016/j.cell.2020.05.025 (2020). 50. Nimmerjahn, F. & Ravetch, J. V. Fcgamma receptors: old friends and new family members. Immunity 24, 19–28 (2006). 51. Frenette, P. S., Pinho, S., Lucas, D. & Scheiermann, C. Mesenchymal stem cell: keystone of the hematopoietic stem cell niche and a stepping-stone for regenerative medicine. Annu Rev. Immunol. 31, 285–316 (2013). 52. Public Health England. Variants distribution of cases. https://www.gov.uk/ government/publications/covid-19-variants-genomically-confirmed-case- numbers/variants-distribution-of-case-data-11-june-2021 (2021). 53. Baum, A. et al. Antibody cocktail to SARS-CoV-2 spike protein prevents rapid mutational escape seen with individual antibodies. Science 369, 1014–1018 (2020). 54. Sun, Y. et al. Structure-based development of three- and four-antibody cocktails against SARS-CoV-2 via multiple mechanisms. Cell Res. 31, 597–600 (2021). 55. Cho, H. et al. Ultrapotent bispecific antibodies neutralize emerging SARS- CoV-2 variants. bioRxiv https://doi.org/10.1101/2021.04.01.437942 (2021). 56. Dogan, M. et al. Targeting SARS-CoV-2 infection through CAR-T-like bispecific T cell engagers incorporating ACE2. Clin. Transl. Immunol. 11, e1421 (2022). 57. Zou, X. et al. Single-cell RNA-seq data analysis on the receptor ACE2 expression reveals the potential risk of different human organs vulnerable to 2019-nCoV infection. Front. Med. https://doi.org/10.1007/s11684-020-0754-0 (2020). Acknowledgements We thank members of the Core Facility of Microbiology and Parasitology of SHMC and the Biosafety Level 3 Laboratory at Shanghai Medical College of Fudan University, especially Qian Wang, Di Qu and Gaowei Hu. X.Y. was supported by the National Natural Science Foundation of China (grant no. 32261143731 and 81971467), and Sheng Yushou Foundation. L.L. was supported by the National Key R&D Program of China under Grant number (2021YFC2300703 and 2022YFC2604102), Shanghai Municipal Science and Technology Major Project (ZD2021CY001 to L.L. and W.X.) and the Pro- gram of Shanghai Academic/Technology Research Leader (grant no. 20XD1420300). Author contributions X.Y., L.L, and YH.X. designed the project. F.L., W.X., X.Z., W.W., S.S., P.H., H.W., YQ.X., M.L., L.F., H.Z., Q.D., H.L., X.Q., J.L, X.W., S.J., and X.Y. performed the experiments. F.L., X.Z., W.X., W.W., and X.Y. analyzed the results and wrote the manuscript with input from all co-authors. Competing interests The authors declare no competing interests. Additional information Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s42003-023-04955-3. Correspondence and requests for materials should be addressed to Youhua Xie, Lu Lu or Xuanming Yang. Peer review information This manuscript has been previously reviewed at another Nature Portfolio journal. Communications Biology thanks luca Varani, Roberto Speck and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editors: Zhijuan Qiu and David Favero. A peer review file is available. This document only contains reviewer comments and rebuttal letters for versions considered at Communications Biology. Reprints and permission information is available at http://www.nature.com/reprints Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. 10 COMMUNICATIONS BIOLOGY | (2023) 6:592 | https://doi.org/10.1038/s42003-023-04955-3 | www.nature.com/commsbio COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-023-04955-3 ARTICLE Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/ licenses/by/4.0/. © The Author(s) 2023 COMMUNICATIONS BIOLOGY | (2023) 6:592 | https://doi.org/10.1038/s42003-023-04955-3 | www.nature.com/commsbio 11
null
10.1093_jncics_pkab068.pdf
Data Availability The data underlying this article cannot be shared publicly due to privacy restrictions of individuals that participated in the study. Aggregated, deidentified data may be shared on reason- able request to the corresponding author.
Data Availability The data underlying this article cannot be shared publicly due to privacy restrictions of individuals that participated in the study. Aggregated, deidentified data may be shared on reasonable request to the corresponding author.
JNCI Cancer Spectrum (2021) 5(5): pkab068 doi: 10.1093/jncics/pkab068 First published online 17 July 2021 Article Project Forward: A Population-Based Cohort Among Young Adult Survivors of Childhood Cancers Joel Milam , PhD,1,2 * David R. Freyer Katherine Y. Wojcik , DO,1,3,4 Kimberly A. Miller , PhD,7,8 Cynthia N. Ramirez, MPH,1 Anamara Ritt-Olson, PhD,1 , MD,9 Lourdes Baezconde-Garbanati, PhD,1 Michael Cousineau, PhD,1 , PhD,1,5 Jessica Tobin Stefanie M. Thomas , PhD,1,6 Denise Modjeski , MS,1 Sapna Gupta , MS,4 Ann S. Hamilton , PhD1 1Department of Preventive Medicine, Keck School of Medicine, University of Southern California, Los Angeles, CA, USA; 2Departments of Medicine and Epidemiology and Biostatistics, Chao Family Comprehensive Cancer Center,; University of California, Irvine, CA, USA; 3Children’s Hospital Los Angeles, Los Angeles, CA, USA; 4USC Norris Comprehensive Cancer Center, Los Angeles, CA, USA; 5Department of Dermatology, Keck School of Medicine, University of Southern California, Los Angeles, CA, USA; 6VA Greater Los Angeles Health Care System, Los Angeles, CA, USA; 7Public Health Sciences, Fred Hutchinson Cancer Center, Seattle, WA, USA; 8Department of Epidemiology, University of Washington, WA, USA; and 9Department of Pediatric Hematology Oncology and Bone Marrow Transplantation, Cleveland Clinic Children’s Hospital, Cleveland, OH, USA *Correspondence to: Joel Milam, PhD, Department of Medicine, Department of Epidemiology and Biostatistics, Chao Family Comprehensive Cancer Center, University of California, Irvine, 653 E. Peltason Dr, Irvine, CA 92697-7550, USA (e-mail: milamj@hs.uci.edu). Abstract Background: Childhood cancer survivors (CCS) face increased risk of morbidity and are recommended to receive lifelong cancer-related follow-up care. Identifying factors associated with follow-up care can inform efforts to support the long-term health of CCS. Methods: Eligible CCS (diagnosed between 1996 and 2010) identified through the Los Angeles County Cancer Surveillance Program responded to a self-report survey that assessed demographic, clinical, health-care engagement, and psychosocial risk and protective factors of recent (prior 2 years) cancer-related follow-up care. Weighted multivariable logistic regression was conducted to identify correlates of care. All statistical tests were 2-sided. Results: The overall response rate was 44.9%, with an analytical sample of n ¼ 1106 (54.2% Hispanic; mean [SD] ages at survey, diagnosis, and years since diagnosis were 26.2 [4.9], 11.6 [5.4], and 14.5 [4.4] years, respectively). Fifty-seven percent reported a recent cancer-related visit, with lower rates reported among older survivors. Having insurance, more late effects, receipt of a written treatment summary, discussing long-term care needs with treating physician, knowledge of the need for long-term care, having a regu- lar source of care, and higher health-care self-efficacy were statistically significantly associated with greater odds of recent follow-up care, whereas older age, Hispanic or Other ethnicity (vs non-Hispanic White), and years since diagnosis were asso- ciated with lower odds of recent care (all Ps < .05). Conclusions: Age and ethnic disparities are observed in receipt of follow- up care among young adult CCS. Potential intervention targets include comprehensive, ongoing patient education; provision of written treatment summaries; and culturally tailored support to ensure equitable access to and the utilization of care. Improvements in childhood cancer treatment regimens have resulted in 5-year survival rates of more than 80% (1,2). Unfortunately, the majority of childhood cancer survivors (CCS) experience late adverse effects of cancer treatment, which often become clinically apparent years after treatment ends (3). Many of these late effects are severe or life-threatening and cause a range of symptomatic health problems, impaired function, and reduced quality of life (3-6). To facilitate prevention, detection, and management of late effects, the Children’s Oncology Group developed the Long-Term Follow-Up Guidelines for Survivors of Childhood, Adolescent and Young Adult Cancer, recommending that all CCS receive lifelong, risk-adapted surveillance and sur- vivorship care (7). Despite these recommendations, rates of health-care en- gagement among CCS decline with age and time since treat- ment, especially as CCS enter their 20s (5,8-10). Because this attrition coincides with the rising cumulative incidence of late effects, it results in multiple missed opportunities for primary and secondary prevention (8). Studies among racial and ethnic minority CCS are also needed (11-14). Disparities in health-care Received: 29 January 2021; Revised: 18 May 2021; Accepted: 14 July 2021 © The Author(s) 2021. Published by Oxford University Press. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. 1 of 10 2 of 10 | JNCI Cancer Spectrum, 2021, Vol. 5, No. 5 utilization among CCS have been observed by ethnicity, with higher proportions of non-Hispanic Whites (vs Hispanic) report- ing receipt of cancer-related follow-up care, an association not explained by health insurance coverage (8). However, such dis- parities are not observed uniformly, suggesting variation in study samples, or individual- and/or system-level factors asso- ciated with health-care access (15-17). Underlying drivers of age- and race- and ethnicity-related disparities in CCS follow- up care need continued investigation, particularly among ethni- cally diverse and more recently treated cohorts, as prior studies of CCS have primarily included non-Hispanic Whites and CCS diagnosed before 1999 and thus have been treated before nu- merous advances in treatment and survivorship care practices (eg, the broader use of survivorship care plans and survivorship clinics) and the impact of the Afforda ble Care Act (ACA) (18,19). Prior research among CCS on access to and utilization of cancer-related follow-up care has focused predominantly on sociodemographic and clinical factors and less on organiza- tional and psychosocial factors. For example, little is known about how many CCS have a regular source of care, the types of providers seen for cancer-related follow-up care, and patient knowledge of their health-care needs, as well as their confi- dence in navigating the health-care system (ie, health-care self- efficacy). Understanding these factors and their association with health-care utilization in early adulthood will clarify op- portunities for intervention to prepare and support young adult CCS for managing their health care independently. To address these gaps, we assessed the prevalence of clinical, demographic, psychosocial, and care-related factors, as well as their associations with receipt of cancer-related follow-up care in a diverse, population-based cohort of young adult CCS. We hy- pothesized that health insurance, greater knowledge of follow- up recommendations, younger age, non-Hispanic White (vs Hispanic) ethnicity, and higher health-care self-efficacy would be associated with greater use of cancer-related follow-up care. Methods Study Population is a cancer The Project Forward Cohort registry–derived, population-based study of risk and protective factors of cancer- related follow-up care among young adult CCS. Data on all cases were obtained from the Los Angeles Cancer Surveillance Program, which is the cancer registry for Los Angeles County (part of the Surveillance, Epidemiology, and End Results pro- gram). Eligible participants included CCS who were diagnosed up to 19 years of age between 1996 and 2010 in Los Angeles County, California, with any cancer diagnosis (stage 2 or greater, except for brain and melanoma, which included stage 1 or greater) and who were at least 5 years postdiagnosis and aged 18-39 years when the study was launched in 2015. Procedure Recruitment methods were based on our pilot work (20) and in- cluded introductory postcards and self-report survey mailings in English and Spanish with the option to complete the survey on- line, over the phone, or in person in either language. Mailings also included a brochure describing the study and an informa- tional brochure concerning the California Cancer Registry. Reminder mailings and calls occurred for those who did not re- spond. Although initial contact information (both recent address and address at diagnosis) is provided by the registry, we per- formed address tracing to improve accuracy of addresses and retraced potential participants who were difficult to contact (eg, in cases of post office returns) before being classified as lost after all efforts (see Figure 1). Participants received $20 cash and a lot- tery entry ($300). Participants who self-reported receiving cancer treatment less than 2 years prior to the study (n ¼ 60) were ex- cluded from analyses, with the exception of those with chronic myeloid leukemia due to the routine use of protracted mainte- nance therapy with tyrosine kinase inhibitors. Procedures were approved by the California State Committee for the Protection of Human Subjects, the institutional review board at the University of Southern California, and the California Cancer Registry. Measures Primary outcome. The primary outcome was receipt of cancer- related follow-up care in the prior 2 years (1 ¼ yes, 0 ¼ no). This was obtained via self-report and defined as any health-care visit where a provider completed an examination or tests to assess health problems from prior cancer or the cancer treatment they received, similar to an item used in the Childhood Cancer Survivor Study (5). Participants also indicated the type of health-care provider seen for this care (9). Demographic and clinical factors. Age at diagnosis, age at survey, cancer diagnosis (site and histology), diagnosing hospital, sex, race and ethnicity (non-Hispanic White, Hispanic, Asian, and other), and quintiles of neighborhood socioeconomic status (nSES) at diagnosis were obtained from the cancer registry. nSES is a census-based composite score (relative to California’s state- wide distribution; 1 ¼ lowest quintile nSES, 5 ¼ highest quintile nSES), reflecting 7 indicators, including education index, per- cent persons above 200% poverty line, percent persons with a blue-collar job, percent persons employed, median rental, me- dian value of owner-occupied housing unit, and median house- hold income (21,22). Current health insurance (private, public, other, or uninsured) was self-reported. As described in Intensity of Treatment Rating Scale 3.0, clini- cal and treatment information collected from medical charts is used to categorize cancer cases into 4 levels of treatment inten- sity (1 ¼ least intensive [eg, surgery only]; 2 ¼ moderately inten- sive [eg, chemotherapy or radiation]; 3 ¼ very intensive [eg, 2 or more treatment modalities]; and 4 ¼ most intensive [eg, relapse regimens]) (23). Because of the prohibitive cost of accessing medical charts for our large sample, we developed a novel method of calculating treatment intensity using cancer registry data as a proxy for chart data. Using our pilot study sample, for which treatment intensity had been previously determined us- ing medical chart data, concordance between treatment inten- sity estimated by our method and treatment intensity estimated by the original chart-based method was assessed with Cohen Kappa statistic to validate this approach showing reasonable agreement between methods. A full description of the validation of this method of estimation of treatment inten- sity using cancer registry and self-reported data is available (24). Self-reported late effects of cancer treatment included 11 items (eg, inability to have children, heart problems, difficulties with learning and memory, eyesight). Items were selected based on the most prevalent chronic conditions previously docu- mented among CCS (3,25-27). Summary scores were categorized as none, 1, or 2 or more late effects. J. Milam et al. | 3 of 10 Registry cases iden(cid:2)fied/screened (n = 2788) Ineligible (n = 196) Incompetent/too ill (54) (cid:2) Deceased (61) (cid:2) (cid:2) Cancer not confirmed at screening (41) (cid:2) MD refused (20) (cid:2) In prison (11) (cid:2) Ineligible a(cid:3)er screening/linkage to registry (9) (cid:2) Denied cancer (n = 47) (cid:2) Did not meet inclusion criteria (n = 21) Eligible & approached (n = 2592) Exclusions (n = 1426) (cid:2) Declined to par(cid:2)cipate (207) (cid:2) Requested no further contact (13) (cid:2) Gatekeeper refusal (64) (cid:2) Passive pa(cid:2)ent refusal (359) (cid:2) Out of the country (9) (cid:2) Lost a(cid:3)er all other efforts (774) Enrolled (n = 1166) Addi(cid:2)onal exclusions (n = 60) (cid:2) Reported on treatment in prior two years (n = 60) Analy(cid:2)c sample (n = 1106) Figure 1. Project forward CONSORT diagram. MD ¼ physician in registry record. Indicators of health-care engagement. These included having dis- cussed future cancer-related health-care needs with any doctor, ever receiving a written cancer treatment summary, having a regular doctor for noncancer care, and ever sharing this written treatment summary with current doctors, which were separate, self-reported variables (1 ¼ yes, 0 ¼ no or not sure) (28). Participants reported whether they believed they needed lifelong follow-up care (1 ¼ yes, 0 ¼ no or not sure). Psychosocial factors. Health-care self-efficacy (HCSE) was mea- sured by 3 items related to perceived confidence in making appointments with physicians: obtaining cancer-related follow- up care; and discussing concerns with physicians, adapted from the Chronic Disease Self-Efficacy Scales from the Stanford Patient Education Research Center (29). Responses included “not at all confident” [0], “somewhat confident” [1], and “totally confident” [2] and were summed to create a total score that could range from 3 to 9 with higher scores representing greater HCSE (Cronbach alpha ¼ 0.72). Family influence was measured using a single item asking how often family has influenced the health-care decisions of the participants (1 ¼ often or occasionally or 0 ¼ never). Depressive symptoms were assessed using the Center for Epidemiologic Studies Depression Scale (30). This scale includes 20 items about how often participants experienced symptoms in the past week, such as negative affect, sleep disruption, and feelings of hopelessness. Response options range from “rarely or none of the time” [0] to “most or all of the time” [3]. Scores were summed with a possible range of 0-60 (Cronbach alpha ¼ 0.80) and dichotomized (1/0) at a score of 16 or greater to indi- cate likely depression. Statistical Analysis Prevalence rates of the different components of survivorship care were examined both individually and cumulatively (to re- flect receipt of multiple follow-up care recommendations) (31). 4 of 10 | JNCI Cancer Spectrum, 2021, Vol. 5, No. 5 This approach is similar to that used previously for identifying gaps in the implementation of recommended care for chronic diseases (eg, HIV, diabetes), using a prevalence-based “cascade of care” to highlight proportions receiving multiple dimensions of care (32,33). Bivariate and multivariable logistic regression analyses were conducted to identify factors associated with receipt of cancer- related follow-up care. The multivariable model was weighted to account for survey response bias (correcting for differences in the distribution of sex, race and ethnicity, and nSES between survey responders and nonresponders) (34). Diagnosing hospital data were obtained from the cancer registry, and we incorpo- rated clustered standard errors in all models to control for within-hospital correlations related to follow-up care. Age at survey, sex, and race and ethnicity (as a proxy for unmeasured cultural and societal factors known to impact health-care access) were adjusted for in the multivariable model (35). The entry criteria for other variables to be retained in the multivariable model were a relationship with follow-up care in bivariate analyses (P (cid:2) .10) (36), which included years since diagnosis, nSES, health insurance, number of late effects, treatment intensity, receipt of a written cancer treatment sum- mary, having a regular doctor for noncancer care, discussion of needed follow-up care with doctor, knowledge of the need for long-term follow-up care, HCSE, and family influence over health-care decisions. Depressive symptoms scores were di- chotomized (0/1) at the clinical cut-point of 16, suggestive of de- pression. Health insurance was dichotomized to insured or uninsured because there was no statistically significant differ- ence in follow-up care between public and private insurance statuses, and less than 2% of the sample reported “other” insur- ance. The variable, “shared a written treatment summary with current doctor,” was included in the cascade of care for descrip- tive purposes but was excluded from the final model because of its linear dependence on receipt of a written treatment sum- mary. Listwise deletion was used to handle missing data (as in- dicated). Statistical significance was determined as a P-value less than .05 for 2-sided hypothesis tests. Data analyses were conducted in SAS statistical software (SAS Institute Inc, version 9.4, Cary, NC). Results Of 2788 eligible cases, 196 were subsequently deemed ineligible (eg, too ill or incompetent, deceased) and 774 were lost (ie, no valid contact information; Figure 1). We recruited 1166 respond- ents. The response rate (denominator excludes confirmed ineli- gible) was 44.9%. Among those successfully contacted (eg, verified address and phone; n ¼ 1764), the participation rate (de- nominator excludes confirmed ineligible and lost) was 64%: 39.3% (n ¼ 434) completed the survey online, 1.2% (n ¼ 13) over the phone, and the rest on paper (n ¼ 659); 1.2% (n ¼ 13) responded in Spanish. Responder analyses were performed us- ing available demographic (at time of sample selection) and clinical variables from the registry data (Table 1). There were no differences between nonresponders (n ¼ 1426) and responders (n ¼ 1166) in age at diagnosis, years since diagnosis, age, cancer diagnosis, or stage of disease. However, those who responded were more likely to be female (vs male) and non-Hispanic White and have higher (vs lower) nSES. Our analyses excluded those who self-reported as on treatment within the prior 2 years (non–chronic myeloid leukemia, n ¼ 60), and the final analytic sample size was 1106 (diagnosed across 68 sites). Participants (54.2% Hispanic; 46.0% female) had a mean age of 11.6 (SD ¼ 5.4) years at diagnosis and a mean age at survey completion of 26.2 (SD ¼ 4.9) years (Table 2). At the time of sur- vey, participants were an average of 14.5 (SD ¼4.4; range ¼ 5-22) years from diagnosis. The most common cancer diagnoses in- cluded leukemia (36.1%), lymphoma (21.7%), and brain (15.2%). Of the participants, 57% reported a cancer-related follow-up care visit in the prior 2 years. The most common health-care providers for cancer-related follow-up care included adult oncologists (41.8%), pediatric oncologists (29.9%), and primary care physicians (15.5%) (not mutually exclusive). Rates of en- dorsement for key components of survivorship care, including discussing follow-up care, knowledge of the need for follow-up care, receiving a written treatment summary, and sharing that summary with doctors, individually ranged from 28.3% to 63.1% (Figure 2). Examining these indicators cumulatively, the survi- vorship cascade decreased at each step of care, resulting in 11.9% reporting yes to all measured follow-up care components (cumulative bars in Figure 2). In the adjusted multivariable model (Table 3), years since di- agnosis, current age, Hispanic and Other ethnicity (vs non- Hispanic White), and age at survey were statistically signifi- cantly negatively associated with follow-up care (all Ps < .05). Health insurance, number of late effects, receipt of a written treatment summary, having a regular doctor for noncancer care, discussion of needed follow-up care with physician, knowledge of the need for long-term follow-up care, and HCSE were all statistically significantly positively associated with re- ceipt of recent care (all Ps< .05). In exploratory models, we examined multivariable models stratified by Hispanic ethnicity (Table 4). These results were largely consistent between Hispanics and non-Hispanics, with the exception of nSES, which showed a stronger positive associ- ation with receipt of recent care among non-Hispanics. Discussion Long-term survivorship care is critical for health maintenance among CCS, but determinants of engagement in care are complex and vary by numerous patient- and system-level fac- tors. This study leveraged a sociodemographically diverse, population-based sample to examine novel correlates of cancer- related health-care engagement, including health-care organi- zational factors. We found each care component to have a sta- tistically significant (all Ps < .01) independent association with follow-up care, suggesting that each represents a unique indica- tor of engagement. However, only 12% of the sample endorsed all components, indicating the critical need for improvement of the full spectrum of survivorship care. Because receiving (43.9%) and sharing (28.1%) a written treatment summary were the least endorsed elements, these represent components amenable to improvement through practical interventions to increase utili- zation of care (37,38). In addition to equipping CCS themselves with a thorough understanding of follow-up recommendations, their future, nononcology physicians who may not be familiar with recom- mended survivorship guidelines must also be supported. In a study of primary care physicians who cared for CCS, 48% had never or almost never received a cancer treatment summary, two-thirds were not comfortable caring for CCS, and few cor- rectly identified guideline-recommended surveillance for senti- nel late effects such as cardiac dysfunction (39). Those providers reported having access to clinical surveillance guidelines and Table 1. Differences between study responders and nonresponders on cancer registry variables (n ¼ 2592 diagnosed in 1996-2010; Los Angeles County) J. Milam et al. | 5 of 10 Characteristic Age at diagnosis, y 0-4 5-9 10-14 15-19 Years since diagnosis (2015) 5-9 10-14 15-22 Sex Male Female Age in 2015, y 18-20 21-25 26-30 31-39 Race and ethnicity Non-Hispanic White Hispanic Asian Other Cancer diagnosis Lymphoma Leukemia Brain and other nervous system Endocrine system Skin Otherd Stage of disease (missing n ¼ 2) Local Regional Distant Socioeconomic status Lowest Low Medium High Highest Nonresponder (n ¼ 1426)a Responder (n ¼ 1166)b 187 (56.8) 281 (56.4) 361 (51.4) 551 (55.5) 358 (54.7) 480 (55.1) 588 (55.1) 834 (59.4) 592 (49.9) 314 (56.7) 538 (55.8) 345 (52.0) 229 (55.9) 309 (47.6) 815 (57.3) 106 (49.5) 196 (63.8) 257 (51.3) 479 (54.1) 260 (58.7) 79 (53.74) 60 (57.14) 291 (57.1) 271 (57.7) 398 (52.4) 755 (55.5) 521 (59.1) 314 (55.1) 225 (55.7) 169 (47.5) 197 (51.8) 142 (43.2) 217 (43.6) 342 (48.7) 442 (44.5) 296 (45.3) 391 (44.9) 479 (44.9) 571 (40.6) 595 (50.1) 240 (43.3) 427 (44.3) 318 (48.0) 181 (44.2) 340 (52.4) 607 (42.7) 108 (50.5) 111 (36.7) 244 (48.7) 407 (45.9) 183 (41.3) 68 (46.26) 45 (42.86) 219 (42.9) 199 (42.3) 361 (47.6) 606 (44.5) 361 (40.9) 256 (44.9) 179 (44.3) 187 (52.5) 183 (48.2) Test statistic v2 4.64 Pc .20 .0268 .99 23.39 <.001 3.32 .34 29.68 <.001 6.69 .24 5.12 .16 15.67 .004 aAmong those eligible and approached. bAmong those initially enrolled. c Two-sided, v2 tests. dOral cavity and pharynx, digestive system, respiratory system, soft tissue including heart, urinary system, eye and orbit, and miscellaneous. receiving patient-specific information would be most likely to improve their quality care for survivors (39). Therefore, care co- ordination and information sharing between oncology and pri- mary care physicians are needed to support survivors. Specialized cancer survivor programs are unlikely to fully sup- port the growing number of CCS, and indeed, more than 15.5% of our sample reported seeing a primary care physician for their follow-up care, underscoring the importance of equipping pri- mary care providers to care for this unique population. Our findings demonstrate that although roughly half the sam- ple reported recent cancer-related follow-up care, rates differed by race and ethnicity, consistent with prior research (8,9,18,40). Among Hispanics, the odds of reporting recent follow-up care were 31% lower compared with non-Hispanic Whites. This dis- parity was not explained by nSES, health insurance, or treatment differences, so additional factors need assessment to inform efforts to improve equity in access to care. For example, failure to adequately account for cultural characteristics and beliefs around health and disease in the provision of care may partially drive ethnic disparities by posing a barrier to patients’ understanding of health-care providers’ instructions (41). Other factors that may underlie ethnic differences in access to care include conceptions about Western medicine, fatalism, or risk perception (41-44). Investigation of sociocultural factors (eg, culturally based beliefs about disease, language, understanding of insurance, neighbor- hood resources) mediating disparities in follow-up care among CCS is underway to clarify subgroups at greater risk of disengage- ment from care and potential areas to target tailored support (14). The observed decline in rates of follow-up care with age (and years since diagnosis) is consistent with prior research showing 6 of 10 | JNCI Cancer Spectrum, 2021, Vol. 5, No. 5 Table 2. Descriptive statistics of registry and self-report data from participants enrolled in the Project Forward Cohort (n ¼ 1106) Variable Cancer registry data Age at diagnosis, y Mean (SD) [range] 0-4 5-9 10-14 15-19 Years since diagnosis, y Mean (SD) [range] 5-9 10-14 15-22 Sex Male Female Age at survey completion, y Mean (SD) [Range] Age group at survey completion, y 18-20 21-25 26-30 31-41 Race and ethnicity Non-Hispanic White Hispanic Asian Other Cancer diagnosis Leukemia Lymphoma Brain and other nervous system Endocrine system Bones and joints Skin Genital system Other Treatment intensityc 1 (least intensive) 2 (moderately intensive) 3 (very intensive) 4 (most intensive) Socioeconomic status at diagnosis Lowest Low Medium High Highest Self-report datad Health insurance (missing n ¼ 35) Private Public Other/Unknown None Health-care self-efficacy (missing n ¼ 20)e Mean (SD) [range] High levels of depressive symptoms (missing n ¼ 93)f Family influence health-care decisions (yes; missing n ¼ 17) Has doctor for regular (noncancer) health checkups (missing n ¼ 19) Had any health-care visit in prior 2 years (missing n ¼ 0) Discussed cancer-related follow-up care needs with a doctor (yes, in the last 2 years; missing n ¼ 20) Knowledge of need of lifelong follow-up care (missing n ¼ 16) No. (Weighted %) 11.60 (5.37) [0-19] 155 (14.3) 214 (19.5) 329 (29.8) 408 (36.5) 14.54 (4.37) [5-22] 174 (15.7) 354 (31.7) 578 (52.6) 544 (54.0) 562 (46.0) 26.15 (4.87) [18-41] 131 (11.7) 422 (38.7) 339 (30.3) 214 (19.3) 324 (27.4) 570 (54.2) 107 (9.2) 105 (9.2)a 392 (36.1) 240 (21.7) 169 (15.2) 60 (5.1) 56 (5.0) 41 (3.5) 56 (5.0) 92 (8.2)b 69 (6.0) 344 (30.9) 544 (49.9) 149 (13.3) 344 (34.8) 238 (21.2) 167 (14.6) 180 (14.6) 177 (14.8) 631 (57.2) 321 (31.1) 17 (1.8) 102 (10.0) 4.83 (1.3) [0-6] 353 (35.0) 935 (85.7) 783 (71.4) 851 (76.4) 561 (51.1) 698 (63.7) (continued) Table 2. (continued) Variable Received cancer-related follow-up care (missing n ¼ 19) Received written cancer treatment summary (missing n ¼ 20) Shared written treatment summary with other doctors (missing n ¼ 1) J. Milam et al. | 7 of 10 No. (Weighted %) 632 (57.7) 481 (43.9) 310 (28.1) aIncluding 53 Black, 39 Middle Eastern, 1 non-Hispanic, American-Indian, and 12 other/unknown. bOral cavity and pharynx, digestive system, respiratory system, soft tissue including heart, urinary system, eye and orbit, and miscellaneous. cIntensity of Treatment Rating (based on both registry and self-report data, see Methods). dBased on self-report data (all missing less than 5%, except for depressive symptoms, which was 8% missing). eExamined as a continuous variable. fCenter for Epidemiological Studies-Depression score of 16 or greater. Figure 2. Cascade of recommended long-term follow-up care. Data reflect 1106 childhood cancer survivors in a population-based cohort of Los Angeles County (diag- nosed in 1996-2010). Raw percentages (white bars) are mutually exclusive, and cascade percentages (black bars) are cumulative, from left to right (eg, the last column indicates that 11.9% of the sample answered yes for all categories). The sequence of care elements is based on clinician feedback and does not represent a prescriptive causal pathway. a notable drop in the period of emerging adulthood (primarily occurring between ages 18 and 25 years) (5,8,18,45). In our study, the odds of recent care among those aged 31-41 years were 65% lower compared with those ages 18-20 years. CCS in their early 20s are especially vulnerable to the effects of interrupted health insurance due to the typical losses of state Children’s Health Insurance Program coverage at age 21 years and of parent- based private insurance coverage at age 26 years. Although pas- sage of the ACA in 2010 expanded health insurance access for young adults, 10% of our cohort was uninsured, and having in- surance was associated with 106% greater likelihood of report- ing recent follow-up care. Because follow-up care remains suboptimal despite the widespread implementation of the ACA, future work should examine discontinuity of coverage, high deductibles, and/or partial coverage for screening as barriers to follow-up care. Declines in health-care engagement with age are likely explained, in part, by competing developmental tasks, as young adulthood is a time marked by major transitions and acquired responsibilities (45). The transition from the pediatric oncology setting to adult-focused care should ideally include interpro- vider communication, involvement of family to discuss the transition of responsibility, and patient education to support health-care independence (eg, information regarding prior treatment exposure, health risks, health insurance, finding a new provider) (46-48). However, survivors often transition by default through simply aging out of pediatric care, which leads to severe attrition to follow-up and reactive medical care (49). Standardized transition assessments and patient navigation systems may enable more CCS to successfully transition to, and remain engaged in, adult survivor–focused care as they age with unique health needs (50). HCSE, the perceived ability to manage one’s health, was a statistically significant (P < .001) independent facilitator of follow-up care. HCSE may promote and be promoted by engage- ment in the health-care system. For example, attendance at a 8 of 10 | JNCI Cancer Spectrum, 2021, Vol. 5, No. 5 Table 3. Univariate and multivariable logistic regression models of receipt of cancer-related follow-up care (within prior 2 years) among child- hood cancer survivors (diagnosed in 1996-2010; Los Angeles County)a Bivariate analyses Multivariable model Characteristic Years since diagnosis Age at survey completion, y 18-20 21-25 26-30 31-39 Female (vs Male) Race and ethnicity (relative to non-Hispanic White) Non-Hispanic White Hispanic Asian Other Socioeconomic status (relative to lowest group) Lowest Low Medium High Highest Health insurance (any vs bone) High levels of depressive symptoms No. of late effects (relative to none) 0 1 (cid:3)2 Treatment intensity Received written cancer treatment summary Has doctor for regular (noncancer) health checkups Discussed cancer-related follow-up care needs with a doctor (in the last 2 years) Knowledge of need of lifelong follow-up care Health-care self-efficacy Family influence health-care decisions OR (95% CI) 0.88 (0.85 to 0.90) 1.00 (referent) 1.69 (1.31 to 2.19) 0.70 (0.54 to 0.92 0.45 (0.33 to 0.61) 1.39 (1.09 to 1.77) 1.00 (referent) 0.81 (0.59 to 1.11) 0.75 (0.43 to 1.29) 0.89 (0.61 to 1.31) 1.00 (referent) 0.89 (0.66 to 1.19) 1.06 (0.75 to 1.49) 1.59 (1.13 to 2.24) 1.09 (0.78 to 1.52) 3.05 (1.96 to 4.75) 0.93 (0.71 to 1.21) 1.00 (referent) 1.23 (0.89 to 1.69) 1.51 (1.10 to 2.07) 1.28 (1.10 to 1.50) 2.72 (2.10 to 3.52) 2.03 (1.54 to 2.67) 3.28 (2.54 to 4.24) 3.53 (2.71 to 4.60) 1.35 (1.23 to 1.48) 1.36 (0.96 to 1.93) P <.001 — <.001 .01 <.001 .01 — .18 .29 .54 — .42 .74 .01 .60 <.001 .57 — .21 .01 .002 <.001 <.001 <.001 <.001 <.001 .08 Adjusted OR (95% CI) 0.88 (0.84 to 0.92) 1.00 (referent) 0.65 (0.50 to 0.85) 0.32 (0.22 to 0.48) 0.35 (0.24 to 0.50) 1.16 (0.86 to 1.58) 1.00 (referent) 0.69 (0.51 to 0.95) 0.83 (0.52 to 1.31) 0.69 (0.48 to 0.99) 1.00 (referent) 0.92 (0.66 to 1.26) 1.12 (0.76 to 1.65) 1.01 (0.67 to 1.52) 0.93 (0.62 to 1.39) 2.06 (1.28 to 3.32) (not included) 1.00 (referent) 1.41 (1.08 to 1.83) 1.54 (1.23 to 1.92) 1.18 (0.92 to 1.52) 1.47 (1.16 to 1.87) 1.47 (1.13 to 1.92) 1.95 (1.49 to 2.55) 3.57 (2.90 to 4.39) 1.23 (1.09 to 1.39) 0.90 (0.59 to 1.38) P <.001 — .002 <.001 <.001 .34 — .02 .42 .04 — .59 .56 .97 .73 .003 — — .01 <.001 .20 .002 .005 <.001 <.001 <.001 .63 aAll logistic regression models adjust for clustering at diagnosing hospital. All variables are included, and mutually adjusted for, in the multivariable model except for depressive symptoms (which was not statistically significant in the bivariate analyses). P values are 2-sided. CI ¼ confidence interval; OR ¼ odds ratio; — indicates no P value. survivorship clinic equips survivors with greater knowledge about their disease, health risks, and preventive behaviors, which may contribute to greater self-efficacy (51). In turn, greater HCSE supports survivors in seeking out follow-up care and maintaining long-term surveillance. Research among adult cancer survivors has shown that receiving a verbal explanation of follow-up care plans was statistically significantly associated with higher HCSE, and higher HCSE was associated with lower rates of hospitalization, possibly because of the improved ability to manage health preventively (52). Enhancing HCSE through comprehensive patient education can support lifelong health management among CCS. Strengths of this study include the ethnically diverse, re- cently diagnosed, population-based sample with rich survey data. Our response rate was similar to other registry-based epi- demiologic studies of cancer survivorship, despite the chal- lenges of recruiting a younger, more geographically mobile population with a longer time since diagnosis (53,54). We were able to address response bias by weighting our analyses on de- mographic factors related to response (eg, sex). Although we were unable to evaluate nonregistry variables associated with likelihood of study participation (eg, current insurance status), as they were unavailable for survey nonresponders, we believe recruitment bias in this cohort is substantially lower than hospital-based studies where study participants generally have greater health-care access. Additional limitations include the cross-sectional nature of the data, which inhibits causal inference. For example, the posi- tive association between late effects and follow-up care may be due to CCS seeking care because of late effects and/or health- care providers effectively identifying late effects. Analyses were restricted to those diagnosed in 1 geographical region and may not be generalizable to other areas (eg, with different geographi- cally related characteristics related to health-care access). Additionally, the cascade of care does not reflect a unidirec- tional, prescriptive causal pathway. Longitudinal data are needed to clarify causal pathways to better understand optimal points of intervention to maximize the long-term health of CCS. Finally, more in-depth assessments of perceived risk, risk-based surveillance, and care received (eg, chart abstract data validat- receipt of guideline-concordant screening ing self-report, exams) can further contextualize CCS knowledge and their health-care utilization and are the focus of ongoing work. Long-term follow-up care is essential to mitigate the height- ened risk of morbidity among CCS. With growing numbers of cancer survivors, greater efforts are needed to increase health- J. Milam et al. | 9 of 10 Table 4. Multivariable logistic regression models of receipt of cancer-related follow-up care (within prior 2 years), stratified by ethnicity (Hispanic, non-Hispanic)a Hispanic (n ¼ 570) Non-Hispanic (n ¼ 536) Characteristic Years since diagnosis Age at survey completion, y 18-20 21-25 26-30 31-39 Female (vs Male) Socioeconomic status (relative to lowest) Lowest Low Medium High Highest Health insurance (any vs none) No. of late effects (relative to none) 0 1 (cid:3)2 Treatment intensity Received written cancer treatment summary Has doctor for regular (non-cancer) health checkups Discussed cancer-related follow-up care needs with a doctor (in the last 2 years) Knowledge of need of lifelong follow up care Health-care self-efficacy Family influence health-care decisions OR (95% CI) 0.87 (0.82 to 0.93) 1.00 (referent) 0.63 (0.37 to 1.07) 0.32 (0.18 to 0.55) 0.30 (0.13 to 0.66) 1.00 (0.66 to 1.51) 1.00 (referent) 0.64 (0.45 to 0.91) 0.92 (0.43 to 2.00) 0.75 (0.42 to 1.36) 1.57 (0.75 to 3.28) 1.53 (1.00 to 2.35) 1.00 (referent) 1.15 (0.78 to 1.69) 1.67 (1.10 to 2.53) 1.12 (0.65 to 1.94) 1.37 (1.04 to 1.79) 1.70 (1.20 to 2.40) 2.40 (1.82 to 3.16) 3.93 (2.78 to 5.56) 1.25 (1.08 to 1.45) 0.69 (0.39 to 1.21) P <.001 — .08 <.001 .004 .99 — .01 .84 .34 .23 .049 — .47 .02 .67 .03 .004 <.001 <.001 .004 .19 Adjusted OR (95% CI) 0.88 (0.84 to 0.93) 1.00 (referent) 0.66 (0.43 to 1.01) 0.31 (0.20 to 0.48) 0.44 (0.20 to 1.00) 1.28 (0.81 to 2.03) 1.00 (referent) 3.09 (1.14 to 8.37) 2.82 (1.50 to 5.31) 2.47 (1.09 to 5.57) 2.00 (0.91 to 4.38) 2.98 (1.06 to 8.32) 1.00 (referent) 2.17 (1.47 to 3.19) 1.72 (1.02 to 2.92) 1.17 (0.98 to 1.41) 1.64 (1.16 to 2.33) 1.24 (0.78 to 1.97) 1.44 (0.93 to 2.23) 3.70 (2.41 to 5.68) 1.25 (1.08 to 1.53) 1.24 (0.77 to 1.98) P <.001 — .06 <.001 .049 .28 — .003 .001 .03 .08 .04 — <.001 .04 .09 .007 .36 .10 <.001 .04 .38 aAll models adjust for clustering at diagnosing hospital. All variables are included, and mutually adjusted for, in each multivariable logistic regression model. P values are 2-sided. CI ¼ confidence interval; OR ¼ odds ratio; — indicates no P value. care engagement as survivors age and to minimize ethnic dis- parities in access. Based on these results, pragmatic approaches for promoting preventive health management among CCS in- clude patient and provider education, written treatment sum- maries, and standardized plans for transitioning CCS from the pediatric to adult care setting. Funding This work was supported by the National Institute on Minority Health and Health Disparities of the National Institutes of Health (grant number 1R01MD007801) and the National Cancer Institute (grant numbers P30CA014089, T32CA009492). Jessica Tobin was also supported by the VA Office of Academic Affiliations through the Advanced Fellowship Program in Health Services Research and Development. The contents do not represent the views of the U.S. Department of Veterans Affairs or the United States Government. Notes Role of the funder: The study funders played no role in the de- sign of the study; the collection, analysis, and interpretation of the data; the writing of the manuscript; and the decision to sub- mit the manuscript for publication. Disclosures: The authors have no conflicts of disclose. interest to Author contributions: Conceptualization- JM, AH, DF. Data cura- tion- JT. Formal analysis- JM, CR, JT. Funding acquisition- JM. Investigation- JM, DM, CR, JT, KW, AH. Methodology- JM, AH, CR. Project administration- JM, AH, DM. Resources- SG. Software- JT. Supervision- JM, AH, LB, MC. Visualization- JM, CR. Writing- original draft- JM, JT, CR. Writing- review & editing- JM DF KM JT KW CR AR ST LB MC DM SG AH. Prior presentations: This research was previously presented in part at the North American Symposium on Late Complications after Childhood Cancer, Atlanta, GA, June 2019. Acknowledgements: We thank the participants for their in- volvement with this study. Data Availability The data underlying this article cannot be shared publicly due to privacy restrictions of individuals that participated in the study. Aggregated, deidentified data may be shared on reason- able request to the corresponding author. References 1. Ward E, DeSantis C, Robbins A, Kohler B, Jemal A. Childhood and adolescent cancer statistics, 2014. CA Cancer J Clin. 2014;64(2):83–103. 2. National Cancer Institute. Cancer Stat Facts: Cancer Among Adolescents [Published online]. Surveillance, and Young Adults (AYAs) Epidemiology, and End Results (SEER) Program; 2020. (Ages 15-39) 10 of 10 | JNCI Cancer Spectrum, 2021, Vol. 5, No. 5 3. Kremer L, Mulder RL, Oeffinger KC, et al.; for the International Late Effects of Childhood Cancer Guideline Harmonization Group. A worldwide collabora- tion to harmonize guidelines for the long-term follow-up of childhood and young adult cancer survivors: a report from the international late effects of Childhood Cancer Guideline Harmonization Group. Pediatr Blood Cancer. 2013; 60(4):543–549. 4. Oeffinger KC, Mertens AC, Sklar CA, et al.; for the Childhood Cancer Survivor Study. Chronic health conditions in adult survivors of childhood cancer. N Engl J Med. 2006;355(15):1572–1582. 5. Nathan PC, Ford JS, Henderson TO, et al. Health behaviors, medical care, and interventions to promote healthy living in the Childhood Cancer Survivor Study cohort. J Clin Oncol. 2009;27(14):2363–2373. 6. Nathan PC, Hayes-Lattin B, Sisler JJ, Hudson MM. Critical issues in transition and survivorship for adolescents and young adults with cancers. Cancer. 2011;117(10 suppl):2335–2341. 7. Landier W, Bhatia S, Eshelman DA, et al. Development of risk-based guide- lines for pediatric cancer survivors: the Children’s Oncology Group Long- term Follow-up Guidelines from the Children’s Oncology Group Late Effects Committee and Nursing Discipline. J Clin Oncol. 2004;22(24):4979–4990. 8. Milam JE, Meeske K, Slaughter RI, et al. Cancer-related follow-up care among Hispanic and non-Hispanic childhood cancer survivors: the Project Forward study: Follow-Up Care Among Cancer Survivors. Cancer. 2015;121(4):605–613. 9. Nathan PC, Greenberg ML, Ness KK, et al. Medical care in long-term survivors of childhood cancer: a report from the childhood cancer survivor study. J Clin Oncol. 2008;26(27):4401–4409. 10. Oeffinger KC, Mertens AC, Hudson MM, et al. Health care of young adult sur- vivors of childhood cancer: a report from the childhood cancer survivor study. Ann Fam Med. 2004;2(1):61–70. 11. Hewitt M, Weiner SL, Simone, JV. Childhood Cancer Survivorship: Improving Care and Quality of Life. Washington, DC: National Academies Press; 2003. 12. Jones BL. Promoting healthy development among survivors of adolescent cancer. Fam Community Health. 2008;31(suppl 1):S61–70. 13. Yanez B, McGinty HL, Buitrago D, Ramirez AG, Penedo FJ. Cancer outcomes in Hispanics/Latinos in the United States: an integrative review and conceptual model of determinants of health. J Lat Psychol. 2016;4(2):114–129. 14. Mobley E, Moke D, Milam J, Ochoa C, Stal J, Osazuwa N, et al. Disparities and Barriers to Pediatric Cancer Survivorship Care. Rockville, MD: Agency for Healthcare Research and Quality; 2021. 15. May L, Schwartz DD, Frug(cid:2)e E, et al. Predictors of suboptimal follow-up in pe- diatric cancer survivors. J Pediatr Hematol Oncol. 2017;39(3):e143–e149. 16. Zheng DJ, Sint K, Mitchell H-R, Kadan-Lottick NS. Patterns and predictors of survivorship clinic attendance in a population-based sample of pediatric and young adult childhood cancer survivors. J Cancer Surviv. 2016;10(3):505–513. 17. Castellino SM, Casillas J, Hudson MM, et al. Minority adult survivors of child- hood cancer: a comparison of long-term outcomes, health care utilization, and health-related behaviors from the childhood cancer survivor study. J Clin Oncol. 2005;23(27):6499–6507. 18. Devine KA, Viola A, Capucilli P, Sahler OZ, Andolina JR. Factors associated with non-compliance with long-term follow-up care among pediatric cancer survivors. J Pediatr Hematol Oncol. 2017;39(3):167–173. 19. Gibson TM, Mostoufi-Moab S, Stratton KL, et al. Temporal patterns in the risk of chronic health conditions in survivors of childhood cancer diagnosed 1970-99: a report from the Childhood Cancer Survivor Study cohort. Lancet Oncol. 2018;19(12):1590–1601. 20. Hamilton AS, Zhuang X, Modjeski D, Slaughter R, Ritt-Olson A, Milam J. Population-based survey methods for reaching adolescent and young adult survivors of pediatric cancer and their parents. J Adolesc Young Adult Oncol. 2019;8(1):40–48. 21. Yost K, Perkins C, Cohen R, Morris C, Wright W. Socioeconomic status and breast cancer incidence in California for different race/ethnic groups. Cancer Causes Control. 2001;12(8):703–711. 22. Yang J, Schupp CW, Harrati A, Clarke C, Keegan THM, Gomez SL. Developing an Area-Based Socioeconomic Measure from American Community Survey Data. Fremont, CA: Cancer Prevention Institute of California; 2014. 23. Kazak AE, Hocking MC, Ittenbach RF, et al. A revision of the intensity of treat- ment rating scale: classifying the intensity of pediatric cancer treatment. Pediatr Blood Cancer. 2012;59(1):96–99. 24. Tobin JL, Thomas SM, Freyer DR, Hamilton AS, Milam JE. Estimating cancer treatment intensity from SEER cancer registry data: methods and implica- tions for population-based registry studies of pediatric cancers. Cancer Causes Control. 2020;31(10):881–810. 25. Kadan-Lottick NS, Robison LL, Gurney JG, et al. Childhood cancer survivors’ knowledge about their past diagnosis and treatment: childhood cancer survi- vor study. JAMA. 2002;287(14):1832–1839. 26. Whelan KF, Stratton K, Kawashima T, et al. Ocular late effects in childhood and adolescent cancer survivors: a report from the childhood cancer survivor study. Pediatr Blood Cancer. 2010;54(1):103–109. 27. Skinner R, Mulder RL, Kremer LC, et al. Recommendations for gonadotoxicity surveillance in male childhood, adolescent, and young adult cancer survivors: a report from the International Late Effects of Childhood Cancer Guideline Harmonization Group in collaboration with the PanCareSurFup Consortium. Lancet Oncol. 2017;18(2):e75–e90. 28. Desmond RA, Jackson BE, Waterbor JW. Disparities in cancer survivorship indicators in the deep south based on BRFSS data: recommendations for sur- vivorship care plans. South Med J. 2017;110(3):181–187. 29. Lorig K, Stewart A, Ritter P, Gonz(cid:2)alez V, Laurent D, Lynch J. Outcome Measures for Health Education and Other Health Care Interventions. Thousand Oaks, CA: SAGE; 1996. 30. Radloff LS. The CES-D scale: a self-report depression scale for research in the general population. Appl Psychol Meas. 1977;1(3):385–401. 31. Hudson M, Landlier W, Constine L. Long-Term Follow-up Guidelines for Survivors of Childhood, Adolescent, and Young Adult Cancers Version 5.0. Children’s Oncology Group; 2018. http://www.survivorshipguidelines.org/ 32. Ali MK, Bullard KM, Gregg EW, del Rio C. A cascade of care for diabetes in the United States: visualizing the gaps. Ann Intern Med. 2014;161(10):681–689. 33. Kay ES, Batey DS, Mugavero MJ. The HIV treatment cascade and care contin- uum: updates, goals, and recommendations for the future. AIDS Res Ther. 2016;13(1):35–37. 34. Brick JM. Unit nonresponse and weighting adjustments: a critical review. J Off Stat. 2013;29(3):329–353. 35. Boyd RW, Lindo EG, Weeks LD, McLemore MR. On racism: a new standard for publishing on racial health inequities. Health Aff Blog. Published July 2, 2020. doi: 10.1377/hblog20200630.939347. 36. Hosmer DW Jr, Lemeshow, S Sturdivant, RX. Applied Logistic Regression vol 398. Hoboken, NJ: John Wiley & Sons; 2013. 37. Bhatia S, Meadows AT. Long-term follow-up of childhood cancer survivors: future directions for clinical care and research. Pediatr Blood Cancer. 2006; 46(2):143–148. 38. Blaes AH, Adamson PC, Foxhall L, Bhatia S. Survivorship Care Plans and the Commission on Cancer Standards: The Increasing Need for Better Strategies to Improve the Outcome for Survivors of Cancer. Alexandria, VA: American Society of Clinical Oncology; 2020. 39. Nathan PC, Daugherty CK, Wroblewski KE, et al. Family physician preferen- ces and knowledge gaps regarding the care of adolescent and young adult survivors of childhood cancer. J Cancer Surviv. 2013;7(3):275–282. 40. Rokitka DA, Curtin C, Heffler JE, Zevon MA, Attwood K, Mahoney MC. Patterns of loss to follow-up care among childhood cancer survivors. J Adolesc Young Adult Oncol. 2017;6(1):67–73. 41. Shaw SJ, Huebner C, Armin J, Orzech K, Orzech K, Vivian J. The role of culture in health literacy and chronic disease screening and management. J Immigr Minor Health. 2009;11(6):460–467. 42. Jun J, Oh KM. Asian and Hispanic Americans’ cancer fatalism and colon can- cer screening. Am J Health Behav. 2013;37(2):145–154. 43. Shelton RC, Jandorf L, Ellison J, Villagra C, DuHamel KN. The influence of so- ciocultural factors on colonoscopy and FOBT screening adherence among low-income Hispanics. J Health Care Poor Underserved. 2011;22(3):925–944. 44. Cheng EM, Chen A, Cunningham W. Primary language and receipt of recom- mended health care among Hispanics in the United States. J Gen Intern Med. 2007;22(suppl 2):283–288. 45. Arnett JJ. Emerging adulthood: a theory of development from the late teens through the twenties. Am Psychol. 2000;55(5):469–480. 46. Henderson TO, Friedman DL, Meadows AT. Childhood cancer survivors: tran- sition to adult-focused risk-based care. Pediatrics. 2010;126(1):129–136. 47. Frederick NN, Bober SL, Berwick L, Tower M, Kenney LB. Preparing childhood cancer survivors for transition to adult care: the young adult perspective. Pediatr Blood Cancer. 2017;64(10):e26544. 48. Sadak KT, Neglia JP, Freyer DR, Harwood E. Identifying metrics of success for transitional care practices in childhood cancer survivorship: a qualitative study of survivorship providers. Pediatr Blood Cancer. 2017;64(11):e26587. 49. Freyer DR. Transition of care for young adult survivors of childhood and ado- lescent cancer: rationale and approaches. J Clin Oncol. 2010;28(32):4810–4818. 50. Kenney LB, Melvin P, Fishman LN, et al. Transition and transfer of childhood cancer survivors to adult care: a national survey of pediatric oncologists. Pediatr Blood Cancer. 2017;64(2):346–352. 51. Szalda D, Pierce L, Hobbie W, et al. Engagement and experience with cancer- related follow-up care among young adult survivors of childhood cancer af- ter transfer to adult care. J Cancer Surviv. 2016;10(2):342–350. 52. Kenzik KM, Kvale EA, Rocque GB, et al. Treatment summaries and follow-up care instructions for cancer survivors: improving survivor self-efficacy and health care utilization. Oncologist. 2016;21(7):817–824. 53. Harlan LC, Lynch CF, Keegan TH, et al.; for the AYA HOPE Study Collaborative Group. Recruitment and follow-up of adolescent and young adult cancer sur- vivors: the AYA HOPE Study. J Cancer Surviv. 2011;5(3):305–314. 54. Gallicchio L, Elena JW, Fagan S, et al. Utilizing SEER cancer registries for population-based cancer survivor epidemiologic studies: a feasibility study. Cancer Epidemiol Biomarkers Prev. 2020;29(9):1699–1709.
null
10.1016/j.jbc.2022.102361
Data availability Data available upon request. Contact anthony.koleske@yale. edu for more information. The limited proteolysis mass spectrometry data have been deposited to the ProteomeXchange Consortium via the PRIDE (46) partner repository with the dataset identifier PXD034393 (http://www.ebi.ac.uk/pride). The cross-linking raw mass spectrometry data and peak lists are available in the massIVE repository (https://massive.ucsd. edu) with accession number: MSV000089621 Annotated spectra supporting the cross-linked identifica- tions are published on MS-Viewer (https://msviewer.ucsf.edu/ cgi-bin/msform.cgi?form=msviewer) with the following search keys: Trio SR6-GEF1-WT: l4abvtas5a
Data availability Data available upon request. Contact anthony.koleske@yale. edu for more information. The limited proteolysis mass spectrometry data have been deposited to the ProteomeXchange Consortium via the PRIDE (46) partner repository with the dataset identifier PXD034393 ( http://www.ebi.ac.uk/pride ). The cross-linking raw mass spectrometry data and peak lists are available in the massIVE repository ( https://massive.ucsd. edu ) with accession number: MSV000089621
RESEARCH ARTICLE Autoinhibition of the GEF activity of cytoskeletal regulatory protein Trio is disrupted in neurodevelopmental disorder-related genetic variants Received for publication, January 11, 2022, and in revised form, August 4, 2022 Published, Papers in Press, August 10, 2022, https://doi.org/10.1016/j.jbc.2022.102361 Josie E. Bircher1,‡ From the 1Department of Molecular Biophysics and Biochemistry, and 2Keck MS & Proteomics Resource, Yale University, New Haven, Connecticut, USA; 3Department of Pharmaceutical Chemistry, University of California at San Francisco, San Francisco, California, USA; 4Department of Neuroscience, Yale University, New Haven, Connecticut, USA , Michael J. Trnka3, and Anthony J. Koleske1,4,* , Ellen E. Corcoran1,‡ , TuKiet T. Lam1,2 Edited by Kirill Martemyanov TRIO encodes a cytoskeletal regulatory protein with three catalytic domains—two guanine exchange factor (GEF) do- mains, GEF1 and GEF2, and a kinase domain—as well as several accessory domains that have not been extensively studied. Function-damaging variants in the TRIO gene are known to be enriched in individuals with neurodevelopmental disorders (NDDs). Disease variants in the GEF1 domain or the nine adjacent spectrin repeats (SRs) are enriched in NDDs, suggesting that dysregulated GEF1 activity is linked to these disorders. We provide evidence here that the Trio SRs interact intramolecularly with the GEF1 domain to inhibit its enzymatic activity. We demonstrate that SRs 6-9 decrease GEF1 catalytic activity both in vitro and in cells and show that NDD- associated variants in the SR8 and GEF1 domains relieve this autoinhibitory constraint. Our results from chemical cross- linking and bio-layer interferometry indicate that the SRs pri- marily contact the pleckstrin homology region of the GEF1 domain, reducing GEF1 binding to the small GTPase Rac1. Together, our findings reveal a key regulatory mechanism that is commonly disrupted in multiple NDDs and may offer a new target for therapeutic intervention for TRIO-associated NDDs. The TRIO gene encodes a large (>300 kDa) multidomain protein with three catalytic domains (hence the name, Trio): two guanine nucleotide exchange factor (GEF) domains, each composed of Dbl homology (DH) and pleckstrin homology (PH) regions, and a putative serine/threonine kinase domain. The two GEF domains exhibit distinct substrate specificities: the more N-terminal GEF domain (GEF1) promotes GTP loading onto Rac1 and RhoG GTPases (1–3), while the more C-terminal GEF domain (GEF2) activates RhoA (1, 4, 5). Trio also contains an N-terminal lipid-binding Sec14 domain, nine spectrin repeat (SR) domains, and Src homology 3 and immunoglobulin-like domains (1, 6–9). Beyond the potential for protein–lipid and protein–protein interactions, the func- tions of these accessory domains remain poorly understood. ‡ These authors contributed equally to this work. * For correspondence: Anthony J. Koleske, anthony.koleske@yale.edu. De novo mutations and ultra-rare variants in TRIO are enriched in neurodevelopmental disorders (NDDs) (10–14) and the pattern of these variants differs in different disorders. For example, de novo missense and rare damaging variants in the GEF1 domain and adjacent regulatory SRs are enriched in autism, intellectual disability, and developmental delay, sug- gesting that dysregulated GEF1 activity contributes to the pathophysiology of these disorders. Indeed, our lab and others have shown that some of these variants disrupt the ability of GEF1 to catalyze Rac1 activation (12–15). Clusters of variants in the SR8 and GEF1 domains impacted cellular Rac1 activity in different ways and were associated with distinct endophenotypes in heterozygous carriers: SR8 domain variants were linked to developmental delay, mac- rocephaly, and hyperactive Rac1 activity in cells, whereas GEF1 domain variants were linked to mild intellectual disability, microcephaly, and reduced Rac1 activity in cells (15). However, the role of the SRs in Trio function and the mechanism of SR8 variant-mediated increase in Rac1 activity are unclear. Previous studies demonstrated that expression of Trio GEF1 increased Rac1 activity in cells and resulted in dominant gain-of-function pathfinding defects in fly retinal axons (16, 17). Appending additional regions of Trio, including the SRs, to GEF1 attenuated both Trio GEF1-dependent processes. These observations strongly suggest that the SRs reduce GEF1 activity in Trio. However, it remains unknown whether the SRs autoinhibit GEF1 activity directly or via the recruitment of cellular cofactor(s). It is also unclear how variants in the SRs would impact this regulatory mechanism in vitro and in cells. We provide evidence here that SRs 6-9 directly inhibit Trio GEF1 activity in vitro and in cells. Using a GDP-fluorescein (FL)-BODIPY nucleotide exchange assay (18), we show that inclusion of SRs 6-9 is sufficient to inhibit GEF1 activity in vitro, suggesting an autoinhibitory mechanism. We then find that NDD-associated variants in the SR8 and GEF1 domains increase GEF1 activity by relieving autoinhibition, whereas an NDD-associated variant in SR6 reinforces autoinhibition. Using interferometry, we chemical cross-linking and bio-layer J. Biol. Chem. (2022) 298(9) 102361 1 © 2022 THE AUTHORS. Published by Elsevier Inc on behalf of American Society for Biochemistry and Molecular Biology. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). Trio GEF autoinhibition by spectrin repeats demonstrate that the SRs make contact with the PH region of the GEF1 domain and reduce the affinity of GEF1 for Rac1. Together, our findings provide a novel RhoGEF regulatory mechanism by which SRs disrupt Trio GEF1 activation by reducing the interaction of Trio GEF1 with Rac1 and impairing catalytic efficiency. This mechanism appears to be commonly disrupted by NDD-associated variants in TRIO, making it a potential target for therapeutic intervention. Results Inclusion of SRs 6-9 reduces Trio GEF1 activity Genetic variants in SRs 6-9 are associated with NDDs (15), some of which were previously shown to affect Trio-mediated Rac1 activation in cells. To measure the impact of the SRs on GEF1 activity in vitro, we generated and purified Trio GEF1 alone (42 kDa) and a Trio fragment containing SRs 6-9 appended to the GEF1 domain (SR6-GEF1, 99 kDa) (Fig. 1A). Both proteins were monodisperse upon size-exclusion chro- matography and eluted at a position consistent with being monomers (estimated Stokes radius was 3.8 nm for GEF1, 5.6 nm for SR6-GEF1) (Fig. 1B). Using a fluorescence-based guanine nucleotide exchange assay, we measured the cata- lytic activity of GEF1 and SR6-GEF1. Purified 100 nM GEF1 efficiently catalyzed exchange of BODIPY-FL-GDP for GTP on Rac1, with a first-order dissociation rate constant kobs = 2.4 ± −1 (Fig. 1, C and D). Measurement of the rate 0.6 × 10 constant, kobs, as a function of GEF1 concentration yielded a −1 (Fig. 1, E and F). SR6-GEF1 kcat/KM = 1.9 × 104 M similarly promoted GTP exchange onto Rac1 but with a significantly reduced ((cid:1)20 fold and 6-fold, respectively) kobs = −1 (Fig. 1, C, D −1 and kcat/KM = 3.1 × 103 M 1.2 ± 1.8 ×10 and F). These data indicate that inclusion of SRs 6-9 inhibits Trio GEF1 activity for Rac1 in vitro. −1 s −3 s −1 s −4 s Figure 1. Inclusion of SRs 6-9 reduces Trio GEF1 activity on Rac1. A, schematic of Trio proteins: full-length Trio, SR6-GEF1, and GEF1. B, Trio SR6-GEF1 and GEF1 were purified and size-exclusion chromatography was performed to verify that proteins were monodisperse. Dotted lines indicate peak elution volume, which is used to calculate Stokes radii. Samples (approximately 5 μg) of purified components were analyzed by SDS-PAGE and stained with Coomassie Blue to assess purity. Gel images were spliced from separate lanes of the same gel, original gel shown in Figure 3B. C, 100 nM of Trio GEF proteins were incubated with 12.8 μM Rac1 preloaded with 3.2 μM BODIPY-FL-GDP, and nucleotide exchange was tracked via the decrease in fluorescence over time. Representative trace is shown here; traces in color, exponential fits overlaid in black. D, Trio SR6-GEF1 had approximately 20-fold lower exchange activity, kobs, than GEF1 alone. N = 21 independent kobs measurements for overall quantification of rates per group. Bars represent average ± SD; ****p ≤ 0.0001 in a two-tailed t test. E, GEF1 catalytic efficiency was determined by measuring the kobs of GEF1 at multiple concentrations (top) and extracting a linear fit from the plot of kobs versus GEF concentration. Sample traces shown with exponential fits overlaid in black. F, the catalytic efficiency of SR6-GEF1 was 6-fold lower than GEF1 (n = 4). Bars represent average ± SD of four experimental replicates; **p ≤ 0.005 in a two-tailed t test. DH1, Dbl homology domain; FL, fluorescein; GEF, guanine exchange factor; Ig, Ig-like domains; PH1, pleckstrin homology domain; SH3-1, Src homology 3 domain; SR, spectrin repeat. 2 J. Biol. Chem. (2022) 298(9) 102361 Trio GEF autoinhibition by spectrin repeats NDD-associated variants in SR8 increase Trio GEF1 activity in the context of SR6-GEF1 GEF1 variant D1368V increases GEF activity only in the context of SR6-GEF1 We generated and purified SR6-GEF1 expression constructs containing single NDD-associated variants in SR8 and measured their ability to catalyze nucleotide exchange on Rac1 (Fig. 2, A and B). When tested at 100 nM, all SR8 variants, except N1080I, increased the kobs by 4 to 8 fold over that of WT SR6-GEF1 (Fig. 2, C and D). In agreement with these findings, one representative SR8 variant, SR6-GEF1R1078Q, −1, which had a significantly increased kobs = 1.0 ± 0.5 × 10 −1, a 1.5-fold increase in cat- −1 s had a kcat/KM = 4.7 × 103 M alytic efficiency over WT SR6-GEF1 (Fig. 2E). These findings indicate that NDD-associated variants in SR8 are sufficient to relieve SR autoinhibition. −3 s NDD-associated variants in SR6 decrease GEF1 activity in the context of SR6-GEF1 We also generated two SR6-GEF1 constructs harboring individual disease variants in the SR6 domain. While the rate constant (kobs) values obtained for each construct did not significantly decrease compared to WT SR6-GEF1, measure- ment of catalytic efficiency, kcat/KM, of both WT SR6-GEF1 and SR6-GEF1E883D revealed that SR6-GEF1E883D had a significantly decreased catalytic efficiency of a kcat/KM = 1.7 × −1, 1.8-fold lower than WT SR6-GEF1 (Fig. 2E). This 103 M suggests that NDD-associated variants in SR6 decrease GEF1 activity. −1 s Hypothesizing that the SRs might contact GEF1 to impact catalytic activity, we searched for GEF1 domain variants that might impact potential autoinhibition of GEF1 activity by SRs. lie in the GEF1:Rac1 Unlike GEF1 disease variants that interface and decrease GEF1 activity (12–14), D1368V lies in the DH domain but is distal to the GEF1:Rac1 interface, so its impact is less well understood (Fig. 3A). However, introduc- tion of the D1368V variant greatly potentiates the ability of the Trio9 splice isoform, which contains all of the SRs, to increase activity of a Rac1 reporter in cells (14). We intro- duced D1368V into SR6-GEF1 and found that it significantly −1 increased catalytic activity, with a kobs = 1.4 ± 0.3 × 10 −1 (Fig. 3, B–E), a 1.5-fold in- −1 s and kcat/KM = 4.8 × 103 M crease over the kcat/KM for WT SR6-GEF1. In contrast, introducing D1368V into GEF1 alone did not impact its ac- tivity compared to GEF1 (Fig. 3, B–E), indicating that the activating effects of D1368V require SRs 6-9. Together with data reported above, these are consistent with a model in which NDD-associated variants in SR8 and GEF1 relieve in- hibition of GEF1 activity by the SRs. −3 s The SRs and GEF1 form distinct stable interacting domains We used AlphaFold (19, 20) to model human Trio SR6- GEF1 (Fig. 4, A and B). Strikingly, this model suggests that SRs interact with the GEF1 domain, with SR8 closely apposed Figure 2. Mutations in SR6 and SR8 differentially impact GEF1 activity. A, schematic of disease associated mutations in the SRs used in this study. B, mutants were generated in the context of SR6-GEF1 and purified. C, sample GEF assay traces of SR6-GEF1E883D and SR6-GEF1R1078Q. Traces in color, exponential fits overlaid in black. D, SR8 variants in SR6-GEF1 have significantly enhanced catalytic rates, kobs, at equal molar amounts (100 nM) (except N1080I). **p ≤ 0.005; ***p ≤ 0.001; ****p ≤ 0.001 for a significant difference compared to SR6-GEF1 in a one-way ANOVA adjusted for multiple comparisons (n ≥ 9). E, catalytic efficiency (kcat/KM) of representative SR6/8 mutants was determined by measuring the kobs values at different concentrations of GEF, as shown in Figure 1D. The catalytic efficiency of SR6-GEF1R1078Q is (cid:1)1.5-fold greater than that of SR6-GEF1, while the catalytic efficiency of SR6-GEF1E883D is (cid:1)1.8-fold slower (n = 3). Data for GEF1 and SR6-GEF1 from Figure 1 are shown again for reference, and all are reported as an average ± SD of three or more experimental replicates. * = significantly different from SR6-GEF1, p ≤ 0.05 in a one-way ANOVA adjusted for multiple comparisons. GEF, guanine exchange factor; SR, spectrin repeat. J. Biol. Chem. (2022) 298(9) 102361 3 Trio GEF autoinhibition by spectrin repeats Figure 3. GEF1 variant D1368V increases GEF1 activity in the context of SR6-GEF1. A, crystal structure of Trio GEF1 (light and dark blue) and Rac1 (gray), accessed in PDB, ID = 2NZ8 (5). D1368, identified in the box, is distal to the Rac1-binding interface. B, samples (approximately 5 μg) of purified components were analyzed by SDS-PAGE and stained with Coomassie Blue R250 to assess purity. Gel bands for WT SR6-GEF1 and WT GEF1 are the same as shown spliced in Figure 1B. C, sample GEF assay traces of D1368V in the context of SR6-GEF1 and GEF1. Traces in color, exponential fits overlaid in black. D, D1368V in SR6-GEF1 increases catalytic rate, kobs, at equal molar amounts of GEF but has no impact when inserted into GEF1 alone (****p ≤ 0.0001, unpaired t test for mutant versus WT in respective GEF1 or SR6-GEF1, n = 3). E, catalytic efficiency (kcat/KM) of SR6-GEF1D1368V was determined by measuring the kobs values at different concentrations of GEF, as in Figure 1D. Data for GEF1 and SR6-GEF1 shown again for reference. The catalytic efficiency, kcat/KM, of SR6- GEF1D1368V is (cid:1)1.5-fold greater than that of SR6-GEF1 (n = 3). * = significantly different from SR6-GEF1, p ≤ 0.05 in a one-way ANOVA adjusted for mul- tiple comparisons. GEF, guanine exchange factor; SR, spectrin repeat. to GEF1 and the NDD-associated mutations concentrated at this SR8:GEF1 interface. This model of SR6-GEF1 and addi- tional analysis using DISOPRED predicted the existence of an unstructured loop between SR9 and GEF1, suggesting this flexible region may connect the SRs and GEF1 domain (Fig. 4C) (21). We used limited proteolysis to probe for the presence of a flexible linker between SR9 and the GEF1 domain that might be susceptible to partial proteolysis. Treatment of SR6-GEF1 at intermediate levels of trypsin yielded two major bands, identified by mass spectrometry as composed of SRs 6-9 and GEF1, respectively. This observation indicates that SRs 6-9 and the GEF1 domain each make up distinct folding units with increased relative resistance to protease (Fig. 4D). Together, these findings support a model in which the SRs make contact with GEF1. To test directly for possible interactions between the SRs and GEF1 domain, we incubated SR6-GEF1 with an 11.4 Å spacer lysine cross-linker, BS3 (bis(sulfosuccinimidyl)suberate), and analyzed cross-linked peptides via mass spectrometry to identify sites in close enough proximity to cross-link. Several long-distance cross-links were observed between the SRs and the GEF1 domain (Fig. 5A). Specifically, the SR:GEF1 interface includes a peptide in DH domain which is directly at the Rac1 binding interface (1429–1438, green in Fig. 5A) and a peptide in the PH domain important for stabilizing the Rac1 interaction (1529–1537, orange in Fig. 5A) (Fig. 5A) (5). Multiple regions originating in SR6-9 contact these peptides in the GEF domain. This suggests that SR6-GEF1 may be dynamic, with multiple conformational states captured by cross-linking. We hypothe- size that these SR:GEF1 contacts likely disrupt Rac1 binding to GEF1 We also performed chemical cross-linking on three variants in SR6-GEF1 to understand how intramolecular contacts may change in the variants. The SR6-GEF1 variants that display activated GEF activity, R1078Q and D1368V, both exhibited a loss of contact between SR6, 7, 9, and the GEF1 domain (Fig. 5B). In addition, R1078Q, but not D1368V, also reduced SR8:GEF1 contacts (Table S2). In contrast, the SR6 variant, E883D, which reduced GEF activity, did not reduce intra- molecular contacts with GEF1; in fact, new contacts appeared (SR7 and SR9 contacts, blue and purple arrowheads, Fig. 5B), suggesting this variant may reinforce intramolecular SR:GEF contacts (Fig. 5B). These data are consistent with a model in which specific intramolecular contacts between the SRs and GEF1 are altered in genetic variants with increased GEF1 activity. 4 J. Biol. Chem. (2022) 298(9) 102361 Trio GEF autoinhibition by spectrin repeats Figure 4. AlphaFold predicts an interaction between the SRs and GEF1, which form independent folding units. A, AlphaFold model of human Trio SR6-GEF1. SR6, 8 in light pink, SR7, 9 in dark pink, linker region in gray, and GEF1 in blue. Sites of mutations used in this study are modeled as black spheres, with amino acids labeled. This model predicts an interaction between SR8 and GEF1. B, SR6-GEF1 from AlphaFold model, rotated to view flexible linker region between GEF1 and SR9. C, probability of disorder was predicted using DISOPRED. The region between SR9 and DH1 has a high probability of being disordered (cutoff > 0.5). D, limited proteolysis of SR6-GEF1. His-SR6-GEF1 was incubated with increasing concentrations of trypsin and select bands were identified using mass spectrometry. Relative abundance of identified peptides was plotted to determine composition of each band. The y-axis displays relative abundance of peptides and x-axis is ‘amino acid position’, which refers to the location in SR6-GEF1 that the peptide covers (with SR6-GEF1 diagram below). Band 1 (pink box around gel band at (cid:1)60 kDa) comprises SR6-9 and Band 2 (blue box around band at (cid:1)40 kDa) comprises GEF1. Therefore, SR6-9 and GEF1 form distinct stable domains. DH1, Dbl homology domain; GEF, guanine exchange factor; SR, spectrin repeat. The SRs reduce GEF1 binding to Rac1 Based on our cross-linking data, we hypothesized that an interaction between SRs 6-9 and PH1 may impair the ability of GEF1 to bind Rac1. We used bio-layer interferometry to measure the association of nucleotide-free Rac1 with His- GEF1 or His-SR6-GEF1 immobilized on a nitrilotriacetic acid (Ni-NTA) affinity chip. GEF1 bound to Rac1 with a Kd = 151 ± 49 nM in nucleotide-free conditions (Fig. 6, A–C). SR6- GEF1 had a reduced affinity for Rac1, with a Kd = 316 ± 87 nM (Fig. 6, A–C). Taken together with the cross-linking data, this supports a model where the SRs contact the PH domain to impair GEF1 binding to Rac1, which likely contributes to the reduction in observed GEF1 activity. SRs 6-9 inhibit GEF1-induced cell spreading Trio GEF1 activates Rac1 and RhoG to coordinate down- stream cytoskeletal changes and mediate changes in cell morphology (1–3, 22). We first expressed Trio GEF1-GFP in HEK293 cells and quantified its impact on cell morphology (Fig. 7, A–C). When matched for GFP expression levels, GEF1 expressing cells had significantly increased cell area compared to GFP controls (Fig. 7, A–C). Cells expressing GEF1 appeared to be more spread with round lamellipodia encompassing the cell edge, a common result of Rac1 activation (23) (Fig. 7B). The area of cells expressing a catalytic-dead mutant of GEF1, GEF1 ND/AA (N1465A/D1466A), were similar to GFP con- trols, indicating a key role for GEF1 catalytic activity in this morphological change (24). In contrast to GEF1, SR6-GEF1 expressing cells had no measurable effect on cell area, but the SR8 mutant, SR6-GEF1R1078Q, increased cell area over that of GFP and SR6-GEF1 WT (Fig. 7, B and C). Cells expressing SR6-GEF1R1078Q also appeared qualitatively in morphology to those cells expressing GEF1 alone, with more full, rounded edges (Fig. 7B). Therefore, inclusion of SRs 6-9 inhibits Trio GEF1-dependent changes in cell morphology, and disease-associated variants can disrupt this inhibitory regulation. similar We then expressed GFP-Trio9s, a predominant neuronal isoform throughout neurodevelopment, in HEK293 cells and quantified its impact on cell morphology (25) (Fig. 7, A, D and E). Interestingly, when matched for GFP expression levels, GFP-Trio9s expressing cells had significantly decreased cell area compared to GFP controls. Expressing two variants of Trio9s, the most activated SR8 mutant, GFP-Trio9sR1078Q, and a catalytic-dead mutant of GEF1, GFP-Trio9s ND/AA (N1465A/D1466A), decreased cell area compared to GFP alone (Fig. 7, D and E). Cells expressing any variant of J. Biol. Chem. (2022) 298(9) 102361 5 Trio GEF autoinhibition by spectrin repeats Figure 5. The SRs interact with GEF1. A, SR6-GEF1 was incubated with lysine cross-linker BS3 and cross-linked peptides were identified using mass spectrometry. Crystal structure of GEF1 alone (gray, left panel) and with Rac1 (black, right panel) (from PDB, ID = 2NZ8 (5)) with cross-linked peptides between SR6-9 and GEF1 (in WT case) shown in green (1429–1438), pink (1503–1506), orange (1529–1537), purple (1562–1588), and light blue (1574–1588). SR6-9 contacts the DH domain at a peptide that likely interferes with Rac1 binding (1429–1438) and a region in the PH domain critical for stabilizing the Rac1 interaction (1529–1537) (5). B, representative activating mutants (R1078Q and D1368V) display fewer contacts between SR6-9 and GEF1 (lost contacts shown with dotted lines). Representative inactivating mutant (E883D) displays increased contacts between SR6-9 and GEF1 (New contacts shown with blue or purple arrows). Cross-links were categorized based on their N-terminal cross-link site (in SR6, 7, or 9) and their C-terminal GEF1 contacts were visualized. For the activating mutants, the peptides that were mutually lost for both activating mutants were visualized here. For table of all mutant cross-links between SR6-9 and GEF1, see Table S2. BS3, bis(sulfosuccinimidyl)suberate; DH1, Dbl homology domain; GEF, guanine exchange factor; PH1, pleckstrin homology domain; SR, spectrin repeat. GFP-Trio9s appeared very round, completely lacking lamelli- podia or cell edge protrusions (Fig. 7, D and E). We speculate that activity of the Trio GEF2 domain, which targets RhoA to promote cytoskeleton contractility (26), may dominate in this context, making it difficult to discern specific effects on GEF1 activity. interferometry, we show that the SRs contact regions of GEF1 important for Rac1 binding and that inclusion of the SRs is associated with reduced binding affinity for Rac1 in vitro. We present a model for how Trio GEF1 activity is regulated, and how this regulation is disrupted by disorder-associated variants. Discussion Inclusion of Trio SRs autoinhibits GEF1 activity in vitro We provide evidence here that the Trio SRs 6-9 directly inhibit GEF1 activity via intramolecular interactions in vitro and in cells. We demonstrate that NDD-associated variants in the SR8 and GEF1 domains release this autoinhibitory constraint, strongly suggesting that disruption of this GEF1 regulatory mechanism contributes to the pathophysiology of these disorders. Using chemical cross-linking and bio-layer Previous cell-based studies have shown that removing the SRs is associated with increased downstream Rac1 activity and Trio gain-of-function phenotypes in vivo, suggesting that the Trio SRs function to inhibit GEF1 activity (16, 17, 27). This hypothesis is supported by evidence that other RhoGEFs, like Tiam1, contain autoinhibitory N-terminally adjacent accessory domains (8, 28, 29). In most cases, how inhibition occurs and 6 J. Biol. Chem. (2022) 298(9) 102361 Trio GEF autoinhibition by spectrin repeats Figure 6. Inclusion of SRs 6-9 reduce binding to Rac1. A, His-GEF1 or His-SR6-GEF1 were immobilized on an Ni-NTA biosensor and the association of different concentrations of Rac1 was measured. Representative traces shown, with data in color and one phase exponential fits in black. Full concentration gradients (4–5 Rac1 concentrations) were performed at least three independent times. B, kobs values were extracted from each association curve and plotted against Rac1 concentration to calculate a Kd of GEF1 or SR6-GEF1 binding to Rac1. C, SR6-GEF1 has a 2-fold weaker affinity for Rac1 than GEF1 (*p ≤ 0.05, unpaired t test). GEF, guanine exchange factor; Ni-NTA, nitrilotriacetic acid; SR, spectrin repeat. how it is released to activate GEF activity is unknown. Our results show that SR6-GEF1 is monomeric in solution and that inclusion of SRs significantly decreases GEF1 catalytic activity in vitro. Collectively, these observations suggest that the SRs are sufficient to inhibit GEF1 activity via intramolecular in- teractions in cis. SRs make direct contact with GEF1 and impair interactions with Rac1 Within GEF1, the DH1 domain catalyzes GTP exchange onto Rac1 and serves as the main Rac1-binding interface. The PH domain plays a regulatory role in catalysis but also serves to stabilize the Rac1:DH1 interaction (30, 31). Using chemical cross-linking, we demonstrate that SRs 6-9 make extensive contacts with the GEF1 domain, including at sites critical for Rac1 binding, suggesting that SR6-9 sterically blocks contact with Rac1. In addition, NDD-associated variants that activate GEF1 exhibit reduced contacts between the SRs and GEF1 and those that impair GEF1 activity exhibit increased contacts. Hence, altering the interaction between the SRs and GEF1 impacts catalytic activity (5). We found that inclusion of SRs 6-9 reduces the affinity of GEF1 for Rac1 by 2-fold, compared to GEF1 alone. Whereas our catalytic rate measurements suggest the presence of SRs 6-9 results in a 6-fold decrease in activity, the reduction in affinity that we observed was smaller in magnitude. It is likely that engagement of the SRs with GEF1 impairs other steps in the catalytic cycle, as demonstrated by our catalytic efficiency data, in addition to impacting Rac1-binding affinity. Future studies will elucidate whether other components of the nucleotide exchange process are impacted by the SRs. NDD-associated mutations in SR8 and GEF1 disrupt SR-mediated GEF1 inhibition Two rare variant clusters in TRIO, one in SR8 (Fig. 2A) and one in GEF1, have been linked to distinct endophenotypes in individuals with NDDs (15). For example, TRIO SR8 variants are linked to developmental delay and macrocephaly in humans and cause increased Rac1 (GEF1) activity in cells, whereas most mutations in the GEF1 domain are linked to mild intellectual disability, microcephaly, and reduced Rac1 activity in cells. However, how SR8 variants increased Rac1 activity was completely unknown. We hypothesized that the increased Rac1 activity associated with SR8 domain variants resulted from disruption of SR-mediated GEF1 inhibition. We generated mutant SR6-GEF1 constructs harboring distinct disorder-associated variants and found that nearly all SR8 mutants increased SR6-GEF1 catalytic activity 4 to 8 fold. Interestingly, the one exception, N1080I, disrupts binding to neuroligin-1 and blocks neuroligin-1–mediated synapto- genesis (32). We hypothesize that other sites, including N1080I, in the SRs serve as convergence points for upstream activators to regulate GEF1 activity and discuss this in a following section. Together, these data demonstrate that many J. Biol. Chem. (2022) 298(9) 102361 7 Trio GEF autoinhibition by spectrin repeats Figure 7. SRs 6-9 reduce the impact of GEF1 on cell spreading. A, schematic of constructs used, with mutants shown below. B, constructs in (A) were transfected into HEK293 cells and plated on fibronectin. Cells were fixed and stained using anti-GFP to visualize GFP expression and cell morphology. Cells expressing GEF1 and SR6-GEF1R1078Q appeared to have more rounded edges and circular shapes. The scale bar represents 10 μm. Contrast was adjusted between images shown to best visualize cell edge; cell edge is outlined with a white dashed line. C, cell area, normalized to protein expression on a cell-by- cell basis, was quantified. Cell area increased upon expression of GEF1 and SR6-GEF1R1078Q, while expression of a catalytic-dead GEF1 mutant (ND/AA) or SR6-GEF1 had no effect compared to GFP alone. D, cells visualized and analyzed as in (B). The scale bar represents 10 μm. Cells expressing Trio9s constructs all appeared rounder and lacked cell edge protrusions. E, cell area quantified as in (C). Cell area was decreased upon expression of all GFP-Trio9s constructs compared to GFP alone. Trio9sR1078Q did not increase cell area to levels seen with GFP alone. Two biological replicates were performed for each set of constructs, with 25 to 40 cells analyzed per group per replicate (*p ≤ 0.05, ****p ≤ 0.0001, one-way ANOVA between GFP control and each group and adjusted for multiple comparisons). GEF, guanine exchange factor; SR, spectrin repeat. NDD variants in SR8 are sufficient to relieve SR-mediated GEF1 inhibition. We also found that a GEF1 domain variant associated with Rac1 activation in cells likely impacts SR-mediated GEF1 in- hibition. Unlike GEF1 disease variants that lie at the Rac1- binding interface and decrease GEF1 activity, this variant, D1368V, is distal to the Rac1 interface and hyperactivates Rac1 activity in cells when introduced in the Trio9 splice isoform (12–14, 32). Our results indicate that D1368V significantly increases GEF1 activity in the context of SR6-GEF1 but has no effect on GEF1 alone. We propose that D1368V enhances SR6- GEF1 activity by disrupting SR autoinhibition. Indeed, our cross-linking data suggests that contacts between the SRs and GEF1 are reduced for the D1368V variant. NDD-associated variants in SR6 may reinforce SR-mediated GEF1 inhibition We also generated two SR6-GEF1 constructs harboring individual disease variants in the SR6 domain, whose impact on Trio function remains completely unknown. The catalytic efficiency (kcat/KM) of SR6-GEF1E883D was significantly slower than SR6-GEF1, suggesting that SR6 mutants decrease SR6-GEF1 catalytic activity. While the mechanism for this is unclear, one possibility is that SR6 acts as a hinge region allosterically governing the flexibility of the helices surround- ing SR8 and that SR6 variants may decrease the ability for the SRs to release their inhibitory lock on the GEF1 domain. Indeed, we observed more contacts between SR7 and SR9 and the GEF1 domain in SR6-GEF1E883D, suggesting that the intramolecular contacts are more stable or extensive in the variant case. This observation underscores the importance of understanding how dysregulation of Trio GEF1 activity con- tributes to NDDs. The SRs may serve as a target for activators of Trio GEF1 activity We demonstrated that the SRs inhibit Trio GEF1 activity, but it is unclear how inhibition may be released in a cellular context. SR domains are widely accepted as scaffolding pro- teins that coordinate cytoskeletal interactions with high spatial precision. Considering that Trio is known to act downstream of cell surface receptors to coordinate cytoskeletal rearrange- ments, we anticipate that the Trio SRs serve as a target of interaction partners to engage and activate Trio GEF1 activity in cells. Trio SRs interact with diverse cellular partners, including synaptic scaffolding proteins (Piccolo and Bassoon) (33), cell-adhesion molecules (VE-cadherin and Intercellular Adhesion Molecule 1 (ICAM1)) (34, 35), and membrane 8 J. Biol. Chem. (2022) 298(9) 102361 trafficking proteins (RABIN8) (36). These SR-binding partners may engage Trio to coordinate GEF1 activation and/or deac- tivation in a spatiotemporal manner. Indeed, several studies have shown that Trio interactions with binding partners im- pacts Rac1 activity in cells (32, 34, 35, 37, 38). For example, VE-cadherin binds Trio SR5 and SR6, and this interaction locally increases Rac1 activity in cells (34). Similarly, the ICAM1 intracellular tail binds Trio GEF1, and the Trio/ ICAM1 interaction potentiates ICAM1 clustering at adhesion sites, promoting Rac1 activation in cells (35). Finally, the in- tegral membrane protein Kidins220 regulates Rac1-dependent neurite outgrowth via interactions with the Trio SRs (37). While these studies suggest that the Trio signaling partners may engage and activate Trio GEF1 activity, the specific interaction interfaces and binding stoichiometry that mediates GEF1 activation and how they are impacted by disorder- associated variants is presently unknown. Based on our evi- dence that SR8 variants relieve autoinhibitory constraint, we anticipate that SR8 may be a convergence point for upstream activators and coordinated regulation of GEF1 activity. Conclusions TRIO has emerged as a significant risk gene for NDDs. Using biochemical and genetic tools, we identified a novel regulatory mechanism by which Trio SRs inhibit GEF1 activity and showed that disorder-associated variants are sufficient to relieve this autoinhibitory constraint. This discovery will serve as a model to understand how Trio GEF1 is regulated by physiological signals and how its disruption leads to NDDs. This mechanism may also offer a new target for therapeutic interventions for TRIO-associated NDDs. Experimental procedures Expression construct cloning and protein purification Human Trio SR6-GEF1 was PCR amplified and inserted into the pFastBac1 HTa vector (Invitrogen). Site-directed mutagenesis was used to insert point mutations into pFast- Bac1-Hta-SR6-GEF1 construct and confirmed by DNA sequencing. Primers used for cloning are included in Table S1. Recombinant baculoviruses were generated using Sf9 cells (Bac-to-Bac expression system, Thermo Fisher Scientific). Baculoviruses were used to infect Hi5 cells at an estimated multiplicity of infection = 1 for 48 h before lysis in lysis buffer (20 mM Hepes pH 7.25, 500 mM KCl, 5 mM β-mercaptoe- thanol, 5% glycerol, 1% TritonX-100, 20 mM imidazole, 1 mM DTT, 1 mM PMSF, 1× Roche cOmplete protease inhibitors EDTA free) for 20 min at 4 (cid:3)C. Lysates were affinity purified using Ni-NTA resin (Qiagen) and eluted with 250 mM imid- azole. Elution fractions were further purified over an Sephadex 200 (S200) Increase 10/300 GL column into assay buffer (20 mM Hepes pH 7.25, 150 mM KCl, 5% glycerol, 0.01% TritonX-100, 1 mM DTT), aliquoted, and flash frozen for long-term storage. Human Trio GEF1 and Rac1 were generated and affinity purified from bacterial cells as described in Blaise et al. (18). Point mutants were generated using site-directed mutagenesis. Trio GEF autoinhibition by spectrin repeats Following affinity purification, eluted protein was further pu- rified over an S200 Increase column into assay buffer, ali- quoted, and flash frozen for long-term storage. Stokes radii of proteins were estimated based on the elution volume from the S200 Increase column, calculated based on a standard curve generated by running protein standards (Pro- tein Standard Mix 15–600 kDa, Supelco). BODIPY-FL-GDP nucleotide exchange assays 12.8 μM Rac1 was loaded with 3.2 μM BODIPY-FL-GDP (Invitrogen) in 1× assay buffer (20 mM Hepes pH 7.25, 150 mM KCl, 5% glycerol, 1 mM DTT, 0.01% TritonX-100) plus 2 mM EDTA to a total volume of 25 μl per reaction, then incubated for 1 h at room temperature. BODIPY-FL-GDP loading onto Rac1 was halted by the addition of 5 μl of MgCl2, for a total reaction volume of 30 μl with a final MgCl2 con- centration of 5 mM. Prior to initiating the reaction with 100 nM Trio GEF, 30 μl of GTPase (12.8 μM) plus MgCl2 (5 mM) mix or blank (3.2 μM BODIPY-FL-GDP, 2 mM EDTA, and 1× assay buffer) was added to appropriate wells. During the BODIPY-FL-GDP loading incubation period, GEF1- containing proteins were prepared in 1× assay buffer, 4 mM GTP, and 2 mM MgCl2. Exchange reactions were initiated by adding 10 μl of 100 nM Trio GEF mixture (as stated above) to each well, for a total reaction volume of 40 μl. Real-time fluorescence data was measured every 10 s for 30 min moni- toring BODIPY-FL fluorescence by excitation at 488 nm and emission at 535 nm, as per Blaise et al. (18). All kobs measurements of GEF1 activity represent at least three experimental replicates with three technical replicates per experiment. Results are shown as the mean ± SD from multiple experiments. A one-way ANOVA was used to determine statistical significance between SR6-GEF1 and all other variants (two-tailed p-value < 0.05) and adjusted using Dunnett’s multiple comparisons test. Catalytic efficiencies (kcat/KM) of selected SR6-GEF1 constructs were extracted −1) versus GEF1 con- from a linear fit of catalytic rate (kobs, s centration (nM). Three experimental replicates were per- formed for each SR6-GEF1 construct, and the catalytic efficiency values were averaged. Results are shown as the mean ± SD. A one-way ANOVA was used to determine sta- tistical significance between SR6-GEF1 and all other variants (two-tailed p-value < 0.05) and adjusted using Dunnett’s multiple comparisons test. Protein structure predictions AlphaFold was used to access the predicted structure of human Trio spectrin repeats 1-GEF1 (amino acids 201–1600), entry number AF-O75962-F2 (19, 20). Swiss pdb Viewer was used to model SR6-GEF1, amino acids 788 to 1599 (39). DISOPRED was used to predict the probability of disorder of Trio SR6-GEF1, amino acids 788 to 1599 (21). Limited proteolysis SR6-GEF1 in assay buffer plus 10 mM CaCl2 was diluted to 0.4 mg/ml and incubated with increasing concentrations of J. Biol. Chem. (2022) 298(9) 102361 9 Trio GEF autoinhibition by spectrin repeats trypsin (0.001 mg/ml–0.11 mg/ml) for 1 h at room tempera- ture in a 25 μl total reaction volume. Reactions were quenched with 8 μl quench buffer (50 mM Tris–HCl pH 6.8, 4% SDS, 10% glycerol, 0.1% bromophenol blue, 5% β-mercaptoethanol, 1 mM PMSF, 4 mM EGTA, 4 mM EDTA) and immediately boiled for 10 min. Samples were immediately run on a 12% SDS-PAGE gel, and proteins were visualized by Coomassie R250 staining. Major gel bands were excised and washed with 50:50 ace- tonitrile:water buffer containing 100 mM ammonium bicar- bonate. Proteins in the gel were reduced with 4.5 mM DTT at 37 (cid:3)C for 20 min and alkylated with 10 mM iodoacetamide at room temperature for 20 min in the dark. Gel bands were washed twice with 50:50 acetonitrile:water containing 100 mM bicarbonate and dried for 10 min in a SpeedVac. Trypsin digestion was carried out (1:100 M ratio of trypsin to protein) by incubation with the gel piece at 37 (cid:3)C overnight. The digest samples were analyzed by LC–MS/MS using a Q-Exactive Plus mass spectrometer equipped with a Waters nanoACQUITY ultra-performance liquid chromatography system using a Waters Symmetry C18 180 μm by 20 mm trap column and a 1.7 μm (75 μm inner diameter by 250 mm) nanoACQUITY ultra-performance liquid chromatography column (35 (cid:3)C) for peptide separation. Trapping was done at 15 μl/min with 99% buffer A (100% water, 0.1% formic acid) for 1 min. Peptide separation was performed at 300 nl/min with buffer A and buffer B (100% acetonitrile, 0.1% formic acid) over a linear gradient. High-Energy collisional dissociation was utilized to fragment peptide ions via data-dependent acquisition. Mass spectral data were processed with Proteome Discoverer (v. 2.3) and protein database search was carried out in Mascot search engine (Matrix Science, LLC; v. 2.6.0). Protein searches were conducted against the Trichoplusia ni protein database and the human Trio SR6-GEF1 sequence. Mascot search parameters included the following: parent peptide ion tolerance of 10.0 ppm; peptide fragment ion mass tolerance of 0.020 Da; strict trypsin fragments (enzyme cleavage after the C terminus of K or R, but not if it is followed by P); fixed modification of carbamidomethyl (C); and variable modification of phospho (S, T, Y), oxidation (M), and propioamidation (C), and dea- midation (NQ). Peptide identification confidence was set at 95% confidence probability based on Mascot MOWSE score. Results were transferred to Scaffold software (Proteome Soft- ware; v. 4) for further data analysis to look at peptide abun- dances in reference to their start position. These were utilized to plot in a frequency distribution to determine band identity. Cross-linking mass spectrometry Cross-linking experiments were performed as in Sanchez et al. (40) with deviations noted below. Twenty five micro- grams of protein was incubated in assay buffer with 100 μM BS3 (Thermo Fisher) for 30 min on ice. The reaction was quenched by adding Tris pH 7.25 to 10 mM final concentra- tion. Protein was then acetone precipitated and the pellet was alkylated with iodoacetamide and digested with trypsin. Pep- tides were desalted on a 100 μl Omix C18 tip (Agilent), dried, 10 J. Biol. Chem. (2022) 298(9) 102361 and reconstituted in 100 μl of 0.1% formic acid. Mass spec- trometry was performed on an Orbitrap Exploris 480 equipped with an EasySpray nanoESI source, an EasySpray 75 μm × 15 cm C18 column, and a FAIMS Pro ion mobility interface coupled with an UltiMate 3000 RSLCnano system (Thermo Scientific). Each sample was analyzed at four different FAIMS compensation voltages (CV = −40 V, −50 V, −60 V, −70 V) to provide gas-phase enrichment/fractionation of cross-linked peptide ions (41). Each analysis was a separate injection (2.5 μl sample). The sample was loaded at 2% B at 600 nl/min for 35 min followed by a multisegment elution gradient to 35% B at 200 nl/min over 70 min with the remaining time used for column washing and reequilibration (buffer A: 0.1% formic acid (aq); buffer B: 0.1% formic acid in acetonitrile). Precursor ions were acquired at 120,000 resolving power, and ions with charges 3 to 8+ were isolated in the quadrupole using a 1.6 m/z unit window and dissociated by HCD at 30% NCE. Product ions were measured at 30,000 resolving power. Peak lists were generated using PAVA (in house Python app), searched with Protein Prospector v6.3.23 (42), and classified as unique res- idue pairs using Touchstone (an in-house R library) at SVM.score ≥1.5 corresponding to a residue pair level FDR < 0.1% and then further summarized and presented as domain-domain pairs using Touchstone. A custom database consisting of the human Trio construct and a 10× longer decoy database (11 sequences total) was used in the Prospector search, using tryptic specificity with 2-missed cleavages and tolerance of 10/25 ppm (precursor/product). DSS/BS3 cross- linking was specified. Bio-layer interferometry Kinetic binding assays were performed using a ForteBio BLItz instrument. Ni-NTA biosensors were prehydrated in assay buffer for 10 min prior to the experiment. Biosensors were first measured for a baseline signal for 30 s before loading His-GEF1 (0.5 μM) or SR6-GEF1 (2 μM) in assay buffer for 5 min (con- centrations were optimized for reproducible biosensor loading and signal change). Biosensors were then re-equilibrated in assay buffer for 30 s before introducing varying concentrations of Rac1 (at least four concentrations per experiment) in assay buffer for 5 min to measure association. Association curves were fit to a one phase exponential curve to obtain a kobs value and these values were plotted against Rac1 concentration to calculate a Kd from the linear fit of this line, where the y-intercept = koff and slope = kon (Kd = koff/kon). Concentration gradients were replicated at least three times independently, and the Kd measurements of each interaction were compared using an unpaired t test. Reported values are mean ± SD. Measurement of GEF and SR6-GEF1 impact on cell morphology PEI was used to transfect HEK293 cells with 0.5 to 4 μg of DNA in 6-well dishes at a density of 3 × 105 cells per well. Twenty four hours after transfection, cells were trypsinized and replated at a density of 2.5 × 104 cells per coverslip on fibronectin-coated coverslips (10 μg/ml fibronectin). Twenty four hours post plating, cells were fixed and stained as in Lim et al. (43). Cells were fixed for 5 min in 2% paraformaldehyde in cytoskeleton buffer (10 mM MES pH 6.8, 138 mM KCl, 3 mM MgCl2, 2 mM EGTA, 320 mM sucrose). Cells were rinsed three times in Tris Buffered Saline (TBS) (20 mM Tris pH 7.4, 150 mM NaCl) and incubated with 5 μg/ml Alexa Fluor Wheat Germ Agglutinin 555 in TBS (Thermo Fisher) for 10 min to visualize the cell membrane when imaging. Cells were washed another three times in TBS, then permeabilized for 10 min in 0.3% TritonX-100/TBS and washed another three times in 0.1% TritonX-100/TBS. Cells were blocked for 30 min in antibody dilution buffer (ADB) (0.1% TritonX-100, 2% bovine serum albumin, 0.1% NaN3, 10% fetal bovine serum, TBS) and incu- bated with primary antibody (ADB containing a 1:2000 dilution of Goat Anti-GFP, Rockland) at 4 (cid:3)C overnight. The next morning, cells were washed in 0.1% TritonX-100/TBS three times and incubated in secondary antibody for 1 h at room temperature (in ADB, 1:2000 Alexa Fluor 488 Donkey Anti- Goat, Abcam). Cells were washed once in 0.1% TritonX-100/ TBS, once in TBS, and then mounted onto glass slides using AquaMount (Lerner Laboratories). After drying, coverslips were sealed using clear nail polish and imaged using a 40× objective on a spinning disk confocal microscope (UltraVIEW VoX spinning disk confocal (PerkinElmer) Nikon Ti-E-Eclipse), collecting a full z-stack of images for each cell. Identical mi- croscope settings were used between imaging samples. After imaging cells, images were processed using Fiji/ImageJ (44) to generate a sum projection of the GFP channel for quantifying fluorescence as a proxy for total protein expres- sion. Images were then analyzed using CellProfiler to semi- automatically detect cell edges and compute cell area (45). Cell area was normalized for protein expression on a single cell basis by dividing the total area of the cell by the total GFP fluorescence of the cell (a proxy for total protein expression). Two biological replicates were performed, with 25 to 40 cells quantified per group per replicate. Statistical significance of differences in the normalized cell area was determined using a one-way ANOVA between the GFP control and all other groups (two-tailed p-value < 0.05) and adjusted using Dun- nett’s multiple comparisons test. Data availability Data available upon request. Contact anthony.koleske@yale. edu for more information. The limited proteolysis mass spectrometry data have been deposited to the ProteomeXchange Consortium via the PRIDE (46) partner repository with the dataset identifier PXD034393 (http://www.ebi.ac.uk/pride). The cross-linking raw mass spectrometry data and peak lists are available in the massIVE repository (https://massive.ucsd. edu) with accession number: MSV000089621 Annotated spectra supporting the cross-linked identifica- tions are published on MS-Viewer (https://msviewer.ucsf.edu/ cgi-bin/msform.cgi?form=msviewer) with the following search keys: Trio SR6-GEF1-WT: l4abvtas5a Trio GEF autoinhibition by spectrin repeats Trio SR6-GEF1-E883D: mmmpkfzwvo Trio SR6-GEF1-R1078Q: paout3qryt Trio SR6-GEF1-D1368V: 7xhepmd94b Supporting information—This article contains supporting informa- tion (18). Institutes of Health Acknowledgments—The mass spectrometers and the accompany biotechnology tools within the MS & Proteomics Resource at Yale University (used for limited proteolysis experiments) were funded in part by the Yale School of Medicine and by the Office of The Di- rector, National (S10OD02365101A1, S10OD019967, and S10OD018034). Crosslinking Mass spectrom- etry experiments were supported by the Adelson Medical Research Foundation and the University of California, San Francisco Program for Breakthrough Biomedical Research. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the article. We thank Daisy Duan, Amanda Jeng, and Wanqing Lyu for helpful comments on the article and Titus Boggon and Kimmie Vish for helpful insights on structure modeling and protein purification. We also thank Florine Collin and Jean Kanyo for help with mass spectrometry sample preparation and data collection, respectively. Author contributions—J. E. B., E. E. C., and A. J. K. conceptualiza- tion; J. E. B. and E. E. C. methodology; J. E. B., E. E. C., T. T. L., and M. J. T. formal analysis; J. E. B., E. E. C., T. T. L., and M. J. T. investigation; J. E. B., E. E. C., and A. J. K. writing–original draft; J. E. B. and E. E. C. visualization; J. E. B., E. E. C., and A. J. K. funding acquisition; J. E. B., E. E. C., and A. J. K. project administration; T. T. L. and M. J. T. writing–review and editing; A. J. K. supervision. Funding and additional information—This work was supported by the National Institute of Health (NIH) grants R56MH122449, R01 MH115939, and R01 NS105640 to A. J. K., F31MH127891-01 to E. E. C., and F31 NS113511-03 to J. E. B. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. Conflict of interest—The authors declare no competing financial conflicts of interest. Abbreviations—The abbreviations used are: ADB, antibody dilution buffer; BS3, bis(sulfosuccinimidyl)suberate; DH1, Dbl homology domain; FL, fluorescein; GEF, guanine exchange factor; NDD, neurodevelopmental disorder; Ni-NTA, nitrilotriacetic acid; PH1, pleckstrin homology domain; SR, spectrin repeat; TBS, Tris Buff- ered Saline. References 1. Debant, A., Serra-Pages, C., Seipel, K., O’Brien, S., Tang, M., Park, S. H., et al. (1996) The multidomain protein Trio binds the LAR trans- membrane tyrosine phosphatase, contains a protein kinase domain, and has separate rac-specific and rho-specific guanine nucleotide exchange factor domains. Proc. Natl. Acad. Sci. U. S. A. 93, 5466–5471 2. Steven, R., Kubiseski, T. J., Zheng, H., Kulkarni, S., Mancillas, J., Ruiz Morales, A., et al. (1998) UNC-73 activates the Rac GTPase and is required for cell and growth cone migrations in C. elegans. Cell 92, 785–795 3. Penzes, P., Johnson, R. C., Kambampati, V., Mains, R. E., and Eipper, B. A. (2001) Distinct roles for the two Rho GDP/GTP exchange factor domains J. Biol. Chem. (2022) 298(9) 102361 11 Trio GEF autoinhibition by spectrin repeats of kalirin in regulation of neurite growth and neuronal morphology. J. Neurosci. 21, 8426–8434 4. Bellanger, J. M., Lazaro, J. B., Diriong, S., Fernandez, A., Lamb, N., and Debant, A. (1998) The two guanine nucleotide exchange factor domains of Trio link the Rac1 and the RhoA pathways in vivo. Oncogene 16, 147–152 5. Chhatriwala, M. K., Betts, L., Worthylake, D. K., and Sondek, J. (2007) The DH and PH domains of Trio coordinately engage Rho GTPases for their efficient activation. J. Mol. Biol. 368, 1307–1320 6. Rabiner, C. A., Mains, R. E., and Eipper, B. A. (2005) Kalirin: a dual rho guanine nucleotide exchange factor that is so much more than the sum of its many parts. Neuroscientist 11, 148–160 7. Saito, K., Tautz, L., and Mustelin, T. (2007) The lipid-binding SEC14 domain. Biochim. Biophys. Acta 1771, 719–726 8. Schiller, M. R., Ferraro, F., Wang, Y., Ma, X. M., McPherson, C. E., Sobota, J. A., et al. (2008) Autonomous functions for the Sec14p/spectrin- repeat region of Kalirin. Exp. Cell Res. 314, 2674–2691 9. Vishwanatha, K. S., Wang, Y. P., Keutmann, H. T., Mains, R. E., and Eipper, B. A. (2012) Structural organization of the nine spectrin repeats of Kalirin. Biochemistry 51, 5663–5673 10. Bircher, J. E., and Koleske, A. J. (2021) Trio family proteins as regulators of cell migration and morphogenesis in development and disease - mechanisms and cellular contexts. J. Cell Sci. 134, jcs248393 11. Paskus, J. D., Herring, B. E., and Roche, K. W. (2020) Kalirin and trio: RhoGEFs in synaptic transmission, plasticity, and complex brain disor- ders. Trends Neurosci. 43, 505–518 12. Katrancha, S. M., Wu, Y., Zhu, M., Eipper, B. A., Koleske, A. J., and Mains, R. E. (2017) Neurodevelopmental disease-associated de novo mutations and rare sequence variants affect TRIO GDP/GTP exchange factor activity. Hum. Mol. Genet. 26, 4728–4740 13. Pengelly, R. J., Greville-Heygate, S., Schmidt, S., Seaby, E. G., Jabalameli, M. R., Mehta, S. G., et al. (2016) Mutations specific to the Rac-GEF domain of TRIO cause intellectual disability and microcephaly. J. Med. Genet. 53, 735–742 14. Sadybekov, A., Tian, C., Arnesano, C., Katritch, V., and Herring, B. E. (2017) An autism spectrum disorder-related de novo mutation hotspot discovered in the GEF1 domain of Trio. Nat. Commun. 8, 601 24. Debreceni, B., Gao, Y., Guo, F., Zhu, K., Jia, B., and Zheng, Y. (2004) Mechanisms of guanine nucleotide exchange and Rac-mediated signaling J. Biol. Chem. 279, revealed by a dominant negative trio mutant. 3777–3786 25. McPherson, C. E., Eipper, B. A., and Mains, R. E. (2005) Multiple novel isoforms of Trio are expressed in the developing rat brain. Gene 347, 125–135 26. van Rijssel, J., Hoogenboezem, M., Wester, L., Hordijk, P. L., and Van Buul, J. D. (2012) The N-terminal DH-PH domain of Trio induces cell spreading and migration by regulating lamellipodia dynamics in a Rac1- dependent fashion. PLoS One 7, e29912 27. Schiller, M. R., Chakrabarti, K., King, G. F., Schiller, N. I., Eipper, B. A., and Maciejewski, M. W. (2006) Regulation of RhoGEF activity by intra- molecular and intermolecular SH3 domain interactions. J. Biol. Chem. 281, 18774–18786 28. Schmidt, A., and Hall, A. (2002) Guanine nucleotide exchange factors for rho GTPases: turning on the switch. Genes Dev. 16, 1587–1609 29. Xu, Z., Gakhar, L., Bain, F. E., Spies, M., and Fuentes, E. J. (2017) The Tiam1 guanine nucleotide exchange factor is auto-inhibited by its pleckstrin homology coiled-coil extension domain. J. Biol. Chem. 292, 17777–17793 30. Bellanger, J. M., Estrach, S., Schmidt, S., Briancon-Marjollet, A., Zugasti, O., Fromont, S., et al. (2003) Different regulation of the Trio Dbl- homology domains by their associated PH domains. Biol. Cell 95, 625–634 31. Kubiseski, T. J., Culotti, J., and Pawson, T. (2003) Functional analysis of the Caenorhabditis elegans UNC-73B PH domain demonstrates a role in activation of the Rac GTPase in vitro and axon guidance in vivo. Mol. Cell. Biol. 23, 6823–6835 32. Tian, C., Paskus, J. D., Fingleton, E., Roche, K. W., and Herring, B. E. (2021) Autism spectrum disorder/intellectual disability-associated muta- tions in trio disrupt neuroligin 1-mediated synaptogenesis. J. Neurosci. 41, 7768–7778 33. Terry-Lorenzo, R. T., Torres, V. I., Wagh, D., Galaz, J., Swanson, S. K., Florens, L., et al. (2016) Trio, a rho family GEF, interacts with the pre- synaptic active zone proteins Piccolo and Bassoon. PLoS One 11, e0167535 15. Barbosa, S., Debant, A., Schmidt, S., and Baralle, D. (2020) Opposite modulation of RAC1 by mutations in TRIO is associated with distinct, domain-specific neurodevelopmental disorders. Am. J. Hum. Genet. 106, 338–355 34. Timmerman, I., Heemskerk, N., Kroon, J., Schaefer, A., van Rijssel, J., Hoogenboezem, M., et al. (2015) A local VE-cadherin and Trio-based signaling complex stabilizes endothelial junctions through Rac1. J. Cell Sci. 128, 3514 16. Newsome, T. P., Schmidt, S., Dietzl, G., Keleman, K., Asling, B., Debant, A., et al. (2000) Trio combines with dock to regulate Pak ac- tivity during photoreceptor axon pathfinding in Drosophila. Cell 101, 283–294 35. van Rijssel, J., Kroon, J., Hoogenboezem, M., van Alphen, F. P., de Jong, R. J., Kostadinova, E., et al. (2012) The Rho-guanine nucleotide exchange factor Trio controls leukocyte transendothelial migration by promoting docking structure formation. Mol. Biol. Cell 23, 2831–2844 17. Estrach, S., Schmidt, S., Diriong, S., Penna, A., Blangy, A., Fort, P., et al. (2002) The human Rho-GEF trio and its target GTPase RhoG are involved in the NGF pathway, leading to neurite outgrowth. Curr. Biol. 12, 307–312 18. Blaise, A. M., Corcoran, E. E., Wattenberg, E. S., Zhang, Y. L., Cottrell, J. R., and Koleske, A. J. (2022) In vitro fluorescence assay to measure GDP/ GTP exchange of guanine nucleotide exchange factors of Rho family GTPases. Biol. Methods Protoc. 7, bpab024 19. Jumper, J., Evans, R., Pritzel, A., Green, T., Figurnov, M., Ronneberger, O., et al. (2021) Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589 20. Varadi, M., Anyango, S., Deshpande, M., Nair, S., Natassia, C., Yorda- nova, G., et al. (2021) AlphaFold protein structure database: massively expanding the structural coverage of protein-sequence space with high- accuracy models. Nucleic Acids Res. 50, D439–D444 21. Ward, J. J., McGuffin, L. J., Bryson, K., Buxton, B. F., and Jones, D. T. (2004) The DISOPRED server for the prediction of protein disorder. Bioinformatics 20, 2138–2139 22. Hall, A. (1998) Rho GTPases and the actin cytoskeleton. Science 279, 509–514 36. Tao, T., Sun, J., Peng, Y., Li, Y., Wang, P., Chen, X., et al. (2019) Golgi- resident TRIO regulates membrane trafficking during neurite outgrowth. J. Biol. Chem. 294, 10954–10968 37. Neubrand, V. E., Thomas, C., Schmidt, S., Debant, A., and Schiavo, G. (2010) Kidins220/ARMS regulates Rac1-dependent neurite outgrowth by direct interaction with the RhoGEF Trio. J. Cell Sci. 123, 2111 38. Son, K., Smith, T. C., and Luna, E. J. (2015) Supervillin binds the rac/rho- GEF trio and increases trio-mediated Rac1 activation. Cytoskeleton (Hoboken) 72, 47–64 39. Guex, N., and Peitsch, M. C. (1997) Swiss-Model and the Swiss- pdbViewer: an environment for comparative protein modeling. Electro- phoresis 18, 2714–2723 40. Sanchez, N. A., Kallweit, L. M., Trnka, M. J., Clemmer, C. L., and Al- Sady, B. (2021) Heterodimerization of H3K9 histone methyltransferases G9a and GLP activates methyl reading and writing capabilities. J. Biol. Chem. 297, 101276 41. Schnirch, L., Nadler-Holly, M., Siao, S. W., Frese, C. K., Viner, R., and Liu, F. (2020) Expanding the depth and sensitivity of cross-link identifi- cation by differential ion mobility using high-field asymmetric waveform ion mobility spectrometry. Anal. Chem. 92, 10495–10503 23. Wells, C. M., Walmsley, M., Ooi, S., Tybulewicz, V., and Ridley, A. J. (2004) Rac1-deficient macrophages exhibit defects in cell spreading and membrane ruffling but not migration. J. Cell Sci. 117, 1259–1268 42. Trnka, M. J., Baker, P. R., Robinson, P. J., Burlingame, A. L., and Chalkley, R. J. (2014) Matching cross-linked peptide spectra: only as good as the worse identification. Mol. Cell. Proteomics 13, 420–434 12 J. Biol. Chem. (2022) 298(9) 102361 Trio GEF autoinhibition by spectrin repeats 43. Lim, C., Berk, J. M., Blaise, A., Bircher, J., Koleske, A. J., Hochstrasser, M., et al. (2020) Crystal structure of a guanine nucleotide exchange factor encoded by the scrub typhus pathogen Orientia tsutsugamushi. Proc. Natl. Acad. Sci. U. S. A. 117, 30380–30390 44. Schindelin, J., Arganda-Carreras, I., Frise, E., Kaynig, V., Longair, M., Pietzsch, T., et al. (2012) Fiji: an open-source platform for biological- image analysis. Nat. Methods 9, 676–682 45. McQuin, C., Goodman, A., Chernyshev, V., Kamentsky, L., Cimini, B. A., Karhohs, K. W., et al. (2018) CellProfiler 3.0: next-generation image processing for biology. PLoS Biol. 16, e2005970 46. Perez-Riverol, Y., Bai, J., Bandla, C., Hewapathirana, S., García-Seisdedos, D., Kamatchinathan, S., et al. (2022) The PRIDE database resources in 2022: a hub for mass spectrometry-based proteomics evidences. Nucleic Acids Res. 50, D543–D552 J. Biol. Chem. (2022) 298(9) 102361 13
null
10.1038_s41586-023-06415-8.pdf
Data availability Design structures, AF2 models and experimental measurements are available at https://figshare.com/s/439fdd59488215753bc3. Cryo-EM maps and corresponding atomic models for the Influenza HA binder in Fig. 6d–h have been deposited in the PDB and the Electron Microscopy Data Bank under accession codes 8SK7 and EMDB-40557, respectively. Electron microscopy data collected for the HE0537 oligomer are avail- able at EMDB-40602. Code availability Code for running RFdiffusion has been released on GitHub, free for academic, personal and commercial use at https://github.com/Rosetta- Commons/RFdiffusion. It is also available as a Google Colab notebook, accessible through GitHub.
Data availability Design structures, AF2 models and experimental measurements are available at https://figshare.com/s/439fdd59488215753bc3 . Cryo-EM maps and corresponding atomic models for the Influenza HA binder in Fig. 6d -h have been deposited in the PDB and the Electron Microscopy Data Bank under accession codes 8SK7 and EMDB-40557, respectively. Electron microscopy data collected for the HE0537 oligomer are available at EMDB-40602 . Code availability Code for running RFdiffusion has been released on GitHub, free for academic, personal and commercial use at https://github.com/Rosetta- Commons/RFdiffusion . It is also available as a Google Colab notebook, accessible through GitHub.
De novo design of protein structure and function with RFdiffusion https://doi.org/10.1038/s41586-023-06415-8 Received: 14 December 2022 Accepted: 7 July 2023 Published online: 11 July 2023 Open access Check for updates Joseph L. Watson1,2,15, David Juergens1,2,3,15, Nathaniel R. Bennett1,2,3,15, Brian L. Trippe2,4,5,15, Jason Yim2,6,15, Helen E. Eisenach1,2,15, Woody Ahern1,2,7,15, Andrew J. Borst1,2, Robert J. Ragotte1,2, Lukas F. Milles1,2, Basile I. M. Wicky1,2, Nikita Hanikel1,2, Samuel J. Pellock1,2, Alexis Courbet1,2,8, William Sheffler1,2, Jue Wang1,2, Preetham Venkatesh1,2,9, Isaac Sappington1,2,9, Susana Vázquez Torres1,2,9, Anna Lauko1,2,9, Valentin De Bortoli8, Emile Mathieu10, Sergey Ovchinnikov11,12, Regina Barzilay6, Tommi S. Jaakkola6, Frank DiMaio1,2, Minkyung Baek13 & David Baker1,2,14 ✉ There has been considerable recent progress in designing new proteins using deep- learning methods1–9. Despite this progress, a general deep-learning framework for protein design that enables solution of a wide range of design challenges, including de novo binder design and design of higher-order symmetric architectures, has yet to be described. Diffusion models10,11 have had considerable success in image and language generative modelling but limited success when applied to protein modelling, probably due to the complexity of protein backbone geometry and sequence–structure relationships. Here we show that by fine-tuning the RoseTTAFold structure prediction network on protein structure denoising tasks, we obtain a generative model of protein backbones that achieves outstanding performance on unconditional and topology- constrained protein monomer design, protein binder design, symmetric oligomer design, enzyme active site scaffolding and symmetric motif scaffolding for therapeutic and metal-binding protein design. We demonstrate the power and generality of the method, called RoseTTAFold diffusion (RFdiffusion), by experimentally characterizing the structures and functions of hundreds of designed symmetric assemblies, metal- binding proteins and protein binders. The accuracy of RFdiffusion is confirmed by the cryogenic electron microscopy structure of a designed binder in complex with influenza haemagglutinin that is nearly identical to the design model. In a manner analogous to networks that produce images from user-specified inputs, RFdiffusion enables the design of diverse functional proteins from simple molecular specifications. De novo protein design seeks to generate proteins with specified structural and/or functional properties, for example, making a bind- ing interaction with a given target12, folding into a particular topology13 or containing a catalytic site4. Denoising diffusion probabilistic models (DDPMs), a powerful class of machine learning models recently dem- onstrated to generate new photorealistic images in response to text prompts14,15, have several properties well suited to protein design. First, DDPMs generate highly diverse outputs, as they are trained to denoise data (for instance, images or text) that have been corrupted with Gauss- ian noise. By learning to stochastically reverse this corruption, diverse outputs closely resembling the training data are generated. Second, DDPMs can be guided at each step of the iterative generation process towards specific design objectives through provision of conditioning information. Third, for almost all protein design applications it is neces- sary to explicitly model three-dimensional (3D) structures; rotation- ally equivariant DDPMs can do this in a global representation frame independent manner. Recent work has adapted DDPMs for protein monomer design by conditioning on small protein ‘motifs’5,9 or on sec- ondary structure and block-adjacency (‘fold’) information8. Although promising, these attempts have shown limited success in generating sequences that fold to the intended structures in silico5,16, probably due to the limited ability of the denoising networks to generate realistic protein backbones, and have not been tested experimentally. We reasoned that improved diffusion models for protein design could be developed by taking advantage of the deep understanding of protein structure implicit in powerful structure prediction methods 1Department of Biochemistry, University of Washington, Seattle, WA, USA. 2Institute for Protein Design, University of Washington, Seattle, WA, USA. 3Graduate Program in Molecular Engineering, University of Washington, Seattle, WA, USA. 4Columbia University, Department of Statistics, New York, NY, USA. 5Irving Institute for Cancer Dynamics, Columbia University, New York, NY, USA. 6Massachusetts Institute of Technology, Cambridge, MA, USA. 7Paul G. Allen School of Computer Science and Engineering, University of Washington, Seattle, WA, USA. 8National Centre for Scientific Research, École Normale Supérieure rue d’Ulm, Paris, France. 9Graduate Program in Biological Physics, Structure and Design, University of Washington, Seattle, WA, USA. 10Department of Engineering, University of Cambridge, Cambridge, UK. 11Faculty of Applied Sciences, Harvard University, Cambridge, MA, USA. 12John Harvard Distinguished Science Fellowship, Harvard University, Cambridge, MA, USA. 13School of Biological Sciences, Seoul National University, Seoul, Republic of Korea. 14Howard Hughes Medical Institute, University of Washington, Seattle, WA, USA. 15These authors contributed equally: Joseph L. Watson, David Juergens, Nathaniel R. Bennett, Brian L. Trippe, Jason Yim, Helen E. Eisenach, Woody Ahern. ✉e-mail: dabaker@uw.edu Nature | Vol 620 | 31 August 2023 | 1089 Article such as AlphaFold2 (ref. 17) (AF2) and RoseTTAFold18 (RF). RF has prop- erties well suited for use in a protein design DDPM (Fig. 1a): it gener- ates protein structures with high precision, operates on a rigid-frame representation of residues with rotational equivariance and has an architecture enabling conditioning on design specifications at the individual residue, inter-residue distance and orientation, and 3D coordinate levels. In previous work, we fine-tuned RF to complete protein backbones around input functional motifs in a single step (RFjoint Inpainting4). Experimental characterization showed that the method can scaffold a wide range of protein functional motifs with atomic accuracy19, but the approach fails on minimalist site descrip- tions that do not sufficiently constrain the overall fold and, because it is deterministic, can produce only a limited diversity of designs for a given problem. We reasoned that by fine-tuning RF as the denoising net- work in a generative diffusion model instead, we could overcome both problems: because the starting point is random noise, each denoising trajectory yields a different solution, and because structure is built up progressively through many denoising iterations, little to no starting structural information should be required. In this study, we used an updated version of RF18 as the basis for the denoising network archi- tecture (Supplementary Methods), but other equivariant structure prediction networks (AF2 (ref. 17), OmegaFold20, ESMFold21) could in principle be substituted into an analogous DDPM. We construct a RF-based diffusion model, RFdiffusion, using the RF frame representation that comprises a Cα coordinate and N-Cα-C rigid orientation for each residue. We generate training inputs by noising structures sampled from the Protein Data Bank (PDB) for up to 200 steps22. For translations, we perturb Cα coordinates with 3D Gaussian noise. For residue orientations, we use Brownian motion on the mani- fold of rotation matrices (building on refs. 23,24). To enable RFdiffusion to learn to reverse each step of the noising process, we train the model by minimizing a mean-squared error (m.s.e.) loss between frame pre- dictions and the true protein structure (without alignment), averaged across all residues (Supplementary Methods). This loss drives denoising trajectories to match the data distribution at each timestep and hence to converge on structures of designable protein backbones (Extended Data Fig. 2a). The m.s.e. contrasts to the loss used in RF structure predic- tion training (frame aligned point error or FAPE) in that, unlike FAPE, m.s.e. loss is not invariant to the global reference frame and therefore promotes continuity of the global coordinate frame between timesteps (Supplementary Methods). To generate a new protein backbone, we first initialize random resi- due frames and RFdiffusion makes a denoised prediction. Each residue frame is updated by taking a step in the direction of this prediction with some noise added to generate the input to the next step. The nature of the noise added and the size of this reverse step is chosen such that the denoising process matches the distribution of the noising process (Supplementary Methods and Extended Data Fig. 2a). RFdiffusion initially seeks to match the full breadth of possible protein structures compatible with the purely random frames with which it is initialized, and hence the denoised structures do not initially seem protein-like (Fig. 1c, left). However, through many such steps, the breadth of pos- sible protein structures from which the input could have arisen narrows and RFdiffusion predictions come to closely resemble protein struc- tures (Fig. 1c, right). We use the ProteinMPNN network1 to subsequently design sequences encoding these structures, typically sampling eight sequences per design in line with previous work5,16 (but see Supplemen- tary Fig. 2a). We also considered simultaneously designing structure and sequence within RFdiffusion, but given the excellent performance of combining ProteinMPNN with the diffusion of structure alone, we did not extensively explore this possibility. Figure 1a highlights the similarities between RF structure predic- tion and an RFdiffusion denoising step: in both cases, the networks transform coordinates into a predicted structure, conditioned on inputs to the model. In RF, sequence is the primary input, with extra 1090 | Nature | Vol 620 | 31 August 2023 structural information provided as templates and initial coordinates to the model. In RFdiffusion, the primary input is the noised coordinates from the previous step. For specific design tasks, a range of auxiliary conditioning information, including partial sequence, fold informa- tion or fixed functional-motif coordinates can be provided (Fig. 1b and Supplementary Methods). We explored two different strategies for training RFdiffusion: (1) in a manner akin to ‘canonical’ diffusion models, with predictions at each timestep independent of predictions at previous timesteps (as in previous work5,8,9,16), and (2) with self-conditioning25, in which the model can condition on previous predictions between timesteps (Fig. 1a, bottom row and Supplementary Methods). The latter strategy was inspired by the success of ‘recycling’ in AF2, which is also central to the more recent RF model used here (Supplementary Methods). Self-conditioning within RFdiffusion notably improved performance on in silico benchmarks encompassing both conditional and uncondi- tional protein design tasks (Fig. 2e and Extended Data Fig. 1e). Increased coherence of predictions within self-conditioned trajectories may, at least in part, explain these performance increases (Extended Data Fig. 1h). Fine-tuning RFdiffusion from pretrained RF weights was far more successful than training for an equivalent length of time from untrained weights (Extended Data Fig. 1f,g, also Supplementary Fig. 1) and the m.s.e. loss was also crucial for unconditional generation (Extended Data Fig. 1d). For all in silico benchmarks in this paper, we use the AF2 structure prediction network17 for validation and define an in silico ‘success’ as an RFdiffusion output for which the AF2 structure predicted from a single sequence is (1) of high confidence (mean pre- dicted aligned error (pAE), less than five), (2) globally within a 2 Å back- bone root mean-squared deviation (r.m.s.d.) of the designed structure and (3) within 1 Å backbone r.m.s.d. on any scaffolded functional site (Supplementary Methods). This measure of in silico success has been found to correlate with experimental success4,7,26 and is significantly more stringent than template modelling (TM)-score-based metrics used elsewhere5,16,27–29 (Supplementary Fig. 2c,d). Unconditional protein monomer generation As shown in Fig. 2a–c and Supplementary Fig. 3c,d, starting from ran- dom noise, RFdiffusion can readily generate elaborate protein struc- tures with little overall structural similarity to structures seen during training, indicating considerable generalization beyond the PDB (see Supplementary Table 1 for a comparison of all designs in the paper to the PDB). The designs are diverse (Supplementary Fig. 3a), spanning a wide range of alpha, beta and mixed alpha–beta topologies, with AF2 and ESMFold (Fig. 2c, Extended Data Fig. 1b,c and Supplemen- tary Fig. 2b) predictions very close to the design structure models for de novo designs with as many as 600 residues. RFdiffusion generates plausible structures for even very large proteins, but these are difficult to validate in silico as they are probably generally beyond the single sequence prediction capabilities of AF2 and ESMFold. The quality and diversity of designs that are sampled are inherent to the model, and do not depend on any auxiliary conditioning input (for example, second- ary structure information8). We experimentally characterized six of the 300 amino acid designs and three of the 200 amino acid designs, and found that they have circular dichroism spectra consistent with the mixed alpha–beta topologies of the designs and are extremely thermostable (Extended Data Fig. 3). Physics-based protein design methodologies have struggled in unconstrained generation of diverse protein monomers because of the difficulty of sampling on the very large and rugged conformational landscape30, and overcoming this limitation has been a primary test of deep-learning based protein design approaches5,6,8,16,27,31. RFdiffusion strongly outperforms (based on the AF2 success metric described above) Hallucination with RF, an experimentally validated method using Monte Carlo search or gradient descent to identify sequences predicted to fold into stable structures Article a Diffusion model b XT X0 Unconditional Forward (noising) process N(0,1) Gaussian noise ... Single step Protein structure XT Xt Xt–1 X0 Reverse (generative) process MADHTI?DTREE RF RoseTTAFold Input sequence Homologous templates Initial/recycled coordinates RFdiffusion Masked input sequence ˆX0 (self- conditioning) t+1 Xt Diffused coordinates Predicted structure ???????????? RF Single RFdiffusion step Xt RF ˆX0 ˆ interp(Xt, X0) + ε Xt–1 Self-conditioning Symmetric noise Symmetric oligomers Binding target Binder design ˆ X0 Functional motif Motif scaffolding Symmetric motif Symmetric scaffolding c t = 200 t = 175 t = 150 t = 125 t = 100 t = 1 ) t u p n i ( t X ) n o i t c d e r p i ( 0 X ˆ Fig. 1 | Protein design using RFdiffusion. a, Diffusion models for proteins are trained to recover corrupted (noised) protein structures and to generate new structures by reversing the corruption process through iterative denoising of initially random noise XT into a realistic structure X0 (top panel). The RF structure prediction network (middle panel, left side) is fine-tuned with minimal architectural changes into RFdiffusion (middle panel, right side); the denoising network of a DDPM is also shown. In RF, the primary input to the model is the sequence. In RFdiffusion, the primary input is diffused residue frames (coordinates and orientations). In both cases, the model predicts final 3D coordinates (denoted X0 in RFdiffusion). The bottom panel shows that in RFdiffusion, the model receives its previous prediction as a template input (‘self-conditioning’, Supplementary Methods). At each timestep t of a trajectory (typically 200 steps), RFdiffusion takes X 0 from the previous step and Xt and +1 t t  ). The next coordinate input to then predicts an updated X0 structure (X 0 t the model (Xt−1) is generated by a noisy interpolation (interp) towards X 0. b, RFdiffusion is broadly applicable for protein design. RFdiffusion generates protein structures either without further input (top row) or by conditioning on (top to bottom): symmetry specifications; binding targets; protein functional motifs or symmetric functional motifs. In each case random noise, along with conditioning information, is input to RFdiffusion, which iteratively refines that noise until a final protein structure is designed. c, An example of an unconditional design trajectory for a 300-residue chain, depicting the input to  prediction. At early timesteps (high t), the model (Xt) and the corresponding X0  bears little resemblance to a protein but is gradually refined into a realistic X0 protein structure. (Fig. 2d). RFdiffusion generation is also more compute efficient than unconstrained Hallucination with RF, and efficiency can be greatly improved by taking larger steps at inference time and by truncating tra- jectories early, which is possible because RF predicts the final structure at each timestep (Extended Data Fig. 2b,c). For example, a 100-residue protein can be generated in as little as 11 s on an NVIDIA RTX A4000 Graphical Processing Unit, in contrast to RF Hallucination, which takes around 8.5 min. It is often desirable to be able to specify a protein fold during design (such as triose-phosphate isomerase (TIM) barrels or cavity-containing Nature | Vol 620 | 31 August 2023 | 1091 300 amino acids 600 amino acids a b f B D P o t e r o c s M T t s e h g H i 0.9 0.8 0.7 0.6 0.5 0.4 0.3 3 _ a a 0 0 3 8 _ a a 0 0 3 Top TMAlign match to PDB 100 200 300 400 600 800 1,000 c ) Å ( . d . s . m . r i n g s e d s u s r e v 2 F A 40 35 30 25 20 15 10 5 0 ) Å ( . d . s . m . r r.m.s.d. design vs AlphaFold2 Zoom 5.0 2.5 0 100 400 300 200 No. of amino acids 100 200 300 400 600 800 1,000 Number of amino acids Number of amino acids CD spectra ×104 300aa_3 300aa_8 1 0 –1 –2 ) 1 – l o m d 2 m c g e d ( E R M 200 225 250 Wavelength (nm) ×104 CD melts 300aa_3 300aa_8 ) 1 – l o m d 2 m c g e d ( 2 2 2 E R M 1 0 –1 –2 Design AF2 25 35 45 55 65 75 85 95 Temperature (ºC) e y t i s n e D 0.25 0.20 0.15 0.10 0.05 0 0 30 25 20 15 10 5 0 d ) Å ( . d . s . m . r i n g s e d s u s r e v 2 F A g r.m.s.d. AlphaFold2 vs design Zoom ) Å ( . d . s . m . r 2 1 0 100 70 No. of amino acids Design method Hallucination RFdiffusion 70 100 200 300 Number of amino acids TIM barrel 6WVS ) 1 – l o m d 2 m c g e d ( Design AF2 Ablations reveal the determinants of RFdiffusion performance RFdiffusion No pretraining (eq. compute) No fine-tuning (RF weights) No MSE Loss No self-conditioning 10 20 30 r.m.s.d. AF2 versus design (Å) CD spectra ×104 6WVS TIM_barrel_6 1.0 0.5 0 –0.5 E R M –1.0 200 225 Wavelength (nm) 250 TIM_barrel_6 Design AF2 ) 1 – l o m d 2 m c g e d ( 2 2 2 E R M 1.0 0.5 0 –0.5 –1.0 –1.5 ×104 CD melts 6WVS TIM_barrel_6 25 35 45 55 65 75 85 95 Temperature (ºC) Fig. 2 | Outstanding performance of RFdiffusion for monomer generation. a, RFdiffusion can generate new monomeric proteins of different lengths (left 300, right 600) with no conditioning information. Grey, design model; colours, AF2 prediction. r.m.s.d. AF2 versus design (Å), left to right: 0.90, 0.98, 1.15, 1.67. b, Unconditional designs from RFdiffusion are new and not present in the training set as quantified by highest TM-score to the PDB; the divergence from previously known structures increases with length. c, Unconditional samples are closely repredicted by AF2 up to about 400 amino acids. d, RFdiffusion significantly outperforms Hallucination (with RF) at unconditional monomer generation (two-proportion z-test of in silico success: n = 400 designs per condition, z = 9.5, P = 1.6 × 10−21). Although Hallucination successfully generates designs up to 100 amino acids in length, in silico success rates rapidly deteriorate beyond this length. e, Ablating pretraining (by starting from untrained RF), RFdiffusion fine-tuning (that is, using original RF structure prediction weights as the denoiser), self-conditioning or m.s.e. losses (by training with FAPE) each notably decrease the performance of RFdiffusion. r.m.s.d. between design and AF2 is shown, for the unconditional generation of 300 amino acid proteins (Supplementary Methods). f, Two example 300 amino acid proteins that expressed as soluble monomers. Designs (grey) overlaid with AF2 predictions (colours) are shown on the left, alongside circular dichroism (CD) spectra (top) and melt curves (bottom) on the right. The designs are highly thermostable. g, RFdiffusion can condition on fold information. An example TIM barrel is shown (bottom left), conditioned on the secondary structure and block adjacency of a previously designed TIM barrel, PDB 6WVS (top left). Designs have very similar circular dichroism spectra to PDB 6WVS (top right) and are highly thermostable (bottom right). See also Extended Data Fig. 3 for further traces. Boxplots represent median ± interquartile range; tails are minimum and maximum excluding outliers (±1.5× interquartile range). NTF2s for small molecule binder and enzyme design32,33), and thus we further fine-tuned RFdiffusion to condition on secondary structure and/or fold information, enabling rapid and accurate generation of diverse designs with the desired topologies (Fig. 2g and Extended Data Fig. 4). In silico success rates were 42.5 and 54.1% for TIM barrels and NTF2 folds, respectively (Extended Data Fig. 4d), and experimental 1092 | Nature | Vol 620 | 31 August 2023 Article characterization of 11 TIM barrel designs indicated that at least eight designs were soluble, thermostable and had circular dichroism spectra consistent with the design model (Fig. 2g and Extended Data Fig. 4e,f). Design of higher-order oligomers There is considerable interest in designing symmetric oligomers, which can serve as vaccine platforms34, delivery vehicles35 and catalysts36. Cyclic oligomers have been designed using structure prediction net- works with an adaptation of Hallucination that searches for sequences predicted to fold to the desired cyclic symmetry, but this approach fails for higher-order dihedral, tetrahedral, octahedral and icosahedral symmetries, probably in part because of the much lower representation of such structures in the PDB7. We set out to generalize RFdiffusion to create symmetric oligomeric structures with any specified point group symmetry. Given a specifica- tion of a point group symmetry for an oligomer with n chains, and the monomer chain length, we generate random starting residue frames for a single monomer subunit as in the unconditional generation case, and then generate n − 1 copies of this starting point arranged with the specified point group symmetry. Because RFdiffusion is equivariant (inherited from RF) with respect to rotation and relabelings of chains, symmetry is largely maintained in the denoising predictions; we explic- itly resymmetrize at each step but this changes the structures only slightly (compare grey and coloured chains in Extended Data Fig. 5a and Supplementary Methods). For octahedral and icosahedral archi- tectures, we explicitly model only the smallest subset of monomers required to generate the full assembly (for example, for icosahedra, the subunits at the five-, three- and twofold symmetry axes) to reduce the computational cost and memory footprint. Despite not being trained on symmetric inputs, RFdiffusion is able to generate symmetric oligomers with high in silico success rates (Extended Data Fig. 5b), particularly when guided by an auxiliary inter- and intrachain contact potential (Extended Data Fig. 5c). As illustrated in Fig. 3 and Extended Data Fig. 5e, RFdiffusion designs are nearly indis- tinguishable from AF2 predictions of the structures adopted by the designed sequences, and many show little resemblance to previously solved protein structures (Extended Data Fig. 5d and Supplementary Table 1). Several of the oligomeric topologies are not seen in the PDB, including two-layer beta barrels (Fig. 3a, C10 symmetry) and complex mixed alpha/beta topologies (Fig. 3a, C8 symmetry; closest TM align in PDB 6BRP, 0.47, and PDB 6BRO, 0.43, respectively). We selected 608 designs for experimental characterization and found using size-exclusion chromatography (SEC) that at least 87 had oligomerization states closely consistent with the design mod- els (within the 95% confidence interval, 126 designs within the 99% confidence interval, as determined by SEC calibration curves; Sup- plementary Figs. 4 and 5). We took advantage of the increased size of these oligomers (compared to the smaller unconditional and fold-conditioned monomers described above) and collected nega- tive stain electron microscopy (nsEM) data on a subset of these designs across different symmetry groups. For most, distinct particles were evident with shapes resembling the design models in both the raw micrographs and subsequent two-dimensional (2D) classifications (Fig. 3 and Extended Data Fig. 5f). nsEM characterization of a C3 design (HE0822) with 350 residue subunits (1,050 residues in total) suggests that the actual structure is very close to the design, both over the 350 residue subunits and the overall C3 architecture. 2D class averages are clearly consistent with both top and side views of the design model, and a 3D reconstruction of the density has key features consistent with the design, including the distinctive pinwheel shape (Fig. 3b, top row). Electron microscopy 2D class averages of C5 and C6 designs with more than 750 residues (HE0794, HE0789, HE0841) were also consistent with the respective design models (Extended Data Fig. 5f). RFdiffusion also generated cyclic oligomers with alpha and/or beta barrel structures that resemble expanded TIM barrels and provide an interesting comparison between innovation during natural evolution and innovation through deep learning. The TIM barrel fold, with eight strands and eight helices, is one of the most abundant folds in nature37. nsEM confirmed the structure of two RFdiffusion designed cyclic oli- gomers, which considerably extend beyond this fold (Fig. 3b, bottom rows). HE0626 is a C6 alpha–beta barrel composed of 18 strands and 18 helices, and HE0675 is a C8 octamer composed of an inner ring of 16 strands and an outer ring of 16 helices arranged locally in a very similar repeating pattern to the TIM barrel (1:1 helix:strand). For both HE0626 and HE0675 we obtained nsEM 3D reconstructions that are in agree- ment with the computational design models. The HE0600 design is also an alpha–beta barrel (Extended Data Fig. 5f), but has two strands for every helix (24 strands and 12 helices in total) and hence is locally different from a TIM barrel. Whereas natural evolution has extensively explored structural variations of the classic eight-strand or eight-helix TIM barrel fold, RFdiffusion can more readily explore global changes in barrel curvature, enabling discovery of TIM barrel-like structures with many more helices and strands. RFdiffusion also readily generated structures with dihedral, tet- rahedral and icosohedral symmetries (Fig. 3c,d and Extended Data Fig. 5e,f). SEC characterization indicated that 38 D2, seven D3 and three D4 designs had the expected molecular weights (these have four, six and eight chains, respectively) (Supplementary Fig. 5). Although the D2 dihedrals are too small for nsEM, 2D class averages—and for some, 3D reconstructions of D3 and D4 designs—were congruent with the overall topologies of the design models (Fig. 3c and Extended Data Fig. 5f). Similarly, 3D reconstruction (Fig. 3c) and cryogenic electron microscopy (cryo-EM) 2D class averages (Extended Data Fig. 5g and Sup- plementary Fig. 6) of the D4 HE0537 closely match the design model, recapitulating the roughly 45° offset between tetramic subunits. 2D nsEM class averages for a 12-chain tetrahedron (HE0964) were consist- ent with the design model (Extended Data Fig. 5f). Forty-eight icosa- hedra were selected for experimental validation, and one, HE0902, a 15 nm (diameter) highly porous assembly (Fig. 3d, left) was observed in nsEM micrographs to form homogeneous particles. 2D class averages and a 3D reconstruction very closely match the design model (Fig. 3d), with triangular hubs arrayed around the empty C5 axes. Designs such as HE0902 (and future similar large assemblies) should be useful as new nanomaterials and vaccine scaffolds, with robust assembly and (in the case of HE0902) the outward facing N and C termini offering many possibilities for antigen display. Functional-motif scaffolding We next investigated the use of RFdiffusion for scaffolding protein structural motifs that carry out binding and catalytic functions, in which the role of the scaffold is to hold the motif in precisely the 3D geometry needed for optimal function. In RFdiffusion, we input motifs as 3D coordinates (including sequence and sidechains) both during conditional training and inference, and build scaffolds that hold the motif atomic coordinates in place. Many deep-learning methods have been developed recently to address this problem, including RFjoint Inpainting4, constrained Hallucination4 and other DDPMs5,8,29. To rigorously evaluate the performance of these methods in comparison to RFdiffusion across a broad set of design challenges, we established an in silico benchmark test (Supplementary Table 9) comprising 25 motif-scaffolding design problems addressed in six recent publications encompassing several design methodologies4,5,29,38–40. The challenges span a broad range of motifs, including simple ‘inpainting’ problems, viral epitopes, receptor traps, small molecule binding sites, binding interfaces and enzyme active sites. RFdiffusion solves 23 of the 25 benchmark problems, compared to 15 for Hallucination and 19 for RFjoint Inpainting (Fig. 4a,b). For 19 out Nature | Vol 620 | 31 August 2023 | 1093 a D2 C6 C8 C10 O d m n 5 1 b RFdiffusion AF2 2D class averages 3D reconstruction HE0822 C3 HE0626 C6 HE0675 C8 c HE0490 D3 HE0537 D4 90º 90º 90º 90º 90º HE0902 HE0902 Representative Representative micrograph micrograph C3 C3 C2 C2 C5 C5 50 nm 90 A Fig. 3 | Design and experimental characterization of symmetric oligomers. a, RFdiffusion-generated assemblies overlaid with the AF2 structure predictions based on the designed sequences; in all five cases they are nearly indistinguishable (for the octahedron (bottom), the prediction was for the C3 substructure). Symmetries are indicated to the left of the design models. b,c, Designed assemblies characterized by nsEM. Model symmetries are as follows: cyclic, C3 (HE0822, 350 amino acids (AA) per chain), C6 (HE0626, 100 AA per chain) and C8 (HE0675, 60 AA per chain) (b); dihedral, D3 (HE0490, 80 AA per chain) and D4 (HE0537, 100 AA per chain) (c). From left to right: (1) symmetric design model, (2) AF2 prediction of design following sequence design with ProteinMPNN, (3) 2D class averages showing both top and side views (scale bar, 60 Å for all class averages) and (4) 3D reconstructions from class averages with the design model fit into the density map. The overall shapes are consistent with the design models, and confirm the intended oligomeric state. As in a, AF2 predictions of each design are nearly indistinguishable from the design model (backbone r.m.s.d.s (Å) for HE0822, HE0626, HE0490, HE0675 and HE0537, are 1.33, 1.03, 0.60, 0.74 and 0.75, respectively). d, nsEM characterization of an icosahedral particle (HE0902, 100 AA per chain). The design model, including the AF2 prediction of the C3 subunit are shown on the left. nsEM data are shown on the right: on top, a representative micrograph is shown alongside 2D class averages along each symmetry axis (C3, C2 and C5, from left to right) with the corresponding 3D reconstruction map views shown directly below overlaid on the design model. of 23 of the problems solved by RFdiffusion, the fraction of successful designs is higher than either Hallucination or RFjoint Inpainting. The excellent performance of RFdiffusion required no hyperparameter tun- ing or external potentials; this contrasts with Hallucination, for which problem-specific optimization can be required. In 17 out of 23 of the problems, RFdiffusion-generated successful solutions with higher in silico success rates when noise was not added during the reverse diffu- sion trajectories (see Extended Data Fig. 1i for further discussion on the 1094 | Nature | Vol 620 | 31 August 2023 Article a ) % ( e t a r s s e c c u s o c i l i s n I 1.0 0.8 0.6 0.4 0.2 0 c Mdm2 p53 f RFdiffusion outperforms hallucination and RFjoint 5TRV long 7MRX 128 b Diffusion, noise = 0 Diffusion, noise = 1 Hallucination, MPNN Hallucination, no MPNN RFjoint, MPNN RFjoint, no MPNN 6E6R long 5TPN R C Y 1 1 W V 6 N P T 5 F C B 1 d e m _ R 6 E 6 8 L K 2 l g n o _ R 6 E 6 l g n o _ Z X E 6 P Y Z 4 d e m _ Z X E 6 t r o h s _ R 6 E 6 l g n o _ V R T 5 T X 3 I W R P 1 5 8 _ X R M 7 d e m _ V R T 5 8 2 1 _ X R M 7 0 6 _ X R M 7 t r o h s _ V R T 5 t r o h s _ Z X E 6 S U 5 I I U Y 5 9 N W 5 W H J 4 G J Q 1 d ) m n ( s t i n u e s n o p s e R 0.25 0.20 0.15 0.10 0.05 0 ) m n ( s t i n u e s n o p s e R 0.30 0.25 0.20 0.15 0.10 0.05 0 KD = 0.7 nM 0 250 500 750 1,000 Time (s) 0 250 750 500 Time (s) 1,000 Native enzyme Input Design Zoom Oxidoreductase (EC1) g ) % ( e t a r s s e c c u s o c i l i s n I 5 4 3 2 1 0 e 0.2 ) m n ( e s n o p s e R 0.1 [Binder] (nM) 333 111 37 12 4 1 KD = 0.5 nM 0 p53/MDM2 Enzyme active site scaffolding EC1 EC2 EC3 EC4 EC5 Fig. 4 | Scaffolding of diverse functional sites with RFdiffusion. a, RFdiffusion outperforms other methods across 25 benchmark motif-scaffolding problems collected from six recent publications (Supplementary Table 9). In silico success is defined as AF2 r.m.s.d. to design model less than 2 Å, AF2 r.m.s.d. to the native functional motif less than 1 Å and AF2 pAE less than five. One hundred designs were generated per problem, with no previous optimization on the benchmark set (some optimization was necessary for Hallucination). Supplementary Table 10 presents full results. In silico success rates on the problems are correlated between the methods, and RFdiffusion can still struggle on challenging problems in which all methods have low success. b, Four examples of designs in which RFdiffusion significantly outperforms existing methods. Teal, native motif; colours, AF2 prediction of a design. Metrics (r.m.s.d. AF2 versus design/versus native motif (Å), AF2 pAE): 5TRV long, 1.17/0.57; 4.73; 6E6R long, 0.89/0.27, 4.56; 7MRX long, 0.84/0.82 4.32; 5TPN, 0.59/0.49 3.77. c, RFdiffusion can scaffold the p53 helix that binds MDM2 (left) and makes extra contacts with the target (right, average 31% increased surface area. Design was p53_design_89). Designs were generated with an RFdiffusion model fine-tuned on complexes. d, BLI measurements indicate high-affinity binding to MDM2 (p53_design_89, 0.7 nM; p53_design_53, 0.5 nM); the native affinity is 600 nM (ref. 42). e, Out of 95 designs, 55 showed binding to MDM2 (more than 50% of maximum response). Thirty-two of these were monomeric (Supplementary Fig. 10h). f, After fine-tuning (Supplementary Methods), RFdiffusion can scaffold enzyme active sites. An oxidoreductase example (EC1) is shown (PDB 1A4I); catalytic site (teal); RFdiffusion output (grey, model; colours, AF2 prediction); zoom of active site. AF2 versus design backbone r.m.s.d. 0.88 Å, AF2 versus design motif backbone r.m.s.d. 0.53 Å, AF2 versus design motif full-atom r.m.s.d. 1.05 Å, AF2 pAE 4.47. g, In silico success rates on active sites derived from EC1-5 (AF2 Motif r.m.s.d. versus native: backbone less than 1 Å, backbone and sidechain atoms less than 1.5 Å, r.m.s.d. AF2 versus design less than 2 Å, AF2 pAE less than 5). effect of noise on design quality, and Supplementary Fig. 8 for analysis of design diversity). The ability of RFdiffusion to scaffold functional motifs is not related to their presence in the RFdiffusion training set (Supplementary Fig. 7). One of the benchmark problems is the scaffolding of the p53 helix that binds MDM2. Inhibiting this interaction through high-affinity competitive inhibition by scaffolding the p53 helix and making further interactions with MDM2 is a promising therapeutic avenue41. In silico Nature | Vol 620 | 31 August 2023 | 1095 success has been described elsewhere4, but experimental success has not been reported. We used an RFdiffusion model fine-tuned on protein complexes (Supplementary Methods) to generate 96 designs scaffolding this helix. We scaffolded the p53 helix in the presence of MDM2, so extra interactions could be designed by RFdiffusion and experimentally identified 0.5 and 0.7 nM binders (Fig. 4c,d), three orders of magnitude higher affinity than the reported 600 nM affinity of the p53 peptide alone42. The overall success rate was quite high: out of the 96 designs, 55 showed some detectable binding at 10 μM (Fig. 4e and Supplementary Fig. 10h). Scaffolding enzyme active sites A grand challenge in protein design is to scaffold minimal descriptions of enzyme active sites comprising a few single amino acids. Whereas some in silico success has been reported previously4, a general solu- tion that can readily produce high-quality, orthogonally validated outputs remains elusive. Following fine-tuning on a task mimicking this problem (Supplementary Methods), RFdiffusion was able to scaf- fold enzyme active sites comprising many sidechain and backbone functional groups with high accuracy and in silico success rates across a range of enzyme classes (Fig. 4f and Extended Data Fig. 6a–d; in silico success required fine tuning). Although RFdiffusion is unable to explicitly model bound small molecules at present (however, see our conclusions), the substrate can be implicitly modelled using an exter- nal potential to guide the generation of ‘pockets’ around the active site. As a demonstration, we scaffold a retroaldolase active site triad while implicitly modelling the reaction substrate (Extended Data Fig. 6e–h). Symmetric functional-motif scaffolding Several important design challenges involve the scaffolding of several copies of a functional motif in symmetric arrangements. For example, many viral glycoproteins are trimeric and symmetry matched arrange- ments of inhibitory domains can be extremely potent43–46. Conversely, symmetric presentation of viral epitopes in an arrangement that mimics the virus could induce new classes of neutralizing antibodies47,48. To explore this general direction, we sought to design trimeric multiva- lent binders to the SARS-CoV-2 spike protein. In previous work, flex- ible linkage of a binder to the ACE2 binding site (on the spike protein receptor binding domain) to a trimerization domain yielded a high-affinity inhibitor that had potent and broadly neutralizing anti- viral activity in animal models43. Ideally, however, symmetric fusions to binders would be rigid, so as to reduce the entropic cost of binding while maintaining the avidity benefits from multivalency. We used RFdiffusion to design C3-symmetric trimers that rigidly hold three bind- ing domains (the functional motif in this case) such that they exactly match the ACE2 binding sites on the SARS-CoV-2 spike protein trimer. The designs were confidently predicted by AF2 to both assemble as C3-symmetric oligomers, and to scaffold the AHB2 SARS-CoV-2 binder interface with high accuracy (Fig. 5a). The ability to scaffold functional sites with any desired symmetry opens up new approaches to designing metal-coordinating protein assemblies49,50. Divalent transition metal ions show distinct prefer- ences for specific coordination geometries (for example, square planar, tetrahedral and octahedral) with ion-specific optimal sidechain–metal bond lengths. RFdiffusion provides a general route to building up sym- metric protein assemblies around such sites, with the symmetry of the assembly matching the symmetry of the coordination geometry. As a first test, we sought to design square-planar Ni2+ binding sites. We designed C4 protein assemblies with four central histidine imida- zoles arranged in an ideal Ni2+-binding site with square-planar coor- dination geometry (Fig. 5b). Diverse designs starting from distinct C4-symmetric histidine square-planar sites had good in silico success 1096 | Nature | Vol 620 | 31 August 2023 with the histidine residues in near ideal geometries for coordinating metal in the AF2-predicted structures (Supplementary Fig. 9). We expressed and purified 44 designs in Escherichia coli, and found that 37 had SEC chromatograms consistent with the intended oligo- meric state (Extended Data Fig. 7b). Of the designs, 36 were tested for Ni2+ coordination by isothermal titration calorimetry, and 18 were found to bind Ni2+ with dissociation constants ranging from low nanomolar to low micromolar (Fig. 5c,d and Extended Data Fig. 7a). The inflection points in the wild-type isotherms indicate binding with the designed stoichiometry, a one to four ratio of ion to monomer. Although most of the designed proteins showed exothermic metal coordination, in a few cases binding was endothermic (Fig. 5d, left and Extended Data Fig. 7a: NiB2.9, NiB2.10, NiB2.15 and NiB2.23), suggesting that Ni2+ coordination is entropically driven in these assemblies. To confirm that Ni2+ binding was indeed mediated by the scaffolded histidine 52, we mutated this residue to alanine, which abolished or notably reduced binding in 17 out of 17 cases with successful expression (Extended Data Figs. 7a,c and Fig. 5c,d; one mutant did not express). We structurally charac- terized by nsEM a subset of the designs—NiB1.12, NiB1.15, NiB1.17 and NiB1.20—that showed histidine-dependent binding. All four designs showed clear fourfold symmetry both in the raw micrographs and in 2D class averages (Fig. 5c,d), with design NiB1.17 also clearly showing twofold axis side views with a measured diameter approximating the design model. A 3D reconstruction of NiB1.17 was in close agreement with the design model (Fig. 5c). Design of protein-binding proteins The design of high-affinity binders to target proteins is a grand chal- lenge in protein design, with numerous therapeutic applications51. A general method for de novo binder design from target structure infor- mation alone using the physically based Rosetta method was recently described12, and subsequently, using ProteinMPNN for sequence design and AF2 for design filtering was found to improve design success rates26. However, experimental success rates were low, still requiring many thousands of designs to be screened for each design campaign12, and the approach relied on prespecifying a particular set of protein scaf- folds as the basis for the designs, inherently limiting the diversity and shape complementarity of possible solutions12. To our knowledge, no deep-learning method has yet demonstrated experimental general success in designing completely de novo binders. We reasoned that RFdiffusion might be able to address this chal- lenge by directly generating binding proteins in the context of the target. For many therapeutic applications, for example, blocking a protein–protein interaction, it is desirable to bind to a particular site on a target protein. To enable this, we fine-tuned RFdiffusion on protein complex structures, providing a feature as input indicating a subset of the residues on the target chain (called ‘interface hotspots’) to which the diffused chain binds (Fig. 6a and Extended Data Fig. 8a,b). For design challenges in which a particular binder fold might be especially compatible, we enabled coarse-grained control over binder scaffold topology by fine-tuning an extra model to condition binder diffusion on secondary structure and block-adjacency information, in addition to conditioning on interface hotspots (Extended Data Fig. 8c,d and Supplementary Methods). To compare RFdiffusion to previous binder design methods, we performed binder design campaigns against five targets: Influenza A H1 Haemagglutinin (HA)52, Interleukin-7 Receptor-α (IL-7Rα)12, Programmed Death-Ligand 1 (PD-L1)12, Insulin Receptor (InsR) and Tropomyosin Receptor Kinase A (TrkA)12. We designed putative binders to each target, both with and without conditioning on compatible fold information, with high in silico success rates (Extended Data Fig. 8e,f). Designs were filtered by AF2 confidence in the interface and mono- mer structure26, and 95 were selected for each target for experimental characterization. Article C3 axis C3 motif + noise a Spike trimer b C4 motif C4 motif + noise c s e g a r e v a l s s a c D 2 –1 –3 –5 –7 –9 ) 1 – l o m l a c k ( H Δ 3D reconstruction WT H52A KD < 20 nM 0 0.1 0.2 0.3 Molar ratio 0.4 0.5 d ) 1 – l o m l a c k ( H Δ 3 1 –1 WT H52A KD < 20 nM ) 1 – l o m l a c k ( H Δ –2 –4 –6 –8 –10 WT H52A KD < 20 nM ) 1 – l o m l a c k ( H Δ 0 –2 –4 WT H52A KD ≈ 77 nM 0 0.1 0.2 0.3 Molar ratio 0.4 0.5 0 0.1 0.2 0.3 Molar ratio 0.4 0.5 0 0.1 0.2 0.3 Molar ratio 0.4 0.5 Fig. 5 | Symmetric motif scaffolding with RFdiffusion. a, Design of symmetric oligomers scaffolding the binding interface of ACE2 mimic AHB2 (left, teal) against the SARS-CoV-2 spike trimer (left, grey). Three AHB2 copies are input to RFdiffusion along with C3 noise (middle); output are C3-symmetric oligomers holding the three AHB2 copies in place to engage all spike subunits. AF2 predictions (right) recapitulate the AHB2 structure with 0.6 Å r.m.s.d. over the assymetric unit and 2.9 Å r.m.s.d. over the C3 assembly. b, Design of C4- symmetric oligomers to scaffold a Ni2+ binding motif (left). Starting from square-planar histidine rotamers within helical fragments (Supplementary Methods), RFdiffusion generates a C4 oligomer scaffolding the binding domain (middle). AF2 predictions (colour) agree closely with the design model (grey), with backbone r.m.s.d. less than 1.0 Å (right). c, nsEM 2D class averages (scale bar, 60 Å) and 3D reconstruction density are consistent with the symmetry and structure of the NiB1.17 design model shown superimposed on the density in ribbon representation (top). Isothermal titration calorimetry binding isotherm of design NiB1.17 (blue) indicates a dissociation constant less than 20 nM at a metal:monomer stoichiometry of 1:4. The H52A mutant isotherm (pink) ablates binding, indicating scaffolded histidine residues are critical for metal binding. d, Additional experimentally characterized Ni2+ binders NiB2.15 (left), NiB1.12 (middle) and NiB1.20 (right). Metal-coordinating sidechains in the design models (top, teal) are closely recapitulated in the AF2 predictions (colours). 2D nsEM class averages (middle; scale bar, 60 Å) are consistent with design models. Binding isotherms for wild-type (WT) and H52A mutant (bottom) indicate Ni2+ binding mediated directly by the scaffolded histidines at the designed stoichiometry. Note that for ITC plots, points represent single measurements. The designed binders were expressed in E. coli and purified, and binding was assessed through single point biolayer interferometry (BLI) screening at 10 μM binder concentration (Extended Data Fig. 8g). The overall experimental success rate, defined as binding at or above 50% of the maximal response for the positive control, was 19% (this is a conservative estimate as some designs that showed binding had insufficient material to permit screening at 10 μM: Extended Data Fig. 8g); an increase of roughly two orders of magnitude over our previous Rosetta-based method on the same targets (Fig. 6b). Bind- ers were identified for all five targets, with fewer than 100 designs tested per target compared to thousands in previous studies. Full BLI titrations for a subset of the designs showed nanomolar affini- ties with no further experimental optimization, including HA and IL-7Rα binders with affinities of roughly 30 nM (Fig. 6c). Binding Nature | Vol 620 | 31 August 2023 | 1097 a b ) % ( I L B y b e t a r s s e c c u s l a t n e m i r e p x E d 40 35 30 25 20 15 10 5 0 ) m n ( s t i n u e s n o p s e R 0.5 0.4 0.3 0.2 0.1 0 Interface hotspots c IL-7Ra InsR PD-L1 TrkA ) % ( D S Y y b e t a r s s e c c u s l a t n e m i r e p x E 40 35 30 25 20 15 10 5 0 ) m n ( s t i n u e s n o p s e R 0.12 0.10 0.08 0.06 0.04 0.02 0 RFdiffusion plus AF2 filtering has orders-of-magnitude higher experimental success rates than previous methods Rosetta pipeline → RFdiffusion plus AF2 filtering ← % 4 1 . 0 % 3 4 . 0 % 7 0 . 0 HA IL-7Ra INSR PD-L1 TrkA Influenza HA e KD = 28 nM f KD = 28 nM KD1 = 80 nM KD2 = 27 nM KD = 30 nM 0.08 0.06 0.04 0.02 0 ) m n ( s t i n u e s n o p s e R 0 100 Time (s) 200 0 200 400 600 Time (s) 5,000 1,000 200 40 [Binder] (nM) 1,000 333 111 37 12 [Binder] (nM) 0.30 0.25 0.20 0.15 0.10 0.05 ) m n ( s t i n u e s n o p s e R 0 0 g 0.4 0.3 0.2 0.1 ) m n ( s t i n u e s n o p s e R 0 0 KD = 1.4 μM 100 Time (s) 200 10,000 2,000 400 80 [Binder] (nM) KD = 328 nM 100 Time (s) 200 10,000 2,000 400 80 [Binder] (nM) r.m.s.d. = 0.63 Å h r.m.s.d. = 0.60 Å 90º 90º 0 100 200 300 Time (s) [Binder (nM) 5,000 1,000 200 40 8 Fig. 6 | De novo design of protein-binding proteins. a, RFdiffusion generates protein binders given a target and specification of interface hotspot residues. b, De novo binders were designed to five protein targets; Influenza A H1 HA, IL-7Rα, InsR, PD-L1 and TrkA and hits with BLI response greater than or equal to 50% of the positive control were identified for all targets. For IL-7Rα, InsR, PD-L1 and TrkA, RFdiffusion has success rates roughly two orders of magnitude higher than the original design campaigns. We attribute one order of magnitude to RFdiffusion, and the second to filtering with AF2 (estimated success rates for previous campaigns if AF2 filtering had been used: HA, 0%; IL-7Rα, 2.2%; InsR, 5.5%; PD-L1, 3.7%; TrkA, 1.5%). c, For IL-7Rα, InsR, PD-L1 and TrkA, the highest affinity binder is shown above a BLI titration series. Reported KD values are based on global kinetic fitting with fixed global Rmax. d, The highest affinity HA binder, HA_20, binds with a KD of 28 nM. c,d, Yellow or orange, target or hotspot residues; grey, design model; purple, AF2 prediction (r.m.s.d. AF2 versus design). Binders: IL7Ra_55 (2.1 Å), InsulinR_30 (2.6 Å), PDL1_77 (1.5 Å), TrkA_88 (1.4 Å) (left to right in c) and HA_20 (1.7 Å) (d). e, Cryo-EM 2D class averages of HA_20 bound to influenza HA, strain A/USA:Iowa/1943 H1N1 (scale bar, 10 nm). f, 2.9 Å cryo-EM 3D reconstruction of the complex viewed along two orthogonal axes. HA_20 (purple) is bound to H1 along the stem of all three subunits. g, The cryo-EM structure of the HA_20 binder in complex closely matches the design model (r.m.s.d. to RFdiffusion design, 0.63 Å; yellow, influenza HA). h, Structure of the HA_20 binder alone superimposed on the design model viewed along two orthogonal axes. For cryo-EM panels, yellow, Influenza H1 map and/or structure; grey, HA_20 binder design model; purple, HA_20 binder map or structure. interfaces were often highly distinct from interfaces to these tar- gets in the PDB (Supplementary Figs. 11 and 12). To assess binder specificity, six of the highest affinity IL-7Rα binders were assessed by means of competition BLI, and all six competed for binding with a structurally validated positive control binding to the same site (Supplementary Fig. 10a; further work is required to fully characterize proteome-wide specificity). We solved the structure of the highest affinity Influenza binder, HA_20, in complex with Iowa43 HA using cryo-EM  (Extended Data Table 1). Raw electron micrographs revealed a well-folded HA 1098 | Nature | Vol 620 | 31 August 2023 Article glycoprotein with clearly discernible side, top and tilted view orienta- tions suspended in a thin layer of vitreous ice (Extended Data Fig. 9a). The 2D class averages further show clear secondary structure elements corresponding to both Iowa43 HA (Extended Data Fig. 9b), as well as the HA_20 binder bound to the stem (Fig 6e). The 3D heterogenous refinement without symmetry revealed full occupancy of all three HA stem epitopes by the HA_20 binder. A final non-uniform 3D refinement reconstruction with C3 symmetry yielded a 2.9 Å map of the HA/HA_20 protein–protein complex (Fig 6f) and corresponding 3D structure that almost perfectly matches the computational design model (0.63 Å, Fig 6f,g; the sidechain interactions at the interface are very different from the closest structure in the PDB; Extended Data Fig. 9h). Over the binder alone, the experimental structure deviates from the RFdiffusion design by only 0.6 Å (Fig. 6h). These results demonstrate the ability of RFdiffusion to generate new proteins with atomic level accuracy, and to precisely target functionally relevant sites on therapeutically important proteins. Discussion RFdiffusion is a comprehensive improvement over current protein design methods. RFdiffusion readily generates diverse uncondi- tional designs up to 600 residues in length that are accurately pre- dicted by AF2, far exceeding the complexity and accuracy achieved by most previous methods (a recent Hallucination-based approach also achieved high unconditional performance53). Half of our tested unconditional designs express in a soluble way,  and have circular dichroism spectra consistent with the design models and high ther- mostability. Despite their substantially increased complexity, the ideality and stability of RFdiffusion designs is akin to that of de novo protein designs generated using previous methods such as Rosetta. RFdiffusion enables generation of higher-order architectures with any desired symmetry, unlike Hallucination methods, which have so far been limited to cyclic symmetries. Electron microscopy confirmed that the structures of these oligomers are very similar to the design mod- els, which in many cases show little global similarity to known protein oligomers. There has been recent progress in scaffolding protein functional motifs using deep-learning methods (RF Hallucination, RFjoint Inpainting and diffusion), but Hallucination is slow for large systems, Inpainting fails when insufficient starting information is provided and previous diffusion methods had low accuracy. RFdiffusion outperforms these previous methods in the complexity of the motifs that can be scaf- folded, the precision with which sidechains are positioned (for cataly- sis and other functions), and the accuracy of motif recapitulation by AF2. The design of MDM2 binding proteins with three orders of magni- tude higher affinities than the scaffolded P53 motif demonstrates the robustness of RFdiffusion motif scaffolding. Combining accurate motif scaffolding with the design of symmetric assemblies enabled consist- ent and atomically precise positioning of sidechains to coordinate Ni2+ ions across diverse tetrameric assemblies For binder design from target structural information alone, previous work required testing tens of thousands of sequences12. RFdiffusion, when combined with improved filtering26 raises experimental success rates by two orders of magnitude; high-affinity binders can be identi- fied from dozens of designs, in many cases eliminating the require- ment for slow and expensive high-throughput screening (at least for the non-polar sites targeted here; further studies will be required to assess success rates on more polar target sites and sites without native binding partners). A high-resolution cryo-EM structure of one of these designs in complex with influenza HA shows that RFdiffusion can design functional proteins with atomic accuracy. Vázquez Torres et al. demonstrate the ability of RFdiffusion to design picomolar affin- ity binders to flexible helical peptides54, further highlighting its use for de novo binder design. Vázquez Torres et al. also show how RFdiffusion can be extended for protein model refinement by partial noising and denoising, which enables tuneable sampling around a given input structure. For peptide binder design, this enabled increases in affin- ity of nearly three orders of magnitude without high-throughput screening. The breadth and complexity of problems solvable with RFdiffusion and the robustness and accuracy of the solutions far exceeds what has been achieved previously. In a manner reminiscent of the generation of images from text prompts, RFdiffusion makes possible, with mini- mal specialist knowledge, the generation of functional proteins from minimal molecular specifications (for example, high-affinity binders to a user-specified target protein, and diverse protein assemblies from user-specified symmetries). The power and scope of RFdiffusion can be extended in several directions. RF has recently been extended to nucleic acids and protein–nucleic acid complexes55, which should enable RFdiffusion to design nucleic acid binding proteins and perhaps folded RNA struc- tures. Extension of RF to incorporate ligands should similarly enable extension of RFdiffusion to explicitly model ligand atoms, and allow the design of protein–ligand interactions. The ability to customize RFdif- fusion to specific design challenges by addition of external potentials and by fine-tuning (as illustrated here for catalytic site scaffolding, binder-targeting and fold specification), along with continued improve- ments to the underlying methodology, should enable de novo protein design to achieve still higher levels of complexity, to approach and, in some cases, surpass what natural evolution has achieved. Online content Any methods, additional references, Nature Portfolio reporting summa- ries, source data, extended data, supplementary information, acknowl- edgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-023-06415-8. 1. 2. Dauparas, J. et al. Robust deep learning-based protein sequence design using ProteinMPNN. Science 378, 49–56 (2022). Ferruz, N., Schmidt, S. & Höcker, B. ProtGPT2 is a deep unsupervised language model for protein design. Nat. Commun. 13, 4348 (2022). 3. Singer, J. M. et al. Large-scale design and refinement of stable proteins using sequence-only models. PLoS ONE 17, e0265020 (2022). 4. Wang, J. et al. Scaffolding protein functional sites using deep learning. Science 377, 5. 387–394 (2022). Trippe, B. L. et al. Diffusion probabilistic modeling of protein backbones in 3D for the motif-scaffolding problem. in The Eleventh International Conference on Learning Representations (2023). 6. Anishchenko, I. et al. De novo protein design by deep network hallucination. Nature 600, 547–552 (2021). 7. Wicky, B. I. M. et al. Hallucinating symmetric protein assemblies. Science 378, 56–61 (2022). 8. Anand, N. & Achim, T. Protein structure and sequence generation with equivariant 9. denoising diffusion probabilistic models. Preprint at https://doi.org/10.48550/arXiv.2205. 15019 (2022). Luo, S. et al. Antigen-specific antibody design and optimization with diffusion-based generative models. in Adv. Neural Information Processing Systems Vol. 35 (eds Koyejo, S. et al.) 9754–9767 (Curran Associates, 2022). 10. Sohl-Dickstein, J., Weiss, E. A., Maheswaranathan, N. & Ganguli, S. Deep unsupervised learning using nonequilibrium thermodynamics. in Proc. 32nd International Conference on Machine Learning Vol. 37 (eds Bach, Francis and Blei, David) 2256–2265 (PMLR, 2015). 11. Ho, J., Jain, A. & Abbeel, P. Denoising diffusion probabilistic models. in Adv. Neural Information Processing Systems Vol. 33 (eds Larochelle, H. et al.) 6840–6851 (Curran Associates, 2020). 12. Cao, L. et al. Design of protein-binding proteins from the target structure alone. Nature 605, 551–560 (2022). 13. Kuhlman, B. et al. Design of a novel globular protein fold with atomic-level accuracy. Science 302, 1364–1368 (2003). 14. Ramesh, A. et al. Zero-shot text-to-image generation. in Proc. 38th International Conference on Machine Learning Vol. 139 (eds Meila, M. & Zhang, T.) 8821–8831 (PMLR, 2021). 15. Saharia, C. et al. Photorealistic text-to-image diffusion models with deep language understanding. in Adv. Neural Information Processing Systems Vol. 35 (eds Koyejo, S. et al.) 36479–36494 (Curran Associates, 2022). 16. Wu, K. E. et al. Protein structure generation via folding diffusion. Preprint at https://doi. 17. org/10.48550/arXiv.2209.15611 (2022). Jumper, J. et al. Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589 (2021). Nature | Vol 620 | 31 August 2023 | 1099 18. Baek, M. et al. Accurate prediction of protein structures and interactions using a three-track 41. Chène, P. Inhibiting the p53-MDM2 interaction: an important target for cancer therapy. neural network. Science 373, 871–876 (2021). Nat. Rev. Cancer 3, 102–109 (2003). 19. Watson, J. L., Bera, A., Juergens, D., Wang, J. & Baker, D. X-ray crystallographic validation 42. Kussie, P. H. et al. Structure of the MDM2 oncoprotein bound to the p53 tumor suppressor of design from this paper. Science 377, 387–394 (2022). transactivation domain. Science 274, 948–953 (1996). 20. Wu, R. et al. High-resolution de novo structure prediction from primary sequence. 43. Hunt, A. C. et al. Multivalent designed proteins neutralize SARS-CoV-2 variants of Preprint at https://doi.org/10.1101/2022.07.21.500999 (2022). 21. Lin, Z. et al. Language models of protein sequences at the scale of evolution enable concern and confer protection against infection in mice. Sci. Transl. Med. 14, eabn1252 (2022). accurate structure prediction. Science 379, 1123–1130 (2023). 44. Silverman, J. et al. Multivalent avimer proteins evolved by exon shuffling of a family of 22. Berman, H. M. et al. The Protein Data Bank. Nucleic Acids Res. 28, 235–242 (2000). 23. De Bortoli, V. et al. Riemannian score-based generative modelling. in Adv. Neural Information Processing Systems Vol. 35 (eds Koyejo, S. et al.) 2406–2422 (Curran Associates, 2022). 24. Leach, A., Schmon, S. M., Degiacomi, M. T. & Willcocks, C. G. Denoising diffusion probabilistic models on SO(3) for rotational alignment. In Proc. ICLR 2022 Workshop on Geometrical and Topological Representation Learning (2022). 25. Chen, T., Zhang, R. & Hinton, G. Analog bits: generating discrete data using diffusion models with self-conditioning. in The Eleventh International Conference on Learning Representations (2023). 26. Bennett, N.R. et al. Improving de novo protein binder design with deep learning. Nat. Commun. 14, 2625 (2023). 27. Anand, N. & Huang, P. Generative modeling for protein structures. in Adv. Neural Information Processing Systems Vol. 31 (eds Bengio, S. et al.) (Curran Associates, 2018). Ingraham, J. et al. Illuminating protein space with a programmable generative model. Preprint at bioRxiv https://doi.org/10.1101/2022.12.01.518682 (2022). 28. human receptor domains. Nat. Biotechnol. 23, 1556–1561 (2005). 45. Detalle, L. et al. Generation and characterization of ALX-0171, a potent novel therapeutic nanobody for the treatment of respiratory syncytial virus infection. Antimicrob. Agents Chemother. 60, 6–13 (2016). 46. Strauch, E.-M. et al. Computational design of trimeric influenza-neutralizing proteins targeting the hemagglutinin receptor binding site. Nat. Biotechnol. 35, 667–671 (2017). 47. Boyoglu-Barnum, S. et al. Quadrivalent influenza nanoparticle vaccines induce broad protection. Nature 592, 623–628 (2021). 48. Walls, A. C. et al. Elicitation of potent neutralizing antibody responses by designed protein nanoparticle vaccines for SARS-CoV-2. Cell 183, 1367–1382.e17 (2020). 49. Salgado, E. N., Lewis, R. A., Mossin, S., Rheingold, A. L. & Tezcan, F. A. Control of protein oligomerization symmetry by metal coordination: C2 and C3 symmetrical assemblies through CuII and NiII coordination. Inorg. Chem. 48, 2726–2728 (2009). 50. Salgado, E. N. et al. Metal templated design of protein interfaces. Proc. Natl Acad. Sci. USA 107, 1827–1832 (2010). 51. Quijano-Rubio, A., Ulge, U. Y., Walkey, C. D. & Silva, D.-A. The advent of de novo proteins 29. Lee, J. S. & Kim, P. M. ProteinSGM: Score-based generative modeling for de novo protein for cancer immunotherapy. Curr. Opin. Chem. Biol. 56, 119–128 (2020). design. Preprint at bioRxiv https://doi.org/10.1101/2022.07.13.499967 (2022). 52. Chevalier, A. et al. Massively parallel de novo protein design for targeted therapeutics. 30. Onuchic, J. N., Luthey-Schulten, Z. & Wolynes, P. G. Theory of protein folding: the energy Nature 550, 74–79 (2017). 31. landscape perspective. Annu. Rev. Phys. Chem. 48, 545–600 (1997). Jendrusch, M., Korbel, J. O. & Sadiq, S. K. AlphaDesign: a de novo protein design framework based on AlphaFold. Preprint at bioRxiv https://doi.org/10.1101/2021.10.11. 463937 (2021). 32. Basanta, B. et al. An enumerative algorithm for de novo design of proteins with diverse pocket structures. Proc. Natl Acad. Sci. USA 117, 22135–22145 (2020). 33. Pan, X. et al. Expanding the space of protein geometries by computational design of 53. Frank, C. et al. Efficient and scalable de novo protein design using a relaxed sequence space. Preprint at bioRxiv https://doi.org/10.1101/2023.02.24.529906 (2023). 54. Torres, S. V. et al. De novo design of high-affinity protein binders to bioactive helical peptides. Preprint at bioRxiv https://doi.org/10.1101/2022.12.10.519862 (2022). 55. Baek, M., McHugh, R., Anishchenko, I., Baker, D. & DiMaio, F. Accurate prediction of nucleic acid and protein-nucleic acid complexes using RoseTTAFoldNA. Preprint at bioRxiv https://doi.org/10.1101/2022.09.09.507333 (2022). de novo fold families. Science 369, 1132–1136 (2020). 34. Marcandalli, J. et al. Induction of potent neutralizing antibody responses by a designed protein nanoparticle vaccine for respiratory syncytial virus. Cell 176, 1420–1431.e17 (2019). 35. Butterfield, G. L. et al. Evolution of a designed protein assembly encapsulating its own RNA genome. Nature 552, 415–420 (2017). 36. Goodsell, D. S. & Olson, A. J. Structural symmetry and protein function. Annu. Rev. Biophys. Biomol. Struct. 29, 105–153 (2000). 37. Sterner, R. & Höcker, B. Catalytic versatility, stability, and evolution of the (βα)8-barrel enzyme fold. Chem. Rev. 105, 4038–4055 (2005). 38. Sesterhenn, F. et al. De novo protein design enables the precise induction of RSV-neutralizing antibodies. Science 368, eaay5051 (2020). 39. Yang, C. et al. Bottom-up de novo design of functional proteins with complex structural features. Nat. Chem. Biol. 17, 492–500 (2021). Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. 40. Glasgow, A. et al. Engineered ACE2 receptor traps potently neutralize SARS-CoV-2. Proc. Natl Acad. Sci. USA 117, 28046–28055 (2020). © The Author(s) 2023 1100 | Nature | Vol 620 | 31 August 2023 Article Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Data availability Design structures, AF2 models and experimental measurements are available at https://figshare.com/s/439fdd59488215753bc3. Cryo-EM maps and corresponding atomic models for the Influenza HA binder in Fig. 6d–h have been deposited in the PDB and the Electron Microscopy Data Bank under accession codes 8SK7 and EMDB-40557, respectively. Electron microscopy data collected for the HE0537 oligomer are avail- able at EMDB-40602. Code availability Code for running RFdiffusion has been released on GitHub, free for academic, personal and commercial use at https://github.com/Rosetta- Commons/RFdiffusion. It is also available as a Google Colab notebook, accessible through GitHub. 56. Yeh, A. H.-W. et al. De novo design of luciferases using deep learning. Nature 614, 774–780 (2023). 57. Ribeiro, A. J. M. et al. Mechanism and Catalytic Site Atlas (M-CSA): a database of enzyme reaction mechanisms and active sites. Nucleic Acids Res. 46, D618–D623 (2018). 58. Leaver-Fay, A. et al. ROSETTA3: an object-oriented software suite for the simulation and design of macromolecules. Methods Enzymol. 487, 545–574 (2011). Design (B.L.T., I.S., J.Y., H.E. and D.B.), the Washington State General Operating Fund supporting the Institute for Protein Design (P.V. and I.S.), grant no. INV-010680 from the Bill and Melinda Gates Foundation (W.B.A., D.J., J.W. and D.B.), grant no. DE-SC0018940 MOD03 from the US Department of Energy Office of Science (A.J.B. and D.B.), grant no. 5U19AG065156-02 from the National Institute for Aging (S.V.T. and D.B.), an EMBO long-term fellowship no. ALTF 139-2018 (B.I.M.W.), the Open Philanthropy Project Improving Protein Design Fund (R.J.R. and D.B.), The Donald and Jo Anne Petersen Endowment for Accelerating Advancements in Alzheimer’s Disease Research (N.R.B.), a Washington Research Foundation Fellowship (S.J.P.), a Human Frontier Science Program Cross Disciplinary Fellowship (grant no. LT000395/2020-C, L.F.M.), an EMBO Non-Stipendiary Fellowship (grant no. ALTF 1047-2019, L.F.M.), the Defense Threat Reduction Agency grant nos. HDTRA1-19-1-0003 (N.H. and D.B.) and HDTRA12210012 (F.D.), the Institute for Protein Design Breakthrough Fund (A.C. and D.B.), an EMBO Postdoctoral Fellowship (grant no. ALTF 292-2022, J.L.W.) and the Howard Hughes Medical Institute (A.C., W.S., R.J.R. and D.B.), an NSF-GRFP (J.Y.), an NSF Expeditions grant (no. 1918839, J.Y., R.B. and T.S.J.), the Machine Learning for Pharmaceutical Discovery and Synthesis consortium (J.Y., R.B. and T.S.J.), the Abdul Latif Jameel Clinic for Machine Learning in Health (J.Y., R.B. and T.S.J.), the DTRA Discovery of Medical Countermeasures Against New and Emerging threats program (J.Y., R.B. and T.S.J.), EPSRC Prosperity Partnership grant no. EP/T005386/1 (E.M.) and the DARPA Accelerated Molecular Discovery program and the Sanofi Computational Antibody Design grant (J.Y., R.B. and T.S.J.). We thank Microsoft and AWS for generous gifts of cloud computing resources. Author contributions J.L.W., D.J., N.R.B., B.L.T., J.Y. and D.B. conceived the study. J.L.W., D.J., N.R.B., W.A., B.L.T. and J.Y. trained RFdiffusion. B.L.T. and J.Y., with assistance from V.D.B. and E.M., extended diffusion to residue orientations. H.E.E., D.J., J.L.W., N.R.B., N.H., W.S., P.V. and I.S. generated experimentally characterized designs. W.A., B.L.T., J.Y., D.J., J.L.W. and N.R.B. generated computational designs. H.E.E., A.J.B., R.J.R., L.F.M., B.I.M.W., S.J.P., N.H., A.C., S.V.T., J.L.W. and B.L.T. experimentally characterized designs. J.W., A.L. and W.S. contributed additional code. S.O. implemented RFdiffusion on Google Colab. M.B. and F.D. trained RF. D.B., T.S.J. and R.B. offered supervision throughout the project. J.L.W., D.J., B.L.T., N.R.B., J.Y., H.E. and D.B. wrote the manuscript. All authors read and contributed to the manuscript. J.L.W. and D.J. agree that the order of their respective names may be changed for personal pursuits to best suit their own interests. Competing interests The authors declare no competing interests. Acknowledgements We thank N. Anand and D. Tischer for helpful discussions, and I. Kalvet and Y. Kipnis for providing helpful Rosetta scripts. We thank A. Dosey for the provision of purified influenza HA protein. We thank R. Wu, J. Mou, K. Choi, L. Wu and D. Blei for valuable feedback during writing. We thank I. Haydon for help with graphics. We also thank L. Goldschmidt and K. VanWormer, respectively, for maintaining the computational and wet laboratory resources at the Institute for Protein Design. This work was supported by gifts from Microsoft (D.J., M.B. and D.B.), Amgen (J.L.W.), the Audacious Project at the Institute for Protein Additional information Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41586-023-06415-8. Correspondence and requests for materials should be addressed to David Baker. Peer review information Nature thanks Arne Elofsson, Giulia Palermo, Alex Pritzel and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Reprints and permissions information is available at http://www.nature.com/reprints. Extended Data Fig. 1 | See next page for caption. Article Extended Data Fig. 1 | Training ablations reveal determinants of RFdiffusion success. A–C) RFdiffusion can generate high quality large unconditional monomers. Designs are routinely accurately recapitulated by AF2 (see also Fig. 2c), with high confidence (A) for proteins up to approximately 400 amino acids in length. B) Further orthogonal validation of designs by ESMFold. C) Recapitulation of the design structure is often better with ESMFold compared with AF2. For each backbone, the best of 8 ProteinMPNN sequences is plotted, with points therefore paired by backbone rather than sequence. D) Comparing RFdiffusion trained with MSE loss on Cα atoms and N-Cα-C backbone frames (Methods 2.5), rather than with FAPE loss8,17. The MSE loss is not invariant to the global coordinate frame, unlike FAPE loss, and is required for good performance at unconditional generation (left, two-proportion z-test of in silico success rate, n = 400 designs per condition, z = 4.1, p = 4.1e-5). For motif scaffolding problems, where the ‘motif’ provides a means to align the global coordinate frame between timesteps, FAPE loss performs approximately as well as MSE loss, suggesting the L2 nature of MSE loss (as opposed to the L1 loss in FAPE) is not empirically critical for performance. E) Allowing the model to condition on its X0 prediction at the previous timestep (see Supplementary Methods 2.4) improves designs. Designs with self-conditioning (pink) have improved recapitulation by AF2 (left) and better AF2 confidence in the prediction (right). Two-proportion z-test of in silico success rate, n = 800 designs per condition z = 11.4, p = 6.1e-30. F) RFdiffusion leverages the representations learned during RF pre-training. RFdiffusion fine-tuned from pre-trained RF (pink) comprehensively outperforms a model trained for an equivalent amount of time, from untrained weights (gray). For context, sequences generated by ProteinMPNN on these output backbones are little better than sampling ProteinMPNN sequences from random Gaussian-sampled coordinates (white). Two-proportion z-test of in silico success rate, pre-training vs without pre-training (or vs random noise; both have zero success rate), n = 800 designs per condition, z = 23.0, p = 3.1e-117. Note that the data in pink in D–F is the same data, reproduced in each plot for clarity. G) The median (by AF2 r.m.s.d. vs design) 300 amino acid unconditional sample highlighting the importance of self-conditioning and pre-training. Without pre-training (at least when trained with equivalent compute), RFdiffusion outputs bear little resemblance to proteins (gray, left). Without self-conditioning, outputs show characteristic protein secondary structures, but lack core-packing and ideality (gray, middle). With pre-training and self-conditioning, proteins are diverse and well-packed (pink, right). H) Greater coherence during unconditional denoising may partly explain the effect of self-conditioning. Successive X0 predictions are more similar when the model can self-condition (lower r.m.s.d. between X0 predictions, pink curve). Data are aggregated from unconditional design trajectories of 100, 200 and 300 residues. I) During the reverse (generation) process, the noise added at each step can be scaled (reduced). Reducing the noise scale improves the in silico design success rates (left, middle; two- proportion z-test of in silico success rate, n = 800 designs per condition, 0 vs 0.5: z = 1.7, p = 0.09, 0 vs 1: z = 6.5, p = 6.8e-11; 0.5 vs 1: z = 4.8, p = 1.4e-6). This comes at the expense of diversity, with the number of unique clusters at a TM-score cutoff of 0.6 reduced when noise is reduced (right). Note throughout this figure the 6EXZ_long benchmarking problem is abbreviated to 6EXZ for brevity. Boxplots represent median±IQR; tails: min/max excluding outliers (±1.5xIQR). Extended Data Fig. 2 | RFdiffusion learns the distribution of the denoising process, and inference efficiency can be improved. A) Analysis of simulated forward (noising) and reverse (denoising) trajectories shows that the distribution of Cα coordinates and residue orientations closely match, demonstrating that RFdiffusion has learned the distribution of the denoising process as desired. Left to right: i) average distance between a Cα coordinate at Xt and its position in X0; ii) average distance between a Cα coordinate at Xt and Xt-1; iii) average distance between adjacent Cα coordinates at Xt; iv) average rotation distance between a residue orientation at Xt and X0; v) average rotation distance between a residue orientation at Xt and Xt-1. B-C) While RFdiffusion is trained to generate samples over 200 timesteps, in many cases, trajectories can be shortened to improve computational efficiency. B) Larger steps can be taken between timesteps at inference. Decreasing the number of timesteps speeds up inference, and often does not decrease in silico success rates (left) (for example, on an NVIDIA A4000 GPU, 100 amino acid designs can be generated with 15 steps, in ~11s, with an in silico success rate of over 60%). When normalized for compute budget (center) it is often much more efficient to run more trajectories with fewer timesteps. This can be done without loss of diversity in samples (right). For harder problems (e.g. unconditional 300 amino acids), one must strike an intermediate number of total timesteps (e.g., T = 50) for optimal compute efficiency. Note that for all other analyses in the paper, 200 inference steps were used, in line with how RFdiffusion is trained. C) An alternative to taking larger steps is to stop trajectories early (possible because RFdiffusion predicts X0 at every timestep). In many cases, trajectories can be stopped at timestep 50–75 with little effect on the final in silico success rate of designs (left), and when normalized by compute budget (center), success rates per unit time are typically higher generating more designs with early-stopping. Again, this can be done without a significant loss in diversity (right). Article Extended Data Fig. 3 | Unconditionally-generated designs are folded and thermostable. A) Four 200 amino acid and fourteen 300 amino acid proteins were tested for expression and stability. 9/18 designs expressed, with a major peak at the expected elution volume. Blue: 300 amino acid proteins; Purple: 200 amino acid proteins. B) Colored AF2 predictions overlaid on gray design models (left), circular dichroism spectra at 25 °C (blue) and 95 °C (pink) (middle) and circular dichroism melt curves (right) for all 9 designs passing expression thresholds. In all cases, proteins remain well folded even at 95 °C. Note that data on 300aa_3 and 300aa_8 are duplicated from Fig. 2f, reproduced here for clarity. Extended Data Fig. 4 | See next page for caption. Article Extended Data Fig. 4 | RFdiffusion can condition on fold information to generate specific, thermostable folds. A) 6WVS is a previously-described de novo designed TIM barrel (left). A fine-tuned RFdiffusion model can condition on 1D and 2D inputs representing this protein fold, specifically secondary structure (middle, bottom) and block-adjacency information (middle, top) (see Supplementary Methods 4.3.2). RFdiffusion then generates proteins that closely recapitulate this course-grained fold information (right). B) Outputs are diverse with respect to each other. With this coarse-grained fold specification, in silico successful designs are much more diverse (as quantified by pairwise TM-scores) compared to diversity generated through simply sampling many sequences for the original PDB backbone (6WVS). C) NTF2 folds are useful scaffolds for de novo enzyme design56, and can also be readily generated with fold-conditioning in RFdiffusion. Designs are diverse and closely recapitulated by AF2. D) In silico success rates are high with fold- conditioned diffusion. TIM barrels are generated with an AF2 in silico success rate of 42.5% (left bar, pink) with in silico success incorporating both AF2 metrics and a TM-score vs 6WVS > 0.5. NTF2 folds are generated with an AF2 in silico success rate of 54.1% (right bar, pink), with in silico success incorporating both AF2 metrics and a TM-score vs PDB: 1GY6 > 0.5. In silico success was further validated with ESMFold (blue bars), where a pLDDT > 80 was used as the confidence metric for success. Gray: RFdiffusion design, colors: AF2 prediction. E) 11 TIM barrel designs were purified alongside the 6WVS positive control. Ten of these express and elute predominantly as monomers (note that the designs are approximately 4kDa larger than 6WVS). F) Eight designs expressed sufficiently for analysis by circular dichroism. All designs are folded, with circular dichroism spectra consistent with the designed structure (middle), and similar to 6WVS. Designs were also all highly thermostable, with CD melt analyses demonstrating designs were folded even at 95 °C (right). Designs are shown in gray, with the AF2 predictions overlaid in colors (left). Note that data on 6WVS and TIM_barrel_6 are duplicated from Fig. 2g, reproduced here for clarity. Extended Data Fig. 5 | See next page for caption. Article Extended Data Fig. 5 | Symmetric oligomer design with RFdiffusion. A) Due to the (near-perfect - see Supplementary Methods 3.1) equivariance properties of RFdiffusion, X0 predictions from symmetric inputs are also symmetric, even at very early timepoints (and becoming increasingly symmetric through time; r.m.s.d. vs symmetrized: t = 200 1.20 Å; t = 150 0.40 Å; t = 50 0.06 Å; t = 0 0.02Å). Gray: symmetrized (top left) subunit; colors: RFdiffusion X0 prediction. B) In silico success rates for symmetric oligomer designs of various cyclic and dihedral symmetries. In silico success is defined here as the proportion of designs for which AF2 yields a prediction from a single sequence that has mean pLDDT > 80 and backbone r.m.s.d. over the oligomer between the design model and AF2 < 2Å. Note that 16 sequences per RFdiffusion design were sampled. C) Box plots of the distribution of backbone r.m.s.d.s between AF2 and the RFdiffusion design model with and without the use of external potentials during the trajectory. The external potentials used are the ‘inter-chain’ contact potential (pushing chains together), as well as the ‘intra-chain’ contact potential (making chains more globular). Using these potentials dramatically improves in silico success (Two-proportion z-test of in silico success rate: n = 100 designs per condition, z = 4.3, p = 1.9e-5). D) Designs are diverse with respect to the training dataset (the PDB). While the monomers (typically 60–100 AA) show reasonable alignment to the PDB (median 0.72), the whole oligomeric assemblies showed little resemblance to the PDB (median 0.50). E) Additional examples of design models (left) against AF2 predictions (right) for C3, C5, C12, and D4 symmetric designs (the symmetries not displayed in Fig. 3) with backbone r.m.s.d.s (Å) against their AF2 predictions of 0.82, 0.63, 0.79, and 0.78 with total amino acids 750, 900, 960, 640. F) Additional nsEM data for symmetric designs. The model is shown on the left and the 2D class averages on the right for each design. G) Two orthogonal side views of HE0537 by cryo-EM. Representative 2D class averages from the cryo-EM data are shown to the right of 2D projection images of the computational design model (lowpass filtered to 8 Å), which appear nearly identical to the experimental data. Scale bars shown (white) are 60 Å. Boxplot represents median ± IQR; tails: min/max excluding outliers (±1.5xIQR). Extended Data Fig. 6 | See next page for caption. Article Extended Data Fig. 6 | External potentials for generating pockets around substrate molecules. A–D) Example in silico successful designs for enzyme classes 2–5 (ref. 57, see also Fig. 4). Native enzyme (PDB: 1CWY, 1DE3, 1P1X, 1SNZ); catalytic site (teal); RFdiffusion output (gray: model, colors: AF2 prediction). Metrics (AF2 vs design backbone r.m.s.d., AF2 vs design motif backbone r.m.s.d., AF2 vs design motif full-atom r.m.s.d., AF2 pAE): EC2: 0.93 Å, 0.50 Å, 1.29 Å, 3.51; EC3: 0.92 Å, 0.60 Å, 1.07 Å, 4.59; EC4: 0.93 Å, 0.80 Å, 1.03 Å, 4.41; EC5: 0.78 Å, 0.44 Å, 1.14 Å, 3.32. E–H) Implicit modeling of a substrate while scaffolding a retroaldolase active site triad [TYR1051-LYS1083-TYR1180] from PDB: 5AN7. E) The potential used to implicitly model the substrate, which has both a repulsive and attractive field (see Supplementary Methods 4.4). F) Left: Kernel densities demonstrate that without using the external potential (pink), designs often fall into two failure modes: (1) no pocket, and (2) clashes with the substrate. Right: clashes (substrate < 3 Å of the backbone) & pockets (no clash and > 16 Cα within 3–8 Å of substrate) with and without the potential. Two- proportion z-test: n = 71/51 +/− potential; clashes z = −2.05, p = 0.02, pocket z = −2.27, p = 0.01. Each datapoint represents a design already passing the stringent in silico success metrics (AF2 motif r.m.s.d. < 1 Å, AF2 backbone r.m.s.d. < 2 Å, AF2 pAE < 5). Note that the potential and clash definition pertain only to backbone Cα atoms, and do not currently include sidechain atoms. G) Designs close to the labeled local maxima of the kernel density estimate. Without the potential, the catalytic triad is predominantly (1) exposed on the surface with no residues available to provide substrate stabilization or (2) buried in the protein core, preventing substrate access. With the potential, the catalytic triad is predominantly (3), partially buried in a concave pocket with shape complementary to the substrate. Backbone atoms within 3 Å of the substrate are shown in red. H) A variety of diverse designs with pockets made using the potential, with no clashes between the substrate and the AF2- predicted backbone. The functional form and parameters used for the pocket potential are detailed in Supplementary Methods 4.4. In each case the substrate is superimposed on the AF2 prediction of the catalytic triad. Extended Data Fig. 7 | See next page for caption. Article Extended Data Fig. 7 | Additional Ni2+ binding C4 oligomers. A) AF2 predictions of a subset of the experimentally verified Ni2+ binding oligomers, with corresponding isothermal titration calorimetry (ITC) binding isotherms for the wild-type (blue) and H52A mutant (pink) below. Note that these, with Fig. 5, encompass all of the experimentally validated outputs deriving from unique RFdiffusion backbones. Wild-type dissociation constants are displayed in each plot. We observe a mixture of endothermic (NiB2.10, NiB2.23, NiB2.15) and exothermic isotherms. For all cases displayed we observe no binding to the ion for H52A mutants, indicating the scaffolded histidine at position 52 is critical for ion binding. KD values in the isotherms indicate binding of the ion with the designed stoichiometry (1:4 Ni2+:protein). Note that each backbone depicted is from a unique RFdiffusion sampling trajectory, and that models and data for designs NiB2.15, NiB1.12, NiB1.20 and NiB1.17 from Fig. 5 are duplicated here for ease of viewing. B) Size exclusion chromatograms for elutions from the 44 purifications suggest the vast majority of designs are soluble and have the correct oligomeric state. C) Size exclusion chromatograms for 20 H52A mutants show that the mutants remain soluble and retain the intended oligomeric state. Note that only 18 of these 20 had wild-type sequences that definitively bound nickel. Note also that for ITC plots, points represent single measurements. Extended Data Fig. 8 | See next page for caption. Article Extended Data Fig. 8 | Targeted unconditional and fold-conditioned protein binder design. A-B) The ability to specify where on a target a designed binder should bind is crucial. Specific “hotspot” residues can be input to a fine-tuned RFdiffusion model, and with these inputs, binders almost universally target the correct site. A) IL-7Rα (PDB: 3DI3) has two patches that are optimal for binding, denoted Site 1 and Site 2 here. For each site, 100 designs were generated (without fold-specification). B) Without guidance, designs typically target Site 1 (left bar, gray), with contact defined as Cα-Cα distance between binder and hotspot reside < 10 Å. Specifying Site 1 hotspot residues increases further the efficiency with which Site 1 is targeted (left bar, pink). In contrast, specifying the Site 2 hotspot residues can completely redirect RFdiffusion, allowing it to efficiently target this site (right bar, pink). C-D) As well as conditioning on hotspot residue information, a fine-tuned RFdiffusion model can also condition on input fold information (secondary structure and block-adjacency information - see Supplementary Methods 4.5). This effectively allows the specification of a (for instance, particularly compatible) fold that the binder should adopt. C) Two examples showing binders can be specified to adopt either a ferredoxin fold (left) or a particular helical bundle fold (right). D) Quantification of the efficiency of fold-conditioning. Secondary structure inputs were accurately respected (top, pink). Note that in this design target and target site, RFdiffusion without fold-specification made generally helical designs (right, gray bar). Block-adjacency inputs were also respected for both input folds (bottom, pink). E) Reducing the noise added at each step of inference improves the quality of binders designed with RFdiffusion, both with and without fold-conditioning. As an example, the distribution of AF2 interaction pAEs (known to indicate binding when pAE < 1026) is shown for binders designed to PD-L1. In both cases, the proportion of designs with interaction pAE < 10 is high (blue curve), and improved when the noise is scaled by a factor 0.5 (pink curve) or 0 (yellow curve). F) Full in silico success rates for the protein binders designed to five targets. In each case, the best fold- conditioned results are shown (i.e. from the most target-compatible input fold), and the success rates at each noise scale are separated. In line with current best practice26, we tested using Rosetta FastRelax58 before designing the sequence with ProteinMPNN, but found that this did not systematically improve designs. In silico success is defined in line with current best practice26: AF2 pLDDT of the monomer > 80, AF2 interaction pAE < 10, AF2 r.m.s.d. monomer vs design < 1 Å. G) Experimentally-validated de novo protein binders were identified for all five of the targets. Designs that bound at 10 μM during single point BLI screening with a response equal to or greater than 50% of the positive control were considered binders. Concentration is denoted by hue for designs that were screened at concentrations less than 10 μM and thus may be false negatives. Extended Data Fig. 9 | See next page for caption. Article Extended Data Fig. 9 | Cryo-electron microscopy structure determination of designed Influenza HA binder. A) Representative raw micrograph showing ideal particle distribution and contrast. B) 2D Class averages of Influenza H1+HA_20 binder with clearly defined secondary structure elements and a full- sampling of particle view angles (scale bar = 10 nm). C) Cryo-EM local resolution map calculated using an FSC value of 0.143 viewed along two different angles. Local resolution estimates range from ~2.3 Å at the core of H1 to ~3.4 Å along the periphery of the N-terminal helix of the HA_20 binder. D) Cryo-EM structure of the full H1+HA_20 binder complex (purple: HA_20; yellow: H1; teal: glycans). E) Global resolution estimation plot. F) Orientational distribution plot demonstrating complete angular sampling. G) 3D ab initio (left) and 3D heterogenous refinement (right - unsharpened) outputs, performed in the absence of applied symmetry, and showing clear density of the HA_20 binder bound to all three stem epitopes of the Iowa43 HA glycoprotein trimer, in all maps. H) The designed binder has topological similarity to 5VLI, a protein in the PDB, but binds with very different interface contacts. Extended Data Table 1 | Cryo-EM data collection, refinement and validation statistics Article Corresponding author(s): David Baker Last updated by author(s): June 22nd, 2023 Reporting Summary Nature Portfolio wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency in reporting. For further information on Nature Portfolio policies, see our Editorial Policies and the Editorial Policy Checklist. Statistics For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section. n/a Confirmed The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly The statistical test(s) used AND whether they are one- or two-sided Only common tests should be described solely by name; describe more complex techniques in the Methods section. A description of all covariates tested A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient) AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals) For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted Give P values as exact values whenever suitable. For Bayesian analysis, information on the choice of priors and Markov chain Monte Carlo settings For hierarchical and complex designs, identification of the appropriate level for tests and full reporting of outcomes Estimates of effect sizes (e.g. Cohen's d, Pearson's r), indicating how they were calculated Our web collection on statistics for biologists contains articles on many of the points above. Software and code Policy information about availability of computer code Data collection RFdiffusion 1.0.0 (this study), ProteinMPNN, AlphaFold2, TMalign, Protein-Protein BLAST 2.11.0+, SerialEM Data analysis Matplotlib 3.6.2, ScIPy 1.9.3, Seaborn 0.11.2, PyMOL 2.5.0, ForteBio Data Analysis Software Version 9.0.0.14, pycorn 0.19, CryoSparc v4.0.3, Microcal PEAQ-ITC Analysis Software For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors and reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Portfolio guidelines for submitting code & software for further information. Data Policy information about availability of data All manuscripts must include a data availability statement. This statement should provide the following information, where applicable: - Accession codes, unique identifiers, or web links for publicly available datasets - A description of any restrictions on data availability - For clinical datasets or third party data, please ensure that the statement adheres to our policy Design structures, AlphaFold2 models and experimental measurements are available at https://figshare.com/s/439fdd59488215753bc3. Cryo-EM maps and corresponding atomic models for the Influenza HA binder in Figure 6D-H have been deposited in the PDB and the Electron Microscopy Data Bank under accession n a t u r e p o r t f o l i o | r e p o r t i n g s u m m a r y A p r i l 2 0 2 3 1 codes 8SK7 and EMDB-40557, respectively. Electron microscopy data collected for the HE0537 oligomer is available at EMDB-40602. Cryo-EM data collection, refinement and validation statistics are supplied in Extended Data Table 1. Research involving human participants, their data, or biological material Policy information about studies with human participants or human data. See also policy information about sex, gender (identity/presentation), and sexual orientation and race, ethnicity and racism. Reporting on sex and gender N/A Reporting on race, ethnicity, or other socially relevant groupings Population characteristics Recruitment Ethics oversight N/A N/A N/A N/A Note that full information on the approval of the study protocol must also be provided in the manuscript. Field-specific reporting Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection. Life sciences Behavioural & social sciences Ecological, evolutionary & environmental sciences For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf Life sciences study design All studies must disclose on these points even when the disclosure is negative. Sample size Variable depending on analysis performed. Detailed in figure legends. Sample sizes were chosen prior to the experiment, and were decided arbitrarily by the experimenter (rather than by statistical test), but were large enough to draw meaningful conclusions from the experiment. Data exclusions None Replication Each dataset contains many (n reported in figure legends) independent measurements. Randomization N/A (all analysis was automated, so each datapoint was generated computationally under controlled and uniform settings) Blinding N/A (all analysis was automated, so there was no user intervention that could have introduced bias) Behavioural & social sciences study design All studies must disclose on these points even when the disclosure is negative. Study description Briefly describe the study type including whether data are quantitative, qualitative, or mixed-methods (e.g. qualitative cross-sectional, quantitative experimental, mixed-methods case study). Research sample Sampling strategy State the research sample (e.g. Harvard university undergraduates, villagers in rural India) and provide relevant demographic information (e.g. age, sex) and indicate whether the sample is representative. Provide a rationale for the study sample chosen. For studies involving existing datasets, please describe the dataset and source. Describe the sampling procedure (e.g. random, snowball, stratified, convenience). Describe the statistical methods that were used to predetermine sample size OR if no sample-size calculation was performed, describe how sample sizes were chosen and provide a rationale for why these sample sizes are sufficient. For qualitative data, please indicate whether data saturation was considered, and what criteria were used to decide that no further sampling was needed. Data collection Provide details about the data collection procedure, including the instruments or devices used to record the data (e.g. pen and paper, computer, eye tracker, video or audio equipment) whether anyone was present besides the participant(s) and the researcher, and whether the researcher was blind to experimental condition and/or the study hypothesis during data collection. Timing Indicate the start and stop dates of data collection. If there is a gap between collection periods, state the dates for each sample cohort. n a t u r e p o r t f o l i o | r e p o r t i n g s u m m a r y A p r i l 2 0 2 3 2 Data exclusions If no data were excluded from the analyses, state so OR if data were excluded, provide the exact number of exclusions and the rationale behind them, indicating whether exclusion criteria were pre-established. Non-participation State how many participants dropped out/declined participation and the reason(s) given OR provide response rate OR state that no participants dropped out/declined participation. Randomization If participants were not allocated into experimental groups, state so OR describe how participants were allocated to groups, and if allocation was not random, describe how covariates were controlled. Ecological, evolutionary & environmental sciences study design All studies must disclose on these points even when the disclosure is negative. Study description Briefly describe the study. For quantitative data include treatment factors and interactions, design structure (e.g. factorial, nested, hierarchical), nature and number of experimental units and replicates. Research sample Describe the research sample (e.g. a group of tagged Passer domesticus, all Stenocereus thurberi within Organ Pipe Cactus National Monument), and provide a rationale for the sample choice. When relevant, describe the organism taxa, source, sex, age range and any manipulations. State what population the sample is meant to represent when applicable. For studies involving existing datasets, describe the data and its source. Sampling strategy Note the sampling procedure. Describe the statistical methods that were used to predetermine sample size OR if no sample-size calculation was performed, describe how sample sizes were chosen and provide a rationale for why these sample sizes are sufficient. Data collection Describe the data collection procedure, including who recorded the data and how. Timing and spatial scale Indicate the start and stop dates of data collection, noting the frequency and periodicity of sampling and providing a rationale for these choices. If there is a gap between collection periods, state the dates for each sample cohort. Specify the spatial scale from which the data are taken Data exclusions If no data were excluded from the analyses, state so OR if data were excluded, describe the exclusions and the rationale behind them, indicating whether exclusion criteria were pre-established. Reproducibility Randomization Blinding Describe the measures taken to verify the reproducibility of experimental findings. For each experiment, note whether any attempts to repeat the experiment failed OR state that all attempts to repeat the experiment were successful. Describe how samples/organisms/participants were allocated into groups. If allocation was not random, describe how covariates were controlled. If this is not relevant to your study, explain why. Describe the extent of blinding used during data acquisition and analysis. If blinding was not possible, describe why OR explain why blinding was not relevant to your study. Did the study involve field work? Yes No Field work, collection and transport Field conditions Describe the study conditions for field work, providing relevant parameters (e.g. temperature, rainfall). Location State the location of the sampling or experiment, providing relevant parameters (e.g. latitude and longitude, elevation, water depth). Access & import/export Describe the efforts you have made to access habitats and to collect and import/export your samples in a responsible manner and in compliance with local, national and international laws, noting any permits that were obtained (give the name of the issuing authority, the date of issue, and any identifying information). Disturbance Describe any disturbance caused by the study and how it was minimized. Reporting for specific materials, systems and methods We require information from authors about some types of materials, experimental systems and methods used in many studies. Here, indicate whether each material, system or method listed is relevant to your study. If you are not sure if a list item applies to your research, read the appropriate section before selecting a response. n a t u r e p o r t f o l i o | r e p o r t i n g s u m m a r y A p r i l 2 0 2 3 3 Materials & experimental systems n/a Involved in the study Methods n/a Involved in the study Antibodies Eukaryotic cell lines Palaeontology and archaeology Animals and other organisms Clinical data Dual use research of concern Plants ChIP-seq Flow cytometry MRI-based neuroimaging Antibodies Antibodies used Validation Describe all antibodies used in the study; as applicable, provide supplier name, catalog number, clone name, and lot number. Describe the validation of each primary antibody for the species and application, noting any validation statements on the manufacturer’s website, relevant citations, antibody profiles in online databases, or data provided in the manuscript. Eukaryotic cell lines Policy information about cell lines and Sex and Gender in Research Cell line source(s) State the source of each cell line used and the sex of all primary cell lines and cells derived from human participants or vertebrate models. Authentication Describe the authentication procedures for each cell line used OR declare that none of the cell lines used were authenticated. Mycoplasma contamination Confirm that all cell lines tested negative for mycoplasma contamination OR describe the results of the testing for mycoplasma contamination OR declare that the cell lines were not tested for mycoplasma contamination. Commonly misidentified lines (See ICLAC register) Name any commonly misidentified cell lines used in the study and provide a rationale for their use. Palaeontology and Archaeology Specimen provenance Provide provenance information for specimens and describe permits that were obtained for the work (including the name of the issuing authority, the date of issue, and any identifying information). Permits should encompass collection and, where applicable, export. Specimen deposition Indicate where the specimens have been deposited to permit free access by other researchers. Dating methods If new dates are provided, describe how they were obtained (e.g. collection, storage, sample pretreatment and measurement), where they were obtained (i.e. lab name), the calibration program and the protocol for quality assurance OR state that no new dates are provided. Tick this box to confirm that the raw and calibrated dates are available in the paper or in Supplementary Information. Ethics oversight Identify the organization(s) that approved or provided guidance on the study protocol, OR state that no ethical approval or guidance was required and explain why not. Note that full information on the approval of the study protocol must also be provided in the manuscript. Animals and other research organisms Policy information about studies involving animals; ARRIVE guidelines recommended for reporting animal research, and Sex and Gender in Research Laboratory animals For laboratory animals, report species, strain and age OR state that the study did not involve laboratory animals. Wild animals Provide details on animals observed in or captured in the field; report species and age where possible. Describe how animals were caught and transported and what happened to captive animals after the study (if killed, explain why and describe method; if released, say where and when) OR state that the study did not involve wild animals. Reporting on sex Indicate if findings apply to only one sex; describe whether sex was considered in study design, methods used for assigning sex. Provide data disaggregated for sex where this information has been collected in the source data as appropriate; provide overall n a t u r e p o r t f o l i o | r e p o r t i n g s u m m a r y A p r i l 2 0 2 3 4 numbers in this Reporting Summary. Please state if this information has not been collected. Report sex-based analyses where performed, justify reasons for lack of sex-based analysis. Field-collected samples For laboratory work with field-collected samples, describe all relevant parameters such as housing, maintenance, temperature, photoperiod and end-of-experiment protocol OR state that the study did not involve samples collected from the field. Ethics oversight Identify the organization(s) that approved or provided guidance on the study protocol, OR state that no ethical approval or guidance was required and explain why not. Note that full information on the approval of the study protocol must also be provided in the manuscript. Clinical data Policy information about clinical studies All manuscripts should comply with the ICMJE guidelines for publication of clinical research and a completed CONSORT checklist must be included with all submissions. Clinical trial registration Provide the trial registration number from ClinicalTrials.gov or an equivalent agency. Study protocol Note where the full trial protocol can be accessed OR if not available, explain why. Data collection Describe the settings and locales of data collection, noting the time periods of recruitment and data collection. Outcomes Describe how you pre-defined primary and secondary outcome measures and how you assessed these measures. Dual use research of concern Policy information about dual use research of concern Hazards Could the accidental, deliberate or reckless misuse of agents or technologies generated in the work, or the application of information presented in the manuscript, pose a threat to: No Yes Public health National security Crops and/or livestock Ecosystems Any other significant area Experiments of concern Does the work involve any of these experiments of concern: No Yes Demonstrate how to render a vaccine ineffective Confer resistance to therapeutically useful antibiotics or antiviral agents Enhance the virulence of a pathogen or render a nonpathogen virulent Increase transmissibility of a pathogen Alter the host range of a pathogen Enable evasion of diagnostic/detection modalities Enable the weaponization of a biological agent or toxin Any other potentially harmful combination of experiments and agents Plants Seed stocks Report on the source of all seed stocks or other plant material used. If applicable, state the seed stock centre and catalogue number. If plant specimens were collected from the field, describe the collection location, date and sampling procedures. Novel plant genotypes Describe the methods by which all novel plant genotypes were produced. This includes those generated by transgenic approaches, gene editing, chemical/radiation-based mutagenesis and hybridization. For transgenic lines, describe the transformation method, the number of independent lines analyzed and the generation upon which experiments were performed. For gene-edited lines, describe the editor used, the endogenous sequence targeted for editing, the targeting guide RNA sequence (if applicable) and how the editor n a t u r e p o r t f o l i o | r e p o r t i n g s u m m a r y A p r i l 2 0 2 3 5 was applied. Authentication Describe any authentication procedures for each seed stock used or novel genotype generated. Describe any experiments used to assess the effect of a mutation and, where applicable, how potential secondary effects (e.g. second site T-DNA insertions, mosiacism, off-target gene editing) were examined. ChIP-seq Data deposition Confirm that both raw and final processed data have been deposited in a public database such as GEO. Confirm that you have deposited or provided access to graph files (e.g. BED files) for the called peaks. Data access links May remain private before publication. For "Initial submission" or "Revised version" documents, provide reviewer access links. For your "Final submission" document, provide a link to the deposited data. Files in database submission Provide a list of all files available in the database submission. Genome browser session (e.g. UCSC) Provide a link to an anonymized genome browser session for "Initial submission" and "Revised version" documents only, to enable peer review. Write "no longer applicable" for "Final submission" documents. Methodology Replicates Describe the experimental replicates, specifying number, type and replicate agreement. Sequencing depth Describe the sequencing depth for each experiment, providing the total number of reads, uniquely mapped reads, length of reads and whether they were paired- or single-end. Antibodies Describe the antibodies used for the ChIP-seq experiments; as applicable, provide supplier name, catalog number, clone name, and lot number. Peak calling parameters Specify the command line program and parameters used for read mapping and peak calling, including the ChIP, control and index files used. Describe the methods used to ensure data quality in full detail, including how many peaks are at FDR 5% and above 5-fold enrichment. Describe the software used to collect and analyze the ChIP-seq data. For custom code that has been deposited into a community repository, provide accession details. Data quality Software Flow Cytometry Plots Confirm that: The axis labels state the marker and fluorochrome used (e.g. CD4-FITC). The axis scales are clearly visible. Include numbers along axes only for bottom left plot of group (a 'group' is an analysis of identical markers). All plots are contour plots with outliers or pseudocolor plots. A numerical value for number of cells or percentage (with statistics) is provided. Methodology Sample preparation Describe the sample preparation, detailing the biological source of the cells and any tissue processing steps used. Instrument Software Identify the instrument used for data collection, specifying make and model number. Describe the software used to collect and analyze the flow cytometry data. For custom code that has been deposited into a community repository, provide accession details. Cell population abundance Describe the abundance of the relevant cell populations within post-sort fractions, providing details on the purity of the samples and how it was determined. Gating strategy Describe the gating strategy used for all relevant experiments, specifying the preliminary FSC/SSC gates of the starting cell population, indicating where boundaries between "positive" and "negative" staining cell populations are defined. Tick this box to confirm that a figure exemplifying the gating strategy is provided in the Supplementary Information. n a t u r e p o r t f o l i o | r e p o r t i n g s u m m a r y A p r i l 2 0 2 3 6 Magnetic resonance imaging Experimental design Design type Indicate task or resting state; event-related or block design. Design specifications Specify the number of blocks, trials or experimental units per session and/or subject, and specify the length of each trial or block (if trials are blocked) and interval between trials. Behavioral performance measures State number and/or type of variables recorded (e.g. correct button press, response time) and what statistics were used to establish that the subjects were performing the task as expected (e.g. mean, range, and/or standard deviation across subjects). Acquisition Imaging type(s) Specify: functional, structural, diffusion, perfusion. Field strength Specify in Tesla Sequence & imaging parameters Specify the pulse sequence type (gradient echo, spin echo, etc.), imaging type (EPI, spiral, etc.), field of view, matrix size, slice thickness, orientation and TE/TR/flip angle. Area of acquisition State whether a whole brain scan was used OR define the area of acquisition, describing how the region was determined. Diffusion MRI Used Not used Preprocessing Preprocessing software Provide detail on software version and revision number and on specific parameters (model/functions, brain extraction, segmentation, smoothing kernel size, etc.). Normalization If data were normalized/standardized, describe the approach(es): specify linear or non-linear and define image types used for transformation OR indicate that data were not normalized and explain rationale for lack of normalization. Normalization template Describe the template used for normalization/transformation, specifying subject space or group standardized space (e.g. original Talairach, MNI305, ICBM152) OR indicate that the data were not normalized. Noise and artifact removal Describe your procedure(s) for artifact and structured noise removal, specifying motion parameters, tissue signals and physiological signals (heart rate, respiration). Volume censoring Define your software and/or method and criteria for volume censoring, and state the extent of such censoring. Statistical modeling & inference Model type and settings Specify type (mass univariate, multivariate, RSA, predictive, etc.) and describe essential details of the model at the first and second levels (e.g. fixed, random or mixed effects; drift or auto-correlation). Effect(s) tested Define precise effect in terms of the task or stimulus conditions instead of psychological concepts and indicate whether ANOVA or factorial designs were used. Specify type of analysis: Whole brain ROI-based Both Statistic type for inference (See Eklund et al. 2016) Specify voxel-wise or cluster-wise and report all relevant parameters for cluster-wise methods. Correction Describe the type of correction and how it is obtained for multiple comparisons (e.g. FWE, FDR, permutation or Monte Carlo). Models & analysis n/a Involved in the study Functional and/or effective connectivity Graph analysis Multivariate modeling or predictive analysis Functional and/or effective connectivity Report the measures of dependence used and the model details (e.g. Pearson correlation, partial correlation, mutual information). Graph analysis Report the dependent variable and connectivity measure, specifying weighted graph or binarized graph, n a t u r e p o r t f o l i o | r e p o r t i n g s u m m a r y A p r i l 2 0 2 3 7 Graph analysis subject- or group-level, and the global and/or node summaries used (e.g. clustering coefficient, efficiency, etc.). Multivariate modeling and predictive analysis Specify independent variables, features extraction and dimension reduction, model, training and evaluation metrics. n a t u r e p o r t f o l i o | r e p o r t i n g s u m m a r y A p r i l 2 0 2 3 8
null
10.1088_1361-6501_ad0e9d.pdf
Data availability statement The data cannot be made publicly available upon publication because the cost of preparing, depositing and hosting the data would be prohibitive within the terms of this research project. The data that support the findings of this study are available upon reasonable request from the authors.
Data availability statement The data cannot be made publicly available upon publication because the cost of preparing, depositing and hosting the data would be prohibitive within the terms of this research project. The data that support the findings of this study are available upon reasonable request from the authors.
Meas. Sci. Technol. 35 (2024) 035002 (15pp) Measurement Science and Technology https://doi.org/10.1088/1361-6501/ad0e9d Attention features selection oversampling technique (AFS-O) for rolling bearing fault diagnosis with class imbalance Zhongze Han2, Haoran Wang2, Chen Shen1, Xuewei Song2, Longchao Cao1 and Lianqing Yu1,∗ 1 Hubei Key Laboratory of Digital Textile Equipment, Wuhan Textile University, Wuhan 430200, People’s Republic of China 2 School of Mechanical Engineering & Automation, Wuhan Textile University, Wuhan 430200, People’s Republic of China E-mail: yulq@wtu.edu.cn Received 9 May 2023, revised 11 September 2023 Accepted for publication 21 November 2023 Published 5 December 2023 Abstract When using data-driven methods for fault diagnosis of mechanical rotating components such as gears and bearings, there is a problem of class imbalance in the lifecycle data collected by sensors. The most commonly used method to address this issue is the synthetic minority over-sampling technique, which synthesizes samples in the feature space, but its blind synthesis may lead to redundant features in the synthetic samples. To avoid this deficiency, this paper proposes a feature-weighted oversampling method called AFS-O (Attention Features Selection Oversampling Technique). First, time–frequency domain features are extracted from the full lifecycle data of bearings to construct an initial subset of features, which serves the input to AFS. Then, AFS is then used to obtain the distribution of feature selection patterns and generate feature weights to determine the inclusion or exclusion of each feature, thereby constructing an optimal subset of features. Finally, the optimal feature subset is synthetically oversampled to achieve class-balanced data, which is then fed into a classifier. AFS-O is applied to the rolling bearing accelerated lifetime dataset from Xi’an Jiaotong University. Experimental results demonstrate that AFS-O outperforms other state-of-the-art synthetic oversampling algorithms in terms of Gmean, F2score, and Recall, confirming the effectiveness of the proposed method. Keywords: rolling bearing fault diagnosis, class imbalance, attention mechanism, oversampling technique 1. Introduction Rotating machinery is one of the most important equipment in modern industrial applications. Rotating machinery such as gears and bearings work under many complex working conditions resulting in their susceptibility to failure, which ∗ Author to whom any correspondence should be addressed. reduces the transmission accuracy, productivity and safety of the equipment [1, 2]. Therefore, fault diagnosis of rotating machinery is crucial to ensure the reliability, productivity and economic efficiency of industrial systems. The existing fault diagnosis methods based on the pro- gnostics and health management (PHM) [3] framework can be classified into three categories: model-based methods, data-driven methods, and hybrid methods. Different from the 1361-6501/24/035002+15$33.00 1 © 2023 IOP Publishing Ltd Meas. Sci. Technol. 35 (2024) 035002 Z Han et al model-based method, the data-driven method does not rely on expert knowledge, which makes it the most widely used method in the PHM framework. The performance of data- driven method largely depends on the quality of the extrac- ted features, that is, how to extract the features that effectively represent the health state of rotating machinery is the core of PHM technology. Deep Learning (DL) models, as one of the mainstream methods in PHM technology, utilize multi-layered neural net- work structures to perform hierarchical abstraction of input data, automatically extracting complex features. This effect- ively compensates for the drawback of manual feature extrac- tion methods, which lack the ability to adaptively learn features [4–6]. The mainstream DL models in PHM include deep belief network, autoencoder (AE), convolutional neural network (CNN), recurrent neural network, and generative adversarial network (GAN) [7]. These classic models have been widely applied in various fields within the domain of PHM. Zhou et al [8] applied channel fusion mechanism to con- volutional AE, giving the model a more stable feature repres- entation capability, thus improving its accuracy and efficiency in fault diagnosis. Yao and Han [9] proposed a deep transfer CNN for accurately estimating the capacity of lithium-ion bat- teries to ensure their safety and reliability. Mitici [10] intro- duced a CNN-based multi-component predictive maintenance framework and validated its advantages in maintenance cost and reliability using a real engineering dataset. However, in actual production processes, the full life- cycle data of rotating machinery exhibits class imbalance characteristics [11], where the number of fault samples is much smaller than that of normal samples. DL models are affected by learning bias [12] when trained in a class- imbalanced setting. In other words, when using a class- imbalanced dataset to train a DL model, the majority class samples have a larger influence on the loss function com- pared to the minority class samples. As a result, the model may perform well on majority class samples but poorly on minority class samples, which is unfavorable for fault dia- gnosis. Due to its prevalence and importance in industrial scenarios, the class imbalance problem has attracted much attention from researchers in various fields. Zhao et al [13] proposed a method based on wavelet packet distortion and CNN to address the class imbalance issue in mechanical fault diagnosis. Wang et al [14] combined adaptive variational AE with GAN to generate new fault samples. Oversampling technique, as one of the main methods for class imbalance, aims to balance the class sample quantities by randomly replicating minority class samples. Tao et al [15] defined a hypersphere region for minority class samples based on the imbalance ratio and distance metrics, and performed adaptive oversampling with varying radius for the minority class samples contained within different hyperspheres. Liu et al [16] proposed a method that effectively suppresses the generation of noisy samples during the oversampling process. Synthetic minority over-sampling technique (SMOTE) [17], as one of the oversampling methods, differs from the conventional random oversampling replication mechanism. Instead of directly replicating minority class samples, SMOTE performs linear interpolation between minority class samples and randomly selected samples from their neighborhood. This effectively alleviates the overfitting issue caused by over- sampling. Li et al [18] proposed an improved SMOTE method that adaptively selects the value of k based on the distribu- tion of minority class samples to enhance the generalization of synthetic samples. Meng and Li [19] defined positive regions using a central offset factor and performed synthetic over- sampling in sparsely distributed regions. However, the blindness of SMOTE in synthesizing samples in feature space results in a high dependency on the fea- tures present in the minority class samples. Therefore, it is necessary to perform feature selection on the minority class samples themselves before synthetic oversampling. By remov- ing redundant features and constructing an optimal feature subset, the quality of the synthetic samples can be improved significantly. Feature selection is commonly considered as a necessary preprocessing step for classification tasks in the context of class imbalance [20–23] to construct an optimal feature subset. The commonly used feature selection methods can be mainly categorized into three types: Filter, Wrapper and Embedded [24]. Among them, Filter methods are widely used due to its low computational cost and simplicity of algorithms. Filter methods individually evaluate the correlation between each feature and the class labels, assigning weights to the features. The features are then selected based on the weight magnitude to construct an optimal feature subset. This process usually requires evaluating feature relevance using scoring functions and thresholding methods. However, even though the selected features have the highest correlation with the class labels, they may still be correlated with each other and contain redundant information, leading to the loss of useful information from the original feature subset [25]. Therefore, to obtain the best fea- ture subset with strong representational capabilities, DL mod- els, due to their powerful adaptive information extraction cap- abilities, have been widely applied in feature selection across various scenarios [26–28]. To further enhance the perform- ance of DL models in specific tasks, the attention mechan- ism (AM) [29] has been introduced into the field of DL. AM aims to guide DL models to focus on more relevant inform- ation for the current task by assigning feature weights, mak- ing it similar to feature selection. Unlike common filter meth- ods such as Fisher Score and Relief [30], AM utilizes its own network structure to adaptively extract more complex rela- tionships among features, in addition to their correlations, to assign feature weights. It continuously adjusts these feature weights using the backpropagation algorithm to achieve the goal of feature selection. This means that AM can be used to improve the minority class samples themselves, thereby avoid- ing the problem of poor-quality synthetic samples caused by the blindness of SMOTE in feature space. This paper proposes a fault diagnosis method named AFS-O (Attention Feature Selection with Oversampling), 2 Meas. Sci. Technol. 35 (2024) 035002 Z Han et al which combines the attention feature selection (AFS) mech- anism with synthetic oversampling technique. Firstly, time- frequency domain features are extracted from the raw vibra- tion signals of the bearings to construct an initial feature set, which serves as the input to AFS. Next, the attention block in AFS treats the correlation between each input feature and the labels as a binary classification problem and generates feature weights based on the distribution of the feature selection pat- terns. Then, the learning block in AFS continuously adjusts the feature weights to make choices among features and construct the optimal feature subset. Finally, the optimal feature subset undergoes synthetic oversampling and is fed into the classifier for fault diagnosis. The main contributions of this paper are as follows: 1. A fault diagnosis model based on an improved SMOTE algorithm is proposed. Unlike the improvements on SMOTE using neighborhood partitioning methods [31, 32], AFS-O focuses on enhancing the quality of synthetic samples by improving the minority class samples them- selves. This property allows AFS-O to avoid the problem of poor synthetic samples quality caused by the blindness of the SMOTE synthesis rules. 2. Compared to traditional feature selection methods, AFS, as a DL model, considers complex relationships among fea- tures by introducing the AM to construct the optimal feature subset. 3. AFS-O was compared with the latest improved SMOTE method on real engineering datasets, demonstrating the superiority of AFS-O in addressing class imbalance issues. Section 2 provides a review of synthetic oversampling methods and feature selection methods in the context of DL. Section 3 elaborates on the working principles of the proposed approach. In section 4, comparative experiments are designed on the rolling bearing accelerated life test dataset from Xi’an Jiaotong University to validate the performance of AFS-O. Finally, section 5 summarizes the work of this paper. 2. Related works 2.1. Overview of synthetic oversampling The focus of this study is on binary classification problems under class imbalance, where minority class samples often carry more classification-relevant information [11]. We refer to majority class samples as negative class and minority class samples as positive class. Although SMOTE effectively alle- viates the overfitting issue caused by oversampling, it also has certain limitations. Problem 1: poor-quality of synthetic samples. The positive samples may contain noise samples, which directly affects the quality of the synthetic samples. Problem 2: blurred class boundaries. SMOTE does not consider the distribution of negative samples when synthes- izing positive samples. Positive samples at the class boundary have their k-nearest neighbors also at the boundary, leading to synthetic samples being in the class overlap region, which further blurs the class boundaries. Problem 3: uneven distribution of positive samples. When the distribution of positive samples contains both dense and sparse regions, the synthetic samples generated by SMOTE follow the proximity principle and are distributed accordingly. This means that after applying SMOTE, the dense regions of the positive class will still remain relatively dense, while the sparse regions will remain relatively sparse. Researchers have made corresponding improvements to SMOTE as for the problems. Borderline-SMOTE [33] was proposed to address Problem 2, while ADASYN [34] (Adaptive Synthetic Sampling) [35] assigns adaptive weights to different distributions of positive samples, determining the number of synthetic samples based on the weight magnitude to address Problem 3. In recent years, several SMOTE variants have also been proposed to tackle Problem 1. For example, FW-SMOTE [31] uses weighted Minkowski distance to define the neighborhood of positive samples. This approach ensures that the partitioned neighborhood contains positive samples that are more relevant to the classification task, thereby improving the quality of synthetic samples. SMOTIFIED-GAN [35] inputs both synthetic samples and positive samples into a GAN and obtains realistic synthetic samples after its convergence. Deep-SMOTE [36] success- fully embeds SMOTE into a deep AE, obtaining high-quality synthetic samples. Geometric-SMOTE [37] defines a flexible geometric region around positive samples as the neighborhood and performs synthesis at the boundary of positive samples. However, for problem 1, the above methods do not consider the negative impact of redundant features that may exist in pos- itive samples on the classifier’s performance. Therefore, it is necessary to perform appropriate feature selection on positive samples before conducting synthetic oversampling to improve the quality of synthetic samples. 2.2. AM for feature selection In fact, traditional feature selection methods often perform poorly in class-imbalanced scenarios because each feature may not be independent and could have complex relation- ships with others. Thanks to the network architecture of DL, the unique global information extraction capability of the AM gives it great potential in the field of feature engineering [38– 40], and researchers have successfully applied AM to fea- ture selection. For example, Paul et al [41] proposed a GAN- based multi-label feature selection method. Zhuang et al [42] designed a residual convolution module for feature learning to enhance classification and suppress redundant features. Zhang et al [43] added attention to the frequency bands obtained through wavelet packet transformation, highlighting key fea- tures in the wavelet coefficients to improve the performance of ResNet in wind turbine gearbox fault diagnosis. Therefore, we believe that AM can be used in the feature selection process before synthetic oversampling to improve the positive samples and construct the optimal feature sub- set. This approach addresses the issue of poor-quality synthetic samples caused by the blindness of SMOTE’s synthesis rules. 3 Meas. Sci. Technol. 35 (2024) 035002 Z Han et al Figure 1. Overall framework of AFS-O. 3. The proposed method This section presents the details of the proposed method. Section 3.1 illustrates the overall architecture of the fault diagnosis model based on AFS-O. Section 3.2 explains the content of time-frequency domain feature extraction using examples and equations. Sections 3.3 and 3.4 describe the working principles of AFS and SMOTE, respectively. 3.1. Fault diagnosis module based on AFS-O Figure 1 illustrates the rolling bearing full-lifecycle fault dia- gnosis model based on AFS-O, with the following steps: Step 1. Obtain the vibration signals of the rolling bearing throughout its full lifecycle and extract time-frequency domain features to create an initial feature subset. Step 2. Input the initial feature subset into AFS. The atten- tion block generates feature weights based on the distribution of feature selection patterns, and the learning block continu- ously adjusts these feature weights to make choices among features and construct the optimal feature subset. Step 3. Apply synthetic oversampling to the positive class samples within the optimal feature subset to obtain a class- balanced feature subset. Step 4. Input the class-balanced feature subset into the clas- sifier for classification, and evaluate the performance of AFS- O based on the Recall, Gmean, and F2score of the classifier. have conducted in-depth research in the field of fault diagnosis and prediction [45–48]. In their work [49], they presented a total of 24 time-frequency domain features for identify- ing rotating machinery faults. Table 1 provides the calcula- tion methods for 11 time-domain features and 13 frequency- domain features, and these 24 time–frequency domain features form the initial feature subset. For example, p7 represents kurtosis, which is sensitive to impulsive signals in bearing vibration data. p4 represents the root mean square (RMS) value, reflecting the energy intensity and stability of the vibration signal. p16 represents the centroid frequency, describing the frequency of the dominant vibration signal in the spectrum. p18 represents the frequency RMS, reflecting the frequency distribution of the vibration signal. In the table, N represents the number of samples, x (n) denotes the input vibration signal, s (k) represents the power spectral density, and fk corresponds to the frequency amp- litude of each sample point. As shown in figure 2, after extract- ing the time–frequency domain features from the source vibra- tion signal, a preliminary feature dataset X is constructed from each individual feature: X = {xi |i = 1, 2, . . . , n} ∈ Rm×n where m represents the number of samples, and n represents the dimensionality of the features. 3.3. AFS 3.2. Time–frequency domain features extraction Time-domain and frequency-domain indicators to some extent represent the health status of rotating machinery [44]. Lei et al Although the above features can reflect rotational machinery faults from different perspectives, there are differences in the sensitivity of different features to faults. Therefore, it is neces- sary to select key features that carry more fault information 4 Meas. Sci. Technol. 35 (2024) 035002 Z Han et al Table 1. Time–frequency domain features for bearings. Time-domain features ∑ N n=1 x(n) N ( ∑ √ N n=1 |x(n)| ) 2 N p1 = p3 = p5 = max |x (n)| ∑ N n=1 (x(n)−p1)4 (N−1)p4 2 p7 = p9 = p5 p3 p11 = Np5∑ N n=1 |x(n)| √ ∑ √ ∑ N n=1 (x(n)−p1) N−1 N n=1 (x(n))2 N ∑ N n=1 (x(n)−p1) (N−1)p3 2 p2 = p4 = p6 = p8 = p5 p4 p10 = Np4∑ N n=1 |x(n)| Frequency-domain features p12 = p14 = p16 = p18 = p20 = p22 = p24 = ∑ ∑ K k=1 s(k) K K k=1 (s(k)−p12)3 p13)3 K( √ ∑ K k=1 s(k)fk ∑ K k=1 s(k) √ ∑ K k=1 s(k)f 2 ∑ k K k=1 s(k) ∑ √∑ K k=1 s(k)f 2 ∑ k ∑ ∑ K k=1 s(k)f 4 k K k=1 s(k) ( fk K k=1 −p16)3s(k) Kp3 17 −p16)1/2s(k) √ K K=1 ( fk K p17 p13 = p15 = p17 = p19 = ∑ ∑ K k=1 (s(k)−p12) K−1 K k=1 (s(k)−p12)4 Kp2 13 √ ∑ √ ∑ ∑ K k=1 ( fk −p16)2s(k) K K k=1 s(k)f 4 k K k=1 s(k)f 2 k p21 = p17 p16 ∑ p23 = K k=1 ( fk −p16)4s(k) Kp4 17 Figure 2. Process of features extraction. while removing redundant features to improve the overall per- formance of the model. Figure 3 shows the framework of AFS. AFS is composed of two main components: the attention block and the learning block. The attention block is respons- ible for generating feature weights based on the association between features and class labels. It transforms this associ- ation into a binary classification problem and assigns a shal- low attention network to each feature. The learning block, on the other hand, establishes the relationship between the labels and the feature weights using a deep neural network. During the training process, the learning block continuously corrects the feature weights to find the optimal correlation between the weighted features and the class labels. The initial feature sub- set X serves as the input to the attention block. Firstly, the Dense Net in the atten- 3.3.1. Attention block. tion block is used to extract the intrinsic correlation within the input X which is represented by equation (1). E compresses the original feature space and remove noise or outliers, E = Tanh (XW1 + b1) (1) 5 where the W1 and b1 are the parameters to be learned in the Dense Net. Then, the attention block provides a shallow attention net- work for each feature to determine the probability of the fea- ture being selected. AFS proposes a novel soft AM in this attention block. The shallow attention network considers the correlation between features and labels as a binary classifica- tion problem, where the attention network outputs two values pk and uk (k is the kth feature), representing selection and non- selection, respectively.    pk = wk phk L + bk p (2) uk = wk uhk L + bk u L is the Lth hidden layer in the kth shallow attention u are the parameters to be learned in p and wk u, bk p, bk where hk network. wk pk and uk. Meas. Sci. Technol. 35 (2024) 035002 Z Han et al Figure 3. Framework of AFS. Furthermore, pk and uk are converted to the probability values through Softmax, respectively. Here, only the probab- ility ak of the kth feature selected by pk is concerned. ak is cal- culated as follows: ( ) exp pk vector S through backpropagation, constantly considering whether to select certain features. After the loss function con- verges, the feature matrix obtained by removing redundant features is the optimal feature subset G before the synthesis oversampling process G: ak = exp (pk) + exp (uk) . (3) G = {gi|i = 1, 2, . . . , d} ∈ Rm×d And all of above- ak constitute the attention matrix A: { ak i A = |i = 1, 2, . . . , m; k = 1, 2, . . . , n } ∈ Rm×n where m is the number of samples, while n is the number of features. Finally, the weight factors sk for the corresponding features are calculated from the ak obtained by the kth shallow attention network using equation (5) : sk = 1 m m∑ i =1 ak i (4) and the weight vector S is formed by all sk: S = (si |i = 1, 2, . . . , n) ∈ Rm×n. In the learning block, the Hadamard 3.3.2. Learning block. product between the weight vector S and input X results in the matrix G with weighted features: G = X ⊙ S. (5) For classification tasks, the learning block solves the cross entropy loss function using a DNN and updates the weight    D = ReLu (GW2 + b2) arg min S loss [ fD (Gθs − Y)] + λR (θ) (6) where d represents the feature dimension after removing redundant features. W2 and b2 are the parameters that need to be learned in the DNN. fD (·) denotes the loss function of the DNN, which is commonly used as the cross-entropy loss function for binary classification tasks. λR (·) represents the regularization term used to prevent overfitting. 3.4. Synthetic oversampling process The fundamental concept behind SMOTE is presented shown in figure 4. To generate a synthetic sample, firstly, a minor- ity sample is selected as the root sample. Then, a sample is randomly selected from the neighborhood (k nearest neigh- bor sample regions of the same class) of xi with the upward sampling rate N as the auxiliary sample of the synthetic sample, and it is repeated n times. Lastly, linear interpolation is carried out on each dimensional feature in the feature space between the root sample and the auxiliary sample, utilizing through equation (7), and n synthetic samples are ultimately produced generated, 6 Meas. Sci. Technol. 35 (2024) 035002 Z Han et al Figure 4. Linear synthesis rules of SMOTE. xnew = xi + γ (xij − xi) (7) is the root sample of ith (i = 1, 2, 3, . . . , n), where xi jth xij ( j = 1, 2, . . . , k) is the the ith selection, γ is a random number between [0,1], xnew presents the synthetic samples after linear interpolation. auxiliary sample of After obtaining the optimal feature subset G, the positive samples it contains need to be synthetically oversampled. In this paper, the k value of SMOTE is set to 5 by default, and the balanced class feature subset ˆG is obtained after SMOTE processing and is fed into the classifier. ˆG = {ˆgi|i = 1, 2, . . . , d} ∈ Rˆm×d where ˆm represents the number of samples in ˆG, which is cal- culated as follows: ˆm = m × 2IR + 1 IR + 1 . (8) 4. Experiment and results 4.1. Description of proposed datasets The dataset used in this study is the rolling bearing accelerated life test dataset from Xi’an Jiaotong University (XJTU-SY) [50]。As presented in figure 5. The experimental setup com- prises an AC motor, a motor speed controller, a rotating shaft, support bearings, a hydraulic loading system, and test bear- ings. The experimental platform allows for adjustment of oper- ating conditions, primarily including radial force and speed, and establish three different operating conditions by varying the radial force and speed. The radial force is generated by the hydraulic loading system and is applied to the bearing seat of the test bearing, while the speed is set and adjusted by the AC motor’s speed controller. The test bearing used in the experi- ment is LDK UER204, and its detailed parameters are listed in table 2. Two PCB 352C33 unidirectional accelerometers were util- ized to acquire vibration signals in the horizontal and ver- tical directions of the test bearings. As shown in figure 6, the sampling frequency was set to 25.6 kHz, with a sampling dur- ation of 1.28 s per sampling and a sampling interval of 1 min. The amount of data acquired from a single sampling was 32 768. Table 3 presents detailed information on each bearing dataset under three operating conditions, including the corres- ponding operating condition, the number of sampling times, the class imbalance ratio IR (which is the negative samples divided by the positive samples), and the actual life and fault location of the bearing. Figure 7 shows three types of bearing fault contained in XJTU-SY. 4.2. Evaluation metrics In the full lifecycle data of rolling bearings, the number of negative samples is much larger than that of positive samples. Therefore, accuracy is not a suitable metric for eval- uating the classification performance of classifiers under class imbalance. Instead, Gmean, Fβ Score, and recall based on the confusion matrix (as shown in table 4) are selected as evaluation metrics to measure the performance of classifiers. Among these, Gmean and Fβ Score are suitable for measuring the overall performance of classifiers. A classifier must per- form well on both classes to obtain larger values of Gmean and Fβ Score. Recall is suitable for measuring the local perform- ance of classifiers as it pays more attention to positive class samples. The calculation formulas of the Gmean and Fβ Score are as follows:    √ Gmean = ( FβScore = 1 + β2 Recall ∗ Precision ) Recall ∗ Precision β2·Precision + Recall where recall and precision are calculated as follows: Recall = TP TP+FN Precision = TP TP+FP . (9) (10) (11) Recall represents the proportion of correctly predicted pos- itive samples out of all positive samples, while precision rep- resents the proportion of true positive samples in the pre- dicted positive samples. In fault diagnosis problems, detecting 7 Meas. Sci. Technol. 35 (2024) 035002 Z Han et al Figure 5. Bearing accelerated life test platform. Table 2. LDK UER204 parameters. Parameters Values Parameters Values Inner race diameter /mm Outer race diameter /mm Bearing mean diameter /mm Load rating (dynamic) /kN 29.30 39.80 34.55 12.82 Ball diameter/mm Number of balls ◦ Contact angle/( Load rating (static) /kN ) 7.92 8 0 6.65 Figure 6. Settings for data acquisition. Table 3. Information of XJTU-SY datasets. Operating condition Datasets Sampling times IR Actual life Fault location Condition 1 (35 Hz/12kN) Condition 2 (37.5 Hz/11kN) Condition 3 (40 Hz/10kN) Bearing1_1 Bearing1_2 Bearing1_3 Bearing2_1 Bearing2_2 Bearing2_3 Bearing3_1 Bearing3_2 Bearing3_3 29.69 39.57 11.97 14.83 11.99 9.77 64.86 93.42 38.41 2 h3 min 2 h2 min 52 min 8 h11 min 8 h53 min 5 h39 min 42 h18 min 41 h36 min 25 h15 min Outer race Cage Inner race Inner race Cage Outer race Outer race Cage Inner race 123 122 52 491 533 339 2538 2496 1515 8 Meas. Sci. Technol. 35 (2024) 035002 Z Han et al Figure 7. Type of faults in XJTU bearings: (a) Outer race fracture. (b) Inner race wear. (c) Cage fracture. Table 4. Confusion matrix. True positive class True negative class Predicted Positive class Predicted Negative class TP FN FP TN Table 5. Parameter settings for AFS and VGG16. AFS Parameters Batch size Learning rate Learning rate decay Optimizer Loss function Activation Epoch Values 100 0.001 StepLR (gamma = 0.99) Adam CrossEntropyLoss ReLu 3000 VGG16 Parameters Batch size Learning rate Learning rate decay Optimizer Loss function Activation Epoch Values 16 0.001 ReduceLRonPlateau Adam CrossEntropyLoss ReLu/Sigmoid 100 positive samples is of utmost importance, so the value of β is set to 2 in equation (9) to give more weight to recall. 4.3. Comparison experiments To evaluate the performance of AFS as a feature selection method, the experimental section compares four common feature selection methods: maximum information coefficient (MIC), variance selection, support vector machine (SVM) with L1 penalty, and random forest (RF) on the XJTU-SY dataset. FW-SMOTE, SMOTIFIED-GAN, deep-SMOTE, and geometric-SMOTE are four oversampling methods that have been proven effective in various class-imbalanced scenarios. Therefore, AFS-O is compared with these four latest methods on the XJTU-SY dataset to validate its performance. To facilitate the description of the subsequent experimental results, we consider the classifier’s scores as the scores for each synthetic oversampling method. The selected comparison meth- 4.3.1. Parameter settings. ods use the parameters as set in the original paper. For AFS, the initialization parameters are set to a truncated normal distribution with a mean of 0 and a standard deviation of 0.1. The dataset is divided using stratified k-fold cross-validation to prevent classifier overfitting, and the default value for k is set to 10. For consistency in comparison, all methods use the VGG16 CNN as the classifier. Table 5 shows the parameter settings for AFS andVGG16 CNN training. feature selection 4.3.2. Comparison with well-known Figure 8 presents the Recall, Gmean, methods on XJTU-SY. and F2score of AFS and the other four common feature selection methods. From figure 8(a), it can be observed that AFS achieves the highest Recall on all datasets compared to the other feature selection methods, with variance being the lowest. Particularly, AFS achieves the highest Recall of 0.948 on the Bearing2_3 dataset. Similarly, AFS also obtains the highest Gmean and F2score on all datasets, with the maximum values of 0.945 and 0.965 on the Bearing2_3, respectively. In summary, the performance of the machine learning- based feature selection methods (RF and L1-SVM) is compar- able to AFS, while methods solely relying on feature correla- tions like Variance and MIC score lower than the three afore- mentioned methods. 9 Meas. Sci. Technol. 35 (2024) 035002 Z Han et al Figure 8. Comparison of AFS and well-known feature selection methods on XJTU-SY. 4.3.3. Comparison with mentioned oversampling methods Tables 6–8 present the Recall, F2score, and on XJTU-SY. Gmean of AFS-O and the other four oversampling meth- ods on the XJTU-SY dataset, respectively. The comparison experiment results consist of the average and standard error of ten sets of Gmean, F2 Score and Recall for each oversampling method. The maximum in each row is highlighted in bold. Figure 9 shows the distribution of scores for AFS-O and other oversampling algorithms on the XJTU-SY dataset. As shown in figure 9(a), Recall of AFS-O has a more concen- trated score range and the highest mean value (0.914). In comparison, FW-SMOTE has the most concentrated score range and the second-highest mean value (0.878). Conversely, SMOTIFIED-GAN exhibits larger fluctuations in Recall and the lowest mean value (0.812). Deep-SMOTE and Geometric- SMOTE show similar overall performance, with Recall mean values of 0.846 and 0.855, respectively. In figures 9(b) and (c), AFS-O achieves the highest mean values of Gmean (0.932) and F2score (0.928). The trends in Gmean and F2score for other oversampling methods are sim- ilar to recall: Gmean and F2score of FW-SMOTE are more concentrated compared to recall, while SMOTIFIED-GAN has the lowest mean values for Gmean and F2score, at 0.744 and 0.753, respectively. In conclusion, AFS-O has the highest recall mean, indic- ating that it can identify more positive samples compared 10 Meas. Sci. Technol. 35 (2024) 035002 Z Han et al Table 6. Comparison results of each oversampling method on recall. Methods Dataset No. AFS-O FW-SMOTE SMOTIFIED-GAN Deep-SMOTE Geometric-SMOTE 1_1 1_2 1_3 2_1 2_2 2_3 3_1 3_2 3_3 Avg 0.913 ± 0.01 0.896 ± 0.01 0.929 ± 0.02 0.927 ± 0.02 0.934 ± 0.02 0.948 ± 0.01 0.887 ± 0.03 0.874 ± 0.02 0.907 ± 0.01 0.914 ± 0.02 0.875 ± 0.02 0.902 ± 0.01 0.872 ± 0.02 0.891 ± 0.02 0.909 ± 0.02 0.908 ± 0.02 0.890 ± 0.02 0.858 ± 0.03 0.879 ± 0.03 0.878 ± 0.02 0.772 ± 0.04 0.804 ± 0.05 0.835 ± 0.02 0.878 ± 0.02 0.857 ± 0.03 0.862 ± 0.02 0.739 ± 0.03 0.720 ± 0.05 0.811 ± 0.04 0.812 ± 0.03 0.864 ± 0.02 0.879 ± 0.02 0.850 ± 0.01 0.904 ± 0.02 0.873 ± 0.02 0.889 ± 0.02 0.824 ± 0.03 0.805 ± 0.04 0.844 ± 0.03 0.846 ± 0.02 0.863 ± 0.02 0.866 ± 0.02 0.832 ± 0.01 0.912 ± 0.02 0.872 ± 0.02 0.879 ± 0.01 0.817 ± 0.03 0.784 ± 0.03 0.870 ± 0.02 0.855 ± 0.02 Note: Bold highlights the best performance method in the certain dataset. Table 7. Comparison results of each oversampling method on F2score . Methods Dataset No. AFS-O FW-SMOTE SMOTIFIED-GAN Deep-SMOTE Geometric-SMOTE 1_1 1_2 1_3 2_1 2_2 2_3 3_1 3_2 3_3 Avg 0.927 ± 0.01 0.913 ± 0.01 0.941 ± 0.02 0.943 ± 0.01 0.936 ± 0.02 0.965 ± 0.01 0.922 ± 0.03 0.903 ± 0.02 0.926 ± 0.01 0.928 ± 0.02 0.895 ± 0.01 0.916 ± 0.01 0.877 ± 0.01 0.894 ± 0.02 0.882 ± 0.02 0.920 ± 0.01 0.873 ± 0.02 0.859 ± 0.02 0.890 ± 0.01 0.888 ± 0.02 0.804 ± 0.04 0.833 ± 0.04 0.839 ± 0.02 0.870 ± 0.03 0.862 ± 0.04 0.897 ± 0.03 0.792 ± 0.02 0.744 ± 0.05 0.820 ± 0.03 0.829 ± 0.03 0.880 ± 0.02 0.876 ± 0.01 0.863 ± 0.02 0.916 ± 0.02 0.878 ± 0.02 0.910 ± 0.01 0.845 ± 0.02 0.836 ± 0.03 0.854 ± 0.01 0.873 ± 0.02 0.879 ± 0.01 0.886 ± 0.01 0.914 ± 0.02 0.945 ± 0.02 0.886 ± 0.02 0.902 ± 0.02 0.844 ± 0.02 0.821 ± 0.03 0.871 ± 0.01 0.883 ± 0.02 Note: Bold highlights the best performance method in the certain dataset. Table 8. Comparison results of each oversampling method on Gmean. Methods Dataset No. AFS-O FW-SMOTE SMOTIFIED-GAN Deep-SMOTE Geometric-SMOTE 1_1 1_2 1_3 2_1 2_2 2_3 3_1 3_2 3_3 Avg 0.934 ± 0.01 0.917 ± 0.01 0.946 ± 0.02 0.936 ± 0.01 0.945 ± 0.02 0.976 ± 0.01 0.918 ± 0.02 0.899 ± 0.03 0.923 ± 0.01 0.932 ± 0.02 0.892 ± 0.01 0.914 ± 0.01 0.893 ± 0.01 0.910 ± 0.02 0.908 ± 0.02 0.919 ± 0.02 0.872 ± 0.02 0.865 ± 0.03 0.891 ± 0.01 0.901 ± 0.02 0.807 ± 0.04 0.831 ± 0.04 0.856 ± 0.03 0.882 ± 0.02 0.871 ± 0.03 0.885 ± 0.02 0.778 ± 0.02 0.753 ± 0.05 0.824 ± 0.03 0.843 ± 0.03 0.874 ± 0.02 0.883 ± 0.01 0.877 ± 0.02 0.920 ± 0.02 0.894 ± 0.02 0.907 ± 0.01 0.842 ± 0.02 0.831 ± 0.03 0.860 ± 0.01 0.876 ± 0.02 0.887 ± 0.01 0.892 ± 0.01 0.911 ± 0.02 0.940 ± 0.02 0.893 ± 0.02 0.913 ± 0.02 0.831 ± 0.02 0.835 ± 0.03 0.879 ± 0.01 0.882 ± 0.03 Note: Bold highlights the best performance method in the certain dataset. to other oversampling methods. Additionally, as shown in equation (8), the highest Gmean and F2score indicate that AFS-O also achieves the highest precision. Therefore, in most cases, AFS-O outperforms the other four oversampling meth- ods. It is worth noting that while FW-SMOTE has a slightly lower upper bound in terms of scores compared to AFS- O, it exhibits better robustness. On the other hand, during the training process of SMOTIFIED-GAN, we observed that the GAN had difficulty converging to a Nash equilibrium [51], which may lead to unstable synthetic sample quality and poorer performance when faced with datasets of different distributions. 4.3.4. Comparison with mentioned oversampling methods on This section compares the performance of different IRs. 11 Meas. Sci. Technol. 35 (2024) 035002 Z Han et al Figure 9. Distribution of each oversampling method on the evaluation metrics. each oversampling method under different IRs. We select the feature subsets of Bearing3_2 (IR = 93.42), Bearing1_2 (IR = 39.57), and Bearing2_3 (IR = 9.77) to highlight the performance differences of each oversampling method when facing different IRs. From figure 10, it can be observed that AFS-O has the highest Recall (0.948, 0.874), F2score (0.965, 0.903), and Gmean (0.976, 0.899) on Bearing2_3 and Bearing3_2, while SMOTIFIED-GAN has the lowest scores, with val- ues of (0.862, 0.720), (0.885, 0.753), and (0.897, 0.744), respectively. It can be seen that the scores of each over- sampling method show a certain degree of decline as the IR increases. Table 9 presents the decline in three evalu- ation metrics for each oversampling method. Among them, FW-SMOTE shows the smallest decline in each metric, with percentages of 4.5% (Recall), 4.9% (F2score), and 7.1% (Gmean). On the other hand, SMOTIFIED-GAN exhibits the most significant declines, with percentages of 14.2% (Recall), 13.2% (F2score), and 15.3% (Gmean). Among the remaining three oversampling methods, AFS-O demonstrates the best robustness, with decreases of 7.4% (Recall), 6.2% (F2score), and 7.7% (Gmean). In summary, when the IR increases significantly, both FW- SMOTE and AFS-O remain effective in helping the classifica- tion model learn the features of positive samples to improve the performance of the classifier. In fact, FW-SMOTE also involves a feature selection process. Unlike AFS-O, it uses the induced ordered weighted average operator to weight the fea- tures of positive samples in the distance metric, and introduces a feature ranking method to remove features with weights below a specified threshold for feature selection. Therefore, the excellent performance of AFS-O and FW-SMOTE on feature subsets with high IR indicates the effectiveness and superiority of establishing an optimal feature subset through feature selection methods to address the class imbalance problem. 12 Meas. Sci. Technol. 35 (2024) 035002 Z Han et al Figure 10. Comparison of each oversampling method on different IRs. Table 9. The decline of each oversampling method on the evaluation metrics. AFS-O FW-SMOTE SMOTIFIED-GAN Deep-SMOTE Geometric-SMOTE Recall 7.4% 4.5% 14.2% 8.4% 9.5% F2-score 6.2% 4.9% 13.2% 7.4% 8.1% Gmean 7.7% 7.1% 15.3% 7.6% 7.8% 5. Conclusions and future works This paper proposes a data preprocessing method called AFS- O, which combines attention-based feature selection with synthetic oversampling techniques. The aim is to improve the quality of synthetic samples and address the class imbalance problem in the fault diagnosis of rolling bearings through- out their entire lifecycle. Attention block treats the correlation 13 Meas. Sci. Technol. 35 (2024) 035002 Z Han et al between each input feature and the labels as a binary classi- fication problem and generates feature weights based on the distribution of the feature selection pattern. The learning block continuously adjusts these weights to select the most inform- ative features and establish the optimal feature subset. The main conclusions drawn from applying AFS-O and four other latest oversampling methods to the XJTU-SY dataset are as follows: 1. AFS outperforms common feature selection methods, demonstrating the powerful potential of the AM in feature selection. Similarly, AFS-O shows higher average scores on all metrics across the nine feature subsets compared to the other four oversampling methods, confirming the effective- ness of AFS-O in addressing the class imbalance problem in binary classification tasks. 2. From the performance of AFS-O on feature subsets with high IR, it is evident that establishing an optimal feature subset through feature selection is a viable preprocessing step for extreme class imbalance scenarios in classification tasks. 3. We infer that further improvements in classifier perform- ance in class imbalance scenarios can be achieved by com- bining the construction of an optimal subset of features with improvements in the SMOTE’s neighborhood partitioning or synthesis rules. 4. AFS-O is proposed specifically for binary classification tasks in class imbalance scenarios. Therefore, in future work, we will consider investigating the performance of AFS-O in multi-class tasks or prediction tasks under class imbalance conditions. Data availability statement The data cannot be made publicly available upon publication because the cost of preparing, depositing and hosting the data would be prohibitive within the terms of this research project. The data that support the findings of this study are available upon reasonable request from the authors. Acknowledgments Innovation Program of Wuhan-Shuguang Knowledge Project under Grant No. 2022010801020252, Guidance Project of Science and Technology Research Program of Hubei Provincial Department of Education under Grant No. B2021107. Open Project of State Key Laboratory of Intelligent Manufacturing Equipment and Technology IMETKF2023011. ORCID iD Zhongze Han  https://orcid.org/0009-0004-2953-971X References [1] Cheng C, Liu W, Wang W and Pecht M 2021 A novel deep neural network based on an unsupervised feature learning method for rotating machinery fault diagnosis Meas. Sci. Technol. 32 095013 [2] Xu S, Yuan R, Lv Y, Hu H, Shen T and Zhu W 2023 A novel fault diagnosis approach of rolling bearing using intrinsic feature extraction and CBAM-enhanced InceptionNet Meas. Sci. Technol. 34 105111 [3] Zio E 2022 Prognostics and health management (PHM): where are we and where do we (need to) go in theory and practice Reliab. Eng. Syst. Saf. 218 108119 [4] Yang D, Karimi H R and Gelman L 2023 An explainable intelligence fault diagnosis framework for rotating machinery Neurocomputing 541 126257 [5] Liu Y, Wen W, Bai Y and Meng Q 2023 Self-supervised feature extraction via time–frequency contrast for intelligent fault diagnosis of rotating machinery Measurement 210 112551 [6] Ma S, Han Q and Chu F 2023 Sparse representation learning for fault feature extraction and diagnosis of rotating machinery Expert Syst. Appl. 232 120858 [7] Zhu Z, Lei Y, Qi G, Chai Y, Mazur N, An Y and Huang X 2023 A review of the application of deep learning in intelligent fault diagnosis of rotating machinery Measurement 206 112346 [8] Zou L, Zhuang K J, Zhou A and Hu J 2023 Bayesian optimization and channel-fusion-based convolutional autoencoder network for fault diagnosis of rotating machinery Eng. Struct. 280 115708 [9] Yao J and Han T 2023 Data-driven lithium-ion batteries capacity estimation based on deep transfer learning using partial segment of charging/discharging data Energy 271 127033 [10] Mitici M, de Pater I, Barros A and Zeng Z 2023 Dynamic predictive maintenance for multiple components using data-driven probabilistic RUL prognostics: the case of turbofan engines Reliab. Eng. Syst. Saf. 234 109199 [11] Haixiang G, Yijing L, Shang J, Mingyun G, Yuanyue H and Bing G 2017 Learning from class-imbalanced data: review of methods and applications Expert Syst. Appl. 73 220–39 [12] Jamal M A, Brown M, Yang M H, Wang L and Gong B (eds) 2020 Rethinking class-balanced methods for long-tailed visual recognition from a domain adaptation perspective 2020 IEEE/CVF Conf. on Computer Vision and Pattern Recognition (CVPR) 13–19 June 2020 [13] Zhao M, Fu X, Zhang Y, Meng L and Tang B 2022 Highly imbalanced fault diagnosis of mechanical systems based on wavelet packet distortion and convolutional neural networks Adv. Eng. Inf. 51 101535 [14] Wang X, Jiang H, Wu Z and Yang Q 2023 Adaptive variational autoencoding generative adversarial networks for rolling bearing fault diagnosis Adv. Eng. Inf. 56 102027 [15] Tao X, Guo X, Zheng Y, Zhang X and Chen Z 2023 Self-adaptive oversampling method based on the complexity of minority data in imbalanced datasets classification Knowl.-Based Syst. 277 110795 [16] Liu Y, Liu Y, Yu B X B, Zhong S and Hu Z 2023 Noise-robust oversampling for imbalanced data classification Pattern Recognit. 133 109008 [17] Bowyer K W, Chawla N V, Hall L O and Kegelmeyer W P 2011 SMOTE: synthetic minority over-sampling technique CoRR (arXiv:1106.1813v1) [18] Li J, Zhu Q, Wu Q and Fan Z 2021 A novel oversampling technique for class-imbalanced learning based on SMOTE and natural neighbors Inf. Sci. 565 438–55 14 Meas. Sci. Technol. 35 (2024) 035002 Z Han et al [19] Meng D and Li Y 2022 An imbalanced learning method by combining SMOTE with center offset factor Appl. Soft Comput. 120 108618 [20] Kim J, Kang J and Sohn M 2021 Ensemble learning-based filter-centric hybrid feature selection framework for high-dimensional imbalanced data Knowl.-Based Syst. 220 106901 [21] Hosseini E S and Moattar M H 2019 Evolutionary feature subsets selection based on interaction information for high dimensional imbalanced data classification Appl. Soft Comput. 82 105581 [22] Liu J and Zio E 2019 Integration of feature vector selection and support vector machine for classification of imbalanced data Appl. Soft Comput. 75 702–11 [23] Wang Z, Jia P, Xu X, Wang B, Zhu Y and Li D 2021 Sample and feature selecting based ensemble learning for imbalanced problems Appl. Soft Comput. 113 107884 [24] Chandrashekar G and Sahin F 2014 A survey on feature selection methods Comput. Electr. Eng. 40 16–28 [25] Taherkhani A, Cosma G and McGinnity T M 2018 Deep-FS: a feature selection algorithm for deep Boltzmann machines Neurocomputing 322 22–37 [26] Ahmeto˘glu H and Das¸ R (eds) 2020 Analysis of feature selection approaches in large scale cyber intelligence data with deep learning 2020 28th Signal Processing and Communications Applications Conf. (SIU) 5–7 October 2020 [27] Haueise T, Liebgott A and Yang B (eds) 2022 A comparative study on the potential of unsupervised deep learning-based feature selection in radiomics 2022 44th Annual Int. Conf. IEEE Engineering in Medicine & Biology Society (EMBC) 11–15 July 2022 [28] Lakshmanarao A, Srisaila A and Kiran T S R (eds) 2022 Machine learning and deep learning framework with feature selection for intrusion detection 2022 Int. Conf. on Communication, Computing and Internet of Things (Ic3iot) 10–11 March 2022 [29] Lv H, Chen J, Pan T, Zhang T, Feng Y and Liu S 2022 Attention mechanism in intelligent fault diagnosis of machinery: a review of technique and application Measurement 199 111594 [30] Robnik-ˇSikonja M and Kononenko I 2003 Theoretical and Empirical analysis of ReliefF and RReliefF Mach. Learn. 53 23–69 [31] Maldonado S, Vairetti C, Fernandez A and Herrera F F W 2022 SMOTE: a feature-weighted oversampling approach for imbalanced classification Pattern Recognit. 124 108511 [32] Li Y, Wang Y, Li T, Li B and Lan X S P 2021 SMOTE: a novel space partitioning based synthetic minority oversampling technique Knowl.-Based Syst. 228 107269 [33] Han H, Wang W-Y and Mao B-H (eds) Borderline-SMOTE: a new over-sampling method in imbalanced data sets learning Advances in Intelligent Computing; 2005 2005 (Springer) [34] Haibo H, Yang B, Garcia E A and Shutao L (eds) 2008 ADASYN: adaptive synthetic sampling approach for imbalanced learning 2008 IEEE Int. Joint Conf. on Neural Networks (IEEE World Congress on Computational Intelligence) 1–8 June 2008 [35] Sharma A, Singh P K and Chandra R 2022 SMOTified-GAN for class imbalanced pattern classification problems IEEE Access 10 30655–65 [36] Dablain D, Krawczyk B and Chawla N V 2022 DeepSMOTE: fusing deep learning and SMOTE for imbalanced data IEEE Trans. Neural Netw. Learn. Syst. 34 1–15 [37] Douzas G and Bacao F 2019 Geometric SMOTE a geometrically enhanced drop-in replacement for SMOTE Inf. Sci. 501 118–35 [38] Wang T, Liu Z, Ou W and Huo H 2023 Domain adaptation based on feature fusion and multi-attention mechanism Comput. Electr. Eng. 108 108726 [39] Zheng J, Zhu J and Xi H 2023 Short-term energy consumption prediction of electric vehicle charging station using attentional feature engineering and multi-sequence stacked gated recurrent unit Comput. Electr. Eng. 108 108694 [40] Jiang Y, Chen Y, Yang F and Peng W 2023 State of health estimation of lithium-ion battery with automatic feature extraction and self-attention learning mechanism J. Power Sour. 556 232466 [41] Paul D, Bardhan S, Saha S and Mathew J M L 2023 KnockoffGAN: deep online feature selection for multi-label learning. Knowl.-Based Syst. 271 110548. [42] Ye Z and Yu J 2021 Deep morphological convolutional network for feature learning of vibration signals and its applications to gearbox fault diagnosis Mech. Syst. Signal Process. 161 107984 [43] Zhang K, Tang B, Deng L and Liu X 2021 A hybrid attention improved ResNet based fault diagnosis method of wind turbines gearbox Measurement 179 109491 [44] Zhou H, Huang X, Wen G, Lei Z, Dong S, Zhang P and Chen X 2022 Construction of health indicators for condition monitoring of rotating machinery: a review of the research Expert Syst. Appl. 203 117297 [45] Miao Y, Zhao M, Lin J and Lei Y 2017 Application of an improved maximum correlated kurtosis deconvolution method for fault diagnosis of rolling element bearings Mech. Syst. Signal Process. 92 173–95 [46] Lei Y, Yang B, Jiang X, Jia F, Li N and Nandi A K 2020 Applications of machine learning to machine fault diagnosis: a review and roadmap Mech. Syst. Signal Process. 138 106587 [47] Lei Y, He Z and Zi Y 2011 EEMD method and WNN for fault diagnosis of locomotive roller bearings Expert Syst. Appl. 38 7334–41 [48] Yang B, Lei Y, Jia F and Xing S 2019 An intelligent fault diagnosis approach based on transfer learning from laboratory bearings to locomotive bearings Mech. Syst. Signal Process. 122 692–706 [49] Lei Y, He Z, Zi Y and Hu Q 2007 Fault diagnosis of rotating machinery based on multiple ANFIS combination with GAs Mech. Syst. Signal Process. 21 2280–94 [50] Wang B, Lei Y, Li N, Li N and Hybrid Prognostics A 2020 Approach for estimating remaining useful life of rolling element bearings IEEE Trans. Reliab. 69 401–12 [51] Goodfellow I, Pouget-Abadie J, Mirza M, Xu B, Warde-Farley D, Ozair S, Courville A and Bengio Y (eds) 2014 Generative adversarial nets Neural Inf. Process. Syst. 15
null
10.1371_journal.pone.0255730.pdf
Data Availability Statement: All relevant data are within the manuscript.
All relevant data are within the manuscript.
RESEARCH ARTICLE Glycemic profile and associated factors in indigenous Munduruku, Amazonas Hanna Lorena Moraes GomesID Oliveira Cordeiro1, Zilmar Augusto de Souza Filho1, Noeli das Neves Toledo1, Evelyne Marie Therese MainbourgID 2, Anto´ nio Manuel Sousa3, Gilsirene Scantelbury de Almeida1 1*, Neuliane Melo Sombra1, Eliza Dayanne de 1 Manaus School of Nursing, Federal University of Amazonas, Manaus, Brazil, 2 Leoˆ nidas & Maria Deane Institute / FIOCRUZ Amazoˆ nia, Manaus, Brazil, 3 Amazonas State University, Manaus, Brazil a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 * hannahlorena.mg@gmail.com Abstract Objective OPEN ACCESS Citation: Gomes HLM, Sombra NM, Cordeiro EDdO, Filho ZAdS, Toledo NdN, Mainbourg EMT, et al. (2021) Glycemic profile and associated factors in indigenous Munduruku, Amazonas. PLoS ONE 16(9): e0255730. https://doi.org/10.1371/journal. pone.0255730 Editor: Fernando Guerrero-Romero, Mexican Social Security Institute, MEXICO Received: January 9, 2021 Accepted: July 22, 2021 Published: September 3, 2021 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0255730 Copyright: © 2021 Gomes et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the manuscript. Funding: This study received funding approved by the National Council for Scientific and To evaluate the glycemic profile and its association with sociodemographic, anthropometric, clinical and lifestyle factors of Munduruku indigenous people. Method Cross-sectional study with a quantitative and analytical approach, a total of 459 indigenous people (57.1% men, aged 36.3 ± 14.7 years old) belonging to the Munduruku ethnic group from the Kwata´ -Laranjal Indigenous Land, in Amazonas, Brazil, were selected by probabilis- tic sampling in all households in the four most populous villages. Sociodemographic and anthropometric variables, blood pressure levels and lipid profile were evaluated. Fasting capillary blood glucose was measured with a digital device. The associations were assessed by multinomial logistic regression, and p-values�0.05 were considered significant. Results For pre-diabetes, prevalence was 74.3% and, for diabetes, 12.2%. The variables associated with the risk for pre-diabetes were the following: age (OR = 1.03; 95% CI = 1.00 – 1.06) and obesity (OR = 9.69; 95% CI = 1.28 – 73.58). The positive associations indicating risk for dia- betes were as follows: age (OR = 1.05; 95% CI = 1.03 – 1.08), overweight (OR = 4.17; 95% CI = 1.69 – 10.32) and obesity (OR = 35.26; 95% CI = 4.12 – 302.08). Conclusions The risks associated with pre-diabetes and diabetes among the Munduruku indigenous peo- ple revealed a worrying index. It is necessary to consider changes in eating habits and life- style, as well as possible environmental and social changes that can affect this and other groups, with emphasis on those who live in vulnerable conditions. PLOS ONE | https://doi.org/10.1371/journal.pone.0255730 September 3, 2021 1 / 16 PLOS ONE Technological Development (CNPq) (Proc. 424053 / 2016-0) and with funding from the Scientific Article Publication Support Program (PAPAC) and the Post Support Program -Graduation (PROSGRAD), both of these are programs of the Amazonas Research Support Foundation - FAPEAM. Funders had no role in the study design, data collection, analysis, decision to publish or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Glycemic profile and associated factors in indigenous Munduruku, Amazonas. Introduction The changes in the globalized world, as a result of the urbanization and industrialization pro- cesses, brought about changes in habits and lifestyles, contributing to the increase of chronic non-communicable diseases, among which we can highlight cardiovascular diseases. These same impacts permeate the indigenous populations, through transitions in life, economic and sociocultural habits, and in their own lifestyle [1, 2]. The destruction of the ecosystems that the Brazilian Indigenous Lands are facing, together with the acceleration of the urbanization process, sedentary lifestyle, changes in the diet, obe- sity and easy access to cities, contribute significantly to the transformations of the daily lives of indigenous populations, leaving them more vulnerable to certain diseases, which contributes to the increase of Chronic Non-communicable Diseases (CNCDs) [3, 4]. Social indicators of a national scope classify the North Region as belonging to Class “E” of social vulnerability, as it consists of extensive rural areas, low demographic density, with a very low human development index, precarious access to treated water, sewage and electricity, among other negative results. Compared with the South and Southeast regions of the country, the North has less capacity to respond to health problems, in terms of Health Care Network structure [5]. Deaths due to non-communicable diseases (NCDs) represented the highest percentage: 73.4% (95% uncertainty interval [UI] = 72.5 – 74.1) in 2017. In relation to 2007, there was a 22.7% (21.5 – 23.9) increase, equivalent to 7.21 million (7.20 – 8.01) of estimated additional deaths. There was a major increase in years of life lost due to neoplasms and cardiovascular diseases. In the general population, cardiovascular diseases (CVDs) are part of the group of main causes of mortality. In 2016, approximately 17.9 million people died due to CVDs worldwide. From this perspective, diabetes mellitus (DM) stands out as a highly prevalent health problem and one of the main risk factors for CVDs [6–8]. DM is configured as a "metabolic disorder" characterized by persistent hyperglycemia, resulting from a deficit in the production of insulin or in its action, or even in both mecha- nisms, leading to long-term complications” (SBD, pg. 19). Data from the International Diabe- tes Federation point out that, in the world, 8% of adults lived with DM in 2017. DM is a growing and important health problem that affects the population of all countries, being responsible for 4 million deaths worldwide in a single year [9, 10]. It is believed that changes in the social, economic and political scopes of indigenous Brazi- lians may have favored changes in their lifestyle and in their epidemiological profile [1]. In the Brazilian indigenous population, the first cases of DM began to be investigated from the 1970s, when the prevalence of diabetes was non-existent [1]. In the state of Mato Grosso do Sul, several studies were carried out with the Terena, Gua- rani and Kaiowa´ indigenous peoples, where it was found that 4.5% had DM in 2007 and 2008 [11]. Another two studies carried out in the same population found a prevalence rate of 5.8% in the period from 2009 to 2011, and of 4.5% in 2008 and 2009 [12]. In 2013, among 385 Ter- ena and Guarani women from the same region, 7% presented altered capillary glycaemia sug- gestive of DM [13]. Among the Guarani and Tupinikin (ES), in 2003 and 2004, the prevalence of DM was 4.5% [14]. In Khisêdjê in 2010 and 2011, prevalence was 3.8% [15]. The highest prevalence rate of DM among indigenous people in Brazil was found among the Xavante in the state of Mato Grosso (n = 948): 25.9% [16]. The data presented show that diabetes has been growing in indigenous populations [17] and that is worsened by the increased consumption of industrialized food products, social problems linked to the economy and the increasingly frequent contact with the non- PLOS ONE | https://doi.org/10.1371/journal.pone.0255730 September 3, 2021 2 / 16 PLOS ONE Glycemic profile and associated factors in indigenous Munduruku, Amazonas. indigenous population [1, 17]. Considering that most of the studies refer to ethnicities in the Brazilian Midwest Region, the objective of the study was to assess the glycemic profile and its association with sociodemographic, anthropometric, clinical and lifestyle factors of Mundur- uku indigenous people from the state of Amazonas, Brazilian North Region. Method Study locus and population The study was carried out in the Kwata´-Laranjal indigenous land (Fig 1), located in the munic- ipality of Borba, state of Amazonas, in the Brazilian North Region. The study population con- sisted of individuals from the Munduruku ethnic group who live in the villages of Laranjal, Mucaja´, Kwata´ and Fronteira, members of the Kwata´-Laranjal Indigenous Land, aged between 18 and 80 years old, and of both genders. According to population data, released by the Special Indigenous Sanitary District of Manaus in 2018, the total population over 18 years old of both genders living in these four villages consisted in 635 inhabitants, divided as follows: 195 in Mucaja´, 118 in Laranjal, 186 in Kwata´ and 136 in Fronteira. Fig 1. Geographic location of the Kwata´-Laranjal Indigenous Land. https://doi.org/10.1371/journal.pone.0255730.g001 PLOS ONE | https://doi.org/10.1371/journal.pone.0255730 September 3, 2021 3 / 16 PLOS ONE Glycemic profile and associated factors in indigenous Munduruku, Amazonas. Study participants The following was accepted for sample calculation: 50.0% proportion of the indigenous popu- lation and the prevalence values of diabetes pointed out by the Guidelines of the Brazilian Dia- betes Society and by the study by Soares et al. [9, 18]. The error margin adopted was 5%, 95% confidence interval, and 10% for losses. The sample consisted of 459 individuals belonging to the Munduruku ethnicity, from the villages of Mucaja´ (n = 129), Laranjal (n = 93), Kwata´ (n = 136) and Fronteira (n = 101). The four most populous villages in the Kwata´-Laranjal Indigenous Land (Mucaja´, Laranjal, Kwata´ and Fronteira) were chosen. Probabilistic sampling of individuals per household was carried out, in which all members had an equal chance of participating in the study. The study included indigenous people belonging to the Munduruku ethnic group, aged � 18 years old and living in the selected villages. It is noted that all the Munduruku indigenous people drawn to participate in this study were able to fluently communicate in the Portuguese language. Only those who were ill and pregnant were excluded from the sample. Data collection Before starting data collection in the Kwata´-Laranjal Indigenous Land, the team of women researchers visited the four villages included in this study, which allowed for previous contact with the local indigenous leaders, closer contact with the health professionals who served in those villages, and holding a meeting with the indigenous people to present the research objec- tives and method. For the data collection stage, the team underwent specific training in order to standardize the procedures for: measuring blood glucose and capillary lipids after fasting for a minimum of eight hours, measuring blood pressure, taking anthropometric measurements and conduct- ing the interview. At the beginning of data collection, the residents were invited again to be informed about how the participants would be selected and the procedures for data collection. For each house- hold, the research participants were selected by means of a draw. The indigenous health agent assisted the team in locating the homes of the selected participants. The guidelines for data col- lection were given the day before, with reinforcement regarding the location, day, time and, mainly, the need for at least 8-hour fasting. The collection of the anthropometric data, blood pressure, blood glucose and lipids was always performed at dawn. Before starting the collection of blood drops from the digit pulp, the indigenous people were asked at what time they had their last meal. Those who reported breaking the fast were rescheduled for the following day and re-oriented. In relation to the tests of capillary blood glucose and lipid levels, the equipment used were as follows: Active portable digital device from the Accu-Chek1 manufacturer for the measure- ment of capillary blood glucose and the Accutrend1 Plus device for the measurement of cho- lesterol and triglycerides, both manufactured by Roche Diagno´stica, with their respective test strips (Accutrend1 Cholesterol and Accutrend1 Triglycerides). The cut-off points used to assess and classify fasting capillary glucose were as follows: normal < 100 mg/dL, pre- diabetes � 100 mg/dL and < 126 mg/dL and diabetes � 126 mg/dL [9]. For the lipid levels, the classification was the following: hypercholesterolemia when � 240 mg/dL and hypertriglyceri- demia when � 175 mg/dL [19]. The following was used for the evaluation of the anthropometric measures: digital bioimpe- dance scale (OMRON HBF-514C), portable stadiometer (ALTURA EXATA) and inelastic measuring tape. The neck circumference measurement was taken at the smallest neck circum- ference, just above the laryngeal prominence. The waist circumference measurement was PLOS ONE | https://doi.org/10.1371/journal.pone.0255730 September 3, 2021 4 / 16 PLOS ONE Glycemic profile and associated factors in indigenous Munduruku, Amazonas. taken at the midpoint between the last rib and the lateral iliac crest, around the narrowest part of the trunk. The taper index was determined, according to its definition, from the measure- ments of weight, height and waist circumference. Both BMI and Body Fat Percentage were assessed using the bioimpedance technique. The cut-off points adopted to classify neck circumference measurements were as follows: � 37 cm for men and � 34 cm for women; and those for waist circumference were: � 102 cm for men and � 88 cm for women [20]. For the taper index, the adopted values were: � 1.25 for men and � 1.18 for women. As for the Body Mass Index (BMI), it was classified as: low weight (< 18.5 kg/m2), normal weight (18.5 kg/m2-24.9 kg/m2), overweight (25.0 kg/m2-29.9 kg/m2) and obesity (� 30.0 kg/m2) [21]. The classification of body fat percentage considered the fol- lowing stratification by age group and gender: low (< 8.0%-< 13.0% for men and < 21.0%-< 30.0% for women), normal (13.0%-24.9% for men and 30.0%-� 35.9% for women) and high (� 25.0% for men and � 36.0% for women). Blood pressure levels were measured on the left arm, using an automatic professional blood pressure monitor (OMRON/Model HBP-1100), properly calibrated. The procedures to per- form the measurement and classification of blood pressure were conducted according to the Brazilian Hypertension Directive. The following cut-off points were considered: pre-hyperten- sion when systolic blood pressure levels are between 140 mmHg and 159 mmHg and/or when the diastolic blood pressure is between 90 mmHg and 99 mmHg; hypertension when the value is � 180 mmHg in systolic pressure and/or � 110 mmHg in diastolic pressure. Alternatively, hypertension could be self-reported, if the indigenous participants reported having been diag- nosed with hypertension by a physician or if they were taking some antihypertensive medica- tion, regardless of the blood pressure values measured in the interview [22]. For the assessment of lifestyle, the level of physical activity was investigated using the IPAC (International Physical Activity Questionnaire), in its short version, an instrument validated with translation into the Portuguese language. The IPAQ allows quantifying the total minutes spent in weekly physical activities and surveying the distribution of time by intensity of the physical activity practiced. The level of physical activity was classified according to the instru- ment. To assess the intake of alcoholic beverages, the Alcohol Use Disorder Identification Test (AUDIT) questionnaire was used, allowing the identification of risk and harmful consumption and of probable dependence on alcohol in the past 12 months. A form consisting of closed questions related to the following variables was applied: gender, age, marital status, schooling, paid work, social benefit received, monthly family income, self- reported hypertension and/or consumption of antihypertensive medications, smoking, level of physical activity, alcohol consumption and family history of cardiovascular diseases. The participants who presented changes in capillary glycaemia, triglycerides, total choles- terol or/and blood pressure, as well as those who were obese were referred directly to the care provided by the health team at the Base Center (reference health unit, belonging to the Indige- nous Health Sub-System) for evaluation and monitoring. For the changes in the anthropomet- ric markers, this information was passed on to the health professionals working in the respective Base Center. Statistical analysis The analysis of the data collected was performed by means of the R software, version 3.5.1. The Kolmogorov-Smirnov test was used to verify normal distribution of the data. In this way, the continuous variables were presented using means and standard deviations; and the categorical variables, with absolute and relative frequencies. For the continuous variables, the Kruskal- Wallis test was used; and for categorical ones, Fisher’s Exact test. The significance level was set PLOS ONE | https://doi.org/10.1371/journal.pone.0255730 September 3, 2021 5 / 16 PLOS ONE Glycemic profile and associated factors in indigenous Munduruku, Amazonas. at 5%. The Wald test was used for the multinomial logistic regression analysis. To verify the association between the dependent variables (pre-diabetes and diabetes) and the independent variables of the study, Odds Ratios (ORs) were estimated based on the multinomial regression model and the respective 95% confidence interval (CI). For this being a multifactorial phe- nomenon, the independent variables were grouped in blocks (sociodemographic, lifestyle and anthropometric and clinical factors) and analyzed hierarchically. Ethical aspects The data were collected from August to September 2018, after the consent of the leaders of the Kwata´-Laranjal Indigenous Land, approval by the National Research Ethics Commission (CAAE 74361617.2.0000.5020), and authorization for entry into indigenous lands of the Min- istry of Justice National Indian Foundation (43/AAEP/PRES/2018). All the indigenous people who agreed to participate in the study signed the Free and Informed Consent Form. Results As shown in Table 1, the profile of the glycemic levels of the 459 indigenous Munduruku indi- cates that 86.5% had high serum levels of fasting capillary glycaemia, with 74.3% being sugges- tive of pre-diabetes and 12.2% of diabetes. As for the sociodemographic factors, it was observed that 57.1% were men, with a mean age of 36.6 years old, most with a partner, and 9.6% not having any schooling level. A little over half of them were unemployed and 61.7% received some social benefit from the Brazilian fed- eral government. In this way, most of the Munduruku indigenous people had a monthly family income of up to US$ 470.67. The general anthropometry assessment allowed identifying that the indigenous people had high mean values of neck circumference, waist circumference and taper index. The mean BMI indicated excess weight, in addition to the majority presenting high body fat percentages. In relation to the clinical factors of the Munduruku indigenous people, the mean pressure levels indicated normality, but 10.2% presented high levels of systolic and diastolic blood pres- sure, suggestive of hypertension. Regarding the serum triglyceride levels, the indigenous popu- lation presented a high mean value but, for total cholesterol, the mean remained within normal limits. Regarding the indigenous people’s lifestyle, there was a high prevalence of alcohol con- sumption (71.2%) and smoking (54.2%), as well as a low prevalence of sedentary lifestyle (7.6%). It is worth mentioning that most of the indigenous people reported having a family his- tory of hypertension and diabetes. Table 1 also shows that the group of Munduruku indigenous people with diabetes presented statistically significant differences when compared to the other groups, in greater proportion having some paid work and, in a smaller proportion, receiving some social benefit. The group of diabetics presents higher values regarding age, mean in the anthropometric markers, preva- lence of obesity and body fat, prevalence of pre-hypertension and hypertension, mean of tri- glycerides and total cholesterol, as well as family members with diabetes or/and stroke. Table 2 shows the unadjusted multinomial logistic regression model. The association of pre-diabetes with age showed that, for every one-year-old increase in the age of the indigenous Munduruku, their chance of becoming pre-diabetic increases by 4%. It is also worth noting that the indigenous people without a partner had a lower risk of being pre-diabetic (OR = 0.55 [95% CI = 0.32 – 0.96]). As for the association of pre-diabetes with the anthropometric factors, it was observed that, with a one-centimeter increase in waist circumference (OR = 1.07 [95% CI = 1.03 – 1.10]), in PLOS ONE | https://doi.org/10.1371/journal.pone.0255730 September 3, 2021 6 / 16 PLOS ONE Table 1. Categorization of the glycemic profile of indigenous Munduruku according to the sociodemographic and anthropometric variables, clinical factors, habits and lifestyle, and family history. Glycemic profile and associated factors in indigenous Munduruku, Amazonas. Variables Sociodemographic Factors Gender Female Male Age (years old), mean (SD) Marital Status Has a partner No partner Schooling Illiterate Elementary School High School Higher Education or Postgraduate Paid work Yes No Social benefit Yes No Monthly family income (minimum wagea) Does not have <1 minimum wage (US$ 235.34) 1 - 2 minimum wages (US$ 235.35 – US$ 470.67) 3 - 4 minimum wages (US$ 706.01 – US$ 941.35) � 5 minimum wages (US$ 1,176) Anthropometric Factors Neck circumference (cm), mean (SD) Waist circumference (cm), mean (SD) Taper index, mean (SD) BMI (kg/m2), mean (SD) BMI classification Low weight (< 18.5 kg/m2) Normal weight (18.5–24.9 kg/m2) Overweight (25.0–29.9 kg/m2) Obesity (�30 kg/m2) Body fat classification Low (<8.0%-<13.0% men/<21.0%-<30.0% women) Normal (13.0%-24.9% men/30.0%-�35.9% women) High (�25.0% men; �36.0% women) Clinical Factors Systolic blood pressure, SBP (mmHg), mean (SD) Diastolic blood pressure, DBP (mmHg), mean (SD) Blood pressure classification Normal N (%) 62 (13.5) Glycemic Profile Pre-diabetes N (%) 337 (74.3) Diabetes N (%) 60 (12.2) Total N (%) 459 (100) 22 (35.5) 40 (64.5) 147 (43.6) 190 (56.4) 28 (46.7) 32 (53.3) 197 (42.9) 262 (57.1) 30.2 (±11.2) 36.5 (±14.8) 44.1 (±14.0) 36.6 (±14.7) 35 (56.5) 27 (43.5) 3 (4.8) 19 (30.6) 30 (48.4) 10 (16.1) 22 (35.5) 40 (64.5) 43 (69.4) 19 (30.6) 21 (26.2) 30 (37.5) 22 (27.5) 6 (7.5) 1 (1.3) 35.5 (±3.3) 79.5 (±7.9) 1.20 (±0.08) 23.6 (±2.8) 1 (1.6) 42 (67.7) 18(29.0) 1 (1.6) 1 (1.6) 35 (56.5) 26 (41.9) 236 (70.0) 101 (30.0) 30 (8.9) 134 (39.8) 128 (38.0) 45 (13.4) 138 (40.9) 199 (59.1) 211 (62.6) 126 (37.4) 7 (2.2) 134 (41.7) 115 (35.8) 49 (15.3) 16 (5.0) 36.2 (±3.3) 85.1 (±10.1) 1.24 (±0.09) 25.7 (±4.0) 4 (1.2) 158 (46.9) 127 (37.7) 48 (14.2) 7/337 (2.1) 126 (37.4) 204 (60.5) 41 (68.3) 19 (31.7) 11 (18.3) 22 (36.7) 19 (31.7) 8 (13.3) 34 (56.7) 26 (43.3) 29 (48.3) 31 (51.7) 1 (1.7) 21 (36.2) 24 (41.4) 11 (19.0) 1 (1.7) 312 (68.0) 147 (32.0) 44 (9.6) 175 (38.1) 177 (38.6) 63 (13.7) 194 (42.3) 265 (57.7) 283 (61.7) 176 (38.3) 29 (6.3) 185 (40.3) 161 (35.1) 66 (14.4) 18 (3.9) 37.7 (±3.2) 92.2 (±8.7) 1.29 (±0.07) 28.0 (±3.6) 36.3 (±3.3) 85.3 (±10.2) 1.24 (±0.09) 25.8 (±4.0) 0 (0) 12 (20.0) 31 (51.7) 17 (28.3) 0 (0) 7 (11.7) 53 (88.3) 5 (1.1) 212 (46.2) 176 (38.3) 66 (14.4) 8 (1.7) 168 (36.6) 283 (61.7) p-value 0.404 <0.001 0.109 0.116 0.039 0.045 0.708 0.001 <0.001 <0.001 <0.001 <0.001 < 0.001 <0.001 110.0 (±12.2) 63.6 (±8.2) 113.6 (±15.0) 66.5 (±8.4) 121.2 (±16.9) 70.4 (±8.8) 114.1 (±15.2) 66.6 (±8.6) <0.001 <0.001 0.001 (Continued ) PLOS ONE | https://doi.org/10.1371/journal.pone.0255730 September 3, 2021 7 / 16 PLOS ONE Table 1. (Continued) Variables Normal (SBP of �120–129 mmHg/DBP �80–84 mmHg) Pre-hypertension (SBP of �130 mmHg-139 mmHg/BPD �80–89 mmHg) Hypertension (SBP of �140 mmHg/DBP �90 mmHg) Triglycerides (mg/dL) Total cholesterol (mg/dL) Lifestyle Smoker Yes No Level of physical activity Sedentary Irregularly active Active Very active Alcohol Consumption Low risk consumption Risk intake, harmful or probable dependence Family history Hypertension Diabetes Stroke Glycemic profile and associated factors in indigenous Munduruku, Amazonas. Normal N (%) 62 (13.5) 59 (95.2) 1 (1.6) 2 (3.2) Glycemic Profile Pre-diabetes N (%) 337 (74.3) 294 (87.2) 12 (3.6) 31 (9.2) Diabetes N (%) 60 (12.2) 40 (66.7) 6 (10.0) 14 (23.3) Total N (%) 459 (100) 393 (85.6) 19 (4.1) 47 (10.2) p-value 131.9 (±65.8) 171.3 (±25.5) 149.3 (±86.3) 176.5 (±32.3) 206.8 (±124.1) 189.7 (±35.4) 154.5 (±92.1) 177.5 (±32.2) <0.001 0.003 36 (58.1) 26 (41.9) 2 (3.2) 13 (21.0) 21 (33.9) 26 (41.9) 7 (25.9) 20 (74.1) 43 (84.3) 31 (63.3) 20 (45.5) 184 (54.6) 153 (45.4) 26 (7.7) 92 (27.3) 114 (33.8) 105 (31.2) 35 (30.2) 81 (69.8) 211 (77.0) 161 (61.7) 75 (31.6) 29 (48.3) 31 (51.7) 7 (11.7) 18 (30.0) 24 (40.0) 11 (18.3) 4 (23.5) 13 (76.5) 43 (86.0) 38 (82.6) 19 (52.8) 249 (54.2) 210 (45.8) 35 (7.6) 123 (26.8) 159 (34.6) 142 (30.9) 46 (28.8) 114 (71.2) 297 (79.2) 230 (64.6) 114 (36.0) 0.542 0.125 0.800 0.222 0.023 0.018 Kwata´-Laranjal Indigenous Land, Borba, Amazonas, Brazil, 2018. a Current minimum wage of R$ 954.00, equivalent to approximately US$ 235.34 in August 2018. https://doi.org/10.1371/journal.pone.0255730.t001 the taper index (OR = 1.06 [95% CI = 1.02 – 1.09]) and in the BMI (OR = 1.20 [95% CI = 1.10 – 1.32]), the indigenous people have a chances to develop pre-diabetes of 7%, 6% and 20%, respectively. Excess weight among the indigenous people also presented an associa- tion with pre-diabetes, since the chance of the indigenous person who presented overweight to become pre-diabetic is 87%; and, among those who were obese, the chance becomes 12 times greater (OR = 12.76 [95% CI = 1.71 – 95.26]). For the indigenous people with high body fat, the risk of becoming pre-diabetics also increases the chance, but two-fold (OR = 2.18 [95% CI = 1.25 – 3.79]). The unadjusted analysis also indicated the association of diabetes with age, schooling, paid work and any social benefits received. All the anthropometric variables were associated with diabetes among the indigenous people. It is worth noting that, among the Munduruku indige- nous people who presented overweight (OR = 6.17 [95% CI = 2.60 – 4.64]) and obesity (OR = 61.03 [95% CI = 7.34 – 507.08]), the chances increased significantly. The clinical factors were also associated with diabetes, such as: pre-hypertension, hypertension and an increase in the total serum cholesterol level. On the other hand, the fact of having a Very Active level of physical activity (OR = 0.12 [95% CI = 0.02 – 0.68]) reduces by 88% the chance of the indige- nous Munduruku developing diabetes. Table 3 shows the Odds Ratio adjusted for gender and age of the variables that presented statistical significance (p�0.05) in the analyses from Table 2, considering the two outcomes PLOS ONE | https://doi.org/10.1371/journal.pone.0255730 September 3, 2021 8 / 16 PLOS ONE Table 2. Unadjusted odds ratio and Confidence Interval (CI) for sociodemographic and anthropometric variables, clinical factors, lifestyle and family history asso- ciated with pre-diabetes and diabetes among the Munduruku indigenous people. Glycemic profile and associated factors in indigenous Munduruku, Amazonas. Variables Sociodemographic Factors Gender (Ref. Female) Male Age (years old) Marital Status (Ref. Has a partner) Without partner Schooling (Ref. Illiterate) Elementary School High School Higher Education or Postgraduate Paid work (Ref. Yes) No Social benefits (Ref. Yes) No Monthly family income (Ref. Does not have) <1 minimum wage (US$: 235.34) 1–2 minimum wages (US$: 235.35–470.67) 3–4 minimum wages (US$: 706.01–941.35) � 5 minimum wages (US$: 1,176) Anthropometric Factors Neck circumference (cm) Waist circumference (cm) Taper index BMI (kg/m2) BMI classification (Ref. Low weight/Normal weight Overweight Obesity Body Fat Classification (Ref. Normal) Low High Clinical Factors Systolic blood pressure Diastolic blood pressure Blood Pressure Classification (Ref. Normal) Pre-hypertension Hypertension Triglycerides Total cholesterol Habits and lifestyle Smoker (Ref. No) Yes Physical activity level (Ref. Sedentary) Irregularly active Active Pre-Diabetes vs Normal Gross OR (95% CI) 0.71 (0.40–1.25) 1.04 (1.02–1.07) 0.55 (0.32–0.96) 0.71 (0.20–2.54) 0.43 (0.12–1.49) 0.45 (0.11–1.77) 0.79 (0.45–1.39) 1.35 (0.75–2.42) 1.28 (0.25–6.45) 1.49 (0.29–7.67) 2.33 (0.39–13.91) 4.57 (0.35–59.12) 1.06 (0.98–1.15) 1.07 (1.03–1.10) 1.06 (1.02–1.09) 1.20 (1.10–1.32) 1.87 (1.03–3.40) 12.76 (1.71–95.26) 1.96 (0.23–1.65) 2.18 (1.25–3.79) 1.02 (1.00–1.04) 1.04 (1.01–1.08) 2.41 (0.31–18.88) 3.11 (0.72–13.35) 1.00 (1.00–1.01) 1.01 (1.00–1.02) 1.15 (0.67–1.99) 0.54 (0.12–2.57) 0.42 (0.09–1.89) p-value 0.235 0.002 0.037 0.593 0.182 0.253 0.420 0.311 0.768 0.631 0.353 0.245 0.164 <0.001 0.002 <0.001 0.040 0.013 0.538 0.006 0.065 0.013 0.403 0.127 0.122 0.222 0.616 0.442 0.258 Diabetes vs Normal Gross OR (95% CI) 0.63 (0.30–1.30) 1.07 (1.04–1.11) 0.60 (0.29–1.26) 0.32 (0.08–1.30) 0.17 (0.04–0.70) 0.22 (0.04–1.06) 0.42 (0.20–0.87) 2.42 (1.15–5.07) 1.40 (0.12–16.47) 2.18 (0.18–25.78) 3.67 (0.27–49.30) 2.00 (0.05–78.31) 1.22 (1.09–1.37) 1.14 (1.10–1.19) 1.14 (1.09–1.19) 1.38 (1.24–1.53) 6.17 (2.60–14.64) 61.03 (7.34–507.08) - 10.19 (3.99–26.00) 1.05 (1.02–1.08) 1.10 (1.05–1.15) 8.85 (1.03–76.36) 10.32 (2.22–47.92) 1.01 (1.00–1.01) 1.02 (1.01–1.03) 1.48 (0.72–3.02) 0.40 (0.07–2.22) 0.33 (0.06–1.75) p-value 0.211 <0.001 0.177 0.111 0.014 0.059 0.020 0.019 0.789 0.536 0.327 0.711 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 - <0.001 <0.001 <0.001 0.047 0.003 <0.001 0.004 0.283 0.292 0.191 (Continued ) PLOS ONE | https://doi.org/10.1371/journal.pone.0255730 September 3, 2021 9 / 16 PLOS ONE Glycemic profile and associated factors in indigenous Munduruku, Amazonas. Table 2. (Continued) Variables Very active Pre-Diabetes vs Normal Gross OR (95% CI) 0.31 (0.07–1.39) Consumption of Alcohol Beverages (Ref. Low risk consumption) Risk intake, harmful or probable dependence 1.10 (0.46–2.60) Family History Hypertension (Ref. No) Yes Diabetes (Ref. No) Yes Stroke (Ref. No) Yes 0.62 (0.28–1.39) 0.93 (0.50–1.76) 0.56 (0.29–1.07) Kwata´-Laranjal Indigenous Land, Borba, Amazonas, Brazil, 2018. https://doi.org/10.1371/journal.pone.0255730.t002 p-value 0.127 0.831 0.250 0.834 0.078 Diabetes vs Normal Gross OR (95% CI) 0.12 (0.02–0.68) 1.78 (0.49–6.43) 1.14 (0.38–3.43) 2.76 (1.06–7.19) 1.34 (0.55–3.24) p-value 0.016 0.378 0.812 0.038 0.515 (pre-diabetes and diabetes). Thus, it is noteworthy that pre-diabetes was associated with increasing age, BMI and obesity. And diabetes remained associated with increasing age, BMI, overweight and obesity. Discussion The prevalence of diabetes among the Munduruku indigenous people (12.2%) was higher than that found in other studies with indigenous populations, such as the Guarani, Kaiowa´ and Ter- ena, from Dourados (Mato Grosso do Sul) (4.5%), Aymara, in Chile (1.5%) and was lower when compared to the Xavante indigenous people (25.9%) from Mato Grosso and to the Pima indigenous people from the state of Arizona (USA) [11, 16, 23, 24]. The largest participation in the study corresponded to the male gender (57.1%), unlike studies on cardiovascular risk carried out with other indigenous populations, such as: Xavante (49.2%) [18], Mura (42.2%) [25], Guarani-Kaiowa´ and Terena (44.2%) [11]. The mean age revealed that the Munduruku indigenous people were young adults: 36.6 years old (±14.7). A number of studies indicate that age is an important indicator for cardio- vascular risk factors, especially for diabetes [18, 26, 27]. This study revealed that age presented a positive and significant association with the glycemic profile and, under this perspective, a study carried out with the Terena and Guarani indigenous peoples in 2016 also presented the same association [13]. Table 3. Odds ratio adjusted for gender and age and confidence interval (CI) for sociodemographic and anthropometric variables, clinical factors, habits and life- style and family history associated with pre-diabetes and diabetes among the Munduruku indigenous people. Variables Age (years old) BMI (kg/m2) Overweight Obesity Pre-Diabetes vs Normal Adjusted OR (95% CI) 1.03 (1.00–1.06) 1.16 (1.06–1.28) 1.48 (0.79–2.77) 9.26 (1.22–70.45) p-value 0.032 0.002 0.226 0.032 Diabetes vs Normal Adjusted OR (95% CI) 1.05 (1.02–1.08) 1.28 (1.14–1.43) 4.07 (1.65–10.04) 29.14 (3.38–251.04) p-value 0.004 <0.001 0.002 0.002 Kwata´-Laranjal Indigenous Land, Borba, Amazonas, Brazil, 2018. https://doi.org/10.1371/journal.pone.0255730.t003 PLOS ONE | https://doi.org/10.1371/journal.pone.0255730 September 3, 2021 10 / 16 PLOS ONE Glycemic profile and associated factors in indigenous Munduruku, Amazonas. In relation to the socioeconomic conditions, the findings show a high proportion of low- income individuals: 46.6% with a family income of less than US$ 235.34, while 57.73% of the participants had no paid work and 61.66% were receiving social benefits from the Brazilian federal government. A study carried out with Mura de Autazes indigenous people (Amazonas) also revealed that 60.2% received income from some social benefits program of the Brazilian federal government and 59.4% had a family income of less than US$ 237.00 [25]. Another study carried out with the Guarani-Kaiowa´ and Terena indigenous peoples from Dourados (Mato Grosso do Sul) presented a percentage of 84.2% of families benefited by the Bolsa Famí- lia program, highlighting the conditions of social vulnerability experienced by the group and the possibility of social benefits improving the living conditions of the indigenous people [28]. In this context, it is worth noting that the Munduruku indigenous population presented risk for diabetes associated with low income. The anthropometric data presented significant differences, revealing higher mean values among the indigenous people classified as diabetic compared to pre-diabetics and to those with normal blood glucose. For the Body Mass Index, the global mean revealed excess weight [25.8 (±4.0) kg/m2] among the Munduruku indigenous people, 38.3% of them with overweight and 14% with obe- sity. A study carried out in 2016 with the Mura de Autazes indigenous people (Amazonas), showed excess weight, with a BMI of 26.6 (±4.7) kg/m2 [25]. Similar results were found among the indigenous women from the municipality of Dourados (Mato Grosso do Sul), who pre- sented a mean BMI of 27.8 (±5.0) kg/m2 [13]. When it comes to the Xavante Indigenous Reserves of São Marcos and Sangradouro, in the municipality of Volta Grande (Mato Grosso), the mean BMI indicates obesity among these indigenous people [30.3 (±5.1) kg/m2] [18]. Overweight and obesity are worrisome conditions, as they increase the risk of developing car- diovascular diseases [18]. Among the Munduruku considered diabetic, the percentage of obesity was 28.3%. Flor et al. showed that, in 2008, the percentage attributable to obesity associated with diabetes melli- tus was, for men, 37.3% in the Brazilian North Region against 45.4% in the entire country; and, for women, 55.1% in the North Region against 58.3% throughout Brazil, and the Brazilian mean was higher than the mean values found in the international literature [29]. When it comes to indigenous peoples, data for comparative analysis between diabetes and neck circumference are scarce. In our study, the mean neck circumference was 36 cm (±3.3), slightly below the national mean for the Brazilian male population (39.5±3.6) and slightly above the national mean for the Brazilian female population (34.0±2.9) [30]. In relation to other ethnic groups, such as Asian groups living in different cultural contexts, the mean found was 33 cm (±4.16), indicating that the increase in fat in the neck region had a greater indica- tion of cardiometabolic disease when compared to the increase in the body and visceral mass index [31]. Another two studies conducted with the general American population suggest that increased neck circumference was associated with hypertension, diabetes, metabolic syndrome and dyslipidemia [32, 33]. In relation to waist circumference, the Munduruku presented a lower mean [85.3 cm (±10.2)], when compared to the Xavante indigenous people (Mato Grosso) [95.1 (±8.3) [34], but higher when compared to the Yanomami (Roraima) [76.3 (±46.8)] [35]. With regard to the Taper Index, the mean was 1.24 (± 0.9) among the Munduruku, similar to the one found among the Mura (municipality of Autazes, Amazonas) [1.27 (±0.08)] [25]. However, these findings are much lower when compared to the mean of the Brazilian popula- tion that varies between 1.35 (±0.08) and 1.34 (±0.09) [36, 37]. In relation to the blood pressure levels, the results of this study show that the prevalence of people with blood pressure levels suggestive of arterial hypertension was 10.2%. A systematic PLOS ONE | https://doi.org/10.1371/journal.pone.0255730 September 3, 2021 11 / 16 PLOS ONE Glycemic profile and associated factors in indigenous Munduruku, Amazonas. review study with meta-analysis and meta-regression, conducted with indigenous people from the North Region (Ianomaˆmi, Suruı´, Tembe´, Amondaua, Parkatêjê, Suruı´), from the Midwest Region (Terena, Zoro´, Suya´, Kalapalo, Kuikuro, Matipus, Nahukwa´, Mehina´ku, Waura´, Yawa- lapitı´, Guaranı´, Tupinikin, Xavante, Khisêdjê and indigenous people from the Jaguapiru vil- lage), and from the Southeast and South Regions (Guaranı´-Mbya´, Kaingang), showed a 12% increase in the chance of hypertension, in any indigenous person in Brazil, for each year stud- ied [38]. The meta-analysis of this study showed that there was an increase in the prevalence of arterial hypertension, since in 1970 it was non-existent in the indigenous population, 0.1% (95% CI = 0.0% – 0.6%), when compared to 2014, when the highest prevalence of arterial hypertension was identified: 29.7% (95% CI = 26.1% – 44.4%) [38]. A study that investigated cardiovascular risk factors among different ethnic groups, living in the same urban area of Manaus (Amazonas), identified that, although the prevalence of SAH among the indigenous people was lower than in white-skinned (62.5%) and brown-/ black-skinned (60.7%) individuals, that for pre-hypertension and hypertension was 28.6% among the Satere´-Mawe´ and 46.5% among ethnic groups from the upper Rio Negro [39]. During the assessment of the lipid levels, this study presented a mean of triglycerides of 165.5 (±86.5) mg/dL. In turn, 21.1% of the participants had high levels of triglycerides. These data are similar to those of the Mura de Autazes indigenous people (Amazonas) [163.5 (±104.7) mg/dL] [25] and Xavante of the São Marcos and Sangradouro Indigenous Reserves (Mato Grosso) [199.1 (±171.2) mg/dL] [18], differing from the mean among the Guarani- Mbya´ indigenous people (Rio de Janeiro), which was 116.0 (±74.9) mg/dL [3]. Regarding the total cholesterol levels, the mean was 177.5 (± 32.2) mg/dL, considered within the boundary range and indicating that the Munduruku presented higher levels when compared to other ethnicities, such as the Sangradouro and the Guarani-Mbya´ indigenous peoples, whose mean total cholesterol values were 145.8 (±4.7) mg/dL [16] and 143.8 (±28.8) mg/dL, respectively [3]. In relation to the diabetics indigenous individuals, 82.6% of them reported having a family history of diabetes and 52.8%, a family history of stroke. Indigenous people under the age of 55, who live in remote areas of Australia, were 14 times more likely to have an ischemic stroke, when compared to non-indigenous people belonging to the same age group. It is worth men- tioning that the prevalence of diabetes found was 70.3% among indigenous people versus 34% among non-indigenous people [40]. With regard to the findings obtained through Odds Ratio adjusted for gender and age, it is possible to assert an increase in the chance of developing Pre-diabetes and Diabetes in relation to age in the group under study. Australian indigenous peoples had a 7% chance of developing diabetes each year of life [41]. A similar percentage was identified among the Munduruku, in which at each one year of life increase, there is a 3% chance of having pre-diabetes, and 5% for diabetes. As for the BMI, for each increase in the unit of this ratio, the chance for the indigenous per- son becoming a pre-diabetic is 16%; and 28%, for diabetes. Data found in a comparative study between the population of the Aracruz Indigenous Reserve (Brazil) and the population of Espı´- rito Santo (Brazil) showed that obese non-indigenous men and women were twice as likely to have DM but, when it comes to the indigenous people in this study, no significant differences were found [42]. Among the overweight indigenous people, the chance of having diabetes is four times higher, respectively. For the obese, on the other hand, the chances substantially increase, both for pre-diabetes, which increases to nine times, and for diabetes, which can reach 29 times. In the study with Guarani, Kaiowa´ and Terena, from the Jaguapiru village (Mato Grosso do Sul), the prevalence of diabetes among women was 9% and among men, 5%. The study also PLOS ONE | https://doi.org/10.1371/journal.pone.0255730 September 3, 2021 12 / 16 PLOS ONE Glycemic profile and associated factors in indigenous Munduruku, Amazonas. indicated a positive and significant association between obesity and diabetes (PR = 1.88; 95% CI = 1.45 – 2.43; p<0.001). A population-based study carried out in different Brazilian regions showed that obese individuals had 35% [95% CI = 1.35 – 1.86; p<0.001] chances of developing diabetes [43]. These findings show that the Munduruku, although still distant from the national mean of the general Brazilian population, are in an unfavorable condition toward the development of diabetes in relation to other ethnic groups living in a similar cultural context. Study limitations In the absence of specific cut-off points for indigenous populations, those used for the general population were considered, also adopted in other studies on different ethnic groups. As it was impossible to apply a dietary recall, it was not possible to verify how much the eat- ing habits are associated with the values found for glucose, cholesterol and triglycerides. The instruments adopted in the interview were not specific to indigenous peoples. How- ever, since it is an essential requirement to achieve the proposed objectives, the adequacy of language to the understanding of the group under study constituted a task that demanded dif- ferent moments of planning and evaluation by the team. Conclusion The 12% prevalence of glycaemia found among the Munduruku indigenous people is sugges- tive of diabetes mellitus, and that of 74.3%, revealing pre-diabetes, configure themselves as worrying indexes, as well as the chance of pre-diabetes, which increases by 20% when the BMI increases by one unit. It is necessary to consider changes in the eating habits and lifestyle, as well as environmental and social changes that can affect the health of the Munduruku, and consider the stress levels, with the possibility of each of these elements contributing or not to the results of this study. Consequently, it becomes indispensable to develop strategies combin- ing early diagnosis and treatment actions with actions to reduce the risk factors, in order to meet the needs and singularities of the Munduruku indigenous people. It is also suggested to develop new research studies on the topic in order to consolidate these findings in other Mun- duruku indigenous contexts. Author Contributions Conceptualization: Zilmar Augusto de Souza Filho, Noeli das Neves Toledo, Anto´nio Manuel Sousa, Gilsirene Scantelbury de Almeida. Data curation: Hanna Lorena Moraes Gomes, Neuliane Melo Sombra. Formal analysis: Anto´nio Manuel Sousa. Funding acquisition: Noeli das Neves Toledo. Investigation: Hanna Lorena Moraes Gomes, Neuliane Melo Sombra. Methodology: Hanna Lorena Moraes Gomes, Neuliane Melo Sombra. Project administration: Zilmar Augusto de Souza Filho, Noeli das Neves Toledo, Gilsirene Scantelbury de Almeida. Supervision: Zilmar Augusto de Souza Filho, Evelyne Marie Therese Mainbourg, Gilsirene Scantelbury de Almeida. Visualization: Hanna Lorena Moraes Gomes, Eliza Dayanne de Oliveira Cordeiro, Zilmar Augusto de Souza Filho, Evelyne Marie Therese Mainbourg, Gilsirene Scantelbury de Almeida. PLOS ONE | https://doi.org/10.1371/journal.pone.0255730 September 3, 2021 13 / 16 PLOS ONE Glycemic profile and associated factors in indigenous Munduruku, Amazonas. Writing – original draft: Hanna Lorena Moraes Gomes, Neuliane Melo Sombra, Eliza Day- anne de Oliveira Cordeiro. Writing – review & editing: Hanna Lorena Moraes Gomes, Neuliane Melo Sombra, Eliza Dayanne de Oliveira Cordeiro, Evelyne Marie Therese Mainbourg, Gilsirene Scantelbury de Almeida. References 1. Coimbra CEA Jr., Santos RV, Escobar AL. Epidemiologia e sau´de dos povos indı´genas no Brasil. Fio Druz. Abrasco, editor. Epidemiologia e sau´de dos povos indı´genas no Brasil. Rio de Janeiro; 2005. 2. Coimbra CEA Jr. Sau´de e povos indı´genas no Brasil: reflexões a partir do I Inque´rito Nacional de Sau´ de e Nutric¸ão Indı´gena. Cad Saude Publica [Internet]. 2014; 30(4):855–9. Available from: http:// www.ncbi.nlm.nih.gov/pubmed/24896060 https://doi.org/10.1590/0102-311x00031214 PMID: 24896060 3. Cardoso AM, Mattos IE, Koifman RJ. doenc¸as cardiovasculares na populac¸ão Guaranı´-Mbya´ do Estado do Rio de Janeiro Prevalenc. Cad Sau´de Pu´blica. 2001; 17(2):345–54. https://doi.org/10.1590/ s0102-311x2001000200009 PMID: 11283765 4. Cardoso AM, Horta BL, Coimbra Ju´ nior C, Folle´ r M-L, Souza MC de, Santos RV. Inque´rito Nacional de Sau´ de e Nutric¸ão dos Povos Indı´genas [Internet]. Abrasco, editor. Rio de Janeiro; 2009. 496 p. Avail- able from: https://goo.gl/nvzKjZ%5Cnecos-redenutri.bvs.br/tiki-download_file.php?fileId=1284 5. Codec¸o CT, Vilela D, Coelho F, Bastos LS, Carvalho LM, Gomes MFC, et al. Estimativa de risco de espalhamento da COVID-19 no Brasil e avaliac¸ão da vulnerabilidade socioeconoˆmica nas microrre- giões brasileiras. FGV Reposito´ rio Digit. 2020; 2:1–17. 6. Cunha E del BB, Fagundes RP, Scalabrin EE, Herai RH. Avaliac¸ão do perfil lipı´dico de adolescentes. Int J Cardiovasc Sci. 2018; 31(4):367–73. 7. Schmidt MI, Duncan BB, E Silva GA, Menezes AM, Monteiro CA, Barreto SM, et al. Doenc¸as croˆnicas não transmissı´veis no Brasil: carga e desafios atuais. Lancet. 2011; 377(9781):1949–61. https://doi. org/10.1016/S0140-6736(11)60135-9 PMID: 21561658 8. Cardoso FN, Domingues TAM, Silva SS, Lopes J de L. Fatores de risco cardiovascular modifica´veis em pacientes com hipertensão arterial sistêmica. Reme Rev Min Enferm [Internet]. 2020; 24:1–8. Avail- able from: https://cdn.publisher.gn1.link/reme.org.br/pdf/e1275.pdf 9. Lyra R, Oliveira M, Lins D, Cavalcanti N, Gross JL, Maia FFR, et al. Diretrizes Sociedade Brasileira de Diabetes 2019–2020. Vol. 5, Clannad. 2019. 709–717 p. 10. IDF. IDF DIABETES ATLAS. 2019. 168 p. 11. Oliveira GF De, Oliveira TRR De, Rodrigues FF, Corrêa LF, Ikejiri AT, Casulari LA. Prevalência de dia- betes melito e toleraˆ ncia à glicose diminuı´da nos indı´genas da Aldeia Jaguapiru, Brasil. Rev Panam Salud Publica. 2011; 29(5):315–21. https://doi.org/10.1590/s1020-49892011000500003 PMID: 21709935 12. Oliveira GF, Oliveira TRR, Ikejiri AT, Andraus MP, Galvao TF, Silva MT, et al. Prevalence of hyperten- sion and associated factors in an indigenous community of Central Brazil: A population-based study. PLoS One. 2014; 9(1):1–6. 13. Freitas GA de, Souza MCC de, Lima R da C. Prevalência de diabetes mellitus e fatores associados em mulheres indı´genas do Municı´pio de Dourados, Mato Grosso do Sul, Brasil. Cad Saude Publica. 2016; 32(8):1–12. 14. Meyerfreund D, Gonc¸alves CP, Cunha Roberto S, Pereira AC, Krieger JE, Mill JG. Age-dependent increase in blood pressure in two different Native American communities in Brazil. J Hypertens. 2009; 27(9):1753–60. https://doi.org/10.1097/hjh.0b013e32832e0b2b PMID: 19701999 15. Santos KM dos, Tsutsui ML da S, Galvão PP de O, Mazzucchetti L, Rodrigues D, A SG, et al. Grau de atividade fı´sica e sı´ndrome metabo´ lica: um estudo transversal com indı´genas Khisêdjê do Parque Indı´- gena do Xingu, Brasil. Cad Sau´ de Pu´ blica. 2012; 28(12):2327–38. https://doi.org/10.1590/s0102- 311x2012001400011 PMID: 23288065 16. Dal Fabbro AL, Franco LJ, Silva AS da, Sartorelli DS, Soares LP, Franco LF, et al. High Prevalence of type 2 Diabetes Mellitus in Xavante indians from Mato Grosso, Brazil. Ethn Dis. 2014; 24:35–40. PMID: 24620446 PLOS ONE | https://doi.org/10.1371/journal.pone.0255730 September 3, 2021 14 / 16 PLOS ONE Glycemic profile and associated factors in indigenous Munduruku, Amazonas. 17. Freitas GA de, Souza MCC de, Lima R da C. Prevalência de diabetes mellitus e fatores associados em mulheres indı´genas do Municı´pio de Dourados, Mato Grosso do Sul, Brasil. Cad Saude Publica. 2016; 32(8):e00023915. https://doi.org/10.1590/0102-311X00023915 PMID: 27626648 18. Soares LP, dal Fabbro AL, Silva AS, Sartorelli DS, Franco LF, Kuhn PC, et al. Cardiovascular risk in Xavante indigenous population. Arq Bras Cardiol. 2018; 110(6):542–50. https://doi.org/10.5935/abc. 20180090 PMID: 30226913 19. Faludi AA, Maria Cristina de Oliveira Izar, Kerr Saraiva Jose´ Francisco, Ana Paula Marte Chacra HTB, Neto AA, Bertolami A, et al. Atualizac¸ão da diretriz brasileira de dislipidemias e prevenc¸ão da atero- sclerose—2017. Arq Bras Cardiol. 2017; 109(1):76. 20. WHO. Obesity: preventing and managing the global epidemic. Who Tech Rep Ser. 2000; 37(10):253. PMID: 11234459 21. Gallagher D, Heymsfield SB, Heo M, Jebb SA, Murgatroyd PR, Sakamoto Y. Healthy percentage body fat ranges: An approach for developing guidelines based on body mass index. Am J Clin Nutr [Internet]. 2000; 72(3):694–701. Available from: https://pubmed.ncbi.nlm.nih.gov/10966886/ https://doi.org/10. 1093/ajcn/72.3.694 PMID: 10966886 22. Barroso WKS, Cibele Isaac Saad Rodrigues LAB, Mota-Gomes MA, Brandão AA, Feitosa AD de M, Machado CA, et al. Diretrizes Diretrizes Brasileiras de Hipertensão Arterial– 2020. Vol. 116. 2021. 516–658 p. 23. Santos JL, Pe´ rez-Bravo F, Carrasco E, Calvilla´ n M, Albala C. Low prevalence of type 2 diabetes despite a high average body mass index in the aymara natives from Chile. Nutrition. 2001; 17(4):305–9. https:// doi.org/10.1016/s0899-9007(00)00551-7 PMID: 11369169 24. Bennett PH, Burch TA, Miller M. Diabetes Mellitus in American (Pima) Indians. Lancet. 1971; 298 (7716):125–8. 25. De Souza Filho ZA, Ferreira AA, Dos Santos J, Meira KC, Pierin AMG. Cardiovascular risk factors with an emphasis on hypertension in the Mura Indians from Amazonia. BMC Public Health. 2018; 18(1):1– 12. https://doi.org/10.1186/s12889-018-6160-8 PMID: 30424745 26. Brasil MDS. Diretrizes e Recomendac¸ões para o Cuidado Integral de Doenc¸as Croˆnicas Diretrizes e Recomendac¸ões para o Cuidado Integral de Doenc¸as Croˆ nicas Não-Transmissı´veis [Internet]. Preven- tion. 2008. Available from: http://bvsms.saude.gov.br/bvs/publicacoes/diretrizes_recomendacoes_ cuidado_doencas_cronicas.pdf 27. WHO. Technical package for cardiovascular disease management in primary health care. Report. 2016;76. 28. Quermes PA de A, Carvalho JA de. Os impactos dos benefı´cios assistenciais para os povos indı´genas. Servic¸o Soc e Soc [Internet]. 2013; 116:769–91. Available from: http://www.planalto.gov.br/ccivil_03/ leis/ 29. Flor LS, Campos MR, de Oliveira AF, Schramm JM de A. Diabetes burden in Brazil: Fraction attribut- able to overweight, obesity, and excess weight. Rev Saude Publica. 2015;49. https://doi.org/10.1590/ S0034-8910.2015049005391 PMID: 26270011 30. Silva AAG de O, de Araujo LF, Diniz M de FHS, Lotufo PA, Bensenor IM, Barreto SM, et al. Neck cir- cumference and 10-year cardiovascular risk at the baseline of the elsa-brasil study: Difference by sex. Arq Bras Cardiol. 2020; 115(5):840–8. https://doi.org/10.36660/abc.20190289 PMID: 33295446 31. Fu W, Zou L, Yin X, Wu J, Zhang S, Mao J, et al. Association between neck circumference and cardio- metabolic disease in Chinese based cross- - adults: a community- - sectional study. BMJ Open. 2019;1–7. https://doi.org/10.1136/bmjopen-2019-030833 PMID: 31273011 32. Preis SR, Massaro JM, Hoffmann U, D’Agostino RB, Levy D, Robins SJ, et al. Neck Circumference as a Novel Measure of Cardiometabolic Risck: The Framingham Heart Study. J Clin Endocr Metab. 2010; 95 (8):3701–10. https://doi.org/10.1210/jc.2009-1779 PMID: 20484490 33. Preis SR, Pencina MJ, D’Agostino RB, Meigs JB, Vasan RS, Fox CS. Neck Circumference and the development of Cardiovascular Disease Risck Factors in the Framingham heart Study. Diabetes Care. 2013;36. 34. Welch JR, Ferreira AA, Tavares FG, Lucena JRM, Oliveira MVG de, Santos R V., et al. The Xavante Longitudinal Health Study in Brazil: Objectives, design, and key results. Am J Hum Biol. 2019;1–12. 35. Bloch K V., Coutinho E da SF, Loˆ bo MS de C, Oliveira JEP de, Milech A. Pressão arterial, glicemia capi- lar e medidas antropome´ tricas em uma populac¸ão Yanoma´ mi. Cad Saude Publica. 1993; 9(4):428–38. 36. Barroso TA, Marins LB, Alves R, Gonc¸alves ACS, Barroso SG, Rocha G de S. Associac¸ão Entre a Obe- sidade Central e a Incidência de Doenc¸as e Fatores de Risco Cardiovascular. Int j Cardiovasc sci. 2017; 30(5):416–24. PLOS ONE | https://doi.org/10.1371/journal.pone.0255730 September 3, 2021 15 / 16 PLOS ONE Glycemic profile and associated factors in indigenous Munduruku, Amazonas. 37. Fontela PC, Winkelmann ER, C PRNV. Estudo do ı´ndice de conicidade, ı´ndice de massa corporal e cir- cunferência abdominal como preditores de doenc¸a arterial coronariana. Rev Port Cardiol [Internet]. 2017; 36(5):357–64. Available from: https://doi.org/10.1016/j.repc.2016.09.013 PMID: 28449975 38. Souza Filho ZA de, Ferreira AA, dos Santos B, Pierin AMG. Hypertension prevalence among indige- nous populations in Brazil: A systematic review with meta-analysis. Rev da Esc Enferm [Internet]. 2015; 49(6):1012–22. Available from: https://www.scielo.br/pdf/reeusp/v49n6/0080-6234-reeusp-49-06- 1016.pdf 39. Toledo N das N, Almeida GS de, Matos MMM, Balieiro AA da S, Martin LC, Franco RJ da S, et al. Car- diovascular risk factors: differences between ethnic groups. Rev Bras Enferm. 2020; 73(4):e20180918. https://doi.org/10.1590/0034-7167-2018-0918 PMID: 32578730 40. Balabanski AH, Goldsmith K, Giarola B, Buxton D, Castle S, McBride K, et al. Stroke incidence and sub- types in Aboriginal people in remote Australia: A healthcare network population-based study. BMJ Open. 2020; 10(10):1–9. https://doi.org/10.1136/bmjopen-2020-039533 PMID: 33033097 41. Brown A, Carrington MJ, McGrady M, Lee G, Zeitz C, Krum H, et al. Cardiometabolic risk and disease in Indigenous Australians: The heart of the heart study. Int J Cardiol [Internet]. 2014; 171(3):377–83. Available from: https://doi.org/10.1016/j.ijcard.2013.12.026 PMID: 24388543 42. Alvim RDO, Mourao CA, De Oliveira CM, Krieger JE, Mill JG, Pereira AC. Body mass index, waist cir- cumference, body adiposity index, and risk for type 2 diabetes in two populations in Brazil: General and Amerindian. PLoS One. 2014; 9(6). 43. Flor LS, Campos MR. The prevalence of diabetes mellitus and its associated factors in the Brazilian adult population: evidence from a population-based survey. Rev Bras Epidemiol. 2017; 20(1):16–29. https://doi.org/10.1590/1980-5497201700010002 PMID: 28513791 PLOS ONE | https://doi.org/10.1371/journal.pone.0255730 September 3, 2021 16 / 16 PLOS ONE
null
10.1371_journal.pbio.3002512.pdf
Data Availability Statement: Codes and preprocessed data are available at https://osf.io/ m7dta/. Note that raw SEEG and neuroimaging (T1-MPRAGE) data are protected and cannot be shared (CPP Sud-Est V, 2009-A00239-48).
Codes and preprocessed data are available at https://osf.io/ m7dta/ . Note that raw SEEG and neuroimaging (T1-MPRAGE) data are protected and cannot be shared (CPP Sud-Est V, 2009-A00239-48).
RESEARCH ARTICLE Cross-frequency coupling in cortico- hippocampal networks supports the maintenance of sequential auditory information in short-term memory Arthur Borderie1,2, Anne Caclin3, Jean-Philippe Lachaux3, Marcela Perrone-Bertollotti4, Roxane S. Hoyer1, Philippe Kahane5, He´ lène Catenoix3,6, Barbara Tillmann3,7, Philippe AlbouyID 1,2,3* 1 CERVO Brain Research Center, School of Psychology, Laval University, Que´ bec, Canada, 2 International Laboratory for Brain, Music and Sound Research (BRAMS), CRBLM, Montreal, Canada, 3 Universite´ Claude Bernard Lyon 1, CNRS, INSERM, Centre de Recherche en Neurosciences de Lyon CRNL U1028 UMR5292, Bron, France, 4 Univ. Grenoble Alpes, Univ. Savoie Mont Blanc, CNRS, LPNC, Grenoble, France, 5 Univ. Grenoble Alpes, Inserm, U1216, CHU Grenoble Alpes, Grenoble Institut Neurosciences, Grenoble, France, 6 Department of Functional Neurology and Epileptology, Lyon Civil Hospices, member of the ERN EpiCARE, and Lyon 1 University, Lyon, France, 7 Laboratory for Research on Learning and Development, LEAD– CNRS UMR5022, Universite´ de Bourgogne, Dijon, France * philippe.albouy@psy.ulaval.ca Abstract AU : Pleaseconfirmthatallheadinglevelsarerepresentedcorrectly: It has been suggested that cross-frequency coupling in cortico-hippocampal networks enables the maintenance of multiple visuo-spatial items in working memory. However, whether this mechanism acts as a global neural code for memory retention across sensory modalities remains to be demonstrated. Intracranial EEG data were recorded while drug- resistant patients with epilepsy performed a delayed matched-to-sample task with tone sequences. We manipulated task difficulty by varying the memory load and the duration of the silent retention period between the to-be-compared sequences. We show that the strength of theta-gamma phase amplitude coupling in the superior temporal sulcus, the infe- rior frontal gyrus, the inferior temporal gyrus, and the hippocampus (i) supports the short- term retention of auditory sequences; (ii) decodes correct and incorrect memory trials as revealed by machine learning analysis; and (iii) is positively correlated with individual short- term memory performance. Specifically, we show that successful task performance is asso- ciated with consistent phase coupling in these regions across participants, with gamma bursts restricted to specific theta phase ranges corresponding to higher levels of neural excitability. These findings highlight the role of cortico-hippocampal activity in auditory short-term memory and expand our knowledge about the role of cross-frequency coupling as a global biological mechanism for information processing, integration, and memory in the human brain. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Borderie A, Caclin A, Lachaux J-P, Perrone-Bertollotti M, Hoyer RS, Kahane P, et al. (2024) Cross-frequency coupling in cortico- hippocampal networks supports the maintenance of sequential auditory information in short-term memory. PLoS Biol 22(3): e3002512. https://doi. org/10.1371/journal.pbio.3002512 Academic Editor: Timothy D. Griffiths, Newcastle University Medical School, UNITED KINGDOM Received: May 23, 2023 Accepted: January 22, 2024 Published: March 5, 2024 Copyright: © 2024 Borderie et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: Codes and preprocessed data are available at https://osf.io/ m7dta/. Note that raw SEEG and neuroimaging (T1-MPRAGE) data are protected and cannot be shared (CPP Sud-Est V, 2009-A00239-48). Funding: This work was conducted in the framework of the LabEx CeLyA ("Centre Lyonnais d’Acoustique", ANR-10-LABX-0060, https://celya. universite-lyon.fr/labex-celya-151124.kjsp) and of the LabEx Cortex ("Construction, Function and PLOS Biology | https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 1 / 24 Cognitive Function and Rehabilitation of the Cortex", ANR-11-LABX-0042, https://labex-cortex. universite-lyon.fr/) of Universite´ de Lyon, within the program "Investissements d’avenir" (ANR-11-IDEX- 0007, https://anr.fr/) operated by the French National Research Agency (ANR, https://anr.fr/). This work was supported a NSERC Discovery grant (https://www.nserc-crsng.gc.ca/) and a FRQS Junior 1 and 2 grants (https://frq.gouv.qc.ca/sante/ ) and a Brain Canada Future leaders Grant (https:// braincanada.ca/) to P.A. A.B. and R.S.H are funded by the CERVO Foundation (https://fondationcervo. com/, FRQS, https://frq.gouv.qc.ca/sante/). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Cross-frequency coupling enables integration and memory of auditory information in the human brain Introduction It is well established that the medial temporal lobe, in particular the hippocampus, is involved in the formation of long-term memories (LTM; [1]). Notably, hippocampal lesions consis- tently entail LTM deficits (i.e., anterograde amnesia [2]). In contrast, numerous empirical data obtained with a variety of materials, such as words [3], digits [4,5], tones [5], or single-dot loca- tions [4], have led to the hypothesis that hippocampal lesions do not impact working memory (WM) and short-term memory (STM) functions [6,7]. These findings suggest that WM and STM functions rely on distinct processes from LTM (e.g., [8,9]; see also [10,11] for neuroimag- ing studies). However, this hypothesis has been challenged by (i) neuropsychological studies reporting that patients with hippocampal lesions experience difficulties in maintaining items in WM or STM [12–14]; and (ii) fMRI [15–17], intracranial EEG [18–21], or single-unit recordings [22,23] in humans reporting persistent, load-dependent, hippocampal activity during WM maintenance of visual information (see also [15] for evidence of hippocampal involvement during auditory STM and [24] for a review about hippocampal activity during general auditory processing). honest significant difference; Abbreviations: HSD, AU : Anabbreviationlisthasbeencompiledforthoseusedthroughoutthetext:Pleaseverifythatallentriesarecorrectlyabbreviated: IES, inverse efficiency score; IFG, inferior frontal gyrus; ITG, inferior temporal gyrus; LMM, linear mixed model; LTM, long-term memory; PAC, phase amplitude coupling; PLV, phase locking value; RT, response time; STM, short-term memory; STS, superior temporal sulcus; SVM, support vector machine; WM, working memory. Hippocampal activity during WM and STM has been originally associated with mainte- nance-related increase of theta and gamma power [21,25–28]. Interestingly, recent studies went a step further by showing that successful visual memory performance requires the cou- pling of gamma activity to specific phases of the hippocampal theta (theta-gamma phase amplitude coupling (PAC) [29–32]). Theta-gamma PAC consists in gamma subcycles (local neural activity associated to the processing of each encoded item) that occur at specific theta phase ranges. It has been suggested that theta-gamma PAC plays a critical role in the mainte- nance of different items in memory and as well as their serial order [31–33]. To date, theta- gamma PAC has been observed in cortico-thalamo-cortical, cortico-cortical, and cortico-hip- pocampal networks for episodic, working, and long-term memory consolidation in the visual modality [28,34,35]. For the specific case of STM, hippocampal theta-gamma PAC has first been isolated with SEEG in a visual word recognition paradigm in humans: an increased syn- chronization between the phase of the theta band, and the power changes in the beta and gamma bands were observed when patients successfully remembered previously presented words [36]. Several studies have since confirmed the implication of PAC in STM and WM by showing that the simultaneous maintenance and/or manipulation of multiple visual items in memory is implemented under the form of hippocampal theta-gamma PAC [18,20,37,38]. Overall, previous results suggest that WM or STM maintenance, in which different items must be separately and sequentially maintained over a short period of time, is represented by an ordered activity of cell assemblies implemented under the form of theta-gamma PAC in human cortico-hippocampal networks [31]. However, to date, these studies have mainly focused on visuo-spatial processing, and very little is known about the potential role of theta- gamma PAC in auditory and hippocampal regions during the short-term retention of sequen- tial auditory information. Coupling across cortical oscillations of distinct frequencies in the auditory cortex has been assumed to enable the multiscale sensory analysis of speech (pho- nemes and syllables [39–41]). However, the direct contribution of auditory-hippocampal cross-frequency coupling for the short-term maintenance of sequential auditory information has not yet been demonstrated. In the present study, we recorded intracranial EEG data while drug-resistant patients with epilepsy performed a delayed matched-to-sample task with tone sequences. If theta-gamma PAC is a predictor of successful memory maintenance, its strength in the auditory and hippocampal regions should (i) be increased during short-term retention of tone sequences (as compared to simple perception); (ii) decode correct and incorrect PLOS Biology | https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 2 / 24 PLOS BIOLOGY Cross-frequency coupling enables integration and memory of auditory information in the human brain responses in the STM task using machine learning analysis; and, finally, (iii) be positively cor- related with individual auditory STM performance. Results Intracranial EEG recordings were obtained from 16 neurosurgical patients with focal drug- resistant epilepsy. The participants performed an auditory STM task, consisting in the compar- ison of tone sequences presented in pairs and separated by a silent retention period. In each block of the task, in 50% of the trials, the tone sequences were identical (expected response “same”) and 50% differed by one note (expected response “different”). To manipulate task dif- ficulty, in different conditions, we varied the memory load (3 or 6 to-be-encoded tones, with a tone duration of 250 ms) and the duration of the silent retention period between the to-be- compared sequences (2 s, 4 s, and 8 s; see Table 1 for a detailed description of the conditions and number of participants tested in each condition). Participants also performed a block of listening of the same trials with the instruction to not compare the tone sequences and were simply required to press a button as fast as possible at the end of the last tone of the second sequence (Perception task, 6 tones, 2 s silent period between the tone sequences; see Methods). Accuracy Task performance was evaluated using d prime (signal detection theory). To evaluate the impact of the duration of the silent retention period for 6-tone sequences, we performed a nonparametric repeated measures ANOVA (Friedman test) with duration (2 s, 4 s, and 8 s) as a within-participants factor (n = 6 participants, note that all participants did not perform all the tasks—see Table 1). The main effect of duration was significant χ2 (2) = 7.00, p = .03. Post hoc tests performed with Durbin–Conover pairwise comparisons revealed that performance in the 2 s duration condition was significantly better than performance in the 2 other duration conditions (4 s, p = 0.004; and 8 s, p = .03). Performance in the 4 s and 8 s conditions did not differ significantly (p = 0.24, Fig 1B, left panel). To evaluate the impact of memory load on accuracy (3 versusAU : PleasenotethatasperPLOSstyle; donotuse}vs:}exceptintablesandcaptions:Hence; allinstanceof }vs:}havebeenspelledoutto}versus}throughoutthetext: a Wilcoxon rank test revealing, as expected, that performance was increased for the 3-tone condition as compared to the 6-tone condition (W [5] = 21.0, p = 0.031; Fig 1B, right panel). 6 tones with a 4 s silent retention period, n = 6 participants), we performed Response times The same analyses were performed for response times of correct responses (RTsAU : PleasenotethatasperPLOSstyle; abbreviateanyinstanceofthefullword=phraseafterthefirstmention:Hence; allinstancesof }responsetime}or}responsetimes}havebeenchangedto}RT}or}RTs; }respectively: ; Fig 1C) in the same participants (n = 6). Nonparametric repeated measures ANOVA (Friedman test) Table 1. Description of the conditions. Conditions 6 tones—short retention 6 tones—medium retention 6 tones—long retention 3 tones—medium retention Task STM STM STM STM 6 tones -perception task Do not compare sequences and press 1 key at the end of the second sequence STM, short-term memory. https://doi.org/10.1371/journal.pbio.3002512.t001 Memory load Retention duration (s) Number of patients tested 6 tones (total sequence duration 1.5 s) 6 tones (total sequence duration 1.5 s) 6 tones (total sequence duration 1.5 s) 3 tones (total sequence duration 0.75 s) 6 tones (total sequence duration 2 4 8 4 2 16 6 16 6 16 1.5 s)AU : Pleaseconfirmthattheitalicized}6tonesðtotalsequenceduration1:5sÞ}underthe}Memoryload}columninTable1canbechangedtoregulartext: PLOS Biology | https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 3 / 24 PLOS BIOLOGY Cross-frequency coupling enables integration and memory of auditory information in the human brain Fig 1. Paradigm, behavioral performance, and brain oscillations. (A) Auditory tasks (here with 6-tone sequences, 2 s retention): “Same” trials: After a delay, the first melody was repeated. “Different” trials: One tone was changed in the second melody of the pair in comparison to the first melody (red rectangle). Memory load (3 or 6 tones) and duration of the retention period (2, 4, 8 s) varied in separate blocks. Source data can be found at https://osf.io/m7dta/. (B) Accuracy in terms of d prime presented as a function of the duration of the retention period (left panel; N = 6) and memory load (right panel; N = 6). Colored circles depict participants (one color per participant). Asterisks indicate significance (p < 0.05, nonparametric tests; see text for details); NS, nonsignificant. Source data can be found at https:// osf.io/m7dta/. (C) Response time (s) presented as a function of the duration of the retention period (left panel; N = 6) and memory load (right panel; N = 6). Colored circles depict participants (one color per participant; same color coding as in Fig 1B). NS, nonsignificant. Source data can be found at https://osf.io/m7dta/. (D) Left panel: T-values in the time-frequency domain (t test relative to baseline −1,000 to 0 before stimulus onset, FDR corrected in time and frequency domains) of SEEG contacts located in the right and left Heschl’s gyrus (displayed on the single subject T1 in the MNI space provided by SPM12) for a trial time window (−1,000 to 6,000 ms) for the condition 6-tone memory load, 2 s retention period (n = 5). Right panel shows the PSD, power spectrum density (zscore) average over a trial time window (0 to 5,000 ms) that was used to define frequency for phase and frequency for amplitude for the PAC analysis. Shaded error bars indicate SEM. Source data can be found at https://osf.io/m7dta/. (E) Left panel: T-values in the time- frequency domain (t test relative to baseline −1,000 to 0 before stimulus onset, FDR corrected in time and frequency domains) of SEEG contacts located in the right and left hippocampus (displayed on the single subject T1 in the MNI space provided by SPM12) for a trial time window (−1,000 to 6,000 ms) for the condition 6-tone memory load, 2 s retention period (n = 14). Right panel shows the PSD, power spectrum density (zscore) average over a trial time window (0 to 5,000 ms) that was used to define frequency for phase for the PAC analysis. Shaded error bars indicate SEM. Source data can be found at https://osf.io/m7dta/. (F) SEEG contacts modelled with 4 mm radius spheres (see Methods) in the MRI volume showing a significant increase in oscillatory power (FDR corrected) relative to baseline in theta (4 Hz) and gamma (30–90 Hz) ranges (Hilbert transform averaged over time) during encoding, retention, and retrieval in all memory conditions in all participants (n = 16). All results are displayed on the single subject T1 in the MNI space provided by SPM12. Source data can be found at https://osf.io/m7dta/. https://doi.org/10.1371/journal.pbio.3002512.g001 PLOS Biology | https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 4 / 24 PLOS BIOLOGY Cross-frequency coupling enables integration and memory of auditory information in the human brain with duration (2 s, 4 s, and 8 s) as a within-participants factor revealed that the main effect of duration of the silent retention period was not significant χ2 (2) = 0.33, p = .84. In addition, Wilcoxon rank test revealed no significant difference of RTs between the 3-tone condition and the 6-tone condition (4 s silent retention period, W [5] = 4.00, p = .21; Fig 1C, right panel). Spectral fingerprints of perception and short-term memory of auditory sequences Fig 1D and 1E show the oscillatory activity (t test relative to the baseline −1,000 to 0 ms before stimulus onset, FDR corrected in time and frequency) in the time-frequency domain for SEEG contacts located in the left and right Heschl’s Gyri (according to the AAL3 atlas; see Methods, Fig 1D, 9 SEEG contacts, n = 5 participants with one electrode in this area, S1 Table) and bilat- eral hippocampal and para-hippocampal regions (Fig 1E, 72 SEEG contacts, n = 14 partici- pants with one electrode in these areas, S2 Table) for a trial time window for the 6-tone condition, 2 s retention period. Note that the same figures using a logarithmic scale for the fre- quency axis are presented in S1 Fig. In the auditory cortex, for each tone during the encoding and retrieval periods, transient gamma activity (30 to 90 Hz) was observed. As expected, the encoding of the entire sequence in the auditory cortex was associated with sustained theta oscillations at 4 Hz (tone presentation rate) and at 8 Hz (harmonic; Fig 1D). Moreover, a sig- nificant alpha/beta (10 to 20 Hz) desynchronization (relative to baseline) was observed in the auditory cortex during encoding, retrieval, and at the beginning of the retention period (Fig 1D). In the hippocampal and para-hippocampal regions, sustained theta oscillations (4 to 8 Hz) were observed during the entire trial time window (Figs 1E and S1). We then aimed to evaluate the fluctuations of power relative to baseline in these frequency bands for all SEEG contacts in all participants and all memory conditions. We used Hilbert’s transform (to reduce the dimension of the data) to extract the magnitude of theta (4 Hz) and gamma (30 to 90 Hz) oscillations during encoding, retention, and retrieval periods of the dif- ferent conditions (averaged in time; see Table 1 for the relevant time periods) for each partici- pant, each SEEG contact, and each trial. A contrast with baseline (FDR corrected) revealed that gamma activity was increased bilaterally in primary and secondary auditory regions and in the hippocampus during encoding retention and retrieval (Fig 1F, top panel; see SupportingAU : PleasenotethatPLOSusestheterm}Supportinginformation:}Hence; }supplementaryinformation}hasbeenreplacedwith}Supportinginformation}throughoutthetext: information for details and coordinates). During memory retention, an increase in theta activity was observed in a distributed net- work including the hippocampal/para-hippocampal regions, inferior frontal gyrus, and several regions of the ventral auditory stream (see Supporting information for details and coordinates; Fig 1F, bottom panel). To investigate whether these fluctuations of oscillatory power were specific to the memory task, we contrasted memory trials (6 tones, 2 s silent retention delay) with perception trials (6 tones, 2 s silent delay) for each frequency band (theta, gamma) and for all time periods (encod- ing, retention, retrieval; note that period names apply to the memory task) with nonparametric permutation tests (see Methods and supporting results). To assess significance, we applied a cluster-based approach: We defined SEEG contacts as significant only when they were overlap- ping for at least 2 participants or 2 SEEG contacts (overlap estimated on an MRI volume where SEEG contacts are represented by spheres with a radius of 4 mm; see Methods). This analysis did not reveal any significant effect for the contrast memory versus perception for each of the periods of the task (encoding, retention, retrieval), all p-values > .05 (see S2 Fig plotting theta and gamma power for memory and perception conditions in all SEEG contacts located in regions showing increased theta and gamma power relative to baseline during the retention period). PLOS Biology | https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 5 / 24 PLOS BIOLOGY Cross-frequency coupling enables integration and memory of auditory information in the human brain Theta-gamma PAC is associated with auditory STM retention Notwithstanding the fact that no effect was observed for the memory versus perception con- trast on theta and gamma power, we investigated whether theta-gamma PAC during memory retention could rather be a more specific marker of STM retention. For all PAC analyses, we adopted the following strategy: All analyses, except the memory versus perception contrast (see Table 1 and Fig 2), were done within subject, for all participants, using all data of the memory conditions. We then report only the significant SEEG contacts that were overlapping between participants or between electrodes using a cluster procedure (see below and Meth- ods). As expected, during encoding, clear transient gamma oscillations were nested in the theta cycle (Fig 2A for illustration) in the auditory cortex (Heschl’s gyrus, 9 SEEG contacts, n = 5 participants, S1 Table). To investigate whether this mechanism played a functional role during retention, we contrasted the theta-gamma PAC strength values of memory trials (6 tones, 2 s retention) with the theta-gamma PAC strength values of perception trials (6 tones, 2 Fig 2. Theta-gamma PAC during encoding and retention. (A) Top: Time-frequency plot of mean gamma power modulation time- locked to a 4-Hz (theta) oscillation during encoding in the right and left median belt (n = 7). Bottom: Theta (4 Hz) cycles for a 1-s time window. Source data can be found at https://osf.io/m7dta/. (B) Memory vs. perception contrast during retention. Top: SEEG contacts (left hippocampus (2 SEEG contacts, n = 2) and right auditory areas (15 SEEG contacts, n = 1)) showing a significant increase of theta (4 Hz)–gamma (30–90 Hz) PAC strength for memory trials as compared to perception trials during the silent (retention) delay (6 tones, 2 s retention period). All results are displayed on the single subject T1 in the MNI space provided by SPM12. Source data can be found at https://osf.io/m7dta/. (C) Bar plot shows theta-gamma PAC values averaged over trials and participants for memory and perception conditions for the significant SEEG contacts displayed in (B). Circles show individual trials. Source data can be found at https://osf.io/ m7dta/. (D). T-values for the co-modulogram (in SEEG contacts identified in B) for memory versus perception contrast (p < .05, FDR corrected). Source data can be found at https://osf.io/m7dta/. https://doi.org/10.1371/journal.pbio.3002512.g002 PLOS Biology | https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 6 / 24 PLOS BIOLOGY Cross-frequency coupling enables integration and memory of auditory information in the human brain s retention) during the retention period (permutation testing, 10,000 permutations), for each participant and each of their SEEG contacts (Fig 2B). After computing this analysis for each participant, we used the same cluster-based approach as for the analysis of oscillatory power (see Methods). This analysis revealed a clear increase in theta-gamma PAC in the left hippo- campus (2 SEEG contacts, n = 2) and right auditory regions (15 SEEG contacts, n = 1) in the memory condition compared to the perception condition (Fig 2B and 2C, all ps < 0.001; see S3 Table for coordinates). However, one can question whether this coupling was specific to theta and gamma oscilla- tions as theta-beta, alpha-gamma, and alpha-beta PAC have previously been reported during working memory [42]. To test whether this effect was specific to the phase of the theta and the amplitude of the gamma oscillations, we computed the same analysis in the SEEG contacts showing significant PAC increase in the memory versus perception contrast (displayed Fig 2B; see S3 Table for details and coordinates), but using multiple low frequencies as frequency for phase (4 to 11 Hz, i.e., theta to alpha) and multiple high frequencies as frequency for amplitude (15 to 140 Hz, i.e., beta to high gamma; see Fig 2D). Interestingly, the memory versus percep- tion contrast performed on these co-modulograms (p < .05, FDR corrected) revealed that the maximum increase in PAC strength for memory trials as compared to perception trials was observed between theta (4 to 6 Hz) as frequency for phase and gamma as frequency for ampli- tude (35 to 105 Hz). Note that we performed the same analysis in all SEEG contacts located in regions showing increased theta and gamma power relative to baseline during retention (Fig 1F, middle panel, coordinates in the Supporting information). This analysis revealed no significant difference of PAC strength between memory and perception trials after FDR cor- rection (see S3 Fig for illustration of the difference of PAC strength values between memory and perception trials) Theta-gamma PAC in fronto-temporal areas and hippocampus decodes correct and incorrect memory trials and correlates with auditory STM performance We then investigated whether the strength of theta-gamma PAC during memory retention can decode correct and incorrect memory trials and predict STM performance. To do so, we used the SEEG data and the behavioral data of all memory conditions for each participant. We first used a support vector machine (SVM) classifier with 3-fold cross-validation to classify correct and incorrect trials in all memory conditions, using only PAC strength in each SEEG contact as input features (see Methods). This approach was implemented for each participant: The model is trained only on data from 2/3 of the trials to predict whether a trial is correct or incorrect in the remaining 1/3 of the trials. The procedure is repeated 3 times, and the sum- mary of the SVM’s performance (average of all models) reflects, for each participant, the degree to which correct and incorrect STM trials can be discriminated based on PAC strength. As all participants had more correct than incorrect trials for all memory conditions, we made a random selection of the correct trials (to match the number of incorrect trials for each condi- tion) to train and test the classifier. Then, we repeated this analysis 100 times with 100 different random selection of correct trials for each participant. SVM’s performance was evaluated using the output of the 100 models (accuracy minus chance) for each participant. The models significantly classified correct and incorrect memory trials above chance in 12/ 16 participants (all ps < .03 as measured by a Wilcoxon rank test; Fig 3A; ROC curves for each participant are presented in Fig 3B). We then aimed to define the SEEG features (i.e., SEEG contacts) the models relied upon to discriminate correct and incorrect STM trials. For each participant with significant above chance decoding accuracy, we extracted the feature weights PLOS Biology | https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 7 / 24 PLOS BIOLOGY Cross-frequency coupling enables integration and memory of auditory information in the human brain Fig 3. PAC as markers of correct vs. incorrect memory retention identified with machine learning. (A) SVM decoding accuracy (accuracy minus chance—chance level: 0%) for a 2-class decoding analysis of PAC strength and SEEG contacts as features (correct vs. incorrect memory retention in all memory conditions). The colored bars represent accuracy minus chance for each participant (sorted as a function of accuracy with a jet colormap). Orange shaded rectangle overlaps with participants showing decoding accuracy significantly above chance. Blue shaded rectangle overlaps with participants with decoding accuracy not significantly different from chance. Asterisk: significant, ns: nonsignificant. Source data can be found at https://osf.io/m7dta/ (B) ROC for each participant (same color code as in A). Black dashed line represents the chance level. Source data can be found at https://osf.io/m7dta/. (C) Normalized feature weights showing features (SEEG contacts) with the largest influence (z-score) for each participant with significant decoding accuracy. Source data can be found at https://osf.io/m7dta/. PAC, phase amplitude coupling; ROC, receiver operating characteristic curve; SVM, support vector machineAU : AbbreviationlistshavebeencompiledforthoseusedinFigs3 (cid:0) 5:Pleaseverifythatallentriesarecorrectlyabbreviated: . https://doi.org/10.1371/journal.pbio.3002512.g003 to estimate their relative importance (z-scored, normalized across features for each partici- pant) in the classification. We then extracted the SEEG contact showing the maximum zscore value (i.e., contributing more to the classification) for each participant and represented it on a MRI volume (Fig 3C). This analysis revealed that the right and left hippocampus, the right IFG, the right and left primary auditory cortices, the left STS, and the left ITG (see S4 Table for details) were the brain regions where PAC strength allowed to classify correct and incorrect memory trials. It is relevant to note, however, that this analysis does not allow to infer whether PAC strength in the identified brain regions was associated to good or poor performance. Indeed, the features weights shown in Fig 3C can be used only to infer that PAC strength in these given SEEG contacts can decode correct and incorrect memory trials. We thus investigated whether theta-gamma PAC during memory retention can be corre- lated to STM performance. To do so, we used the SEEG data and the behavioral data of all memory conditions for each participant. This allowed us to benefit from the variability in PLOS Biology | https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 8 / 24 PLOS BIOLOGY Cross-frequency coupling enables integration and memory of auditory information in the human brain Fig 4. Theta-gamma PAC in the hippocampus and ventral auditory stream correlates with behavior. (A) Left panel: SEEG contacts showing a positive correlational relationship between theta-gamma PAC and performance (negative correlation with IES). Results are displayed on the single subject T1 in the MNI space provided by SPM12. Right panel: Scatter plot of IES (note that the scale is inverted for clarity: 5 corresponding to poor performance and 0 corresponding to good performance) against theta-gamma PAC strength for each significant SEEG contact. Each color depicts a different participant (N = 6). Source data can be found at https://osf.io/m7dta/. (B) Left panel: SEEG contacts showing a negative correlational relationship between theta-gamma PAC and performance (positive correlation with IES). Results are displayed on the single subject T1 in the MNI space provided by SPM12. Right panel: Scatter plot of IES (note that the scale is inverted for clarity: 5 corresponding to poor performance and 0 corresponding to good performance) against theta-gamma PAC strength for each significant SEEG contact. Colors show the different participant (N = 4). Source data can be found at https://osf.io/m7dta/. IES, inverse efficiency score; PAC, phase amplitude coupling. https://doi.org/10.1371/journal.pbio.3002512.g004 behavioral performance associated with the manipulation of the memory load and of the dura- tion of the retention period. As a significant effect of condition emerged for the accuracy data (Fig 1B), but not for the RT data (Fig 1C), we computed for each trial the inverse efficiency score (IES; correct RT at the single trial scale/percent correct in the corresponding condition; see [43] and Methods). This behavioral metric increased the variability of behavioral scores between memory conditions with a low score representing a rapid RT and a high percentage of correctness. We then performed a Pearson’s correlation between IES and PAC strength val- ues for each SEEG contact and each participant (across all conditions). This analysis revealed, after cluster correction, that theta-gamma PAC values in the left hippocampus (4 SEEG con- tacts, n = 2), left superior temporal sulcus (STS; 2 SEEG contacts, n = 2), right inferior tempo- ral gyrus (ITG; 2 SEEG contacts, n = 2), and left inferior frontal gyrus/insula (IFG; 2 SEEG contacts, n = 2) had a positive correlational relationship with performance (i.e., negatively cor- related with the IES; Fig 4A and see S5 Table). Moreover, this analysis also revealed that theta- gamma PAC in the left Heschl’s gyrus (4 SEEG contacts, n = 4) had a negative relationship with performance (positively correlated with the IES; Fig 4B and S6 Table). Note that we per- formed the same analysis only with the conditions that were performed by all 16 participants (see Table 1) and obtained similar results (see S4 Fig). Coupling phase is consistent across participants and trials The analyses presented in Figs 2 to 4 evaluated PAC strength for each participant (coupling consistent across trials, within participant). However, these analyses do not guarantee that the coupling occurred at the same phase for all participants: Different participants could show a preferred coupling at different phases of the theta oscillations. To investigate this question, we PLOS Biology | https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 9 / 24 PLOS BIOLOGY Cross-frequency coupling enables integration and memory of auditory information in the human brain further evaluated whether gamma bursts were consistently restricted to specific phase ranges of the theta oscillations across participants in regions identified in Fig 4A (using data of all conditions available for the participants showing significant effects in Fig 4A). We first com- puted the theta-gamma phase consistency across trials, for the SEEG contacts where the PAC strength was correlated with behavioral performance (see Fig 4A and S5 Table). For each trial, and each SEEG contact, we extracted the magnitude of gamma oscillations (30 to 90 Hz) as a function of the phase of the theta oscillation (4 Hz) (average over the entire retention period, theta phase divided into 8 bins; see Methods). In both memory (correct trials) and perception trials separately, we computed the intertrial phase locking value (PLV) as a measure of inter- trial phase consistency of the coupling. Then, this metric was contrasted between memory and perception trials (Wilcoxon rank test) for each region (grouping SEEG contacts as a function of their location in the AAL atlas; Fig 5A). As expected, this analysis revealed greater consis- tency in theta-gamma PAC for memory as compared to perception trials for all regions (all p- values < .0001; Fig 5B). Finally, we aimed to identify whether a specific coupling phase range between the phase of the theta oscillations and the amplitude of gamma oscillations can be identified in these regions across trials and participants. To do so, we used linear mixed models (LMM) and mod- eled the variability between participants by defining by-participant random intercepts. This analysis was done for each region with theta phase bin as fixed factors and participants as a random factor (using data of all memory conditions available for the participants showing sig- nificant effects in Fig 4A). For all regions, we observed a main effect of theta phase (all χ2 (7) > 18.7; all ps < .01) on the gamma power. Post hoc Tukey analysis revealed increased gamma power between −π/2 and 0 of the theta cycle as compared to other bins in all regions (Fig 5C, see S7–S10 Tables for detailed statistics). Discussion Using intracranial electrophysiological recordings in humans, we showed that (i) the strength of theta-gamma PAC in temporal regions and hippocampus was increased during the short-term retention of auditory sequences as compared to simple perception; (ii) the strength of theta-gamma PAC in STS, ITG, IFG, and hippocampus decode correct and incorrect memory trials as evaluated with machine learning; (iii) the strength of theta- gamma PAC in these regions was positively correlated with individual STM performance; and, finally, that (iv) the coupling phase was highly consistent in these regions across indi- vidual participants to enable successful memory performance (high-frequency oscillations consistently restricted to specific phase ranges of the theta oscillations). The implications of these findings are discussed below. Increasing memory load and duration of the silent retention period decrease performance In line with previous studies, the present behavioral findings indicated that participants’ STM abilities (as also observed for other materials, such as verbal or visuo-spatial) decreased with increasing duration of the silent retention period [44] and increasing memory load ([45]; see Fig 1B). In the present study, we used these manipulations to increase the variability in task difficulty (and, consequently, modulate participants’ behavioral performance) across condi- tions. By combining information from accuracy and response times, we extracted a behavioral measure for each trial (IES; see methods and [43]) that we used to investigate the link between PAC strength values and behavior for each participant. PLOS Biology | https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 10 / 24 PLOS BIOLOGY Cross-frequency coupling enables integration and memory of auditory information in the human brain Fig 5. Theta gamma PAC is consistent across trials and participants. (A) SEEG contacts identified in Fig 4A and grouped as a function of their location according to the AAL Atlas: green, left STS; red, left hippocampus; blue, right ITG; yellow, left IFG/insula. Regions are displayed on the single subject T1 in the MNI space provided by SPM12. Source data can be found at https://osf.io/m7dta/. (B) PAC intertrial phase consistency computed for each region. Bar plot shows intertrial phase locking values across participants and SEEG contacts for memory trials (correct responses, colored as a function of the regions) and perception trials in the same region. Error bars indicate SEM. Asterisk indicates significance. Source data can be found at https://osf.io/m7dta/. (C) Preferred coupling phase: gamma power presented as a function of theta phase bins for each region. Shading represents the standard deviation across trials and participants. Asterisks (*** p < .001; * p < .05) and grey shading indicate significance. Note that for clarity, we show only the results for the post hoc tests performed for the peak of gamma power for each region. Detailed post hoc statistics are reported in S7–S10 Tables. Source data can be found at https://osf.io/m7dta/. IFG, inferior frontal gyrus; ITG, inferior temporal gyrus; PAC, phase amplitude coupling; STS, superior temporal sulcus. https://doi.org/10.1371/journal.pbio.3002512.g005 Brain networks of auditory perception and short-term memory Time-frequency analyses revealed that transient gamma activity was evoked by each tone of the sequence in the auditory cortex, secondary auditory regions, hippocampus, and several PLOS Biology | https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 11 / 24 PLOS BIOLOGY Cross-frequency coupling enables integration and memory of auditory information in the human brain areas of the ventral pathway during the encoding and retrieval periods of the STM task and the equivalent periods of the perception task (see Fig 1C and 1D). It is well established that gamma oscillations are marking bottom-up and local (intraregional) processes during both passive and active sensory integration [46,47]. Observing such transient bursts after each tone of the to-be-encoded sequence can thus be considered as a marker of the integration of tones’ fea- tures by the sensory system (bottom-up). In addition, sustained theta oscillations were observed in distributed regions of the ventral pathway, including STS, STG, IFG, and hippocampus (see Supporting information) during encoding, retention, and retrieval. Theta oscillations (4 to 8 Hz) are typically considered as markers of attention, arousal, or memory during demanding cognitive tasks [48–50]. Notably, theta oscillations are known to play a key role in ordering items that are presented sequentially in STM or WM [51]. Moreover, theta oscillations have been associated to long-range commu- nication between distant brain regions during memory maintenance [49,50,52–54]. In the present study, an increase relative to baseline in theta power was observed in the hippocampus, inferior frontal regions, and secondary auditory regions, a brain network that has been consis- tently reported as being recruited during auditory STM tasks [15,55–57] (Fig 1F). However, during all phases of the task (referred to as encoding, retention, and retrieval periods for the memory task and their equivalent for the perception task), we did not observe any significant differences of gamma and theta magnitude between memory and perception trials. This result contrasts with the studies reported above [49,50,52–54]. A possible interpre- tation would be that the participants have been carrying out a form of WM during the percep- tion task (always performed after the memory condition; see Methods) even if they were not instructed to do so. An alternative interpretation would be that the fluctuations in oscillatory magnitude in the theta and gamma frequency ranges extracted in the present study were not specific to memory and might rather be associated with the perception of the sequence and attention towards the auditory input (note that even in the perception task, participants had to pay attention to the sound sequences to push a button at the end of S2).We thus aimed to define whether more fine-grained oscillatory markers related to memory retention can be identified with the investigation of theta-gamma PAC. Theta-gamma PAC in auditory and hippocampal regions is associated to auditory short-term memory retention During encoding, we observed that gamma oscillations were nested in the theta cycle in the auditory cortex (see Fig 2A for illustration). This effect was expected as each tone of the sequence induced a time-locked (or evoked) increase in gamma power, and the phase of the theta oscillation was entrained by the tone presentation rate (4 Hz; see [49,54] for basic princi- ples of sensory entrainment). We then investigated whether this statistical dependency between the phase of theta oscillations and the amplitude of gamma oscillations was still pres- ent during the retention period, a time window for which no stimuli were presented. More specifically, we investigated whether PAC signals were increased during memory retention as compared to perception. In the left hippocampus and right temporal regions, the strength of theta-gamma PAC was indeed significantly higher during the retention delay in the memory condition compared to the perception condition (see Fig 2B and S3 Table). It is relevant to note that this effect was observed in a limited number of SEEG contacts and participants. This is related to the cluster correction procedure we have used that keep only SEEG contacts that overlap between participants or contacts. One possible interpretation is that PAC during memory retention could result from sustained PAC signals that originally emerged during encoding (see Fig 2A; PAC coming from bottom-up entrainment at 4 Hz). It can thus be PLOS Biology | https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 12 / 24 PLOS BIOLOGY Cross-frequency coupling enables integration and memory of auditory information in the human brain argued that the significant effect observed between memory retention and perception could result from attentional differences for memory and perception trials during encoding (partici- pants paying more attention during memory than perception trials). However, one can argue that attentional effects could not only be observed in PAC measures but could also affect theta and gamma magnitude [58]. As the contrast between memory trials and perception trials for theta and gamma magnitude was not significant in the present study, we propose that these PAC effects were specific to memory. These results thus suggest a role of the hippocampus in auditory STM. This is in line with several neuroimaging studies in the visual modality [16,18,19,38] and also with recent single- unit recording studies in humans reporting increased neural firing in the hippocampus during the maintenance of visual representations [22,23,59]. For auditory STM, hippocampal involve- ment has, however, been less frequently described in previous research. Using an auditory STM task during fMRI recordings, Kumar and colleagues [15] have shown sustained activity in both ventral and dorsal parts of the hippocampus during an auditory STM task. Here, we observed activity mainly in its ventral part (y = −4), a finding fitting well with the fact that the anterior portion of the hippocampus is anatomically and functionally connected to auditory areas [60,61]. Interestingly, Kumar and colleagues [15] also reported that the pattern of fMRI activity in hippocampal areas allows the decoding of the different sounds maintained in mem- ory. Our present study goes beyond these findings by identifying the neurophysiological mech- anism by which the hippocampus supports retention of auditory information in memory. Indeed, here we showed that theta-gamma PAC in the hippocampus and temporal regions (STS, ITG) decodes correct and incorrect memory trials (Fig 3A and S4 Table) and was posi- tively correlated with behavioral performance (negative correlation with IES; Fig 4A and 4B and S5 Table). This finding is well aligned with previous research showing that hippocampal theta-gamma PAC plays a functional role during memory retention for visual material [18,20,37,38]. In the present study, we show that the temporal and hippocampal regions imple- ment the same electrophysiological mechanism to allow for the maintenance of sequential auditory information, a finding that has, to our knowledge, never been reported before. This finding is also well aligned with a recent study showing cortico-hippocampal interplay in the theta range during both encoding and retention of a STM task with visually presented words [62]. Taken together, our results suggest a clear role of theta-gamma PAC in the temporal and hippocampal regions during auditory STM in the human brain. In addition to auditory and hippocampal regions, we observed that theta-gamma PAC strength in the left IFG decodes correct and incorrect memory trials (Fig 3A and S4 Table) and was positively correlated with behavioral performance (negative correlation with IES; Fig 4A and S5 Table). This is in line with the well-established role of the IFG in STM maintenance in humans [15,50,55–57,63–69].Interestingly, we also observed that theta-gamma PAC in Heschl’s gyrus during memory retention was negatively correlated with behavioral performance (positive corre- lational relationship with IES; Fig 4B). This result suggests that to perform successfully the STM task, PAC signals need to reach higher-level regions, namely, STS, ITG, hippocampus, and infe- rior frontal regions, to allow for efficient maintenance of the information. This hypothesis receives support in a recent study showing that theta and gamma activity in the human hippo- campus is associated with successful recall when extrahippocampal activation patterns shifted from perceptual toward mnemonic representations. This study also suggests that recurrent hip- pocampal–cortical interactions are then implemented to support memory processing [70]. From a more global perspective, our results are in agreement with the theta-gamma neural code hypothesis developed by Lisman and Jensen [31], proposing that cross-frequency signal- ing in cortico-hippocampal networks is a sophisticated mechanism implanted by the brain to hold sequentially organized information in memory [20,25,31]. This hypothesis assumes that PLOS Biology | https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 13 / 24 PLOS BIOLOGY Cross-frequency coupling enables integration and memory of auditory information in the human brain representations of individual encoded items (via high-frequency oscillations) do not occur during the entire cycle of low-frequency oscillations. Instead, these high-frequency oscillations are thought to be restricted to specific phase ranges of the slow oscillation that correspond to higher levels of neural excitability [20,31,71]. To test the validity of this model, we investigated for each region whether the gamma bursts in the present data were consistently restricted to a specific phase range of the theta oscillations across trials and participants. Consistent phase coupling across participants during successful memory performance We extracted the PAC consistency across trials and participants in the brain regions where PAC strength was positively predicting behavioural performance (see Fig 5A and S5 Table). Inter- trial-phase locking analysis on these signals revealed greater consistency in theta-gamma PAC for memory trials than for perception trials in all regions (Fig 5B). We then aimed to identify whether a preferred coupling phase range could be identified. We observed that, for correct memory trials, the gamma bursts were occurring consistently at a specific phase range of the theta cycle in the left STS, right ITG, left IFG, and the left hippocampus (see Fig 5C and S7–S10 Tables). This preferred phase is of interest because it suggests that similar mechanisms are implemented in this network across trials and participants. Interestingly, the gamma burst occurred from the trough of the theta cycle to its peak. As shown in earlier research, the phase of theta oscillation reflects rhythmic fluctuations of neural excitability [72]. Such cycles, occur- ring several times per second, represent fluctuations between (high-excitability) phases during which relevant information is amplified and (low-excitability) phases during which information is suppressed. Here, we observed high coupling consistency between −π/2 and 0 of the theta cycle, a phase range corresponding to a high-excitability period of the oscillation where infor- mation processing can be amplified [25,31,72]. Observing this effect only for correct memory trials is another important cue suggesting that fronto-auditory-hippocampal theta-gamma PAC allows successful integration and the retention of sequential auditory information in STM. Overall, our study provides new information about the neurophysiological mechanisms by which the fronto-temporal-hippocampal network encodes and maintains sequential auditory information. The findings provide crucial insights into the networks and brain dynamics involved in this fundamental process in the auditory modality. Methods Participants Intracranial recordings were obtained from 16 neurosurgical patients with drug-resistant focal epilepsy (8 females and 8 males, mean age: 32.6 +/− 8.73 years) at the Epilepsy Department of the Grenoble Neurological Hospital (Grenoble, France) and the Epilepsy Department of Lyon Neurological Hospital (Lyon, France). All patients were stereotactically implanted with multi- lead EEG depth electrodes. Data from all electrodes exhibiting pathological waveforms were discarded from the present study. All participants provided written informed consent, and the experimental procedures were approved by the appropriate regional ethics committee (CPP Sud-Est V, 2009-A00239-48). The study has been conducted according to the principles expressed in the Declaration of Helsinki. Task and conditions The participants were asked to perform an auditory STM task, consisting in the comparison of tone sequences presented in pairs and separated by a silent retention period. Participants also PLOS Biology | https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 14 / 24 PLOS BIOLOGY Cross-frequency coupling enables integration and memory of auditory information in the human brain performed a block of passive listening of these trials in which they were required to ignore the content of tone sequences and press a button as fast as possible at the end of S2. To manipulate task difficulty (only for the memory task), in different blocks, we varied the memory load (3 or 6 to-be-encoded items) as well as the duration of the silent retention period between the to-be- compared sequences (2 s, 4 s, and 8 s; see Table 1 for a detailed description of the conditions). All tone sequences were composed of 250-ms-long piano tones presented sequentially without interstimulus interval. The 2 sequences could be either the same or different (50% of each trial type). For “different” trials, the second sequence differed by a single tone altering the melodic contour (Fig 1A). For the 6-tone melodies, 120 different tone sequences were created using 8 piano tones differing in pitch height (Cubase software, melodies from [55]); all used tones belonged to the key of C Major (C3, D3, E3, F3, G3, A3, B3, C4). For the 3-tone sequences, 60 different tone sequences were created using the same pool of piano tones (material from [55,56]). Procedure Presentation software (Neurobehavioral Systems, Albany, CA, USA) was used for the delivery of the experimental protocol to present the auditory stimuli and to register button presses. For each trial, participants listened binaurally (presented with headphone at a comfortable listen- ing level) to the first 3- or 6-tone sequence with a total respective duration of 750 or 1,500 ms (encoding, S1), followed by a silent retention period (2, 4, or 8 s), and then the second sequence (retrieval, S2, 750 or 1,500 ms duration). Conditions were counterbalanced across participants. Participants were informed of the block order and were asked to indicate their answers by pressing one of 2 keys with their right hand after the end of S2. Their responses were recorded during the first 2 s of the intertrial interval, whose random duration was com- prised between 2.5 and 3 s. No feedback was given during the experiment. Each block of the task included 30 trials (15 “same” trials and 15 “different” trials for each condition). Within each block, the trials were presented in a pseudorandomized order; the same trial type (i.e., “same” or “different”) could not be repeated more than 3 times in a row. Before the first ses- sion, participants performed a set of 10 practice trials (with melodies not used in the main experiment). Analysis of behavioral data Task performance was measured with d prime (Signal Detection Theory). RTs were measured from the end of S2. Behavioral data were analyzed with nonparametric repeated measures ANOVA (Friedman) and Wilcoxon rank test (see Results). The IES was calculated for each trial. IES is computed by normalizing, at the single trial scale, the participant RT by their respective percentage of correct responses in each condition. As compared to RTs, this beha- vioural metric increases the variability of behavioural scores with a low score representing a short RT and a high percentage of correctness [43]. Correlation analysis between performance at the single trial level and brain data (PAC values; see below) were performed using IES. Localization of depth electrodes In each patient’s brain, 10 to 16 semirigid, multilead electrodes were stereotactically implanted. The SEEG electrodes had a diameter of 0.8 mm and, depending on the target structure, consist of 10 to 15 contact leads 2.0 mm wide and 1.5 mm apart (DIXI Medical Instruments). All par- ticipants underwent two 3D anatomical MPRAGE T1-weighted MRI scan on a 1.5T Siemens Sonata scanner or on a 3T Siemens Trio (Siemens AG, Erlangen, Germany) before implanta- tion and just after the SEEG implantation. The anatomical volume consisted of 160 sagittal PLOS Biology | https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 15 / 24 PLOS BIOLOGY Cross-frequency coupling enables integration and memory of auditory information in the human brain slices with 1 mm3 voxel, covering the whole brain. The scalp and cortical surfaces were extracted from the T1-weighted anatomical MRI. All electrode contacts were identified on the post-implantation MRI showing the electrodes and coregistered on a pre-implantation MRI (ImaGIN toolbox; https://f-tract.eu/software/imagin/). MNI coordinates were computed using the SPM (http://www.fil.ion.ucl.ac.uk/spm/) toolbox. In addition to MNI coordinates, we computed the localization of the SEEG contacts in the AAL3 atlas [73]. Intracranial recordings Intracranial recordings were conducted using a video-SEEG monitoring system (Micromed), which allowed the simultaneous data recording from 128 depth EEG electrode sites (identical acquisition system and acquisition parameters in the 2 recording sites). The data were band- pass filtered online from 0.1 to 200 Hz and sampled at 512 Hz for all patients. At the time of acquisition, data were recorded using a reference electrode located in white matter, and each electrode trace was subsequently re-referenced to its immediate neighbour (bipolar deriva- tions). This bipolar montage has several advantages over common referencing. It helps elimi- nating signal artifacts common to adjacent electrode contacts (such as the 50 Hz mains artifact or distant physiological artifacts) and achieves a high local specificity by cancelling out effects of distant sources that spread equally to both adjacent sites through volume conduction. The spatial resolution achieved by the bipolar SEEG is estimated to be on the order of 3 mm [74]. Preprocessing SEEG data were preprocessed and visually checked to reject contacts contaminated by patho- logical epileptic activity or environmental artifacts. Powerline contamination of the raw data (main 50 Hz, harmonics 100 and 150 Hz) was reduced using notch filtering. Then, data were epoched to create trials with a window of 1,000 ms before the onset of S1 and 500 ms after the end of the last stimulus of the S2 sequence. SEEG contacts showing signal values exceeding 1,500 μV during the trial time window were excluded from the analysis: As a result, between 17 and 30 trials were kept for each participant and condition. Time-frequency analysis in Heschl’s gyrus and hippocampus We first performed time-frequency Morlet analysis for the SEEG contacts located in the right and left Heschl’s gyrus and bilateral hippocampus (according to the AAL atlas). This analysis was done to define the frequency bands of interest for the whole brain Hilbert’s analysis and to define the frequency for phase and frequency for amplitude for the PAC analysis. Time-fre- quency Morlet analysis was computed based on a wavelet transform of the signals [75]. The wavelet family was defined by (f0 /sf) = 7 with f0 ranging from 1 to 150 Hz in 1 Hz steps. The time-frequency wavelet transform was applied to each SEEG contact, each trial, and then power was averaged across trials, resulting in an estimate of oscillatory power at each time sample and each frequency bin between 1 and 150 Hz. Note that both evoked and induced activity were estimated. We then performed a normalization (z-scoring) with −1,000 to 0 ms preceding the presentation of the S1 sequence as baseline. Time-frequency plots of SEEG con- tacts were regrouped in left and right Heschl’s gyrus and bilateral hippocampus across partici- pants using the AAL3 brain atlas. By doing so, we were able to investigate the data of several participants on one time-frequency map per area. Normalized and averaged time-frequency maps of the auditory cortex and hippocampus were used to define the frequency for phase and frequency for amplitude for the PAC analysis (see below). Frequency for amplitude was defined from 30 Hz to 90 Hz as it matched with the amplitude of time-frequency maps gamma bursts in the auditory cortex (see also [18] for similar parameters). Frequency for phase was PLOS Biology | https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 16 / 24 PLOS BIOLOGY Cross-frequency coupling enables integration and memory of auditory information in the human brain defined at 4 Hz because sustained theta power at 4 Hz was observed in the auditory cortex dur- ing encoding (Fig 1D), and this frequency matched the frequency of presentation of the stimuli. Hilbert transform Once the frequency bands of interest were defined, we aimed to investigate if fluctuation of theta and gamma power were associated to memory processes (as compared to perception). In order to perform this analysis at the whole brain level and to reduce the dimension of the data, we computed for each participant the Hilbert transform for correct trials for each period of the STM task (encoding, retention, and retrieval, average in time for each time period; see Table 1) and the corresponding periods of the perception task. We extracted the magnitude of theta activity at 4 Hz and gamma activity between 30 to 90 Hz for each trial for each SEEG contact. These data were then used to contrast brain activity in the memory conditions and baseline and to contrast brain activity in the memory and perception conditions using permutation tests as implemented in MATLAB. Contrasts with baseline were corrected for multiple com- parison using FDR corrections. Memory versus perception contrast were corrected with a cluster procedure (see below). Phase amplitude coupling Theta-gamma PAC was computed using the method developed by [76]. Frequency for phase and frequencies for amplitudes were defined by a power spectrum density analysis on SEEG contacts located in the auditory cortex and in the hippocampus and computed over the total duration of a trial time window (0 to 5.5 s for the 6 tones, 2 s memory condition as this condi- tion was performed by all 16 participants). Frequency for phase was selected as the frequency showing the highest peak in the theta band (4 to 8 Hz) in the auditory cortex and hippocampus (see Fig 1D and 1E) and frequency for amplitude was defined as a 60-Hz-width frequency band centered on the highest peak in the gamma band (peak at 60 Hz ± 30 Hz resulting in a band between 30 and 90 Hz) in the auditory cortex. Based on these results (see Fig 1D and 1E), we used 4 Hz as the frequency for phase (frequency of presentation of stimuli) and 30 to 90 Hz as the frequency for amplitude for the PAC analyses. As no high gamma peak emerged in this PSD analysis, we did not investigate PAC for frequencies above 90 Hz. 3D representation and cluster procedure For all PAC analyses and Hilbert data, significant SEEG contacts were plotted on a MNI MRI volume using marsbar and SPM functions [77]. To do so, we extracted the MNI coordinate of each SEEG contact and represent the oscillatory magnitude and PAC values on spheres of 4 mm radius in the MRI volume. PAC plots were corrected with a cluster approach: by consider- ing as significant only the contacts that were overlapping across at least 2 participants or 2 SEEG contacts in the MRI volume. Multivariate analyses Multivariate analyses were performed using MATLAB and SVM implementation (https:// www.mathworks.com/help/stats/fitcecoc.html). A linear classifier was chosen as SEEG data contains many more features than examples, and classification of such data is generally suscep- tible to overfitting. One way of alleviating the danger of overfitting is to choose a simple func- tion (such as a linear function) for classification, where each feature affects the prediction solely via its weight and without interaction with other features (rather than more complex PLOS Biology | https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 17 / 24 PLOS BIOLOGY Cross-frequency coupling enables integration and memory of auditory information in the human brain classifiers, such as nonlinear SVMs or artificial neural networks, which can let interactions between features and nonlinear functions thereof drive the prediction). Our strategy was to use the SVM classifier with 3-fold cross-validation to classify correct and incorrect memory trials of all memory conditions, using the SEEG contact as features. For each participant, the 1/ model is trained only on data 2/3 of the trials to predict whether each trial in the remainingAU : Pleasecheckandconfirmthattheeditto}Foreachparticipant; themodelistrainedonlyondata:::}didnotaltertheintendedmeaningofthesentence: 3 set of trials is correct or incorrect. The procedure is repeated 3 further times to estimate the classification performance across the full set folds. As all participants had more correct than incorrect trials for all memory conditions, we made a random selection of the correct trials (to match the number of incorrect trials for each condition) to train and test the classifier. Then, we repeated this analysis 100 times with 100 different random selection of correct trials for each participant. SVM’s performance was evaluated using the output of the 100 models (accu- racy minus chance) for each subject. For each subject with above chance decoding accuracy, we extracted the features weights (zscore) to evaluate the relative contribution of each feature (SEEG contact) in the classification. Phase consistency analysis We extracted the PAC consistency across trials and participants in the brain regions where the PAC strength was correlated with behavioural performance (see Figs 4A and 5A and S5 Table). For each trial, we extracted the magnitude of gamma oscillations (30 to 90 Hz) as a function of the phase of the theta oscillation (4 Hz; phase divided into 8 bins). We then extracted the intertrial phase locking (PLV) on these signals using PLV functions available in Brainstorm. To identify whether significant preferred coupling phase could be identified, we extracted for each SEEG contact the gamma power for 8 different phase bins of the theta cycle. To define if a preferred coupling phase can be identified across trials and participant for each region, we used LMMs. The variability between participants was modeled by defining by-par- ticipant random intercepts. LMMs were performed in R 3.4.1 using the lme4 [78] and car [79] packages. Both fixed and random factors were considered in statistical modeling. Wald chi- squared tests were used for fixed effects in LMM [79]. The fixed effect represents the mean effect across all participants after accounting for variability. We considered the results of the main analyses significant at p < .05. When we found a significant main effect, post hoc honest significant difference (HSD) tests were systematically performed using the R emmeans pack- age (emmeans version 1.6.3). P values were considered as significant at p < .05 and were adjusted for the number of comparisons performed. More precisely, to avoid increased Type I error when multiple comparisons were performed, the p-value of the Tukey HSD test was adjusted using the Tukey method for comparing the given number of estimates. Supporting information S1 Fig. Brain oscillations displayed with a logarithmic scale for the frequency axis. (A) T- values in the time-frequency domain (t test relative to baseline −1,000 to 0 before stimulus onset, FDR corrected in time and frequency domains) of SEEG contacts located in the right and left Heschl’s gyrus (displayed on the single subject T1 in the MNI space provided by SPM12) for a trial time window (−1,000 to 6,000 ms) for the condition 6-tone memory load, 2 s retention period (n = 5). (B) T-values in the time-frequency domain (t test relative to baseline −1,000 to 0 before stimulus onset, FDR corrected in time and frequency domains) of SEEG contacts located in the right and left hippocampus (displayed on the single subject T1 in the MNI space provided by SPM12) for a trial time window (−1,000 to 6,000 ms) for the condition 6-tone memory load, 2 s retention period (n = 14). (PDF) PLOS Biology | https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 18 / 24 PLOS BIOLOGY Cross-frequency coupling enables integration and memory of auditory information in the human brain S2 Fig. Theta (orange) and gamma (red) magnitude averaged over SEEG contacts located in regions showing increased power relative to baseline during retention presented as a function of task (memory, perception). NS, nonsignificant. (PDF) S3 Fig. Memory minus perception (the colormap represents the difference in PAC strength between memory and perception trial—note that the contrast is not significant) for the co- modulogram in SEEG contacts that had previously shown an increase in theta and gamma power identified in Fig 1F, retention period). (PDF) S4 Fig. Theta-gamma PAC in the hippocampus and ventral auditory stream correlates with behavior. Left panel: SEEG contacts showing a positive (hot colormap) and negative (blue colormap) relationship between theta-gamma PAC and performance using data from conditions performed by all 16 participants (6 tones encoding 2 s retention and 6 tones encod- ing 8 s retention). Results are displayed on the single subject T1 in the MNI space provided by SPM12. (PDF) S1 Table. Regions and coordinates Fig 1D: Heschl’s gyrus. (PDF) S2 Table. Regions and coordinates Fig 1E: Hippocampal regions. (PDF) S3 Table. Regions and coordinates Fig 2B: PAC memory vs. perception L, Left; R, Right; Sup, Superior; Mid, Middle; Inf, Inferior. (PDF) S4 Table. Coordinates of the maximum value (zscore) of the features weights for each par- ticipant with significant above chance decoding accuracy—Fig 3C, L, Left; R, Right; Sup, Superior; Mid, Middle; Inf, Inferior; Tri, Triangular. (PDF) S5 Table. Regions and coordinates Fig 4A: Correlation between PAC and IES, L, Left; R, Right; Sup, Superior; Mid, Middle; Inf, Inferior; Oper, Opercular. (PDF) S6 Table. Regions and coordinates Fig 4B: Correlation between PAC and IES. (PDF) S7 Table. Post hoc tests of Fig 5C: Left STS. (PDF) S8 Table. Post hoc tests of Fig 5C: Left IFG. (PDF) S9 Table. Post hoc tests of Fig 5C: Left hippocampus. (PDF) S10 Table. Post hoc tests of Fig 5C: Right ITG. (PDF) PLOS Biology | https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 19 / 24 PLOS BIOLOGY Cross-frequency coupling enables integration and memory of auditory information in the human brain Acknowledgments We thank Luc H. Arnal for his insightful comments on a previous version of this manuscript. Author Contributions Conceptualization: Anne Caclin, Jean-Philippe Lachaux, Barbara Tillmann, Philippe Albouy. Data curation: Arthur Borderie, Anne Caclin, Marcela Perrone-Bertollotti, Barbara Tillmann, Philippe Albouy. Formal analysis: Arthur Borderie, Roxane S. Hoyer, Philippe Albouy. Funding acquisition: Barbara Tillmann, Philippe Albouy. Investigation: Arthur Borderie, Marcela Perrone-Bertollotti, Philippe Albouy. Methodology: Arthur Borderie, Jean-Philippe Lachaux, Philippe Kahane, He´lène Catenoix, Philippe Albouy. Project administration: Anne Caclin, Jean-Philippe Lachaux, Philippe Kahane, He´lène Cate- noix, Barbara Tillmann, Philippe Albouy. Resources: Anne Caclin, Jean-Philippe Lachaux, Philippe Kahane, He´lène Catenoix, Barbara Tillmann, Philippe Albouy. Software: Philippe Albouy. Supervision: Anne Caclin, Barbara Tillmann, Philippe Albouy. Validation: Philippe Albouy. Visualization: Arthur Borderie, Philippe Albouy. Writing – original draft: Arthur Borderie, Philippe Albouy. Writing – review & editing: Arthur Borderie, Anne Caclin, Jean-Philippe Lachaux, Roxane S. Hoyer, Philippe Kahane, He´lène Catenoix, Barbara Tillmann, Philippe Albouy. References 1. Scoville WB, Milner B. Loss of recent memory after bilateral hippocampal lesions. J Neurol Neurosurg Psychiatry. 1957; 20(1):11–21. https://doi.org/10.1136/jnnp.20.1.11 PMID: 13406589 2. Spiers HJ, Maguire EA, Burgess N. Hippocampal amnesia. Neurocase. 2001; 7(5):357–382. https://doi. org/10.1076/neur.7.5.357.16245 PMID: 11744778 3. Baddeley AD, Warrington EK. Amnesia and the distinction between long-and short-term memory. J Verb Learning Verb Behav. 1970; 9(2):176–189. 4. Cave CB, Squire LR. Intact verbal and nonverbal short-term memory following damage to the human hippocampus. Hippocampus. 1992; 2(2):151–163. https://doi.org/10.1002/hipo.450020207 PMID: 1308180 5. Wickelgren WA. Sparing of short-term memory in an amnesic patient: Implications for strength theory of memory. Neuropsychologia. 1968; 6(3):235–244. 6. Baddeley A, Jarrold C, Vargha-Khadem F. Working memory and the hippocampus. J Cogn Neurosci. 2011; 23(12):3855–3861. https://doi.org/10.1162/jocn_a_00066 PMID: 21671734 7. Jeneson A, Squire LR. Working memory, long-term memory, and medial temporal lobe function. Learn Mem. 2012; 19(1):15–25. https://doi.org/10.1101/lm.024018.111 PMID: 22180053 8. Atkinson RC, Shiffrin RM. Human memory: a proposed system and its control processes. In: Spence KW, editor. The Psychology of Learning and Motivation: Advances in Research and Theory. 2. New York: Academic Press; 1968. p. 89–195. 9. James W. The Principles of Psychology. Holt H, editor. New York; 1890. PLOS Biology | https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 20 / 24 PLOS BIOLOGY Cross-frequency coupling enables integration and memory of auditory information in the human brain 10. Michels L, Bucher K, Luchinger R, Klaver P, Martin E, Jeanmonod D, et al. Simultaneous EEG-fMRI during a working memory task: modulations in low and high frequency bands. PLoS ONE. 2010; 5(4): e10298. https://doi.org/10.1371/journal.pone.0010298 PMID: 20421978 11. Zarahn E, Rakitin B, Abela D, Flynn J, Stern Y. Positive evidence against human hippocampal involve- ment in working memory maintenance of familiar stimuli. Cereb Cortex. 2005; 15(3):303–316. https:// doi.org/10.1093/cercor/bhh132 PMID: 15342440 12. Buffalo EA, Reber PJ, Squire LR. The human perirhinal cortex and recognition memory. Hippocampus. 1998; 8(4):330–339. https://doi.org/10.1002/(SICI)1098-1063(1998)8:4<330::AID-HIPO3>3.0.CO;2-L PMID: 9744420 13. Holdstock JS, Mayes AR, Gong QY, Roberts N, Kapur N. Item recognition is less impaired than recall and associative recognition in a patient with selective hippocampal damage. Hippocampus. 2005; 15 (2):203–215. https://doi.org/10.1002/hipo.20046 PMID: 15390152 14. Olson IR, Moore KS, Stark M, Chatterjee A. Visual working memory is impaired when the medial tempo- ral lobe is damaged. J Cogn Neurosci. 2006; 18(7):1087–1097. https://doi.org/10.1162/jocn.2006.18.7. 1087 PMID: 16839283 15. Kumar S, Joseph S, Gander PE, Barascud N, Halpern AR, Griffiths TD. A Brain System for Auditory Working Memory. J Neurosci. 2016; 36(16):4492–4505. https://doi.org/10.1523/JNEUROSCI.4341-14. 2016 PMID: 27098693 16. Ranganath C D’Esposito M. Medial temporal lobe activity associated with active maintenance of novel information. Neuron. 2001; 31(5):865–873. 17. Nichols EA, Kao YC, Verfaellie M, Gabrieli JD. Working memory and long-term memory for faces: Evi- dence from fMRI and global amnesia for involvement of the medial temporal lobes. Hippocampus. 2006; 16(7):604–616. https://doi.org/10.1002/hipo.20190 PMID: 16770797 18. Axmacher N, Henseler MM, Jensen O, Weinreich I, Elger CE, Fell J. Cross-frequency coupling supports multi-item working memory in the human hippocampus. Proc Natl Acad Sci U S A. 2010; 107(7):3228– 3233. https://doi.org/10.1073/pnas.0911531107 PMID: 20133762 19. Axmacher N, Mormann F, Fernandez G, Cohen MX, Elger CE, Fell J. Sustained neural activity patterns during working memory in the human medial temporal lobe. J Neurosci. 2007; 27(29):7807–7816. https://doi.org/10.1523/JNEUROSCI.0962-07.2007 PMID: 17634374 20. Bahramisharif A, Jensen O, Jacobs J, Lisman J. Serial representation of items during working memory maintenance at letter-selective cortical sites. PLoS Biol. 2018; 16(8):e2003805. https://doi.org/10.1371/ journal.pbio.2003805 PMID: 30110320 21. van Vugt MK, Schulze-Bonhage A, Litt B, Brandt A, Kahana MJ. Hippocampal gamma oscillations increase with memory load. J Neurosci. 2010; 30(7):2694–2699. https://doi.org/10.1523/JNEUROSCI. 0567-09.2010 PMID: 20164353 22. Boran E, Fedele T, Klaver P, Hilfiker P, Stieglitz L, Grunwald T, et al. Persistent hippocampal neural fir- ing and hippocampal-cortical coupling predict verbal working memory load. Sci Adv. 2019; 5(3): eaav3687. https://doi.org/10.1126/sciadv.aav3687 PMID: 30944858 23. Kornblith S, Quian Quiroga R, Koch C, Fried I, Mormann F. Persistent Single-Neuron Activity during Working Memory in the Human Medial Temporal Lobe. Curr Biol. 2017; 27(7):1026–1032. https://doi. org/10.1016/j.cub.2017.02.013 PMID: 28318972 24. Billig AJ, Lad M, Sedley W, Griffiths TD. The hearing hippocampus. Prog Neurobiol. 2022; 218:102326. https://doi.org/10.1016/j.pneurobio.2022.102326 PMID: 35870677 25. 26. 27. Lisman J, Buzsaki G, Eichenbaum H, Nadel L, Ranganath C, Redish AD. Viewpoints: how the hippo- campus contributes to memory, navigation and cognition. Nat Neurosci. 2017; 20(11):1434–1447. https://doi.org/10.1038/nn.4661 PMID: 29073641 Fell J, Klaver P, Lehnertz K, Grunwald T, Schaller C, Elger CE, et al. Human memory formation is accompanied by rhinal-hippocampal coupling and decoupling. Nat Neurosci. 2001; 4(12):1259–1264. https://doi.org/10.1038/nn759 PMID: 11694886 Fell J, Ludowig E, Staresina BP, Wagner T, Kranz T, Elger CE, et al. Medial temporal theta/alpha power enhancement precedes successful memory encoding: evidence based on intracranial EEG. J Neurosci. 2011; 31(14):5392–5397. https://doi.org/10.1523/JNEUROSCI.3668-10.2011 PMID: 21471374 28. Colgin LL, Moser EI. Gamma oscillations in the hippocampus. Physiology (Bethesda). 2010; 25 (5):319–329. https://doi.org/10.1152/physiol.00021.2010 PMID: 20940437 29. Canolty RT, Edwards E, Dalal SS, Soltani M, Nagarajan SS, Kirsch HE, et al. High gamma power is phase-locked to theta oscillations in human neocortex. Science. 2006; 313(5793):1626–1628. https:// doi.org/10.1126/science.1128115 PMID: 16973878 30. Canolty RT, Knight RT. The functional role of cross-frequency coupling. Trends Cogn Sci. 2010; 14 (11):506–515. https://doi.org/10.1016/j.tics.2010.09.001 PMID: 20932795 PLOS Biology | https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 21 / 24 PLOS BIOLOGY Cross-frequency coupling enables integration and memory of auditory information in the human brain 31. 32. 33. Lisman JE, Jensen O. The theta-gamma neural code. Neuron. 2013; 77(6):1002–1016. Lisman JE, Idiart MA. Storage of 7 +/- 2 short-term memories in oscillatory subcycles. Science. 1995; 267(5203):1512–1515. https://doi.org/10.1126/science.7878473 PMID: 7878473 Fuentemilla L, Penny WD, Cashdollar N, Bunzeck N, Duzel E. Theta-coupled periodic replay in working memory. Curr Biol. 2010; 20(7):606–612. https://doi.org/10.1016/j.cub.2010.01.057 PMID: 20303266 34. Bergmann TO, Born J. Phase-Amplitude Coupling: A General Mechanism for Memory Processing and Synaptic Plasticity? Neuron. 2018; 97(1):10–13. https://doi.org/10.1016/j.neuron.2017.12.023 PMID: 29301097 35. Helfrich RF, Mander BA, Jagust WJ, Knight RT, Walker MP. Old Brains Come Uncoupled in Sleep: Slow Wave-Spindle Synchrony, Brain Atrophy, and Forgetting. Neuron. 2018; 97(1):221–30 e4. https:// doi.org/10.1016/j.neuron.2017.11.020 PMID: 29249289 36. Mormann F, Fell J, Axmacher N, Weber B, Lehnertz K, Elger CE, et al. Phase/amplitude reset and theta-gamma interaction in the human medial temporal lobe during a continuous word recognition mem- ory task. Hippocampus. 2005; 15(7):890–900. https://doi.org/10.1002/hipo.20117 PMID: 16114010 37. Chaieb L, Leszczynski M, Axmacher N, Hohne M, Elger CE, Fell J. Theta-gamma phase-phase cou- pling during working memory maintenance in the human hippocampus. Cogn Neurosci. 2015; 6 (4):149–157. https://doi.org/10.1080/17588928.2015.1058254 PMID: 26101947 38. 39. Leszczynski M, Fell J, Axmacher N. Rhythmic Working Memory Activation in the Human Hippocampus. Cell Rep. 2015; 13(6):1272–1282. https://doi.org/10.1016/j.celrep.2015.09.081 PMID: 26527004 Fontolan L, Morillon B, Liegeois-Chauvel C, Giraud AL. The contribution of frequency-specific activity to hierarchical information processing in the human auditory cortex. Nat Commun. 2014; 5:4694. https:// doi.org/10.1038/ncomms5694 PMID: 25178489 40. Hyafil A, Fontolan L, Kabdebon C, Gutkin B, Giraud AL. Speech encoding by coupled cortical theta and gamma oscillations. Elife. 2015; 4:e06213. 41. 42. Lakatos P, Shah AS, Knuth KH, Ulbert I, Karmos G, Schroeder CE. An oscillatory hierarchy controlling neuronal excitability and stimulus processing in the auditory cortex. J Neurophysiol. 2005; 94(3):1904– 1911. https://doi.org/10.1152/jn.00263.2005 PMID: 15901760 van der Meij R, Kahana M, Maris E. Phase-amplitude coupling in human electrocorticography is spa- tially distributed and phase diverse. J Neurosci. 2012; 32(1):111–123. https://doi.org/10.1523/ JNEUROSCI.4816-11.2012 PMID: 22219274 43. Bruyer R, Brysbaert M. Combining Speed and Accuracy in Cognitive Psychology: Is the Inverse Effi- ciency Score (IES) a Better Dependent Variable than the Mean Reaction Time (RT) and the Percentage Of Errors (PE)? Psychologica Belgica. 2011; 51(1):1–5. 44. Williamson VJ, McDonald C, Deutsch D, Griffiths TD, Stewart L. Faster decline of pitch memory over time in congenital amusia. Adv Cogn Psychol. 2010; 6:15–22. https://doi.org/10.2478/v10053-008- 0073-5 PMID: 20689638 45. Albouy P, Cousineau M, Caclin A, Tillmann B, Peretz I. Impaired encoding of rapid pitch information underlies perception and memory deficits in congenital amusia. Sci Rep. 2016; 6:18861. https://doi.org/ 10.1038/srep18861 PMID: 26732511 46. Siegel M, Donner TH, Engel AK. Spectral fingerprints of large-scale neuronal interactions. Nat Rev Neu- rosci. 2012; 13(2):121–134. https://doi.org/10.1038/nrn3137 PMID: 22233726 47. Fries P. Rhythms for Cognition: Communication through Coherence. Neuron. 2015; 88(1):220–235. https://doi.org/10.1016/j.neuron.2015.09.034 PMID: 26447583 48. Albouy P, Baillet S, Zatorre RJ. Driving working memory with frequency-tuned noninvasive brain stimu- lation. Ann N Y Acad Sci. 2018. https://doi.org/10.1111/nyas.13664 PMID: 29707781 49. Albouy P, Martinez-Moreno ZE, Hoyer RS, Zatorre RJ, Baillet S. Supramodality of neural entrainment: Rhythmic visual stimulation causally enhances auditory working memory performance. Sci Adv. 2022; 8(8):eabj9782. https://doi.org/10.1126/sciadv.abj9782 PMID: 35196074 50. Albouy P, Weiss A, Baillet S, Zatorre RJ. Selective Entrainment of Theta Oscillations in the Dorsal Stream Causally Enhances Auditory Working Memory Performance. Neuron. 2017; 94(1):193–206 e5. https://doi.org/10.1016/j.neuron.2017.03.015 PMID: 28343866 51. Roux F, Uhlhaas PJ. Working memory and neural oscillations: alpha-gamma versus theta-gamma codes for distinct WM information? Trends Cogn Sci. 2014; 18(1):16–25. 52. Backus AR, Schoffelen JM, Szebenyi S, Hanslmayr S, Doeller CF. Hippocampal-Prefrontal Theta Oscil- lations Support Memory Integration. Curr Biol. 2016; 26(4):450–457. https://doi.org/10.1016/j.cub.2015. 12.048 PMID: 26832442 PLOS Biology | https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 22 / 24 PLOS BIOLOGY Cross-frequency coupling enables integration and memory of auditory information in the human brain 53. Violante IR, Li LM, Carmichael DW, Lorenz R, Leech R, Hampshire A, et al. Externally induced fronto- parietal synchronization modulates network dynamics and enhances working memory performance. Elife. 2017:6. https://doi.org/10.7554/eLife.22001 PMID: 28288700 54. Hanslmayr S, Axmacher N, Inman CS. Modulating Human Memory via Entrainment of Brain Oscilla- tions. Trends Neurosci. 2019; 42(7):485–499. https://doi.org/10.1016/j.tins.2019.04.004 PMID: 31178076 55. Albouy P, Mattout J, Bouet R, Maby E, Sanchez G, Aguera PE, et al. Impaired pitch perception and memory in congenital amusia: the deficit starts in the auditory cortex. Brain. 2013; 136(Pt 5):1639– 1661. https://doi.org/10.1093/brain/awt082 PMID: 23616587 56. Albouy P, Peretz I, Bermudez P, Zatorre RJ, Tillmann B, Caclin A. Specialized neural dynamics for ver- bal and tonal memory: fMRI evidence in congenital amusia. Hum Brain Mapp. 2019; 40(3):855–867. https://doi.org/10.1002/hbm.24416 PMID: 30381866 57. Malinovitch T, Albouy P, Zatorre RJ, Ahissar M. Training allows switching from limited-capacity manipu- lations to large-capacity perceptual processing. Cereb Cortex. 2023; 33(5):1826–1842AU : Pleasenotethatdetailshavebeenaddedtoref :57:Pleasecheckandconfirmthatthesearecorrect: 10.1093/cercor/bhac175 PMID: 35511687 . https://doi.org/ 58. Keller AS, Payne L, Sekuler R. Characterizing the roles of alpha and theta oscillations in multisensory attention. Neuropsychologia. 2017; 99:48–63. https://doi.org/10.1016/j.neuropsychologia.2017.02.021 PMID: 28259771 59. Kaminski J, Sullivan S, Chung JM, Ross IB, Mamelak AN, Rutishauser U. Erratum: Persistently active neurons in human medial frontal and medial temporal lobe support working memory. Nat Neurosci. 2017; 20(8):1189. https://doi.org/10.1038/nn0817-1189d PMID: 28745722 60. Poppenk J, Evensmoen HR, Moscovitch M, Nadel L. Long-axis specialization of the human hippocam- pus. Trends Cogn Sci. 2013; 17(5):230–240. https://doi.org/10.1016/j.tics.2013.03.005 PMID: 23597720 61. Strange BA, Witter MP, Lein ES, Moser EI. Functional organization of the hippocampal longitudinal axis. Nat Rev Neurosci. 2014; 15(10):655–669. https://doi.org/10.1038/nrn3785 PMID: 25234264 62. Dimakopoulos V, Megevand P, Stieglitz LH, Imbach L, Sarnthein J. Information flows from hippocam- pus to auditory cortex during replay of verbal working memory items. Elife. 2022:11. https://doi.org/10. 7554/eLife.78677 PMID: 35960169 63. Albouy P, Caclin A, Norman-Haignere SV, Leveque Y, Peretz I, Tillmann B, et al. Decoding Task- Related Functional Brain Imaging Data to Identify Developmental Disorders: The Case of Congenital Amusia. Front Neurosci. 2019; 13:1165. https://doi.org/10.3389/fnins.2019.01165 PMID: 31736698 64. Albouy P, Mattout J, Sanchez G, Tillmann B, Caclin A. Altered retrieval of melodic information in con- genital amusia: insights from dynamic causal modeling of MEG data. Front Hum Neurosci. 2015; 9:20. https://doi.org/10.3389/fnhum.2015.00020 PMID: 25698955 65. Samiee S, Vuvan D, Florin E, Albouy P, Peretz I, Baillet S. Cross-frequency brain network dynamics support pitch change detection. J Neurosci. 2022; 42(18):3823–3835. https://doi.org/10.1523/ JNEUROSCI.0630-21.2022 PMID: 35351829 66. 67. Zatorre RJ, Belin P, Penhune VB. Structure and function of auditory cortex: music and speech. Trends Cogn Sci. 2002; 6(1):37–46. https://doi.org/10.1016/s1364-6613(00)01816-7 PMID: 11849614 Zatorre RJ, Evans AC, Meyer E. Neural mechanisms underlying melodic perception and memory for pitch. J Neurosci. 1994; 14(4):1908–1919. https://doi.org/10.1523/JNEUROSCI.14-04-01908.1994 PMID: 8158246 68. Gaab N, Gaser C, Zaehle T, Jancke L, Schlaug G. Functional anatomy of pitch memory—an fMRI study with sparse temporal sampling. Neuroimage. 2003; 19(4):1417–1426. https://doi.org/10.1016/s1053- 8119(03)00224-6 PMID: 12948699 69. Schulze K, Gaab N, Schlaug G. Perceiving pitch absolutely: comparing absolute and relative pitch pos- sessors in a pitch memory task. BMC Neurosci. 2009; 10:106. https://doi.org/10.1186/1471-2202-10- 106 PMID: 19712445 70. Treder MS, Charest I, Michelmann S, Martin-Buro MC, Roux F, Carceller-Benito F, et al. The hippocam- pus as the switchboard between perception and memory. Proc Natl Acad Sci U S A. 2021; 118(50): e2114171118. https://doi.org/10.1073/pnas.2114171118 PMID: 34880133 71. Alekseichuk I, Turi Z, Amador de Lara G, Antal A, Paulus W. Spatial Working Memory in Humans Depends on Theta and High Gamma Synchronization in the Prefrontal Cortex. Curr Biol. 2016; 26 (12):1513–1521. https://doi.org/10.1016/j.cub.2016.04.035 PMID: 27238283 72. Buzsaki G, Draguhn A. Neuronal oscillations in cortical networks. Science. 2004; 304(5679):1926– 1929. https://doi.org/10.1126/science.1099745 PMID: 15218136 73. Tzourio-Mazoyer N, Landeau B, Papathanassiou D, Crivello F, Etard O, Delcroix N, et al. Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI PLOS Biology | https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 23 / 24 PLOS BIOLOGY Cross-frequency coupling enables integration and memory of auditory information in the human brain 74. 75. single-subject brain. Neuroimage. 2002; 15(1):273–289. https://doi.org/10.1006/nimg.2001.0978 PMID: 11771995 Jerbi K, Ossandon T, Hamame CM, Senova S, Dalal SS, Jung J, et al. Task-related gamma-band dynamics from an intracerebral perspective: review and implications for surface EEG and MEG. Hum Brain Mapp. 2009; 30(6):1758–1771. https://doi.org/10.1002/hbm.20750 PMID: 19343801 Tallon-Baudry C, Bertrand O. Oscillatory gamma activity in humans and its role in object representation. Trends Cogn Sci. 1999; 3(4):151–162. https://doi.org/10.1016/s1364-6613(99)01299-1 PMID: 10322469 76. Ozkurt TE, Schnitzler A. A critical note on the definition of phase-amplitude cross-frequency coupling. J Neurosci Methods. 2011; 201(2):438–443. https://doi.org/10.1016/j.jneumeth.2011.08.014 PMID: 21871489 77. Brett M, Anton JL, Valabregue R, Poline JB. Region of interest analysis using the MarsBar toolbox for SPM 99. Neuroimage. 2002; 16(Suppl 1:S497). 78. Bates D, Ma¨chler M, Bolker B, Walker S. Fitting Linear Mixed-Effects Models Using lme4. J Stat Softw. 2015; 67(1):1–48. 79. Fox J, Weisberg S. An R Companion to Applied Regression. Thousand Oaks, CA: Sage; 2019. PLOS Biology | https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 24 / 24 PLOS BIOLOGY
null
10.12688_f1000research.16224.3.pdf
Data availability Pediococcus acidilactici strain DS32 16S ribosomal RNA gene, partial sequence, obtained during this study. GenBank accession http://identifiers.org/ncbigi/ GI:1481059229.
all the source data underlying the results available to ensure full reproducibility? Yes Are the conclusions drawn adequately supported Data availability Pediococcus acidilactici strain DS32 16S ribosomal RNA gene, partial sequence, obtained during this study. GenBank accession number MH938236: http://identifiers.org/ncbigi/ GI:1481059229 . Grant information This research was supported by Ministry of Research, Technology and Higher Education Republic of Indonesia through Master of Education Towards Doctoral Scholarship Program for Excellence Undergraduate and the support through World Class Professor Program Scheme-B No. 123.57/D2.3/KP/2018. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
F1000Research 2019, 7:1663 Last updated: 12 MAR 2020 RESEARCH ARTICLE    Molecular identification and phylogenetic analysis of GABA-producing lactic acid bacteria isolated from indigenous dadih of West Sumatera, Indonesia [version 3; peer review: 2 approved, 1 approved with reservations, 1 not approved] Lili Anggraini 1 Yetti Marlida , Wizna Wizna , Jamsari Jamsari , 4 5,6 5,6 2 2 Frederick Adzitey , Nurul Huda 3 2 , Mirzah Mirzah , 1 2 3 4 5 6 Graduate Program, Andalas University, Padang, West Sumatera, Indonesia Department of Nutrition and Feed Technology, Faculty of Animal Science, Andalas University, Padang, West Sumatera, Indonesia Department of Plant Breeding, Faculty of Agriculture, Andalas University, Padang, West Sumatera, Indonesia Department of Veterinary Science, University for Development Studies, Temale, Ghana School of Food Industry, Universiti Sultan Zainal Abidin, Kuala Nerus, Terengganu, 21300, Malaysia Faculty of Food Science and Nutrition, Universiti Malaysia Sabah, Kota Kinabalu, Sabah, 88400, Malaysia v3 First published:  19 Oct 2018, https://doi.org/10.12688/f1000research.16224.1 ) :1663 ( 7 Second version:  06 Feb 2019, https://doi.org/10.12688/f1000research.16224.2 ) :1663 ( 7 Latest published: https://doi.org/10.12688/f1000research.16224.3 )  17 Oct 2019, :1663 ( 7  Dadih (fermented buffalo milk) is a traditional Indonesian Abstract Background: food originating from West Sumatra province. The fermentation process is carried out by lactic acid bacteria (LAB), which are naturally present in buffalo milk.  Lactic acid bacteria have been reported as one of potential producers of γ-aminobutyric acid (GABA). GABA acts as a neurotransmitter inhibitor of the central nervous system.  In this study, molecular identification and phylogenetic analysis Methods: of GABA producing LAB isolated from indigenous dadih of West Sumatera were determined. Identification of the GABA-producing LAB DS15 was based on conventional polymerase chain reaction. 16S rRNA gene sequence analysis was used to identify LAB DS15. Results: approximately 1400 bp amplicon.  Phylogenetic analysis showed that LAB DS15 was query coverage to Conclusions: indigenous dadih was , with high similarity of 99% at 100% strain DSM 20284.  PCR of the 16S rRNA gene sequence of LAB DS15 gave an  It can be concluded that GABA producing LAB isolated from Pediococcus acidilactici Pediococcus acidilactici Pediococcus acidilactici . Keywords indigenous dadih, GABA, LAB, 16S rRNA gene, phylogenetic analysis Open Peer Review Reviewer Status Invited Reviewers 1 2 3 4 report report report report report version 3 (revision) 17 Oct 2019 version 2 (revision) 06 Feb 2019 version 1 19 Oct 2018 1 2 3 Qinglong Wu , Baylor College of Medicine, Houston, USA Jagadish Mahanta , Indian Council of Medical Research (ICMR), Dibrugarh, India Sahilah Abd Mutalib , Universiti Kebangsaan Malaysia (UKM), Selangor, Malaysia Page 1 of 15 F1000Research 2019, 7:1663 Last updated: 12 MAR 2020 4 Usman Pato , Riau University, Pekanbaru, Indonesia Any reports and responses or comments on the article can be found at the end of the article. Corresponding author:  Yetti Marlida ( yettimarlida@ansci.unand.ac.id ) Author roles: Anggraini L Conceptualization; Adzitey F : Investigation; Marlida Y : Writing – Review & Editing; : Supervision; Huda N Wizna W : Writing – Review & Editing : Conceptualization; Jamsari J : Conceptualization; Mirzah M : Competing interests:  No competing interests were disclosed.  This research was supported by Ministry of Research, Technology and Higher Education Republic of Indonesia through Grant information: Master of Education Towards Doctoral Scholarship Program for Excellence Undergraduate and the support through World Class Professor Program Scheme-B No. 123.57/D2.3/KP/2018. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Copyright: which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. . This is an open access article distributed under the terms of the  © 2019 Anggraini L et al Creative Commons Attribution License , How to cite this article: lactic acid bacteria isolated from indigenous dadih of West Sumatera, Indonesia [version 3; peer review: 2 approved, 1 approved with reservations, 1 not approved] et al. Molecular identification and phylogenetic analysis of GABA-producing https://doi.org/10.12688/f1000research.16224.3  Anggraini L, Marlida Y, Wizna W  F1000Research 2019, :1663 ( 7 ) First published:  19 Oct 2018, 7 :1663 ( https://doi.org/10.12688/f1000research.16224.1 ) Page 2 of 15 REVISED   Amendments from Version 2 Additional information on the reference of forward primer 63F (5’- CAG GCC TAA CAC ATG CAA GTC-3’) and reverse primer 1387R (5’-GGG CGG GGT GTA CAA GGC-3’). Updating the sentence of 1% agarose electrophoresis to 1 % (w/v) agarose electrophoresis. Mentioning the marker used 1 Kb Plus DNA ladder (ThermoFisher Scientific). Any further responses from the reviewers can be found at the end of the article acid amino γ-aminobutyric Introduction The non-proteinogenic acid (GABA) is widely found in bacteria, animals, plants, and fungi (Dhakal et al., 2012; Nonaka et al., 2017). GABA acts as a neurotransmitter inhibitor of the central nervous system (Olsen & Li, 2012). It is formed by decarboxylation of L-glutamate, a reaction catalyzed by an enzyme that depends on the peridoxal phosphate of decarboxylated L-glutamate (Murray et al., 2003). Lactic acid bacteria (LAB) have been reported as a potential producer of GABA (Seo et al., 2013; Wu & Shah, 2017). LAB are generally regarded as safe and non-pathogenic microbes, and has been referred to as ‘generally recognized as safe’. Therefore, GABA-producing LAB can be used directly in functional foods (Zhao et al., 2017). Some LAB can be found in the dairy industry for the production of cheese, yogurt, and other fermented milk products (Yamada et al., 2018). Dadih (fermented buffalo milk) is an Indonesian traditional food originating from West Sumatra Province; it is an extremely popular dairy product in Bukittinggi, Padangpanjang, Solok, Lima Puluh Kota, and Tanah Datar, Indonesia (Surono, 2015). Dadih is made from buffalo milk which is fermented in bamboo for 24–48 hours. The fermentation process is carried out by LAB which are naturally present in buffalo milk (Rizqiati et al., 2015) and the environment (Wirawati et al., 2017). Studies have found that, the LAB strains present in dadih are generally Lactobacillus, Streptococcus, Leuconostoc and Lactococcus (Collado et al., 2007; Surono, 2003). Extraction of DNA is a basic principle in molecular analysis and it is one of the success factors in DNA amplification that is used in the analysis of genetic characters (Mustafa et al., 2016). Polymerase chain reaction (PCR) and phylogenetic analysis based on 16S rRNA gene sequences have been used for successful identification of isolates from various fermented food products (Malik et al., 2015). These molecular approaches have allowed Lactobacillus species to be reliably identified (Henry et al., 2015). This research was conducted to identify and isolated from indigenous dadih of West Sumatera based on 16 S rRNA gene sequence analysis. to characterize GABA producing LAB Methods Sample This study used lactic acid bacteria (LAB) DS15, a GABA- producing LAB isolated from dadih of West Sumatera origin. This bacterium was isolated previously according to the method described by Ali et al. (2009). The experiment was carried out at the Feed Technology Industry Laboratory, Faculty of Animal F1000Research 2019, 7:1663 Last updated: 12 MAR 2020 Science, Andalas University, West Sumatra, Indonesia. LAB DS15 was grown anaerobically in MRS medium (Merck, Darmstadt, Germany) at 30°C and stored for further analysis. Isolation of bacterial genomic DNA Isolation of the total genome of LAB DS15 was done using Genomic DNA Mini Kit purchased from Invitrogen (Pure- LinkTM, USA) by following the manufacturer’s instructions. We used Lysozyme (PureLinkTM, USA) at a concentration of 20 mg/ml to break down the bacterial cell wall to improve protein or nucleic acid extraction efficiency. 16S rRNA gene amplification Genomic DNA of LAB DS15 was used for amplification of 16S rRNA gene. Amplification was done using forward primer 63F (5’-CAG GCC TAA CAC ATG CAA GTC-3’) and reverse primer 1387R (5’-GGG CGG GGT GTA CAA GGC-3’). of Laboratory of Medical Molecular Biology and Diagnos- tic, Indonesian Institute of Sciences. The reaction was car- ried out in a volume of 50 μl. The PCR mixture contained 22 μl of MQ, 25 μl DreamTaq Green DNA Polymerase (Thermo Fisher Scientific, USA), 1 μl of each forward and reverse primer (10 μM each, IDT synthesized) and 1 μl template. Amplification conditions were 5 minutes of preheat- ing at 95°C, 30 seconds denaturation at 95°C, 30 seconds of primer annealing at 58°C, 1 minute extension step at 72°C and post cycling extension of 5 minutes at 72°C for 35 cycles. The reactions were carried out in a thermal cycler (Biometra’s T-Personal Thermal Cycler, USA). Electrophoresis PCR products were stored at 4°C for further examination using 1% (w/v) agarose electrophoresis in 1x TAE, 100 V for 30 minutes. The DNA bands formed from electrophoresis process was visualized using UV transluminator. The marker used was 1 Kb Plus DNA ladder (ThermoFisher Scientific). Sequence alignment of the 16S rRNA gene Sequencing of the 16S rRNA gene was performed at the Laboratory of Medical Molecular Biology and Diagnostic, Indonesian Institute of Sciences, Jakarta. Sequencing results were edited (contig and peak chromatogram verification) using the SeqManTM II program. Analysis of 16S rRNA sequences of LAB DS15 was performed using NCBI BLAST. Multiple alignment was done using the ClustalX 2.1 program. BioEdit version 7.2.5 in edit mode to see the absence of an inverted sequence and align the sequence length. Kinship visualization was done using the combined phylogenetic tree of the MEGA 7.0.20 program with the Neighbor-Joining hood method (Saitou & Nei, 1987). Results and discussion The identification of LAB DS15 to determine the strain was done based on 16S rRNA gene. The first step was amplification using PCR method, with the electrophoresis image shown in Supplementary File 1. The amplification process was carried out to obtain more copies of the 16S rRNA gene for the sequencing process. Analysis of sequencing results begun by aligning the base sequence of the 63F forward sequence and reverse 138R using the SegMan program. PCR of the 16S rRNA gene of LAB DS15 gave an approximately 1400 bp amplicon (Figure 1). Page 3 of 15 Saitou & Nei (1987) indicated that the evolutionary history of organisms can be known using the neighbour-joining method. Organisms within the same taxa are normally clustered together in the phylogenetic tree and have better bootstrap values (Felsenstein, 1985). In this study, we drew a phylogenetic tree to scale and determined the evolutionary distances using the p-distance method. A total of 26 nucleotide sequences and codon positions 1st + 2nd + and 3rd + noncoding were con- sidered, using MEGA 7.0 as reported by Kumar et al. (2016) for evolutionary analyses. Figure  1.  Agarose  gel  (1%)  electrophoresis  showing  amplified 16S rRNA gene of LAB DS18. M, DNA marker; 1, PCR product of LAB DS18. F1000Research 2019, 7:1663 Last updated: 12 MAR 2020 DNA sequencing results were analyzed using NCBI BLAST. According to Willey et al. (2009), 16S rRNA sequencing looks at the similarity of isolates to those already available in GenBank; this is one molecular detection method that is ideal enough to know the kinship relationship between bacteria because the 16S rRNA sequence is a gene found in all microbes and is indispensable in maintain life. The 16S rRNA gene sequencing identified the LAB DS15 to belong to the genus Pediococcus, forming a well-defined cluster with Pediococcus acidilactici. This cluster was recovered in 100% of bootstrap analysis. Pediococcus spp. are widely described as probiotics (Porto et al., 2017). Abbasiliasi et al. (2012) also found Pediococcus acidilactici in fermented milk products. Pediococcus acidilactici are important LAB which have been used as starter cultures in meat, vegetable and dairy fermentation causing charac- teristic flavor changes, improving hygiene and extending the shelf life of these products (Mora et al., 1997; Porto et al., 2017). A phylogenetic tree (Figure 2) was constructed to determine the kinship relationship of LAB DS15. The phylogenetic tree is known to show a high consistency of the relationships between organisms. In this study, the isolate showed similarity of 99% at 100% query coverage to Pediococcus acidilactici strain DSM 20284. A value of 99% indicates that the isolate can be considered as the same species with Pediococcus acidilac- tici strain DSM 20284. The sequence of homology levels was high, as shown by the red color with a score of ≥200 (Figure 3). Figure 2. Phylogenetic tree of 16S rRNA gene of LAB DS18 using the neighbor-joining method. Page 4 of 15 F1000Research 2019, 7:1663 Last updated: 12 MAR 2020 Figure 3. Graphic summary of DS18 and Pediococcus acidilactici strain DSM 20284. From the results of this homology it can be concluded that the two sequences are the same and have an evolutionary relationship. LAB DS15 was Pediococcus acidilactici, with 99% similarity to Pediococcus acidilactici strain DSM 20284. The next closest species for which a sequence alignment of at least 100% query coverage was observed were Pediococcus pentosaceus strain DSM 20336, Pediococcus acidilactici strain NGRI 0510Q and Pediococcus argentini strain CRL 776 at 98% similarity to the DS15 isolate. Pediococcus stilesi strain FAIR-E 180 showed 98% similarity with 99% query coverage. An alignment query result of 100% indicates a significant alignment, which means the search sequence in this study was identical with the species level. identified genus, even at the Conclusion The PCR of 16S rRNA gene sequence gave an approximately 1400 bp amplicon for LAB DS15, isolated from indigenous dadih of West Sumatera. Phylogenetic analysis showed that Data availability Pediococcus acidilactici strain DS32 16S ribosomal RNA gene, partial sequence, obtained during this study. GenBank accession http://identifiers.org/ncbigi/ GI:1481059229. number MH938236: Grant information This research was supported by Ministry of Research, Technology and Higher Education Republic of Indonesia through Master of Education Towards Doctoral Scholarship Program for Excellence Undergraduate and the support through World Class Professor Program Scheme-B No. 123.57/D2.3/KP/2018. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Supplementary material Supplementary File 1. Electrophoresis image of the PCR amplification product. Click here to access the data. Page 5 of 15 References Abbasiliasi S, Tan JS, Ibrahim TA, et al.: Isolation of Pediococcus acidilactici Kp10 with ability to secrete bacteriocin-like inhibitory substance from milk products for applications in food industry. BMC Microbiol. 2012; 12: 260. PubMed Abstract | Publisher Full Text | Free Full Text Ali FWO, Abdulamir AS, Mohammed AS, et al.: Novel, practical and cheap source for isolating beneficial γ-aminobutyric acid-producing leuconostoc NC5 bacteria. Res J Med Sci. 2009; 3(4): 146–153. Reference Source Collado CM, Surono IS, Meriluoto J, et al.: Potential probiotic characteristics of Lactobacillus and Enterococcus strains isolated from traditional dadih fermented milk against pathogen intestinal colonization. J Food Prot. 2007; 70(3): 700–705. PubMed Abstract | Publisher Full Text Dhakal R, Baipai VK, Baek KH: Production of gaba (γ - Aminobutyric acid) by microorganisms: a review. Braz J Microbiol. 2012; 43(4): 1230–41. PubMed Abstract | Publisher Full Text | Free Full Text Felsenstein J: Confidence Limits On Phylogenies: An Approach Using The Bootstrap. Evolution. 1985; 39(4): 783–791. PubMed Abstract | Publisher Full Text Henry DE, Halami PM, Prapulla SG: Lactobacillus plantarum mcc2034, a novel isolate from traditional Indian lactic fermented preparation: molecular identification and evaluation of its in vitro probiotic potential. J Microbiol Biotechnol Food Sci. 2015; 4(4): 328–331. Publisher Full Text Kumar S, Stecher G, Tamura K: MEGA7: Molecular Evolutionary Genetics Analysis version 7.0 for bigger datasets. Mol Biol Evol. 2016; 33(7): 1870–1874. PubMed Abstract | Publisher Full Text Malik V, Devi U, Yadav RNS, et al.: 16s rRNA based phylogenetic analysis of lactobacillus plantarum isolated from various fermented food products of Assam. J Microbiol Biotechnol Food Sci. 2015; 5(1): 20–22. Publisher Full Text Mora D, Fortina MG, Parini C, et al.: Identification of Pediococcus acidilactici and Pediococcus pentosaceus based on 16s rRNA and ldhD gene-targeted multiplex PCR analysis. FEMS Microbiol Lett. 1997; 151(2): 231–236. PubMed Abstract | Publisher Full Text Murray RK, Granner DK, Rodwell VW, et al.: Harper’s Illustrated Biochemistry 23th edn. McGraw-Hill Companies, Inc. USA. 2003. Mustafa H, Rachmawati I, Udin Y: Genomic DNA concentration and purity measurement of Anopheles barbirostris. Journal of Disease Vector. 2016; 1: 7–10. Publisher Full Text Nonaka S, Arai C, Takayama M, et al.: Efficient increase of γ-aminobutyric acid (GABA) content in tomato fruits by targeted mutagenesis. Sci Rep. 2017; 7(1): 7057. PubMed Abstract | Publisher Full Text | Free Full Text F1000Research 2019, 7:1663 Last updated: 12 MAR 2020 Olsen RW, Li GD: Gaba. In Brady ST, Siegel GJ, Albers LW, Price DL, (Editors). Basic Neurochemistry (Eight edition): Principles of molecular, cellular, and medical neurobiology. 2012; 367–376. Publisher Full Text Porto MC, Kuniyoshi TM, Azevedo PO, et al.: Pediococcus spp.: An important genus of lactic acid bacteria and pediocin producers. Biotechnol Adv. 2017; 35(3): 361–374. PubMed Abstract | Publisher Full Text Rizqiati H, Sumantri C, Noor RR, et al.: Isolation and identification of indigenous lactic acid bacteria from North Sumatra river buffalo milk. Indonesian Journal of Animal and Veterinary Sciences. 2015; 20(2): 87–94. Publisher Full Text Saitou N, Nei M: The neighbor-joining method: a new method for reconstructing phylogenetic trees. Mol Biol Evol. 1987; 4(4): 406–425. PubMed Abstract | Publisher Full Text Seo MJ, Lee JY, Nam YD, et al.: Production of γ-Aminobutyric Acid by Lactobacillus brevis 340G Isolated from Kimchi and Its Application to Skim Milk. Food Eng Prog. 2013; 17(4): 418–423. Publisher Full Text Surono IS: In vitro probiotic properties of indigenous dadih lactic acid bacteria. Asian-Australas J Anim Sci. 2003; 16(5): 726–731. Publisher Full Text Surono IS: Traditional Indonesian dairy foods. Asia Pac J Clin Nutr. 2015; 24 Suppl 1: S26–S30. PubMed Abstract | Publisher Full Text Willey JM, Sherwood LM, Woolverton CJ: Prescott’s Principles of Microbiology. Boston: McGraw-Hill Higher Education. 2009. Reference Source Wirawati CU, Sudarwanto MB, Lukman DW, et al.: Characteristic and development of cow’s milk dadih as an alternate of buffalo’s milk dadih. WARTAZOA Indonesian Bulletin of Animal and Veterinary Sciences. 2017; 27(2): 95–103. Publisher Full Text Wu Q, Shah NP: High γ-aminobutyric acid production from lactic acid bacteria: Emphasis on Lactobacillus brevis as a functional dairy starter. Crit Rev Food Sci Nutr. 2017; 57(17): 3661–3672. PubMed Abstract | Publisher Full Text Yamada Y, Endou M, Morikawa S, et al.: Lactic Acid Bacteria Isolated from Japanese Fermented Fish (Funa-Sushi) Inhibit Mesangial Proliferative Glomerulonephritis by Alcohol Intake with Stress. J Nutr Metab. 2018; 2018: 6491907. PubMed Abstract | Publisher Full Text | Free Full Text Zhao A, Hu X, Wang X: Metabolic engineering of Escherichia coli to produce gamma-aminobutyric acid using xylose. Appl Microbiol Biotechnol. 2017; 101(9): 3587–3603. PubMed Abstract | Publisher Full Text Page 6 of 15 F1000Research 2019, 7:1663 Last updated: 12 MAR 2020 Open Peer Review Current Peer Review Status: Version 3 Reviewer Report 12 March 2020 https://doi.org/10.5256/f1000research.22366.r56703 © 2020 Pato U. This is an open access peer review report distributed under the terms of the Creative Commons Attribution License work is properly cited. , which permits unrestricted use, distribution, and reproduction in any medium, provided the original Usman Pato Faculty of Agriculture, Riau University, Pekanbaru, Indonesia INTRODUCTION 1. In general, the introduction is relatively good but needs to be added by the results of research from Hosono et al 1989  and Wirawati et al., 2019  about micflora in dadih 1 2 2. In the introduction, the author needs to explain in more detail the role of GABA produced by LAB and other organisms  METHODS 1. An explanation should be added as to why only to choose the DS15 strain producing GABA in this study. 2. It is necessary to add the reference methods used in the 16S rRNA gene amplification analysis and electrophoresis process  RESULTS AND DISCUSSION The results are well presented and discussed systematically because the authors used only one strain. REFERENCES The author needs to add references as a follow-up to suggestions for improvement in the introduction and method of this paper The strength of this paper The strength of study is the first research to report on GABA-producing LAB from dadih and local fermented milk products from Indonesia The weakness of this paper One of the weaknesses of this study is that the authors only used one LAB dadih isolate (strain DS15) so that no comparative data were produced and the discussion was relatively limited. References 1. Hosono A, Wardojo R, Otani H: Microbial flora in dadih a traditional fermented milk in indonesia. Life, Earth . Page 7 of 15 F1000Research 2019, 7:1663 Last updated: 12 MAR 2020 . Earth 2. Wirawati CU, Sudarwanto MB, Lukman DW, Wientarsih I, et al.: Diversity of lactic acid bacteria in dadih produced by either back-slopping or spontaneous fermentation from two different regions of West Sumatra, Indonesia. PubMed Abstract Publisher Full Text  |  (6): 823-829 Vet World . 2019; 12 Is the work clearly and accurately presented and does it cite the current literature? Yes Is the study design appropriate and is the work technically sound? No Are sufficient details of methods and analysis provided to allow replication by others? Yes If applicable, is the statistical analysis and its interpretation appropriate? No Are all the source data underlying the results available to ensure full reproducibility? Yes Are the conclusions drawn adequately supported by the results? Yes Competing Interests:  No competing interests were disclosed. Reviewer Expertise: Food Microbiology, Probiotic I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard. Reviewer Report 25 November 2019 https://doi.org/10.5256/f1000research.22366.r55312 © 2019 Wu Q. This is an open access peer review report distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Qinglong Wu Texas Children's Microbiome Center, Baylor College of Medicine, Houston, TX, USA I did not see any improvements of scientific value that have been made in the revision. Is the work clearly and accurately presented and does it cite the current literature? Yes Is the study design appropriate and is the work technically sound? Yes Page 8 of 15 F1000Research 2019, 7:1663 Last updated: 12 MAR 2020 Are sufficient details of methods and analysis provided to allow replication by others? Yes If applicable, is the statistical analysis and its interpretation appropriate? Yes Are all the source data underlying the results available to ensure full reproducibility? Yes Are the conclusions drawn adequately supported by the results? Yes Competing Interests:  No competing interests were disclosed. I confirm that I have read this submission and believe that I have an appropriate level of expertise to state that I do not consider it to be of an acceptable scientific standard, for reasons outlined above. Version 2 Reviewer Report 11 June 2019 https://doi.org/10.5256/f1000research.19627.r48482 © 2019 Mutalib S. This is an open access peer review report distributed under the terms of the Creative Commons Attribution License work is properly cited. , which permits unrestricted use, distribution, and reproduction in any medium, provided the original Sahilah Abd Mutalib Centre for Biotechnology and Functional Food, Faculty of Science and Technology, Universiti Kebangsaan Malaysia (UKM), Selangor, Malaysia 1. Introduction - fairly good and can be improved i. Dadih from Indonesia has- Lactobacillus Streptococcus Leuconostoc identify the bacteria? Biochemical tests or using molecular approaches? ii. Is there any data on dadih from Malaysia as well for comparison. , ,  and Lactococcus - How did they 2. Methods - can be improved 2.1 Sample - Subtopic sample suggested to change - Bacterial strain The month and year of the bacterium should be mentioned for ex: in June 2009. Why too long to continue the partial sequence 16s rRNA analysis? 2. 2 Isolation of bacterial genomic DNA We used lysozyme-change to "Twenty(20) mg/ml of lysozyme was used to break down ...." Please state where did you keep the genomic DNA. Example in -20 C freezer or 4 C refrigerator prior o o analysis Page 9 of 15 analysis F1000Research 2019, 7:1663 Last updated: 12 MAR 2020 2.3 16S rRNA gene amplification Please state the reference  after forward and reverse primers is mentioned. 2.4 Electrophoresis 1% change to 1% (w/v) in 1x change to1x TAE, 100 V The marker should be mentioned in this section, 1 Kb ladder? What kind of dye did you used? Red dye, syber green, ethidium bromide? State their brand as well Results and discussion Good - due to a single strain/isolate was studied, thus, the explanation is straight forward. Is the work clearly and accurately presented and does it cite the current literature? Yes Is the study design appropriate and is the work technically sound? Yes Are sufficient details of methods and analysis provided to allow replication by others? Yes If applicable, is the statistical analysis and its interpretation appropriate? No Are all the source data underlying the results available to ensure full reproducibility? Yes Are the conclusions drawn adequately supported by the results? Yes Competing Interests:  No competing interests were disclosed. Reviewer Expertise: Food microbiology, Halal Science, biomass degradation (EFB and POME) I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard. Author Response 14 Aug 2019 Nurul Huda , Universiti Malaysia Sabah, Malaysia, Malaysia 1. Introduction - fairly good and can be improved i. Dadih from Indonesia has- did they identify the bacteria? Biochemical tests or using molecular approaches? Lactobacillus Streptococcus Leuconostoc and , , Lactococcus - They identify bacteria with a molecular approach using the 16SsRNA technique ii. Is there any data on dadih from Malaysia as well for comparison. How Page 10 of 15 F1000Research 2019, 7:1663 Last updated: 12 MAR 2020 ii. Is there any data on dadih from Malaysia as well for comparison. No, we don’t have data on dadih from Malaysia. 2. Methods - can be improved 2.1 Sample - Subtopic sample suggested to change - Bacterial strain The month and year of the bacterium should be mentioned for ex: in June 2009. Why too long to continue the partial sequence 16s rRNA analysis? Bacterial strains isolated in July 2017. We did this isolation based on the method of Ali et al., (2009), not isolates from the author. 2. 2 Isolation of bacterial genomic DNA We used lysozyme-change to "Twenty (20) mg/ml of lysozyme was used to break down ...." Please state where did you keep the genomic DNA. Example in -20 C freezer or 4 C refrigerator prior Analysis We keep the genomic DNA in 4 C refrigerator. o 2.3 16S rRNA gene amplification Please state the reference after forward and reverse primers is mentioned. We got reference for forward primer 63F (5'-CAG GCC TAA CAC ATG CAA GTC-3') and reverse primer 1387R (5'-GGG CGG GGT GTA CAA GGC-3') from Laboratory of Medical Molecular Biology and Diagnostic, Indonesian Institute of Sciences, Jakarta, Indonesia. 2.4 Electrophoresis 1% change to 1% (w/v) Ok we will change in 1x change to1x TAE, 100 V Ok we will change The marker should be mentioned in this section, 1 Kb ladder? What kind of dye did you used? Red dye, syber green, ethidium bromide? State their brand as well We used 1 Kb Plus DNA Ladder (ThermoFisher Scientific) Results and discussion Good - due to a single strain/isolate was studied, thus, the explanation is straight forward. Thank You. Competing Interests:  No competing interests were disclosed. Reviewer Report 08 April 2019 https://doi.org/10.5256/f1000research.19627.r46293 Page 11 of 15 F1000Research 2019, 7:1663 Last updated: 12 MAR 2020 © 2019 Mahanta J. This is an open access peer review report distributed under the terms of the Creative Commons Attribution License work is properly cited. , which permits unrestricted use, distribution, and reproduction in any medium, provided the original Jagadish Mahanta Regional Medical Research Centre, Indian Council of Medical Research (ICMR), Dibrugarh, Assam, India Authors wanted to identify and characterize GABA producing LAB isolated from “Dadih”. 1. 2. 3. 4. However, authors have taken a strain already isolated and identified in 2009. Authors have not mentioned anything about the gap in the previous research that necessitated undertaking the present exercise. Authors may clarify the issue. Authors have done elaborate molecular testing and phylogenetic analysis of the bacteria taken from the stock. Authors should elaborate the achievement of this exercise. As emphasized by the authors, they should elaborate, about characterization and GABA production potential of the strain Authors should elaborate on the novelty of the study. Is the work clearly and accurately presented and does it cite the current literature? Yes Is the study design appropriate and is the work technically sound? No Are sufficient details of methods and analysis provided to allow replication by others? Yes If applicable, is the statistical analysis and its interpretation appropriate? Not applicable Are all the source data underlying the results available to ensure full reproducibility? Partly Are the conclusions drawn adequately supported by the results? Yes Competing Interests:  No competing interests were disclosed. I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above. Author Response 10 Apr 2019 Nurul Huda , Universiti Malaysia Sabah, Kota Kinabalu, Malaysia Authors wanted to identify and characterize GABA producing LAB isolated from “Dadih”. 1. However, authors have taken a strain already isolated and identified in 2009. Authors have not mentioned anything about the gap in the previous research that necessitated undertaking the Page 12 of 15 F1000Research 2019, 7:1663 Last updated: 12 MAR 2020 1. However, authors have taken a strain already isolated and identified in 2009. Authors have not mentioned anything about the gap in the previous research that necessitated undertaking the present exercise. Authors may clarify the issue. Exploration of isolates from dadih has been carried out, but no studies have used these isolates as GABA producers. In this study we obtained DS15 isolate from the isolation of various fermented foods which had the highest GABA production. We did this isolation based on the method of Ali et al.,  (2019), not isolates from the author. 2. Authors have done elaborate molecular testing and phylogenetic analysis of the bacteria taken from the stock. Authors should elaborate the achievement of this exercise. The result of BLAST at the NCBI GenBank site from the sequences showed that DS15 isolate were Pediococcus acidilactici with P. acidilactici P. acidilactici similarity.  DSM 20284, with the difference of one base pair. The next closest species were . Based on the phylogenetic tree, DS15 has a 99% similarity or homology  FAIR-E 180 shows 98% similarity with 99% query coverage.  CRL 776 with 98%  DSM 20336 and P. pentosaceus P. argentinicus  NGRI 0510Q, P. stilesi 3.  As emphasized by the authors, they should elaborate, about characterization and GABA production potential of the strain. We have carried out quantitative screening on some of the isolates we obtained from dadih, and we found that DS15 isolates produced the highest amount of GABA. Data and discussion are used in another publications. 4. Authors should elaborate on the novelty of the study. The novelty of this study was the use of bacterial isolates from dadih as a GABA producer. Competing Interests:  No competing interests were disclosed. Reviewer Report 01 April 2019 https://doi.org/10.5256/f1000research.19627.r46425 © 2019 Wu Q. This is an open access peer review report distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Qinglong Wu Texas Children's Microbiome Center, Baylor College of Medicine, Houston, TX, USA The authors detailed the 16S rRNA gene-based to be a good lab protocol without demonstrating any scientific value. There is no experimental data to support the GABA production from isolate DS15. They have to present the GABA data in terms of GABA yield under defined fermentation conditions. Meanwhile, they have to demonstrate the pathway in isolate DS15 that is responsible for GABA biosynthesis and GABA export in this strain. Page 13 of 15 F1000Research 2019, 7:1663 Last updated: 12 MAR 2020 Secondly, the authors just use one isolate to achieve the claim "GABA producing LAB isolated from indigenous dadih was . This is not a rigorous way. Pediococcus acidilactici" Here are my questions: 1. 2. 3. What is the level of GABA in dadih? How is GABA production capacity of DS15? There is no massive bacterial isolation from dadih; neither no microbial community profiling for dadih, nor pathway identification of GABA production for microbial community of dadih; so one isolate from dadih does not mean anything. Is the work clearly and accurately presented and does it cite the current literature? Partly Is the study design appropriate and is the work technically sound? No Are sufficient details of methods and analysis provided to allow replication by others? Yes If applicable, is the statistical analysis and its interpretation appropriate? Not applicable Are all the source data underlying the results available to ensure full reproducibility? Partly Are the conclusions drawn adequately supported by the results? No Competing Interests:  No competing interests were disclosed. Reviewer Expertise: Food microbiology, microbiome science, microbial genomics, functional genomics, microbial GABA biosynthesis, biochemistry I confirm that I have read this submission and believe that I have an appropriate level of expertise to state that I do not consider it to be of an acceptable scientific standard, for reasons outlined above. Author Response 10 Apr 2019 Nurul Huda , Universiti Malaysia Sabah, Kota Kinabalu, Malaysia 1. What is the level of GABA in dadih? We don't count the amount of GABA on dadih. We did not count the amount of GABA produced in dadih.  GABA is produced by lactic acid bacteria of dadih origin but not the dadih, so we don’t count or determine GABA level of dadih 1. How is GABA production capacity of DS15? GABA production capacity of DS15 was 49.365 mg/L 1. There is no massive bacterial isolation from dadih; neither no microbial community profiling Page 14 of 15 F1000Research 2019, 7:1663 Last updated: 12 MAR 2020 1. There is no massive bacterial isolation from dadih; neither no microbial community profiling for dadih, nor pathway identification of GABA production for microbial community of dadih; so, one isolate from dadih does not mean anything. In this study, we isolated bacteria from various fermented food products (dadih, ikan budu, asam durian and tape singkong), determined their GABA producing ability and selected the isolate with the highest GABA production for further identification. The results of isolation and characterization are explained in other articles. The distribution of LAB isolates from the indigenous West Sumatera fermented food (dadih only) is; 1. 2. 3.  Origin from Aiadingin area.  Number of isolate 131; Number of LAB isolate 125; Number of GABA producing LAB isolate 23. Origin from Sijunjung area.  Number of isolate 166; Number of LAB isolate 93; Number of GABA producing LAB isolate 19. Origin from Solok area.  Number of isolate 100; Number of LAB isolate 96; Number of GABA producing LAB isolate 19. In total, from 3 areas, number of isolate 397; number of LAB isolate 314; number of GABA producing LAB isolate 62. Competing Interests:  No competing interests were disclosed. The benefits of publishing with F1000Research: Your article is published within days, with no editorial bias You can publish traditional articles, null/negative results, case reports, data notes and more The peer review process is transparent and collaborative Your article is indexed in PubMed after passing peer review Dedicated customer support at every stage For pre-submission enquiries, contact research@f1000.com Page 15 of 15
null
10.1038_s41598-022-24860-9.pdf
Data availability All data presented here can be found online in Supplementary Information 1 (includes Methods S1–S6; Fig- ures S1–S3; Tables S1–S5), and Supplementary Data S1.
Data availability All data presented here can be found online in Supplementary Information 1 (includes Methods S1-S6; Figures S1-S3; Tables S1-S5 ), and Supplementary Data S1.
OPEN Quantification of biological nitrogen fixation by Mo‑independent complementary nitrogenases in environmental samples with low nitrogen fixation activity Shannon J. Haynes 1,3*, Romain Darnajoux 1,3, Eunah Han 1, Sergey Oleynik 1, Ezra Zimble 2 & Xinning Zhang 1,2* Biological nitrogen fixation (BNF) by canonical molybdenum and complementary vanadium and iron‑only nitrogenase isoforms is the primary natural source of newly fixed nitrogen. Understanding controls on global nitrogen cycling requires knowledge of the isoform responsible for environmental BNF. The isotopic acetylene reduction assay (ISARA), which measures carbon stable isotope (13C/12C) fractionation between ethylene and acetylene in acetylene reduction assays, is one of the few methods that can quantify isoform‑specific BNF fluxes. Application of classical ISARA has been challenging because environmental BNF activity is often too low to generate sufficient ethylene for isotopic analyses. Here we describe a high sensitivity method to measure ethylene δ13C by in‑line coupling of ethylene preconcentration to gas chromatography‑combustion‑isotope ratio mass spectrometry (EPCon‑GC‑C‑IRMS). Ethylene requirements in samples with 10% v/v acetylene are reduced from > 500 to ~ 20 ppmv (~ 2 ppmv with prior offline acetylene removal). To increase robustness by reducing calibration error, single nitrogenase‑isoform Azotobacter vinelandii mutants and environmental sample assays rely on a common acetylene source for ethylene production. Application of the Low BNF activity ISARA (LISARA) method to low nitrogen‑fixing activity soils, leaf litter, decayed wood, cryptogams, and termites indicates complementary BNF in most sample types, calling for additional studies of isoform‑specific BNF. Nitrogen (N) fundamentally sets the limits of biological productivity, likely constraining natural ecosystem responses to global environmental change1–3. Biological nitrogen fixation (BNF), the prokaryotic process that converts atmospheric dinitrogen (N2) into ammonia, is the primary biological input of new bioavailable N to global and regional N budgets. It thus plays a key biogeochemical function in diverse ecosystems includ- ing tropical, temperate, and high latitude forests, montane grass and shrublands, as well as benthic and open ocean environments4,5. Nitrogenase, the metalloenzyme responsible for BNF, exists in three primary isoforms, characterized by the transition metal present at the active site: the canonical nitrogenase and the ‘alternative’, or more recently termed ‘complementary’ vanadium (V)-only and iron (Fe)-only nitrogenases6,7. The V- and Fe-only nitrogenases are Mo-independent, containing the more abundant crustal-sourced trace metals V and Fe in place of Mo8. Determining the contribution of the different nitrogenase isoforms to environmental BNF is critical for understanding the mechanistic controls on ecosystem BNF, particularly how the coupled biogeochemical cycles of macronutrients and biologically active trace metals respond to anthropogenic perturbations. Because calcula- tions of BNF rate based on traditional methods (i.e., acetylene reduction assays and 15N/14N natural abundance 1Department of Geosciences, Princeton University, Guyot Hall, Princeton, NJ 08544, USA. 2High Meadow Environmental Institute, Princeton University, Guyot Hall, Princeton, NJ 08544, USA. 3These authors contributed equally: Shannon J. Haynes and Romain Darnajoux. *email: sjhaynes@princeton.edu; xinningz@princeton.edu Scientific Reports | (2022) 12:22011 | https://doi.org/10.1038/s41598-022-24860-9 1 Vol.:(0123456789)www.nature.com/scientificreports methods) are sensitive to the nitrogenase isoform9–11, incorrect attribution of the nitrogenase isoforms active in BNF can alter N budget estimates by as much as 50%12,13, influencing ecosystem N status and management. The metal specificity of environmental BNF fluxes can now be assessed with the application of isoform-specific flux tracking via the Isotopic Acetylene Reduction Assay (ISARA) and ethane yield methods combined with nitrogenase gene sequence analyses6,12–14. These approaches have identified significant contributions of comple- mentary V- and Fe-only nitrogenases to non-rhizobial BNF in diverse samples ranging from temperate Everglade mangrove leaf litter, temperate coastal salt marsh sediments, and boreal cyanolichens12,13,15,16. Most recently, a study of cyanolichen BNF across a 600 km boreal forest nutrient gradient provided the first ecosystem-scale evidence for the role of V-nitrogenase in sustaining BNF inputs under Mo-limited conditions13, validating a long held hypothesis on the “backup” role of complementary nitrogenases originally suggested by laboratory studies17. Additionally, low ratios of acetylene to nitrogen reduction activity (i.e., R ratios), suggestive of complementary BNF, have been observed for temperate soil9, boreal moss11,18, and decaying wood19. Further, complementary and uncharacterized nitrogenase genes have been detected in wood mulch20, termite hindguts21, soil9, moss22, and cyanolichens23,24. These studies along with accumulating examples of Mo-limited BNF in boreal18,25,26, temper- ate, and tropical forest biomes27–33 suggest that Mo-independent, complementary BNF could play a global role. Nevertheless, quantification of Mo-independent BNF rates in environmental samples, which often have low BNF activity, has been challenging as the most reliable method for complementary BNF attribution, ISARA 12, requires much higher ethylene yields than are typically observed (e.g., soil, moss, leaf litter typically generate < 300 ppmv ethylene over 1–2 day acetylene reduction assay incubations). Broader study of complementary BNF and its controls within important ecosystems necessitate methodological improvements of ISARA. The ISARA method, based on the widely used acetylene reduction assay (ARA) proxy for BNF activity, relies on natural abundance carbon stable isotope 13C/12C fractionation of acetylene reduction to ethylene (13εAR = δ13Cacetylene – δ13Cethylene, where δ13C (‰) = ([(13C/12C)sample/(13C/12C)standard) − 1] × 1,000 ) to quantify the activity of the different nitrogenase isozymes12. Headspace samples from ARA incubations are analyzed by manual injection into a gas chromatograph-combustion reactor-isotope ratio mass spectrometer (GC-C-IRMS, Fig. 1a). Ethylene (C2H4) is separated from other constituents in headspace [typically, carbon dioxide (CO2), water (H2O), methane (CH4), and acetylene (C2H2)] by gas chromatography, the combustion reactor then con- verts ethylene into CO2, followed by IRMS measurement of the 13C/12C ratio of the produced CO2, which is equivalent to the 13C/12C of ethylene. A similar process yields the 13C/12C of acetylene. Several technical limita- tions and difficulties are associated with the methods as they are currently implemented. Firstly, there is a trade- off between analytical sensitivity (i.e., the magnitude of signal obtained per unit of ethylene concentration) and good chromatographic separation of ethylene (i.e., yielding sharp, well-defined peaks that do not overlap with other headspace constituents) required for accurate and reproducible analyses. This phenomenon primarily results from the conditions of sample injection into the system (e.g., injection volume, flow rate, dilution “split” ratio in the GC injector). Precise δ13Cethylene measurements accommodate maximum injection volumes of ~ 1 mL and thus require samples yielding high ethylene concentrations in ARAs (> 500 ppmv). Secondly, acetylene meas- urements (δ13Cacetylene) often have large uncertainties due to peak tailing and memory effects, which necessitates frequent GC column conditioning (i.e., a brief increase of temperature to remove water, acetylene, and any other analytes accumulated on the column) and combustion reactor oxidations in which pure O2 is flushed into the reactor at high temperature to regenerate the reactor’s oxidative capacity. Finally, ethylene and acetylene isotope measurements are calibrated to the VPDB international carbon isotope reference scale using methane isotope standards because no ethylene standards with NIST traceable δ13C values exist. Deviations in chemical behavior between the methane standard and target analytes, ethylene and acetylene, during chromatographic separation and combustion can lead to biases during drift correction along and across multiple sample runs comprised of replicate measurements. The classical ISARA method is thus relatively time-consuming and limited to samples with high BNF activity. Here, we describe a highly sensitive ISARA method targeted at low nitrogen-fixing activity samples (Low BNF activity ISARA, LISARA). It includes instrumental and methodological improvements to the classical ISARA method that enable precise quantification of Mo-independent BNF rates in samples in an automated fashion. The novel analytical design relies on interfacing a commercially available GC-C-IRMS system used in traditional ISARA analyses with an in-house fabricated, fully automated on-line gas ethylene pre-concentration system (EPCon) developed from Weigand et al.35. The EPCon removes acetylene, a headspace constituent with the greatest peak interference with ethylene, and concentrates ethylene in samples to levels that enable high preci- sion isotope analyses at the part-per-million level with little analytical interference from non-target molecules. In this updated method, ISARA sample requirements have been reduced from ~ 500 ppmv ethylene down to ~ 2 ppmv. To reduce calibration-based uncertainties, we propose the use of commercially available and micro- bially-derived in-house ethylene standards, thus removing the need for acetylene measurements and enabling better within and across laboratory comparisons. To demonstrate environmental applicability, we use LISARA to survey low activity BNF in wood-feeding termites as well as leaf litter, soil, moss, lichens, and decayed wood samples from suburban forests of the Northeastern US. The results suggest significant complementary BNF activity in diverse samples. Material and methods Direct injection method for ethylene and acetylene δ13C analyses by GC‑C‑IRMS. Following the direct injection approach of classical ISARA 12 with a few modifications, ARA samples with high ethylene yield (> 500 ppmv) in 10% v/v acetylene were manually injected into a Thermo Scientific Trace GC Ultra-Isolink with an Agilent HP-PLOT/Q  capillary GC column (30 m, i.d. = 0.32 mm, f.t. = 20 μm) and a combustion reactor connected to a Thermo Scientific Delta V Plus isotope ratio mass spectrometer (GC-C-IRMS; Fig. 1a). Modifica- Scientific Reports | (2022) 12:22011 | https://doi.org/10.1038/s41598-022-24860-9 2 Vol:.(1234567890)www.nature.com/scientificreports/ Chemical precipitation < 20 ppmv (10% v/v Ac) (c) (b) EPCon > 2 ppmv (No Ac) or > 20 ppmv (10% v/v Ac) ETHYLENE (a) GC-C-IRMS > 500 ppmv (10% v/v Ac) Moisture removal Sample collection n o i f a N l 2 ) 4 O C ( g M V1 Vent He backflush Ethanol chiller Trap 1 Liquid N Trap 2 2 High flow e H l e p m a S Sealed vial with sample Sample injection Low flow GC 2 Oxidation ethylene to CO2 Thermo Delta V IRMS Thermo GC Isolink He + sample Separation of analytes Vent GC 1 He carrier flow V2 He backflush Cryofocus Liquid N2 Trap 3 He + sample Low He flow V3 Open for future testing V4 vent out acetylene Acetylene removal (b) EPCon δ13C measurement (a) Direct injection Figure 1. Analytical methodology for δ13C measurement of ethylene in a background matrix containing 10% v/v acetylene or no acetylene based on (a) classical ISARA methods involving direct injection12, (b) the EPCon system, which adds ethylene preconcentration and acetylene removal steps, and (c) an optional chemical precipitation to remove acetylene34 prior to sample loading on EPCon. EPCon development and schematic is adapted from Weigand et al.35. Abbreviations: Ac – acetylene. tions include the replacement of silver ferrules in the GC oven with Valcon polymide (graphite reinforced poly- mer) ferrules to limit memory effects from acetylene. The combustion reactor was oxidized with pure oxygen for 1 h before each run and brief (15 min) seed oxidations were performed between measurement batches (i.e., required every ~ 6–8 ethylene injections, ~ 4–6 acetylene injections) to regenerate reactor oxidation capacity and minimize drift of δ13C values. See Supplementary Table S1a online for additional instrument settings. Ethylene Pre‑Concentration (EPCon) method. ARA samples with < 500 ppmv ethylene were ana- lyzed using an ethylene pre-concentration system developed based on Weigand et al.35 and fabricated in-house (EPCon, Fig. 1b). The EPCon is a fully automated on-line gas preparation system that uses a series of precisely timed valves, cryogenic traps, and a gas chromatograph (GC) to remove background components (particularly water and acetylene) and concentrate ethylene before it is introduced into the GC-C-IRMS. The EPCon was developed through modification of a similar in-house system designed by Weigand et al.35 to measure nitrogen and oxygen isotopes in seawater and freshwater nitrate35–37 and optimized for measurement of low concentration ethylene δ13C. Differences from its direct predecessor35 include direct connection between valve 4 in the EPCon (“V4” on Fig. 1) to the GC column in the commercial GC-C-IRMS system, by-passing the injection chamber to eliminate associated problems (e.g., decreased sensitivity, peak broadening). Flow rates, pressures, valve and trap timings were adjusted to effectively separate ethylene and acetylene such that acetylene could be removed from the analyte stream, and ethylene could be cryogenically focused into a small volume prior to introduction into the GC-C-IRMS. See Supplementary Methods S1 and Supplementary Table S1b-c online for detailed instrument information and settings. Chemical precipitation of background headspace acetylene. For ISARA samples with less than ~ 20 ppmv ethylene, complete GC separation of acetylene and ethylene within the EPCon system was unachievable under our laboratory working conditions due to extreme mass imbalance in analytes. Prior to EPCon δ13Cethylene analysis of these samples, we performed off-line acetylene removal from sample headspace by chemical precipi- tation of acetylene with silver nitrate (AgNO3) in ammonia, producing a silver carbide salt34 (Chemical precipi- tation, Fig. 1c). Ammoniacal AgNO3 solution (0.5 g AgNO3 in 10 mL water) was added to each sample (0.5 mL AgNO3/10 mL headspace containing 10% v/v acetylene). Once the reaction was complete (~ 10 min), sample headspace was transferred to an autosampler vial for EPCon analysis (Fig. 1b), and the remaining carbide salt solution was neutralized (1 mL of 5 N HCl). Complete acetylene removal was verified by analyzing it on a gas chromatograph with a flame ionization detector (GC-FID). We estimated the influence of chemical precipita- tion of acetylene on δ13Cethylene values using control samples made with 2000 ppmv ethylene (from tank EY-4) with and without the addition of 10% v/v acetylene (n = 3, Table 1). Given the highly reactive nature of the silver carbide salt product of precipitation when dry, acetylene precipitation needs to be handled with great care34 and it was only performed as necessary in this study (e.g., sample ethylene < 20 ppmv). Scientific Reports | (2022) 12:22011 | https://doi.org/10.1038/s41598-022-24860-9 3 Vol.:(0123456789)www.nature.com/scientificreports/ Quality controls and data processing. To ensure continuity between our sample analyses within-runs and in the long term (between runs), we used commercially available ethylene and acetylene gas tanks as in- house tank standards (ethylene EY-4, EY-8, acetylene AY-1, AY-4, Table 1) for drift correction and daily quality assurance checks. Quality control standards to test IRMS and EPCon performance were analyzed before each batch of samples that were run. All δ13Cethylene measurements produced by the EPCon-GC-C-IRMS during long (~ 30 h) runs were corrected for drift in instrumental response over time relative to the drift correction standard (EY-4) that was measured at uniform intervals throughout sample runs using linear interpolation between drift correction standards. A second standard (EY-8 or a separate batch of EY-4 standards) was used to indepen- dently validate the drift correction process. Data from direct injections were processed according to the classical method described by Zhang et al., 201612, and did not require drift correction due to the frequent seed oxidations of the reactor. See Supplementary Table S2 online for sample loading details with placement of quality control check standards and Supplementary Data S1 online for data processing calculations. Analytical method validation. For each measurement method (i.e., direct injection, EPCon, and chemi- cal precipitation + EPCon), we determined the sensitivity, limit of quantification, linearity range, intraday repeat- ability, and within laboratory reproducibility (as defined in Carter and Barwick, 201138) by repeated analysis of the main in-house ethylene tank standard (EY-4) under various conditions (Table 1). Sensitivity was determined by linear regression of the IRMS response mass 44 signal (area in volt seconds [Vs]) relative to the amount of ethylene carbon (C) loaded (in nmols C). Linearity range was defined by the lowest and highest quantities of ethylene C that could be directly injected into the GC or loaded into the EPCon autosampler to obtain a mass 44 peak amplitude of 1–6 V (typical conservative analytical range). Samples were loaded with a goal of ~ 2 V for the mass 44 signal. Repeatability (i.e., intraday variability) was estimated as the average of the standard devia- tions for each day over 26 days for the EPCon, and 6 days for direct injection. Within lab reproducibility was calculated using the standard deviation of average δ13C measurements for each day over 26 days for the EPCon and 6 days for direct injection. The limit of quantification (LOQ) was determined based on the minimum ethylene concentration (in ppmv) that could be measured using each method. The technical LOQ, based on ethylene standards and samples with no acetylene, is bounded by the minimum accepted peak amplitude (1 V for mass 44) and the maximum loading volumes for each method (direct injection, 1 mL as constrained by injector and GC column loading; EPCon and chemical precipitation methods, 20 mL as constrained by autosampler vial volume). The methodological LOQ for samples with a 10% matrix, set by the maximum loading volume that avoids overloading the system with acetylene, is 0.5 mL for acetylene and 1.5 mL for EPCon. The methodological LOQ when chemical precipitation was used is ~ 2 ppmv, the lowest sample concentration before the background ethylene concentration carried over in acetylene generated from calcium carbide is greater than ethylene from sample acetylene reduction. Bacterial cultures. Azotobacter vinelandii mutants utilizing only Mo-nitrogenase (‘MoNase’ mutant, strain CA70.139) or only V-nitrogenase (‘VNase’ mutant, strain CA11.7040) for nitrogen fixation were grown aerobi- cally at 30 °C in a modified Burks medium12,41 with 100 nM to 1 µM NaMoO4 (strain CA70.1) or NaVO3 (strain CA11.70). CA70.1 is a double gene deletion mutant (ΔvnfDGK::spc, ΔanfHD70::kan) that expresses only the nif genes (Mo-nitrogenase). CA11.70 is also a double gene deletion mutant (ΔnifHDK, ΔanfHD70::kan) that expresses only the vnf genes (V-nitrogenase). Exponential phase cells (OD620nm ~ 0.3–0.8) were sampled to initi- ate acetylene reduction assays. See Supplementary Methods S2 online for details. Environmental samples. Natural surface samples (moss, cyanolichens, leaf litter, topsoil, decaying wood) and wood-feeding termites with low BNF activity were assessed for complementary nitrogenase activity. Sam- ples were collected from forested sites in central New Jersey (Institute of Advance Studies, Stony Ford Reserve, Pine Barrens, Watershed Institute) and New Hampshire (Mount Moosilauke) from 2019 to 2021. At each site, triplicates of each sample type were collected from one or more stations (10  m × 10  m per station separated by 500–1000  m). Samples, stored at room temperature, were assessed by ARAs within 5  days of collection. Wood-feeding termites (genus Zootermopsis) were obtained from Ward Scientific (https:// www. wards ci. com) and maintained within controlled laboratory habitats for 2–16 days prior to ARA. See Supplementary Methods S3 and Supplementary Table S3 online for details. Acetylene reduction assays. Acetylene reduction assays42 (ARAs) were performed on Azotobacter cul- tures and environmental samples using 10% v/v acetylene generated from calcium carbide. Headspace ethyl- ene concentration was monitored by GC-FID. See Supplementary Methods S2, S3 and Supplementary Table S3 online for ARA details. Azotobacter ARAs were conducted at 30 °C, 200–250 rpm shaking in 25–240 mL serum bottles sealed with 20 mm blue butyl stoppers (Bellco), containing 10% by volume of cell culture and a starting headspace composi- tion of 90% v/v air and 10% v/v acetylene. Headspace gas was transferred to evacuated serum vials (10 mL) with 20 mm blue butyl stoppers (Bellco) to be saved for later IRMS analysis once headspace ethylene concentrations reached 100–2200 ppmv (MoNase strain, typically within 4 h of incubation) and 50–200 ppmv (VNase strain, within 6 h of incubation), yielding in-house ethylene scaling standards EY-Mo-1 and EY-V-1 (Table 1, Fig. 2). Field sample ARAs were conducted in 100–500 mL glass canning jars (Mason, Ball) with metal lids fitted with 20 mm blue butyl stoppers (leaf litter, soil, and wood, Supplementary Table S3); in 30 mL glass vials with screw caps fitted with PTFE/silicone septa (moss, lichens, soil, Supplementary Table S3); or in 15 mL serum vials sealed with 20 mm butyl stoppers (termites, Supplementary Table S3). Control incubations (no acetylene added) were performed with leaf litter, soil, decaying wood (Mt. Moosilauke, Pine Barrens), moss and lichens (Mt. Scientific Reports | (2022) 12:22011 | https://doi.org/10.1038/s41598-022-24860-9 4 Vol:.(1234567890)www.nature.com/scientificreports/ Moosilauke), and termite samples to assess natural endogenous ethylene production independent of acetylene reduction. ARA incubation times for environmental samples varied from ~ 2 to 300 h (Supplementary Table S3) depending on the rate of ethylene production, with a goal of obtaining at least 20 ppmv ethylene. Sample weights for ARA incubations were variable due to sample availability and estimated ethylene production rate, and are listed for each location and sample type in Supplementary Table S3. ARA headspace was subsampled (≤ 3 mL) to measure ethylene concentration by GC-FID, and the remaining headspace was transferred to evacuated sealed vials (10 mL) for later isotopic analysis. Background ethylene correction. Due to low BNF activity, δ13Cethylene was corrected for isotopic influ- ence of background ethylene (~ 2 ppmv) carried over into ARAs by 10% v/v source acetylene (See Supplemen- tary Methods S4, Eqn. S1 online). Background correction was required for ARA samples producing < 20 ppmv ethylene; no quantitative information on nitrogenase could be derived from samples producing < 5 ppmv ethyl- ene due to the isotopic influence of background ethylene. For ARAs yielding ethylene > 5000 ppmv (i.e., 5% of acetylene concentration), δ13Cethylene was also corrected for Rayleigh fractionation12,43. Direct δ13Cethylene and 13εAR scaling methods to quantify complementary nitrogenase contribu‑ tion. One of three methods (Fig. 2) was used to quantify the contribution of complementary nitrogenase to acetylene reduction (as %VNase or %FeNase) in ARAs using δ13Cethylene and δ13Cacetylene. The scaling method used was dependent on whether precise measurements of source δ13Cacetylene values were achievable, given sample availability and technical difficulties in chromatography and combustion. EPCon-GC-IRMS was used to meas- ure δ13Cethylene. All δ13Cacetylene measurements were made using the direct injection approach. See Supplementary Methods S5, Supplementary Table S4, and Supplementary Data S1 online for expanded calculation details. Method 1- The direct scaling approach (Fig. 2), which circumvents the need to measure δ13Cacetylene, was used to calculate complementary nitrogenase contribution when the same source of acetylene stock was used in envi- ronmental sample ARAs as a set of calibration ARAs performed with MoNase and VNase strains of Azotobacter vinelandii. Measured δ13Cethylene in environmental sample ARAs is converted to %VNase using endmember δ13Cethylene values (e.g., ethylene scaling standards, Table 1, Fig. 2) diagnostic of 0% and 100% VNase activity generated, respectively, by MoNase and VNase Azotobacter calibration ARAs (See Supplementary Methods, S4, Eqn. S3 online). See Supplementary Methods S2 online for details on setup and analyses of Azotobacter ARAs. When source acetylene stock used in sample ARAs was not processed in Azotobacter calibration ARAs, we quantified complementary nitrogenase contribution using classical ISARA approaches12 (Fig. 2, methods 2 and 3), which require knowledge of both sample δ13Cacetylene and δ13Cethylene to account for isotopic variation in dif- ferent acetylene stocks in calculations of 13εAR (= δ13Cacetylene – δ13Cethylene). Method 2- The δ13Cacetylene of different acetylene stocks used in sample and Azotobacter ARAs, measured with the direct injection method, was used with sample δ13Cethylene to calculate 13εAR, followed by calibration to the %VNase scale using 13εV and 13εMo of Azotobacter and other diazotrophs, Rhodopseudomonas palustris and Anabaena variabilis (Fig. 2, Supplementary Methods S5, Supplementary Table S4, Supplementary Data S1; calculation modified from Zhang et al., 201612). Method 3- When precise measurement of δ13Cacetylene by direct injection for the specific stock of acetylene within an ARA was unachievable, we used the mean and standard deviation of δ13Cacetylene for seven different batches of acetylene generated from calcium carbide over the past 4 years (δ13Cacetylene = 14.9 ± 0.9 ‰, n = 8; Supplementary Fig. S1; Eqn. S5) in 13εAR calculations. %VNase was calculated using 13εV and 13εMo values from Azotobacter and other diazotrophs as in method 2 (Fig. 2, Supplementary Methods S5, Supplementary Table S4, Supplementary Data S1). Unstable growth of the A. vinelandii Fe-only nitrogenase strain (RP1.1144, ‘FeNase’ mutant) precluded cal- culations of %FeNase based on Azotobacter. Calculations %FeNase (Fig. 2, Supplementary Method S5, Table S4, Data S1) used EPCon derived 13εFe = 5.2 ± 0.7‰ (s.d.) from Rhodopseudomonas palustris using only FeNase12 in ARAs . Because 13εFe < 13εV < 13εMo 12, significant FeNase activity can lead to %VNase values > 100% (i.e., 100% FeNase is equivalent to ~ 140% VNase; Supplementary Table S4). Estimated uncertainty on the %FeNase scale is at most ~ 20%. Complementary nitrogenase contributions to N2 fixation and isoform adjusted total N2 fixation rates can be calculated using %VNase or %FeNase contribution to AR (see above) and R ratios specifying the rate of AR to N2 fixation for each nitrogenase (e.g., RMoNase = 4 , RVNase = 2, RFeNase = 0.5)12. Results Increase in sensitivity and linearity range with the EPCon‑GC‑C‑IRMS system. Measurement sensitivity of δ13Cethylene by GC-C-IRMS is ~ 40-times higher with the addition of the EPCon peripheral than by direct injection (4.3 vs. 0.1 Vs  nmolC−1, Table  1). By removing acetylene and condensing ethylene prior to on-column introduction into the GC-C-IRMS, the EPCon-GC-C-IRMS system produces reliable δ13Cethylene measurements with as little as 1.1 nmol C ethylene, whereas the direct injection method requires > 23.6 nmol C. The larger volume allowance of the EPCon autosampler (20 mL) relative to the GC-C-IRMS sample inlet port (≤ 1  mL) and increased sensitivity enables measurement of gases with ≥ 0.7 ppmv ethylene in the absence of background acetylene. The minimum ethylene concentration for samples with a background of 10% v/v acety- lene (typical ARA samples) is 9 ppmv, or 2 ppmv if background acetylene is removed by chemical precipitation prior to EPCon analyses. Conservatively, minimum working ethylene concentrations for ARA samples are 500 ppmv (direct injection), 20 ppmv (EPCon-GC-C-IRMS), and 5 ppmv (chemical precipitation + EPCon-GC-C- IRMS). The lower sensitivity of the direct injection GC-C-IRMS method is partly due to the necessary use of a high split-ratio within the sample injector port (40:1 – the proportion of sample and He carrier gas flow that is Scientific Reports | (2022) 12:22011 | https://doi.org/10.1038/s41598-022-24860-9 5 Vol.:(0123456789)www.nature.com/scientificreports/ Method 1 Methods 2, 3 10% acetylene VNase mutant ethylene sample ethylene MoNase mutant ethylene 10% acetylene sample ethylene (cid:31)13C 100 sample 0 %VNase VNase est. sample MoNase est. (cid:31)13C 13(cid:30) AR %VNase 1. %VNase = ( (cid:31)13CMo - (cid:31)13Csample (cid:31)13CMo - (cid:31)13CV )x 100 2. 3. %VNase = ( %VNase = ( 100 sample 0 13(cid:30) Mo est - ((cid:31)13Csource acetylene - (cid:31)13Csample ) 13(cid:30) Mo est - 13(cid:30) V est ) x 100 13(cid:30) Mo est - ((cid:31)13C acetylene est - (cid:31)13Csample ) 13(cid:30) Mo est - 13(cid:30) V est ) x 100 Figure 2. Overview of direct δ13Cethylene and 13εAR (= δ13Csource acetylene – δ13Cethylene) scaling methods for converting sample δ13Cethylene values to percent acetylene reduction by V-nitrogenase (%VNase). In the direct scaling method 1, the same batch of source acetylene is used in ARA incubations of Azotobacter mutants expressing only Mo or VNase as for the environmental samples, precluding the need to measure δ13Csource acetylene and enabling %VNase to be calculated based solely on δ13Cethylene. Following 13εAR scaling methods12, different batches of acetylene can be used for sample and single nitrogenase calibration (e.g., mutant) ARAs; measured 13Csource acetylene along with measured δ13Cethylene for each batch of (method 2) or estimated (method 3) values of ARAs are used to calculate 13εAR values, which are then converted to %VNase by comparison with 13εMo and 13εv. ẟ See Method section above and Supplementary Methods S5 online for equation details. vented from the injection port relative to the proportion that enters the GC column) to fully resolve ethylene (~ a few to several hundred ppmv) and acetylene (~ 100,000 ppmv) peaks with our capillary GC column. Comparison of precision and accuracy between direct injection and EPCon‑GC‑C‑IRMS meth‑ ods. We used tank ethylene with constant δ13C compositions as internal standards over the course of ~ 30-h runs (~ 75 samples and ~ 45 quality controls, typical run setup in Supplementary Table S2 online) for intra-day drift corrections (2–4‰-range; Supplementary Fig.  S3) caused by reactor aging over time (without frequent seed-oxidations), ensuring the comparability of results over multiple days (long term s.d. = 0.2‰, Table  1). Repeatability and within-lab reproducibility of δ13Cethylene from tank EY-4 are similar for both direct injection and EPCon-GC-C-IRMS methods (repeatability 0.11‰ and 0.20‰; reproducibility 0.27‰ and 0.17‰, respec- tively). In addition to high reproducibility of δ13Cethylene from the EPCon system, δ13Cethylene values obtained by EPCon and direct injection methods for all ethylene standards were in good agreement (Table 1). We conclude that the EPCon system does not introduce substantial bias into the accuracy of the results, and EPCon data are directly comparable with published results obtained using direct injection methods12,13,15,16. Several sources of uncertainty and bias for ethylene and acetylene δ13C measurements were identified using tank standards. At times, the automatic integration proposed by the software under-estimated the expected δ13C value of the standard, likely due to substantial tailing of 13C relative to 12C linked to the combustion reactor (Supplementary Fig. S3). This phenomenon was more pronounced with δ13Cacetylene analyses, possibly due to stronger interactions between acetylene and combustion reactor metals (CuO, NiO, Pt) as well as the GC column itself. Excess acetylene (i.e., peak amplitude of mass 44 > 5 V) apparent during δ13Cethylene analyses exacerbated peak tailing problems, causing decreased δ13Cethylene precision (by as much as 5‰). GC column and combustion reactor reconditioning was required when excess acetylene was inadvertently introduced into the GC-C-IRMS (e.g., incomplete venting within EPCon). Robustness of the EPCon system to background components in headspace matrices. We tested the interferences of different gases commonly present in environmental samples and of background gases generated during ARA. The EPCon system successfully removes most background gases (Air/N2, CH4, CO2, acetylene) and minimizes their peak interferences with ethylene (Table 1). Only ethane, produced at < 3% of the rate of ethylene by nitrogenase in ARAs12,14, is retained by the EPCon, but its isotopic interference with ethyl- ene is minimal as ethane and ethylene peaks are well-separated in the GC-C-IRMS. Scientific Reports | (2022) 12:22011 | https://doi.org/10.1038/s41598-022-24860-9 6 Vol:.(1234567890)www.nature.com/scientificreports/ Analytical parameters Direct injection EPCon Chem Precip + EPCon Methods Performance Sensitivity (Vs.nmol C ethylene−1) Technical limit of quantificationa (ppmv ethylene) Methodological limit of quantification for samples in 10% v/v acetylene matrixb (ppmv ethylene) Linearity rangec (nmol C ethylene) 0.1 320 600 4.3 0.7 9 23.9 to 143.5 1.1 to 6.7 s.d.‰ (n days) 0.11 (6) 0.27 (6) 0.20 (26) 0.17 (26) N/A 0.7 2 N/A 0.13 ND Composition, vendor (part no.) δ13Clab CO2 δ13Clab CO2 δ13Clab CO2 Average ± s.d. ‰ (n days) Precision Repeatability ethylene d Within lab reproducibility ethylene d Accuracy Standard ID, usage Ethylene: EY-4, drift correction, daily QC 100% ethylene, Airgas (EY R35) 10.0 ± 0.3 (6) 10.1 ± 0.3 (26) 10.4 ± 0.3 (2) EY-8, drift validation, daily QC EY-Mo-1e,f, relative scaling EY-V-1e, relative scaling Acetylene: AY-1, daily QC AY-4, daily QC 1000 ppmv ethylene in He, Airgas (custom mix) ethylene from Azotobacter vinelandii MoNase mutant ethylene from Azotobacter vinelandii VNase mutant 10.6 ± 0.2 (2) 10.6 ± 0.2 (6) 0.6 ± 0.2 (2) 0.2 ± 0.4 (9) 7.0 ± 0.2 (2) 6.9 ± 0.3 (7) 100% acetylene, Airgas (spe- cialty gas) 1,000 ppm acetylene in He, Airgas (custom mix) 14.2 ± 0.9 (11) 15.9 ± 0.8 (3) N/A N/A N/A N/A N/A N/A N/A Chromatographic interference of background components Acetylene (C2H2) Air (N2) Carbon dioxide (CO2) Ethane (C2H6) Methane (CH4) Nitrous oxide (N2O) Water (H2O) If Vinj > 0.5 mL If Vinj > 1.5 mL N/A Interference w/ methane Yes Vented at V1 Vented at V4 Vented at V1 Vented at V4 No interference No interference No interference No interference w/ ethylene Vented at cryotrap Vented at cryotrap Reduced to N2 in combustion reactor Reduced to N2 in combustion reactor Reduced to N2 in combustion reactor Accum. leads to instability Trapped and/or flushed Trapped and/or flushed Table 1. Analytical performance of Direct Injection, EPCon, and Chemical Precipitation + EPCon methods for δ13C measurements of ethylene. All parameters reported here were obtained under conditions typical of a controlled laboratory environment (e.g. relative humidity between ~ 20 and 60%, temperature at 21 ± 2 °C). Accuracy statistics are reported only for days when a particular standard was measured at least three times. See Supplementary DATA 1, tab “Supporting data for Table 1” for the full data set. a constrained by the minimum accepted peak amplitude (1 V for mass 44) and the maximum loading volumes for each method, 1 mL for direct injection, and 20 mL for EPCon and Chemical precipitation. b set by the maximum loading volume without overloading the system with acetylene (0.5 mL for direct injection and 1.5 mL for EPCon), and for Chemical precipitation, by the overprinting of sample δ13C with background ethylene carried over in acetylene generated from calcium carbide (~ 2 ppm). c conservative range of acceptable mass 44 peak amplitudes is 1–6 V. d corrected for instrumental drift. e corrected for background ethylene in acetylene generated from calcium carbide. f corrected for Rayleigh fractionation. LISARA analyses of ARA incubations from environmental samples. The measurement of δ13Cethylene by EPCon-GC-C-IRMS and δ13Cacetylene by direct injection GC-C-IRMS for sample ARAs forms the basis of the Low BNF activity Isotopic Acetylene Reduction Assay method (LISARA). We applied LISARA to diverse envi- ronmental samples (soil, leaf litter, decayed wood, moss, and cyanolichens from sites in the Northeastern US, and laboratory raised wood-feeding termites) with a wide range of ethylene yields in ARAs (5–1000  ppmv). One (or more) of three calculation methods (Fig. 2) was used to obtain %VNase (or %FeNase) contributions to acetylene reduction (AR; methods 1, 2, 3, Figs. 2 and 3). The classical ISARA method12 uses 13εAR, the carbon stable isotope fractionation due to acetylene reduction in ARAs (i.e., δ13Cacetylene– δ13Cethylene), and diagnostic 13εAR values for AR by each nitrogenase isoform (13εMo, 13εV, and 13εFe, Supplementary Table S4) to determine %VNase or %FeNase (Figs. 2 and 3). To circumvent acetylene δ13C measurement, which has typical uncertainties 3–4 times higher than of ethylene (Table 1) and is often retained in the system to necessitate frequent GC-C reconditioning, we developed a direct scaling approach to Scientific Reports | (2022) 12:22011 | https://doi.org/10.1038/s41598-022-24860-9 7 Vol.:(0123456789)www.nature.com/scientificreports/ 200 150 100 50 0 200 150 100 50 0 ) % ( e l a c s e s a N V % 200 150 100 50 0 Av MoNase Av VNase Rp FeNase Method #1 Method #2 Method #3 Lichens Leaf litter Moss Sample Type Termites Soil Decaying wood Sample Identifiers Laboratory grown 100 Mt. Moosilauke, NH IAS Woods, NJ Stony Ford, NJ Pine Barrens, NJ Watershed Institute, NJ 2x total uncertainty 50 0 ) % ( e l a c s e s a N e F % 100 50 0 100 50 0 Figure 3. Complementary nitrogenase contribution to acetylene reduction (as %VNase or %FeNase) within ARAs on environmental samples with low BNF activity and single nitrogenase utilizing diazotroph cultures. Sample summary statistics (Avg ± s.d. %VNase, Range %VNase, no. samples): Leaf litter (32.4% ± 45.4%, − 19.9 to 195.4%, 30), Lichens (− 0.8% ± 4.7%, − 8.9 to 5.6%, 6), Moss (65.3% ± 37.9%, − 14.5 to 123.0%, 31), Soil (123.9% ± 37.2%, 25.4 to 177.8%, 21), Termites (130.1% ± 22.0%, 104.8 to 156.6%, 7), and Decaying wood (125.9% ± 32.6%, 40.6 to 167.6%, 43). Abbreviations are as follows: Av MoNase – Azotobacter vinelandii MoNase strain, Av VNase – A. vinelandii VNase strain, and Rp FeNase – Rhodopseudomonas palustris FeNase strain. See Supplementary Methods S5 and S6, and Supplementary Data S1 online for details of scaling and uncertainty calculations. calculate complementary nitrogenase contribution based solely on δ13Cethylene (Fig. 2, method 1). This is achieved by comparing δ13Cethylene generated from a common source of acetylene stock used within environmental sample ARAs (δ13Csample) and sets of isotopic calibration ARAs performed with MoNase and VNase strains of Azoto- bacter vinelandii (δ13CMo and δ13CV, Fig. 2, Supplementary Methods S5). We could not determine δ13CFe values for %FeNase calculations with A. vinelandii by direct scaling approaches as the growth of the FeNase strain RP1.1144 was unstable. Ideally, all samples isotope values would be scaled to % complementary nitrogenase using method 1, the direct scaling approach because it is associated with the least amount of uncertainty. Methods 2 and 3, applied to samples that were analyzed before direct scaling standards and associated protocols were developed, can also be used in cases where direct scaling procedures could not be completed (e.g., insufficient acetylene, failed Azotobacter ARA experiments). While method 3 is associated with the highest uncertainty, it provides the fastest means to estimate the contribution of complementary nitrogenases to BNF. With the exception of cyanolichens, all sample types exhibited an isotopic signal consistent with comple- mentary nitrogenase activity (Fig. 3, summary statistics in legend, note that 140% VNase is equivalent to 100% Scientific Reports | (2022) 12:22011 | https://doi.org/10.1038/s41598-022-24860-9 8 Vol:.(1234567890)www.nature.com/scientificreports/ FeNase, see Methods above). Potential contributions of complementary nitrogenase to AR in leaf litter and moss samples ranged from 0 to 100% VNase, with the exception of one leaf litter sample with 195% VNase. Potential contributions in decaying wood and termites ranged from 40 to 160% VNase. Soil data are also highly variable, ranging from 30 to 180% VNase. A few of the 114 samples analyzed are outliers with greater than ~ 200% VNase (1 moss, 2 soil samples, data not shown in Fig. 3), which we attribute to isotopic fractionation of gas due to stopper leakage. The estimated uncertainty for %VNase contributions to AR for environmental samples quantified by direct scaling of δ13Cethylene values is lower than uncertainties from 13εAR–based methods: ~ 9% for direct scaling method 1, ~ 15% for 13εAR method 2, ~ 20% for 13εAR method 3 (Fig. 3, Supplementary Method S5; Supplementary Data S1). The increased precision obtained by directly scaling δ13Cethylene to %VNase (method 1) avoids the uncer- tainty associated with δ13Cacetylene measurements. This is evident in Fig. 3, where %VNase values for the single nitrogenase mutants (i.e., values for Av MoNase, and Av VNase in Fig. 3) are more tightly clustered for method 1 (uses direct scaling of sample and mutant δ13Cethylene values) than those for methods 2 and 3 (uses explicit 13εAR values). Most complementary nitrogenase attributions for single nitrogenase culture ARAs cluster within 15% of their expected values (i.e., 0% VNase for A. vinelandii MoNase strain, 100% VNase for A. vinelandii VNase strain, 100% FeNase = 140% VNase for R. palustris FeNase strain) however a few samples show ~ 20–30% errors (e.g., ~ 130% VNase for A. vinelandii VNase strain, ~ − 20% VNase for A. vinelandii MoNase strain). Uncertain- ties for %FeNase quantified using 13εAR and 13εFe from Rhodopseudomonas palustris FeNase are ~ 15–20% (Sup- plementary Method S6, Supplementary Data S1). Thus, samples requiring the highest precision (very low BNF activity) for quantifications of complementary nitrogenase contribution should use the δ13Cethylene-based direct scaling method (method 1). Given the uncertainties, ARA samples with > 160% in the %VNase scale and > 120% in the %FeNase scale must be strongly influenced by processes unrelated to BNF (e.g. natural endogenous ethylene cycling—production or consumption, gas leakage from stoppers). Non-BNF related fractionation may explain the > 160% VNase values observed for certain samples of litter (n = 1), moss (n = 1), soil (n = 4), decayed wood (n = 2). Discussion Analytical improvements of δ13Cethylene measurement in environmental samples with trace levels of ethylene. Increased sensitivity in the EPCon-GC-C-IRMS system compared to GC-C-IRMS (Table 1) results in ~ 120–450 times lower sample ethylene requirements (depending on whether acetylene is also present in the sample, Table 1). Importantly, EPCon analyses of ARA samples results in limited exposure of the GC column and combustion reactor to the acetylene in ARA samples, which causes substantial drift in analytical outputs due to acetylene degradation of GC column and reactor performance. Combined with intra-day drift corrections based on tank gas with a constant δ13C, use of the EPCon system yields more reproducible results, limits the number of reactor oxidations, and extends the lifetime of the reactor and capillary GC column, thus reducing time and long-term cost per IRMS analysis. Further, all of these instrumental and analytical improve- ments ensure comparable results with much lower ethylene concentrations across analytical runs and experi- ments, without compromising reproducibility. As a result, 120 measurements of δ13Cethylene can be achieved in an automated fashion over a 30-h run in the EPCon system compared to 7–10 days of full-time work for one person using direct injection into the GC-C-IRMS. The LISARA method is a key analytical improvement necessary to studies of nitrogen fixation by complemen- tary nitrogenases in the global environment. The EPCon-GC-C-IRMS analytical upgrade allows for the reliable and reproducible isotopic characterization of ethylene in ARA samples at virtually any ethylene concentration. In practice, samples with as low as 20 ppmv ethylene can be routinely measured before the capacity for acety- lene removal by the EPCon is reached (Fig. 1). Very low yield ARA samples (5–20 ppmv ethylene) can also be measured by the EPCon system following the complete removal of acetylene from the headspace using chemical precipitation (see Methods section). However, the presence of background ethylene carried over in acetylene used for ARAs and potential natural endogenous ethylene production (i.e. unassociated with BNF) can affect δ13Cethylene values, complicating interpretations of complementary nitrogenase contribution in very low BNF activity samples assessed using LISARA. The development of a direct scaling approach to calculate complementary nitrogenase contributions based solely on δ13Cethylene from LISARA analyses circumvents several limitations associated with traditional ISARA, such as needing ethylene concentrations > 500 ppmv and the requirement for δ13Cacetylene measurements. Further, the offline precipitation step to completely remove acetylene from ARA sample headspace would enable any microbiology or wet-chemistry lab to outsource ISARA analyses of ethylene from sample and single nitrogenase calibration ARAs run with the same source acetylene to other stable isotope analytical laboratories for δ13Cethylene measurements. We note that the comparability of absolute δ13C values across research groups may vary and be difficult to assess as multiple factors (e.g., type of GC column and oxidation reactor state) can influence absolute δ13C values. This makes the simultaneous analyses of single nitrogenase calibration ARAs and environmental samples particularly important. Aside from studies of complementary nitrogenase, the EPCon system also has applications in other fields, including plant biology. For example, EPCon analysis of the isotopic composition of endogenously produced ethylene (e.g., by soil bacteria and plants), a phytohormone involved in stress response and seed germination45, could help identify its sources, and track its cycling in complex soil environments. LISARA suggests widespread complementary nitrogenase activity in environmental samples from the Northeastern United States. Of the diverse terrestrial samples characterized using LISARA, 5 of the 6 sample types exhibited δ13Cethylene values consistent with some contribution of complementary nitro- Scientific Reports | (2022) 12:22011 | https://doi.org/10.1038/s41598-022-24860-9 9 Vol.:(0123456789)www.nature.com/scientificreports/ genase to BNF activity (Fig. 3). These results add to growing evidence suggestive of widespread complementary nitrogenase activity in terrestrial ecosystems. Contrary to a previous study on the cyanolichen species Peltigera in boreal forests, which revealed high levels of complementary nitrogenase activity13, we found no evidence of VNase activity in samples of the same genus collected in the temperate Northeastern US (Fig. 3, lichens). Complementary nitrogenase activity in this cyanolichen genus has been found to be primarily controlled by the −1)13, which reflects atmospheric quantity of molybdenum in lichen thalli (Mo thalli content < 300 µgMo.gdry_lichen deposition. The higher atmospheric deposition rate of Mo in the temperate Northeastern US46 may provide suf- ficient Mo to these lichen samples to obviate the need for complementary nitrogenase BNF. The consistent complementary nitrogenase activity observed here for Northeastern leaf litter, soil, decaying wood samples, and for wood feeding termites likely reflects different and more complex Mo controls on BNF than in cyanolichens and moss, which are more directly connected to Mo-rich atmospheric deposition. Differ- ences likely exist in the Mo requirements among diazotrophs, possibly reflecting variations in organism-level metal management strategies47 and in the physicochemical properties of environments around diazotroph cells, which can modulate Mo bioavailability. For example, higher levels of certain forms of organic matter with strong Mo binding capacity (catechol moieties) in samples48 could result in higher total Mo requirements for Mo BNF, influencing Mo and complementary nitrogenase relationships across samples49. Collectively, our results highlight the need for more detailed studies on complementary BNF and its controls in many common sample types. Remaining analytical limitations and future methodological improvements. The presence of background ethylene in the source acetylene (~ 2 ppmv ethylene per 10% v/v acetylene from calcium carbide) remains a challenge when quantifying complementary nitrogenase in environmental samples with very low activity (< 20 ppmv ethylene generated in ARAs). While the isotopic signal of background ethylene in source acetylene can be determined easily using the present EPCon methods, there is significant variability (8.4 ± 1.9‰, n = 8 acetylene batches). Isotopic corrections for the background ethylene do not lead to much loss in precision and accuracy within ARAs containing 10–20 ppmv ethylene yield but would result in a large increase in uncer- tainty for samples with ethylene < 10 ppmv. Hence the LISARA method can only provide qualitative information on complementary nitrogenase for samples with 2–5 ppmv ethylene. When probing environmental samples, natural cycling of endogenous ethylene by soil bacteria and plants can also interfere with the quantification of complementary nitrogenase contribution (see Hendrickson 198950 and references therein). Because low-oxygen conditions favor ethylene production and inhibit ethylene oxidation50, long incubations are likely to increase this phenomenon. In this study, we observed significant endogenous pro- duction of ethylene (i.e., > 5% of the ARA produced ethylene concentration for a given sample type and location) in 12 out of 93 “no acetylene added” control samples. All 12 samples that contained endogenous ethylene were incubated for 290 to 300 h (27 samples were incubated for that long). Another batch of 27 samples incubated for 165 to 175 h did not show signs of endogenous ethylene production. Acetylene has been reported to inhibit ethylene oxidation, thus “no acetylene added” control samples might not be sufficient to assess endogenous eth- ylene production in ARAs with very low ethylene yield (< 20 ppmv)50. Thus, we recommend incubating samples for ideally less than 60 h when conducting an ISARA or LISARA surveys. Overall, the low endogenous ethylene production rate from our samples during incubations (Supplementary Table S3), and the similarity among iso- topic signatures obtained for each sample type over four sites, 2 years, and various incubation times (2–300 h) indicates that natural ethylene cycling has minimal influence on our reported results. Our updated analytical procedure and methodology now allows for the investigation of the contribution and environmental controls on complementary nitrogenase in most N2 fixing samples, notwithstanding the remaining limitations in LISARA analysis of very low-yield ethylene samples from ARA. The LISARA method decreases uncertainty and bias associated with acetylene measurements and allows for the broader use of the ISARA methodology with pure cultures and high yield organisms. Finally, our study of common sample types in several temperate ecosystems of the Northeastern US provides further evidence for the ecological importance of complementary nitrogenase to the cycling of nitrogen and trace metals in terrestrial ecosystems. Sample-specific differences in contribution, as suggested by our results, calls for more investigation into the controls on isozyme specific nitrogenase in natural environments. Data availability All data presented here can be found online in Supplementary Information 1 (includes Methods S1–S6; Fig- ures S1–S3; Tables S1–S5), and Supplementary Data S1. Received: 22 August 2022; Accepted: 22 November 2022 References 1. Elser, J. J. et al. Global analysis of nitrogen and phosphorus limitation of primary producers in freshwater, marine and terrestrial ecosystems. Ecol. Lett. 10, 1135–1142. https:// doi. org/ 10. 1111/j. 1461- 0248. 2007. 01113.x (2007). 2. LeBauer, D. S. & Treseder, K. K. Nitrogen limitation of net primary productivity in terrestrial ecosystems is globally distributed. Ecology 89, 371–379 (2008). 3. Zhang, X., Ward, B. B. & Sigman, D. M. Global nitrogen cycle: critical enzymes, organisms, and processes for nitrogen budgets and dynamics. Chem. Rev. 120, 5308–5351. https:// doi. org/ 10. 1021/ acs. chemr ev. 9b006 13 (2020). 4. Cleveland, C. C. et al. Patterns of new versus recycled primary production in the terrestrial biosphere. Proc. Natl. Acad. Sci. 110, 12733–12737. https:// doi. org/ 10. 1073/ pnas. 13027 68110 (2013). 5. Du, E. et al. Global patterns of terrestrial nitrogen and phosphorus limitation. Nat. Geosci. 13, 221–226. https:// doi. org/ 10. 1038/ s41561- 019- 0530-4 (2020). Scientific Reports | (2022) 12:22011 | https://doi.org/10.1038/s41598-022-24860-9 10 Vol:.(1234567890)www.nature.com/scientificreports/ 6. Bellenger, J. P., Darnajoux, R., Zhang, X. & Kraepiel, A. M. L. Biological nitrogen fixation by alternative nitrogenases in terrestrial ecosystems: a review. Biogeochemistry 149, 53–73. https:// doi. org/ 10. 1007/ s10533- 020- 00666-7 (2020). 7. Harwood, C. S. Iron-only and vanadium nitrogenases: Fail-safe enzymes or something more?. Annu. Rev. Microbiol. 74, 247–266 (2020). 8. Wedepohl, H. K. The composition of the continental crust. Geochim. Cosmochim. Acta 59, 1217–1232. https:// doi. org/ 10. 1016/ 0016- 7037(95) 00038-2 (1995). 9. Bellenger, J.-P., Xu, Y., Zhang, X., Morel, F. M. M. & Kraepiel, A. M. L. Possible contribution of alternative nitrogenases to nitrogen fixation by asymbiotic N2-fixing bacterial in soils. Soil Biol. Biochem. 69, 413–420 (2014). 10. Zhang, X., Sigman, D., Morel, F. & Kraepiel, A. Nitrogen isotope fractionation by alternative nitrogenases and past ocean anoxia. Proc. Natl. Acad. Sci. 111, 4782–4787. https:// doi. org/ 10. 1073/ pnas. 14029 76111 (2014). 11. Soper, F. M., Simon, C. & Jauss, V. Measuring nitrogen fixation by the acetylene reduction assay (ARA): Is 3 the magic ratio?. Biogeochemistry 152, 345–351. https:// doi. org/ 10. 1007/ s10533- 021- 00761-3 (2021). 12. Zhang, X. et al. Alternative nitrogenase activity in the environment and nitrogen cycle implications. Biogeochemistry 127, 189–198 (2016). 13. Darnajoux, R. et al. Molybdenum threshold for ecosystem scale alternative vanadium nitrogenase activity in boreal forests. Proc. Natl. Acad. Sci. 116, 24682. https:// doi. org/ 10. 1073/ pnas. 19133 14116 (2019). 14. Dilworth, M. J., Eady, R. R., Robson, R. L. & Miller, R. W. Ethane formation from acetylene as a potential test for vanadium nitro- genase in vivo. Nature 327, 167–168 (1987). 15. McRose, D., Zhang, X., Kraepiel, A. & Morel, F. Diversity and activity of alternative nitrogenases in sequenced genomes and coastal environments. Front. Microbiol. https:// doi. org/ 10. 3389/ fmicb. 2017. 00267 (2017). 16. Darnajoux, R. et al. Biological nitrogen fixation by alternative nitrogenases in boreal cyanolichens: importance of molybdenum availability and implications for current biological nitrogen fixation estimates. New Phytol. 213, 680–689. https:// doi. org/ 10. 1111/ nph. 14166 (2017). 17. Robson, R. L., Eady, R. R. & Richardson, T. H. The alternative nitrogenase of Azotobacter chroococcum is a vanadium enzyme. Nature 322, 388–390 (1986). 18. Rousk, K., Degboe, J., Michelsen, A., Bradley, R. & Bellenger, J.-P. Molybdenum and phosphorus limitation of moss-associated nitrogen fixation in boreal ecosystems. New Phytol. 214, 97–107. https:// doi. org/ 10. 1111/ nph. 14331 (2017). 19. Benoist, A., Houle, D., Bradley, R. & Bellenger, J.-P. Evaluation of biological nitrogen fixation in coarse woody debris from Eastern Canadian boreal forests. Soil Biol. Biochem. 165, 108531–108531. https:// doi. org/ 10. 1016/j. soilb io. 2021. 108531 (2021). 20. Betancourt, D. A., Loveless, T. M., Brown, J. W. & Bishop, P. E. Characterization of diazotrophs containing Mo-independent nitrogenases, isolated from diverse natural environments. Appl. Environ. Microbiol. 74, 3471–3480. https:// doi. org/ 10. 1128/ aem. 02694- 07 (2008). 21. Desai, M. S. & Brune, A. Bacteroidales ectosymbionts of gut flagellates shape the nitrogen-fixing community in dry-wood termites. The ISME J. 6, 1302–1313. https:// doi. org/ 10. 1038/ ismej. 2011. 194 (2012). 22. Nelson, J. M. et al. Complete genomes of symbiotic cyanobacteria clarify the evolution of vanadium-nitrogenase. Genome Biol. Evolut. 11, 1959–1964. https:// doi. org/ 10. 1093/ gbe/ evz137 (2019). 23. Hodkinson, B. P. et al. Lichen-symbiotic cyanobacteria associated with Peltigera have an alternative vanadium-dependent nitrogen fixation system. Eur. J. Phycol. 49, 11–19. https:// doi. org/ 10. 1080/ 09670 262. 2013. 873143 (2014). 24. Villarreal, A. J. C., Renaudin, M., Beaulieu-Laliberté, A. & Bellenger, J.-P. Stigonema associated with boreal Stereocaulon possesses the alternative vanadium nitrogenase. Lichenologist 53, 215–220. https:// doi. org/ 10. 1017/ S0024 28292 10000 62 (2021). 25. Rousk, K. & Rousk, J. The responses of moss-associated nitrogen fixation and belowground microbial community to chronic Mo and P supplements in subarctic dry heaths. Plant Soil 451, 261–276. https:// doi. org/ 10. 1007/ s11104- 020- 04492-6 (2020). 26. Scott, D. L. et al. Anthropogenic deposition of heavy metals and phosphorus may reduce biological N2 fixation in boreal forest mosses. Sci. Total Environ. 630, 203–210. https:// doi. org/ 10. 1016/j. scito tenv. 2018. 02. 192 (2018). 27. Horstmann, J. L., Denison, W. C. & Silvester, W. B. 15N2 fixation and molybdenum enhancement of ARA by Lobaria Spp. New Phytol. 92, 235–241 (1982). 28. Barron, A. R. et al. Molybdenum limitation of asymbiotic nitrogen fixation in tropical forest soils. Nat. Geosci. 2, 42–45 (2009). 29. Wurzburger, N., Bellenger, J. P., Kraepiel, A. M. L. & Hedin, L. O. Molybdenum and Phosphorus Interact to Constrain Asymbiotic Nitrogen Fixation in Tropical Forests. PLoS One 7, e33710. https:// doi. org/ 10. 1371/ journ al. pone. 00337 10 (2012). 30. Jean, M. E., Phalyvong, K., Forest-Drolet, J. & Bellenger, J.-P. Molybdenum and phosphorus limitation of asymbiotic nitrogen fixation in forests of Eastern Canada: influence of vegetative cover and seasonal variability. Soil Biol. Biochem. 67, 140–146. https:// doi. org/ 10. 1016/j. soilb io. 2013. 08. 018 (2013). 31. Reed, S. C., Cleveland, C. C. & Townsend, A. R. Relationships among phosphorus, molybdenum and free-living nitrogen fixation in tropical rain forests: results from observational and experimental analyses. Biogeochemistry 114, 135–147. https:// doi. org/ 10. 1007/ s10533- 013- 9835-3 (2013). 32. Dynarski, K. A. & Houlton, B. Z. Nutrient limitation of terrestrial free-living nitrogen fixation. New Phytol. 217, 1050–1061. https:// doi. org/ 10. 1111/ nph. 14905 (2018). 33. Stanton, D. E., Batterman, S. A., Von Fischer, J. C. & Hedin, L. O. Rapid nitrogen fixation by canopy microbiome in tropical forest determined by both phosphorus and molybdenum. Ecology 100, 1–8. https:// doi. org/ 10. 1002/ ecy. 2795 (2019). 34. David, K. A. V., Apte, S. K., Banerji, A. & Thomas, J. Acetylene reducton assay for nitrogenase activity: gas chromatographic determination of ethylene per sample in less than 1 min. Appl. Environ. Microbiol. 39, 1078–1080. https:// doi. org/ 10. 1128/ aem. 39.5. 1078- 1080. 1980 (1980). 35. Weigand, M. A., Foriel, J., Barnett, B., Oleynik, S. & Sigman, D. M. Updates to instrumentation and protocols for isotopic analysis of nitrate by the denitrifier method. Rapid Commun. Mass Spectrom. 30, 1365–1383. https:// doi. org/ 10. 1002/ rcm. 7570 (2016). 36. Sigman, D. M. et al. A bacterial method for the nitrogen isotopic analysis of nitrate in seawater and freshwater. Anal. Chem. 73, 4145–4153. https:// doi. org/ 10. 1021/ ac010 088e (2001). 37. Casciotti, K. L., Sigman, D. M., Hastings, M. G., Böhlke, J. K. & Hilkert, A. Measurement of the oxygen isotopic composition of nitrate in seawater and freshwater using the denitrifier method. Anal. Chem. 74, 4905–4912. https:// doi. org/ 10. 1021/ ac020 113w (2002). 38. Carter, J. F. & Barwick, V. J. Good practice guide for isotope ratio Mass Spectrometry. (2011). 39. Paulsen, D. M., Paerl, H. W. & Bishop, P. E. Evidence that molybdenum-dependent nitrogen fixation is not limited by high sulfate concentrations in marine environments. Limnol. Oceanogr. 36, 1325–1334 (1991). 40. Joerger, R. D., Jacobson, M. R., Premakumar, R., Wolfinger, E. D. & Bishop, P. E. Nucleotide sequence and mutational analysis of the structural genes (anfHDGK) for the second alternative nitrogenase from Azotobacter vinelandii. J. Bacteriol. 171, 1075–1086. https:// doi. org/ 10. 1128/ jb. 171.2. 1075- 1086. 1989 (1989). 41. Bellenger, J. P., Wichard, T., Xu, Y. & Kraepiel, A. M. L. Essential metals for nitrogen fixation in a free-living n2-fixing bacterium: chelation, homeostasis and high use efficiency. Environ. Microbiol. 13, 1395–1411. https:// doi. org/ 10. 1111/j. 1462- 2920. 2011. 02440.x (2011). 42. Hardy, R. W. F., Holsten, R. D., Jackson, E. K. & Burns, R. C. Acetylene-ethylene assay for N2 fixation - laboratory and field evalu- ation. Plant Physiol. 43, 1185–1207. https:// doi. org/ 10. 1104/ Pp. 43.8. 1185 (1968). 43. Hayes, J. M. An introduction to isotopic calculations. 1–10 (2004). Scientific Reports | (2022) 12:22011 | https://doi.org/10.1038/s41598-022-24860-9 11 Vol.:(0123456789)www.nature.com/scientificreports/ 44. Joerger, R. D. et al. Nucleotide sequences and mutational analysis of the structural genes for nitrogenase 2 of Azotobacter vinelandii. J. Bacteriol. 172, 3400–3408. https:// doi. org/ 10. 1128/ jb. 172.6. 3400- 3408. 1990 (1990). 45. Iqbal, N. et al. Ethylene role in plant growth, development and senescence: interaction with other phytohormones. Front. Plant Sci. 8, 475–475. https:// doi. org/ 10. 3389/ fpls. 2017. 00475 (2017). 46. Wong, M. Y., Neill, C., Marino, R., Silvério, D. & Howarth, R. W. Molybdenum, phosphorus, and pH do not constrain nitrogen fixation in a tropical forest in the southeastern Amazon. Ecology 102, e03211. https:// doi. org/ 10. 1002/ ecy. 3211 (2021). 47. Darnajoux, R., Constantin, J., Miadlikowska, J., Lutzoni, F. & Bellenger, J.-P. Is vanadium a biometal for boreal cyanolichens?. New Phytol. 202, 765–771. https:// doi. org/ 10. 1111/ nph. 12777 (2014). 48. Kraepiel, A. M. L., Bellenger, J. P., Wichard, T. & Morel, F. M. M. Multiple roles of siderophores in free-living nitrogen-fixing bacteria. Biometals 22, 573–581. https:// doi. org/ 10. 1007/ s10534- 009- 9222-7 (2009). 49. Jouogo Noumsi, C. et al. Effect of organic matter on nitrogenase metal cofactors homeostasis in Azotobacter vinelandii under diazotrophic conditions. Environ. Microbiol. Rep. 8, 76–84. https:// doi. org/ 10. 1111/ 1758- 2229. 12353 (2016). 50. Hendrickson, O. Q. Implications of natural ethylene cycling processes for forest soil acetylene reduction assays. Can. J. Microbiol. 35, 713–718. https:// doi. org/ 10. 1139/ m89- 116 (1989). Acknowledgements We gratefully acknowledge funding support from the Watershed Institute (https:// thewa tersh ed. org), the Carbon Mitigation Initiative (Princeton High Meadows Environmental Institute to XZ), the Tuttle Fund (Princeton Geo- science Department to XZ), NSF EAR grant 1631814 (to XZ), and a Simons Foundation Life Science Research Postdoctoral Fellowship (to RD). We thank collaborators at the Watershed institute, the Rutgers Pineland Field Station, the Stony Ford Center for Ecological Studies at Princeton University, the Institute of Advanced Studies, and the Dartmouth College-Mt. Moosilauke Advisory Committee for permission and access to field samples; Katja Luxem and Linta Reji for help with field work, Anne Kraepiel, F. Morel, J.P. Bellenger, and all members of the Zhang laboratory for discussions. Author contributions S.J.H., R.D., and X.Z. wrote the manuscript text. S.J.H. and R.D. prepared the figures. S.O. built the EPCon instrument utilized in this study. S.H. led methodological development with contributions by R.D., S.O., and X.Z., S.J.H., R.D., E.H., and E.Z. collected field samples and contributed to data acquisition. S.J.H., R.D, and E.H. performed data analysis. X.Z. provided technical expertise, project guidance, and financial support. All authors reviewed the manuscript. Competing interests The authors declare no competing interests. Additional information Supplementary Information The online version contains supplementary material available at https:// doi. org/ 10. 1038/ s41598- 022- 24860-9. Correspondence and requests for materials should be addressed to S.J.H. or X.Z. Reprints and permissions information is available at www.nature.com/reprints. Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/. © The Author(s) 2022 Scientific Reports | (2022) 12:22011 | https://doi.org/10.1038/s41598-022-24860-9 12 Vol:.(1234567890)www.nature.com/scientificreports/
null
10.1371_journal.pone.0279220.pdf
null
All relevant data are within the manuscript and its Supporting Information files. All files are available from thefigshare database ( 10.6084/m9.figshare. 22578787 10.6084/m9.figshare.22578808 ).
RESEARCH ARTICLE Do environmental, social, and governance scores improve green innovation? Empirical evidence from Chinese-listed companies Chunlian Zhang1,2‡, Danni ChenID 3‡* 1 School of Economics and Trade, Nanchang Institute of Technology, Nanchang, Jiangxi, China, 2 The Water Economy and Water Rights Research Center, Nanchang Institute of Technology, Nanchang, Jiangxi, China, 3 School of Finance, Jiangxi University of Finance and Economics, Nanchang, China ‡ DC and CZ have contributed equally to this work and share first authorship. * 2202121925@stu.jxufe.edu.cn Abstract Environmental, social, and governance (ESG) has become a buzzword in investment circles as ecological damage and climate warming occur. ESG assessment is one of the important institutions of the green financial system, which plays a significant part in boosting corporate green development. We use the number of green patent applications and green patent cita- tions to measure corporate green innovation and analyze the micro-green effects of the ESG score system using the panel fixed effects models, which means that we explore the impact of the ESG scores on corporate green innovation performance, the specific mecha- nism of this effect, and the asymmetry of this impact under different moderation effects by using Chinese listed A-shares in Shanghai and Shenzhen from 2010–2019 as our research sample. We find that ESG positively affects corporate green innovation; the higher the ESG evaluation, the more it improves firms’ green innovation performance. The promotion effect is reflected quantitatively and qualitatively and remains valid after several robustness tests. In addition, the contribution of ESG to corporate green innovation is achieved through two main paths improving corporate investment efficiency and government-enterprise relations. Corporate black attributes inhibit the contribution of ESG to green innovation, while green attributes strengthen the contribution of ESG to green innovation performance. Our study demonstrates the importance of corporate participation in environmental, social, and gover- nance practices for corporate green innovation, which is beneficial for achieving win-win environmental, social, and economic results. Furthermore, our research completes the research on the effects of corporate green performance and green finance. It can provide empirical references for promoting corporate green development and improving the ESG evaluation system. Introduction Green innovation mainly emphasizes the sustainability of innovation, which describes novel products, processes, and techniques that might minimize the dangers to the environment, a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Zhang C, Chen D (2023) Do environmental, social, and governance scores improve green innovation? Empirical evidence from Chinese-listed companies. PLoS ONE 18(5): e0279220. https://doi.org/10.1371/journal. pone.0279220 Editor: Jose´ Antonio Clemente Almendros, Universidad Internacional de La Rioja, SPAIN Received: December 2, 2022 Accepted: May 3, 2023 Published: May 25, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0279220 Copyright: © 2023 Zhang, Chen. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the manuscript and its Supporting Information files. All files are available from PLOS ONE | https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 1 / 24 PLOS ONE thefigshare database (10.6084/m9.figshare. 22578787 10.6084/m9.figshare.22578808). Funding: The authors acknowledge the financial support from the project of the Water Economy and Water Rights Research Center, a school-level platform in Nanchang Institute of Technology: An empirical study on the Microeconomics of ESG performance under the ’Dual-carbon’ vision (22ZXZD01). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Do Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies pollution, and resource consumption throughout their life cycles [1–3]. Moreover, green inno- vation is an essential form for companies to practice the environmental, social, and governance concepts and an important tool to drive the green transformation of enterprises [4]. Green innovations for companies are low-carbon, energy efficient, and effective [5], but it also has characteristics such as long-term riskiness, public goods, and positive environmental externali- ties [6]. With the increasing economic globalization and industrialization, the natural world is subject to significant adverse impacts. Environmental pollution problems are becoming more prominent, severe climate problems are becoming more powerful, and green innovation may be essential to reconcile the contradictions between man and nature [7, 8]. Several elements drive corporate green innovation performance, as businesses need to maintain a competitive edge and increase corporate value. The literature has classified the fac- tors influencing green innovation into four categories. The first is market factors, including market pressure, green consumer demand, capital market opening, and environmental label- ing certification [9–14]. Corporate green innovation will be aided by the news media and pub- lic social supervision [15]. The second is environmental policy issues. Some studies have indicated that these rules can encourage corporate green innovation [16]. Some studies have discovered a link between environmental restrictions that first inhibit corporate green innova- tion and later promote it [17, 18]. Others have discovered pilot policies of emissions trading [19], low-carbon pilot policies [20, 21], emission permit systems [22], carbon emission trading systems [23, 24], green credit policies [25–27], clean production audit (CPA) program [28] and environmental information disclosure system [29–32] can stimulate enterprises to make green innovation. The third is the political-enterprise relationship, which manifests as political affiliation and government subsidies. Political affiliation inhibits firms’ green innovation, espe- cially when the market degree is low [33]. And subsidies have a driving influence on corporate green innovation performance. However, political affiliation encourages enterprises’ green innovation by raising R&D spending and organizational capital [34, 35], but some studies show no significant relationship between corporate subsidies and green innovation [36]. The fourth is internal corporate factors, CEO responsible leadership [3], executive academic expe- rience [37], sustainability goals [38], CSR performance [4], and internal controls of institu- tional investors all contribute to encouraging green business innovation. In a broad sense, environmental, social, and governance (ESG) can be seen as an extension of corporate social responsibility because it uses the three criteria of environmental, social, and internal governance to evaluate businesses [39], which reflects the degree of green transforma- tion, and environmental image of enterprises [40]. ESG is also an ESG investment concept pursued by investors and becomes an investment basket of ESG factors. Hence, the ESG con- cept gradually becomes the consensus of global enterprises, investors, and financial institutions [41, 42]. ESG is an essential indicator of corporate green development, which is gradually incorporated into corporate development strategies [43, 44]. Current research has focused on the link between ESG and company performance. Some argue that ESG is unrelated to corpo- rate profitability, cost of capital, or ESG deteriorates corporate performance [45]. Other researchers find that ESG scores can alleviate firms’ financing constraints and improve their business performance [46–49]. In addition, better business stock returns with correspondingly higher stock liquidity and a dampening effect on crash risk are linked to higher ESG ratings [50, 51]. ESG ratings can help firms improve innovation performance and corporate value [52, 53]. However, other research believes ESG has a detrimental effect on corporate value [54]. Furthermore, a study found ESG investors in the Asia-Pacific region and the US perform simi- larly to the market. ESG investments are more suitable for ’value-driven investors’ (VDI). It also found that European investors will pay the price for making ESG investments, which is not conducive to improving company performance [55]. PLOS ONE | https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 2 / 24 PLOS ONE Do Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies As market competition becomes increasingly fierce, green innovation capability is becom- ing increasingly widely concerned by society. Companies’ protection and attention to the envi- ronment have been strengthened by deepening their ESG practices. Although companies have gradually paid attention to carrying out ESG practices and focus on the sustainable develop- ment route of enterprises, there aren’t many studies about how the ESG performance of com- panies affects corporate green innovation. A portion of the literature has focused solely on the social responsibility component, contending that social responsibility institutions and perfor- mance favorably influence the quantity and caliber of corporate green innovation [4, 56]. The ESG performance of a firm is not well represented by the social responsibility perspective, which is simply one component of that performance. Some scholars have pointed out the posi- tive association of environmental, social, and governance practices on corporate green innova- tion from three ESG dimensions. However, the sample is heterogeneous and covers different research settings [57]. Even though there is research that specifically investigates the influence of ESG on green innovation using ESG rating data for Chinese business channels [58], the paper has the following shortcomings: On the one hand, their main regression using whether firms receive ratings as a quasi-natural test is not very plausible because the sample includes unrated firms and the financial and green performance of firms that receive ratings is naturally higher than those of non-rated firms. Green innovation output and quality are correspond- ingly higher. Hence their empirical models are highly endogenous and cannot be considered a quasi-natural experiment, and the grouping method is not clean. On the other hand, they also discuss the effect of ESG-specific ratings on firms’ green innovation, with a much smaller sam- ple size than the stated DID regression, and the small difference in ratings does not reflect the difference in the refinement of corporate ESG performance, which therefore does not support their conclusions. It is significant to recognize that ESG performance can influence corporate innovations, specifically how it affects business performance, share price, and corporate value. Therefore, to understand how ESG performance affects corporate innovation activities, business perfor- mance, share price, and enterprise value, we must first understand how it affects those activities. Understanding the impact of ESG scores on corporate green innovation activities, the specific mechanisms, and the asymmetry of the impact in different circumstances is of utmost practical importance in the context of environmental pollution and resource depletion to realize green economic development and corporate green transformation. To empirically analyze the rela- tionship between corporate ESG scores and corporate green innovation, as well as its role mech- anism and moderating effect, we use the number of green patent applications and the number of green patent citations to measure corporate green innovation, build an empirical model using ESG scores from 2010 to 2019 and data on the quantity and quality of green innovation from 2011 to 2020. We also make an empirical model with a sample of listed Chinese companies in Shanghai and Shenzhen. Our findings confirm our research hypothesis by demonstrating a favorable relationship between company ESG performance and green innovation. Meanwhile, corporate ESG scores promote corporate green innovation activities mainly through two paths: improving investment efficiency and improving political and business rela- tions. In addition, the stronger the green attributes of firms, the stronger the ESG’s contribu- tion to green innovation performance; the stronger the black attributes of firms, the weaker the positive impact of ESG on green innovation performance. We use Bloomberg ESG Disclo- sure Scores published by Bloomberg as a proxy variable for ESG for the following reasons. On the one hand, the scores data is published by a non-Chinese organization. Thus, it is more independent in evaluating the ESG of enterprises. On the other hand, the variable is score data, which overcomes the original rating problems that are the non-refined and non-accurate evaluation of the ESG of enterprises. PLOS ONE | https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 3 / 24 PLOS ONE Do Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies The main contributions of our study are as follows. Firstly, to overcome the lack of refine- ment and precision of ESG performance measurement in previous studies [58], we use the ESG score variables instead of the original ESG ratings. Secondly, we tested the effectiveness of the ESG evaluation system from two perspectives of green innovation quantity and quality. Existing studies on ESG only analyze from three perspectives of corporate performance, stock price, and value ignoring the environmental and green attributes of ESG. Studies on green innovation explore from a quantitative standpoint ignoring the quality of green innovation. The ESG concept is better reflected in our study’s green innovation elements, and the green innovation quality is better reflected in the quantitative indicators used to measure green inno- vation. Thirdly, we examine the possible mechanisms of ESG influence on green innovation and analyze the asymmetry of ESG influence on green innovation in terms of the green and black markets. The remainder of the paper is as follows. Section 2 introduces the relevant theoretical foun- dations and then presents the relevant research hypotheses. Section 3 outlines the data selec- tion and model design. Section 4 presents and analyzes the empirical results. Section 5 provides robustness tests. In Section 6, the final section, we conclude with a discussion. Theory and hypotheses development ESG and green innovation Enterprises pay more attention to the relationship with corporate stakeholders, whether it is the green innovation activities based on technology or market-oriented business models [59], and put more emphasis on the creation of multiple integrated values based on innovation-led economic, social, and environmental [9] which are all associated with the ESG performance. Green innovation is a type of innovation where businesses try to use resources more efficiently and use less energy, and employ cutting-edge techniques to accomplish the twin objectives of economic and environmental performance [1]. Through green process product innovation [2, 60], businesses can reduce emissions and save energy. The advantages of companies’ environ- mental, social, and governance practices favorably increase the intensity of green technology innovation [57], so the impact of ESG on corporate green innovation is mainly reflected in the following three aspects. First of all, the environmental responsibility of enterprises contributes to the promotion of green innovation activities. Businesses’ production and operation activities are under signifi- cant pressure from the legal limits of environmental rules and the informal constraints of pub- lic environmental expectations. Enterprises engage in environmental responsibility while compelled to take steps to enhance their environmental performance to preserve a positive environmental reputation. Consequently, for environmental performance, energy saving, and emission reduction, enterprises must use green innovation technology to improve production technology to achieve clean production. And financial institutions consider companies’ envi- ronmental compliance when making investment and financing decisions. Therefore good environmental performance can alleviate financing constraints by reducing the financing cost of enterprises [61]. Furthermore, the environment is an essential external stakeholder for com- panies, and active corporate environmental responsibility helps to promote environmental cooperation. Companies are more likely to gain ideas about environmental management from their partners, including suppliers, to drive responsible green innovation projects [62]. Second, corporate social responsibility will promote green innovation by improving the relationship with stakeholders and providing them with the resources and information needed for green innovation activities. According to the stakeholder theory, actively undertaking social responsibility can assist companies in establishing broader and stronger relationships PLOS ONE | https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 4 / 24 PLOS ONE Do Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies with multiple stakeholders, such as customers, investors, regulators, and the public. These stakeholders will support enterprises’ green innovation activities by increasing consumption and investment [63]. Based on the resource-based theory, corporate social responsibility is conducive to gaining the trust of stakeholders, including investors and consumers, and getting the market and financial resources needed for green innovation. Companies increase their green innovation investment, thus promoting green corporate innovation [30, 58]. According to signaling theory, on the one hand, CSR has the "information effect" of alleviating informa- tion asymmetry and principal-agent problems, thus providing information to help enterprises make long-term decisions on green innovation activities, which enables them to obtain more green patent outputs [64]. On the other hand, the active fulfillment of social responsibility can send positive signals to the market about the good business performance of the company, which indicates that the company is capable of participating in social responsibility activities with its resources and helps to attract positive media attention and improve their reputation and brand image [65, 66], and enterprises faceless media pressure to enhance their risk toler- ance for innovation failure and stimulate their innovation energy, which in turn drives them to conduct green innovation activities with high uncertainty [67, 68]. In addition, higher par- ticipation in socially responsible activities enhances firms’ product market recognition [69]. As green markets develop and consumer demand soars dramatically, firms are more willing to engage in environmentally friendly green innovation activities to increase corporate value. Third, the better the internal governance, the higher the performance of corporate green innovation [70]. As green innovation activities have the characteristics of higher risk and lon- ger cycle, enterprises do not tend to make innovation investment decisions, thus hindering green innovation, but good corporate governance alleviates principal-agent conflicts through incentive and constraint mechanisms, prompting corporate management to increase corpo- rate R&D and innovation investment to achieve long-term sustainable corporate development [71, 72]. In addition, better internal governance can improve corporate performance, thus pro- viding continuous and stable financial support for long-term corporate green innovation activ- ities by mobilizing internal and external resources. Furthermore, ESG can promote corporate green innovation by optimizing corporate governance structures [73]. Gender diversity in the board of directors and executive management promotes more ESG practices in firms [74]. Board diversity can help companies become green organizations by promoting corporate ESG practices to stimulate green creativity, which drives companies to engage in green innovation [75]. Otherwise, ESG can help businesses adopt the ideas of sustainable development and crea- tive growth [76]. On the one hand, these ideas encourage companies to pursue energy conser- vation, emission reduction, and clean production goals, as well as to increase their innovation spending and adopt technology that reduces energy consumption and protects the environ- ment [77] to apply in their production and operation processes. On the other hand, enterprises with the guidance of green concepts invest their funds in green projects through green financ- ing to achieve a pro-environmental and pro-climate transformation of internal capital flows or make investment funds further greener to provide strong support for green activities, includ- ing green innovation activities, which can encourage corporate innovation in green. Accordingly, we obtained the research hypothesis: H1: ESG is positively correlated with corporate green innovation. Mechanism of investment efficiency The ESG evaluation system provides information and resources to support corporate invest- ment and forms constraints on corporate investment. In general, the higher the ESG score, the PLOS ONE | https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 5 / 24 PLOS ONE Do Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies better performance of the company, the better the relationship between the firm and its stake- holders, and the more willing they are to provide information and resources to support the firm. At the same time, the ESG score also imposes constraints on enterprises, or to sustain the current score, enterprises have to make more green investments to maintain their corporate image, and the relevant investments are by the policy call and public expectation, so ESG enhances the efficiency of enterprises’ investment through both resource support and invest- ment constraints. In addition, it has been shown that corporate social responsibility can reduce agency costs and information asymmetry, so ESG firms have a low cost of equity [78]. The higher the ESG score of a firm, the lower the cost of equity, which is conducive to further enhancing the firm’s investment efficiency. The higher the investment efficiency, the lower the inefficient investment, the closer to the optimal investment level, the more the resource utiliza- tion rate is about sufficient, the more the innovation output, and the more the green innova- tion output, that is, the more the green innovation patents obtained by the enterprise. Therefore, the higher the ESG score, the more efficient the firm’s investment and green inno- vation output. Accordingly, we propose the research hypothesis: H2: ESG can promote corporate green innovation by promoting investment efficiency. Mechanism of government-business relations The higher the ESG score, the better the relationship between the company and its stakehold- ers. In Asia, many government-backed investment funds inject large amounts of money into ESG activities to reflect the importance of ESG practices for social development [79]. In China, companies pay particular attention to their relationship with the government because a good relationship with the government provides them with government support, such as govern- ment subsidies and tax breaks, and facilitates their financing, production, and management by obtaining government approval. In recent years, the local ecological environment has been related to the performance of the local government. Government regulation, technology push, and market pull are the three major influencing factors on carbon technology innovation activities. Government regulation is the only factor positively influencing carbon technology innovation activities [80]. The promotion of green technology innovation in China cannot be achieved without the power of the government, and the connection between the government and firms will impact enterprises’ green technology innovation activities. Therefore, the better the ESG performance of a company, the more the government will support it, and conversely, its development will be restricted by the government. Since green innovation is long-term and risky [6], this greatly constrains the willingness and confidence of firms to make green innova- tion decisions. However, firms that maintain a good relationship with the government can gain more government support to share innovation risks and losses [34], encouraging firms to engage in green innovation activities. In summary, ESG scores can improve the relationship between government and firms, provide more resources for green innovation, and thus pro- mote innovation. Therefore, we develop the following research hypothesis. H3: ESG scores promote corporate green innovation through improving government-busi- ness relations. Moderation effect of green and black attributes The ESG evaluation, one of the critical components of the green financial system, can contrib- ute to green finance by promoting the effectiveness of financial resource allocation through the green flow of funds, thereby addressing the issue of environmental externality. This system primarily affects the financing of small and medium-sized businesses. And the companies internal and external environmental variables impact the green micro effect of the ESG system. PLOS ONE | https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 6 / 24 PLOS ONE Do Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies The stronger the black attribute, the stronger the environmental information asymmetry, the greater the environmental risk, and the more inclined enterprises are to make green bleaching behavior to cover up the poor environmental performance, thus maintaining the false ESG score and the external regulation will identify the green bleaching behavior of enterprises and thus inhibit the role of ESG. Nevertheless, when green attributes are stronger, environmental information asymmetry is smaller, and the environmental risks faced by firms are reduced, prompting ESG scores to be more objective to the ESG performance of firms, thus making full utilization of the green micro effects of the ESG system. In summary, we obtain the following research hypotheses. As a green attribute of a com- pany, an increase in environmental disclosure is conducive to promoting corporate ESG prac- tices. Such environmental and ethical practices can promote the legitimization of corporate activities, improve corporate image and thus increase corporate financial performance [81]. Companies increase their investment in green technology innovation and enhance their inno- vation capabilities. H4: Black attributes can weaken the positive effect of ESG scores on the green innovation of firms, and green attributes can enhance the promotion effect of ESG scores on the green inno- vation of a company. Methodology Sample and data The sample of this study is a research sample of Chinese listed businesses in Shanghai and Shenzhen A-shares from 2010–2019 to analyze the impact of ESG on corporate green innova- tion performance. We conduct the following treatment for the sample: firstly, we remove the samples that were ST, PT, and *ST; secondly, we remove listed companies in the financial sec- tor; thirdly, we remove companies listed before 2010; fourthly, we remove the samples with missing main variables. After processing, we finally obtained 8258 annual observations of 1090 listed companies. We use a 1% and 99% tail reduction (Winsorize) for the primary variables. The data green on patents is from the China Research Data Service Platform. The data (CNRDS) on corporate finance is from the CSMAR and Wind databases, data on environmen- tal disclosure from social responsibility reports published by Hexun.com, data on corporate ESG scores are from Bloomberg’s Corporate Social Responsibility Disclosure Index (Bloom- berg ESG Disclosure Scores), regional environmental data and economic data are from pro- vincial statistical yearbooks, and macroeconomic data are from CEINet. We declare that we have no human participants, human data, or human issues. We do not have any individual person’s data in any form. Variables Explained variable. The explanatory variables in this paper are corporate green innova- tion. We define firms’ green innovation performance as quantitative and qualitative to obtain two explanatory variables for the number of green innovations (GI) and green patent citations (GC). The green patent is the most widely selected indicator to measure the green innovation ability of enterprises. The number of green patents granted reflects an enterprise’s green inno- vation level more than the number of green patent applications, so we add one to the number of green patents granted and take the logarithm to measure the quantity of green innovation (GI) of enterprises. For the quality of green innovation, most existing scholars choose to mea- sure the number of green invention patents and the number of green patents cited, among which the number of patents cited is more convincing than the invention patents [58], so in PLOS ONE | https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 7 / 24 PLOS ONE Do Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies this paper, we choose the number of green patents cited plus one and take the natural loga- rithm to measure the quality of green innovation (GC) of enterprises. Explanatory variable. The core explanatory variable in this paper is the ESG score of firms. The ESG data is derived from the Bloomberg ESG Disclosure Scores, which consists of the ESG composite score and the ESG sub-scores with 122 sub-scores on 21 topics in three major categories. Intermediary variables. (1) Efficacy of investments comes first (IE). We utilize the absolute value of the residuals from the subsequent regression to measure inefficient investment as Model (1) [82]. The larger the indicator, the less efficient the firm’s investment. CIi;t ¼ b0 þ b1SGi;t(cid:0) 1 þ εi;t ð1Þ In Model (1), CIi,t represents the investment level of an enterprise, which is the proportion of fixed and intangible assets to total assets. ESGi,t represents the investment opportunity of an enterprise, which is the growth rate of sales revenue. The residual term represents the propor- tion of inefficient investment in the total investment, and the absolute value is taken to obtain the investment efficiency index IE. The larger the value, the less efficient the investment. (2) Government subsidies (Subsidy). We use the normalized government subsidy (Subsidy) as a proxy variable for the government-enterprise relationship, which reflects the characteristics of the sample. The larger value indicates that means, the more government subsidy a firm receives, the better the relationship between the firm and the government Control variables. By previous studies [4, 15, 56, 83], we take into account variables such as the firm’s age (year of foundation), gearing (leverage), return on total assets (ROA), and Tobin’s Q. (Q), net cash from investing activities (ICF), fixed assets (Fix), foreign ownership (QFII), dual employment (Dual), and audit opinion (Opinion). The key variables used in the empirical analysis are shown in Table 1. Model Baseline model. Our data are short panel data, so a baseline regression model can repre- sent the significant relationship between the independent variable ESG score and the depen- dent variable green innovation level. We use this model to control for year-fixed, industry- fixed, and province-fixed effects to control for the effect of unobservable factors at the industry and province levels overtime on the relationship between ESG score firms and green innova- tion level, and to city-level clustering. In addition, we can use the model to further examine the mechanisms and moderators of ESG scores affecting firms’ green innovation. Based on the prior analysis and variable definitions, we use Model (2) for testing hypothesis H1. GIi;tþ1 ¼ a0 þ a1ESGi;t þ gXi;t þ lt þ Zj þ εi;t ð2Þ Where GIi,t+1 repents the firm i’s level of green innovation in year t+1, ESGi,t denotes the firm i’s Bloomberg ESG score in that year, Xi,t suggests a series of control variables, λt denotes time fixed effects, ηj denotes industry fixed effects; and εi,t represents the random disturbance term. Intermediation model. To test H2 and H3, the mediating effects of investment efficiency (IE) and government-enterprise relationship (Subsidy), this paper further sets up the following mediation model and sets up the following testing steps [84, 85]. First, Model (1) shows the results of the regression model of corporate green innovation on ESG score. If β1 is significant, PLOS ONE | https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 8 / 24 PLOS ONE Do Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies Table 1. Descriptive statistics of the variables. Variable classification Variable name Variable symbol Variable definition Explained variables Quantity of Green Innovation Quality of Green Innovation Core explanatory variables ESG Score Intermediate variables Investment efficiency Control variables Years of Establishment Government Grants Gearing Ratio Total Return on Assets Tobin’s Q Net cash from investing activities Fixed Assets Foreign equity holdings Two positions in one GIt+1 GCIt+1 ESG IE Subsidy Age Leverage ROA Q ICF Fix QFII Dual The logarithm of the number of green patents granted plus one to take the logarithm The logarithm of the number of green patent citations plus one to take the logarithm The logarithm of Bloomberg ESG Estimated from the Model (1) Normalized government grants Ln(year—year of establishment) Total liabilities/total assets Total profit/total assets Total market capitalization/total assets Net cash from investing activities/total assets Fixed Assets/Total Assets Foreign shareholding ratio The value is 1 if the chairman is also the general manager; otherwise, it is 0 https://doi.org/10.1371/journal.pone.0279220.t001 Audit opinion Opinion The standard unqualified opinion takes the value of 1; otherwise, it is 0 the second step is carried out. Second, the regression equation of ESG score and mediating variables (IE and Subsidy) on corporate green innovation is constructed. The mediating mech- anism exists if μ2 is significant and the signs of μ2 and β1 are the same. IEi;t=Subsidyi;t ¼ b0 þ b1ESGi;t þ gXi;t þ lt þ Zj þ εi;t GIi;tþ1 ¼ m0 þ m1ESGi;t þ m2IEi;t=Subsidyi;t þ gXi;t þ lt þ Zj þ εi;t ð3Þ ð4Þ Where IEi,t, and Subsidyi,t represent the investment efficiency and government subsidies, respectively, and the rest of the variables are consistent with the baseline model. Moderating effect model. To test H4, the moderating effect of the environmental attri- butes of firms, the following regression Model(5) was set up based on the baseline model. GIi;tþ1 ¼ a0 þ a1ESGi;t þ a2ESGi;t � Rit þ gXi;t þ lt þ Zj þ εi;t ð5Þ Where R consists of the black and green attributes of the company. Black attributes include regional, industry, and company pollution attributes. We use the high pollution region dummy variable HPP (The regional pollution index for the current year takes a value of 1 if it is higher than the average value, and 0 otherwise.), the high pollution industry dummy variable HPI (high pollution industry takes a value of 1 otherwise it takes a value of 0) and the high pol- lution company dummy variable HPC (If the enterprise is a key pollution monitoring unit take the value of 1, otherwise it takes the value of 0) separately to measure black attributes. Green attributes include provincial, city, and firm environmental attributes. We employ pro- vincial green finance DGF (normalized green finance index), city green innovation DGI (ratio of the total number of green patents in the city to the current year’s average), and corporate environmental disclosure (the number of quantitative disclosures of environmental liability items as a proportion of the total number of items) as green attributes. And the remaining var- iables are consistent with Model 2. PLOS ONE | https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 9 / 24 PLOS ONE Do Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies Table 2. Descriptive statistics of the main variables. Variables GIt+1 GCt+1 ESG E S G Age Leverage ROA Q ICF Fix QFII Dual Opinion N 8258 8258 8258 6950 8035 8258 8258 8258 8258 8258 8258 8258 8258 8258 8258 Mean S. D. 0.25 0.41 2.97 2.16 3.07 3.80 2.87 0.47 7.29 1.90 -0.06 0.23 0.17 0.20 0.99 0.62 0.92 0.31 0.67 0.41 0.11 0.32 0.20 6.17 1.24 0.08 0.18 0.54 0.40 0.12 Max 2.48 4.56 3.77 3.72 4.03 4.05 3.53 0.89 36.44 8.78 0.17 0.70 2.79 1.00 1.00 Median 0.00 0.00 2.99 2.23 3.13 3.80 2.89 0.48 5.95 1.48 -0.05 0.19 0.00 0.00 1.00 Min 0.00 0.00 2.21 0.73 1.95 3.52 1.61 0.05 -8.26 0.88 -0.39 0.00 0.00 0.00 0.00 https://doi.org/10.1371/journal.pone.0279220.t002 Descriptive statistics. The results of the descriptive statistics for the primary variables are shown in Table 2, where the mean value of green patents (GI) is 0.25, the standard deviation is 0.62, the maximum value is 0.48, and the minimum value is 0. This data suggests that the sam- ple enterprises’ average level of green innovation is low and that there is significant enterprise- level variation in their capacity for green innovation. ESG scores (ESG) vary significantly among businesses; the mean value is 2.97, the standard deviation is 0.31, the maximum is 3.77, the minimum is 2.21, and the median value is 2.99. Correlation test. The Pearson correlation coefficient test matrix is displayed in Table 3. We can infer from Table 3 that there is a significant positive association between ESG score and corporate green innovation, which supports H1 preliminarily. Panel unit root test. The existence of unit roots in panel data can have serious conse- quences, such as pseudo-regression, so we use both the Im-Pesaran-Shin test and Levin-Lin- Chu test to perform unit root tests to ensure the smoothness of each variable. Table 4 shows the results of the panel unit root tests. It can be seen that all variables are stationary at the 1% level, which means no unit root exists in the series. The results strongly reject the null Table 3. Pearson correlation coefficient test. GIt+1 1.000 0.513*** 0.118*** 0.110*** 0.101*** 0.028*** GCt+1 1.000 0.200*** 0.182*** 0.154*** 0.098*** ESG E S G 1.000 0.833*** 0.820*** 0.514*** 1.000 0.508*** 0.260*** 1.000 0.306*** 1.000 GIt+1 GCt+1 ESG E S G Note ***p < 0.01 **p < 0.05 *p < 0.1. https://doi.org/10.1371/journal.pone.0279220.t003 PLOS ONE | https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 10 / 24 PLOS ONE Do Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies Table 4. Panel unit root test. Variables GIt+1 GCt+1 ESG Age Leverage ROA Q ICF Fix QFII Dual Opinion Im-Pesaran-Shin test Levin-Lin-Chu test t-bar -1.814 -1.898 -1.657 -2.509 -1.850 -2.044 -1.838 -2.033 -1.984 -6.425 -4.737 -1.988 W[t-bar] -6.819 -8.581 -3.540 -21.349 -7.574 -11.631 -7.325 -11.404 -10.378 -103.261 -67.945 -10.455 P-value 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** t-value -49.167 -47.758 -45.468 -71.791 -56.996 -57.075 -55.392 -50.308 -57.179 -161.793 -391.658 -283.569 t-star -9.414 -37.025 -24.492 -65.239 -44.343 -38.191 -33.445 -28.543 -41.668 -164.627 -415.238 -300.026 P-value 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** https://doi.org/10.1371/journal.pone.0279220.t004 hypothesis of unit root, so we can argue that the data are stable and there is no biased informa- tion in the panel. Empirical results and analysis Baseline regression The ESG benchmark regression results are shown in Table 5. The explanatory variables in col- umns (1)-(4) are the quantity of green innovation. In column (1), the coefficient of ESG on the number of green innovation patents is 0.300, which is significant at the 1% level, indicating that ESG can increase the number of green innovation patents for companies. Based on the three sub-items of the ESG evaluation, we replace ESG with the natural logarithm of the corre- sponding scores for Environmental E, Social S, and Corporate Governance G. In column (2)- (4), the coefficient estimates of E and S are significantly positive at the 1% level, and the coeffi- cient estimates of G is significantly positive at the 10% level, indicating that E, S, and G scores all promote the level of green innovation in companies. The explanatory variables in columns (5)-(8) are the quality of green innovation. In column (5), the regression coefficient of ESG is 0.610, which is significant at the 1% level, which suggests that ESG encourages business cita- tion of green innovation patents. In columns (6)-(8), E, S, and G coefficient estimates are all significantly positive at the 1% level. The coefficient values are increasing in order, demon- strating that the positive effects of E, S, and G on the quality of green patents are in the order of G, S, E. The result above indicates that E, S, and G scores all promote the quality of green inno- vation in companies. The regression results show that the amount and quality of green innovation output increase with increasing ESG score, supporting H1. In addition, our regression results indicate that all three subcategories of ESG can promote the quantity and quality of green innovation in enterprises. For the subscores of corporate ESG scores, we find that the E score has the most significant impact on corporate green innovation, and the G score has the least significant impact on corporate green innovation. Still, overall, the subscores of ESG all drive the quantity and quality of corporate green innovation. The descriptive statistics of the remaining control variables are generally consistent with existing studies [35, 55, 58]. The results illustrate that ESG scores can increase the quantity and quality of green innova- tion and that ESG is a sustainable "substantive innovation" rather than a "masked innovation" PLOS ONE | https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 11 / 24 PLOS ONE Do Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies Table 5. Baseline regression results. Variables ESG E S G Age Leverage ROA Q ICF Fix QFII Dual Opinion Constant Y/I/P FE Observations Adj R2 (1) GI_1 0.300*** (7.04) -0.193** (-2.09) 0.183** (2.35) 0.002 (0.89) -0.028** (-2.30) -0.079 (-0.63) -0.035 (-0.26) 0.026 (0.83) 0.041 (1.16) 0.081* (1.80) -0.128 (-0.40) YES 8,258 0.114 (2) GI_1 (3) GI_1 (4) GI_1 0.123*** (6.68) -0.194** (-2.00) 0.198** (1.98) 0.002 (0.95) -0.037*** (-2.62) -0.125 (-0.78) -0.069 (-0.49) 0.034 (1.08) 0.047 (1.26) 0.087 (1.61) 0.498 (1.63) YES 6,950 0.120 0.167*** (5.75) -0.195** (-2.11) 0.201*** (2.66) 0.003 (1.22) -0.032*** (-2.72) -0.063 (-0.48) 0.018 (0.13) 0.029 (0.91) 0.035 (1.03) 0.079* (1.76) 0.231 (0.73) YES 8,035 0.107 0.312* (1.75) -0.203** (-2.27) 0.209*** (2.83) 0.003 (1.17) -0.034*** (-2.86) -0.066 (-0.50) 0.035 (0.25) 0.029 (0.90) 0.029 (0.76) 0.102** (2.19) -0.446 (-0.55) YES 8,258 0.0975 Note: T-statistics calculated for city-level clusters in parentheses. https://doi.org/10.1371/journal.pone.0279220.t005 (5) GC_1 0.610*** (7.32) -0.223 (-1.33) 0.334** (2.16) -0.006* (-1.78) 0.012 (0.68) -0.440* (-1.86) -0.387*** (-2.74) 0.032 (0.75) 0.096 (1.27) -0.123 (-1.18) -0.712 (-1.28) YES 8,258 0.0993 (6) GC_1 (7) GC_1 (8) GC_1 0.267*** (6.87) -0.217 (-1.12) 0.390** (2.01) -0.007* (-1.87) -0.006 (-0.28) -0.590** (-1.99) -0.478*** (-2.91) 0.044 (0.98) 0.103 (1.25) -0.080 (-0.68) 0.373 (0.66) YES 6,950 0.0997 0.322*** (6.39) -0.226 (-1.34) 0.379** (2.56) -0.005 (-1.54) 0.004 (0.24) -0.411* (-1.67) -0.290* (-1.92) 0.042 (0.95) 0.077 (1.04) -0.122 (-1.14) 0.017 (0.03) YES 8,035 0.0830 0.981*** (3.11) -0.249 (-1.52) 0.367** (2.40) -0.006 (-1.55) 0.003 (0.16) -0.424* (-1.71) -0.260* (-1.74) 0.038 (0.86) 0.076 (0.96) -0.087 (-0.78) -2.645* (-1.80) YES 8,258 0.0754 to simply whitewash financial statements. It is worth mentioning that the G score affects the number of green innovations less significantly than the E and S scores, probably because green innovation projects crowd out the firm’s inherent resources and conflict with its short-term financial performance. We also find that when the explanatory variable is replaced with the number of green patents cited, all three aspects of ESG significantly improve the quality of green innovation at the 1% level. The coefficient of the G score is the largest. This result indi- cates that executives value the strategic perspective of the company’s long-term development and choose to make high-quality green innovations to improve the company’s competitive- ness, so companies with good green strategies significantly improve the quality of green innovation. Our results affirm the positive significance of ESG practices for green innovation, which positively affect companies’ green transformation. The results also demonstrate the critical PLOS ONE | https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 12 / 24 PLOS ONE Do Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies role of the ESG scores of companies in influencing their green innovation decisions and that favorable practices in environmental, social, and governance aspects of companies will jointly promote corporate green innovation, achieve a sustainable development path for enterprises, and promote the integration of environmental, social and economic effects of enterprises. Intermediary mechanism analysis Mechanism of investment efficiency. The regression results for the mediating influence of investment efficiency are shown in columns (1)-(3) of Table 5. The coefficient estimate of ESG in column (1) is -3.183 and significantly negative at the 1% level, which means businesses with higher ESG scores make better investors. The coefficient estimates of IEt+1 in columns (2) and (3) are -0.003 and -0.005, respectively, and are statistically negative at the 5% level, indicat- ing the existence of this mediating effect and the relationship between investment efficiency and green innovation performance, At the 1% level, both of the coefficient estimates of ESG in columns (2) and (3)—0.291 and 0.594, respectively—are significantly positive. Both columns (2) and (3) coefficient values of ESG 0.291 and 0.594, respectively—are statistically significant at the 1% level. The regression results suggest that ESG performance contributes to green inno- vation by improving firms’ investment efficiency. As a result, H3 should be accepted. The results suggest that the fulfillment of ESG responsibilities will drive companies to make green investments to cater to investors’ preference for environmentally friendly companies and that ESG practices are conducive to improving the efficiency of investments and the utili- zation of internal and external resources, which in turn will make companies willing to engage in more green innovation activities and improve their green technological innovation capabilities. Mechanism of government-business relations. Columns (4) to (6) of Table 6 show the regression results for the mediating effect of the government-firm relationship. The coefficient estimate of ESG in column (4) is 0.027 and significantly positive at the 1% level, indicating that the higher the ESG score, the more government subsidies the firm receives. In other words, the ESG score significantly improves the relationship between the government and the firm; the coefficient estimates of Subsidy in columns (5) and (6) are respectively 0.2289 and 5.407, and both are positive at the 1% level, which means that government subsidies significantly pro- mote green innovation, so the better the relationship with the government, the more govern- ment subsidies the enterprises receive, and the more funds they have to engage in green Table 6. Regression results for mediating mechanisms. (1) IEt+1 -3.184*** (-4.65) Variables ESG IEt+1 Subsidy Constant 13.907*** Controls Y/I/P FE Observations Adj R2 (3.50) YES YES 8,258 0.082 https://doi.org/10.1371/journal.pone.0279220.t006 (2) GIt+1 0.291*** (7.16) -0.003** (-2.53) -0.109 (-0.34) YES YES 8,258 0.139 (3) GCt+1 0.594*** (7.33) -0.005** (-2.38) -0.673 (-1.25) YES YES 8,258 0.163 (4) Subsidy 0.027*** (3.23) -0.036* (-1.69) YES YES 8,258 0.252 (5) GIt+1 0.239*** (5.14) 2.289*** (10.80) -0.065 (-0.21) YES YES 8,258 0.115 (6) GCt+1 0.465*** (5.50) 5.407*** (7.58) -0.541 (-1.04) YES YES 8,258 0.101 PLOS ONE | https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 13 / 24 PLOS ONE Do Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies innovation behavior. Thus this mediating effect exists. The coefficient value decreases com- pared to the baseline regression. Therefore the mediation effect is partial. The above regression results suggest that ESG performance promotes corporate green innovation by improving the relationship between government and business, which supports hypothesis H2. The results indicate the important role of government-business relationships in mediating the impact of ESG performance on corporate green innovation. The government encourages and supports ESG practice projects, so companies that participate in ESG practice projects can build a good corporate image, maintain good relations with the government, and gain political resources, including government subsidies and the economic resources they bring. These com- petitive resources can be regarded as a kind of external risk protection, which can reduce the cost of green innovation and reduce the risk of R&D, and enhance the motivation of enter- prises to invest in green innovation projects. Moderation effects analysis The regression results of Panel A in Table 7 show that the interaction coefficients of ESG with HPP, HPI, and HPC decrease in significance and coefficient values compared with the esti- mated values of the baseline regression ESG, which indicates that the stronger the black attri- butes of the firm, the weaker the promotion effect of ESG on green innovation. The regression Table 7. Regression results for moderating effects of black and green attributes. Variables ESG×HPP ESG×HPI ESG×HPC Constant Controls Y/I/P FE Observations Adj R2 ESG×DGF ESG×CGI ESG×EDG Constant Controls Y/I/P FE Observations Adj R2 (1) GIt+1 0.026 (1.40) 0.693** (2.40) YES YES 8,258 0.095 0.386*** (3.99) -0.469* (-1.66) YES YES 8,258 0.104 https://doi.org/10.1371/journal.pone.0279220.t007 (2) GCt+1 0.101** (2.54) 0.977** (1.99) YES YES 8,258 0.068 0.826*** (3.83) -1.521*** (-2.64) YES YES 8,258 0.083 (3) GIt+1 Panel A: Black Features 0.015* (1.78) 0.677** (2.31) YES YES 8,258 0.096 Panel B: Green Features 0.572* (1.93) 0.820** (2.58) YES YES 5,439 0.127 (4) GCt+1 0.038** (2.24) 0.930* (1.86) YES YES 8,258 0.068 1.306** (2.48) 1.042** (2.13) YES YES 5,439 0.125 (5) GIt+1 (6) GCt+1 0.011 (1.43) 0.698** (2.38) YES YES 8,258 0.095 0.856*** (4.80) 0.673** (2.34) YES YES 8,258 0.010 0.067*** (4.14) 1.024** (2.05) YES YES 8,258 0.072 1.360*** (4.31) 0.933* (1.91) YES YES 8,258 0.071 PLOS ONE | https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 14 / 24 PLOS ONE Do Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies results of Panel B show that the coefficient values of ESG and DGF, CGI, and EDG are all signif- icant at the 1% level and are greater than the baseline regression coefficient values, which sug- gests that the stronger the green attributes of firms, the greater the positive impact of ESG on green innovation. The results demonstrate the opposed effects of corporate green and black attributes on the relationship between ESG scores and corporate green innovation. At the black attribute level, the sample, whether a highly polluting industry or a highly polluting firm, exacerbates environ- mental information asymmetry and exposes firms to higher environmental risks. Firms will mask the inherent risks through green bleaching practices. Thus ESG scores are more biased towards a false reflection of ESG performance and will weaken the positive effect of ESG scores on green innovation. At the level of green attributes, whether at the province, city, or firm level, green attributes can reduce environmental information asymmetry, make ESG scores more realistic and reliable reflections of firms’ true ESG performance, and enhance the effec- tiveness of ESG scores in promoting corporate green innovation. The government should increase the punishment for polluting enterprises, increase the cost of polluting enterprises through environmental regulation pressure, and consciously promote the transformation of enterprises from black attributes to green attributes. And enterprises should increase the dis- closure of environmental information to reduce the uncertainty of environmental information and enhance their green attributes, and at the same time, reduce emissions and environmental pollution by improving production processes and greening production to reduce their black attributes, to better utilize the positive effect of ESG performance on green innovation. Robustness tests Replacing measures of core variables In the robustness test section, we use the number of green patent applications to measure the quantity of green innovation of the firm (GGI_1) and the number of green invention patents to measure the quality of green innovation of the firm (INNO_1). Table 8 reports the regres- sion results for replacing the core variable measures. The results are consistent with the bench- mark regression, where both the composite corporate ESG score and sub-scores contribute to the quantity and quality of corporate green innovation. Table 8. Regression results for replacing core variables. Variables ESG E S G Constant Controls Y/I/P FE Observations Adj R2 (1) GGI_1 0.389*** (6.02) -1.058*** (-3.25) YES YES 6,891 0.165 (2) GGI_1 (3) GGI_1 (4) GGI_1 0.186*** (6.28) -0.345 (-1.06) YES YES 5,724 0.169 0.187*** (4.16) -0.530 (-1.59) YES YES 6,675 0.153 0.326* (1.82) -1.179 (-1.45) YES YES 6,891 0.150 https://doi.org/10.1371/journal.pone.0279220.t008 (5) INNO_1 0.264*** (5.48) -0.143 (-0.42) YES YES 8,258 0.0741 (6) (7) (8) INNO_1 INNO_1 INNO_1 0.122*** (4.76) 0.368 (1.29) YES YES 6,950 0.0803 0.153*** (4.40) 0.158 (0.49) YES YES 8,035 0.0682 0.472** (2.26) -1.154 (-1.19) YES YES 8,258 0.0616 PLOS ONE | https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 15 / 24 PLOS ONE Do Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies Fig 1. Placebo test. https://doi.org/10.1371/journal.pone.0279220.g001 Placebo test We used a non-parametric permutation test to perform a placebo test on the baseline regres- sion. The placebo test is illustrated in Fig 1. We find that from the test results that the distribu- tion of the estimated coefficients for the 500 random samples is close to a normal distribution with mean zero and that the coefficients of the benchmark regressions for GIt+1 and GCt+1 green innovation indicated by the dashed lines in the figure, Table 5 columns (3) and (6) are dif- ferent from the correlation coefficients obtained from the non-parametric tests. Therefore, the test results exclude the possibility that the effect of ESG on green innovation performance is dependent on other unobservable factors. In other words, the interference of other events in the benchmark regression is excluded, and the obtained benchmark regression results are robust. Adding variables We next control provincial and national level economic variables that may affect firms’ green innovation based on the baseline regression column (1) to verify the robustness of the baseline regression results. We specifically introduce regional per capita gross product (PerGDP, the logarithm of regional per capita gross product), regional financial development level (FD, regional deposit and loan as a share of GNP), regional pollution level (DPG, industrial pollu- tion investment as a share of GNP), broad money growth rate (M2), and Shanghai interbank lending rate (Shibor, the annual 10-year Shanghai interbank lending average interest rate) to control for regional economic, environmental and macroeconomic effects on the benchmark regressions. Table 9 shows the results of the regressions with the addition of control variables. From the regression results, the coefficient estimates for ESG are all significantly positive at the 1% level. The regression results are generally consistent with the benchmark regression, which means the robustness of the benchmark regression. PLOS ONE | https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 16 / 24 PLOS ONE Do Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies Table 9. Regression results for adding control variables. Variables ESG FD DPG M2 Shibor Constant Controls Y/I/P FE Observations Adj R2 (1) GIt+1 0.301*** (7.10) 0.002 (1.07) -0.208** (-2.56) -5.300*** (-2.73) YES YES 8,258 0.114 (2) GIt+1 0.301*** (7.10) 0.002 (1.07) -0.208** (-2.56) 0.288*** (5.05) 1.043*** (4.90) -10.082*** (-3.48) YES YES 8,258 0.114 (3) GCt+1 0.609*** (7.33) -0.002 (-1.21) 0.089 (0.59) -0.341 (-0.13) YES YES 8,258 0.074 (4) GCt+1 0.609*** (7.33) -0.002 (-1.21) 0.089 (0.59) -0.142* (-1.72) -0.476 (-1.55) 1.829 (0.46) YES YES 8,258 0.074 https://doi.org/10.1371/journal.pone.0279220.t009 Replacement regression models GIt+1 and GCt+1 are discrete variables suitable for Poisson, Tobin, and Negative Binomial regression models. Table 10 shows the results of the substitution regression model. From the regression results, the coefficient estimates of ESG are all significantly positive at the 1% level. The regression results are generally consistent with the baseline regression, which suggests the robustness of the baseline regression. Instrumental variables approach Using green innovation indicators for period t+1 avoids the problems associated with certain simultaneity biases while reducing the estimation error associated with reverse causality issues. However, the relationship between ESG and green innovation is still strongly endogenous, which means that firms with higher green innovation performance also have higher ESG scores. There may also be omitted variables that affect ESG scores. At the same time, there is Table 10. Regression results of the replacement model. Variables ESG Constant Controls Y/I/P FE Observations Loglikelihood Pseudo R2 (1) Poisson 1.181*** (8.24) -4.124*** (-3.71) YES YES 8,258 -4437 0.186 https://doi.org/10.1371/journal.pone.0279220.t010 (2) GIt+1 Tobit 0.300*** (7.10) -0.148 (-0.46) YES YES 8,258 -7176 0.068 (3) NB 1.169*** (7.35) -3.901*** (-3.41) YES YES 8,258 -4338 0.147 (4) Poisson 1.376*** (9.37) -4.029*** (-3.72) YES YES 8,258 -6926 0.118 (5) Tobit GCt+1 0.609*** (7.36) -0.737 (-1.33) YES YES 8,258 -10599 0.042 (6) NB 1.378*** (8.08) -3.857*** (-3.43) YES YES 8,258 -6345 0.067 PLOS ONE | https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 17 / 24 PLOS ONE Do Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies Table 11. 2SLS and GMM results for instrumental variables. Variables ESGMeant-1 ESG Constant Observations Controls Y/I/P FE Adj R2 F statistics Kleibergen-Paaprk LM statistic Cragg-Donald Wald F-statistic Kleibergen-Paaprk Wald F-statistic https://doi.org/10.1371/journal.pone.0279220.t011 (1) ESG 0.679*** (17.29) 2.766*** (54.35) 8,258 YES YES 0.211 35.95 (2) 2SLS GIt+1 (3) GCt+1 (4) GIt+1 (5) GCt+1 GMM 0.757*** (6.89) -1.419*** (-4.45) 8,258 YES YES 0.071 19.64 235.653 262.190 299.097 1.156*** (6.56) -2.260*** (-4.45) 8,258 YES YES 0.072 12.26 235.653 262.190 299.097 0.755*** (6.87) -1.391*** (-4.37) 8,258 YES YES 0.071 1.159*** (6.58) -2.238*** (-4.40) 8,258 YES YES 0.072 simultaneously an impact on firms’ green innovation that makes the benchmark regressions biased and inconsistent. We use an instrumental variable to address this issue to eliminate the effect of potential endogeneity on the benchmark regression. This paper chooses the industry- level mean of ESG (ESGMeant-1) of the previous year as the instrumental variable [84]. The industry influences the ESG score, but the industry-level mean is not directly related to the green performance of individual firms, so ESGMeant-1 meets the requirements of an instru- mental variable. Before conducting the least squares regression of the instrumental variables, we first con- ducted a correlation coefficient test between ESGMeant-1 and ESG. The Pearson correlation coefficient test results showed that the correlation coefficient between the two was 0.194 and significant at the 1% level, so we can initially conclude that the higher the industry ESG means, the higher the ESG performance of the firm. The outcomes of the 2SLS and GMM results for the instrumental variables are shown in Table 11. The first three columns are the estimated results of 2SLS. According to the regression results, the first stage’s coefficient estimates of ESGMeant-1 is 0.679 and significant at the 1% level, suggesting that the industry in which a company operates impacts its ESG performance. The second stage regression shows that the predicted ESG coefficients are considerably positive at the 1% level, demonstrating that ESG improves business performance regarding green innovation. After conducting the main regression, we conduct a series of tests for instrumental variables such as homogeneity of instrumental variables, weak instrumental variables, and over-identification, whose results show that the Model passes all tests. The last two columns are the estimated results of GMM. The regression results also validate the baseline hypothesis of this paper. Propensity score matching To address the problem of sample selection bias, we choose the propensity score matching method (PSM), using a 1:1 nearest neighbor matching with a matching radius of 0.05, with whether it is a highly polluting industry as the grouping variable and all the control variables in column (1) as covariates, inducing age of establishment (Age), gearing (Leverage), return on total assets (ROA), Tobin’s Q (Q), net cash from investing activities (ICF), fixed assets (Fix), PLOS ONE | https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 18 / 24 PLOS ONE Do Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies Table 12. PSM-benchmark regression results. Variables ESG E S G Constant Y/I/P FE Observations Adj R2 (1) GIt+1 0.260*** (4.92) 0.016 (0.05) YES 3,379 0.109 (2) GIt+1 (3) GIt+1 (4) GIt+1 0.099*** (4.06) 0.663*** (2.69) YES 2,894 0.111 0.138*** (3.85) 0.317 (1.16) YES 3,274 0.103 0.284 (1.57) -0.361 (-0.45) YES 3,379 0.098 https://doi.org/10.1371/journal.pone.0279220.t012 (5) GCt+1 0.504*** (5.98) -0.129 (-0.23) YES 3,379 0.082 (6) GCt+1 (7) GCt+1 (8) GCt+1 0.236*** (4.89) 0.867* (1.84) YES 2,894 0.080 0.260*** (6.10) 0.460 (0.96) YES 3,274 0.072 0.456 (1.05) -0.509 (-0.27) YES 3,379 0.069 foreign ownership (QFII), dual employment (Dual) and audit opinion (Opinion). After passing the common support hypothesis and parallel trend hypothesis tests, the benchmark regression was re-run, and the regression results are shown in Table 12. From the results, we find that the regression coefficients of ESG are all significantly positive at the 1% level. Meanwhile, the coef- ficient estimates of E and S are both significantly positive at the 1% level, but the coefficient estimate of G is not statistically significant after eliminating the problem of the sample, which indicates that the short-term corporate governance objectives of the company are contrary to the long-term green innovation activities, consistent with economic theory and experience. Conclusion and discussion Green innovation is a crucial manifestation of corporate applying the ESG concept, which reflects the micro-green effect of the ESG evaluation system. Using panel data and the sample of Chinese listed businesses from 2010 to 2019, we empirically explore the impact of ESG scores on corporate green innovation from corporate investment efficiency and government- enterprise relations perspectives. The results indicate both the composite and sub-scores of a company’s ESG contribute to the quantity and quality of its green innovation. And ESG sup- ports corporate green innovation by increasing businesses’ investment effectiveness and improving their government-business relationship. The results also show that corporate green attributes strengthen the promotion function of ESG on corporate green innovation. In con- trast, black attributes reduce the beneficial effects of ESG on corporate green innovation. According to our research, the following recommendations can be made for enhancing the ESG evaluation system and encouraging the sustainable growth of micro-enterprises. Firms need to implement the ESG concept, manage the various environmental risks they face, increase their level of pro-environmental preference, enhance the environmental disclosure mechanism, pay more attention to the non-financial performance of green performance, and promote business development and green development. The findings of this paper prove the importance of practicing environmental, social, and governance responsibilities and the posi- tive significance of ESG performance for enterprises’ green and sustainable development. The implementation of the ESG concept by enterprises is conducive to promoting the integration of environmental, social, and economic performance and achieving a win-win situation of environmental, social, and economic effects. Moreover, the findings of this paper also point PLOS ONE | https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 19 / 24 PLOS ONE Do Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies out the way and direction for enterprises to promote green innovation development. By actively fulfilling environmental and social responsibilities, enterprises can win the trust of stakeholders, including governments and investors, obtain key political and economic resources that are indispensable for green innovation, alleviate financing constraints, improve resource utilization, enhance the output and quality of green technology innovation, and embark on a sustainable green development transformation path. Moreover, the government should implement green development into practice, create fiscal policies for businesses based on the ESG evaluation system, subsidize green enterprises and restrict black enterprises, and encourage businesses to engage in green innovation activities that adhere to ESG standards. The conclusion of this paper proves the key role played by good political-business relations between ESG scores and corporate green innovation. Therefore, the government should focus on the role of important political and economic resources, including government subsidies and tax incentives, and strongly support enterprises to carry out ESG practice projects that are beneficial to social development and progress to attract enterprises to participate in green innovation activities consciously and actively, thus guiding more enterprises to take the green development path. Regulators should create distinct regulatory policies based on businesses’ environmental risks and enhance the mechanism for exchanging environmental information to encourage companies to engage in green innovation activities. Regulators should pay attention to the environmental information disclosure of enterprises, timely detect the possible "greenwashing" behavior of enterprises and punish these enterprises, to promote the ESG score to reflect the ESG performance of enterprises more truly and let the ESG performance promote the green innovation of enterprises in practice, that is, let the green attributes better promote the positive link between ESG score and green innovation of enterprises, and weaken the inhibiting effect of black attributes on the relationship between the two. Institutional investors need to pay attention to the ESG performance of enterprises and fur- ther incorporate ESG factors into their investment strategies to better identify enterprises’ internal and external environmental risks and provide enterprises with corresponding funds based on ESG evaluation. As an important external stakeholder of enterprises, enterprises will pay attention to the investment tendency of institutional investors to obtain more financing support. Therefore, institutional investors pay attention to ESG investment concepts, environ- mental protection of enterprises, and sustainable development strategies, which are conducive to guiding enterprises to pay attention to ESG practices, fulfilling environmental and social responsibilities, and enhancing their green innovation drive. The limitations of this paper lie in the following two aspects. On the one hand, we only explore the micro-green effect of the ESG evaluation system and do not analyze the role of the ESG evaluation system comprehensively. On the other hand, we ignored the motives of corpo- rate greenwashing and failed to eliminate the part of corporate greenwashing in green innova- tion. Future research can examine the relationship between ESG scores and green innovation from two aspects. First, the research can analyze the role of ESG in greenwashing behaviors such as environmental performance, production performance, and investment efficiency. Sec- ond, future research will have indicators to identify green innovation drifting green motives to better examine the effectiveness of the ESG evaluation system. Author Contributions Conceptualization: Danni Chen. Data curation: Chunlian Zhang, Danni Chen. PLOS ONE | https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 20 / 24 PLOS ONE Do Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies Formal analysis: Chunlian Zhang, Danni Chen. Methodology: Chunlian Zhang. Software: Danni Chen. Supervision: Danni Chen. Validation: Chunlian Zhang, Danni Chen. Visualization: Danni Chen. Writing – original draft: Chunlian Zhang, Danni Chen. Writing – review & editing: Chunlian Zhang. References 1. Xie X., Huo J., Zou H., "Green process innovation, green product innovation, and corporate financial performance: A content analysis method". Journal of Business Research, vol. 101, pp. 697–706, 2019. 2. Roh T., Lee K., Yang J. Y., "How do intellectual property rights and government support drive a firm’s green innovation? The mediating role of open innovation". Journal of Cleaner Production, vol. 317, Arti- cle ID 128422, 2021. 3. Wang Y., Shen T., Chen Y., et al., "CEO environmentally responsible leadership and firm environmental innovation: A socio-psychological perspective". Journal of Business Research, vol. 126, pp. 327–340, 2021. 4. Hao J., He F., "Corporate social responsibility (CSR) performance and green innovation: Evidence from China". Finance Research Letters, vol. 48, Article ID 102889, 2022. 5. Lin R.-J., Tan K.-H., Geng Y., "Market demand, green product innovation, and firm performance: evi- dence from Vietnam motorcycle industry". Journal of Cleaner Production, vol. 40, pp. 101–107, 2013. 6. Rennings K., "Redefining innovation—eco-innovation research and the contribution from ecological economics". Ecological Economics, vol. 32, no. 2, pp. 319–332, 2000. 7. Huang Y.-C., Chen C. T., "Exploring institutional pressures, firm green slack, green product innovation and green new product success: Evidence from Taiwan’s high-tech industries". Technological Fore- casting and Social Change, vol. 174, Article ID 121196, 2022. 8. Razzaq A., Wang Y., Chupradit S., et al., "Asymmetric inter-linkages between green technology innova- tion and consumption-based carbon emissions in BRICS countries using the quantile-on-quantile framework". Technology in Society, vol. 66, Article ID 101656, 2021. 9. 10. Triguero A., Moreno-Monde´ jar L., Davia M. A., "Drivers of different types of eco-innovation in European SMEs". Ecological Economics, vol. 92, pp. 25–33, 2013. Farza K., Ftiti Z., Hlioui Z., et al., "Does it pay to go green? The environmental innovation effect on cor- porate financial performance". Journal of Environmental Management, vol. 300, Article ID 113695, 2021. https://doi.org/10.1016/j.jenvman.2021.113695 PMID: 34649325 11. Cillo V., Petruzzelli A. M., Ardito L., et al., "Understanding sustainable innovation: A systematic literature review". Corporate Social Responsibility & Environmental Management, vol. 26, no. 5, pp. 1012–1025, 2019. 12. Islam T., Islam R., Pitafi A. H., et al., "The impact of corporate social responsibility on customer loyalty: The mediating role of corporate reputation, customer satisfaction, and trust". Sustainable Production and Consumption, vol. 25, pp. 123–135, 2021. 13. Sha Y., Zhang P., Wang Y., et al., "Capital market opening and green innovation——Evidence from Shanghai-Hong Kong stock connect and the Shenzhen-Hong Kong stock connect". Energy Economics, Article ID 106048, 2022. 14. Ren S., He D., Yan J., et al., "Environmental labeling certification and corporate environmental innova- tion: The moderating role of corporate ownership and local government intervention". Journal of Busi- ness Research, vol. 140, pp. 556–571, 2022. 15. Zhao L., Zhang L., Sun J., et al., "Can public participation constraints promote green technological inno- vation of Chinese enterprises? The moderating role of government environmental regulatory enforce- ment". Technological Forecasting and Social Change, vol. 174, Article ID 121198, 2022. 16. Wu B., Fang H., Jacoby G., et al., "Environmental regulations and innovation for sustainability? Moder- ating effect of political connections". Emerging Markets Review, vol. 50, Article ID 100835, 2022. PLOS ONE | https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 21 / 24 PLOS ONE Do Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies 17. Yang C.-H., Tseng Y.-H., Chen C.-P., "Environmental regulations, induced R&D, and productivity: Evi- dence from Taiwan’s manufacturing industries". Resource and Energy Economics, vol. 34, no. 4, pp. 514–532, 2012. 18. Zhong Z., Peng B., "Can environmental regulation promote green innovation in heavily polluting enter- prises? Empirical evidence from a quasi-natural experiment in China". Sustainable Production and Con- sumption, vol. 30, pp. 815–828, 2022. 19. Du G., Yu M., Sun C., et al., "Green innovation effect of emission trading policy on pilot areas and neigh- boring areas: An analysis based on the spatial econometric model". Energy Policy, vol. 156, Article ID 112431, 2021. 20. Hu G., Wang X., Wang Y., "Can the green credit policy stimulate green innovation in heavily polluting enterprises? Evidence from a quasi-natural experiment in China". Energy Economics, vol. 98, Article ID 105134, 2021. 21. Yang S., Wang W., Feng D., et al., "Impact of pilot environmental policy on urban eco-innovation". Jour- nal of Cleaner Production, vol. 341, Article ID 130858, 2022. 22. Ren K., Kong Y., Zhang T., et al., "The impact of the pollution permits system on green innovation: Evidence from the county-level data in China". Journal of Cleaner Production, vol. 344, Article ID 130896, 2022. 23. 24. Zhang W., Li G., Guo F., "Does carbon emissions trading promote green technology innovation in China?". Applied Energy, vol. 315, Article ID 119012, 2022. Zhou F., Wang X., "The carbon emissions trading scheme and green technology innovation in China: A new structural economics perspective". Economic Analysis and Policy, vol. 74, pp. 365–381, 2022. 25. Wang H., Qi S., Zhou C., et al., "Green credit policy, government behavior and green innovation quality of enterprises". Journal of Cleaner Production, vol. 331, Article ID 129834, 2022. 26. Wang Y., Li M., "Credit policy and its heterogeneous effects on green innovations". Journal of Financial Stability, vol. 58, Article ID 100961, 2022. 27. Zhang Y., Li X., Xing C., "How does China’s green credit policy affect the green innovation of high pollut- ing enterprises? The perspective of radical and incremental innovations". Journal of Cleaner Produc- tion, vol. 336, Article ID 130387, 2022. 28. Cui J., Dai J., Wang Z., et al., "Does Environmental Regulation Induce Green Innovation? A Panel Study of Chinese Listed Firms". Technological Forecasting and Social Change, vol. 176, Article ID 121492, 2022. 29. 30. Li H., Zhang X., Zhao Y., "ESG and Firm’s Default Risk". Finance Research Letters, vol., Article ID 102713, 2022. Luo Y., Xiong G., Mardani A., "Environmental information disclosure and corporate innovation: The "Inverted U-shaped" regulating effect of media attention". Journal of Business Research, vol. 146, pp. 453–463, 2022. 31. Wang S., Lin W., Zhang Z., et al., "Does the environment information announcement promote green innovation? A quasi-natural experimental evidence from the city-level of China". Ecological Indicators, vol. 136, Article ID 108720, 2022. 32. 33. 34. Zhang S., Zhang M., Qiao Y., et al., "Does improvement of environmental information transparency boost firms’ green innovation? Evidence from the air quality monitoring and disclosure program in China". Journal of Cleaner Production, vol. 357, no., pp. 131921, 2022. Zhang C., Zhou B., Tian X., "Political connections and green innovation: The role of a corporate entrepreneur- ship strategy in state-owned enterprises". Journal of Business Research, vol. 146, pp. 375–384, 2022. Liu J., Zhao M., Wang Y., "Impacts of government subsidies and environmental regulations on green process innovation: A nonlinear approach". Technology in Society, vol. 63, Article ID 101417, 2020. 35. Xiang X., Liu C., Yang M., "Who is financing corporate green innovation?". International Review of Eco- nomics & Finance, vol. 78, pp. 321–337, 2022. 36. Ren S., Sun H., Zhang T., "Do environmental subsidies spur environmental innovation? Empirical evi- dence from Chinese listed firms". Technological Forecasting and Social Change, vol. 173, Article ID 121123, 2021. 37. He K., Chen W., Zhang L., "Senior management’s academic experience and corporate green innova- tion". Technological Forecasting and Social Change, vol. 166, Article ID 120664, 2021. 38. Ullah S., Nasim A., "Do firm-level sustainability targets drive environmental innovation? Insights from BRICS Economies". Journal of Environmental Management, vol. 294, Article ID 112754, 2021. https:// doi.org/10.1016/j.jenvman.2021.112754 PMID: 34265739 39. Alda M., "The environmental, social, and governance (ESG) dimension of firms in which social responsi- ble investment (SRI) and conventional pension funds invest: The mainstream SRI and the ESG inclu- sion". Journal of Cleaner Production, vol. 298, Article ID 126812, 2021. PLOS ONE | https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 22 / 24 PLOS ONE Do Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies 40. Debbarma J., Choi Y., "A taxonomy of green governance: A qualitative and quantitative analysis towards sustainable development". Sustainable Cities and Society, vol. 79, Article ID 103693, 2022. 41. Kraus S., Rehman S. U., Garcı´a F. J. S., "Corporate social responsibility and environmental perfor- mance: The mediating role of environmental strategy and green innovation". Technological Forecasting and Social Change, vol. 160, Article ID 120262, 2020. 42. Khan S. A. R., Godil D. I., Jabbour C. J. C., et al., "Green data analytics, blockchain technology for sus- tainable development, and sustainable supply chain practices: evidence from small and medium enter- prises". Annals of Operations Research, pp. 1–25, 2021. 43. Tarmuji I., Maelah R., Tarmuji N., et al., "The Impact of Environmental, Social and Governance Prac- tices (ESG) on Economic Performance: Evidence from ESG Score ". International Conference on Eco- nomics & Finance Research; Vol. 37, no. 3, pp. 67–74,2016. 44. Nekhili M., Boukadhaba A., Nagati H., et al., "ESG performance and market value: the moderating role of employee board representation". International Journal of Human Resource Management, vol. 32, no. 14, pp. 3061–3087, 2021. 45. Atan R., Alam M. M., Said J., et al., "The impacts of environmental, social, and governance factors on firm performance". Management of Environmental Quality: An International Journal, vol. 29, no. 2, pp. 182–194, 2018. 46. Wong W. C., Batten J. A., Ahmad A. H., et al., "Does ESG certification add firm value?". Finance Research Letters, vol. 39, Article ID 101593, 2021. 47. Patel P. C., Pearce J. A., Oghazi P., "Not so myopic: Investors lowering short-term growth expectations under high industry ESG-sales-related dynamism and predictability". Journal of Business Research, vol. 128, pp. 551–563, 2021. 48. Alkaraan F., Albitar K., Hussainey K., et al., "Corporate transformation toward Industry 4.0 and financial performance: The influence of environmental, social, and governance (ESG)". Technological Forecast- ing and Social Change, vol. 175, Article ID 121423, 2022. 49. Becker M. G., Martin F., Walter A., "The power of ESG transparency: The effect of the new SFDR sus- tainability labels on mutual funds and individual investors". Finance Research Letters, Article ID 102708, 2022. 50. 51. Luo D., "ESG, liquidity, and stock returns". Journal of International Financial Markets, Institutions and Money, vol. 78, Article ID 101526, 2022. Lo¨ o¨ f H., Sahamkhadam M., Stephan A., "Is Corporate Social Responsibility investing a free lunch? The relationship between ESG, tail risk, and upside potential of stocks before and during the COVID-19 cri- sis". Finance Research Letters, vol. 46, Article ID 102499, 2022. 52. Giese G., Lee L., Melas D., et al., "Foundations of ESG Investing: How ESG Affects Equity Valuation, Risk, and Performance". Journal of Portfolio Management, vol. 45, no. 05, pp. 69–83, 2019. 53. Broadstock D. C., Matousek R., Meyer M., et al., "Does corporate social responsibility impact firms’ innovation capacity? The indirect link between environmental & social governance implementation and innovation performance". Journal of Business Research, vol. 119, pp. 99–110, 2020. 54. Fatemi A., Glaum M., Kaiser S., "ESG performance and firm value: The moderating role of disclosure". Global Finance Journal, vol. 38, pp. 45–64, 2018. 55. Auer B.R., Schuhmacher F., Do socially (ir)responsible investments pay? New evidence from interna- tional ESG data, (2016) Quarterly Review of Economics and Finance, 59, pp. 51–62 56. Mbanyele W., Huang H., Li Y., et al., "Corporate social responsibility and green innovation: Evidence from mandatory CSR disclosure laws". Economics Letters, vol. 212, Article ID 110322, 2022. 57. ESG and corporate financial performance: the mediating role of green innovation: UK common law ver- sus Germany civil law, Chouaibi S., Chouaibi J., Rossi M. EuroMed Journal of Business, 2022, 17 (1), pp. 46–71 58. Tan Y., Zhu Z., "The effect of ESG rating events on corporate green innovation in China: The mediating role of financial constraints and managers’ environmental awareness". Technology in Society, vol. 68, Article ID 101906, 2022. 59. Yoon B., Chung Y., "The effects of corporate social responsibility on firm performance: A stakeholder approach". Journal of Hospitality and Tourism Management, vol. 37, pp. 89–96, 2018. 60. Albort-Morant G., Leal-Milla´n A., Cepeda-Carrio´n G., "The antecedents of green innovation perfor- mance: A model of learning and capabilities". Journal of Business Research, vol. 69, no. 11, pp. 4912– 4917, 2016. 61. Shen H., Ma Z., "Local Economic Development Pressure, Firm Environmental Performance and Debt Financing". Journal of Financial Research, vol., no. 2, pp. 153–166, 2014. PLOS ONE | https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 23 / 24 PLOS ONE Do Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies 62. Adomako S, Tran MD. Environmental collaboration, responsible innovation, and firm performance: The moderating role of stakeholder pressure[J]. Business Strategy and the Environment, 2022, 31(4): 1695–1704. 63. Yuan B, Cao X. Do corporate social responsibility practices contribute to green innovation? The mediat- ing role of green dynamic capability[J]. Technology in Society, 2022, 68: 101868. 64. Yan J., "Corporate Social Responsibility Disclosure and Innovation Performance: An Empirical Study Based on Chinese Listed Firms during "the Post- mandatory Period"". Science of Science and Manage- ment of S.& T., vol. 42, no. 1, pp. 57–75, 2021. 65. Cowan K, Guzman F. How CSR reputation, sustainability signals, and country-of-origin sustainability reputation contribute to corporate brand performance: An exploratory study[J]. Journal of business research, 2020, 117: 683–693. 66. Cahan SF, Chen C, Chen L, et al. Corporate social responsibility and media coverage[J]. Journal of Banking & Finance, 2015, 59: 409–422. 67. Tian X, Wang TY. Tolerance for failure and corporate innovation[J]. The Review of Financial Studies, 2014, 27(1): 211–255. 68. He JJ, Tian X. The dark side of analyst coverage: The case of innovation[J]. Journal of Financial Eco- nomics, 2013, 109(3): 856–878. 69. Bardos KS, Ertugrul M, Gao LS. Corporate social responsibility, product market perception, and firm value[J]. Journal of Corporate Finance, 2020, 62: 101588. 70. Zhu D., Zhou X., "Equity Restriction, Managerial Ownership and Enterprise Innovation Efficiency". Nan- kai Business Review, vol. 19, no. 3, pp. 136–144, 2016. 71. Ross S. A., "The Determination of Financial Structure: The Incentive-Signaling Approach". Bell Journal of Economics, vol. 8, no. 1, pp. 23–40, 1977. 72. Cornell B., "ESG Preferences, Risk and Return". European Financial Management, vol. 27, no. 1, pp. 12–19, 2021. 73. Zhai Y, Cai Z, Lin H, et al. Does better environmental, social, and governance induce better corporate green innovation: The mediating role of financing constraints[J]. Corporate Social Responsibility and Environmental Management, 2022, 29(5): 1513–1526. 74. Buallay A., Hamdan R., Barone E., & Hamdan A. (2022). Increasing female participation on boards: Effects on sustainability reporting. International Journal of Finance and Economics, 27(1), 111–124. 75. Song W. H., & Yu H. Y. (2018). Green innovation strategy and green innovation: The roles of green cre- ativity and green organizational identity. Corporate Social Responsibility and Environmental Manage- ment, 25(2), 135–150. 76. Limkriangkrai M., Koh S., Durand R. B., "Environmental, Social, and Governance (ESG) Profiles, Stock Returns, and Financial Policy: Australian Evidence". International Review of Finance, vol. 17, no. 3, pp. 461–471, 2017. 77. Cheng B. T., Ioannou I., Serafeim G., "Corporate social responsibility and access to finance". Strategic management journal, vol. 35, no. 1, pp. 1–23, 2014. 78. 79. The effect of corporate social responsibility and the executive compensation on implicit cost of equity: Evidence from French ESG data, Chouaibi, Y., Rossi, M., Zouari, G., Sustainability (Switzerland), 2021, 13(20), 11510 Tolliver C, Fujii H, Keeley AR, et al. Green innovation and finance in Asia[J]. Asian Economic Policy Review, 2021, 16(1): 67–87. 80. Bi K., Huang P., Wang X. Innovation performance and influencing factors of low-carbon technological innovation under the global value chain: A case of Chinese manufacturing industry, (2016) Technologi- cal Forecasting and Social Change, 111, pp. 275–284. 81. Chouaibi S, Rossi M, Siggia D, et al. Exploring the moderating role of social and ethical practices in the relationship between environmental disclosure and financial performance: evidence from ESG compa- nies[J]. Sustainability, 2022, 14(1): 209. 82. Gomariz M. F. C., Ballesta J. P. S., "Financial reporting quality, debt maturity and investment efficiency". Journal of Banking & Finance, vol. 40, pp. 494–506, 2014. 83. Yuan B., Cao X., "Do corporate social responsibility practices contribute to green innovation? The medi- ating role of green dynamic capability". Technology in Society, vol. 68, Article ID 101868, 2022. 84. Baron R. M., Kenny D. A., "The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations". Journal of Personality and Social Psy- chology, vol. 51, pp. 1173–1182, 1986. https://doi.org/10.1037//0022-3514.51.6.1173 PMID: 3806354 85. Wu D., Zhao Q., Han J., "Corporate Social Responsibility and Technological Innovation: Evidence from China". Nankai Economic Studies, no. 3, pp. 140–160, 2020. PLOS ONE | https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 24 / 24 PLOS ONE
Could not heal snippet
10.1016_j.jrurstud.2023.01.003.pdf
Data availability Data will be made available on request.
Data availability Data will be made available on request.
Contents lists available at ScienceDirect Journal of Rural Studies journal homepage: www.elsevier.com/locate/jrurstud Rural co-working: New network spaces and new opportunities for a smart countryside Gary Bosworth a, *, Jason Whalley a, Anita Fuzi b, Ian Merrell c, f, Polly Chapman d, Emma Russell e a Newcastle Business School, Northumbria University, Newcastle Upon Tyne, UK b Cushman & Wakefield, London, UK c Rural Policy Centre, Scotland’s Rural University College, Edinburgh, Scotland d Impact Hub Inverness, Inverness, Scotland, UK e DIGIT Research Centre, University of Sussex, UK f National Innovation Centre for Rural Enterprise, Newcastle University, UK A R T I C L E I N F O A B S T R A C T Keywords: Co-working Rural entrepreneurship Digital economy Network-immiscibility Smart countryside Coworking has been a largely urban phenomenon although new initiatives are emerging in rural areas. Rural coworking is partly a response to the growing need for ICT, which is unevenly provided across rural areas, and partly to the social needs of freelancers and home-workers. By combining technological and social functions, coworking spaces can play key roles in the progress of a Smart Countryside, supporting digital, knowledge-based and creative entrepreneurs within rural places, thus reducing the need for extensive commuting and out- migration, particularly among younger and higher-skilled workers. As working practices evolve in the aftermath of Covid-19, these new physical spaces are expected to facilitate new network connections. Castells’ Network Society provides a valuable lens through which to investigate how coworking founders and managers promote a mix of internal and external networks that might create new, and superior, entrepreneurial opportunities. The research highlights strategies to promote collaboration as well as methods of adapting to meet new demands from rural workers in a range of rural settings. As an array of different rural coworking models evolve, we also reflect on the importance of inclusivity and identity in determining their relationship with other actors in the local economy. 1. Introduction The digitalisation of information and communications in the Global Network Society has facilitated working beyond traditional offices, so long as individuals have the requisite network connectivity (Castells, 2004) and the skills required for digital and remote working (Helsper and van Deursen, 2017; OECD, 2019). Remote working offers the po- tential to create a so-called “cyber-utopia” without traffic jams or urban overcrowding (Malecki and Moriset, 2008, p150), but this vision was only unexpectantly realised as a consequence of the lockdown measures adopted during the Covid-19 global pandemic, which were anything but utopian. Despite the earlier, relatively slow development of coworking, particularly in more rural settings, many commentators suggest that elements of these new ways of working will perpetuate in varying forms in a post-Covid economy (Clark, 2020; Kitagawa et al., 2021; Marcus, 2022; Tomaz et al., 2021; Reuschke et al., 2021). In this article, we define coworking spaces as, “flexible, shared, rentable and community-oriented workspaces occupied by professionals from diverse sectors” that are “designed to encourage collaboration, creativity, idea sharing, networking, socializing, and generating new business opportunities for small firms, start-ups and freelancers” (Füzi, 2015, p462). Coworking offers the potential to reverse or slow down the relentless expansion of commuting and other business travel (Fior- entino, 2019; Ohnmacht et al., 2020), which can have major impacts for the environment as well as the economic and social geography of both cities and rural regions. Uncertainty about the future intensity of city-centre office working in the wake of Covid-19 (Glaeser, 2021; Florida et al., 2020; Marcus, 2022; Nathan and Overman, 2020) along with increased investment in rural digital connectivity to address the long-standing “digital divide” (Salemink et al., 2017) and increasing * Corresponding author. E-mail addresses: gary.bosworth@northumbria.ac.uk (G. Bosworth), jason.whalley@northumbria.ac.uk (J. Whalley), anita.fuezi@gmail.com (A. Fuzi), ian. merrell@sruc.ac.uk (I. Merrell), polly.champman@impacthub.net (P. Chapman), emma.russell@sussex.ac.uk (E. Russell). https://doi.org/10.1016/j.jrurstud.2023.01.003 Received 26 February 2022; Received in revised form 23 December 2022; Accepted 9 January 2023 JournalofRuralStudies97(2023)550–559Availableonline13January20230743-0167/©2023TheAuthors.PublishedbyElsevierLtd.ThisisanopenaccessarticleundertheCCBYlicense(http://creativecommons.org/licenses/by/4.0/). G. Bosworth et al. demand for rural living (Property Wire, 2020) make this a critical time to investigate the new entrepreneurial dynamics that might be activated and sustained by rural coworking spaces. We apply the lens of the Network Society (Castells, 2004), which emphasises both social and technological processes, to assess the role of coworking in so called “smart rural futures” that are themselves dependent upon knowledge and innovation supported by advances in communications technology (Naldi et al., 2015). Applying this lens, our analysis focuses on two objectives: Firstly, to examine the new networks that are emerging within rural coworking spaces and the strategies of coworking operators that nurture collaborative communities; and sec- ondly, to examine linkages that are developing between coworking spaces and their wider rural and regional economies. As rural develop- ment is influenced by both internal and external drivers of growth, requiring a similar mix of network connections (Ray, 2006; Bock, 2016), we are fundamentally concerned with the roles that rural coworking spaces can play in integrating local and extra-local economies. Our research examines whether coworking spaces build new con- nections within their local communities and economies (i.e., are highly embedded) to boost the local entrepreneurial ecosystem (Mason and Brown, 2014), or whether they exist more as urban exclaves serving the needs of urban-centric businesses and remote working practices among urban employees. Just as Castells observed the potential for unequal access to networks and resources in his Network Society, a study of a London coworking venue, identified that the value of openness could “constitute new geographies of exclusion, enclosure and exploitation” (Lorne, 2019, p761). The diversity that is championed as a driver of innovation reifies the entrepreneurial personality who is comfortable is that space, but potentially alienates other kinds of diversity. This dilemma helps us to frame the two objectives of this paper around the internal and external dynamics of coworking. In line with these objectives, we developed a qualitative approach to engage with a range of coworking operators located in, and/or serving, rural areas. After an initial review of the literature on the emergence of coworking and the theoretical foundations of the Network Society and Smart Rural Development, we present the full methodology and then report on findings from interviews and focus groups. We finish by of- fering conclusions and recommendations. 1.1. Rural coworking: the story pre-covid Telework centres (Oestmann and Dymond, 2001) or telecottages (Paavonen, 1999) developed through the 1990–2000s with early ver- sions recognising the need of homeworkers to create physical and mental separation between home and work, to access superior tech- nology and to replicate the “buzz” of a traditional office setting (Malecki and Moriset, 2008). Many early examples struggled to transition from public funding into sustainable business models (Mokhtarian and Bag- ley, 2000) but, moving into the 2010s, the number of coworking spaces grew globally (Clifton et al., 2019). Although the sector has evolved more slowly in rural areas, the impact of the Covid-19 pandemic has drawn attention to more peripheral and rural working environments (Akhavan et al., 2021). Coworking spaces take a number of forms and operate with different ownership and management structures (Fiorentino, 2019). Private en- terprises can be single facilities or global companies operating a network of venues. There are also a wide range of publicly-run and community-led initiatives, filling these gaps left by private enterprise or creating alternative spaces tailored to niche user-demands. Focusing on rural regions, venues vary from informal community spaces, often retro-fitted to take up otherwise redundant space, through to dedicated spaces co-located with enterprise hubs or business incubators offering users the option to rent fixed workspace as well as hot-desks (Merrell et al., 2022). The spread of coworking spaces into more rural areas has been enabled by rapid advances in digital technologies and increased coverage of Wi-Fi enabled broadband (Houghton et al., 2018; Nambisan et al., 2019). The range of jobs that can be carried out beyond the traditional workplace is also increasing, so long as the requisite con- nectivity is available (Kane and Clark, 2019). In particular, the indi- vidualisation of work, combined with low-cost software and an explosion of cloud-based and mobile app-based digital services allow co-workers to operate relatively independently (Vallas and Schor, 2020). Sole-traders can streamline a range of administration activities, customer services and accounts (Atherton, 2016; Jordan, 2021), changing the traditional professional service function for both service user and service provider and creating new spaces for innovation. Dig- ital technologies are also accelerating the inception, scaling and evolu- tion of new ventures and leading to some radical re-thinking of creative endeavours that span traditional industry/sectoral boundaries (Nambi- san et al., 2019). Coworking was traditionally most attractive to smaller start-up businesses, creative industries, freelancers and solo consultancies (Füzi, 2015), with only a few examples identifying their appeal to homeworkers employed by larger institutions, including the public sector (Houghton et al., 2018). The essential values of coworking include work-life balance, reduced commuting and new network op- portunities, whether for collaboration and knowledge-sharing or to help homeworkers to overcome isolation (Spinuzzi 2012; Füzi, 2015) and create important markers between work and home life (Russell and Grant, 2020; Merrell et al., 2022). While pre-pandemic research has shown that homeworking can enhance the well-being of many groups of workers, especially employees, isolation of self-employed workers was found to have impacts on the perceived financial situation of the household in addition to feeling of loneliness (Reuschke, 2019). The social value of coworking spaces extends to the provision of a stronger collective voice to their members in local development policy circles with the ability to lobby for better business support and infrastructure improvements (Kolehmainen et al., 2016). Whether just small-talk and companionship or more business focused benefits of knowledge exchange and collaboration, the social functions of coworking spaces have been linked to better time management, personal and psychological health benefits and serendipitous moments that trigger learning and innovations (Kov´acs and Zolt´an 2017). In rural settings, this can extend to community well-being impacts too, partic- ularly as coworking spaces have the potential to engage different com- munity groups as well as businesses (Stojmenova Duh and Kos, 2016). Where coworking spaces develop to become embedded as part of the relational assets (Storper, 1997) of a local innovative milieu (Camagni, 1995) or entrepreneurial ecosystem (MasonandBrown, 2014), their in- fluence can transcend the value to members by enhancing the image of a place, providing a hub of activity to sustain other nearby enterprise and providing support to a range of community initiatives (Hill, 2022). This embedding role of coworking spaces fits with narratives of the influence of social and community factors on rural entrepreneurship practices (Korsgaard et al., 2015; Bosworth and Turner, 2018). The benefits of interacting and collaborating with people from different professions is frequently cited (Houghton et al., 2018; ˇ Sebestov´a et al., 2017), but research suggests that co-location alone is not sufficient to generate cross-fertilization and innovation outcomes (Füzi, 2015; Johns and Hall, 2020). Successful collaboration is depen- dent on internal facilitators and the wider entrepreneurial environments in which they are located (Kov´acs and Zolt´an, 2017; Clifton et al., 2019). In particular, more facilitated models of coworking with skilled hosts/managers were found to be important to support younger entre- preneurs and start-ups, mirroring some of the more established learning from business incubators (Füzi, 2015). This highlights the need to better understand the nature of new network configurations that will form within and beyond coworking spaces and the outcomes that may follow. Predictions that rural coworking will advance through a combination of tailored policies coupled with bottom-up initiatives (Akhavan et al., 2021) lead us to examine these complex relationships through the lenses JournalofRuralStudies97(2023)550–559551 G. Bosworth et al. of the Network Society and “smart” rural development. 2. Smart rural development and the Network Society The likely impact of new connectivity and mobility technologies mean that smart rural futures need to be framed differently from smart cities (Cowie et al., 2020), and need to take account of different rural and remote working patterns and coworking spaces. From a sustain- ability perspective, new technologies within coworking hubs can reduce commuting and carbon footprints and shorten supply chains, offering the potential to revitalise rural economies (Zavratnik et al., 2019) and helping to address the smart vs sustainable growth conundrum (Naldi et al., 2015). To be effective, these technological developments depend on social factors too, which are central to understanding the Network Society. The Network Society is defined as: “The social structure that results from the interaction between social organisation, social change, and a tech- nological paradigm constituted around digital information and communica- tion technologies” (Castells, 2004, xvii). Although most references to Castells’ work focus on the global reach of digital networks and examine his “space of flows” concept (Simonsen, 2004; Zhen et al., 2020), Cas- tells himself recognises the importance of different cultures, power and localised networks being integral to understanding and shaping the Network Society. While the Network Society connects many cultures on one level, people’s local experiences can be “fragmented, customized [and] individualized” (Castells, 2004, p30). The Network Society allows people to participate in multiple net- worked spaces of communication centred around mass media and the Internet, and not necessarily embedded in the local community. This spatial-social dichotomy is not unique to the online world, as shown by research into rural migration and commuting patterns (Champion et al., 2009; Bosworth and Venhorst, 2018), but the proliferation of digital communications exacerbates fragmentation. The irony of framing coworking spaces, which are themselves dependent on digital technol- ogy, as the antidote for rural society to reconnect around “place” is not lost on us, but we see their emergence as a key component of smart rural development (Naldi et al., 2015; Slee, 2019). Just as smart growth is founded on knowledge and innovation supported by advances in com- munications technology (Naldi et al., 2015), the Network Society also views economic growth as being dependent on global flows of infor- mation structured around socio-technological networks (Castells, 2004). Castells makes no particular reference to rural areas, suggesting that rural spaces sit rather low in the hierarchy of network nodes (Murdoch, 2000) and at the periphery of knowledge-based networks (Benneworth and Charles, 2005). However, a more positive outlook is that mecha- nisms to enhance access to these global flows of information could break down old spatial divisions such as the urban-rural divide (Murdoch, 2000). Coworking is one such mechanism, which brings the added advantage that it can help to address the digital divide (Salemink et al., 2017) by providing greater access to new technologies and supporting the digital skills and social networks needed to promote local entre- preneurship and innovation (Gerli and Whalley, 2022). This reinforces the importance of places as mediators of technological change (Cowie et al., 2020) as well as the environments in which meaningful cultural and social existence occurs (Fisker et al., 2021). The global nature of the Network Society demands cultural distinc- tiveness as the cornerstone of communication and knowledge exchange. Castells argues that “cultural identities become the trenches of auton- omy” (2004, p39) offering the potential for “complementarity and reciprocal learning” (2004, p42) between cultures. This requires local actors to have sufficient agency to balance top-down and bottom-up processes and develop a strong voice in dialogues with external orga- nisations. In the language of the Network Society, actors need the means to communicate and understand different cultures with the necessary openness to allow the permeation of new ideas across diverse networks. To advance “smart” forms of place-based development, local actors need to draw upon the value and distinctiveness of local resources, knowledge and traditions when engaging in wider networks (Naldi et al., 2015; OECD, 2018). Castells refers to cultures having their relevance as “nodes of a net- worked system of cultural dialogue” (2004, p42) and Murdoch describes “a constellation of networks that can be found in the contemporary countryside” (2006, p172). While this shows that rural areas have an important place in a global Network Society, we need to understand more about the different types of networks, their resources, their inter-connections and their reach. Where rural nodes become discon- nected from dominant, resource-rich networks, their value is diminished and individuals become excluded (Hacker et al., 2009). Exclusion from networks relegates actors to the space of place alone, bypassed by the network flows that are essential facilitators of social mobility as well as entrepreneurship (Baker et al., 2017). Therefore, the spaces and pro- cesses that create and sustain networks within rural spaces are critical to explaining entrepreneurial and innovative potential. Returning to Cas- tells, “We must place at the centre of the analysis the networking capacity of institutions, organisations, and social actors, both locally and globally. Connectivity and access to networks become essential” (2004, p42). Local social and economic dynamics see rural entrepreneurs draw on a range of resources to create distinctive business opportunities that satisfy both economic and lifestyle goals (Korsgaard et al., 2015). Too much emphasis on high growth, high-tech and innovative entrepre- neurship within the entrepreneurial ecosystem literature constrains our understanding of entrepreneurial enablers and dynamics in rural con- texts (Mu˜noz and Kimmitt, 2019). Instead, capitalising on the value of multiple, heterogenous rural assets requires networks through which their distinctive values can be communicated effectively, thus strengthening the identity of network nodes themselves. As Horlings et al. observe, “The nature of a place is not just a matter of its internal (perceived) features, but a product of its connectivity with other places. Places are nodes in networks, integrating the global and the local” (2020. P.356). The value of networks depends upon the utility of their nodes and the wider access that they provide (Anttiroiko, 2016; Varnelis, 2008). The sparser networks of firms in rural areas may diminish some network advantages, such as access to information, business support or training, but they still motivate innovation and entrepreneurship (Copus and Skuras 2006) and provide conduits through which firms can develop and communicate their distinctive values and capabilities (Malecki 1997). Indeed, the greater propensity for self-employment (Phillipson et al., 2019) and greater overlap of social and economic imperatives among many rural businesses (Steiner and Atterton, 2014) may see rural net- works becoming more start-up oriented and mutually supportive, drawing on a collective identity outside of urban networks. Within this space, new combinations of local and extra-local knowledge and re- lationships can spark new entrepreneurial ideas and opportunities. In rural regions experiencing increased rates of counterurbanisation and return migration, these trends add further to the network diversity (Kalantaridis and Bika, 2011; Mitchell and Madden, 2014). Until now, the economic potential of rural areas has been limited by slower and inferior provision of communications infrastructure compared to urban areas (Grubesic and Mack, 2017). The disadvantages that this created for rural areas are, however, narrowing through the collective impact of policy initiatives, government investment and entrepreneurial activity (Gerli et al., 2020; Sadowski, 2017). As a result, new opportunities are emerging for entrepreneurs to combine distinc- tive features of rurality with the benefits of digital technologies – reaching new markets, interacting more with customers and developing new products and services as well as new working practices and business models that reflect distinctive values attributed to rural places (Hill, 2022; Bosworth and Turner, 2018). Rural coworking spaces form part of this evolution, challenging conventional institutional and organisational cultures and affording greater importance to individuals’ networks in their communities of JournalofRuralStudies97(2023)550–559552 G. Bosworth et al. place (Mazur and Duchlinski, 2020). Recognising that rural coworking is opening up to employees as well as freelancers, the idea that one shares information with one’s coworking neighbour, in another firm or another industry, before sharing it with one’s work colleague may be unsettling for managers but transformative for innovation. With Covid-19 stimu- lating a rapid increase in remote working, the “buzz” of urban locations may be compromised, and the value of rural environments and their community connections are accentuated. The weakening gravitational pull of clusters, especially in the tech- nology sector (Feldman et al., 2020), challenges conventional regional economic theories and represents a major U-turn for firms who have spent years investing in attractive, comfortable and collaborative workplace environments (Dahl and Sorensen, 2020). Echoing calls from Gruber and Soci (2010) a decade ago, such transformation calls for greater attention to be afforded to the local dynamics of peripheral re- gions, not just to dominant (traditionally urban-centric) network nodes. While cities will recover, their functions may change and the new-found acceptance of nomadic forms of working will see different features of local environments attracting workers with the flexibility to work remotely. Just as Castells observed, though, this will have implications for those who are less able to engage in this new labour market and whose jobs require a physical presence in fixed premises (Florida et al., 2020; Marcus, 2022). Reframing the Network Society to consider the uniqueness of rural economies identifies that networks are not just spaces of flows but they are fundamental to shaping and narrating rural places. However, the configuration of networks within a spatially defined node and the extent to which actors are embedded in more locally or externally-oriented networks are essential to understanding the implications for rural pla- ces. For example, more innovative services have been associated with the need for stronger external networks connecting into nodes higher up the urban hierarchy (Shearmur and Doloreux, 2015) yet other creative businesses thrive as a result of their rural locations (Townsend et al., 2017). The new spaces of rural coworking hubs and the increased va- riety of remote-working practices prompted by the Covid-19 pandemic, provide the context for rethinking the meaning and influence of rural places becoming more vibrant and active nodes within the Network Society. The co-location of employees and entrepreneurs across a range of sectors forms part of the entrepreneurial potential of rural coworking, supporting an emerging literature on sector fluidity that views industries sectors being less fixed or bounded (De Massis et al., 2018) and collaborating in a quadruple helix relationship (Kolehmainen et al., 2016). Rather than a sector-focused set of relationships, rural coworking provides a greater emphasis on the social and cultural environment, inspiration and opportunities from where entrepreneurs derive (Anderson et al., 2010; Honig and Samuelsson, 2021). At this hyper-local scale, coworking spaces foster individual relationships and knowledge exchange that erode boundaries between firms and sectors. This is not technology breaking down barriers in the traditional lan- guage of the Network Society but a hybrid space where re-localisation presents a new nexus of opportunities and enterprising actors (Shane and Venkataraman, 2000) combined with networks connecting to external enablers (Davidsson, 2015). To better understand these emerging entrepreneurial spaces, both the internal and external dynamics of rural coworking spaces are investigated. Recognising that digitization is offering the tools to sup- port collective approaches to the pursuit of entrepreneurship (Nambi- san, 2017), and combining this with analysis of the network structures that surround rural coworking spaces, the methodology reflects contemporary understanding of a smart countryside. 3. Methodology Since the research took place during the Covid-19 pandemic, all data collection was conducted online. This included a series of 17 semi- structured video interviews with coworking operators/developers, supplemented by two policy-maker focus groups, an interview with the managing director of the Flexible Workspace Association and a larger online workshop. In total, the research engaged with around 80 discrete participants between September 2020 and June 2021. Additional data was collected from analysis of website content to explore the marketing messages used to describe the advantages of coworking, their key fea- tures and the rationales behind their establishment. This captured the perspectives of operators as well as the representation of rural cow- orking that they seek to communicate externally – mirroring the twin objectives of understanding both internal and external dynamics of rural coworking. The inability to access users of coworking spaces was a limitation of the research project, something which is planned to be addressed in future research. However, the framing of this paper means that the founders and managers are best placed to explain their strategies and give an informed overview of the evolving nature of rural coworking based on their experiences. They were asked to comment on the reasons that their members and customers gave for using their venues as well as explaining their marketing strategies, business models, workspace and technology provision, and the ways that they adapted to stay in contact with their members through the various periods of Covid-19 lockdown. The video interviews were audio-recorded and participants gave their consent to transcribe the conversations. The online workshop was staged on the Collab online conferencing platform (https://collabvirtual world.com) and attracted 60 delegates, mainly coworking operators along with a small number of researchers and policy-makers. This began with a presentation of emerging findings after which participants were asked to join one of a selection of “virtual tables” where members of the research team led structured break-out discussions as one might do in a global caf´e style event. Focus group participants were recruited through an email to members of the Rural Services Network, a membership organisation for rural Local Authorities and associated rural develop- ment stakeholders. Each focus group was conducted on Microsoft Teams with three members of the research team joined by 11 participants split across two sessions. Thematic analysis of the interview transcripts, focus groups and workshop notes focused on key themes of coworking practices, intra- group networks, wider connections within and beyond the rural econ- omy, the impacts of Covid-19 and the role of technology. For this paper, we focused principally on the interview data and analyse the transcripts to draw out references to “internal collaboration and networks” and “external networks and spillover effects”. Quotations were collected that picked up both positive and negative features relating to each broad theme and then arranged according to secondary themes of social or economic factors, formal or informal networks and the degree to which place was important in shaping the activities or networks being analysed. 4. Findings The sample of coworking spaces identified a wide range of organi- sations with different business models, premises, clientele and future aspirations. These ranged from social enterprises focusing on the needs of small local communities through to wholly for-profit ventures with growth plans across multiple settlements. We also spoke to operators of coworking retreats that were more targeted towards digital nomads at the national and even international scale as well as some in larger towns and cities who served a heavily rural region and others in much smaller and more remote locations. A summary of the 16 interviewees is pro- vided in Table 1. Although it is possible to identify a number of different coworking models across the operators we interviewed (Author et al., 2022), this section focuses on common elements of coworking that nurture sup- portive networks and community identities internally, while building extensive connections that help to develop their external profiles. Before JournalofRuralStudies97(2023)550–559553 G. Bosworth et al. Table 1 Interview sample characteristics. Interviewee (pseudonym) Location type Type of Organisation Annie Ben Connie David Ernie Freddy Graham Harriet Ian Julia Kenny Louise Martin Neil Olive Peter Rachel Small city Open countryside Village 2 Market towns Market town Village Market town Village Small city Island town Market town Village Market town Market town Market town 2 village locations Village Non-profit Private limited company Private limited company Private limited company Private limited company and social enterprise Family business Local Authority Family Business Community Interest Company Part of a private limited company Private company Private company Private company Private limited company Private company Private company Opened/ registered 2020 2016 2019 2017 2012 2020 2009/10 2021 2017 2018 2016 2017 2020 2021 2017 2013 Informal group 2020 exploring these networks, it is important to contextualise the research in relation to the importance of the rural location as portrayed among coworking operators. The interviewees identified both nature-based and community-based values for co-workers, for whom connections with the environment has been shown to benefit their wider well-being too (Merrell et al., 2022). Whether moving into rural areas or already embedded in the locality, many operators were very passionate about the location as highlighted in the selected quotations below: “We set it up in the countryside because we had identified … that people actually wanted to not just go [to the countryside] for the weekend or for a holiday but actually spend a longer amount of time, and if they could they’d like to work on their projects outside of the city. So we developed it as a way to help people escape the city” (Louise) “You don’t just get a nice desk. You get an AONB landscape out your window and wetlands habitat and opportunity to plant trees or whatever it might be. I think being out in the countryside around green space can help with productivity [and] creative thinking” (Neil) “One of the advantages that we really have here is that we’re on the coast and that in your lunch hour you can walk down to the beach and have your picnic lunch there” (Harriet) And operators were well aware of the marketing potential that rural locations offered too: “We definitely play on our rustic feel, like we can’t offer sleek city centre kind of facilities. This is very much a country house with views of the [mountains] and I guess it’s the location that sells it but the house itself is rustic … so to be honest it kind of suits my style.” (Connie) Emphasising the distinctiveness of the location as a strong base from which to communicate with the wider world is a good example of how the Network Society can empower rural places to take advantage of their distinctive characteristics. While urban coworking spaces may be rela- tively homogenous, focusing on hi-spec and hi-tech office space that is familiar to mobile workers wherever they happen to be, rural spaces have the scope to position themselves differently. First impressions from our research sample indicated that creating the “buzz” of urban loca- tions requires alternative approaches to community-building as well as efforts to raise awareness about coworking. These differences give rise to a number of questions to explore, in terms of how these distinctive identities are formed and the extent to which they are inclusive and representative of their wider communities. 4.1. Internal networking The literature on networking among rural firms and co-workers in- dicates that simply being close together does not guarantee collabora- tion, but it provides a foundation for new connections to emerge. Therefore, in addition to functional responsibilities, a key role for coworking operators is to promote an entrepreneurial and supportive culture within their organisation. As David observed “We always find that people think they need a desk and Wi-Fi and when people are in what keeps them in is the community.” The consensus among interviewees was that collaboration cannot be forced upon people, only facilitated, but it was very rewarding for founders when this worked: “One of the nicest parts of running a coworking space is seeing those connections being made and facilitating it, or it happening auto- matically. It’s very enjoyable. I love that. I love when people interact and they find each other and it works out and it’s very positive”. (Ben) The value of softer networks was illustrated by interviewees referring to “socializing” more than business networking. Examples included the value of being able to share the success of winning a new contract (online workshop conversation), sharing the frustration of IT problems (Rachel) or simply the need for companionship: “[One member], he comes just for company really. But he needs complete silence to work so he has his own office then comes down for coffee and lunch to meet everyone. We have a couple of people that just like to come in and know that there’s people to speak to if they need to, but they just find their own space. And then the rest of us come in and chat and then we work and then we chat a little bit more and then we work again.” (Connie) This culture was reinforced by another interview with a founder of a high street coworking venue who described one member being “a little bit too pushy” when it came to business networking: “There’s one member … he wants us to have lunches where we talk about what we do and maybe share some presentations, but [among the wider group] it’s quite overwhelmingly an interest in socialising and not talking about your business … and that actually becomes a little bit of a thing because he’s not interested in socialising, he wants to talk business and nobody else wants to.” (Annie) Later in the same interview Annie said: “We always kind of look to who’s in our building first when we look for collaborators. And I also think that very much draws people to us”, highlighting that collaborative working for mutual gain is part of their aspiration – but there is a culturally acceptable way to facilitate it. A second example from Scot- land identified similar collaborations that support members to bid for larger contracts: “we’ve formed a consortium … together we are able to bid for contracts. A lot of these contracts come along and you need to have something like £5 million worth of public liability, or some kind of insurance that is vast sums. And none of these individuals will have it whereas we’ve got it” (Ian). Stimulating this type of collaboration was also important for Local Authority focus group participants who are looking at how cow- orking might translate into rural economic growth. Whether providing a supportive social environment or actively facilitating collaborative working, there is no prescription for what makes an entrepreneurial culture. It might be relaxed, professional, focused, sociable or collaborative, each requiring different combinations of events, branding and spaces to support their members. The selection of furniture, the layout of the venue and d´ecor of rooms all contribute to JournalofRuralStudies97(2023)550–559554 G. Bosworth et al. the identity of the coworking group, often reflecting the attitudes of the founders: “Everything is community for us. We use second-hand furniture as much as possible for environmental reasons [and] … so we don’t spend millions of pounds on fitting out space. We’d much rather spend that money on activities that happen within the space.” (David) “It was important for us to have a variety of workspace types … that’s why we had this caf´e type space. That’s where people can be more social. They can have little meetings, little coffee meetings, either with their colleagues or for a break. The library is also more of a shared space, a little bit more casual. But then we have the really dedicated workspaces” (Louise) “We’re professional but we’re not formal” (Harriet) This focus on “community”, as something over and above the fundamental provision of ICT, is a clear example of Castells’ argument that nodes within the Network Society are defined by their internal cultural identity. The functional or tangible elements of the service are largely homogenous so can be accessed anywhere, but social capital and community identity are seen by the coworking founders/managers as being unique. In the case of founders who work in the space, it is often a personal reflection of their own working culture too. Without this, the homogeneity of a single Global Network Society becomes the dominant trope of how new (digital) technologies influence working practices but the response among coworking operators appears to engender a clear desire for diversity. Following this logic, spaces designed to facilitate different types of behaviour and interaction are paramount to the success of coworking spaces and consistently it was the kitchen area that was most discussed. This is where people are “off-duty” and relaxing as themselves, so the tone of the conversation is different and people become more open and more interested in each other since the pressure of the next task, the next phone call or next email is in another room: “[the kitchen] should be the heart of a coworking space because that’s where everyone collaborates and talks, and that should be right in the middle of the building and it should be where everyone goes and you should base everything around that coffee pod. (Ernie) “In [the local region] you meet people in their kitchens so we designed the front of the office to be a kitchen. So we’ve got a new dishwasher, we’ve got the toaster, we’ve got everything else in there. People come in and have their breakfast … That’s where you learn stuff” (Ian) As well as internal network building, common spaces allow for non- members to see the coworking space and for new users or event at- tendees to interact with established members. Breakfast clubs, caf´e’s open to the public and rooms dedicated to community functions all provided opportunities for events to widen the reach of the venue. Where co-workers were able to host external guests, this also helped to build a sense of community ownership among members (David). So long as external events were not disruptive for co-workers, they become a key foundation for external network connections. 4.2. Building external networks Coworking spaces represent new network nodes that can strengthen connections between rural and urban economies. A particular example was cited in Scotland where bringing together sole-traders or very small businesses allowed them to bid for larger projects outside of their lo- cality (Ian). Not only did this help others realise that a geographically peripheral business location was not a barrier to working further afield, but it is also provides a practical demonstration of how internal net- works can be leveraged externally. While the internal dynamics of the coworking “node” are critical for generating the scale of activity and cultural distinctiveness to engage in complimentary and reciprocal learning within the Network Society (Castells, 2004), interviewees were equally aware of their wider responsibilities. These include business support programmes, networking events, boosting trade for other local businesses and engaging in wider outreach activities. A number of comments capture this mentality: “We actively try and do stuff outside of our four walls which is why we’ve recruited, two years ago we recruited an outreach manager. It was her job to go out and run courses for people, so it’s a big part of what we do.” (Ernie) “We have a lot of partnerships with local businesses … I don’t think it’s a nice thing to have a project in the community where you don’t interact with the community” (Louise) “We don’t just want our spaces being another coworking space, we’re really set on a mission to make our spaces the hub of the ecosystem … we work really hard to try to get that set in people’s minds that it becomes a functional hub for the stakeholders” (David) In some cases, building external networks to support rural economic development was part of the founding principle of establishing a cow- orking space too: “The decision to start a rural hub really came from part of our pur- pose which is to improve the connections between rural and urban entrepreneurs, to see some of their learning spread a little bit further than just within the city, [and] to see the rural entrepreneurs benefiting from what’s happening in the vibrant start-up scene, which is often city based” (Olive). The bridging role of coworking spaces encompasses both the urban- rural scale and more local connections beyond the traditional digital or creative freelancer groups of co-workers. One opportunity at the local scale is presented by the anticipated growth of homeworking among salaried employees who are seeking to reduce their commuting fre- quency following the impacts of the Covid-19 pandemic. This potential new source of demand was a foundation of Neil’s business model and a major topic of conversation in the research workshop sessions. From a Local Authority perspective, potential new demand stimulated enthu- siasm to promote coworking as part of a regeneration strategy to raise the profile and appeal of small towns and failing High Streets. Although there were mixed opinions about the role of the public sector as risk- taking founder or arms’ length facilitator, there was optimism that small town coworking could boost the footfall on the High Street and support other town centre businesses. Despite positive ambitions and rhetoric around the wider value of coworking spaces, only one attempted to quantify their contribution: “It’s bringing people here, has a pretty big impact so I estimate that for the local business every year we generate about €1.2 million for accommodation, for food, for transportation, for stuff that people buy here.” (Kenny) More typical, were comments such as: “These people come here, spend money, spend time, accommoda- tion, other services … I think we are a very good addition to the landscape of [our] area” (Martin) Beyond financial benefits, the research identified a variety of con- tributions yielding more social value. A good example is Peter, the founder of a rural coworking and co-living destination, who explained that they involve local retired people in events because “they don’t need the money … they need conversations.” Peter and his business partner have also set up an educational programme where they “teach the skills of digital nomads to people who want to become digital nomads” because “we want to teach people who don’t want to leave their villages to work, but to stay at home.” In a Network Society sense, the growth of digital JournalofRuralStudies97(2023)550–559555 G. Bosworth et al. nomadism is an illustration that the urban-rural connectivity can be a two-way dynamic where people chose to visit rural locations for certain types of work. Thus, the rural coworking venue is not solely a mecha- nism to reduce out-commuting from rural places but also a location that attracts inward commuters that strengthens its role as a node linking (rural and urban) places together. The chance to support young people was echoed by Neil who felt that they struggle to access to the same training and career development opportunities as people in the big cities and recognised coworking as part of a solution that offers “a stepping-stone to seeing new career op- portunities [and] … a real opportunity for rural areas.” The sense that coworking is a point of connection between places reflects the Network Society but it also extends to a psychological connection where rural places can be perceived as being less isolated and offering greater equality in terms of access to skills and skilled employment. Once the purpose and identity of a rural coworking space is under- stood as something distinctive and place-based, the opportunity for a range of community-focused activities emerge – both promoting the space to other potential users and helping to develop a unique identity. For example, another recent start-up explained her social values in relation to future development plans: “There’s another building that I want to refurbish … we were kind of thinking like a gallery or an exhibition space or something for artists or creatives … they could run workshops there because we’ve already got a link with a local artist and she’s keen to set up chil- dren’s activities and then also do a programme for 16-24 year-olds that aren’t engaging that well with school. So that kind of thing … as well as the desks I’d like to be doing some projects that actually help people as well” (Harriet) While Harriet and her family are firmly embedded in the local area, and approach the community function from that perspective, an in- comer in a similarly remote location gave an interesting perspective on the integrative function that coworking can play. “90 per cent, maybe even 95 per cent of the people who use the Business Hub are incomers. I don’t know whether locals just feel like they don’t need it because they’ve got enough contacts and they know enough places where they can find space to work themselves, so it’s the people who don’t have those connections in the commu- nity who are coming to me. And I’m an incomer myself.” (Julia) These examples highlight the potential for coworking spaces to provide the connectivity and access to networks that are essential to the Network Society. The combined social and technological functions also highlight how this application of Network Society thinking is commensurate with “Smart” rural development. As well as highlighting the local/extra-local connections promoted by coworking, the final quotation also opens up a new set of questions about the inclusiveness of rural coworking. In the early phases of development, and with the need to build communities of users, it ap- pears inevitable that some cliques will emerge and not all people will feel able to participate. This is where the variety of rural coworking models can broaden3 accessibility far more than the corporate struc- tures that have predominated in big cities. Introducing a range of social and community activities that welcome different people into coworking venues offers the potential to build new connections among increasingly mobile, but less cohesive, rural populations. The inclusiveness of indi- vidual coworking spaces is a question for future research with co- workers but the variety of local spaces as interconnected and heterog- enous nodes aligns with Castells’ conceptualization of cultural nodes in the Network Society. 5. Discussion: conceiving diverse impacts for rural places The two areas of findings have highlighted that network relation- ships are critical to the development of rural coworking. In each case, facilitation of soft, informal networks is a key role for coworking oper- ators that was supported by a range of strategies from the design of the space, particularly communal spaces like kitchens, the staging of events (including some that were online during the pandemic) and the creation of a collective identity that engages co-workers. As in urban coworking spaces, collaboration and innovation occur through serendipitous meetings of like-minded people, not through formal networking meet- ings or hard-sell approaches. The difference in rural coworking spaces arises when communities of users develop particular identities, often based around place and nourished by the efforts of managers to create distinctive community identities. As a result, rural coworking venues become more heterogenous, shaped by combinations of social, cultural and environmental factors, and represented through the interactions of co-workers in different settings. The local environment, the character- istics of the building itself, the range of non-business activities, the personal characteristics of the owner and their ambitions to grow or diversify the membership all contribute to a particular feel for each venue. This was evident in the marketing messages of coworking web- sites too, where quotations frequently drew on their location to communicate opportunities to interact with nature, to socialise and to enhance well-being: “Pack your swimming trunks, take your to-do list and then nothing like going out to the country” “There is nowhere else can you surf in the morning and be in central London by lunch time. This is a pure manifestation of the perfect work/ life balance we all strive for” “We want the freelancers that ultimately form the creative group at NAME to feel like family” “With its own garden, high ceilings, lots of light, natural finishes and loads of plants, NAME is an energising, enjoyable place to work” “You will gain inspiration while you work, and exchange experi- ences, tips, ideas and contacts” Through the examples here, aspects of creativity, inspiration and collaboration are evident, but all were presented as part of something more holistic in terms of the work/life experience that coworking can provide. To realise this, coworking operators have to provide the right working spaces, complete with both social and technological in- frastructures – the twin pillars of smart rural development in microcosm. Each pillar has implications for the internal and external network structures, and the communications that evolve within and beyond coworking spaces. In other words, the social and technological context of rural coworking shapes the ways in which co-workers engage in the Network Society and influences the balance of local and external factors that shape business opportunities and identities. The economic spillovers, although hard to quantify, appeared to stem from building a community of co-workers with a sense of connection to their locality. Through this, businesses are able to collaborate with another and recognise opportunities to work with other local firms. Business events and training, as well as more community- focused events in some venues, all expanded the social networks around coworking spaces, increasing their external visibility and often building a sense of identity within the group – the local culture that emerges provides a sense of autonomy and empowerment aligned with that of the Network Society. The importance of the collective, can also be explained in game-theory terms since if all members sought to exploit the group for business growth, the working environment would become a deterrent. In reality, the only way to foster collaboration over time is to prioritise and develop the collective well-being of the group. Shifting the locus of networking from corporate to community spaces raises a number of questions about the agency of individuals within social networks (Taselli and Kilduff, 2021); particularly the extent to which they actively build new connections that spark the potential for innovation and new network configurations. Where home-workers and JournalofRuralStudies97(2023)550–559556 G. Bosworth et al. entrepreneurs interact in rural coworking spaces, the locality affords a common frame of reference and shared identity out of which new ideas can emerge. If these ideas are place-dependent, bringing characteristics of a rural location to the fore, the cultural identities that evolve might become new “trenches of autonomy” (Castells, 2004) that can sustain rural social innovation as well as profit-motivated entrepreneurship. In essence, where agency shifts to the local level, yet the actor remains influentially connected into wider networks, this reflects the philosophy of neo-endogenous development too (Ray, 2006). Re-engaging with Network Society theory is especially timely because of the new connections to ‘place’ deriving from the Covid-19 pandemic (Newman, 2020). In some interpretations, the Network So- ciety emphasises networks to the detriment of places (Zhen et al., 2020) where, rather than being in the right place, being in the right network counts (Anttiroiko, 2016). Here, we argue that such a dichotomy be- tween place and networks can be bridged by new remote-working and coworking practices that build and sustain new network connections within rural places while strengthening and extending connections beyond. Furthermore, creating these new nodes offers significant po- tential innovation, opportunity-creating and professional support networks associated with agglomeration (relatively homogenous) while simultaneously strength- ening heterogeneous, place-based identities and social networks that capture distinctive qualities of their rural context. rural communities replicate the for to The growing diversity of rural businesses in the UK context has been linked with professional incomers and rural returnees (Kalantaridis and Bika, 2011; Stockdale, 2015). These mobile professionals (Keeble and Nachum, 2002) and members of the rural creative class (Herslund, 2012) are better equipped to draw on valuable experience and con- nections beyond the constraints of the local rural context (Bosworth and Bat Finke, 2020); a feature aided by advances in communications technology across rural areas. However, not all forms of employment can benefit from digitalisation and the new ways of working that this enables, with a notable divide between knowledge intensive and manual occupations for example (Dingel and Neiman, 2020). Throughout the Covid pandemic, the housing market has seen increased demand for rural living, indicating that remote working practices are likely to increase in popularity. Combined with the continuing spread of online working and education, this likely to result in further decentralisation of skilled work, with migration more aligned to lifestyle choices and natural amenity values associated with the rural creative class (McGranahan and Wojan, 2007) rather than proximity to workplaces. On one hand, this offers opportunities for coworking, as identified by several research participants, but it also reinforces the perception that coworking is exclusively for mobile professionals and skilled workers. In the Network Society, Castells framed this in terms of differences in education and a person’s ability to work in the informa- tion economy, not as class conflict (Ampuja and Koivsito, 2014). This is reinforced by findings from research into homeworking during the Covid-19 pandemic too, where personal and household factors were key factors determining changes in worker productivity (Felstead and Reuschke, 2021; Hackney et al., 2022; Kitagawa et al., 2021). Given that there are multiple factors that influence workers’ productivity and their ability to participate equally in new ways of working, there is a risk that localised professional networks lead to a two-tier rural society with increased social and economic inequalities. Rural coworking is a possible cause and a possible solution to this problem. The research has identified that many coworking spaces pro- vide opportunities for community activities, training and inclusion. This is essential to avoid the perils of “network immiscibility” (Bosworth and Venhorst, 2018) where, just like the chemical properties of oil and water, networks may co-exist in a place but they require catalysts to stimulate new interactions to bridge between different sub-groups. Where coworking spaces adopt an integrating role, they can facilitate the human, social and financial capital in their networks to contribute to local development. By contrast, if they become exclusive professional spaces more integrated into urban economies, they will exacerbate the marginalisation of other sections of rural society less equipped to participate in the Network Society, perhaps lacking (access to) digital, social or professional skills. As rural coworking evolves, the challenge for operators and policymakers will be to ensure that other parts of the rural economy can benefit, even if they are not active in coworking themselves. 6. Conclusions As creative industries and knowledge-intensive business services continue to grow in rural areas (Townsend et al., 2017; Johnston and Huggins, 2016), facilitated by improved digital connectivity (European Commission, 2020; Ofcom, 2020) and the opportunity to work outside of congested, costly city locations, they are likely to shape the next phase of rural coworking development. In a post-Covid economy, there is every likelihood that rural residential preferences and digitally-enabled homeworking will fuel further demand for coworking too (McKinsey, 2021). Such a shift could challenge certain urban-centric assumptions of the Network Society based on the greater density of flows of people, knowledge and ideas that can fuel urban economic growth. Instead, rural regions can be supported in catching up with their urban coun- terparts if these flows of resources become increasingly accessible to rural entrepreneurs. As evidenced by those participating in our research, this can be facilitated through enhanced communications technologies, personal mobility and extensive networks. Rural coworking spaces can play important roles in elevating their localities to become more significant network nodes, combining local and extra-local networks around a space that depends upon both social and digital infrastructures. Conceptually, this emphasis on social and technological processes confirms that coworking can be an integral component of smart rural development too (Naldi et al., 2015). The potential for innovative mixing between sectors and professions adds a further dimension to rural coworking as a driver of new economic op- portunities. By fulfilling a combination of functions, they can be simultaneously remote network bridges connecting urban centres and urban firms and they can integrate rural economy actors into new networks. If, as a consequence of Covid-19, increased remote working becomes the norm to the extent that we conceive of ‘remote employers’ rather than ‘remote workers’, it is likely that the co-worker with rural business connections will be strongly positioned. Conversely, if the growth of remote working wanes, the potential functions of rural coworking nodes become less clear. We argue that a critical mass of human and social capital operating in rural places is integral to the development of cow- orking spaces as hubs for enterprising businesses. Through improved connectivity, which may take the form of better physical infrastructure or digital networks, rural areas are then better able to draw on a wider array of resources, which, in turn, can be leveraged to enhance the attractiveness of rural places and generate new economic activities. If resulting forms of entrepreneurship are socially embedded and digitally enabled, they can contribute to new dynamics of smart rural develop- ment that valorise spatial diversity (Naldi et al., 2015). Our paper has sought to re-invigorate the Network Society by applying its core ideas in the context of dominant place-based and “smart” rural development paradigms. This has revealed significant opportunities to promote new networks built around the social and technological needs of contemporary ways of working. Moreover, the strategies of rural coworking operators highlight the importance of identity, or “cultural distinctiveness” (Castells, 2004), in addition to the connectivity and openness to engage in heterogenous networks that characterise the Network Society. The research has also identified a challenge for rural policymakers and coworking operators to facilitate networks that bridge spatial, social and skills divides while supporting local cohesion and integration. We suggest that the most promising avenues to achieve this require rural coworking spaces to enhance their JournalofRuralStudies97(2023)550–559557 G. Bosworth et al. place-based distinctiveness by providing services to more isolated and marginalised groups, as well as the essential facilities and network brokerage demanded by rural co-workers. Author statement Gary Bosworth: Funding acquisition, Conceptualization, Method- ology, Investigation, Original draft. Jason Whalley: Conceptualization, Reviewing and editing. Anita Fuzi: Methodology, Investigation. Ian Merrell: Investigation, Reviewing and editing. Polly Chapman: Meth- odology, investigation, Reviewing and editing. Emma Russell: Funding acquisition, Reviewing and editing. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Data availability Data will be made available on request. Acknowledgements We would like to acknowledge the Digital Futures at Work Research Centre (Digit) for their funding and support throughout the project. References Akhavan, M., Mariotti, I., Rossi, F., 2021. The rise of coworking spaces in peripheral and rural areas in Italy. Territorio - Sezione Open Access (97-Supplemento). https://doi. org/10.3280/tr2021-097-Supplementooa12925. Ampuja, M., Koivisto, J., 2014. From “post-industrial” to “network society” and beyond: the political conjunctures and current crisis of information society theory. TripleC 12 (2), 447–463. Anderson, A., Dodd, S., Jack, S., 2010. Network practices and entrepreneurial growth. Scand. J. Manag. 26, 121–133. Anttiroiko, A.-V., 2016. Castells’ network concept and its connections to social, economic and political network analyses. J. Soc. Struct. 16, 1–18. Atherton, B., 2016. The rise and rise of the tech-savvy sole trader, Xero blog. Accessed 23rd February 2022 at: https://www.xero.com/blog/2016/01/the-rise-and-rise-of-th e-tech-savvy-sole-trader/. Baker, S., Warburton, J., Hodgkin, S., Pascal, J., 2017. The supportive network: rural disadvantaged older people and ICT. Ageing Soc. 37, 1291–1309. Benneworth, P., Charles, D., 2005. University spin-off policies and economic development in less successful regions: learning from two decades of policy practice. Eur. Plann. Stud. 13 (4), 537–557. Bock, B., 2016. Rural marginalisation and the role of social innovation; a turn towards nexogenous development and rural reconnection. Sociol. Rural. 56 (4), 552–573, 2016. Bosworth, G., Bat Finke, H., 2020. Commercial counterurbanisation and the changing roles of rural businesses. Environ. Plann. 52 (3), 654–674. Bosworth, G., Turner, R., 2018. Interrogating the meaning of a rural business through a rural capitals framework. J. Rural Stud. 60 (1), 1–10. Bosworth, G., Venhorst, V., 2018. Economic linkages between urban and rural regions what’s in it for the Rural? Reg. Stud. 52 (8), 1075–1085. Camagni, R., 1995. The concept of innovative milieu and its relevance for public policies in European lagging regions. Pap. Reg. Sci. 74 (4), 317–340. Castells, M., 2004. The Network Society: A Cross-Cultural Perspective. Edward Elgar, Cheltenham. Champion, T., Coombes, M., Brown, D., 2009. Migration and longer-distance commuting in rural england. Reg. Stud. 43 (10), 1245–1259. Clark, P., 2020. Here’s what the Office of 2021 Should Look like, 6 December. Financial Times available at: www.ft.com. Clifton, N., Füzi, A., Loudon, G., 2019. Coworking in the Digital Economy: Context, Motivations, and Outcomes Futures, 102439. https://doi.org/10.1016/j. futures.2019.102439. Copus, A., Skuras, D., 2006. Business networks and innovation in selected lagging areas of the European union: a spatial perspective. Eur. Plann. Stud. 14 (1), pp79–93. Cowie, P., Townsend, L., Salemink, K., 2020. Smart rural futures: will rural areas be left behind in the 4th industrial revolution? J. Rural Stud. 79, 169–176. Dahl, M., Sorenson, O., 2020. Will Covid-19 Take the Air Out of Silicon Valley? Entrepreneur & Innovation Exchange, 2020. https://doi.org/10.32617/586- 5fdfae3ca0cbf. Published online at EIX.org on December 20th. Davidsson, P., 2015. Entrepreneurial opportunities and the entrepreneurship nexus: a re- conceptualization. J. Bus. Ventur. 30, 674–695. De Massis, A., Kotlar, J., Wright, M., Kellermanns, E.W., 2018. Sector-based entrepreneurial capabilities and the promise of sector studies in entrepreneurship. Enterpren. Theor. Pract. 42 (1), 47–69. Dingel, J., Neiman, B., 2020. How many jobs can be done at home? National Bureau of Economic Research. Accessed 14th February 2022 at: https://www.nber.org/papers /w26948. European Commission, 2020. Digital Economy and Society Index accessed 26th February 2022 at: https://digital-strategy.ec.europa.eu/en/policies/desi. Feldman, M., Guy, F., Iammarino, S., 2020. Regional income disparities, monopoly and finance. Camb. J. Reg. Econ. Soc. 14 (1), 25–49. Felstead, A., Reuschke, D., 2021. A flash in the pan or a permanent change?. In: The Growth of Homeworking during the Pandemic and its Effect on Employee Productivity in the UK. Information Technology and People, pp. 1–22. https://doi. org/10.1108/ITP-11-2020-0758. Fiorentino, S., 2019. Different typologies of ‘coworking spaces’ and the contemporary dynamics of local economic development in Rome. Eur. Plann. Stud. 27 (9), 1768–1790. Fisker, J.K., Kwiatkowski, G., Hjalager, A.M., 2021. The translocal fluidity of rural grassroots festivals in the network society. Soc. Cult. Geogr. 22 (2), 250–272. Florida, R., Rodrıguez-Pose, A., Storper, M., 2020. Cities in a post-COVID world, papers in evolutionary economic geography (PEEG). Accessed 5th December 2022 at: http://econ.geo.uu.nl/peeg/peeg2041.pdf. Füzi, A., 2015. Coworking spaces for promoting entrepreneurship in sparse regions: the case of South Wales. Regional Studies, Regional Science 2 (1), 462–469. Gerli, P., Whalley, J., 2022. Digital entrepreneurship in a rural context: the implications of the rural-urban digital divide. In: Keyhani, M., Kollmann, T., Ashjari, A., Sorgner, A., Hull, C.E. (Eds.), Handbook of Digital Entrepreneurship. Edward Elgar, Cheltenham, UK. Gerli, P., Matteuci, N., Whalley, J., 2020. Infrastructure provision on the margins: an assessment of Broadband Delivery UK. Int. J. Publ. Adm. 43 (6), 540–551. Glaeser, E., 2021. Urban Resilience. National Bureau of Economic Research, p. WP29261. Accessed 23rd February 2022 at: www.nber.org/papers/w29261. Gruber, S., Soci, A., 2010. Agglomeration, agriculture and the perspective of the periphery. Spatial Econ. Anal. 5 (1), 43–72. Grubesic, T.H., Mack, E.A., 2017. Broadband Telecommunications and Regional Development. Routledge, London, UK. Hacker, K., Mason, S., Morgan, E., 2009. Digital disempowerment in a network society. Int. J. Electron. Govern. Res. 5 (2), 57–71. Hackney, A., Yung, M., Somasundram, K.G., Nowrouzi-Kia, B., Oakman, J., Yazdani, A., 2022. Working in the digital economy: a systematic review of the impact of work from home arrangements on personal and organisational performance and productivity. PLoS One 17 (10), e0274728. Helsper, E.J., van Deursen, A.J.A.M., 2017. Do the rich get digitally richer? Quantity and quality of support for digital engagement. Inf. Commun. Soc. 20 (5), 700–714. Herslund, L., 2012. The rural creative class: counterurbanisation and entrepreneurship in the Danish countryside. Sociol. Rural. 52 (2), 235–255. Hill, I., 2022. Rural Arts Entrepreneurs’ Placemaking – How ‘entrepreneurial Placemaking’ Explains Rural Creative Hub Evolution during COVID-19 Lockdowns. In: Local Economy (in press). Honig, B., Samuelsson, M., 2021. Business planning by intrapreneurs and entrepreneurs under environmental uncertainty and institutional pressure. Technovation 2021, 102124. Horlings, L., Roep, D., Mathijs, E., Marsden, T., 2020. Exploring the transformative capacity of place-shaping practices. Sustain. Sci. 15, 353–362. Houghton, K., Foth, M., Hearn, G., 2018. Working from the other office: trialling coworking spaces for public servants. Aust. J. Publ. Adm. 77 (4), 757–778. Johns, J., Hall, S., 2020. I have so little time [. . .] I got shit I need to do’: critical perspectives on making and sharing in Manchester’s FabLab. Environ. Plann. 52 (7), 1292–1312. Johnshton, Huggins, 2016. Drivers of university-industry links: the case of knowledge- intensive business service firms in rural locations. Reg. Stud. 50 (8), 1330–1345. Jordan, A., 2021. Best Payment Processors for UK Small Businesses – 11 of the Best, Small Business, 18 October available at: https://smallbusiness.co.uk. Kalantaridis, C., Bika, Z., 2011. Entrepreneurial origin and the configuration of innovation in rural areas: the case of Cumbria. In: Environment and Planning A, vol. 43. North West England, 886-884. Kane, K., Clark, 2019. Mapping the landscape of urban work: home-based businesses and the built environment. Environ. Plann. 51 (2), 323–350. Keeble, D., Nachum, L., 2002. Why do business service firms cluster? Small consultancies, clustering and decentralization in London and southern England. Trans. Inst. Br. Geogr. 27 (1), 67–90. Kitagawa, R., Kuroad, S., Okudaira, H., Owan, H., 2021. Working from home and productivity under the COVID-19 pandemic: using survey data of four manufacturing firms. PLoS One 16 (12), e0261761. Kolehmainen, J., Irvine, J., Stewart, L., Karacsonyi, Z., Szab´o, T., Alarinta, J., Norberg, A., 2016. Quadruple Helix, innovation and the knowledge-based development: lessons from remote, rural and less-favoured regions. Journal of the Knowledge Economy 7 (1), 23–42. Korsgaard, S., Muller, S., Tanvig, H., 2015. Rural entrepreneurship or entrepreneurship in the rural – between place and space. Int. J. Entrepreneurial Behavior and Research 21 (1), 5–26. Kov´acs, J., Zolt´an, E., 2017. Rural enterprise hub supporting rural entrepreneurship and innovation – case studies from Hungary. Eur. Countrys. 9 (3), 473-385. Lorne, C., 2019. The limits to openness: coworking, design and social innovation in the neoliberal city. Environ. Plann. 52 (4), 747–765. JournalofRuralStudies97(2023)550–559558 G. Bosworth et al. Malecki, E.J., 1997. Technology and Economic Development: the Dynamics of Local, Regional and National Competitiveness. Longman, Harlow. Malecki, E., Moriset, B., 2008. The Digital Economy; Business Organization, Production Ray, C., 2006. Neo-endogenous rural development in the EU. In: Cloke, P., Marsden, T., Mooney, P. (Eds.), The Handbook of Rural Studies. SAGE Publications Ltd), London, pp. 278–291. Processes and Regional Developments. Routledge, Abingdon, UK. Reuschke, D., 2019. The subjective well-being of homeworkers across life domains. Marcus, S., 2022. COVID-19 and the Shift to Remote Working. Policy Contribution 09/ Environ. Plann. 51 (6), 1326–1349. 22, June, Accessed 15 December at. https://www.bruegel.org. Mason, C., Brown, R., 2014. Entrepreneurial ecosystems and growth oriented entrepreneurship. Final Report to OECD, Paris 30 (1), 77–102. Mazur, P., Duchlinski, P., 2020. Credibility and creativity in network society. Creativity Studies 13 (1), 53–63. https://journals.vgtu.lt/index.php/CS/article/view/6585/ 9626. McGranahan, D., Wojan, T., 2007. Recasting the creative class to examine growth processes in rural and urban counties. Reg. Stud. 41 (2), 197–216. McKinsey, 2021. The future of work after COVID-19. Accessed 2nd February 2022 at: www.mckinsey.com/featured-insights/future-of-work/the-future-of-work-after-c ovid-19. Merrell, I., Fuzi, A., Russell, E., Bosworth, G., 2022. How rural coworking hubs can Reuschke, D., Clifton, N., Long, J., 2021. Remote Working - Spatial Implications in Wales. In: A Report Commissioned by the Welsh Parliament. Accessed 5th December 2022 at: https://research.senedd.wales/research-articles/new-publication-remote- working-spatial-implications-in-wales/. Russell, E., Grant, C., 2020. Agile Working and Well-Being in the Digital Age. Palgrave MacMillan, Cham, Switzerland. Sadowski, B.M., 2017. Advanced users and the adoption of high-speed broadband: results of a living lab in The Netherlands. Technol. Forecast. Soc. Change 1155, 1–14. Salemink, K., Strijker, D., Bosworth, G., 2017. Rural Development in the Digital Age: a systematic literature review on unequal ICT availability, adoption, and use in rural areas. J. Rural Stud. 54, 360–371. ˇ ˇ Sebestov´a, J., Sperka, R., Małecka, J., Łuczka, T., 2017. Coworking centres as a potential facilitate well-being through the satisfaction of key psychological needs. Local Econ 36 (7–8), 606–626. supportive network for cross-border business co-operation. Forum Scientiae Oeconomia 5 (4), 23–34. Mitchell, C., Madden, M., 2014. Re-Thinking commercial counterurbanisation: evidence Shane, S.A., Venkataraman, S., 2000. The promise of entrepreneurship as a field of from rural nova scotia, Canada. J. Rural Stud. 36, 137–148. Mokhtarian, P., Bagley, M., 2000. Modeling employees’ perceptions and proportional preferences of work locations: the regular workplace and telecommuting alternatives. Transport. Res. Pol. Pract. 34, 643–675. Mu˜noz, P., Kimmitt, J., 2019. Rural entrepreneurship in place: an integrated framework. Enterpren. Reg. Dev. 31 (9–10), 842–873. research. Acad. Manag. Rev. 25 (1), 217–226. Shearmur, R., Doloreux, D., 2015. Central places or networks? Paradigms, metaphors, and spatial configurations of innovation-related service use. Environ. Plann. 47 (7), 1521–1539. Simonsen, K., 2004. Networks, flows, and fluids – reimagining spatial analysis? Environ. Plann. 36 (8), 1333–1337. Murdoch, J., 2000. Networks – a new paradigm of rural development? J. Rural Stud. 16, Slee, B., 2019. Delivering on the concept of smart villages – in search of an enabling 407–419. Murdoch, J., 2006. Networking rurality: emergent complexity in the countryside. In: Cloke, P., Marsden, T., Mooney, P.H. (Eds.), Handbook of Rural Studies. Sage, London, pp. 171–184. Naldi, L., Nilsson, P., Westlund, H., Wixe, S., 2015. What is smart rural development? J. Rural Stud. (40) 190-101. Nambisan, S., 2017. Digital entrepreneurship: toward a digital technology perspective of entrepreneurship. Enterpren. Theor. Pract. 41 (6), 1029–1055. Nambisan, S., Wright, M., Feldman, M., 2019. The digital transformation of innovation and entrepreneurship: progress, challenges and key themes. Res. Pol. 48 (8), 103773. Nathan, M., Overman, H., 2020. Will coronavirus cause a big city exodus? Environment and Planning B: Urban Analytics and City Science 47 (9), 1537–1542. Newman, P., 2020. Covid, cities and climate: historical precedents and potential transitions for the new economy. Urban Science 4 (3), 32. OECD, 2018. Rural 3.0: a framework for rural development. Accessed 23rd at: https:// www.oecd.org/rural/rural-development-conference/documents/. OECD, 2019. Skills outlook – thriving in a digital world. Accessed 23rd December at: https://www.oecd-ilibrary.org/education/oecd-skills-outlook-2019_df80bc12-en. Oestmann, S., Dymond, A.C., 2001. Telecentres—experiences, lessons and trends. Telecentres: Case studies and key issues 1, 1–15. theory. Eur. Countrys. 11 (4), 634–650. Spinuzzi, C., 2012. Working alone together: coworking as emergent collaborative activity. J. Bus. Tech. Commun. 26 (4), 399–441. Steiner, A., Atterton, J., 2014. The contribution of rural businesses to community resilience. Local Econ. 29 (3), 228–244. Stockdale, A., 2015. Contemporary and ‘messy’ rural in-migration processes: comparing counterurban and lateral rural migration. Popul. Space Place 22 (6), 599–616. Stojmenova Duh, E., Kos, A., 2016. Fablabs as drivers for open innovation and co- creation to foster rural development. In: International Conference on Identification, Information and Knowledge in the Internet of Things. https://doi.org/10.1109/ IIKI2016.70. Storper, M., 1997. Regional economies as relational assets. In: Lee, R., Wills, J. (Eds.), Geographies of Economics. Arnold, London, pp. 248–258. Tasselli, S., Kilduff, M., 2021. Network agency. Acad. Manag. Ann. 15 (1) https://doi. org/10.5465/annals.2019.0037. Tomaz, E., Moriset, B., Teller, J., 2021. Rural coworking spaces in the Covid-19 era. A window of opportunity? Accessed 16th December 2021 at: https://halshs.archives-ou vertes.fr. Townsend, L., Wallace, C., Fairhurst, G., Anderson, A., 2017. Broadband and the creative industries in rural Scotland. J. Rural Stud. 54, 451–458. Ofcom, 2020. Connected Nations 2020 - UK report, 17 December, available at: www.ofc Vallas, S., Schor, J.B., 2020. What do platforms do? Understanding the gig economy. om.gov.uk. Ohnmacht, T., Z’Rotz, J., Dang, L., 2020. Relationships between Coworking Spaces and CO2 emissions in work-related commuting: first empirical insights for the case of Switzerland with regard to urban-rural differences. Environmental Research Communications 2 (12), 125004. Paavonen, W., 1999. Telecottages and other work centre experiments. Telektronikk 95 (4), 64–68. Phillipson, J., Tiwasing, P., Gorton, M., Maioli, S., Newbery, R., Turner, R., 2019. Shining a spotlight on small rural businesses: how does their performance compare with urban? J. Rural Stud. 68, 230–239. Annu. Rev. Sociol. 46, 273–294. Varnelis, K., 2008. Networked Publics. MIT Press, Cambridge, MA. Wire, Property, 2020. Demand for homes in rural areas rises due to the pandemic. available at: www.propertywire.com. Zavratnik, V., Superina, A., Stojmenova Duh, E., 2019. Living labs for rural areas: contextualization of living lab frameworks, concepts and practices. Sustainability 11 (14), 3797. Zhen, F., Jia, T., Wang, X., 2020. How does castells’s the rise of the network society contribute to research in human geography? A citation content and context analysis. Prof. Geogr. 72 (1), 96–108. JournalofRuralStudies97(2023)550–559559
null
10.1007_s00428-018-2504-0.pdf
null
null
Virchows Archiv (2019) 474:289–296 https://doi.org/10.1007/s00428-018-2504-0 ORIGINAL ARTICLE A novel algorithm for better distinction of primary mucinous ovarian carcinomas and mucinous carcinomas metastatic to the ovary Michiel Simons 1 Leon F. Massuger 4 & Iris D. Nagtegaal 1 & Thomas Bolhuis 1 & Anton F. De Haan 2 & Annette H. Bruggink 3 & Johan Bulten 1 & Received: 16 May 2018 / Revised: 21 November 2018 / Accepted: 3 December 2018 / Published online: 10 January 2019 # The Author(s) 2019 Abstract Primary mucinous ovarian carcinomas (MOC) are notoriously difficult to distinguish from mucinous carcinomas metastatic to the ovary (mMC). Studies performed on small cohorts reported algorithms based on tumor size and laterality to aid in distinguishing MOC from mMC. We evaluated and improved these by performing a large-scale, nationwide search in the Dutch Pathology Registry. All registered pathology reports fulfilling our search criteria concerning MOC in the Netherlands from 2000 to 2011 were collected. Age, histology, laterality, and size were extracted. An existing database covering the same timeline containing tumors metastatic to the ovary was used, extracting all mMC, age, size, laterality, and primary tumor location. Existing algorithms were applied to our cohort. Subsequently, an algorithm based on tumor histology, laterality, and a nomogram based on age and size was created for differentiating MOC and mMC. We identified 735 MOC and 1018 mMC. Patients with MOC were significantly younger and MOC were significantly larger and more often unilateral than mMC. Signet ring cell carcinomas were rarely primary. Our algorithm used signet ring cell histology, bilaterality, and a nomogram integrating patient age and tumor size to diagnose mMC. Sensitivity and specificity for mMC was 90.1% and 59.0%, respectively. Applying existing algorithms on our cohort yielded a far lower sensitivity. The algorithm described here using tumor histology, laterality, size, and patient age has higher sensitivity but lower specificity compared to earlier algorithms and aids in indicating tumor origin, but for conclusive diagnosis, careful integration of morphology, immunohistochemistry, and clinical and imaging data is recommended. Keywords Mucinous ovarian carcinoma . Colorectal carcinoma . Metastasis . Algorithm Electronic supplementary material The online version of this article (https://doi.org/10.1007/s00428-018-2504-0) contains supplementary material, which is available to authorized users. * Michiel Simons Michiel.Simons@radboudumc.nl 1 Department of Pathology, Radboud University Medical Center, Nijmegen 6525, GA, The Netherlands 2 Department for Health Evidence, Radboud University Medical Center, Nijmegen 6525, GA, The Netherlands 3 PALGA, The Nationwide Network and Registry of Histo- and Cytopathology in the Netherlands, 3995, GA Houten, The Netherlands 4 Department of Obstetrics and Gynecology, Radboud University Medical Center, Nijmegen 6525, GA, The Netherlands Introduction It is well known that a considerable part of mucinous ovar- ian carcinomas are in fact metastases, mainly from the gas- trointestinal tract, pancreas, and gallbladder [1–4]. The dis- tinction of primary mucinous carcinomas of the ovary (MOC) and mucinous carcinomas metastatic to the ovary (mMC) might be difficult and misdiagnosis has important consequences for therapy. Chemotherapy regimens differ between tumor types and advanced stage MOC are gener- ally associated with a poor response to treatment [5]. Although certain histological features may be indicative of primary or metastatic origin, these are often inconclu- sive [6–9]. A classic immunohistochemical panel of CK7, CK20, and CDx2 is usually considered helpful in indicat- ing tumor origin, but unfortunately shows overlap in ex- pression patterns in MOC and mMC [10–13]. Also, partic- ularly MOC arising from teratomas are known to express a more gastrointestinal phenotype [14]. 290 Virchows Arch (2019) 474:289–296 Macroscopic features as size and laterality have also been investigated. Unilaterality and large size is indicative of MOC, while bilaterality is more suggestive of mMC [7]. We have shown earlier that colorectal mMC are uni- lateral in almost 60% of cases [1]. Despite this, these macroscopic features have been proposed by various stud- ies as discriminators between MOC and mMC. Seidman et al. proposed an algorithm designating unilateral tumors smaller than 10 cm and bilateral tumors as mMC, and unilateral tumors of at least 10 cm in size as MOC [2]. Another algorithm by Yemelyanova et al. used a different size cut off point of 13 cm [15]. These algorithms classi- fied 90% and 87% of tumors correctly, and yielded a sensitivity for mMC of 94.7% and 82%, respectively. These algorithms should focus on a low rate of false neg- ative patients with mMC, since misdiagnosis will lead to withholding the diagnostic workup to identify a primary tumor elsewhere with important therapeutic and prognos- tic consequences. Previous studies were performed on relatively small co- horts. We aimed to evaluate these algorithms on a larger tumor cohort and improve them where possible. Materials and methods Case selection: primary mucinous ovarian tumors The nationwide network and registry of histopathology and cytopathology in the Netherlands (PALGA) codes and saves pathology reports in the Netherlands from 1971 with nationwide coverage from 1991 [16]. We performed a na- tionwide search for primary (micro-)invasive mucinous ovarian carcinomas diagnosed between 2000 and 2011 in the PALGA database obtained by complete resection. All tumors of non-mucinous, mixed, and uncertain histology were excluded. Tumors labeled Krukenberg tumors were excluded from the MOC group, since this term refers to metastatic signet ring cell carcinomas [17]. Tumors asso- ciated with pseudomyxoma peritonei (PMP) were exclud- ed. Any tumors of which origin was reported to be uncer- tain were excluded. History of eligible patients was re- quested at PALGA, and patients with a history of a gastro- intestinal tumor regardless of histology, a mucinous tumor regardless of location, and an adenocarcinoma NOS locat- ed in the genital tract were excluded. For each patient, we extracted the following items: age at time of diagnosis, origin (primary or metastasis), histological subtype, laterality, and size of the ovarian tumor. In case of bilateral tumors, both largest and smallest sizes were registered if available. In case of unilateral tumors, size was scored as largest. If only one ovary was resected or reported, we considered the tumor to be unilateral. Case selection: mucinous tumors metastatic to the ovary A database containing mMC was created earlier. Details about criteria for this database are described elsewhere [1]. From this database, we extracted all tumors metastatic to the ovary with histological proof of extra-ovarian origin and mucinous histology. Additional macroscopic data were requested at PALGA. Cases were excluded from this data- base if tumor size mentioned in additional macroscopy and conclusion was discrepant or if macroscopy contained any information making it uncertain whether a tumor was pri- mary or metastatic. For cases in this dataset, we addition- ally extracted location of the primary tumor. These data were combined to create a database containing both MOC and mMC. Statistical analysis For cases with no data available for laterality or size, multiple imputation was applied to estimate these values to maintain cohort size and to avoid biased estimates in the regression analyses. With this technique, multiple complete datasets are created by drawing a value for the missing values based on the estimated distribution. Each dataset is analyzed and the results are combined [18]. Imputed variables and variables used for imputation are shown in Online Resource 1. Twenty imputated datasets were created. We applied the algorithms described earlier on our da- tabase to evaluate sensitivity, specificity, and number of correctly classified cases. To identify discriminating factors, logistic regression was carried out for each step in the algorithm creation process. Our approach was based on a high sensitivity for mMC. For creating nomogram scores, regression coefficients B were calculated using logistic regression with the con- tinuous variables age and largest size. For size and age, a score Score(size + age) was calculated for nomogram crea- tion. Details are found in Online Resource 2. Statistical analysis was performed using IBM SPSS statistics version 20.0. For comparison of means, two- tailed t tests were performed, for comparison of frequency distributions between categorical data χ2 tests were per- formed. A p value < 0.05 was considered statistically sig- nificant. ROC curves were used to determine optimal cut- off points. Virchows Arch (2019) 474:289–296 Results Features of primary mucinous ovarian tumors and tumors metastatic to the ovary A total of 735 MOC and 1018 mMC were identified. Laterality data was missing for 52 MOC (7.1%) and 84 mMC (8.3%); largest size data were missing for 129 MOC (17.6%) and 312 mMC (30.6%). Patients with MOC were significantly younger than patients with mMC (54.6 vs. 59.6 years; p < 0.01) and had larger tumors (19.0 vs. 12.0 cm; p < 0.01). Size and age distribution among patients with MOC and mMC are depicted in Fig. 1. Patients with MOC had unilateral tumors in 662 cases (90.1%) vs. 73 (9.9%) bilateral tumors, whereas patients with mMC had bi- lateral tumors in 508 cases (49.9%) and unilateral in 510 cases (50.1%) (p < 0.001). Signet ring cell carcinomas were more often metastatic than primary (122 (98.4%) vs. 2 (1.6%); p < 0.001). Bilateral tumors were more often metastatic than primary (508 (87.4%) vs. 73 (12.6%); p < 0.001), whereas unilateral tumors were primary in 662 cases (56.5%) and were metastatic in 510 cases (43.5%). Characteristics before and after imputation are shown in Tables 1 and 2, showing that this led to no significant changes. Comparison to earlier studies Seidman et al. [2] classified tumors as MOC if they were uni- lateral and ≥ 10 cm. In our cohort, 15.4% of tumors < 10 cm were primary and 84.6% was metastatic. Tumors ≥ 10 cm were primary in 52.5% and were metastatic in 47.5%. MMC were < 10 cm in 41.5% and ≥ 10 cm in 58.5%. Of MOC, this was 10.5% and 89.5%, respectively. On our data, the Seidman al- 291 gorithm has a sensitivity of 72.5% and a specificity of 82.4% and of all 76.6% tumors were classified correctly. Yemelyanova et al. [15] used 13 cm as a size cutoff point. In our cohort, tumors < 13 cm were primary in 26.8% and were metastatic in 73.2%. Tumors ≥ 13 cm were primary in 56.9% and were metastatic in 43.1%. MMC were < 13 cm in 56.8% and ≥ 13 cm in 43.2%. Of MOC, this was 21.1% and 78.9%, respectively. On our data, the Yemelyanova algorithm has a sensitivity of 79.9% and a specificity of 73.6% and of all tumors 77.2% were classified correctly. Further test details for both algorithms are shown in Table 3. Optimizing algorithm Logistic regression identified age, largest size, histology, and laterality as significant independent predicting factors for distinguishing MOC from mMC. Regression coefficients, odds ratios, and 95% confidence intervals are displayed in Online Resource 2. Signet ring cell histology compared to non-signet ring cell histology showed a sensitivity of only 12.0%, but a specificity of 99.7% for indicating metastasis, with a positive predictive value for metastasis of 98.4%. Comparing bilaterality to unilaterality as a next step, after excluding signet ring cell carcinomas, shows a sensitivity of only 48.1%, but a specific- ity of 90.0% for indicating metastasis, with a positive predic- tive value of 85.5%. Based on the remaining cases, areas under the curve (AUC) for largest size and age as a determinant of origin were 0.78 and 0.64, respectively. To test a combination of these two variables, logistic regression including age and largest size was carried out and rendered regression coefficient Bsize Metastasis Primary Metastasis Primary ) m c ( e z i S 60 40 20 0 100 80 60 40 20 0 ) s r a e y ( e g A 60 40 S i z e ( c m ) 20 0 100 80 60 40 20 0 A g e ( y e a r s ) 120 100 80 60 40 20 0 20 40 60 80 100 120 60 40 20 0 0 20 40 60 Frequency Frequency Frequency Frequency Fig. 1 Frequency distribution for largest size (a) and age (b) 292 Virchows Arch (2019) 474:289–296 Table 1 Features of primary and metastatic mucinous ovarian carcinomas before imputation, age, and size expressed as mean Parameter Age Histology Location primary tumor Laterality Size (largest) Total Primary % Metastasis % Mucinous Signet-ring cell Appendix Bladder Breast Cervix Endometrium Colon Duodenum Small intestine Pancreas Bile ducts/gallbladder Esophagus Stomach Urachus Left Right Bilateral Unknown 54.6 ± 15.1 733 2 45.0 1.6 284 330 69 52 18.9 ± 7.9 735 56.6 54.1 13.7 38.2 41.9 59.6 ± 13.1 896 122 97 2 3 2 4 748 1 22 17 14 7 100 1 218 280 436 84 11.6 ± 6.4 1018 55.0 98.4 9.5 0.2 0.3 0.2 0.4 73.5 0.1 2.2 1.7 1.4 0.7 9.8 0.1 43.4 45.9 86.3 61.7 58.1 p value < 0.001 < 0.001 < 0.001 < 0.001 0.154 and Bage − 0.033, respectively (p < 0.001 for both vari- ables). Larger tumors and lower age tended to be associated with primary tumors, although distributions showed too much overlap to be used as a solitary determinant (see Fig. 1). The largest size range was 1 to 60 cm and age range was 15 to 95 years. Exact calculations can be found in Online Resource 3. Final scores for size and age can be found in Online Resources 4 and 5, respectively. The ROC curve for Score(size + age) showed an AUC of 0.81 (see Online Resource 6), and for Score(size) or Score(age) again 0.78 and 0.64, respectively. Based on the AUC, Score(size + age) was considered superior to Score(size) or Score(age) separately. An optimal cutoff point for the sum of these scores was Table 2 carcinomas after imputation Size and laterality of primary and metastatic mucinous ovarian Parameter Primary % Metastasis % Laterality Left Right Bilateral Unknown Size (largest) Total 307 355 73 0 19.0 735 57.8 55.4 12.6 224 286 508 0 42.2 44.6 87.4 12.0 < 0.001 41.9 1018 58.1 determined as 6.1 using the ROC curve coordinates. A nomo- gram based on this score is shown in Fig. 2. The final algo- rithm as depicted in Fig. 3 shows a sensitivity and specificity of 90.1% and 59.0%, respectively, and 77.1% of tumors were classified correctly. Details are shown in Table 3. Table 3 Results of algorithms on current tumor cohort Study Origin Primary Metastasis Seidman et al. Yemelyanova et al. p value < 0.001 Current study Primary Metastasis Sensitivity Specificity Primary Metastasis Sensitivity Specificity Primary Metastasis Sensitivity Specificity 604 131 72.4% 82.2% 541 194 79.9% 73.6% 434 301 90.1% 59.0% 280 738 205 813 101 917 Virchows Arch (2019) 474:289–296 293 Fig. 2 Nomogram based on Score(size + age). By applying patient age en tumor size to the corresponding axes and extrapolating a line through these points to the lower axis, final Score(size + age) can be determined Discussion MOC are often difficult to distinguish from mMC, since mor- phological and immunohistochemical features are unsatisfac- tory differentiators. In the current study, we composed the largest database of MOC and mMC to our knowledge to eval- uate size and laterality as predictors of tumor origin. Patients with MOC were significantly younger, and MOCs were larger and more often unilateral, which is in line with earlier findings [7, 8]. We compared our data to earlier algorithms using these features and optimized the algorithm by adding presence of signet ring cells and patient age. Earlier algorithms, based on small patient cohorts, of only 50, 194, and 68 tumors, respectively, solely used laterality and size of the tumors [2, 15, 19]. Application of a 10-cm cutoff in two studies resulted in a sensitivity of 83–95% [2, 19]; adjust- ment of the cutoff to 13 cm showed a 82% sensitivity [15]. The populations used in these studies were heterogeneous because of diverse inclusion criteria regarding tumors of un- certain primary site and endometrioid and signet ring cell Fig. 3 Final algorithm for distinguishing primary mucinous carcinomas and carcinomas metastatic to the ovary using parameters signet ring cells, laterality, patient age, and tumor size. For calculating Score(size + age), use the nomogram displayed in Fig. 2 histology. Signet ring cell carcinoma can be of primary ovar- ian origin, but this is extremely rare [20]. In our cohort, less than 2 per 100 signet ring cell carcinomas were MOC. Applying the earlier algorithms to our cohort yielded far lower sensitivity compared to our algorithm, suggesting that our algorithm including signet ring cells and a combination of relative values for tumor size and patient age renders superior results. Interestingly, sensitivity was also lower than found in the cohorts used in their original studies. Since the number of correctly classified tumors in general was comparable (ap- proximately 77%), these differences seem to be mainly the consequence of different composition of the cohorts. This can be explained by several factors. Firstly, the distribution of primary tumors in the mMC group differs between study populations, which may be due to geographical differences. Secondly, revision of cases in our cohort is not feasible due to large numbers, but since it concerns a nation-wide population- based cohort, it reflects daily practice. Thirdly, patients from tertiary referral centers include a selection of patients, with unusual cases, as can be observed in the Yemelyanova study, that included as much as 35% consultation cases. No bilateral MOC were observed in the Yemelyanova co- hort, as opposed to both our cohort (8.9% bilateral MOC) and the cohorts of Seidman and Khunarmonpong (17% and 12.5%, respectively) [2, 19]. Bilaterality of MOC might be explained by MOC metastasizing from one ovary to the con- tralateral ovary without this being recognized or reported as such. The possibility of a misdiagnosed mMC cannot be fully excluded. In the current study, the number of bilateral mMC was much lower with 49.9%, most likely due to the large number of colorectal metastases in our cohort which are known for their ability to present as large, unilateral metasta- ses [21]. Another difference is that Yemelyanova et al. also included atypical proliferative mucinous (borderline) tumors (APMTs) and tumors associated with PMP. The latter may have led to a higher number of bilateral metastatic tumors, since we discarded cases associated with PMP. Ovarian in- volvement of pseudomyxoma peritonei has been shown to 294 Virchows Arch (2019) 474:289–296 be almost invariably of appendiceal origin, the only potential but rare exception being a mucinous neoplasm originating in an ovarian teratoma [14, 22–24]. Hence, cases associated with pseudomyxoma peritonei pose less diagnostic problems. Also, pathological classification of pseudomyxoma peritonei remains problematic [25–27]. We also excluded APMTs to prevent contamination of the MOC group with misclassified mMC, since APMTs would not generally trigger workup for metastasis from a primary tumor elsewhere. Especially pan- creatic tumors are known for their capability to mimic APMTs of the ovary. In addition, we ideally wanted to include carci- nomas according to WHO criteria, but for micro-invasion varying criteria are used and the exact proportions of invasive foci were rarely reported. To prevent exclusion of actual inva- sive carcinomas falsely diagnosed as micro-invasive, we did include tumors reported to be micro-invasive. In our cohort, size and patient age were significantly differ- ent between MOC an mMC, but showed too much overlap to be discriminating by themselves. We integrated age in the existing algorithm, using it in direct combination with size. The optimized algorithm based on our own cohort led to a sensitivity and specificity of 90.1% and 59.0%, respectively. Since misdiagnosing an mMC as an MOC has greater conse- quences for further diagnostic workup and therapy than vice versa, our approach was based on a high sensitivity for diag- nosing mMC and yielding a low number of false negative patients. This reduces the possibility of patients ultimately re- ceiving inappropriate treatment for their disease, which differs considerably. Primary mucinous ovarian carcinomas are pri- marily treated surgically, followed by a combination of paclitaxel- and platinum-based chemotherapy in case of ad- vanced stage disease. In case of mMC, patients will be surgi- cally treated if possible, followed by up to triple therapy with platinum-based chemotherapy, fluoropyrimidines, irinotecan, and the addition of targeted therapy if indicated. With a positive predictive value of 75.3% for mMC, almost 25% of patients will undergo unnecessary diagnostic workup. In the intraoper- ative setting the algorithm has limited value, since low speci- ficity might lead to denial of surgical staging of patients with MOC when a mMC is reported. However, in practice, manual exploration of the abdominal cavity is performed, which— given the high incidence of both colorectal and appendiceal metastasis—can lead to clinical confirmation of metastasis. In absence of this clinical confirmation, limited staging can be performed, and the surgeon can consider to perform (limited) surgical staging based on the intraoperative suspicion. In this study, we used multiple imputations to replace values missing at random (MAR) by values drawn from an estimated distribution of the variable in question, a method used frequently in biomedical research [28–30]. This tech- nique is based on the general statistical principle that every subject in a randomly chosen sample can be replaced by a new subject that is randomly chosen from the same source population. Analysis of available cases when values are MAR is no longer based on a random sample from the source population, leading to severely biased study associations and incorrect standard errors. This can be reliably overcome by multiple imputations, rendering this method superior to com- plete cases analysis [31, 32]. The imprecision caused by the fact that the distribution of the variables with missing values is estimated, is taken into account by creating multiple imputated datasets and combining these to obtain a pooled estimate of the parameters and standard errors [32]. The ran- dom subset of which new subjects are chosen or imputed is defined by the already known characteristics (the variables used for imputation). Using as many as six variables for im- putation greatly reduces the influence of the technique on the final result [33]. Our algorithm was not subjected to valida- tion, since there is a lack of large validation sets for this type of patient cohort. This might lead to overestimated accuracy, although due to the large sample size this overestimation will be relatively small. Lack of a gold standard for classifying MOC and mMC causes difficulties in creating study populations in all reported studies to date. We did not revise the cases included in our cohort. We used a proven primary tumor elsewhere as evi- dence for metastatic ovarian disease, which can be considered an objective criterion. The probability of patients presenting with both a MOC and a gastrointestinal tumor simultaneously seems very low. It is conceivable that metastatic disease may have been falsely classified as a primary ovarian tumor, if no diagnostic workup took place because of initial misdiagnosis or if patients did not undergo surgery of the primary tumor. However, the large sample size reduces the influence of these factors to some extent. Evaluation of histological and immunohistochemical features as well as clinical and imaging data was impeded by the large sample size and therefore considered beyond the scope of this study. Microscopic features observed more often in MOC are for example expansive growth patterns or presence of precursor lesions, whereas features such as infiltrative growth, dirty necro- sis, lymph vessel invasion, and surface involvement are seen more often in mMC [7–9]. Multiple studies have shown that MOC and mMC show overlap in classic immunohistochemical expression patterns [10–13]. Despite overlap of these morpho- logical and immunophenotypical features between MOC and mMC, integrating histological and immunohistochemical fea- tures will very probably further optimize the described algo- rithm. Also, recently discovered markers may prove superior to the existing combinations, such as SATB2 which is a promising new marker with high specificity for gastrointestinal origin [34–36]. This algorithm can be useful for frozen section, al- though strictly for patients with unilateral salpingo- oophorectomy it may be misleading since microscopic involve- ment of the contralateral ovary may not be macroscopically vis- ible preoperatively and therefore prevent bilateral resection. Virchows Arch (2019) 474:289–296 295 In conclusion, our algorithm has a higher sensitivity of 90.1% for diagnosing mMC compared to earlier reported al- gorithms, hereby validating these earlier approaches on a large cohort and adding patient age and tumor histology as contrib- uting factors. Macroscopic and demographic features as pro- posed in the current study strongly aid in decision making, but algorithms as described here should be regarded as helpful rather than conclusive tools. Ultimately, differentiating MOC from mMC is a task beyond the responsibility of the patholo- gist alone and should be based on careful integration of pre- operative workup including imaging and laboratory results and macroscopic, histological, and immunophenotypical tu- mor features and requires accurate and thorough multidisci- plinary communication. Authors’ contributions MS obtained funding. MS and TB contributed to the study design, data analysis, and drafting of the manuscript. AB was responsible for performing the national pathology database search. AH contributed to the data analysis. HB, LM, and IN contributed to the study design and data analysis. All authors reviewed the manuscript and ap- proved the final version. Funding This study was funded by the Dutch Cancer Society (grant number KUN 2014–6613). Compliance with ethical standards Conflicts of interest The authors declare that they have no conflict of interest. Open Access This article is distributed under the terms of the Creative C o m m o n s A t t r i b u t i o n 4 . 0 I n t e r n a t i o n a l L i c e n s e ( h t t p : / / creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. Publisher’s Note Springer Nature remains neutral with regard to juris- dictional claims in published maps and institutional affiliations. References 1. Bruls J, Simons M, Overbeek LI, Bulten J, Massuger LF, Nagtegaal ID (2015) A national population-based study provides insight in the origin of malignancies metastatic to the ovary. Virchows Arch 467: 79–86. https://doi.org/10.1007/s00428-015-1771-2 2. Seidman JD, Kurman RJ, Ronnett BM (2003) Primary and meta- static mucinous adenocarcinomas in the ovaries: incidence in rou- tine practice with a new approach to improve intraoperative diag- nosis. Am J Surg Pathol 27:985–993 3. Moore RG, Chung M, Granai CO, Gajewski W, Steinhoff MM (2004) Incidence of metastasis to the ovaries from nongenital tract primary tumors. Gynecol Oncol 93:87–91. https://doi.org/10.1016/ j.ygyno.2003.12.039 4. Antila R, Jalkanen J, Heikinheimo O (2006) Comparison of sec- ondary and primary ovarian malignancies reveals differences in their pre- and perioperative characteristics. Gynecol Oncol 101: 97–101. https://doi.org/10.1016/j.ygyno.2005.09.046 5. Simons M, Ezendam N, Bulten J, Nagtegaal I, Massuger L (2015) Survival of patients with mucinous ovarian carcinoma and ovarian metastases: a population-based Cancer registry study. Int J Gynecol C a n c e r 2 5 : 1 2 0 8 – 1 2 1 5 . h tt p s : / / d o i . o rg / 1 0 . 1 0 9 7 / I G C . 0000000000000473 6. McCluggage WG, Wilkinson N (2005) Metastatic neoplasms in- volving the ovary: a review with an emphasis on morphological and immunohistochemical features. Histopathology 47:231–247. https://doi.org/10.1111/j.1365-2559.2005.02194.x 7. Lee KR, Young RH (2003) The distinction between primary and metastatic mucinous carcinomas of the ovary: gross and histologic findings in 50 cases. Am J Surg Pathol 27:281–292 8. Riopel MA, Ronnett BM, Kurman RJ (1999) Evaluation of diag- nostic criteria and behavior of ovarian intestinal-type mucinous tumors: atypical proliferative (borderline) tumors and intraepithelial, microinvasive, invasive, and metastatic carcinomas. Am J Surg Pathol 23:617–635 9. Lewis MR, Deavers MT, Silva EG, Malpica A (2006) Ovarian involvement by metastatic colorectal adenocarcinoma: still a diag- nostic challenge. Am J Surg Pathol 30:177–184 10. Vang R, Gown AM, Wu LS, Barry TS, Wheeler DT, Yemelyanova A, Seidman JD, Ronnett BM (2006) Immunohistochemical expres- sion of CDX2 in primary ovarian mucinous tumors and metastatic mucinous carcinomas involving the ovary: comparison with CK20 and correlation with coordinate expression of CK7. Mod Pathol 19: 1421–1428. https://doi.org/10.1038/modpathol.3800698 11. Vang R, Gown AM, Barry TS, Wheeler DT, Yemelyanova A, Seidman JD, Ronnett BM (2006) Cytokeratins 7 and 20 in primary and secondary mucinous tumors of the ovary: analysis of coordi- nate immunohistochemical expression profiles and staining distri- bution in 179 cases. Am J Surg Pathol 30:1130–1139. https://doi. org/10.1097/01.pas.0000213281.43036.bb 12. Groisman GM, Meir A, Sabo E (2004) The value of Cdx2 immu- nostaining in differentiating primary ovarian carcinomas from co- lonic carcinomas metastatic to the ovaries. Int J Gynecol Pathol 23: 52–57. https://doi.org/10.1097/01.pgp.0000101141.31270.a0 13. Werling RW, Yaziji H, Bacchi CE, Gown AM (2003) CDX2, a highly sensitive and specific marker of adenocarcinomas of intesti- nal origin: an immunohistochemical survey of 476 primary and metastatic carcinomas. Am J Surg Pathol 27:303–310 14. Vang R, Gown AM, Zhao C, Barry TS, Isacson C, Richardson MS, Ronnett BM (2007) Ovarian mucinous tumors associated with ma- ture cystic teratomas: morphologic and immunohistochemical anal- ysis identifies a subset of potential teratomatous origin that shares features of lower gastrointestinal tract mucinous tumors more com- monly encountered as secondary tumors in the ovary. Am J Surg P a t h o l 3 1 : 8 5 4– 8 6 9 . h t t p s : / / d o i . o r g / 1 0 . 1 0 9 7 / PA S . 0b013e31802efb45 15. Yemelyanova AV, Vang R, Judson K, Wu LS, Ronnett BM (2008) Distinction of primary and metastatic mucinous tumors involving the ovary: analysis of size and laterality data by primary site with reevaluation of an algorithm for tumor classification. Am J Surg P a t h o l 3 2 : 1 2 8– 1 3 8 . h t t p s : / / d o i . o r g / 1 0 . 1 0 9 7 / PA S . 0b013e3180690d2d 16. Casparie M, Tiebosch AT, Burger G, Blauwgeers H, van de Pol A, van Krieken JH, Meijer GA (2007) Pathology databanking and biobanking in the Netherlands, a central role for PALGA, the na- tionwide histopathology and cytopathology data network and ar- chive. Cell Oncol 29:19–24 17. Vang RCA, Kommoss F, Matias-Guiu X, Ronnett BM, Young RH (2014) Secondary Tumours. In: Kurman RJCM, Herrington CS, Young RH (eds) WHO classification of Tumours of female repro- ductive organs, 4th edn. International Agency for Research on Cancer (IARC), Lyon, pp 83–86 296 18. van Buuren S (2007) Multiple imputation of discrete and continu- ous data by fully conditional specification. Stat Methods Med Res 16:219–242. https://doi.org/10.1177/0962280206074463 19. Khunamornpong S, Suprasert P, Pojchamarnwiputh S, Na Chiangmai W, Settakorn J, Siriaunkgul S (2006) Primary and met- astatic mucinous adenocarcinomas of the ovary: evaluation of the diagnostic approach using tumor size and laterality. Gynecol Oncol 101:152–157. https://doi.org/10.1016/j.ygyno.2005.10.008 20. McCluggage WG, Young RH (2008) Primary ovarian mucinous tumors with signet ring cells: report of 3 cases with discussion of so-called primary Krukenberg tumor. Am J Surg Pathol 32:1373– 1379. https://doi.org/10.1097/PAS.0b013e31816b18c1 Judson K, McCormick C, Vang R, Yemelyanova AV, Wu LS, Bristow RE, Ronnett BM (2008) Women with undiagnosed colo- rectal adenocarcinomas presenting with ovarian metastases: clini- copathologic features and comparison with women having known colorectal adenocarcinomas and ovarian involvement. Int J Gynecol Pathol 27:182–190. https://doi.org/10.1097/PGP. 0b013e31815b9752 21. 22. Ronnett BM, Shmookler BM, Diener-West M, Sugarbaker PH, Kurman RJ (1997) Immunohistochemical evidence supporting the appendiceal origin of pseudomyxoma peritonei in women. Int J Gynecol Pathol 16:1–9 23. Young RH, Gilks CB, Scully RE (1991) Mucinous tumors of the appendix associated with mucinous tumors of the ovary and pseudomyxoma peritonei. A clinicopathological analysis of 22 cases supporting an origin in the appendix. Am J Surg Pathol 15: 415–429 24. Ronnett BM, Seidman JD (2003) Mucinous tumors arising in ovar- ian mature cystic teratomas: relationship to the clinical syndrome of pseudomyxoma peritonei. Am J Surg Pathol 27:650–657 25. Bradley RF, Stewart JH, Russell GB, Levine EA, Geisinger KR (2006) Pseudomyxoma peritonei of appendiceal origin: a clinico- pathologic analysis of 101 patients uniformly treated at a single institution, with literature review. Am J Surg Pathol 30:551–559. https://doi.org/10.1097/01.pas.0000202039.74837.7d 26. Ronnett BM (2006) Pseudomyxoma peritonei: a rose by any other name. Am J Surg Pathol 30:1483–1484; author reply 1484-1485. https://doi.org/10.1097/01.pas.0000213357.18380.36 27. Shetty S, Natarajan B, Thomas P, Govindarajan V, Sharma P, Loggie B (2013) Proposed classification of pseudomyxoma peritonei: influence of signet ring cells on survival. Am Surg 79: 1171–1176 Virchows Arch (2019) 474:289–296 28. Allard MA, Adam R, Giuliante F, Lapointe R, Hubert C, Ijzermans JNM, Mirza DF, Elias D, Laurent C, Gruenberger T, Poston G, Letoublon C, Isoniemi H, Lucidi V, Popescu I, Figueras J (2017) Long-term outcomes of patients with 10 or more colorectal liver metastases. Br J Cancer 117:604–611. https://doi.org/10.1038/bjc. 2017.218 30. 29. Lorincz AT, Nathan M, Reuter C, Warman R, Thaha MA, Sheaff M, Vasiljevic N, Ahmad A, Cuzick J, Sasieni P (2017) Methylation of HPVand a tumor suppressor gene reveals anal cancer and precursor lesions. Oncotarget 8:50510–50520. https://doi.org/10.18632/ oncotarget.17984 Jansen IGH, Mulder M, Goldhoorn RB, investigators MCR (2018) Endovascular treatment for acute ischaemic stroke in routine clini- cal practice: prospective, observational cohort study (MR CLEAN registry). BMJ 360:k949. https://doi.org/10.1136/bmj.k949 Janssen KJ, Donders AR, Harrell FE Jr, Vergouwe Y, Chen Q, Grobbee DE, Moons KG (2010) Missing covariate data in medical research: to impute is better than to ignore. J Clin Epidemiol 63: 721–727. https://doi.org/10.1016/j.jclinepi.2009.12.008 31. 32. Donders AR, van der Heijden GJ, Stijnen T, Moons KG (2006) Review: a gentle introduction to imputation of missing values. J Clin Epidemiol 59:1087–1091. https://doi.org/10.1016/j.jclinepi. 2006.01.014 33. Lang KM, Little TD (2018) Principled missing data treatments. Prev Sci 19:284–294. https://doi.org/10.1007/s11121-016-0644-5 34. Moh M, Krings G, Ates D, Aysal A, Kim GE, Rabban JT (2016) SATB2 expression distinguishes ovarian metastases of colorectal and Appendiceal origin from primary ovarian tumors of mucinous or Endometrioid type. Am J Surg Pathol 40:419–432. https://doi. org/10.1097/PAS.0000000000000553 35. Strickland S, Wasserman JK, Giassi A, Djordjevic B, Parra-Herran C (2016) Immunohistochemistry in the diagnosis of mucinous neo- plasms involving the ovary: the added value of SATB2 and bio- marker discovery through protein expression database mining. Int J Gynecol Pathol 35:191–208. https://doi.org/10.1097/PGP. 0000000000000238 36. Perez Montiel D, Arispe Angulo K, Cantu-de Leon D, Bornstein Quevedo L, Chanona Vilchis J, Herrera Montalvo L (2015) The value of SATB2 in the differential diagnosis of intestinal-type mucinous tumors of the ovary: primary vs metastatic. Ann Diagn Pathol 19: 249–252. https://doi.org/10.1016/j.anndiagpath.2015.05.004
null
10.1088_1402-4896_ad07c3.pdf
Data availability statement The data cannot be made publicly available upon publication because they contain sensitive personal information. The data that support the findings of this study are available upon reasonable request from the authors.
Data availability statement The data cannot be made publicly available upon publication because they contain sensitive personal information. The data that support the findings of this study are available upon reasonable request from the authors.
Phys. Scr. 98 (2023) 125924 https://doi.org/10.1088/1402-4896/ad07c3 RECEIVED 25 August 2023 REVISED 24 October 2023 ACCEPTED FOR PUBLICATION 27 October 2023 PUBLISHED 7 November 2023 PAPER Preparation and properties of MAO self-healing anticorrosion film on 5B70 Al alloy Mingjin Wu1 and Feng Jiang1,2,∗ 1 Light Alloy Research Institute, Central South University, Changsha, 410083, People’s Republic of China 2 School of Material Science and Engineering, Central South University, Changsha, 410083, People’s Republic of China ∗ Author to whom any correspondence should be addressed. E-mail: jfeng2@csu.edu.cn Keywords: 5B70 Al alloy, MAO, corrosion inhibitor, anti-corrosion performance, self-healing performance Abstract Micro-arc oxidation (MAO) was a new surface treatment technology for Al alloys. However, the MAO ceramic film could not provide long-term protective performance owing to its inherent porous structure. In this work, a new type of CeO2-sealed GO/Al2O3 composite film on 5B70 Al alloy with excellent corrosion resistance was prepared by integrating MAO and the hole sealing technique. The experimental results indicated that the loose and porous MAO ceramic film could serve as a ‘shield’ and a ‘reservoir’, respectively, to obtain improved impedance and sufficient corrosion inhibitor loading. Compared to the original MAO ceramic film or GO/Al2O3 composite film, The GO/Al2O3 composite film after sealing treatment for 60 min had lower porosity and better corrosion resistance. In addition, CeO2-sealed GO/Al2O3 composite film exhibited a positive self-healing effect in 3.5 wt% NaCl solution. 1. Introduction Al alloys were widely used in civil and aerospace fields due to their low density and high specific strength. For example, the high-strength 5B70 Al alloy with light weight and weldability could be used to manufacture main structures of the large sealed cabin of manned spacecraft [1, 2]. However, the wider application Al alloys were restricted due to the high chemical reactivity and low anti-corrosion performance. Many surface treatment technologies, such as thermal spraying [3], physical vapor deposition [4], chemical vapor deposition [5], laser cladding [6] and MAO [7], have been used to improve the corrosion resistance of Al alloys. Among these technologies, MAO was considered as a high productivity and eco-friendly technology, which could provide a hard ceramic film with strong adhesion and medium corrosion resistance. However, the traditional MAO ceramic film with porous structure could not meet the requirements of anti-corrosion performance, especially in extreme circumstances. To improve anti-corrosion performance of MAO ceramic film, the nanoparticles with special −). It has been confirmed that GO was one of the best corrosion-resistant materials [8]. characteristics, such as ZrO2, CeO2, TiO2 and so on, were added to the electrolyte solution. Recently, we learned that the two-dimensional material-graphite oxide (GO) had strong obstruct capability to molecules (O2 and H2O) and ions (Cl Therefore, we prepared GO/Al2O3 composite film on 5B70 Al alloy by MAO in the silicate electrolyte solution with GO nanoparticles. The results suggested that the addition of GO nanoparticles could reduce the pores and microcracks generated by MAO process, and improve the density of ceramic film. At the same time, GO would form a barrier network in MAO ceramic film to prevent the penetration of corrosive media, and finally enhance the corrosion resistance of MAO ceramic film. However, there were still a small number of micropores and − cracks on the surface of the GO/Al2O3 composite film, which could easily provide a diffusion path for Cl and O2 in corrosive environments. In long-term aggressive environment, once the ceramic film was damaged, the substrate might be susceptible to local corrosion, causing the occurrence of pits, gaps, and the degradation in the composite film [9], ultimately causing film detachment and Al alloy workpiece failure. © 2023 IOP Publishing Ltd Phys. Scr. 98 (2023) 125924 M Wu and F Jiang At present, many researchers were particularly concerned about obtaining a robust self-healing composite film with functional surfaces through defect sealing, which could effectively prevent the occurrence and further expansion of corrosion behavior in damaged region of Al alloys [10, 11]. In terms of GO/Al2O3 composite film, the micropores served as carriers for inhibitors, and corrosion inhibitors could fill the pores to seal the GO/Al2O3 composite film. Gnedenkov et al [12] immersed the MAO ceramic film in a solution containing 8-hydroxyquinoline, which could reduce the current density of the composite film in 3.5 wt% NaCl solution by 30 times and avoid serious damage to the substrate. Liu et al [13] incorporated 2-mercaptobenzothiazole and Na3PO4 into the MAO film, their synergistic effects could effectively enhanced the self-healing performance of the composite film in 3.5 wt% NaCl solution. To further improve the application scope of 5B70 Al alloy, it was important to enhance its anti-corrosion performance or induce its self-healing performance. Nevertheless, there was little research on the self-healing and anti-corrosion MAO films on 5B70 Al alloy. According to reports, Ce-based solutions could introduce corrosion inhibitors into MAO ceramic film to fill the pores [14]. Therefore, in this paper, the self-healing and anti-corrosion film based on MAO film on 5B70 Al alloy with the corrosion inhibitor was prepared by immersing into the sealing solution. The corrosion resistance and self-healing performance of this sealed film in 3.5 wt% NaCl solution was investigated. 2. Experimental section 2.1. Materials The 5B70 Al alloy used in this experiment (Mg:6.02wt%, Sc:0.25wt%, Mn:0.32wt%, Mn:0.1wt%, and Al balance) was provided by Northeast Light Alloy Co., Ltd First, the 5B70 Al alloy sheet was cut into 20 mm × 20 mm × 3 mm samples. These samples were mechanically ground with 180#, 800#, and 1200# SiC sandpaper, polished to a mirror surface, washed with deionized water, and finally dried in flowing air. 2.2. Preparation of the composite film Direct-current power supply (Yisheng Electronic Technology Co., Ltd, China) was used for MAO treatment. The alloy samples served as the anode, and the stainless steel container served as the cathode. The electrolyte solution was composed of Na2SiO3(10 g l −1. GO nanoparticles were purchased Biochemical Co., Ltd The concentration of GO nanoparticles was 0.15 g l from Shenzhen Huiheng Technology Co., Ltd. Before adding to the electrolyte solution, GO nanoparticles were dispersed in deionized water by ultrasonic treatment for 20 min. A constant current mode was used in the MAO process. The current density was 10 A cm was maintained below 30 °C. After MAO treatment, the samples were washed with deionized water and dried in flowing air. −2, and the oxidation time was 15 min. The electrolyte temperature −1), purchased from Shanghai Macklin −1) and NaOH(1 g l 2.3. Micropores sealing process The composite samples were immersed in a sealing solution for the holes sealing. The sealing solution was prepared by dissolving Ce(NO3)3 and H2O2 in deionized water at room temperature and continuously stirring mechanically. The concentration of Ce(NO3)3 and H2O2 were 6 g l addition, these composite samples were immersed for 0 min, 10 min, 30 min, 60 min, and 90 min, respectively, to load corrosion inhibitors, which could meet the requirements of the micropores sealing process (for convenience, the corresponding samples were named MAO-S0, MAO-S10, MAO-S30, MAO-S60, and MAO- S90, respectively). After sealing treatment, these samples were washed with deionized water and dried in air. The entire experimental process was shown in figure 1. −1 [14], respectively. In −1 and 30 ml l 2.4. Electrochemical test Electrochemical workstation (MULTIAUTOLABM204, Switzerland) was used to test the potentiodynamic polarization curve and electrochemical impedance spectroscopy (EIS) of the composite sample and the sealed samples to characterize its electrochemical characteristics in 3.5 wt% NaCl solution. A typical three electrode device was used, in which a saturated calomel electrode (SCE) was used as the reference electrode, a platinum electrode was used as the counter electrode, and the MAO sample with an exposure area of 1cm2 was used as the working electrode. To get a stable open circuit potential, the sample needed to be immersion for 15 min in 3.5 wt % NaCl solution before testing. The frequency range measured by EIS was 105 Hz to 0.1 Hz, with an amplitude of 10 mV. The EIS data was fitted using ZSimDemo software. During the measurement of polarization curves, the scanning speed was designed to 1 mV s three times to ensure the data repeatability. −1. The above tests should be conducted at least 2 Phys. Scr. 98 (2023) 125924 M Wu and F Jiang Figure 1. The experimental process. 2.5. Scratching test To evaluate the self-healing performance of the sealing films, artificial defects were created by scratching the sealing film with a sharp knife. The scratches were observed with an optical microscope to ensure the presence of Al substrate in the visual field. Each sealing sample with artificial scratches were immersed in a 3.5 wt % NaCl solution for 6 h and 12 h, respectively, to simulate the damage of Al alloy workpieces in the actual working condition. Considering the possible failure of the sealing film, the appropriate self-healing time in 3.5 wt% NaCl solution was determined. The optical photos, corrosion morphologies, and corresponding EDS spectra after immersion were recorded. 2.6. Characterization The surface roughness and three-dimensional morphology of composite sample and sealed samples were measured the optical profilometer (Bruker, ContourX-200, USA). To ensure accuracy, three measurements on the roughness of each sample were conducted to obtain an average value. Field emission scanning electron microscope (FESEM, JEOL, Japan) equipped with energy dispersive x-ray spectroscopy (EDS) equipment was used to observe the surface morphologies of composite sample and sealed samples. In addition, the elements distribution on the sealing sample surface was also detected. The porosity of the composite sample and sealing samples was measured using Image-pro plus software. The phase composition of the composite sample and sealed samples was detected by x-ray diffractometer (XRD, Rigaku, Japan), using Cu-Kα radiation at 30 kV and 20 mA. Diffraction data were obtained at a scattering angle 2θ from 10° to 80°, with a scanning speed of 2°/min. The chemical state of elements was detected by x-ray photoelectron spectroscopy (XPS, AXIS Supra, UK) equipped with monochromatic Al-Kα radiation sources (6 mA, 12 kV, and 1486.68 eV). 3. Results and discussion 3.1. Composition analysis To determine the chemical composition of the sealing films, XRD and XPS characterization methods were used. Figure 2 showed the XRD spectra of the composite films after sealing treatment. All samples were composed of γ-Al2O3 and α-Al2O3 and there was no significant difference in phase composition. In order to investigate the chemical state of surface elements on the sealing films, XPS analysis was performed on the MAO-S30 sample, and the results were shown in figure 3. The wide spectrum of the MAO-S30 sample confirmed the presence of Al, Ce, and O. The high-resolution spectrum of Al 2p (figure 3(b)) showed that the binding energy of Al 2p was about 74.7 eV, which belonged to Al2O3. MAO-S30 sample showed obvious Ce 3d peak (figure 3(c)), which indicated that the corresponding compounds had successfully entered the GO/Al2O3 composite film. To be exact, the Ce 3d spectrum was fitted through XPS simulation, and it was found that the original spectrum was highly consistent with the standard specification spectrum of Ce 3d4+ indicated that the corrosion inhibitor contained Ce4+ 898.1 eV, 901.1 eV, 905.9 eV and 916.7 eV indicated the presence of CeO2 in the sealing samples [15]. The high- resolution spectrum of O1s (figure 3(d)) showed that two peaks at 531.05 eV and 532.3 eV, corresponded to the binding energy of Al2O3 and CeO2, respectively. Therefore, it could be concluded that the corrosion inhibitor, which was produced in GO/Al2O3 composite film after sealing treatment, was CeO2. The chemical reactions were as followed: , which . The peaks at the binding energy of 882.5 eV, 886.8 eV, 3 + 2Ce + H O 2 2 + 2H O 2  ( 2Ce OH ) + 2 2 + + 2H ( ) 1 3 Phys. Scr. 98 (2023) 125924 M Wu and F Jiang Figure 2. XRD spectra of the composite films after sealing treatment. Figure 3. (a) Wide scan XPS spectrum and XPS high-resolution spectra of b) Al, c) Ce, and d) O of the MAO-S30 sample. 2H O O + 2 + 4e - - 4OH 2 ( Ce OH ) + 2 2 + - 2OH  ( Ce OH ) 4  CeO 2 + 2H O 2 ( ) 2 ( ) 3 3.2. Morphology observation The optical morphologies of the composite sample and sealing samples were shown in figure 4. From these figures, the color changes of the samples surface could be clearly observed. It was obvious that with the extension of sealing time, the color of these samples surface gradually deepened and changed from gray white to yellow brown. As was well known, Ce4+ [16]. Therefore, it could be reasonably inferred that as the sealing time increasing, the corrosion inhibitors ions could cause solutions or compounds to take on a yellow based appearance 4 Phys. Scr. 98 (2023) 125924 M Wu and F Jiang Figure 4. Optical morphologies of the composite sample ((a)MAO-S0) and the sealing samples ((b) MAO-S10, (c) MAO-S30, (d) MAO-S60 and (e) MAO-S90). should gradually form in the GO/Al2O3 composite film, and the color changes were related to the enrichment of Ce4+ based compounds. The SEM images of GO/Al2O3 composite film and four sealing films were shown in figure 5. From the figures, the surface morphology of MAO-S0 sample exhibited microcracks and pores, which were attributed to the release of gas in the molten oxide and the thermal stress present at the discharge channel during the formation process of GO/Al2O3 composite film. And there were many micropores with larger sizes on the samples surface. Specifically, the surface porosity was 6.09%, and the average size of the micropores was 6.96 μm. When the composite samples were sealed for 10 min, 30 min and 60 min, the surface morphologies were significantly different from that of composite sample, and the porosity of sealing films was 5.99%, 4.16% and 3.93%, respectively. Average pore diameter significantly decreased with the increase of sealing time. This was principally because the micro-pores in ceramic films were filled with corrosion inhibitors. However, after the sealing treatment for 90 min, the surface porosity increased to 4.83%. This was mainly because the formation and propagation of cracks led to an increase in the surface porosity of ceramic film. The chemical reactions involved in the formation of corrosion inhibitors generated thermal stress, and cracks were usually formed at these points. Stress concentration could cause local stresses in ceramic film to exceed their load-bearing capacity, leading to the formation of the cracks, which could propagate under load or stress. The crack propagation could lead to the detachment or destruction of the ceramic films, forming new micro-pores, which in turn led to an increase in surface porosity. Therefore, we found that the porosity did not decrease with the increase of sealing time, which indicated that there was a dynamic equilibrium relationship between the formation and dissolution of corrosion inhibitor. In order to elucidate the dispersion state of corrosion inhibitors in the sealing film, EDS elemental analysis was performed on the sealed samples surface. The elemental distribution of Al, O, Ce, and Si was shown in the figure. It was worth noting that Al and O elements were uniformly distributed on the surface, indicating that Al2O3 was the main phase in the sealed film. Si element was mainly distributed in the protrusion around the discharge hole, and Ce element was locally enriched to fill the micropores to seal the composite film, which indicated that the micropores of composite film became a container for carrying corrosion inhibitor. The three-dimensional morphologies and surface roughness of the GO/Al2O3 composite film and sealing films were shown in figure 6. As shown in the figure, with the extension of sealing time, the surface roughness of the sealing samples gradually increased. This phenomenon was attributed to the fact that the prolonged sealing time led to the generation and deposition of more corrosion inhibitors. The cross-sectional morphologies of GO/Al2O3 composite film and the sealing films were shown in figure 7. It could be seen from the figure that there was an obvious interface between the oxide film and the 5B70 Al alloy substrate. The outer layer of GO/Al2O3 composite film was loose and porous, and the inner layer was compact. After being sealed with the sealing solution for 10–60 min, the cross-section of the sealing films did not show obvious changes. However, after being sealed for 90 min, the obvious cracks occurred on the cross-section, which had a negative impact on the anti-corrosion performance of the sealing films. 3.3. Corrosion evaluation Figure 8(a) showed the polarization curves of GO/Al2O3 composite film and the sealed films after immersion in a 3.5 wt % NaCl solution for 15 min. As shown in the figure, the presence of corrosion inhibitor improved the corrosion resistance of GO/Al2O3 composite film. With the extension of the sealing time, the corresponding curve moved to the left (lower current density) and upward (higher corrosion potential), suggesting that the corrosion resistance of GO/Al2O3 composite film was improved by increasing the sealing time. The fitting data of the polarization curves was displayed in table 1. As seen from the table, the corrosion inhibitor in the pores was beneficial for improving the anti-corrosion performance of GO/Al2O3 composite film. Compared with the MAO-S0 sample, when the sealing time was 60 min, the icorr of the sealing samples increased from 3.72 × −2 and Ecorr increased from −0.590 V to 10 –8A cm2 reduced by an order of magnitude to 1.04 × 10 −9 A cm 5 Phys. Scr. 98 (2023) 125924 M Wu and F Jiang Figure 5. Surface morphologies and the EDS elemental mapping of the samples. (a) MAO-S0, (b) MAO-S10, (c) MAO-S30, (d) MAO- S60 and (e) MAO-S90. −0.528 V, indicating that the corrosion resistance of GO/Al2O3 composite film was increased evidently. The polarization resistance (Rp) and corrosion rate (CR) were calculated using the Stern-Gray equation [17]. The results showed that with the extension of sealing time, Rp increased and CR decreased. The MAO-S60 sample had the optimal corrosion resistance, with Rp = 8.656 × 105 Ω cm2, CR = 1.16 × 10 −6 A cm −2. The Nyquist plots of the GO/Al2O3 composite film and the sealing films were displayed in figure 8(b). It was found that these plots presented a distinct characteristic, which was the low capacitance semicircle that constituting the impedance spectrum. The obvious difference between the five spectra was their arc radius, which meant that the corrosion behavior of the five samples were different. Based on previous research findings, the larger the radius was, the better the corrosion resistance was [18]. When the sealing time was less than 60 min, the curve radius gradually increased with the increase of sealing time. Therefore, the corrosion resistance of these sealing samples gradually increased. At the same time, the change also meant that the charge transfer process was suppressed, and the corrosion inhibitor was effectively loaded into the GO/Al2O3 composite film. As the sealing time prolonged, more Ce4+ ions entered the GO/Al2O3 composite film to form 6 Phys. Scr. 98 (2023) 125924 M Wu and F Jiang Figure 6. 3D morphologies and surface roughness of the GO/Al2O3 composite film(a) and the sealing films (b) MAO-S10, (c) MAO- S30, (d) MAO-S60 and (e) MAO-S90. Figure 7. The cross-sectional morphologies of the GO/Al2O3 composite film after sealing treatment. (a) MAO-S10, (b) MAO-S30, (c) MAO-S60 and (d) MAO-S90. more corrosion inhibitors, leading to a gradual improvement of the sealing effect. When the sealing time was 90 min, the corrosion resistance took on a descendant trend. This phenomenon was attributed to that the dissolution behavior was dominant during the formation process of the sealing film. In addition, the Bode impedance plots and phase plots of GO/Al2O3 composite film and sealing films were displayed in figures 8(c) and (d), respectively. In the low-frequency region, the sealing films with higher |Z| always had better anti-corrosion performance. Obviously, the MAO-S60 sample exhibited the best corrosion resistance. From the analysis of figure 8(d), two time-constants could be observed in the Bode phase plots of the MAO-S0 and MAO-S90 samples, and thus the equivalent circuits shown in figures 8(a) and (c) were used for fitting. while the one time-constant was observed in the Bode phase plots of other sealing samples, and thus the equivalent circuit shown in figure 8(b) was used for fitting. Rs referred to the resistance of the corrosion solution. Previous studies have shown that MAO ceramic films were composed of an external porous layer and an internal dense layer [7]. Therefore, For the equivalent circuit of the MAO-S0 sample (figure 9(a)), the electrical elements were composed of the resistance elements (R1 for porous layer, R2 for dense layer), and constant phase element (Q1 and n1 for porous layer, Q2 and n2 for dense layer). For the equivalent circuit of the MAO-S10, MAO-S30 and MAO-S60 sample (figure 9(b)), Rm referred to the resistance of the sealing film, Qm referred to the capacitance of the sealing film. For the equivalent circuit of MAO-S90 sample (figure 9(c)), the electrical 7 Phys. Scr. 98 (2023) 125924 M Wu and F Jiang Figure 8. (a) Potentiodynamic polarization curves, (b) Nyquist plot, (c) Bode impedance plot, (d) Bode phase plot for the GO/Al2O3 composite sample and the four Sealing samples. elements consisted of the resistance elements (Rf for the sealed film, Ri for the interface between film and substrate), and constant phase element (Qf and nf for the sealed film, Qi and ni for the interface between film and substrate. The parameters obtained from the corresponding equivalent circuit models were shown in table 2. The Bode phase plots of MAO-S10, MAO-S30 and MAO-S60 samples only presented a time-constant. This was because the corrosion inhibitor was successfully introduced into the GO/Al2O3 composite film, improving the anti-corrosion performance of the porous layer and effectively suppressing the invasion of corrosive media. When the sealing time reached 90 min, the Bode phase plot of the MAO-S90 sample once again presented two time-constants. Based on the morphology observation in figure 5(d), long immersion in the sealing solution would cause more cracks to generate on the GO/Al2O3 composite film surface, thereby accelerating the corrosion reaction. In summary, electrochemical measurements and SEM images demonstrated that the hole sealing technique could improve the corrosion resistance of GO/Al2O3 composite films. In addition, the corrosion resistance of sealing film achieved the desired effect after the sealing treatment for 60 min. R P = b a 2.3 ⋅ i corr c ⋅ b b ( a + b c ) C R = 22.85 ⋅ i corr ( ) 4 ( ) 5 Previous studies have shown that GO nanoparticles as an electrolyte additive could change the characteristic of the electrolyte solution and affect the MAO process. In addition, GO nanoparticles had been proved to have high aspect ratio and permeability resistance. The MAO process could effectively wrap GO nanoparticles in the MAO ceramic film, which made the conduction path of corrosion medium in the GO/Al2O3 composite film more tortuous, giving the GO/Al2O3 composite film a labyrinth effect. Finally, the corrosion resistance of the GO/Al2O3 composite film was significantly improved. When the GO/Al2O3 composite film was immersed in the sealing solution, under the action of mechanical agitation, the corrosion inhibitor (CeO2) was formed in the loose structure to fill the micropores and ultimately seal GO/Al2O3 composite film. Meanwhile, GO was lamellar, and had high transparency and wrinkled edges. This undoubtedly indicated that GO could be applied as an excellent support material. CeO2 was adsorbed on the GO matrix to nucleate and grow in different directions to obtain thin sheets with poor crystallinity and small transverse size, which could act as a barrier layer to prevent the invasion of corrosive media, and to some extent, protect 5B70 Al alloy. Meanwhile, CeO2 deposited on the lamellar structure of GO would maintain the high surface and inherent folding properties of GO. In addition, it also exhibited a thicker crystal sheet-like structure, 8 Table 1. Electrochemical data obtained from potentiodynamic polarization tests. 9 Samples Corrosion potential Ecorr/V(SCE) Current density icorr/ (A·cm −2) Anodic slope βa(V/dec.) Cathodic slope -βc(V/dec.) Polarization resistance Rp(Ω·cm2) Corrosion rate CR(A·cm −2) MAO-S0 MAO-S10 MAO-S30 MAO-S60 MAO-S90 −0.590 −0.581 −0.556 −0.528 −0.547 3.72E-8 5.79E-9 2.43E-9 1.04E-9 2.39E-9 0.182 0.175 0.219 0.181 0.155 0.215 0.167 0.166 0.167 0.213 1.15E6 6.42E6 1.69E7 3.63E7 1.63E7 8.50E-7 1.32E-7 5.55E-8 2.38E-8 5.46E-8 P h y s . S c r . 9 8 ( 2 0 2 3 ) 1 2 5 9 2 4 M W u a n d F J i a n g Table 2. The values of electrical element of equivalent data. 1 0 Sample Rs(Ω/cm2) R1(Ω/cm2) Q1(F/cm2) n1 R2(Ω/cm2) Q2(F/cm2) n2 Rm(Ω/cm2) Qm(F/cm2) MAO-S0 MAO-S10 MAO-S30 MAO-S60 MAO-S90 13.93 71.01 11.96 21.80 6.229 6.118E6 — — — — 2.225E-7 — — — — 0.7472 — — — — 8.113E6 — — — — 4.858E-7 — — — — 0.8999 — — — — — 1.126E7 3.008E7 1.519E8 — — 1.759E-7 2.889E-8 1.854E-8 — nm — 0.851 0.7935 0.7404 — Rf(Ω/cm2) Qf(F/cm2) — — — — — — — — nf — — — — Ri(Ω/cm2) Qi(F/cm2) — — — — — — — — ni — — — — 2.516E7 3.354E-7 0.5555 1.678E7 1.252E-7 0.8185 P h y s . S c r . 9 8 ( 2 0 2 3 ) 1 2 5 9 2 4 M W u a n d F J i a n g Phys. Scr. 98 (2023) 125924 M Wu and F Jiang Figure 9. Equivalent electrical circuits used for fitting the EIS data of (a) MAO-S0 sample, (b) MAO-S10, MAO-S30 and MAO-S60 sample, (c) MAO-S90 sample. Figure 10. Corrosion protection mechanism for the sealed films on 5B70 Al alloy. Figure 11. Optical morphologies of the scratches on the MAO-S60 sample after immersion for different times in 3.5 wt% NaCl solution. (a) 0 h, (b) 6 h and (c) 12 h. providing greater possibilities for extending the invasion path of corrosive media. The corrosion mechanism − diagram was shown in figure 10. As the sealing time prolonged, the generated OH effect on the alumina based sealing film, thereby reducing the integrity and anti-corrosion performance of the sealing samples during long-term sealing treatment. ions might have a dissolution 3.4. Scratching test The self-healing performance of MAO-S60 sample with the best sealing effect in 3.5 wt% NaCl solution was evaluated. Optical morphologies of the MAO-S60 sample with the scratches after immersion for different times were displayed in figure 11. It was noteworthy that no obvious self-healing morphology was observed on the MAO-S60 sample surface after immersion for 6 h. However, the scratches on the sample surface could be effectively healed in a 3.5 wt% NaCl solution after immersion for 12 h, indicating the generation of self-healing 11 Phys. Scr. 98 (2023) 125924 M Wu and F Jiang Figure 12. Potentiometric polarization curves of the MAO-S60 sample with scratches after self-healing treatment in 3.5 wt% NaCl solution for different times. Figure 13. Surface morphologies of the scratch area after self-sealing treatment in a 3.5 wt% NaCl solution: the SEM image of the scratch area after self-sealing treatment for 6 h (a), the locally enlarged view of figure 13(a), (b), the locally enlarged view of figures 13(b), (c); the EDS elemental mapping figure 13(c): (d) Al map; (e) O map; the EDS element analysis of P1(f) and P2(g) in figure 13(c); the SEM images of the scratch area after self-sealing treatment for 12 h (h), the locally enlarged view of figure 13(h), (i), and the locally enlarged view of figure 13(i) (j); the EDS elemental mapping figure 13(j): (k) Al map; (l) O map; the EDS element analysis of P3(m) and P4(n) in figure 13(j). products here. In general, as you could see from the optical morphologies of scratches, the sealing film on the MAO-S60 sample surface had good self-healing effect after immersion in 3.5 wt% NaCl solution for 12 h. In order to further determine the self-healing effect, electrochemical tests were conducted on MAO-S60 sample with scratches. Figure 12 displayed the potentiodynamic polarization curves of MAO-S60 samples with scratches after self-healing treatment in a 3.5 wt% NaCl solution for different time. The fitting results were shown in table 3. The Ecorr of MAO-S60 sample without self-sealing treatment was −0.790 V, and the icorr was −2. However, after immersion in 3.5 wt% NaCl solution for 12 h, the Ecorr was increased to 4.29 × 10 −2. The improvement of anti-corrosion performance −0.647 V and the icorr was reduced to 2.92 × 10 proved that scratch area was effectively repaired after self-healing treatment in 3.5 wt% NaCl solution. That was to say, self-healing products should help improve corrosion resistance, which confirmed that the sealing film has self-healing ability in 3.5 wt% NaCl solution. −8 A cm −8A·cm Figure 13 showed the surface morphologies and the EDS element analysis of scratch area after self-sealing treatment for 6 h and 12 h in 3.5 wt% NaCl solution. In figure 13(a), EDS analysis showed that there was no significant self-healing effect, and a small amount of oxide or hydroxide of Al was generated in the scratch area. In figure 13(b), the scratched area presented varying degrees of expansion and coverage of corrosion products. EDS element analysis suggested that pitting corrosion preferentially occurred around Al3(Sc, Zr) particles in the scratch area. There was a large amount of oxide or hydroxide of Al around the corrosion pit. Hence, the 12 1 3 Table 3. Polarization fitting results of the MAO-S60 sample with scratches after self-healing treatment in 3.5 wt% NaCl solution for different times. Immersion time/h Corrosion potential Ecorr/V(SCE) Current density icorr/ (A·cm −2) Anodic slope βa(V/dec.) Cathodic slope -βc(V/dec.) Polarization resistance Rp(Ω·cm2) Corrosion rate CR(A·cm −2) 0 6 12 −0.790 −0.825 −0.647 4.29E-8 5.25E-6 2.92E-9 0.198 0.194 0.168 0.138 0.178 0.142 1.01E7 8.28E5 1.49E8 9.80E-7 1.19E-4 6.67E-8 P h y s . S c r . 9 8 ( 2 0 2 3 ) 1 2 5 9 2 4 M W u a n d F J i a n g Phys. Scr. 98 (2023) 125924 M Wu and F Jiang Figure 14. XPS on the scratch area on MAO-S60 sample after immersion for 12 h in 3.5%wt NaCl solution. appearance of corrosion morphology indicated that the pitting corrosion had an adverse effect on the self- healing performance of sealing film. The scratch area on MAO-S60 sample after immersion for 12 h was detected through XPS technique to determine the composition of self-healing product. As shown from the high-resolution spectrum of Ce 3d, Al 2p and O1s in figure 14, the Ce 3d peak of the scratch area on MAO-S60 sample was similar to that of the sample without scratches. This meant that CeO2 was not involved in the self-healing reaction in 3.5 wt% NaCl solution. The binding energy of Al 2p was about 75.0 eV, 74.4 eV, corresponding to Al2O3 and Al(OH)3, respectively. The O1s peak could be divided into two peaks, corresponding to Al(OH)3 at 531.4 eV and Al2O3 at 532.2 eV, respectively. It could be inferred that the exposed Al substrate could react with O2 and H2O to form Al(OH)3 to repair the film in the scratch area. This dense insoluble precipitate covered the entire scratch area after immersion for 12 h, delaying the invasion of corrosion ions into the alloy surface and achieving self-healing effect. 4. Conclusion (1) GO could better fix the corrosion inhibitor through its inherent lamellar structure and fold state. CeO2 grew readily along the lamellar structure of GO to obtain a thicker lamellar structure, effectively extending the diffusion path of the corrosive medium, thereby preventing the leakage of corrosion inhibitor, and enhancing the corrosion resistance of the sealing films. −8 A cm (2) The GO/Al2O3 composite film after sealing treatment for 60 min had the best corrosion resistance, which was mainly manifested as: the RP and CR of potentiometric polarization curve were 3.63 × 107 Ω cm2 and −2, respectively. The impedance values of the sealing film in EIS were 1.519 × 108 Ω cm 2.38 × 10 −2. (3) Scratch experiments showed that CeO2-sealed GO/Al2O3 composite film had self-healing properties, which could automatically and independently repair defects partially, thereby extending the service life of the workpiece. Acknowledgments This work was supported by the Fundamental Research Funds for the Central Universities of Central South University (NO.2023ZZTS0360). Data availability statement The data cannot be made publicly available upon publication because they contain sensitive personal information. The data that support the findings of this study are available upon reasonable request from the authors. Declaration of interests The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. 14 Phys. Scr. 98 (2023) 125924 ORCID iDs Feng Jiang https://orcid.org/0000-0003-0239-259X References M Wu and F Jiang [1] Wu M and Jiang F 2021 Investigation of alloying element Mg in the near surface layer of micro-arc oxidation coating on Al-Mg-Sc alloy Vacuum 197 110823 [2] Zhongqin Tang F J, Long M, Jiang J, Liu H and Tong M 2020 Effect of annealing temperature on microstructure perties and corrosion behavior of Al-Mg-Mn-Sc-Zr alloy Appl. Surf. Sci. 514 146081 [3] Sakata K et al 2013 Development of nanoporous alumina catalyst support by anodic oxidation of thermally and kinetically sprayed aluminum coatings J. Therm. Spray Technol. 22 138–44 [4] Bashir M I et al 2017 Enhanced surface properties of aluminum by PVD-TiN coating combined with cathodic cage plasma nitriding Surf. Coat. Technol. 327 59–65 [5] Sanjay S and Baskar K 2018 Fabrication of Schottky barrier diodes on clump of gallium nitride nanowires grown by chemical vapour deposition Appl. Surf. Sci. 456 526–31 [6] Li Y and Shi Y 2020 Microstructure and wear resistance of the laser-cladded Al0.8CrFeCoNiCu0.5Bx high-entropy alloy coating on aluminum Mater. Res. Express 7 026517 [7] Wu M and Jiang F 2023 Effect of Na2SiO3 concentration on corrosion resistance and wear resistance of MAO ceramic film on the Al- Mg-Sc alloy Int. J. Appl. Ceram. Technol. 20 1828–45 [8] Askarnia R et al 2022 Effect of graphene oxide on properties of AZ91 magnesium alloys coating developed by micro-arc oxidation process J. Alloys Compd. 892 162106 [9] Yan L et al 2019 One-step in situ synthesis of reduced graphene oxide/Zn-Al Layered double hydroxide film for enhanced corrosion protection of magnesium alloys Langmuir 19 6312–20 [10] Wang T et al 2019 Triple-stimuli-responsive smart nanocontainers enhanced self-healing anticorrosion coatings for protection of aluminum alloy ACS Appl. Mater. Interfaces 11 4425–38 [11] Manasa S et al 2017 Effect of inhibitor loading into nanocontainer additives of self-healing corrosion protection coatings on aluminum alloy A356 J. Alloys Compd. 726 969–77 [12] Gnedenkov A S et al 2016 Localized corrosion of the Mg alloys with inhibitor-containing coatings: SVET and SIET studies Corros. Sci. 102 269–78 [13] Liu D et al 2019 Enhancing the self-healing property by adding the synergetic corrosion inhibitors of Na3PO4 and 2-mercaptobenzothiazole into the coating of Mg alloy Electrochim. Acta 323 134796 [14] Gong Y et al 2021 Self-healing performance and corrosion resistance of novel CeO2-sealed MAO film on aluminum alloy Surf. Coat. Technol. 417 127208 [15] Bêche E et al 2008 Ce 3d XPS investigation of cerium oxides and mixed cerium oxide (CexTiyOz) Surf. Interface Anal. 40 264–7 [16] Sykora R E et al 2004 Isolation of intermediate-valent Ce (III)/Ce (IV) hydrolysis products in the preparation of cerium iodates: electronic and structural aspects of Ce2 (IO3) 6 (OH x)(x ≈ 0 and 0.44) Chem. Mater. 16 1343–9 [17] Wu M, Jiang F and Jiang J 2022 Effect of Na2SiO3 concentration on microstructure and corrosion resistance of MAO coatings prepared on Al-Mg-Sc alloys Anti-Corrosion Methods and Materials 69 417–25 [18] Deng Y et al 2020 Comparative investigations on the electrochemical behaviors among Al and aluminum alloys Mater. Res. Express 7 116510 15
null
10.3390_ijms241210388.pdf
Data Availability Statement: Not applicable.
Data Availability Statement: Not applicable.
Article Quantitative Loop-Mediated Isothermal Amplification Detection of Ustilaginoidea virens Causing Rice False Smut Yu Zhang, Xinyue Li †, Shuya Zhang, Tianling Ma *, Chengxin Mao and Chuanqing Zhang * Department of Plant Pathology, Zhejiang Agriculture and Forest University, Hangzhou 311300, China; zy1335659339@163.com (Y.Z.); lixinyue1992@126.com (X.L.); 15168349553@163.com (S.Z.); zafumaocx@163.com (C.M.) * Correspondence: czipotw@163.com (T.M.); cqzhang@zafu.edu.cn (C.Z.) † Current address: Station of Agriculture Techniques of Zhenhai District, Ningbo 315200, China. Abstract: Rice false smut caused by Ustilaginoidea virens is one of the most devastating diseases in rice worldwide, which results in serious reductions in rice quality and yield. As an airborne fungal disease, early diagnosis of rice false smut and monitoring its epidemics and distribution of its pathogens is particularly important to manage the infection. In this study, a quantitative loop-mediated isothermal amplification (q-LAMP) method for U. virens detection and quantification was developed. This method has higher sensitivity and efficiency compared to the quantitative real-time PCR (q-PCR) method. The species-specific primer that the UV-2 set used was designed based on the unique sequence of the U. virens ustiloxins biosynthetic gene (NCBI accession number: BR001221.1). The q-LAMP assay was able to detect a concentration of 6.4 spores/mL at an optimal reaction temperature of 63.4 ◦C within 60 min. Moreover, the q-LAMP assay could even achieve accurate quantitative detection when there were only nine spores on the tape. A linearized equation for the standard curve, y = −0.2866x + 13.829 (x is the amplification time, the spore number = 100.65y), was established for the detection and quantification of U. virens. In field detection applications, this q-LAMP method is more accurate and sensitive than traditional observation methods. Collectively, this study has established a powerful and simple monitoring tool for U. virens, which provides valuable technical support for the forecast and management of rice false smut, and a theoretical basis for precise fungicide application. Keywords: rice false smut; quantitative loop-mediated isothermal amplification (q-LAMP); detection; ustiloxins biosynthetic gene 1. Introduction Rice false smut is a disease affecting rice spikes that occurs from the flowering to the milking stage [1,2]. Its most typical and visible symptom is the replacement of rice grains with false smut balls [3,4]. It occurs mainly in Asian countries such as China, Japan, Korea, the Philippines, and India, and is one of the most devastating diseases in the world’s major rice producing regions [5–8]. In recent years, due to the promotion of short-stalked compact and high-yielding rice varieties, indica–japonica interspecific hybrid rice combinations, changes in cultivation patterns, and the excessive use of nitrogen fertilizer during the tillering and gestation periods, the occurrence of rice false smut has become increasingly serious and has gradually risen from a previously minor or sporadic disease to become one of the three new major diseases affecting rice in China [9,10]. The damage caused not only results in a decrease in rice quality and yield, but also the generation of mycotoxin ustiloxins on infected rice spikelets [11,12]. As antimitotic cyclopeptide mycotoxins, the ustiloxins produced within a false smut ball can inhibit microtubule assembly and cell skeleton formation, which poses a serious threat to farmland preservation and ecosystems, as well as the health of humans and animals [13]. Strategies to manage this devastating disease are therefore urgently needed. Citation: Zhang, Y.; Li, X.; Zhang, S.; Ma, T.; Mao, C.; Zhang, C. Quantitative Loop-Mediated Isothermal Amplification Detection of Ustilaginoidea virens Causing Rice False Smut. Int. J. Mol. Sci. 2023, 24, 10388. https://doi.org/10.3390/ ijms241210388 Academic Editor: Fucheng Lin Received: 28 April 2023 Revised: 14 June 2023 Accepted: 19 June 2023 Published: 20 June 2023 Copyright: © 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/). Int. J. Mol. Sci. 2023, 24, 10388. https://doi.org/10.3390/ijms241210388 https://www.mdpi.com/journal/ijms International Journal of Molecular Sciences Int. J. Mol. Sci. 2023, 24, 10388 2 of 13 It is widely accepted that Ustilaginoidea virens (teleomorph Villosiclava virens) is the causal agent of rice false smut [14,15]. As a typical airborne disease, virulent pathogen spores land on the surface of a rice spikelet and germinate hyphae as well as false smut balls on the spikelet [3,12,16–18]. Thus, the epidemic of rice false smut is closely related to the amount of U. virens spores in the field, and the diagnosis of rice false smut, combined with accurate detection and spore quantification, is of great importance for its prevention and management [5,19]. Traditionally, the microscopic counting of spores after capture is widely used in rice false smut diagnosis; however, this method requires specialist taxonomic technicians [20]. Given the complexity of environmental samples and human subjectivity, it is difficult to obtain reliable data with high efficiency via microscopic analysis. A quantitative real-time PCR (q-PCR) technique has been applied for the early identification and quantification of pathogens in airborne diseases [21]. However, this technique is susceptible to interference from environmental dust and other pathogens, making it difficult to quantify the low concentrations of spores captured [20]. in Loop-mediated isothermal amplification (LAMP), developed by Notomi et al. 2000, is a non-PCR-based nucleic acid amplification technique that can be used for the molecular detection of various bacteria, viruses, fungi in disease diagnosis [22–24]. The LAMP reaction is carried out at a constant temperature (between 60 and 65 ◦C) in less than an hour through the use of two pairs of primers—an inner primer (FIP/BIP) and an outer primer (F3/B3). These two pairs of primers constitute the basic LAMP primer set for the LAMP reaction, in order to recognize specific nucleic acid sequences of monitored targets [25–27]. An additional pair of LAMP primers, loop primers, can also be used to significantly improve LAMP efficiency. A simple and visual LAMP assay was developed by Yang et al. in 2018 for the rapid diagnosis of U. virens [28]. However, this assay cannot be used directly for quantitative detection of complex DNA samples. The quantitative- LAMP (q-LAMP) assay (DiaSorin S.p.A., Saluggia, Italy) is a technical improvement from the classical LAMP, which combines LAMP technology with the real-time fluorescence quantitative PCR technique [29]. It is based on the addition of nucleic acid fluorescent dyes, such as SYBR Green or SYTO, resulting in a more sophisticated method suitable for the needs of field diagnosis [30,31]. In this study, we aimed to design and develop a specific and sensitive q-LAMP assay for detection and quantification U. virens, which can be applied in the early diagnosis of rice false smut for preventing the spread of this devastating airborne disease. Additionally, this study is the first report to describe a quantitative diagnostic test for the detection of U. virens using q-LAMP. 2. Results 2.1. Design of Primers for U. virens Detection The best LAMP UV-2 primers were designed based on the ustiloxins biosynthetic gene sequence of U. virens (NCBI accession number: BR001221.1) that did not show any simi- larities to other sequences available in the National Center for Biotechnology Information (NCBI) GenBank database, in order to allow specific amplification of U. virens (Figure 1, Table 1) [32–34]. Additionally, the UV-2 primer sets met the requirement that ∆G values must be less than or equal to −4 Kcal/mol at the 3(cid:48)end of F3/B3 and F2/B2 and 5(cid:48)ends of F1c and B1c. 2.2. Optimization of the q-LAMP Assay To optimize the q-LAMP assay system, the q-LAMP assay was carried out using the UV-2 primer sets at temperatures ranging from 61.8 ◦C to 66 ◦C. As shown in Figure 2, the fluorescence quantitative results show that the strongest fluorescence intensity and the shortest reaction time were obtained when the reaction temperature was 63.4 ◦C (which reached the amplification peak at 30 min). Thus, 63.4 ◦C was chosen as the reaction temperature at which to carry out the optimal q-LAMP assays. Int. J. Mol. Sci. 2023, 24, 10388 3 of 13 Figure 1. The species-specific primers for detecting Ustilaginoidea virens in the quantitative loop- mediated isothermal amplification (q-LAMP) and quantitative real-time PCR (q-PCR). The species- specific primers designed based on the sequence of the ustiloxins biosynthetic gene segments for identification and quantification of U. virens in q-LAMP assay and q-PCR assay. The forward and reverse primer sequences were highlighted with shade and arrow for orientation. Table 1. The sequences of species-specific primers used in the quantitative loop-mediated isothermal amplification (q-LAMP) assay and quantitative real-time PCR (q-PCR) assay. Serial Number UV-2 Sequence (5(cid:48)–3(cid:48)) F3 B3 FIP(F1c-F2) BIP(B1c-B2) GGCACAGCATGACAGGATG TGCTCCCACACTGGTAGT CCTGACATGGCCGGTTTCCCGACGCATGGCCAATAACTCC AGCGGGGCACTTAGGTTCTGCCAATCAAGGCAGCTGATCT Figure 2. Optimization of the q-LAMP assay via reaction temperature screening. The influence of temperature ranged from 61.8 ◦C to 66 ◦C in the q-LAMP detection system and showed that the strongest fluorescence intensity and the shortest reaction time were obtained at 63.4 ◦C (black line). Int. J. Mol. Sci. 2023, 24, x FOR PEER REVIEW 3 of 14 Table 1. The sequences of species-specific primers used in the quantitative loop-mediated isother-mal amplification (q-LAMP) assay and quantitative real-time PCR (q-PCR) assay. Serial Number Sequence (5′–3′) UV-2 F3 GGCACAGCATGACAGGATG B3 TGCTCCCACACTGGTAGT FIP(F1c-F2) CCTGACATGGCCGGTTTCCCGACGCATGGCCAATAACTCC BIP(B1c-B2) AGCGGGGCACTTAGGTTCTGCCAATCAAGGCAGCTGATCT Figure 1. The species-specific primers for detecting Ustilaginoidea virens in the quantitative loop-mediated isothermal amplification (q-LAMP) and quantitative real-time PCR (q-PCR). The species-specific primers designed based on the sequence of the ustiloxins biosynthetic gene segments for identification and quantification of U. virens in q-LAMP assay and q-PCR assay. The forward and reverse primer sequences were highlighted with shade and arrow for orientation. 2.2. Optimization of the q-LAMP Assay To optimize the q-LAMP assay system, the q-LAMP assay was carried out using the UV-2 primer sets at temperatures ranging from 61.8 °C to 66 °C. As shown in Figure 2, the fluorescence quantitative results show that the strongest fluorescence intensity and the shortest reaction time were obtained when the reaction temperature was 63.4 °C (which reached the amplification peak at 30 min). Thus, 63.4 °C was chosen as the reaction tem-perature at which to carry out the optimal q-LAMP assays. Int. J. Mol. Sci. 2023, 24, x FOR PEER REVIEW 4 of 14 Figure 2. Optimization of the q-LAMP assay via reaction temperature screening. The influence of temperature ranged from 61.8 °C to 66 °C in the q-LAMP detection system and showed that the strongest fluorescence intensity and the shortest reaction time were obtained at 63.4 °C (black line). 2.3. Specificity Validation of the q-LAMP Assay System The specificity validation of design in the q-LAMP assay system (using UV-2 primer sets and 63.4 °C as reaction temperature) was tested using the U. virens stain and the other nine fungi. The results show that fluorescence signals were detected in the samples with the DNA template of U. virens, while the samples with the DNA template of the other nine fungi or ddH2O (negative control) did not show any fluorescence signal (Figure 3), indi-cating that the design of the q-LAMP assay system was highly specific to the detection of U. virens. Figure 3. Specificity validation of the q-LAMP assay system. The q-LAMP assay system (using UV-2 primer sets and 63.4 °C as reaction temperature) was highly specific for the detection of U. virens. The q-LAMP assay system showed that fluorescence signals were only detected in the samples with DNA template of U. virens (black line), while the samples with DNA template of the other 9 fungi (including Fusarium fujikuroi, F. oxysporum, F. proliferatum, F. solani, F. graminearum, Penicillium sp., Int. J. Mol. Sci. 2023, 24, 10388 4 of 13 2.3. Specificity Validation of the q-LAMP Assay System The specificity validation of design in the q-LAMP assay system (using UV-2 primer sets and 63.4 ◦C as reaction temperature) was tested using the U. virens stain and the other nine fungi. The results show that fluorescence signals were detected in the samples with the DNA template of U. virens, while the samples with the DNA template of the other nine fungi or ddH2O (negative control) did not show any fluorescence signal (Figure 3), indicating that the design of the q-LAMP assay system was highly specific to the detection of U. virens. Figure 3. Specificity validation of the q-LAMP assay system. The q-LAMP assay system (using UV-2 primer sets and 63.4 ◦C as reaction temperature) was highly specific for the detection of U. virens. The q-LAMP assay system showed that fluorescence signals were only detected in the samples with DNA template of U. virens (black line), while the samples with DNA template of the other 9 fungi (including Fusarium fujikuroi, F. oxysporum, F. proliferatum, F. solani, F. graminearum, Penicillium sp., Pyricularia oryzae, Alternaria alternata and Rhizoctonia solani) or negative control (nucleic acid-free water) did not show any fluorescence signal. 2.4. Sensitivity Validation of the q-LAMP Assay System The sensitivity validation of the q-LAMP assay was determined using the genomic DNA of gradient dilution of U. virens spores as templates under optimal conditions (using primer sets UV-2 and 63.4 ◦C as reaction temperature). As shown in Table 2 and Figure 4, the fluo- rescence signals were detected in the samples with 2 × 104 spores/mL, 4 × 103 spores/mL, 8 × 102 spores/mL, 1.6 × 102 spores/mL, 32 spores/mL, and 6.4 spores/mL within 60 min (the spore extracts were used as DNA template in q-LAMP assay system), while no signals were detected in the sample with the DNA template of 1.28 spores/mL. Thus, theoretically, the q-LAMP assay was able to detect the sample with a concentration of 6.4 spores/mL. We also compared the sensitivity of the q-LAMP assay system with quantitative real-time PCR (q-PCR) for U. virens detection. The q-PCR assay was carried out using the FB/B3 primer set and the effective amplification reactions were detected in samples with spore concentrations of 2 × 104, 4 × 103, 8 × 102, 1.6 × 102 spores/mL, but not 32 spores/mL (Supplementary Figure S1). Thus, the q-LAMP assay system is more sensitive and efficient compared to the q-PCR system used in this study. 2.5. Establishment of a Standard Curve for q-LAMP Detection of U. virens A standard curve between the amplification time (x) and the Log10 value of spore number (y) was constructed based on the q-LAMP assay: y = −0.2866x + 13.829 (Figure 5, Supplementary Table S1), the formula used for calculating spore number is 100.65y, and the correlation coefficient R2 = 0.9942, showing a good linear relationship. Int. J. Mol. Sci. 2023, 24, x FOR PEER REVIEW 4 of 14 Figure 2. Optimization of the q-LAMP assay via reaction temperature screening. The influence of temperature ranged from 61.8 °C to 66 °C in the q-LAMP detection system and showed that the strongest fluorescence intensity and the shortest reaction time were obtained at 63.4 °C (black line). 2.3. Specificity Validation of the q-LAMP Assay System The specificity validation of design in the q-LAMP assay system (using UV-2 primer sets and 63.4 °C as reaction temperature) was tested using the U. virens stain and the other nine fungi. The results show that fluorescence signals were detected in the samples with the DNA template of U. virens, while the samples with the DNA template of the other nine fungi or ddH2O (negative control) did not show any fluorescence signal (Figure 3), indi-cating that the design of the q-LAMP assay system was highly specific to the detection of U. virens. Figure 3. Specificity validation of the q-LAMP assay system. The q-LAMP assay system (using UV-2 primer sets and 63.4 °C as reaction temperature) was highly specific for the detection of U. virens. The q-LAMP assay system showed that fluorescence signals were only detected in the samples with DNA template of U. virens (black line), while the samples with DNA template of the other 9 fungi (including Fusarium fujikuroi, F. oxysporum, F. proliferatum, F. solani, F. graminearum, Penicillium sp., Int. J. Mol. Sci. 2023, 24, 10388 5 of 13 Table 2. The time for fluorescence signal to reach the fluorescence threshold and fluorescence signal records in the q-LAMP assays for testing samples with known spore concentration. Spore Concentration (Spores/mL) Time a (min) (Mean ± Standard Deviation) Fluorescence Signals b 2 × 104 4 × 103 8 × 102 1.6 × 102 3.2 × 101 6.4 1.28 CK c 17.90 ± 1.64 31.44 ± 0.71 34.68 ± 1.26 37.92 ± 1.53 41.16 ± 0.98 44.40 ± 2.42 + + + + + + − − a Time for fluorescence signal reaching the fluorescence threshold in the q-LAMP assays. b “+” indicates a successful fluorescence signal detection, “−” indicates no fluorescence signal detected in the q-LAMP assays. c The nucleic acid-free water was used negative control (CK) in the q-LAMP assay. Figure 4. Sensitivity validation of q-LAMP system. The fluorescence signals in q-LAMP assays were detected in the samples with DNA template of 2 × 104 spores/mL, 4 × 103 spores/mL, 8 × 102 spores/mL, 1.6 × 102 spores/mL, 32 spores/mL, and 6.4 spores/mL within 60 min, while no signals were detected in sample with DNA template of 1.28 spores/mL and CK. The bolded dark green line (horizontal) indicates fluorescence threshold. Fluorescence signals above this threshold marked as a successful detection of U. virens in q-LAMP assays. 2.6. Application of q-LAMP Assay for U. virens Spore Calculation The standard curve of q-LAMP was applied to calculate U. virens spore number on tapes, and each tape sample contained 450, 116, 29, and 9 manually added spores, respectively. As shown in Table 3 and Figure 6, the amplification times quantitated using the cycle threshold (Ct) values for the tested samples were 34.03, 37.12, 40.46, and 43.17, corresponding to 446.07, 118.51, 28.29, and 8.85 predicted spores per tape, respectively, which is very close to the actual spore number on each Melinex tape. Thus, this q-LAMP system can efficiently quantitate U. virens spore number with high accuracy. Int. J. Mol. Sci. 2023, 24, x FOR PEER REVIEW 6 of 14 Figure 4. Sensitivity validation of q-LAMP system. The fluorescence signals in q-LAMP assays were detected in the samples with DNA template of 2 × 104 spores/mL, 4 × 103 spores/mL, 8 × 102 spores/mL, 1.6 × 102 spores/mL, 32 spores/mL, and 6.4 spores/mL within 60 min, while no signals were detected in sample with DNA template of 1.28 spores/mL and CK. The bolded dark green line (horizontal) indicates fluorescence threshold. Fluorescence signals above this threshold marked as a successful detection of U. virens in q-LAMP assays. 2.5. Establishment of a Standard Curve for q-LAMP Detection of U. virens A standard curve between the amplification time (x) and the Log10 value of spore number (y) was constructed based on the q-LAMP assay: y = −0.2866x + 13.829 (Figure 5, Supplementary Table S1), the formula used for calculating spore number is 100.65y, and the correlation coefficient R2 = 0.9942, showing a good linear relationship. Figure 5. Standard curve of q-LAMP detection system. A standard curve between logarithmic values of the spore number (y) and the amplification time quantitated using the cycle threshold (Ct) values (x): y = −0.2866x + 13.829. The correlation coefficient (R2) is 0.9942, showing a good linear relation-ship. Int. J. Mol. Sci. 2023, 24, 10388 6 of 13 Figure 5. Standard curve of q-LAMP detection system. A standard curve between logarithmic values of the spore number (y) and the amplification time quantitated using the cycle threshold (Ct) values (x): y = −0.2866x + 13.829. The correlation coefficient (R2) is 0.9942, showing a good linear relationship. Table 3. Quantitative detection of U. virens spores using q-LAMP system. Ct a 34.03 37.12 40.46 43.17 Manually Added Spores (Spores/mL) Predictive Spores (Spores/mL) R2 p Value 450 116 29 9 446.07 118.51 28.29 8.85 0.999 0.639 a the amplification times (x) quantitated using the cycle threshold (Ct) values. Figure 6. Quantitative detection of U. virens spores on Melinex tape using q-LAMP system. Serial numbers 1, 2, 3, and 4 represent 450, 116, 29, and 9 spores, respectively. The green line (horizontal) indicates fluorescence threshold. Fluorescence signals above this threshold marked as a successful detection of U. virens in q-LAMP assays. Int. J. Mol. Sci. 2023, 24, x FOR PEER REVIEW 6 of 14 Figure 4. Sensitivity validation of q-LAMP system. The fluorescence signals in q-LAMP assays were detected in the samples with DNA template of 2 × 104 spores/mL, 4 × 103 spores/mL, 8 × 102 spores/mL, 1.6 × 102 spores/mL, 32 spores/mL, and 6.4 spores/mL within 60 min, while no signals were detected in sample with DNA template of 1.28 spores/mL and CK. The bolded dark green line (horizontal) indicates fluorescence threshold. Fluorescence signals above this threshold marked as a successful detection of U. virens in q-LAMP assays. 2.5. Establishment of a Standard Curve for q-LAMP Detection of U. virens A standard curve between the amplification time (x) and the Log10 value of spore number (y) was constructed based on the q-LAMP assay: y = −0.2866x + 13.829 (Figure 5, Supplementary Table S1), the formula used for calculating spore number is 100.65y, and the correlation coefficient R2 = 0.9942, showing a good linear relationship. Figure 5. Standard curve of q-LAMP detection system. A standard curve between logarithmic values of the spore number (y) and the amplification time quantitated using the cycle threshold (Ct) values (x): y = −0.2866x + 13.829. The correlation coefficient (R2) is 0.9942, showing a good linear relation-ship. Int. J. Mol. Sci. 2023, 24, x FOR PEER REVIEW 7 of 14 2.6. Application of q-LAMP Assay for U. virens Spore Calculation The standard curve of q-LAMP was applied to calculate U. virens spore number on tapes, and each tape sample contained 450, 116, 29, and 9 manually added spores, respec-tively. As shown in Table 3 and Figure 6, the amplification times quantitated using the cycle threshold (Ct) values for the tested samples were 34.03, 37.12, 40.46, and 43.17, cor-responding to 446.07, 118.51, 28.29, and 8.85 predicted spores per tape, respectively, which is very close to the actual spore number on each Melinex tape. Thus, this q-LAMP system can efficiently quantitate U. virens spore number with high accuracy. Table 3. Quantitative detection of U. virens spores using q-LAMP system. Ct a Manually Added Spores (Spores/mL) Predictive Spores (Spores/mL) R2 p Value 34.03 450 446.07 0.999 0.639 37.12 116 118.51 40.46 29 28.29 43.17 9 8.85 a the amplification times (x) quantitated using the cycle threshold (Ct) values. Figure 6. Quantitative detection of U. virens spores on Melinex tape using q-LAMP system. Serial numbers 1, 2, 3, and 4 represent 450, 116, 29, and 9 spores, respectively. The green line (horizontal) indicates fluorescence threshold. Fluorescence signals above this threshold marked as a successful detection of U. virens in q-LAMP assays. 2.7. Field Application of q-LAMP Assay System The q-LAMP system results show that spores of U. virens were first observed on the 27 August 2018, while the results obtained using the microscope show that spores were observed for the first time on the 2nd of September. Then, the number of spores began to rise rapidly and reached its peak on the 20 September and obvious symptoms of rice false trot were found in the field on the 25th of September. In the following year (2019), the q-LAMP system and microscope manual observation were used to monitor the spores of U. virens in the field again. The results showed that the q-LAMP system detected the spores for the first time on the 31st of August, while microscopic observation led to the detection of only a handful of spores on the 6th of September, and the concentration of spores reached its peak on the 30th of September. The symptoms of rice false smut were found Int. J. Mol. Sci. 2023, 24, 10388 7 of 13 2.7. Field Application of q-LAMP Assay System The q-LAMP system results show that spores of U. virens were first observed on the 27 August 2018, while the results obtained using the microscope show that spores were observed for the first time on the 2nd of September. Then, the number of spores began to rise rapidly and reached its peak on the 20 September and obvious symptoms of rice false trot were found in the field on the 25th of September. In the following year (2019), the q-LAMP system and microscope manual observation were used to monitor the spores of U. virens in the field again. The results showed that the q-LAMP system detected the spores for the first time on the 31st of August, while microscopic observation led to the detection of only a handful of spores on the 6th of September, and the concentration of spores reached its peak on the 30th of September. The symptoms of rice false smut were found in the field on the 5th of October. Through monitoring the dynamic changes in the spore number of U. virens in the field for two consecutive years (Figure 7), it was clearly seen that the q-LAMP system was faster and more efficient than the traditional microscopic observation method. Figure 7. Field application of U. virens spores using q-LAMP system. (A) Flow chart of field U. virens spore sample detection, q-LAMP assay system and microscope observation were used for the collected samples, respectively. (B) The results of U. virens spore concentration measured by different methods in rice fields in 2018 (q-LAMP assay system is the gray line; microscope observation method is the blue line). (C) The results of U. virens spore concentration measured by different methods in rice fields in 2019 (q-LAMP assay system is the gray line; microscope observation method is the blue line). The green arrow is the first detection of spores by q-LAMP assay system, the orange arrow is the first observation of spores by microscope observation, and the red arrow is the occurrence time of rice false smut in the field. Int. J. Mol. Sci. 2023, 24, x FOR PEER REVIEW 8 of 14 in the field on the 5th of October. Through monitoring the dynamic changes in the spore number of U. virens in the field for two consecutive years (Figure 7), it was clearly seen that the q-LAMP system was faster and more efficient than the traditional microscopic observation method. Figure 7. Field application of U. virens spores using q-LAMP system. (A) Flow chart of field U. virens spore sample detection, q-LAMP assay system and microscope observation were used for the col-lected samples, respectively. (B) The results of U. virens spore concentration measured by different methods in rice fields in 2018 (q-LAMP assay system is the gray line; microscope observation method is the blue line). (C) The results of U. virens spore concentration measured by different meth-ods in rice fields in 2019 (q-LAMP assay system is the gray line; microscope observation method is the blue line). The green arrow is the first detection of spores by q-LAMP assay system, the orange arrow is the first observation of spores by microscope observation, and the red arrow is the occur-rence time of rice false smut in the field. 3. Discussion Currently, rice false smut disease caused by U. virens is one of the most devastating rice diseases in China, as well as many other countries [35]. The occurrence of rice false smut disease not only results in the decrease in rice quality and the serious loss of rice yield, but also threatens food safety due to its production of toxic mycotoxins within the false smut balls [10,11]. However, it has been found that rice false smut disease is difficult to control. As a typical airborne disease, the epidemic of rice false smut is closely related to the number of U. virens spores in the field; thus, early detection and warning are critical for preventing and mitigating rice false smut. In this study, a q-LAMP assay system was developed. The results show that the species-specific UV-2 primer sets in the q-LAMP assay system could correctly distinguish U. virens from the other nine air-dispersed fungi, including Fusarium fujikuroi, F. oxysporum, F. proliferatum, F. solani, F. graminearum, Penicil-lium sp, Pyricularia oryzae, Alternaria alternata, and Rhizoctonia solani (Figure 3). Addition-ally, sensitivity validation found that the q-LAMP assay was able to detect a concentration Int. J. Mol. Sci. 2023, 24, 10388 8 of 13 3. Discussion Currently, rice false smut disease caused by U. virens is one of the most devastating rice diseases in China, as well as many other countries [35]. The occurrence of rice false smut disease not only results in the decrease in rice quality and the serious loss of rice yield, but also threatens food safety due to its production of toxic mycotoxins within the false smut balls [10,11]. However, it has been found that rice false smut disease is difficult to control. As a typical airborne disease, the epidemic of rice false smut is closely related to the number of U. virens spores in the field; thus, early detection and warning are critical for preventing and mitigating rice false smut. In this study, a q-LAMP assay system was developed. The results show that the species-specific UV-2 primer sets in the q-LAMP assay system could correctly distinguish U. virens from the other nine air-dispersed fungi, including Fusarium fujikuroi, F. oxysporum, F. proliferatum, F. solani, F. graminearum, Penicillium sp., Pyricularia oryzae, Alternaria alternata, and Rhizoctonia solani (Figure 3). Additionally, sensitivity validation found that the q-LAMP assay was able to detect a concentration of 6.4 U. virens spores/mL at an optimal reaction temperature of 63.4 ◦C within 60 min (Figure 4), and the q-LAMP assay could even achieve accurate quantitative detection when there were only nine U. virens spores on the Melinex tape (Figure 6). Moreover, there was a good linear relationship between the spore amount (y) and the amplification time (x) (Figure 5), which enables accurate quantification of U. virens and early diagnosis of U. virens infection via q-LAMP assay. The LAMP primer set consisted of two outer primers (forward primer F3 and backward primer B3), two inner primers (forward inner primer FIP and backward inner primer BIP), and two loop primers (forward loop F and backward loop B) (Supplementary Figure S2). The outer primers (F3 and B3) were used in the initial steps of the LAMP reactions but later, during the isothermal cycling, only the inner primers were used for strand-displacement DNA synthesis. Outer and inner primers are necessary for LAMP primer design, while the loop primers can be used to accelerate amplification reactions and improve the LAMP assay efficiency [36]. In this study, the q-LAMP primer set was designed according to the work of Wang et al. [20] and Li et al. [37], containing a forward inner primer (FIP), a backward inner primer (BIP), and two outer (F3 and B3) primers. The ustiloxins biosynthetic gene sequence was used to design primers to ensure their specificity. Meanwhile, we sequenced the targeted region of ustiloxins biosynthetic gene in 15 U. virens stains and designed the primer sets elaborately to eliminate the interference from nucleotide polymorphisms, ensuring the amplification efficiency in U. virens detection (Figure 3). For U. virens diagnosis, besides traditional disease diagnosis that includes the iden- tification of symptoms, isolation of pathogens, and microscopic techniques, a conven- tional nested-PCR assay has been developed for the detection U. virens in rice [6]. How- ever, the nested-PCR has less sensitivity and cannot be used in accurate quantification of U. virens [38]. Recently, the q-PCR technique and q-LAMP assay have been applied for the identification and quantification of pathogens in disease diagnosis. In this study, we have established these two systems for U. virens quantification. The q-PCR assay was carried out using the F3/FB primer set and the effective amplification reactions were detected in samples with spore concentrations of 2 × 104, 4 × 103, 8 × 102, 1.6 × 102 spores/mL, but not 32 spores/mL (Supplementary Figure S1), indicating a lower sensitivity of q-PCR for U. virens detection compared to the q-LAMP assay system. Rice false smut has no symptoms in the early stage and can only be identified in the late stage when the smut balls appear. Chemical control is the main means of rice false smut prevention and control [39]. The previous study showed that the first 4~15 d of ear bud breakage was the main period of control, and the first 4~7 d of control was the best [40]. If the key window in the infection of U. virens in rice is not grasped, the efficacy of management will be inadequate [16,40]. Therefore, for rice false smut that relies on airborne transmission, early detection and early warning can aid in disease prevention and control. In this study, we collected spore samples of U. virens from the field for two consecutive years using the q-LAMP assay system and microscopic observation. Compared with manual Int. J. Mol. Sci. 2023, 24, 10388 9 of 13 observation, the q-LAMP assay system could detect spores in the air more accurately and quickly, providing a theoretical basis for precise fungicide application (Figure 7). Therefore, the q-LAMP assay, with higher efficiency and sensitivity, is a better choice for the early diagnosis of rice false smut. In conclusion, this is the first assay developed for the detection of U. virens using q-LAMP assays. Compared with other U. virens detection methods, the newly developed LAMP assay has superior operability, specificity, and sensitivity, and is more suitable for the quantitative detection of U. virens and early diagnosis. 4. Materials and Methods 4.1. Fungal Isolates Isolates of Ustilaginoidea virens and the nine other fungal pathogens used in this study were isolated and identified in our lab, and detailed information on each fungus is listed in Table 4. Isolates were maintained on potato dextrose agar (PDA, prepared by 200 g potato, 20 g glucose, and 20 g agar per 1 L pure water) slants at 4 ◦C. Table 4. The information of strains used in the specificity validation of the q-LAMP assay system. Species Fusarium fujikuroi F. oxysporum F. proliferatum F. solani F. graminearum Penicillium sp. Ustilaginoidea virens Pyricularia oryzae Alternaria alternata Rhizoctonia solani Isolate NO. / ACCC30927 a CICC2489 b ACCC37119 ACCC37680 ACCC31507 ACCC2711 ACCC37631 ACCC36843 ACCC36246 Host Rice Rice Rice Rice Wheat Soil Rice Rice Rice Rice Origin Zhejiang, China Hainan, China Anhui, China Hebei, China Jiangxi, China Shandong, China Hunan, China Fujian, China Hainan, China Beijing, China a ACCC (Agricultural Culture Collection of China). b CICC (China Center of Industrial Culture Collection). 4.2. DNA Template Preparation from Mycelium and Spores for q-PCR and q-LAMP Analysis Preparation of mycelial DNA template for optimum conditions and specificity of the q- LAMP assay, after mycelia grew covering two-thirds of the PDA plate surfaces, the hyphae were then transferred to a mortar and ground with liquid nitrogen. The resultant powder was then placed into a 2-mL centrifuge tube and the mycelial DNA of each fungus was extracted using a Genomic DNA Kit (Sangon Biotech Co., Ltd., Shanghai, China) according to the manufacturer’s instructions. The extracted DNA was used as DNA template in q- LAMP analyses and stored at −20 ◦C. For spore DNA template preparation, after growing on PDA medium at 25 ◦C in darkness for 20 days, 5 mm diameter mycelial plugs taken from colony margin were placed into the potato sucrose (PS, prepared by 200 g potato and 20 g sucrose per 1 L pure water) medium at 25 ◦C 150 rpm, in darkness for 7 days. Spores were separated from medium with filtration through four layers of lens tissue and washed twice with distilled water. Then, spores were diluted with 10% sodium dodecylsulfate (SDS) solution into a series of concentration gradients. An amount of 1-mL spore suspension sample of known concentration mixed with 200-µL 10% Chelex-100 solution [20], 50-µL 10% SDS solution and 0.4 g acid-washed glass beads was placed into a 2.0-mL centrifuge tube. The sample was lysed by Fast Prep Apparatus (JXFSTPRP-24L, Jingxin Technology, Shanghai, China) for 40 s at speed of 6 m/s and placed in boiling water bath for 5 min. The grinding and heating steps were repeated three times, after which the sample was placed on ice for 2 min. The cooled lysate was used directly as DNA template in q-PCR and q-LAMP analyses and stored at −20 ◦C. Int. J. Mol. Sci. 2023, 24, 10388 10 of 13 4.3. Design of q-LAMP Primers for U. virens Detection Ustiloxin A and Ustiloxin B of U. virens are synthesized by ustiloxins biosynthetic gene that was found to be species-specific to U. virens [13,41]. Thus, the sequence of ustiloxins biosynthetic gene (NCBI accession number: BR001221.1) was chosen for q-LAMP primer design using Primer Explore V5 (online web service, http://primerexplorer.jp/e/) ensuring the specificity and accuracy of q-LAMP assay system for U. virens detection. The q-LAMP primers contain forward inner primer (FIP), backward inner primer (BIP), and two outer (F3 and B3) primers (Supplementary Figure S2). The primers were designed according to the following rules: ∆G values of less than or equal to −4 Kcal/mol at the 3(cid:48)end of F3/B3 and F2/B2 and 5(cid:48)ends of F1c and B1c. 4.4. Determination of Optimum Condition of the q-LAMP Assay To better facilitate the efficiency of q-LAMP reaction, the LAMP reaction system was improved via screening for the optimal reaction temperature based on a reference from Notomi [42]. The LAMP reaction was carried out in the following reaction mixtures containing 0.25 µM·L−1 of the primers, FIP and BIP; 0.2 µM·L−1 of the primers, F3 and B3; 1.0 mM·L−1 betaine; 2.0 mM·L−1 dNTPs (Takara Bio Inc., 108, San Jose, CA, USA); 25 mM·L−1 Tris-HCl (pH 8.8); 12.5 mM·L−1 KCl, 12.5 mM·L−1 (NH4)2SO4; 10 mM·L−1 MgCl2; 0.125% (v/v) Triton X-100; 0.2 U·L−1 of Bst DNA polymerase (New England Biolabs, 110, Beijing, China); 0.5 µL 1 × SYBR Green I; and 1 µL of DNA template extracted as described above, and the volume was adjusted to 25 µL with nucleic-acid-free water. The screened reaction temperature gradients were 61.8 ◦C, 62.1 ◦C, 62.6 ◦C, 63.4 ◦C, 64.4 ◦C, 65.2 ◦C, 65.6 ◦C, and 66 ◦C. LAMP reactions were performed using a Bio-Rad quantitative fluorescent PCR instrument (Bio-Rad CFX96, Hercules, CA, USA) for 80 cycles each, each cycle for 60 s, and the reaction was terminated at 80 ◦C for 10 min. Optimal reaction temperature screening experiments were repeated three times. 4.5. Validation of the Specificity for q-LAMP Assay Systems The specificity of the reaction system was tested by performing q-LAMP reactions at the optimal reaction temperature with UV-2 primers in above 25-µL reaction mixtures for 70 min. The assay results were compared with the DNA of U. virens and the 9 other fungi listed in Table 4. The nucleic acid-free water was set as negative control. Additionally, the DNA template of U. virens and the 9 other fungi were prepared as descripted in 4.2 mycelial DNA template preparation. The extracted DNA of U. virens and the 9 other fungi were stored at −20 ◦C and their concentration were more than 150 µg·mL−1. The results were rigorously validated with the assessment that the detectable peak of fluorescence signals detected by Bio-Rad CFX96 as positive; no fluorescence signal as negative. The specificity testing experiment was repeated three times. 4.6. Sensitivity Validation of q-LAMP and q-PCR Assay Systems The sensitivity validation of q-LAMP reactions was performed at the optimal reaction temperature with UV-2 primers in reaction mixtures above 25-µL for 60 min. An amount of 1 µL of DNA lysate from U. virens spores of known concentration was used as a DNA template in the LAMP reaction system. The nucleic-acid-free water was used as a DNA template in the negative control (CK). The detectable peak of fluorescence signals detected by Bio-Rad CFX96 was regarded as positive, while no fluorescence signal was regarded as negative. Sensitivity assay experiments were repeated three times. The sensitivity of the q-PCR reaction system was assayed via performing q-PCR amplification using primers, UV-2 F3/B3. The q-PCR reaction system was 12.5 µL SYBR® Premix Ex Taq II (Tli RNaseH Plus, 2×), 1.0 µL of forward primer F3 (10 µM), 1.0 µL of reverse primer B3 (10.0 µM), 1.0 µL DNA template (in CK, nucleic-acid-free water was used as DNA template), and the volume was adjusted to 25 µL with nucleic-acid-free water. The reaction conditions were: pre-denaturation at 95 ◦C for 2 min, denaturation at 95 ◦C for 5 s, annealing at 60 ◦C for 30 s, extension at 72 ◦C for 6 s. The fluorescence signal was collected during the extension Int. J. Mol. Sci. 2023, 24, 10388 11 of 13 for a total of 40 cycles, and finally the amplification curve was plotted. The detectable peak of fluorescence signals detected by Bio-Rad CFX96 was regarded as positive, while no fluorescence signal was regarded as negative. Sensitivity assay experiments were repeated three times. 4.7. Establishment of Standard Curves for q-LAMP Assay Systems A standard curve was constructed using software SPSS 13.0 by analyzing the associa- tion of logarithmic values of the spore number (y) and the amplification time quantitated using the cycle threshold (Ct) values (x). The correlation coefficient R2 was used for assess- ing the linear relationship between the spore number in sample (y) and amplification time (x). The experiments were repeated three times. 4.8. Calculating of U. virens Spore Using q-LAMP System Spores of U. virens were artificially added to each of the four Melinex tape (1 cm × 2 cm) in the ultra-clean bench, with 450, 116, 29 and 9 spores in each tape. The collected spore- adsorbed Melinex tape was cut and placed in 2-mL centrifuge tubes, and the genomic DNA of the spores on the Melinex tape was then extracted according to the method mentioned above. An amount of 1 µL of the cooled lysate was used directly as DNA template. q-LAMP assay was performed with the optimal reaction conditions in reaction mixtures above 25-µL for 60 min, and the time quantitated using the cycle threshold (Ct) values detected by Bio-Rad CFX96 was recorded as the amplification time (x). The linearized equation for the standard curve was used for converting the amplification time to the corresponding spore number. Then, the calculated spore number was compared to the amount of actual added (listed above) to test the accuracy efficiency of this q-LAMP system. 4.9. Field Application of q-LAMP Assay by U. virens An air borne spore catcher (DIANJIANG, DJ-0723) with Melinex tape was established in Yongyou 1540 cultivation area, Jiangtang Village, Jinhua City, Zhejiang Province for the collection of spores of U. virens, and samples were collected at six-day intervals for 11 consecutive times, starting on the 9th of August 2018. Similarly, starting on the 13th of August of the following year (2019), 11 consecutive samples were collected every six days. The spores of U. virens were adsorbed on the Melinex tape and the tapes (1 cm × 1 cm) with spores were cut and placed in a 2-mL centrifuge tube, and the conidial DNA was extracted according to the methods mentioned above. An amount of 1 µL of the cooled lysate was used directly as DNA template. q-LAMP assay was performed with the optimal reaction conditions in the reaction mixtures above 25-µL for 60 min, and the amplification time quantitated using the cycle threshold (Ct) values was recorded. According to the established standard curve, the number of spores was calculated. The spore population of U. virens in the Melinex tape was recorded via q-LAMP assay at six-day intervals. Meanwhile, the spores of U. virens (1 cm × 1 cm) adsorbed on the slide were suspended in 1 mL of ddH2O, and the spore suspension was counted using a hemocytometer under the microscope to determine the spore concentration. There were three spore catchers placed at the collection site, and the data collected by each instrument were used as a repetition. Supplementary Materials: The following supporting information can be downloaded at: https:// www.mdpi.com/article/10.3390/ijms241210388/s1. Author Contributions: Conceptualization, T.M. and C.Z.; methodology, C.Z., Y.Z. and X.L.; software, T.M. and Y.Z.; validation, T.M. and C.Z.; formal analysis, Y.Z. and S.Z.; investigation, S.Z, C.M. and X.L.; writing—original draft preparation, Y.Z. and T.M.; writing—review and editing, T.M. and C.Z.; visualization, C.M. and X.L.; supervision, T.M. and C.Z. All authors have read and agreed to the published version of the manuscript. Funding: This research was funded by Key Research and Development Project of Zhejiang Province, China (2015C02019), Science &Technology Program of Agriculture and Country in Zhenhai District. Institutional Review Board Statement: Not applicable. Int. J. Mol. Sci. 2023, 24, 10388 12 of 13 Informed Consent Statement: Not applicable. Data Availability Statement: Not applicable. Conflicts of Interest: The authors declare no conflict of interest. References 2. 1. Wei, S.; Wang, Y.; Zhou, J.; Xiang, S.; Sun, W.; Peng, X.; Li, J.; Hai, Y.; Wang, Y.; Li, S. The conserved effector UvHrip1 interacts with OsHGW and infection of Ustilaginoidea virens regulates defense- and heading date-related signaling pathway. Int. J. Mol. Sci. 2020, 21, 3376. [CrossRef] Andargie, M.; Li, J. Arabidopsis thaliana: A model host plant to study plant–pathogen interaction using rice false smut isolates of Ustilaginoidea virens. Front. Plant Sci. 2016, 7, 192. [CrossRef] Song, J.-H.; Wei, W.; Lv, B.; Lin, Y.; Yin, W.-X.; Peng, Y.-L.; Schnabel, G.; Huang, J.-B.; Jiang, D.-H.; Luo, C.-X. Rice False smut fungus hijacks the rice nutrients supply by blocking and mimicking the fertilization of rice ovary. Environ. Microbiol. 2016, 18, 3840–3849. [CrossRef] 3. 4. Meng, S.; Xiong, M.; Jagernath, J.S.; Wang, C.; Qiu, J.; Shi, H.; Kou, Y. UvAtg8-mediated autophagy regulates fungal growth, 5. 6. 7. 8. 9. stress responses, conidiation, and pathogenesis in Ustilaginoidea virens. Rice 2020, 13, 56. [CrossRef] [PubMed] Devi, T.K.; Singh, N.I. Aerobiology and epidemiology of false smut disease of rice by Ustilagnoidea virens (Syn. Claviceps oryzae sativae) in Thoubal District. J. Mycopatholog. Res. 2007, 45, 107–108. Zhou, Y.-L.; Izumitsu, K.; Sonoda, R.; Nakazaki, T.; Tanaka, E.; Tsuda, M.; Tanaka, C. PCR-based specific detection of Ustilaginoidea virens and Ephelis japonica. J. Phytopathol. 2003, 151, 513–518. [CrossRef] Ashizawa, T.; Takahashi, M.; Moriwaki, J.; Hirayae, K. Quantification of the rice false smut pathogen Ustilaginoidea virens from soil in Japan using real-time PCR. Eur. J. Plant Pathol. 2010, 128, 221–232. [CrossRef] Tanaka, E.; Ashizawa, T.; Sonoda, R.; Tanaka, C. Villosiclava virens gen. nov., comb. nov., teleomorph of Ustilaginoidea virens, the causal agent of rice false smut. Mycotaxon 2008, 106, 491–501. [CrossRef] Zhang, Y.; Zhang, K.; Fang, A.; Han, Y.; Yang, J.; Xue, M.; Bao, J.; Hu, D.; Zhou, B.; Sun, X.; et al. Specific adaptation of Ustilaginoidea virens in occupying host florets revealed by comparative and functional genomics. Nat. Commun. 2014, 5, 3849. [CrossRef] [PubMed] 10. Zhou, Y.-L.; Xie, X.-W.; Zhang, F.; Wang, S.; Liu, X.-Z.; Zhu, L.-H.; Xu, J.-L.; Gao, Y.-M.; Li, Z.-K. Detection of quantitative resistance loci associated with resistance to rice false smut (Ustilaginoidea virens) using introgression lines. Plant Pathol. 2013, 63, 365–372. [CrossRef] Sun, X.; Kang, S.; Zhang, Y.; Tan, X.; Yu, Y.; He, H.; Zhang, X.; Liu, Y.; Wang, S.; Sun, W.; et al. Genetic diversity and population structure of rice pathogen Ustilaginoidea virens in China. PLoS ONE 2013, 8, e76879. [CrossRef] 11. 12. Zhou, L.; Lu, S.; Shan, T.; Wang, P.; Wang, S. Chemistry and biology of mycotoxins from rice false smut pathogen. In Mycotoxins: Properties, Applications, and Hazards; Melbor, B., Greene, J., Eds.; Nova Science Publishers, Inc.: New York, NY, USA, 2012. Fu, X.; Wang, A.; Wang, X.; Lin, F.; He, L.; Lai, D.; Liu, Y.; Li, Q.X.; Zhou, L.; Wang, B. Development of a monoclonal antibody-based icELISA for the detection of Ustiloxin B in rice false smut balls and rice grains. Toxins 2015, 7, 3481–3496. [CrossRef] 13. 14. Yu, M.; Yu, J.; Cao, H.; Song, T.; Pan, X.; Qi, Z.; Du, Y.; Zhang, R.; Huang, S.; Liu, W.; et al. SUN-family protein UvSUN1 regulates the development and virulence of Ustilaginoidea virens. Front. Microbiol. 2021, 12, 739453. [CrossRef] 15. Yu, J.; Yu, M.; Song, T.; Cao, H.; Pan, X.; Yong, M.; Qi, Z.; Du, Y.; Zhang, R.; Yin, X.; et al. A homeobox transcription factor UvHOX2 regulates chlamydospore formation, conidiogenesis, and pathogenicity in Ustilaginoidea virens. Front. Microbiol. 2019, 10, 1071. [CrossRef] [PubMed] 16. Tang, Y.-X.; Jin, J.; Hu, D.-W.; Yong, M.-L.; Xu, Y.; He, L.-P. Elucidation of the infection process of Ustilaginoidea virens (teleomorph: Villosiclava virens) in rice spikelets. Plant Pathol. 2012, 62, 1–8. [CrossRef] 17. Hu, Y. Infection processes of Ustilaginoidea virens during artificial inoculation of rice panicles. Eur. J. Plant Pathol. 2014, 139, 67–77. [CrossRef] 18. Yong, M.; Deng, Q.; Fan, L.; Miao, J.; Lai, C.; Chen, H.; Yang, X.; Wang, S.; Chen, F.; Jin, L.; et al. The role of Ustilaginoidea virens sclerotia in increasing incidence of rice false smut disease in the subtropical zone in China. Eur. J. Plant Pathol. 2017, 150, 669–677. [CrossRef] 19. Tsukui, T.; Nagano, N.; Umemura, M.; Kumagai, T.; Terai, G.; Machida, M.; Asai, K. Ustiloxins, fungal cyclic peptides, are ribosomally synthesized in Ustilaginoidea virens. Bioinformatics 2014, 31, 981–985. [CrossRef] [PubMed] 20. Wang, Q.W.; Zhang, C.-Q. Q-LAMP assays for the detection of Botryosphaeria dothidea causing Chinese hickory canker in trunk, water, and air samples. Plant Dis. 2019, 103, 3142–3149. [CrossRef] [PubMed] 21. Harrison, N.A.; Womack, M.; Carpio, M.L. Detection and characterization of a lethal yellowing (16SrIV) group phytoplasma in Canary Island date palms affected by lethal decline in Texas. Plant Dis. 2002, 86, 676–681. [CrossRef] 22. Notomi, T.; Okayama, H.; Masubuchi, H.; Yonekawa, T.; Watanabe, K.; Amino, N.; Hase, T. Loop-mediated isothermal amplifica- tion of DNA. Nucleic Acids Res. 2000, 28, E63. [CrossRef] [PubMed] 23. Aryan, E.; Makvandi, M.; Farajzadeh, A.; Huygen, K.; Bifani, P.; Mousavi, S.-L.; Fateh, A.; Jelodar, A.; Gouya, M.-M.; Romano, M. A novel and more sensitive loop-mediated isothermal amplification assay targeting IS6110 for detection of Mycobacterium tuberculosis complex. Microbiol. Res. 2010, 165, 211–220. [CrossRef] [PubMed] Int. J. Mol. Sci. 2023, 24, 10388 13 of 13 24. McKenna, J.P.; Fairley, D.J.; Shields, M.D.; Cosby, S.L.; Wyatt, D.E.; McCaughey, C.; Coyle, P.V. Development and clinical validation of a loop-mediated isothermal amplification method for the rapid detection of Neisseria meningitidis. Diagn. Microbiol. Infect. Dis. 2011, 69, 137–144. [CrossRef] [PubMed] 25. Xie, L.; Xie, Z.; Zhao, G.; Liu, J.; Pang, Y.; Deng, X.; Xie, Z.; Fan, Q.; Luo, S. A loop-mediated isothermal amplification assay for the 26. visual detection of duck circovirus. Virol. J. 2014, 11, 76. [CrossRef] Soleimani, M.; Shams, S.; Majidzadeh, A.K. Developing a real-time quantitative loop-mediated isothermal amplification assay as a rapid and accurate method for detection of Brucellosis. J. Appl. Microbiol. 2013, 115, 828–834. [CrossRef] 27. Tomita, N.; Mori, Y.; Kanda, H.; Notomi, T. Loop-mediated isothermal amplification (LAMP) of gene sequences and simple visual detection of products. Nat. Protoc. 2008, 3, 877–882. [CrossRef] 28. Yang, X.; Al-Attala, M.N.; Zhang, Y.; Zhang, A.-F.; Zang, H.-Y.; Gu, C.-Y.; Gao, T.-C.; Chen, Y.; Ali, F.; Li, Y.-F.; et al. Rapid detection of Ustilaginoidea virens from rice using Loop-Mediated Isothermal Amplification Assay. Plant Dis. 2018, 102, 1741–1747. [CrossRef] Stella, S.; Gottardi, E.M.; Favout, V.; Gonzalez, E.B.; Errichiello, S.; Vitale, S.R.; Fava, C.; Luciano, L.; Stagno, F.; Grimaldi, F.; et al. The q-LAMP method represents a valid and rapid alternative for the detection of the BCR-ABL1 rearrangement in Philadelphia- positive leukemias. Int. J. Mol. Sci. 2019, 20, 6106. [CrossRef] 29. 30. Wang, Y.; Li, K.; Xu, G.; Chen, C.; Song, G.; Dong, Z.; Lin, L.; Wang, Y.; Xu, Z.; Yu, M.; et al. Low-cost and Scalable platform with multiplexed microwell array biochip for rapid diagnosis of COVID-19. Research 2021, 2021, 2813643. [CrossRef] 31. Ku, J.; Chauhan, K.; Hwang, S.-H.; Jeong, Y.-J.; Kim, D.-E. Enhanced specificity in loop-mediated isothermal amplification with poly(ethylene glycol)-engrafted graphene oxide for detection of viral genes. Biosensors 2022, 12, 661. [CrossRef] 32. Huang, Y.; Tang, X.; Zheng, L.; Huang, J.; Zhang, Q.; Liu, H. Development of generic immuno-magnetic bead-based enzyme-linked immunoassay for Ustiloxins in rice coupled with enrichment. Toxins 2021, 13, 907. [CrossRef] 33. Wang, X.; Fu, X.; Lin, F.; Sun, W.; Meng, J.; Wang, A.; Lai, D.; Zhou, L.; Liu, Y. The contents of Ustiloxins A and B along with their 34. distribution in rice false smut balls. Toxins 2016, 8, 262. [CrossRef] Fu, R.; Chen, C.; Wang, J.; Liu, Y.; Zhao, L.; Lu, D. Transcription profiling of rice panicle in response to crude toxin extract of Ustilaginoidea virens. Front. Microbiol. 2022, 13, 701489. [CrossRef] 35. Abbas, H.K.; Shier, W.T.; Cartwright, R.D.; Sciumbato, G.L. Ustilaginoidea virens infection of rice in Arkansas: Toxicity of false smut galls, their extracts and the Ustiloxin fraction. Am. J. Plant Sci. 2014, 05, 3166–3176. [CrossRef] 36. Rovira, A.; Abrahante, J.; Murtaugh, M.; Claudia, M.-Z. Reverse transcription loop-mediated isothermal amplification for the detection of porcine reproductive and respiratory syndrome virus. J. Veter-Diagn. Investig. 2009, 21, 350–354. [CrossRef] 37. Li, L.; Zhang, S.Y.; Zhang, C.-Q. Establishment of a rapid detection method for rice blast fungus based on one-step loop-mediated isothermal amplification (LAMP). Plant Dis. 2019, 103, 1967–1973. [CrossRef] 38. Li, H.; Ni, D.; Duan, Y.; Chen, Y.; Li, J.; Song, F.; Li, L.; Wei, P.; Yang, J. Quantitative detection of the rice false smut pathogen Ustilaginoidea virens by real-time PCR. Genet. Mol. Res. 2013, 12, 6433–6441. [CrossRef] [PubMed] 39. Wang, Z.; Yang, X.; Lyu, L.; Yuan, B.; Chang, X.; Zhang, S. Progress and prospective of Villosiclava virens infection mechanism. Hubei Agric. Sci. 2019, 58, 5. [CrossRef] 40. Zhang, S.; Zhang, Q.; Luo, H. Test pesticides against rice false smut and choose optimum application period. J. Huazhong Agric. 41. Univ. 2007, 26, 178. Fu, X.; Xie, R.; Wang, J.; Chen, X.; Wang, X.; Sun, W.; Meng, J.; Lai, D.; Zhou, L.; Wang, B. Development of colloidal gold-based lateral flow immunoassay for rapid qualitative and semi-quantitative analysis of Ustiloxins A and B in rice samples. Toxins 2017, 9, 79. [CrossRef] 42. Tomlinson, J.; Boonham, N. Real-time LAMP for Chalara fraxinea diagnosis. Methods Mol. Biol. 2015, 1302, 75–83. [CrossRef] [PubMed] Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.
null
10.1038_s41467-021-27769-5.pdf
Data availability The data and code underlying Fig. 2a, c, d are provided in the github repository https:// github.com/ClaMtnez/Ocean_tags. The data underlying Figs. 3, 4 & 5 and Supplementary Figs. 1, 4 & 5 are provided as a Supplementary Data Files. The sequence data generated in this study have been deposited in the EMBL Nucleotide Sequence Database (ENA) database under Bioproject PRJEB35712 (metagenomic and metatranscriptomic raw reads, metagenomic and metatranscriptomic assemblies, metagenomic assembled genomes, and single-cell amplified genomes) and in the NCBI Sequence Read Archive (SRA) under Bioproject PRJNA593264 (16S rRNA gene amplicon reads). The following public databases were used in this study: Swiss-Prot database, https://www.uniprot.org/, release-2018_10; Genome Taxonomy Database, https://gtdb.ecogenomic.org/, release 80; SILVA non-redundant SSU Ref database, https://www.arb-silva.de/, v.138; UniRef 100 VIROME database, http:// virome.dbi.udel.edu; Greening lab metabolic marker gene database, https://doi.org/ 10.26180/c.5230745; CAZyme HMM database, https://bcb.unl.edu/dbCAN2/, v.8.0; Pfam HMM database, http://pfam.xfam.org/, release 32.0; and TIGRFAM HMM database, https://www.ncbi.nlm.nih.gov/genome/annotation_prok/tigrfams/, release 15.0
Data availability The data and code underlying Fig. 2a , c, d are provided in the github repository https:// github.com/ClaMtnez/Ocean_tags . The data underlying Figs. 3 , 4 & 5 and Supplementary Figs. 1, 4 & 5 are provided as a Supplementary Data Files. The sequence data generated in this study have been deposited in the EMBL Nucleotide Sequence Database (ENA) database under Bioproject PRJEB35712 (metagenomic and metatranscriptomic raw reads, metagenomic and metatranscriptomic assemblies, metagenomic assembled genomes, and single-cell amplified genomes) and in the NCBI Sequence Read Archive (SRA) under Bioproject PRJNA593264 (16S rRNA gene amplicon reads). The following public databases were used in this study: Swiss-Prot database, https://www.uniprot.org/ , release-2018_10; Genome Taxonomy Database, https://gtdb.ecogenomic.org/ , release 80; SILVA non-redundant SSU Ref database, https://www.arb-silva.de/ , v.138; UniRef 100 VIROME database, http:// virome.dbi.udel.edu ; Greening lab metabolic marker gene database, https://doi.org/ 10.26180/c.5230745 ; CAZyme HMM database, https://bcb.unl.edu/dbCAN2/ , v.8.0; Pfam HMM database, http://pfam.xfam.org/ , release 32.0; and TIGRFAM HMM database, https://www.ncbi.nlm.nih.gov/genome/annotation_prok/tigrfams/ , release 15.0
ARTICLE https://doi.org/10.1038/s41467-021-27769-5 OPEN Phylogenetically and functionally diverse microorganisms reside under the Ross Ice Shelf 3, Zihao Zhao 3,4, Rachael J. Lappan 1,2,17, Chris Greening 3,4, Sean K. Bay 1, Clara Martínez-Pérez Daniele De Corte5, Christina Hulbe Ramunas Stepanauskas 16✉ Sergio E. Morales 11, José M. González & Federico Baltar 1,10✉ 6, Christian Ohneiser 7, Craig Stevens 8,9, Blair Thomson10, 12, Ramiro Logares 13, Gerhard J. Herndl 1,14,15, ; , : ) ( 0 9 8 7 6 5 4 3 2 1 Throughout coastal Antarctica, ice shelves separate oceanic waters from sunlight by hun- dreds of meters of ice. Historical studies have detected activity of nitrifying microorganisms in oceanic cavities below permanent ice shelves. However, little is known about the microbial In this study, we profiled the composition and pathways that mediate these activities. microbial communities beneath the Ross Ice Shelf using a multi-omics approach. Overall, beneath-shelf microorganisms are of comparable abundance and diversity, though distinct composition, relative to those in the open meso- and bathypelagic ocean. Production of new organic carbon is likely driven by aerobic lithoautotrophic archaea and bacteria that can use ammonium, nitrite, and sulfur compounds as electron donors. Also enriched were aerobic organoheterotrophic bacteria capable of degrading complex organic carbon substrates, likely derived from in situ fixed carbon and potentially refractory organic matter laterally advected by the below-shelf waters. Altogether, these findings uncover a taxonomically distinct microbial community potentially adapted to a highly oligotrophic marine environment and suggest that ocean cavity waters are primarily chemosynthetically-driven systems. 1 Department of Functional and Evolutionary Ecology, University of Vienna, Djerassiplatz 1, 1030 Vienna, Austria. 2 Centre for Microbiology and Environmental Systems Science, Division of Microbial Ecology, University of Vienna, Djerassiplatz 1, 1030 Vienna, Austria. 3 Department of Microbiology, Biomedicine Discovery Institute, Monash University, Clayton, VIC 3800, Australia. 4 Securing Antarctica’s Environmental Future, Monash University, Clayton, VIC 3800, Australia. 5 Institute for Chemistry and Biology of the Marine Environment, Carl von Ossietzky University of Oldenburg, Oldenburg, Germany. 6 School of Surveying, University of Otago, Dunedin, New Zealand. 7 Department of Geology, University of Otago, Dunedin, New Zealand. 8 National Institute of Water and Atmospheric Research, Greta Point, Wellington 6021, New Zealand. 9 Department of Physics, University of Auckland, Auckland, New Zealand. 10 Department of Marine Sciences, University of Otago, Dunedin, New Zealand. 11 Bigelow Laboratory for Ocean Sciences, East Boothbay, ME, USA. 12 Department of Microbiology, University of La Laguna, ES-38200 La Laguna, Spain. 13 Department of Marine Biology and Oceanography, Institut de Ciències del Mar (CSIC), Barcelona, Spain. 14 NIOZ, Department of Marine Microbiology and Biogeochemistry, Royal Netherlands Institute for Sea Research, Utrecht University, PO Box 59, 1790 AB Den Burg, The Netherlands. 15 Vienna Metabolomics Center, University of Vienna, Djerassiplatz 1, A-1030 Vienna, Austria. 16 Department of Microbiology and Immunology, University of Otago, Dunedin, New Zealand. 17Present address: Institute for Environmental Engineering, Department of Civil, Environmental and Geomatic Engineering, Eidgenössische Technische Hochschule (ETH) Zürich, 8093 Zurich, Switzerland. ✉ email: sergio.morales@otago.ac.nz; federico.baltar@univie.ac.at NATURE COMMUNICATIONS | (2022) 13:117 | https://doi.org/10.1038/s41467-021-27769-5 | www.nature.com/naturecommunications 1 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-27769-5 Results The water column under the Ross Ice Shelf is characterized by a steep vertical ammonium gradient. During the Ross Ice Shelf Program in December 2017, an access borehole was created by hot longitude water drilling at site HWD-2 (latitude 80.6577 S, ice shelf Ice shelves are permanent floating extensions of grounded sheets of ice that connect to a landmass. The Ross Ice Shelf, by in the world, floats atop an area the largest ~54,000 km3 ocean cavity that covers about half of the Ross Sea and hugs the coast of Antarctica (Fig. 1a). Generally over 300 m thick1, the ice shelf creates a “lid” that isolates the underlying ocean from the atmosphere and from sunlight, and exerts a direct effect on the chemical composition of the water column beneath it (in general ~700 m deep2). Waters under the permanent ice shelves are influenced by continental ice-sheet melting and are thus an important intermediary between subglacial outflow from the Antarctic continent and the open Ross Sea, and ultimately the Southern Ocean. Despite their oceanographic significance, sub-ice shelf habitats are among the least-studied ecosystems in the world’s oceans. Oceanographic and biogeochemical observations of the water cavity beneath the Ross Ice Shelf have been largely concentrated on the shelf margins, in particular at the McMurdo Ice Shelf (northwestern portion of the Ross Ice Shelf). Here, nutrient- and biomass-rich water advected from eastern McMurdo Sound likely plays an important role in sub-ice biogeochemistry of the dark ecosystem beneath the shelf front3,4. Direct observations in the grounding area have also confirmed a stratified and quiescent ocean setting5. As a result, water below the Ross Ice Shelf is reported to be exchanged with the Ross Sea with an estimated residence time of 0.9–5.4 years6,7. This allows transport of nutrients and organisms from the sea into the cavity. However, unlike other well-ventilated shelves (e.g., Amery shelf8), the proximity to open water is likely a major factor controlling bio- geochemical process in the central basin of the Ross Ice Shelf cavity. Opportunities to directly access the central sub-ice shelf cavity have been greatly limited by logistical constraints and only one expedition to date has sampled the seawater beneath the center of the Ross Ice Shelf. Sampling of the sub-ice water column took place through borehole J9, during the Ross Ice Shelf Project of 19779. The environment beneath the Ross Ice Shelf was described as “similar to the abyssal ocean in being cold and aphotic”. Within these waters, “sparse” populations of bacteria, microbial eukaryotes, and animals were observed10,11. The microbial populations were proven to be heterotrophically active and incorporated radiolabeled organic carbon molecules at very low rates comparable to the abyssal ocean10. Autotrophic activity of subsequently these microbial communities was reported and attributed to “nitrifying bacteria”12. In these aphotic ecosystems lacking photosynthetic primary production, dark carbon fixation by nitrifying microorganisms may be suf- ficient and macrofaunal populations12. Lateral inputs of organic carbon from the Ross Sea may also support these populations. However, given these studies preceded the advent of molecular techniques, the com- position of the microbial communities, their relatedness to open ocean communities, and their possible links to ecosystem function remained unexplored. to sustain observed microbial techniques In this study we accessed the waters beneath the Ross Ice Shelf to uncover the phylogenetic and functional diversity of the microbial communities under the Antarctic ice shelf. We com- bined multi-omics (metagenomics, metatran- scriptomics, single-cell genomics) with supporting biogeochemical measurements (nutrient measurements and heterotrophic bac- terial production). We show that the waters below the shelf harbor a diverse microbial community with a taxonomic composition distinct from other open ocean environments. In addition, we observed the transcription of various genes associated with lithoautotrophic and organoheterotrophic growth, uncovering the basis for previous activities reported in below-shelf waters. Fig. 1 Sampling location. a Map showing the sampling location of this study (HWD-2) and the borehole study site J9 drilled in 197710. Bathymetry and ice thickness are based on the Bedmap-2 data set1. The transparent ice surface image was sourced from the MOA2009 image map119. b (left) Thermohaline structure of the water column at station HWD-2 and defined regions. IBL, Ice basal boundary layer. V-IL, variable intermediate layer, likely modulated by tides and resulting in patches of water with variable temperature and salinity. S-IL, stratified intermediate layer. BBL, benthic boundary layer. (right) Schematic of HWD-2 drilling site depicts the sampling location of seawater samples (red circles) at 30, 180, and 330 m below the ice shelf base. 2 NATURE COMMUNICATIONS | (2022) 13:117 | https://doi.org/10.1038/s41467-021-27769-5 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-27769-5 ARTICLE ) d ( e m i t r e v o n r u T ) 1 − d 3 − m C l o m µ ( P H P ) % ( A N H ) % ( A N L l ) 1 − m s l l e c 5 0 1 × ( e c n a d n u b a l l e C ) M µ ( 2 O S i ) M µ ( – 3 4 O P ) M µ ( x O N ) M µ ( 3 H N y t i n i l a s l a c i t c a r P ) C ° ( p m e T ) m ( h t p e D 1 6 4 9 3 3 6 8 3 ) 2 0 0 ( . 0 3 0 . ) 3 0 0 ( . 0 6 0 . ) 1 . 0 ( . 4 0 . ) 4 0 ( . 0 8 1 . ) 3 0 ( 1 . 5 1 . ) 7 0 ( . 4 3 2 . ) 2 0 ( . 8 2 8 . ) 3 0 ( . 5 5 8 . ) 9 0 ( . 4 7 7 ) 1 . 0 ( . 9 0 ) 7 0 0 ( . 0 2 . 1 ) 7 0 0 ( . 0 8 0 . ) 1 . 2 ( . 0 5 6 1 . ) 7 0 ( . 0 6 6 1 . ) 3 0 ( . 0 5 6 1 ) 3 0 0 0 ( . 0 2 7 0 . ) 4 0 0 0 ( . 0 2 7 0 . ) 1 0 0 0 ( . 0 1 7 0 . ) 3 2 0 ( . 5 3 7 . ) 4 0 0 ( . 2 3 7 . ) 5 0 0 ( . 7 3 7 . ) 2 0 0 ( . 4 4 0 . ) 2 0 0 ( . 5 0 0 . ) 1 0 0 ( . 4 0 0 . ) u s p ( . 7 5 4 3 9 6 4 3 . . 6 7 4 3 6 9 . 1 − 1 9 . 1 − 3 1 . 2 − 0 8 1 0 3 3 0 3 174.4626 W), approximately 300 km from the Ross Sea and 330 km northwest of borehole J9 (Fig. 1a). The shelf ice was 370 m thick, and the underlying waters extended to 750 m below the shelf sur- face (Fig. 1b). Triplicate samples were collected at three depths: 30, 180, and 330 m below the bottom of the shelf (i.e., the ice-water interface). These depths correspond to three regions based on the thermohaline structure of the water column: a basal boundary layer just beneath the ice (IBL), the upper part of an intermediate layer characterized by highly variable temperature and salinity (V-IL), and the lower part of the intermediate layer characterized by linear stratification (S-IL). A homogeneous benthic layer was observed (BBL) but not sampled (Fig. 1b; see13 for a detailed physical oceanographic description of the study site). This structure con- firmed that the cavity is filled southward by thermohaline convec- tion in which dense, high salinity shelf water (HSSW) evolves into very cold (~−2 °C) but relatively fresh Ice Shelf Water (ISW). The temperature and salinity conditions suggest that, other than the boundary layer regions, water properties conform to Deep Ice Shelf Water, a mixture of high and low salinity shelf water and Antarctic Surface Water (AASW)13. Contrary to what previous studies detected at the shelf front3,4, other regional water masses were not present at borehole HWD2. The flow of waters beneath the drilling site was 2 cm s−1 towards the open ocean, suggesting a residence time for these waters of ca. 4 years13. This estimate is within the range of 1–6 years from previous ocean measurements6 and modeling studies2,14. the ice shelf3,4 and in deep waters of Nutrient concentrations beneath the center of the Ross Ice Shelf were generally lower than those measured at the edge of the the Ross Sea15. of Concentrations of SiO2 (165–166 µM), NOx (7.32–7.37 µM) and 3− (0.71–0.72 µM) were relatively constant across the water PO4 column (Table 1) and two- to fourfold lower than in the oceanic cavity of the McMurdo Ice Shelf at the edge of the Ross Ice Shelf3,4. In contrast, we observed a steep gradient of ammonium, with concentrations tenfold higher at the basal layer (440 nM) than in deeper waters (40–50 nM). Such high ammonium concentrations, while lower than those in open waters of the Ross Sea (which peak in summer with values >2 µM;15), were in the same range as deep (400 m) high-salinity shelf waters (HSSW) entering the front of the cavity (~500 nm;4). A similar nutrient profile was reported beneath borehole J912, where ammonium concentrations were higher beneath the ice shelf base and − − and NO2 decreased with depth, whereas values of NO3 remained constant the water column. However, concentrations of ammonium and NOx were 10- and 4-times higher at the J9 borehole than we reported for the HWD-2 3− and SiO2 were not reported)13,16,17. borehole (PO4 Microbial cell abundance ranged from 0.9 to 1.2 × 105 cells mL−1 (Table 1), which is typical for mesopelagic and upper bathypelagic open ocean environments18 and comparable to deep waters at the the McMurdo Ice Shelf4. In contrast, prokaryotic margin of heterotrophic production (PHP, a proxy for growth of hetero- trophic organisms) ranged from 0.3 to 0.6 µmol C m−3 d−1 (Table 1), which is one to two orders of magnitude lower than at the margins of the Ross Ice Shelf (~40 µmol C m−3 d−1;4) and the average global PHP rates in the mesopelagic (24 µmol C m−3 d−1) and bathypelagic (4 µmol C m−3 d−1) open ocean18. Based on these PHP rates, the turnover time of the microbial community in our study ranged between 339 and 461 days, within the same order of magnitude as the approximately 400 days reported previously at borehole J910. throughout Below-shelf microbial communities are distinct from open ocean communities. Microbial community composition beneath the Ross Ice Shelf was determined using a combination of 16S . n o i t c u d o r p c i h p o r t o r e t e h c i t o y r a k o r p P H P , s l l e c t n e t n o c d i c a c i e l c u n h g h i A N H , s l l e c t n e t n o c d i c a l c i e c u n w o l A N L , e r u t a r e p m e t p m e T . ) 3 = n ( n w o h s e r a s e u a v l ) s e s e h t n e r a p n i n o i t a v e d i d r a d n a t s ( n a e M . e l o h e r o b 2 - D W H e h t t a f l e h S e c I s s o R e h t w o l e b s n m u l o c r e t a w e h t m o r f a t a d l a c i m e h c o e g o i B 1 e l b a T NATURE COMMUNICATIONS | (2022) 13:117 | https://doi.org/10.1038/s41467-021-27769-5 | www.nature.com/naturecommunications 3 ARTICLE a e d u t i t a L 50 0 -50 b ) m 30 180 ( h t p e 330D -100 Ocean zone Bathypelagic 0 Longitude 001 002 Epipelagic Mesopelagic Ocean cavity, RIS 0 02 04 06 08 100 Relative abundance (%) Phylum Acidobacteriota Actinobacteriota Bacteroidota Chloroflexota Crenarchaeota Gemmatimonadota Latescibacterota Marinisomatota Myxococcota Nitrospinota Planctomycetota Proteobacteria SAR324 Thermoplasmatota Verrucomicrobiota Unclassified bacteria Other phyla c y t i r a l i m i s s i D 6 4 2 0 Latitude Polar Non polar Ocean zone Bathypelagic Epipelagic Mesopelagic Ocean cavity, RIS NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-27769-5 d Latitude Polar Non polar Ocean zone Bathypelagic Epipelagic Mesopelagic Ocean cavity, RIS Row Z-score 4 2 0 -2 -4 * ****** Phylum Crenarchaeota Thermoplasmatota Gemmatimonadota Myxococcota Chloroflexota Marinisomatota SAR324 clade PAUC34f AncK6 * Halobacterota Verrucomicrobiota Planctomycetota * Nanoarchaeota Hydrothermarchaeota Euryarchaeota Calditrichota Latescibacterota Aerophobota CK-2C2-2 Eremiobacterota Actinobacteriota Sva0485 Desulfobacterota Proteobacteria Poribacteria Cyanobacteria NB1-j RCP2-54 Caldatribacteriota Bacteroidota Entotheonellaeota Dadabacteria Bdellovibrionota Firmicutes Hydrogenedentes Spirochaetota Margulisbacteria Nitrospinota Fusobacteriota Nitrospirota Patescibacteria Deinococcota Acidobacteriota MBNT15 Cloacimonadota Schekmanbacteria Elusimicrobiota Dependentiae Fig. 2 Comparison of bacterial and archaeal communities in the cavity beneath the Ross Ice Shelf with open ocean environments worldwide. a Global map depicting the locations of metagenomic surveys utilized in the analysis and this study. Overlapping of symbols represent locations where multiple depths were sampled. b Phylum-level composition of microbial communities under the Ross Ice Shelf based on 16S rRNA amplicon sequencing (this study). The results for each sequencing triplicate are averaged; results for individual replicates and controls are shown in Supplementary Fig. 2a, b. Comparisons with metagenomic 16S ribosomal RNA genes (miTags) are shown in Supplementary Fig. 2c. c Cluster dendrogram depicting the average linkage hierarchical clustering based on a Bray-Curtis dissimilarity matrix of community compositions, based on the relative abundance of miTags from this study, global ocean expeditions, and Antarctic and Arctic surveys20–23. The dashed box highlights the clustering of communities in the ocean cavity under the Ross Ice Shelf with global deep-sea environments (in detail in 2d). d Heatmap visualization of calculated Z-scores from below-shelf and global deep-sea environments, based on the relative abundance of miTags grouped at phylum level. Column dendrogram shows clustering of samples according to Bray- Curtis dissimilarity index (detailed from 2b). Rows are clustered based on euclidean distance, grouping phyla that are most likely to co-occur in an environment. Asterisks mark phyla that are significantly more abundant under the Ross Ice Shelf (Kruskal-Wallis test, p < 0.05, Supplementary Data 3). Taxonomic assignment is based on the Genome Taxonomy Database (GTDB107). rRNA gene amplicon sequencing and shotgun metagenomic sequencing. The microbial community was dominated by six phyla: Proteobacteria, SAR324, Crenarchaeota (mostly Nitroso- sphaerales), Marinisomatota (formerly Marinimicrobia, SAR406 clade), Chloroflexota (mostly SAR202), and Planctomycetota (Fig. 2b). Consistent with a dark oligotrophic environment, the eukaryotic community was largely comprised of taxa typically found in the meso- and bathypelagic open ocean, including Alveolata, Dinoflagellata, and Rhizaria lineages (Supplementary Fig. 1a, Supplementary Data 1). With respect to viruses, most bacteriophages detected in the metagenomic assemblies (~50%) belonged to uncultured or unclassified taxa (Supplementary Fig. 1b, Supplementary Data 2), with the most abundant classified viruses affiliating with the family Myoviridae (~30%). We used 16S rRNA gene sequences extracted from metage- nomic reads (miTags;19) to profile the relatedness of microbial communities beneath the Ross Ice Shelf to those of marine ecosystems globally (Fig. 2a, c20–23,). This approach enabled comparison of microbial communities from available marine metagenomic datasets, while circumventing potential biases from inter-study community composition comparisons based on amplicon analyses24. In agreement with previous global metage- nomic analyses20, beta diversity analysis (Bray-Curtis dissim- ilarity) showed oceanic microbial communities cluster by depth, in especially though this was less pronounced in polar regions (Fig. 2c, d). In this global context, the communities beneath the Ross Ice Shelf form a cluster that is related to, but distinct from, those of (Fig. 2c, d). When mesopelagic polar open ocean waters compared to deep (>200 m) open ocean communities worldwide, compositional differences between open-ocean and below-shelf microbial communities are evident even at the phylum level (Fig. 2d). For example, the relative abundances of Chloroflexota, Gemmatimonadota, Marinisomatota, Myxococcota, Planctomy- cetota, and SAR324 were significantly higher under the Ross Ice Shelf, test, p = 9.4 × 10−7 − 1.9 × 10−5, full p values shown in Supplemen- tary Data 3). The phyla Halobacterota, Anck6 and PAUC34f, while typically rare in the open dark oceans, showed a tenfold increase in relative abundance in the cavity beneath the Ross Ice Shelf. Analyses restricted to polar environments using MGLM- ANOVA confirmed significant compositional differences between the ocean cavity and deep (>200 m) open-water polar environ- ments (LRT = 17333, p = 0.001, Supplementary Data 3). In addition, Indicator Species Analysis (Indval) congruently identi- fied ‘signature species’ of the ocean cavity (with respect to deep open-water polar communities) belonging to the phyla PAUC34f, Planctomycetota, and SAR324, as well as the classes Lenti- sphaeria, and SAR202 (p = 0.001–0.002, full p values shown in (Kruskal-Wallis deeper layers 4 NATURE COMMUNICATIONS | (2022) 13:117 | https://doi.org/10.1038/s41467-021-27769-5 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-27769-5 ARTICLE Fig. 3 Phylogeny of reconstructed genomes under the Ross Ice Shelf. Phylogenetic genome tree of the 235 metagenome-assembled genomes (MAGs) and single-amplified genomes (SAGs) retrieved from this study. The genomes are labeled by order, shaded by phylum, and numbered as per Supplementary Data 4. Genome characteristics (inner-to-outer circular heatmap): average genome completeness (%) at phylum level, relative abundance expressed as counts per million (CPM) and relative transcriptional activity as transcripts per million (TPM, Log10 + 1 transformed), and presence of marker genes for key metabolic pathways discussed in the main text. Supplementary Data 3). These ‘signature species’ (with IndVal p < 0.05, test statistic >0.5, Supplementary Data 3) represented on average ~10% of the community beneath the Ross Ice Shelf, reaching up to 17% in the mid water column, in comparison to an average abundance of 0.75% in deep polar open waters. Amplicon sequencing analysis provided additional taxonomic resolution of the communities under the ice shelf and confirmed the depth differentiation anticipated from oceanographic and chemical data. Significant differences in community alpha and beta diversity below the Ross Ice Shelf were observed between the basal boundary layer below the ice (30 m) and the deep water column (330 m) (p = 0.028, Supplementary Data 3, Supplemen- tary Figs. 2 and 3). The species driving these differences are described in the Supplementary Notes. Nitrifying archaea and bacteria dominate transcription under the shelf. We used a multi-omics approach to uncover the functional capacity of the microbial community beneath the Ross Ice Shelf, integrating genome-resolved metagenomics, single-cell genomics, and metatranscriptomics. We assembled 235 derepli- cated partial genomes (Fig. 3, Supplementary Figs. 4 and 5; Supplementary Data 4). These comprised 67 SAGs (single- amplified genomes) and 168 manually curated MAGs (meta- genome-assembled genomes), all with completeness >50% and contamination <5%25 (Fig. 3; Supplementary Data 4). These represent on average 50–60% of each sample’s metagenomic and including all phyla with relative metatranscriptomic reads, abundance above 0.5% (Fig. 2) and the top four most abundant genera (Supplementary Fig. 2b). Their phylogenetic diversity, metabolic traits, and relative abundances are depicted in Fig. 3. The presence and transcription of key metabolic genes in assembled and unassembled reads was used to identify prevailing metabolic pathways in the cavity under the Ross Ice Shelf. By far the most highly transcribed genes involved in autotrophic energy conservation pathways were those for oxidation of ammonium (ammonia monoxygenase, amoA) and nitrite (nitrite oxidoreductase, nxrA) (Fig. 4b). Accordingly, ammonium transporters and amoA NATURE COMMUNICATIONS | (2022) 13:117 | https://doi.org/10.1038/s41467-021-27769-5 | www.nature.com/naturecommunications 5 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-27769-5 Fig. 4 Energy conservation and carbon fixation strategies of communities beneath the Ross Ice Shelf. a Dot plot showing the metabolic potential of the 235 metagenome-assembled genomes (MAGs) and single-amplified genomes (SAGs). The size class of each point represents the number of genomes in each class that encode the gene of interest and the shading represents the average genome completeness. b Heatmaps showing the relative abundance of these genes in the three metagenomic and metatranscriptomic unassembled short reads datasets. For metagenome reads, the heatmap shows the abundance of each pathway, expressed as average gene copies per organism (across all genes listed in the pathway) calculated relative to the abundance of 14 universal single-copy ribosomal genes, with scales capped at 1. For metatranscriptome reads, the heatmap shows log10-transformed reads per kilobase million (RPKM). Where genes within the same pathway are collapsed together, the values (community percentage or RPKM) are summed. c Phylogenetic tree of protein sequences of the highly transcribed ammonia monooxygenase subunit A (amoA) gene from archaeal single-amplified genomes and unbinned metagenomic contigs shown in bold compared to reference sequences. See Supplementary Fig. 7 for a detailed version of this tree. were the most transcribed genes overall (Supplementary Fig. 6). Transcription patterns correlated with ammonium concentrations (Table 1) and relative abundance of the archaeal order Nitrosophaer- ales (Supplementary Figs. 2b, 4 and 5). Phylogenetic analysis corroborated that the most numerous amoA genes and transcripts were affiliated with Nitrosopumilus spp. (Fig. 4c, Supplementary Fig. 7), the most abundant and active archaeal lineage beneath the ice shelf (Supplementary Figs. 2b, 4 and 5), with some gammaproteo- bacterial amoA reads also detected (Fig. 4a, Supplementary Fig. 7). The metagenomic and metatranscriptomic reads of the marker gene for nitrite oxidation, nxrA, affiliated with the phyla Nitrospinota and, to a lesser extent Nitrospirota (Supplementary Data 5, Supplementary Fig. 8). In line with an autotrophic lifestyle, we identified the determinants of ammonium- or nitrite-dependent carbon fixation via the archaeal 4-hydroxybutyrate cycle (hbsC, hbsT genes) and Nitrospina reductive tricarboxylic acid cycle (aclB gene) (Fig. 4, Supplementary Figs. 9, 10 and 11; Supplementary Data 3). Consistent with these results, reconstructed genomes from the genera Nitrosopumilus and Nitrospina were among those with highest relative transcriptional activity in our dataset (S4, S5). These groups express a small fraction of their genomes (i.e., ~25% of total genes at 30 m) compared to other community members (Supplementary Fig. 4d–f), devoting most of their transcriptional effort to the key processes of carbon fixation and ammonia and nitrite oxidation, respectively. Despite being well-represented in the metatranscriptomic dataset, the relative abundance of the genus Nitrospina was low in the metagenomic dataset. For instance, the Nitrospina lineage represented by SAG_5 was among the least abundant genomes, but was highly active on the level (RNA/DNA ~270; Supplementary Fig. 5) transcriptional (Supplementary Data 4). These discrepant findings are in line with recent single-cell analyses showing Nitrospinota have high activity despite low abundance;26 it is proposed that the large cell these nitrite oxidizers are size or high mortality rates of responsible for low abundance in metagenomes and amplicon datasets compared to ammonium oxidizers26,27. their Various inorganic and organic energy sources likely support below-shelf bacteria. Many members of the microbial community are capable of supporting or surviving beneath the shelf through a 6 NATURE COMMUNICATIONS | (2022) 13:117 | https://doi.org/10.1038/s41467-021-27769-5 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-27769-5 ARTICLE Genome name Genome Class (Phylum) CAZyme diversity Transcript counts (TPM) MAG_106 SAG_36 MAG_15 MAG_22 MAG_133 MAG_104 MAG_116 MAG_137 MAG_84 MAG_79 MAG_64 MAG_92 MAG_129 MAG_155 MAG_66 MAG_77 MAG_143 MAG_160 MAG_65 MAG_99 MAG_100 MAG_105 MAG_120 MAG_122 MAG_148 MAG_58 MAG_67 MAG_74 MAG_83 MAG_86 MAG_126 MAG_130 MAG_70 MAG_111 MAG_131 MAG_37 MAG_45 MAG_47 MAG_78 MAG_94 MAG_139 MAG_128 MAG_156 MAG_60 MAG_72 MAG_75 MAG_81 MAG_88 MAG_95 SAG_30 Bacteroidia (Bacteroidota) Bacteroidia (Bacteroidota) Dehalococcoidia(Chloroflexota) Dehalococcoidia(Chloroflexota) Hydrogenedentia(Hydrogenedentota) UBA2968 (Latescibacterota) UBA2968 (Latescibacterota) UBA2968 (Latescibacterota) UBA2968 (Latescibacterota) UBA8240 (Latescibacterota) Marinisomatia (Marinisomatota) UBA4248 (Myxococcota) UBA796 (Myxococcota) UBA796 (Myxococcota) UBA796 (Myxococcota) UBA796 (Myxococcota) UBA9160 (Myxococcota) Physciphaerae (Planctomycetota) Physciphaerae (Planctomycetota) Physciphaerae (Planctomycetota) Planctomycetes (Planctomycetota) Planctomycetes (Planctomycetota) Planctomycetes (Planctomycetota) Planctomycetes (Planctomycetota) Planctomycetes (Planctomycetota) Planctomycetes (Planctomycetota) Planctomycetes (Planctomycetota) Planctomycetes (Planctomycetota) Planctomycetes (Planctomycetota) Planctomycetes (Planctomycetota) UBA1135 (Planctomycetota) UBA1135 (Planctomycetota) UBA1135 (Planctomycetota) Gammaproteobacteria (Proteobacteria) Gammaproteobacteria (Proteobacteria) Gammaproteobacteria (Proteobacteria) Gammaproteobacteria (Proteobacteria) Gammaproteobacteria (Proteobacteria) Gammaproteobacteria (Proteobacteria) Gammaproteobacteria (Proteobacteria) Lentisphaeria (Verrucomicrobiota) Verrucomicrobiae (Verrucomicrobiota) Verrucomicrobiae (Verrucomicrobiota) Verrucomicrobiae (Verrucomicrobiota) Verrucomicrobiae (Verrucomicrobiota) Verrucomicrobiae (Verrucomicrobiota) Verrucomicrobiae (Verrucomicrobiota) Verrucomicrobiae (Verrucomicrobiota) Verrucomicrobiae (Verrucomicrobiota) Verrucomicrobiae (Verrucomicrobiota) Nr. of CAZyme genes per genome 20 40 60 80 GH transcription (TPM) 100 50 0 GT transcription (TPM) 200 150 100 50 0 CBM trancription (TPM) 50 40 30 20 10 0 GH GT CBM CAZyme class 30 m 180 m 330 m 30 m 180 m 330 m 30 m 180 m 330 m Fig. 5 Relative abundance and transcription of selected carbohydrate active enzyme (CAZYme) classes. Data is displayed for reconstructed genomes (MAGs and SAGs) where CAZyme diversity was highest (top 50 genomes). Bubble plots represent the number of different genes from each CAZyme class per genome (GH, glycosyl hydrolases; GT, glycosyl transferases; CBD, genes containing carbohydrate binding domains). Heatmaps represent the total gene transcription for each CAZyme class, normalized to total transcripts per sample (transcripts per million, TPM). The data used to construct these plots is provided in Supplementary Data 7. chemoautotrophic or mixotrophic lifestyle. These include gamma- lineages, such as the Thioglobaceae (SUP05 and proteobacterial ARCTIC96BD-19) and UBA10353, which co-encode genes for the Calvin-Benson-Bassham cycle and heterotrophic metabolism. Con- sistently, RuBisCO genes (rbcL) affiliated to sulfur-oxidizing taxa (Supplementary Fig. 10) were transcribed at high levels throughout the water column (Supplementary Fig. 6). The potential of these lineages to fuel chemoautotrophy using reduced sulfur compounds as electron donors is supported by the presence and transcription of marker genes for sulfide oxidation (sqr, r-dsrA) and thiosulfate oxi- dation (soxB) (Fig. 4a, Supplementary Figs. 12, 13 and 14); (Sup- plementary Data 5 and 6). Abundant heterotrophic lineages, such as Marinisomatota and SAR324 (Fig. 2a, Supplementary Fig. 4), also encoded carbon monoxide dehydrogenases (Fig. 4a, Supplementary Fig. 15, Supplementary Data 6); carbon monoxide may serve as an energy source supporting persistence of this community, as we have recently described for other aerobic heterotrophic bacteria28,29. Genes for formate oxidation were also widespread and highly transcribed (Fig. 4b, Supplementary Fig. 6, Supplementary Data 6), whereas few community members are predicted to use H2 (Supplementary Fig. 16, Supplementary Data 6). Metabolic annotations of the derived genomes suggests that many identified taxa in this ecosystem adopt an organohetero- trophic lifestyle. Highly transcribed genes include a wide range of carbohydrate-active enzymes (CAZymes, Fig. 5,30), as well as the substrate-binding protein of the oligopeptide transporter (OppA; Supplementary Fig. 6). The highest enrichment (genes/Mbp), diversity (number of different families), and transcripts of CAZymes were detected in reconstructed genomes of the phyla Hydrogenedentota, Latescibacterota, Myxococcota, Planctomyce- tota, and Verrucomicrobia. The CAZyme-rich genomes were among the most abundant (i.e., with highest coverage) in our study (Supplementary Fig. 4) and belong to the phyla enriched under the Ross Ice Shelf with respect to deep ocean environments (Fig. 2d). These genomes contained glycoside hydrolases, polysaccharide lyases, and glycosyltransferase families required for the utilization of heterogeneous polysaccharide chains, such as alginate, rhamnose, and xylan (Supplementary Data 7). These genomic features are consistent with previous studies describing the capability of these phyla to metabolize recalcitrant organic polymers31–33. Thus, the proportion of the community differen- tially enriched in this ecosystem could be adapted to degrade refractory organic compounds persisting in the advected waters beneath the Ross Ice Shelf. In contrast to their autotrophic counterparts, these heterotrophic populations transcribed a large percentage of their genome (~80%), especially in deeper waters (Supplementary Fig. 4d–f), with transcriptional effort spreading across a variety of substrate-utilization processes. The metatranscriptome also revealed various other processes supporting life beneath the shelf. The heterotrophic majority in this system transcribed genes involved in the acquisition of inorganic and organic nitrogen and phosphorus compounds (e.g., urea, isocyanates, phosphonates, polyphosphonates; Supplemen- tary Fig. 6). Genes encoding for cold adaptation processes (e.g., NATURE COMMUNICATIONS | (2022) 13:117 | https://doi.org/10.1038/s41467-021-27769-5 | www.nature.com/naturecommunications 7 ARTICLE Ross Ice Shelf HWD-2 melting/refreezing ice NH3 ? NH3 NH3 ice basalboundary layer Nitrosopumilus spp. Nitrospina spp. S-oxidizing lithoautotrophs (e.g. Thioglobus spp.) organoheterotrophs (e.g. Latescibacterota ) advected Corg in situ produced Corg NH3 IBL NH3 CO2 HbsC GT AmoA CO2 AclB - NO2 NxrA - NO3 Sred CO2 RbcL 2- SO4 SoxB GH V-IL S-IL BBL GH GH GT Deep cavity circulation Continental Shelf NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-27769-5 estimate that the waters sampled at the borehole location have been in the cavity for as much as four years prior to sampling; this is up to 10-20-fold longer than the time predicted for marine snow from the ocean surface to reach the abyss (~6000 m38,). Likewise, the heterotrophic production rates measured in this study and at borehole J910 were among the lowest measured in environments with similar marine temperatures39. It has been suggested that production rates are highly influenced by the supply and concentration of labile dis- solved organic material39, and thus the water column beneath the ice shelf is predicted to be highly oligotrophic with respect to labile organic matter. ecosystems, including Based on these heterotrophic rates and assuming a heterotrophic prokaryotic growth efficiency of ~5% (typical of deep oceanic waters, e.g.,40.), we estimate a total organic carbon demand (i.e., the combined carbon incorporation into biomass and respiration) of ~6–12 µmol C m−3 d−1. This total carbon demand is in the same range as the carbon fixation rates reported from the environment beneath the J9 borehole (8.3 µmol C m−3 d−112). While the con- tribution of exogenous organic matter remains to be quantified, the close coupling between in situ dark carbon fixation and organic carbon demand suggests that the ecosystem beneath the Ross Ice Shelf is largely sustained by dark carbon fixation. This would differ from deep open ocean environments, where heterotrophic carbon demand significantly relies on the vertical fluxes of particulate organic carbon generated in the euphotic layer41,42. Fig. 6 Schematic illustration of the dominant bacterial and archaeal groups in the water column under the Ross Ice Shelf. Dotted lines represent the three depths sampled below the sea ice in this study (not to scale; for a scaled representation, see Fig. 1). At the lower fringe of the ice basal boundary layer (IBL), high concentrations of ammonium (from a yet unknown source) are likely to drive high relative abundance and transcriptional activity of ammonium oxidizing archaea (Nitrosopumilus ssp.) and nitrite oxidizing bacteria (Nitrospina ssp.). These, together with sulfur-oxidizing chemolithoautrotrophs (belonging to e.g., the genus Thioglobus), are likely the main source of new organic matter to this ecosystem. The representative enzymes for the metabolic pathways are displayed only once for simplicity but were detected at all depths. The heterotrophic majority is characterized by metabolically versatile bacterial lineages (e.g., belonging to the phylum Latescibacterota), encoding and transcribing multiple copies of carbohydrate-active enzymes (CAZymes, such as glycosyl transferases GT, or glycosyl hydrolases, GH). These likely feed on in-situ generated or laterally advected complex organic matter. cold-shock proteins), osmoregulation (e.g., glycine betaine transporters), and motility (i.e., flagellar apparatus) were highly transcribed (Supplementary Fig. 6). The constitutive expression of cold-shock chaperones can protect against cold-induced protein misfolding34 and is likely an adaptive response to maintain protein homeostasis at the very low water temperatures below the shelf. Furthermore, transport of compatible solutes protects the cell against freezing, hyper-osmolality, and desiccation35. Glycine betaine transporters may provide an additional advantage given these transporters were recently shown to be multifunctional, as in addition to the key they transport multiple substrates osmoregulatory compound glycine betaine36. Discussion Collectively, our results provide a detailed insight on the ecolo- gical strategies adopted by communities living in the world’s most extensive sub-ice shelf system. Oceanic cavities below ice shelf systems are uniquely different from open ocean environments in their dependence on in situ chemosynthesis and on lateral advection of food sources from open-water areas, rather than on vertical fluxes of phytoplankton-derived detrital matter37. We Our multi-omic results support this hypothesis, while unco- vering the mediators and pathways responsible for the auto- trophic and heterotrophic activities under the Ross Ice Shelf (Fig. 6). Among the lineages represented by MAGs and SAGs with the highest transcriptional activity are those originating from the chemolithoautotrophic genera Nitrosopumilus and Nitros- pina. Overall, this agrees with previous reports that aerobic ammonium-oxidizing microorganisms are widespread in Ant- arctic marine environments (e.g.,43) and that ammonium oxida- tion occurs beneath Antarctic shelves and sea ice12,44. These and other inferred facultative chemolithoautotrophs (such as facul- tative sulfur-oxidizing bacteria) are likely to be responsible for dark carbon fixation rates previously observed beneath borehole J912 and thus provide a supply of organic carbon to an ecosystem shielded from sunlight. (e.g., Nitrospina, Nitrosopumilus, The importance of dark carbon fixation has been recognized in various oceanic regions during the polar winter. Microbial lineages and Marinisomatota45–47) and enzymes (such as those mediating ammonium, nitrite, and sulfur oxidation48) that mediate che- molithoautotrophy have been observed to increase in Antarctic waters during the transition to the winter season. Likewise, comparable lineages and genes capable of sulfur compound oxi- dation have been detected in winter open waters and the central basin under the Ross Ice Shelf. Together with mounting evidence that sulfur compound oxidizers sustain carbon fixation in the wide dark open ocean (e.g.,49) and the diverse sources of reduced sulfur compounds in marine oxic environments (e.g.,50), it is plausible that these clades can also contribute to chemoauto- trophy in the oceanic cavity beneath the Ross Ice Shelf. SAR324, It is likely that ammonium is a primary energy source sustaining primary production in aphotic Antarctic waters. Consistent with this idea, ammonium oxidation rates have been reported to be higher in Antarctic coastal waters during the austral winter and to significantly support the heterotrophic demand43. In the absence of direct rate measurements in this study, we estimated the ammo- nium oxidation rates potentially supported by the standing ammonium concentrations in the water column. Our estimates for + d−1) are in accordance to rates the basal layer (~90 nM NH4 + d−1) with measured in the Southern Ocean (62 nM NH4 8 NATURE COMMUNICATIONS | (2022) 13:117 | https://doi.org/10.1038/s41467-021-27769-5 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-27769-5 ARTICLE +;43), and comparable ammonium concentrations (0.7 µM NH4 could support the heterotrophic demand in the oceanic cavity under the shelf (Supplementary Notes). These estimates suggests that the microbial communities beneath the Ross Ice Shelf can sustain ammonium oxidation at similar rates to those in the winter Antarctic Ocean and have the potential to be significant primary producers. The ammonium profile beneath the Ross Ice Shelf is intriguing. Contrary to other nutrient concentrations measured (which do not vary significantly through the water column), ammonium concentrations are significantly higher in the ice basal boundary layer compared to the deeper water samples, but comparable to those in the periphery of the shelf4. This profile (exclusive for ammonium with respect to other nitrogen species) is consistent with the reports beneath borehole J912. The proposed circulation model beneath the shelf13, by which the cavity is filled southward by dense water masses that reach its interior via deep cavity circulation, renders it unlikely that the high ammonium con- centrations detected in the fresh, northward flowing waters beneath borehole HW2D or J9 originate from the open Ross Sea. If externally sourced, nutrient concentrations would be expected to be highest in deeper waters, or else be homogenized in the water column as water masses evolve and mix in the cavity. The latter appears to be the case for the other nutrients measured in this and the J9 expedition. The exception observed in the ammonium profile suggests that this compound could be sourced beneath the ice shelf. In particular, terrestrial-origin sediments in the basal ice layer may be a significant source of ammonium to the seawater circulating beneath. Deployment of cameras at HWD2 revealed sedimentary englacial debris in the lower 20 meters of the ice shelf13. While ice melting and freezing can plausibly result in the rainout of the pellets in a sub-ice-shelf cavity, we did not witness this effect; no sediments were retrieved from the pumping samples and the microbial communities sequenced from the englacial debris and the water column were unrelated (Supplementary Fig. 2). However, temperature and salinity data from our study site (Fig. 1b,13) clearly showed ice- shelf basal melting and a supply of freshwater to the upper region of the water column, a phenomenon that could result in the observed replenishment of ammonium concentrations in this system. In free-floating sea ice, as well as in subglacial lakes, ammonium enrichments have been traditionally attributed to wet and dry atmospheric deposition, as well as in situ organic matter regeneration in brine channels, especially within older and thicker ice51–53. The latter may be also a mechanism for ammonium accumulation in deep layers of the ice shelf54, subject to solubi- lization and transport by fresh melt water. If such is the case, the ammonium transported by the ice basal boundary layer could be sourced locally (at borehole HWD2) or elsewhere upstream. Dissolved nutrients in the ice sheet or englacial debris are even- tually diluted as they circulate the interior of the shelf54, which could explain the observed higher concentrations in the water column from borehole J912, 330 km upstream from our study site. While the driving factors of the nutrient profile in the water the tenfold decrease in ammonium column remain unclear, concentrations correlate with changes in relative transcriptional activity of the ammonium-oxidizing genus Nitrosopumilus (Supplementary Fig. 4). As described in Supplementary Notes, we observed depth-related differences in microbial community composition, metabolic capabilities, and gene expression, though additional depth profiles would be required to confirm this. The community members with highest relative abundance and transcriptional activity throughout the water column included nitrifying autotrophic taxa and organoheterotrophic bacteria (Supplementary Figs. 4, 5 and 6). It is likely that the genomes with highest relative transcriptional activity represent two opposite adaptative strategies to the conditions beneath the Ross Ice Shelf. Based on the proportion of their genome expressed, nitrifiers are the surrounding environment by likely to effectively exploit expressing a reduced set of genes encoding a few metabolic pathways. The opposite is observed in the highly expressed het- erotrophic clades (Supplementary Fig. 4). By expressing up to 95% of their genome (e.g., in members of Latescibacterota and Verrucomicrobiota), the transcriptional effort of the latter is spread across a variety of process and in particular, to the exploitation of multiple substrates. These observations are con- sistent with previous studies combining expression and genomic datasets, which suggest that activity levels, substrate utilization and transcriptome diversity may be linked in defining ecological niches of microbial communities55,56. In particular, our results suggest that the most active hetero- trophic organisms are adapted to degrade complex organic com- pounds, including most of the enriched phyla in this environment, such as Myxococcota and Planctomycetota. Their capacity to degrade complex organic material from a range of sources, including potentially of both autochthonous and allochthonous origin, likely confers a major selective advantage in this highly oligotrophic ecosystem. Heterotrophy based on the consumption of recalcitrant dissolved organic carbon has been considered as one possibility for sustaining the oceanic Antarctic winter food web57, and could also be an additional support for life under the Ross Ice Shelf. Unlike organic carbon in Antarctic winter waters, which may have accumulated during the highly productive summer season, organic substrates beneath the Ross Ice Shelf potentially consist of vertically transported exudates and necromass derived from lithoautotrophic primary producers, but also recalcitrant complex organic compounds laterally transported from the Ross Sea into the shelf cavity. Decomposition of phytoplankton entering the shelf cavity is estimated at a scale of ~10 years4. Together with previous reports of diatoms in below-shelf waters9, this indicates that some photoautotrophically-derived organic matter can reach the center of the oceanic cavity. However, the metagenomes suggest that pho- tosynthetic eukaryotes (i.e., class Bacillariophyceae) make a small fraction of the eukaryotic community (0.05 %); this finding is also consistent with undetectable concentrations of chlorophyll a beneath borehole J912. Despite potentially serving as a substrate for organoheterotrophs beneath the ice shelf, phytoplankton are therefore unlikely contributors to the dissolved organic matter pool, whereas detrital sources of bacterial substrates may be more important. Further work is now needed to discriminate organic matter sources and nutrient exchange processes within the shelf. Overall, microorganisms under Antarctica’s ice shelves can thrive in some of the coldest and possibly carbon-limited marine waters, while playing a crucial role in the remineralization of nutrients to the Southern Ocean. Our results not only suggest that the waters below the Ross Ice Shelf are driven by chemo- lithoautotrophic processes, but also uncover the mechanisms responsible for sustaining that activity58. Alongside other recent reports of oceanic dark carbon fixation,27,49,59, this study also emphasizes the importance of inorganic energy sources in driving marine communities in the absence of photosynthesis. Finally, our results suggest that ammonium associated with fresh melt waters at the base of the ice is an important supply of inorganic electron donors supporting chemolithoautotrophy, and thus has a significant the microbial community. Ocean-driven basal melting, a source of freshwater and thus potentially of ammonium in the sub-ice cavity, may increase in a warming climate scenario60. Assuming that our observations are representative of the central region of the cavity under the Ross Ice Shelf, increased basal ice melting could result in an increased vulnerability of communities sup- ported by sub-ice shelf processes61, potentially leading to shifts in influence in the composition and activity of NATURE COMMUNICATIONS | (2022) 13:117 | https://doi.org/10.1038/s41467-021-27769-5 | www.nature.com/naturecommunications 9 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-27769-5 the relative biogeochemical importance of chemolithoautotrophic processes in this extensive ecosystem. These insights emphasize the importance of baseline data from existing sub-ice shelf eco- systems, such as the Ross Ice Shelf, to inform the prediction of biogeochemical impacts of climate change in the Southern Ocean. Methods Site selection and description. Sampling took place in December 2017 and was conducted by members of the Aotearoa New Zealand Ross Ice Shelf Program. Samples were collected from the sub-shelf water column at a site in the central region of the ice shelf, borehole HWD-2 (Latitude -80.6577 N, Longitude 174.4626 W), ~300 km from the shelf front and 330 km northwest of borehole J9 (Fig. 1a). The sampling site is near the glaciological boundary between ice origi- nating from the West Antarctic Ice Sheet and ice flowing from East Antarctica through Transantarctic Mountain glaciers (Fig. 1a). Sediment of terrestrial origin was observed in the lowermost ~60 m of the ice. Hot water drilling and sampling. A hot water drilling system built and operated by the Victoria University of Wellington Drilling Office was used to bore through the ice shelf, creating an access borehole with a maximum diameter of 30 cm. The borehole was used for direct sampling of water and sea floor sediments, and to conduct in situ measurements in the water column. These activities were con- ducted inside a custom-built tent that facilitated 24-h operations in any weather conditions. Seawater samples were obtained from three depths (400 m, 550 m, and 700 m from the top of the shelf, which correspond to 30 m, 180 m, and 330 m deep from the bottom of the ice shelf, respectively). These were chosen to characterize the water column under the Ross Ice Shelf while keeping the sampler ca. 40–50 m away from the seafloor and from ice crystals and sediment in the ice-shelf basal layer. The drilling water was fresh (<15 psu) and relatively warm (between −1 and +1 °C), so it remained stably floating in the borehole and did not sink into deeper layers. This, together with the advection of seawater below the ice shelf, precluded any contamination of collected seawater with the drilling water (Supplementary Fig. 2a, b). The lack of intrusion of the freshwater used for the drilling was rou- tinely checked by salinity and temperature-depth profiles. Samples were collected by in situ filtration using a McLane WTS-LV-Bore Hole filter pump fitted with a 142 mm diameter, 0.22 μm pore-size filter (Supor membrane filters, Pall Corporation). Before and after deployment, the filter holder was thoroughly cleaned to avoid sample cross-contamination. The pump head interior was also flushed after every deployment with fresh water to prevent salt crystal formation and sample contamination. This sampling approach was aimed at obtaining the most realistic representation of the microbial community’s composition and activity with the minimum possible sampling biases. Approximately 200 L of water were filtered at each depth within ca. 2 h. Thereafter, filters were placed in sterile Petri dishes and divided into seven sections using sterile scalpels and transferred to cryovials. The filtered, frozen samples were directly stored in zip lock bags in a 3 m deep borehole drilled into the cold surface snow layer until transported to Scott Base (and further airplane transport to New Zealand). The temperature of the samples deposited in the storage borehole remained stable ranging mostly between –27 °C and –28 °C (Supplementary Fig. 17). These samples were used for 16S rRNA amplicon sequencing, metagenomics, and metatranscriptomics. Water samples (150–300 mL) were also collected at the same three depths using the McLane WTS-LV-Bore Hole pump without a filter-holder in order to further minimize contamination. Once the pump was brought up, it was run in reverse to collect the water, but excluding the first 30–60 mL of water (used for rinsing). Water samples for inorganic nutrient analyses were filtered through combusted Whatman GF/F filters, collected in acid-cleaned HDPE bottles, and stored frozen until analysis in the home laboratory, following procedures recommended by the Joint Global Ocean Flux Study (JGOFS62). The liquid samples for the determination of microbial cell abundance, prokaryotic heterotrophic production, and the generation of single-cell amplified genomes (SAGs) were collected in acid- cleaned Nalgene™ opaque amber HDPE bottles, stored at 2 °C, and transported within 48 h to Scott Base to perform further laboratory analyses. The samples were imported to New Zealand under Ministry for Primary Industry permit number 2017063583 (Permit to import Restricted Biological Products of Animal Origin) issued to the University of Otago Department of Marine Science. To check for potential contamination, samples were also collected from the following sites: freshly melted snow nearby the camp area, drilling water from a reservoir tank, and sediments dislodged from the ice shelf (identified as englacial debris) and collected with the reaming tool. Water samples were filtered onto 0.22 µm polycarbonate filters (47 mm filter diameter, Millipore), and all samples were stored in cryovials and frozen. Physicochemical measurements. A SBE 19plusV2 SeaCAT Profiler CTD (Seabird Electronics, Inc.) was used to measure temperature, salinity and depth within the borehole and in the water under the Ross Ice Shelf for a detailed characterization of the water column. Furthermore, a self-contained single channel logger (RBR Solo) was attached to the frame of the WTS-LV-Bore Hole pump (at the opposite side of the water intake) for an accurate determination of the temperature and depth of the sampling casts. Samples for determining the concentrations of nitrate, dissolved 62 were colorimetrically reactive phosphorus (phosphate), ammonium and SiO2 analyzed using flow-injection analysis on a Lachat Auto-analyzer according to methods described elsewhere63. Measurements of nutrient concentrations were routinely corrected with reference blank solutions in each sample run. No anomalies were detected in the blanks, indicating no source of detectable con- tamination during the measurements. Prokaryotic abundances and heterotrophic production. Prokaryotic abundance was determined by flow cytometry. Samples (1.6 mL) were preserved with glutar- aldehyde (2% final concentration), left at 4 °C in the dark for 15 min, flash-frozen in liquid nitrogen, and stored at −80 °C until analysis. Prior to analysis, the fixed samples were thawed, stained in the dark with a DMS-diluted SYTO-13 dye (Molecular Probes Inc., 2.5 µM final concentration) for 5 min, and run on a BD AccuriTM flow cytometer with a laser emitting at 488 nm wavelength. Samples were run at low or medium speed until 10,000 events were captured. A suspension of yellow–green 1 µm latex beads (105–106 beads mL−1) was added as an internal standard (Polysciences, Inc.). Prokaryotic heterotrophic activity was estimated via the incorporation of 3H-leucine using the centrifugation method64. 3H-leucine (Perkin-Elmer, specific activity 169 Ci mmol−1) was added at saturating concentration (40 nmol L−1) to triplicate 1.2 mL subsamples. Controls were established by adding 120 µL of 50% trichloroacetic acid (TCA) to triplicate control tubes 10 min prior to radioisotope addition. The microcentrifuge tubes were incubated in the dark at 4 °C for 48 h. Incorporation of leucine in the quadruplicate tubes per sample was terminated by adding 120 µL ice-cold 50% TCA. Subsequently, the samples and the controls were kept at –20 °C until centrifugation (at ca. 12,000 × g) for 20 min followed by aspiration of the water. Finally, 1 mL of scintillation cocktail was added to the microcentrifuge tubes before determining the incorporated radioactivity after 24–48 h on a Tri-Carb 2000® Liquid Scintillation Counters scintillation counter (Perkin-Elmer) with quenching correction. The blank-corrected leucine incorporation rates were converted into prokaryotic heterotrophic production (PHP) using the theoretical conversion of 1.55 kg mol−1 leucine incorporated65–67. The rates of leucine incorporation obtained at the incubation temperature (4 °C) were converted to the in situ temperature of -2 °C using an activation energy of 72 kJ mol−1[ 67. Single cell genomics. Sample collection and analyses were performed as described previously27, see Supplementary Methods for full description. Briefly, triplicate seawater samples (1 mL) were transferred to a sterile cryovial containing 100 µL of glyTE (20 mL of 100 × TE buffer pH 8.0, 60 mL Milli-Q water and 100 mL of molecular-grade glycerol), and samples were stored at –80 °C until analysis. SAG generation was performed at the Single Cell Genomic Center at Bigelow Laboratory for Ocean Sciences (SCGC) using fluorescence-activated cell sorting and WGA-X genomic DNA amplification. Paired-end Illumina libraries were created with Nextera XT (Illumina), sequenced with NextSeq 500 (Illumina) and de novo assembled using a workflow based on SPAdes68 as previously described69. The quality of the sequencing reads was assessed using FastQC v0.11.7 (https:// www.bioinformatics.babraham.ac.uk/projects/fastqc/) and the quality of the assembled genomes was determined using CheckM v.1.0.770 and tetramer fre- quency analysis71. This workflow was evaluated for assembly errors using three bacterial benchmark cultures with diverse genome complexity and %GC, indicating no non-target and undefined bases in the assemblies and average frequencies of mis-assemblies, indels and mismatches per 100 kbp: 1.5, 3.0 and 5.069. Functional annotation was first performed using Prokka72 with default Swiss-Prot databases supplied by the software. Prokka was run a second time with a custom protein annotation database built from compiling Swiss-Prot73 entries for Archaea and Bacteria. DNA extraction, 16S rRNA gene amplicon and metagenomic sequencing. DNA was extracted using a PowerSoil® DNA Isolation Kit (MoBio, Carlsbad, CA, USA). The manufacturer’s protocol was modified to use a Geno/Grinder for 2 × 15 s instead of vortexing for 10 min and a final elution of 50 µL solution C6 (sterile elution buffer, 10 mM Tris) was used. DNA concentration was measured using a Nanodrop spectrophotometer (Thermo Fisher). The median 260/280 nm wave- length ratio was 1.5 with a lower quartile of 1.4 and an upper quartile of 1.7. Extractions were performed in triplicate for each depth under the Ross Ice Shelf (total of 9 samples) for subsequent amplicon and metagenomic sequencing. 16S rRNA gene amplicon sequencing was carried out using the Earth Microbiome Project74 protocols and standards (http://earthmicrobiome.org/ protocols-and-standards/16s/), which include the following modifications to the original 515F–806 R primer pair75 (the updated sequences, 5′- 3′, are as follows: 515 F: GTGYCAGCMGCCGCGGTAA; 806 R: GGACTACNVGGGTWTCTAAT). In brief, degeneracy was added to both the forward and reverse primers to remove known biases against Crenarachaeota/Thaumarchaeota (515 F, also called 515F- Y76) and the marine and freshwater Alphaproteobacterial clade SAR11 (806 R77,). All amplicons (independent replicates) were run on an Illumina (Foster City, CA, USA) MiSeq 250 bp × 2 run. For metagenomic sequencing, Thruplex DNA libraries 10 NATURE COMMUNICATIONS | (2022) 13:117 | https://doi.org/10.1038/s41467-021-27769-5 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-27769-5 ARTICLE (~300 bp inserts) were created from each individual DNA extraction and sequenced in an Illumina HiSeq 2500 platform (2 × 125 bp). RNA extraction and metatranscriptomic sequencing. RNA was extracted fol- lowing the RNeasy mini kit (Qiagen, Hilden, Germany) procedure and the ethanol precipitation protocol. The remaining DNA was removed with TurboDNase (Invitrogen, Carlsbad, CA, USA) and the efficiency of removal was tested with PCR. Enrichment of RNA was performed with 20 μL of sample RNA following the procedures of the MICROBEnrich (Ambion, Austin, TX, USA) and MICROBEx- press (Ambion, Austin, TX, USA) kits. Thereafter, the MessageAmp II-Bacteria kit (Invitrogen) was used to improve the subsequent amplification and purification: enriched RNA was reverse transcribed to cDNA, which was in vitro transcribed back to amplified RNA (aRNA) using the mentioned kit. Quantifications were simultaneously run with a Nanodrop spectrophotometer (Thermo Fisher) and a Qubit fluorometer (Invitrogen, Carlsbad, CA, USA) using the RNA HS Assay kit and an RNA profile generated with a Bioanalyzer 2100 (Agilent Technologies, Böblingen, Germany). aRNA was shotgun sequenced directly in an Illumina HiSeq4000 platform (CNAG, Barcelona, Spain), generating between 28–35 Gb of 2 × 101 bp reads per sample. 16S rRNA gene amplicon profiling. Paired-end 16S rRNA gene amplicon sequences were processed on the QIIME2 platform using the DADA2 pipeline to resolve exact amplicon sequence variants78,79. Raw reads were demultiplexed, yielding 302,585 reads across 16 samples. Quality plots were generated and sequences failing to pass an average base call accuracy of 99% (Phred score 20) were excluded. Low quality regions of each sequence were removed by trimming the first 13 bases of the forward and reverse reads and truncating at 150 base pairs before de-noising with DADA2 using the function qiime dada2 denoise-paired with default parameters. The final dataset contained 1228 amplicon sequence variants (ASVs) with a total frequency of 271,736. Taxonomic assignment was performed by using a Genome Taxonomy Database classifier built for the QIIME2 platform, using the SSU sequence files from GTDB ssu_r86.1_20180911 (https://osf.io/25djp/ wiki/home/). The classifier was first spliced to the 515 F/806 R primer pair using the qiime feature-classifier extract-reads, and trained using the qiime feature- classifier fit-classifier-naive-bayes command in QIIME279. The trained classifier was then used to assign the taxonomy to the ASV features using our representative reads via the function feature-classifier classify-sklearn. No sequence overlap was observed between below-shelf waters with those of control samples (e.g., drilling fluid, sediment recovered from basal ice on the shelf, snow at the camp site) (Supplementary Fig. 2), confirming absence of contamination in the water column samples. Metagenomic community profiling. Raw metagenomic and metatranscriptomic paired-end reads were quality-assessed with FastQC v0.11.7 and MultiQC v1.080. BBDuk v38.51 from the BBTools suite (https://sourceforge.net/projects/bbmap/) was used to trim adapter sequences, remove reads corresponding to Illumina’s PhiX sequencing control, trim low-quality bases (minimum quality score 20), and discard short sequences (minimum length 50 bp). The metatranscriptome reads were further processed with SortMeRNA v2.1b81 to remove reads corresponding to prokaryotic and eukaryotic ribosomal RNA, followed by BBDuk to filter low- complexity reads (entropy threshold 0.05). In addition, taxonomic profiling of bacteria, archaeal, and eukaryotic communities was performed with 16S rRNA gene sequences extracted from metagenomic reads (miTags) using a previously described protocol19. miTags were also extracted from bathypelagic samples from the Malaspina Circumnavigation expedition23, metagenomic surveys in the Arctic and Southern Ocean21, as well as metagenomic datasets from polar regions obtained from the TARA Ocean Expedition22. This allowed comparing these datasets to available miTags from epipelagic and mesopelagic samples from the TARA Ocean Expedition20. Extracted 16S and 18S rRNA gene reads were mapped to the SILVA non-redundant SSU Ref database (v.138)82 and assigned to an approximate taxonomic affiliation (nearest taxonomic unit, NTU) using PhyloFlash v3.083 (http://github.com/HRGV/phyloFlash). Bacteriophage prediction was based on identifying viral signals in the metagenomic-assembled contigs (described below) using VirSorter84. In brief, viral-like genes were identified against a curated virome database84 and a set of single-amplified viral genomes85. Abundance of viral contigs was estimated by recruitment of metagenomic reads to viral contigs and calculation of contig coverage. Open reading frames (ORFs) were detected and translated with Prodigal v.2.6.386. Taxonomic classification of the translated sequences was based on sequence homology search87 against the Uniref 100 viral database (http:// virome.dbi.udel.edu; e-value < 10−5) and used to obtain taxonomy classification of viral contigs with the anvi-import-taxonomy function from Anvi’o v.5.288. The metagenomic reads were mapped to the obtained viral contigs using Bowtie 289 (local alignment, sensitive setting). Coverage of viral contigs was calculated by metagenomic read recruitment using Anvi’o. Alpha- and beta-diversity analyses of 16S rRNA amplicons and extracted miTAGs. All statistical analyses were carried out in R v3.5.3. Data manipulation was performed using the R package tidyverse and all visualizations were made using ggplot2. Community richness and beta-diversity was calculated using the R packages Phyloseq90 and Vegan v2.5-691. In total, nine samples representing a triplicate of depth profiles were used for downstream diversity analysis of ASVs (Supplementary Fig. 3, Supplementary Data 3). Rarefaction curves were con- structed to confirm that sequencing depth adequately captured richness in each sample and rarefied using the Phyloseq rarefy_even_depth function with a sample size of 15,400, which represented the minimum sequencing depth to retain 100% of samples used for downstream analysis. Observed richness (counts) and estimated richness (Chao1) was calculated using the estimate richness function in Phyloseq. Normality of the distribution of alpha-diversity estimates was confirmed using a Shapiro-Wilk test and a one-way analysis of variance (ANOVA) to test for sig- nificant differences in richness across depth profiles. As a post-hoc, a Tukey multiple comparison of means was used to confirm which pairs of sites showed significant differences. For beta-diversity analysis on amplicon and miTag data, Bray Curtis distance matrices were calculated in Vegan and visualized using a principal coordinate analysis (PcoA). Independent permutational analysis of var- iance (PERMANOVA) based on the Bray-Curtis dissimilarities values were cal- culated with the adonis function in Vegan (999 random permutations), to test for significant differences in community structure between depth profiles. Finally, a beta-dispersion test (PERMDISP) was applied to confirm that observed differences were not influenced due to dispersion. As a post-hoc evaluation of taxa responsible for differences in microbial community structure, we performed an indicator species analysis. We used the indicator value method92 to calculate indicator values using the R package indicspecies. An individual ASV was considered a valid indicator species if the p value was < 0.05 and the Test statistic (the indicator value) was 0.5 or greater, based on 1000 random permutations93. IndVals were compared between two groups, basal layer (30 m) and mid-column samples (180 m and 330 m), with the multipatt function in the R Indicspecies package (with the option control = how(nperm = 999)). This function uses an extension of the original Indicator Value method: it looks for indicator species of both individual site groups and combinations of site groups94. Counts per NTU (at species-level resolution) of extracted miTAGs were used for comparative analyses between communities under the Ross Ice Shelf and other oceanic samples. Only bacterial and archaeal species with >4 reads per sample were included in the analyses. Samples were divided into four groups, according to sampling depth or location: below-shelf ocean cavity (depth 30–330 m, n = 9), epipelagic (depth <200 m, n = 169), mesopelagic (depth ~200–1000 m, n = 60), and bathypelagic (depth 1000–4000 m, n = 54). The Vegan function vegdist was used to calculate a Bray-Curtis dissimilarity matrix between all samples, which was visualized by hierarchical cluster analysis (average linkage method, function hclust in Vegan). Significant differences (p < 0.05) between relative abundances of taxa from deep (>200 m) open ocean communities worldwide and below-shelf communities were confirmed using a non-parametric one-way analysis of variance (Kruskal-Wallis test, function kruskal.test() in R base). The following comparisons were restricted to two groups from deep, polar environments: samples from mesopelagic and bathypelagic polar environments (n = 42) and samples from the below-shelf cavity (n = 9). As distance-based multivariate methods can confound the within- and between-group effect size and fail to account for the mean variance relationship95, a generalized linear model (GLM) approach was used via the R package mvabund96. A multivariate model was fitted using the manyglm function and negative binominal distribution. To test the multivariate hypothesis of whether species composition varied across sub-ice and open water, the anova function was used which performed an analysis of deviance using likelihood ratio tests (LRT) and PIT-trap resampling of p values using 1000 iterations. To further examine which taxa contribute to compositional changes, a series of univariate tests were performed on each taxon using the p.uni = “adjusted” argument in the anova function. IndVal values were also calculated, using the same parameters described above, to identify which species contributed most to the differences between sub-ice environments and deep open ocean waters, Further, an additional post hoc test for between-group differences was performed with analysis of similarity percentages (simper97,) on a Bray-Curtis dissimilarity matrix calculated as described above. Metagenomic assembly and binning. For assembly, metagenome paired-end reads were error corrected using Bayes Hammer implemented in SPAdes v.3.0.068, merged with BBmerge v.36.3298 and normalized to a kmer depth of 42 with BBnorm v.36.32, from the BBtools program suite. Co-assembly of metagenomes was performed with MEGAHIT v.1.1.199 with merged and unmerged reads. Metagenomic reads were mapped back to the co-assembly (min. length 1 kb) using BBmap v.36.32100 to calculate differential coverage across all samples. Contigs were binned with MetaWatt v.3.5.3101, MaxBin v.2.2.7102 and MetaBAT v.2.12.1103. Bins were automatically de-replicated and aggregated with DasTool104, then manually inspected and refined with Anvi’o v.5.288. Bins classified as Archaea, Gammaproteobacteria, Deltaproteobacteria, Gemmatimonadota, Actinobacteriota, and Chloroflexota were selected from the bulk co-assembly and used for read recruitment with a minimum identity of 70% using BBmap v.36.32. This led to less complex subsets of reads for subsequent re- assembly with a more thorough assembler (SPAdes). For each taxonomic group a separate re-assembly with SPAdes v.3.0.0 was performed followed by a new round of binning as described above and manual refinement in Anvi’o. This procedure NATURE COMMUNICATIONS | (2022) 13:117 | https://doi.org/10.1038/s41467-021-27769-5 | www.nature.com/naturecommunications 11 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-27769-5 improves assembly (i.e., number of scaffolds reduced) and consequently bin metrics such as contig length and purity of bins105. Completeness and quality of final assemblies were assessed by CheckM v.1.0.770, with bins with >50% completeness and <5% contamination (i.e., high and medium quality bins) retained for further analysis25. Genome de-replication, classification, and phylogenetic analysis. Metagenomic bins and single-cell-assembled genomes with >50% completeness were defined as MAGs and SAGs, respectively, and collectively as ‘genomes’ for simplicity. Com- parison and de-replication of genomes were performed with dRep pipeline106. In brief, genomes were grouped at an average nucleotide identity (ANI) of 99%. Representative genomes from each cluster were selected based on the highest ‘genome score’106. This analysis provided a de-replicated genomic database of population genomes. BBmap and samtools were used to recruit reads from the metagenomes (97% identity), and Anvi’o was used to calculate the interquartile (Q2Q3) mean coverage of the de-replicated genomes across samples. On average, 50–60% of each sample’s metagenomic reads mapped to the metagenomic and SAG contigs. MAGs and SAGs were taxonomically assigned using the tool GTDBTk v.0.0.6 (release 80, www.github.com/Ecogenomics/GtdbTk) in accordance to the Genome Taxonomy Database107 (Supplementary Data 4). Phylogenetic tree construction for all 235 MAGS/SAGS was performed using ribosomal protein sequences retrieved from CheckM v.1.0.770 (Fig. 3). The concatenated marker sequence for each genome was aligned using MAFFT108 and an approximate maximum-likelihood phylogenetic tree was generated using FastTree 2109 with default parameters. The tree was then visualized and annotated using the web-based tool iTOL v.6 (https://itol.embl.de). Metabolic profiling of MAGs, SAGs, and assembled unbinned reads. ORFs in binned and unbinned contigs were predicted using Prodigal v.2.6.3.86, with default noise-cut-offs followed by manual filtering using HMM cut-off scores previously described110. The predicted ORFs were automatically annotated with the standard RAST annotation pipeline111, and against the Pfam (release 32.0)112 and TIGRfam (release 15.0)113 HMM models using Interproscan 5114. Phylogenetic trees were constructed to validate findings and to determine which protein classes / lineages were present in the Ross Ice Shelf (Supplementary Figs. 7–16). Trees were constructed for AmoA, NxrA, HbsT, RbcL, AclB, DsrA, Sqr, SoxB, CoxL, and the group 1 h [NiFe]-hydrogenase (HhyL). In all cases, protein sequences retrieved from the MAGs, SAGs, and metagenomic assembled reads by homology-based searches were aligned against a subset of reference sequences from a custom database containing 51 proteins (available at https://doi.org/10.26180/ c.5230745) using ClustalW in MEGA7115. Evolutionary relationships were visualized by constructing maximum-likelihood phylogenetic trees. Specifically, initial trees for the heuristic search were obtained automatically by applying Neighbor-Join and BioNJ algorithms to a matrix of pairwise distances estimated using a JTT model, and then selecting the topology with superior log likelihood value. All residues were used, and trees were bootstrapped with 50 replicates. Annotation of carbohydrate active enzymes (CAZymes) was performed by protein search against the CAZyme HMM database (dbCAN HMMdb release 8.0) following the dbCAN2 CAZyme annotation pipeline116, with stringent parameters for all CAZyme classes (E-value <1e−15 and coverage >0.35). We quantified the number of genes in each genome encoding for different glycosyl hydrolases (GH), glycosyl transferases (GT) and containing carbohydrate binding domains (CBD) (Supplementary Data 7). Heatmaps for the 50 genomes with highest GH diversity were generated in R with ggplot2 (Fig. 6), representing their abundance in the metagenome and the metatranscriptome (as described in the section below). Comparison of abundance and expression of assembled reads. To analyze the expression of annotated ORFs, pre-processed metatranscriptomic paired reads were merged with BBmerge98. Merged and unmerged non-rRNA sequences were mapped to the metagenomic and SAG contigs (99% id) with BBmap (on average, 60% of each sample’s reads were successfully assigned). Quantification of mapped reads per identified gene was performed with the function featureCounts of the R Subread package117. The transcript abundance of each ORF was converted to transcript per million (TPM) (Eq. (1)) for each sampled depth. TPM ¼ A (cid:2) 1=ΣA (cid:2) 106 where A = reads mapped to gene/gene length (kbp). ð1Þ To minimize systematic variability of individual gene abundance, the genome interquartile (Q2Q3) mean coverage (or, for unbinned contigs, the contig’s coverage) was used to define gene abundance in the metagenome. Gene coverage was then converted to counts per million (CPM), to allow for direct comparison with TPM. CPM ¼ B (cid:2) 1=ΣB (cid:2) 106 ð2Þ where B = gene coverage. Data from sample replicates were combined for the above calculations. metatranscriptomic reads were aligned using DIAMOND v0.9.24 to the 1 manually curated protein databases described above and to the predicted ORFs that matched the additional 10 HMMs described above (Supplementary Data 6). DIAMOND mapping was performed with a query coverage threshold of >80% and a gene specific threshold of 40% (RHO), 60% (AtpA, AmoA, MmoA, CoxL, NxrA, NuoF and RbcL), 75% (HbsT), 70% (PsbA, YgfK, ARO, IsoA), (80%) PsaA, or 50% (all other databases), with data further parsed to retain only group 1 and 2 [NiFe]- hydrogenase hits. For the metagenomic data, forward reads with at least 124 bp in length were used. For the metatranscriptomic data, paired-end reads were merged with BBMerge v38.51 and merged reads of at least 124 bp in length were used. Data from sample replicates were combined for this analysis. The abundance of each gene was converted to reads per kilobase million (RPKM). RPKM ¼ X=total sample reads (cid:2) 106 ð3Þ where X = reads aligned to a gene/ gene length (kbp). The gene abundances in RPKM from the metagenomic data were further used to estimate the proportion of the community encoding these functions. The processed metagenomic reads were aligned to each of the 14 universal single-copy ribosomal marker genes available in SingleM (https://github.com/wwood/singlem) with DIAMOND using a query coverage threshold of 80%. Alignments with a bitscore below 40 were removed; the alignment counts were converted to RPKM as described above and averaged across the 14 genes to represent the abundance of a universal single-copy gene. Metabolic gene RPKM values were divided by this value to obtain the average gene copies per organism in each sample (abundance relative to a single-copy gene). Heatmaps representing the community percentage (metagenomic data) and RPKM abundance (metatranscriptomic data) were generated in R with ggplot2 (Fig. 4b). Where genes within the same pathway are collapsed together, the values (community percentage or RPKM) are summed. Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article. Data availability The data and code underlying Fig. 2a, c, d are provided in the github repository https:// github.com/ClaMtnez/Ocean_tags. The data underlying Figs. 3, 4 & 5 and Supplementary Figs. 1, 4 & 5 are provided as a Supplementary Data Files. The sequence data generated in this study have been deposited in the EMBL Nucleotide Sequence Database (ENA) database under Bioproject PRJEB35712 (metagenomic and metatranscriptomic raw reads, metagenomic and metatranscriptomic assemblies, metagenomic assembled genomes, and single-cell amplified genomes) and in the NCBI Sequence Read Archive (SRA) under Bioproject PRJNA593264 (16S rRNA gene amplicon reads). The following public databases were used in this study: Swiss-Prot database, https://www.uniprot.org/, release-2018_10; Genome Taxonomy Database, https://gtdb.ecogenomic.org/, release 80; SILVA non-redundant SSU Ref database, https://www.arb-silva.de/, v.138; UniRef 100 VIROME database, http:// virome.dbi.udel.edu; Greening lab metabolic marker gene database, https://doi.org/ 10.26180/c.5230745; CAZyme HMM database, https://bcb.unl.edu/dbCAN2/, v.8.0; Pfam HMM database, http://pfam.xfam.org/, release 32.0; and TIGRFAM HMM database, https://www.ncbi.nlm.nih.gov/genome/annotation_prok/tigrfams/, release 15.0 Received: 13 January 2021; Accepted: 30 November 2021; References 1. Fretwell, P. et al. Bedmap2: improved ice bed, surface and thickness datasets for Antarctica. Cryosph 7, 375–393 (2013). 2. Holland, D. M., Jacobs, S. S. & Jenkins, A. Modelling the ocean circulation beneath the Ross Ice Shelf. Antarct. Sci. 15, 13–23 (2003). 3. Robinson, N. J., Williams, M. J. M., Barrett, P. J. & Pyne, A. R. Observations of flow and ice-ocean interaction beneath the McMurdo Ice Shelf, Antarctica. J. Geophys. Res. Ocean. 115, 1–10 (2010). 5. 4. Vick-Majors, T. J. et al. Biogeochemistry and microbial diversity in the marine cavity beneath the McMurdo Ice Shelf, Antarctica. Limnol. Oceanogr. 61, 572–586 (2016). Begeman, C. B. et al. Ocean Stratification and Low Melt Rates at the Ross Ice Shelf Grounding Zone. J. Geophys. Res. Ocean. 123, 7438–7452 (2018). Smethie, W. M. Jr & Jacobs, S. S. Circulation and melting under the Ross Ice Shelf: estimates from evolving CFC, salinity and temperature fields in the Ross Sea. Deep Sea Res. Part I Oceanogr. Res. Pap. 52, 959–978 (2005). 7. Michel, R. L., Linick, T. W. & Williams, P. M. Tritium and carbon-14 6. Metabolic profiling of unassembled metagenome and metatranscriptome reads. The abundance of particular metabolic functions independent of assembly was calculated as previously described118. Briefly, pre-processed metagenomic and 8. distributions in seawater from under the Ross Ice Shelf Project ice hole. Sci. (80-.) 203, 445–446 (1979). Post, A. L. et al. Modern sedimentation, circulation and life beneath the Amery Ice Shelf, East Antarctica. Cont. Shelf Res. 74, 77–87 (2014). 12 NATURE COMMUNICATIONS | (2022) 13:117 | https://doi.org/10.1038/s41467-021-27769-5 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-27769-5 ARTICLE 9. Clough, J. W. & Hansen, B. L. The Ross Ice Shelf Project. Sci. (80-.) 203, 433–434 (1979). 10. Azam, F. et al. Occurrence and metabolic activity of organisms under the Ross Ice Shelf, Antarctica, at station J9. Sci. (80-.) 203, 451–453 (1979). 11. Bruchhausen, P. M. et al. Fish, crustaceans, and the sea floor under the Ross Ice Shelf. Sci. (80-.) 203, 449–451 (1979). 12. Horrigan, S. G. Primary production under the Ross Ice Shelf, Antarctica1. Limnol. Oceanogr. 26, 378–382 (1981). 13. Stevens, C. et al. Ocean mixing and heat transport processes observed under the Ross Ice Shelf control its basal melting. Proc. Natl. Acad. Sci. USA 117, 16799–16804 (2020). 14. Reddy, T. E., Holland, D. M. & Arrigo, K. R. Ross ice shelf cavity circulation, residence time, and melting: Results from a model of oceanic chlorofluorocarbons. Cont. Shelf Res. 30, 733–742 (2010). 15. Gordon, L. I. et al. Seasonal evolution of hydrographic properties in the Ross Sea, Antarctica, 1996–1997. Deep Sea Res. Part II Top. Stud. Oceanogr. 47, 3095–3117 (2000). 16. Tovar-Sánchez, A. et al., Impacts of metals and nutrients released from melting multiyear Arctic sea ice. J. Geophys. Res. C Ocean. 115 (2010). 17. Biggs, D. C., Amos, A. F., Holm-Hansen, O., Oceanographic studies of epi-pelagic ammonium distributions: the Ross Sea NH4+ flux experiment. In Antarctic Nutrient Cycles and Food Webs. 93–103 (Springer Berlin Heidelberg, 1985). 18. Arístegui, J., Gasol, J. M., Duarte, C. M. & Herndld, G. J. Microbial oceanography of the dark ocean’s pelagic realm. Limnol. Oceanogr. 54, 1501–1529 (2009). 19. Logares, R. et al. Metagenomic 16S rDNA Illumina tags are a powerful alternative to amplicon sequencing to explore diversity and structure of microbial communities. Environ. Microbiol. 16, 2659–2671 (2014). 20. Sunagawa, S. et al. Structure and function of the global ocean microbiome. Sci. (80-.) 348, 1261359 (2015). 21. Zhang, W. et al. Structure and function of the Arctic and Antarctic marine microbiota as revealed by metagenomics. Microbiome 8, 1–12 (2020). 22. Salazar, G. et al. Gene expression changes and community turnover differentially shape the global ocean metatranscriptome. Cell 179, 1068–1083.e21 (2019). 23. Duarte, C. M. Seafaring in the 21st century: the malaspina 2010 circumnavigation. Exped. Limnol. Oceanogr. Bull. 24, 11–14 (2015). 40. Baltar, F., Arístegui, J., Gasol, J. M. & Herndl, G. J. Prokaryotic carbon utilization in the dark ocean: growth efficiency, leucine-to-carbon conversion factors, and their relation. Aquat. Microb. Ecol. 60, 227–232 (2010). 41. Baltar, F. et al, Significance of non-sinking particulate organic carbon and dark CO2 fixation to heterotrophic carbon demand in the mesopelagic northeast Atlantic. Geophys. Res. Lett. 37, L09602, https://doi.org/10.1029/ 2010GL043105. 42. Reinthaler, T., van Aken, H. M. & Herndl, G. J. Major contribution of autotrophy to microbial carbon cycling in the deep North. Atlantic’s Inter. Deep Sea Res. Part II Top. Stud. Oceanogr. 57, 1572–1580 (2010). 43. Tolar, B. B. et al. Contribution of ammonia oxidation to chemoautotrophy in Antarctic coastal waters. ISME J. 10, 2605–2619 (2016). 44. Priscu, J., Downes, M., Priscu, L., Palmisano, A. & Sullivan, C. Dynamics of ammonium oxidizer activity and nitrous oxide (N20) within and beneath Antarctic sea ice. Mar. Ecol. Prog. Ser. 62, 37–46 (1990). 45. Luria, C. M., Amaral-Zettler, L. A., Ducklow, H. W. & Rich, J. J. Seasonal succession of free-living bacterial communities in coastal waters of the Western Antarctic Peninsula. Front. Microbiol. 7, 1731 (2016). 46. Grzymski, J. J. et al. A metagenomic assessment of winter and summer bacterioplankton from Antarctica Peninsula coastal surface waters. ISME J. 6, 1901–1915 (2012). 2012 610. 47. Signori, C. N., Pellizari, V. H., Enrich-Prast, A. & Sievert, S. M. Spatiotemporal dynamics of marine bacterial and archaeal communities in surface waters off the northern Antarctic Peninsula. Deep Sea Res. Part II Top. Stud. Oceanogr. 149, 150–160 (2018). 48. Williams, T. J. et al. A metaproteomic assessment of winter and summer bacterioplankton from Antarctic Peninsula coastal surface waters. ISME J. 6, 1883–1900 (2012). 2012 610. 49. Swan, B. K. et al. Potential for chemolithoautotrophy among ubiquitous bacteria lineages in the dark ocean. Sci. (80-.). 333, 1296–1300 (2011). 50. Moran, M. A. & Durham, B. P. Sulfur metabolites in the pelagic ocean. Nat. Rev. Microbiol. 17, 665–678 (2019). 2019 1711. 51. Thomas, D. N. & Dieckmann, G. S. Antarctic sea ice-a habitat for extremophiles. Sci. (80-.) 295, 641–644 (2002). 52. Wolff, E. W. Ice sheets and nitrogen. Philos. Trans. R. Soc. B Biol. Sci. 368, 20130127 (2013). 53. Christner, B. C. et al. A microbial ecosystem beneath the West Antarctic ice sheet. Nature 512, 310–313 (2014) . 24. Tremblay, J. et al. Primer and platform effects on 16S rRNA tag sequencing. 54. Smith, J. A. et al, The marine geological imprint of Antarctic ice shelves. Nat. Front. Microbiol. 6, 771 (2015). 25. Bowers, R. M. et al. Minimum information about a single amplified genome (MISAG) and a metagenome-assembled genome (MIMAG) of bacteria and archaea. Nat. Biotechnol. 35, 725–731 (2017). T. G. S. Consortium. 26. Kitzinger, K. et al. Single cell analyses reveal contrasting life strategies of the two main nitrifiers in the ocean. Nat. Commun. 11, 767 (2020). 27. Pachiadaki, M. G. et al. Major role of nitrite-oxidizing bacteria in dark ocean carbon fixation. Sci. (80-.) 358, 1046–1051 (2017). 28. Cordero, P. R. F. et al. Atmospheric carbon monoxide oxidation is a widespread mechanism supporting microbial survival. ISME J. 13, 2868–2881 (2019). Islam, Z. F. et al. Two Chloroflexi classes independently evolved the ability to persist on atmospheric hydrogen and carbon monoxide. ISME J. 13, 1801–1813 (2019). 29. 30. Cantarel, B. L. et al. The Carbohydrate-Active EnZymes database (CAZy): an expert resource for glycogenomics. Nucleic Acids Res 37, D233–D238 (2009). 31. Yilmaz, P., Yarza, P., Rapp, J. Z. & Glöckner, F. O. Expanding the world of marine bacterial and archaeal clades. Front. Microbiol. 6, 1–29 (2016). 32. Cabello-Yeves, P. J. et al, Reconstruction of diverse verrucomicrobial genomes from metagenome datasets of freshwater reservoirs. Front. Microbiol. 8, 2131 (2017). 33. Youssef, N. H. et al, In silico analysis of the metabolic potential and niche specialization of candidate phylum “Latescibacteria” (WS3). PLoS One. 10, e0127499 (2015). 34. De Maayer, P., Anderson, D., Cary, C. & Cowan, D. A. Some like it cold: understanding the survival strategies of psychrophiles. EMBO Rep. 15, 508–517 (2014). 35. Barria, C., Malecki, M. & Arraiano, C. M. Bacterial adaptation to cold. Microbiology 159, 2437–2443 (2013). 36. Noell, S. E. & Giovannoni, S. J. SAR11 Bacteria have a high affinity and multifunctional glycine betaine transporter. Environ. Microbiol. 21, 2559–2575 (2019). Commun. 10, 5635 (2019). 55. Gifford, S. M., Sharma, S., Booth, M. & Moran, M. A. Expression patterns reveal niche diversification in a marine microbial assemblage. ISME J. 7, 281–298 (2013). 56. Polz, M. F., Hunt, D. E., Preheim, S. P. & Weinreich, D. M. Patterns and mechanisms of genetic and phenotypic differentiation in marine microbes. Philos. Trans. R. Soc. B Biol. Sci. 361, 2009–2021 (2006). 57. Azam, F., Smith, D. C. & Hollibaugh, J. T. The role of the microbial loop in Antarctic pelagic ecosystems. Polar Res 10, 239–244 (1991). Ji, M. et al. Atmospheric trace gases support primary production in Antarctic desert surface soil. Nature 552, 400–403 (2017). 58. 59. Shah, V., Chang, B. X. & Morris, R. M. Cultivation of a chemoautotroph from the SUP05 clade of marine bacteria that produces nitrite and consumes ammonium. ISME J. 11, 263–271 (2017). 60. Timmermann, R. & Hellmer, H. H. Southern Ocean warming and increased ice shelf basal melting in the twenty-first and twenty-second centuries based on coupled ice-ocean finite-element modelling. Ocean Dyn. 63, 1011–1026 (2013). Ingels, J. et al, Antarctic ecosystem responses following ice-shelf collapse and iceberg calving: Science review and future research. Wiley Interdiscip. Rev. Clim. Chang. 12, 12:e682 (2021). 61. 62. Knap, A. H., Michaels, A., Close, A. R., Ducklow, H. and Dickson, A. G. (eds) (1996): Protocols for the Joint Global Ocean Flux Study (JGOFS) Core Measurements. JGOFS Report Nr. 19, vi+170 pp. Reprint of the IOC Manuals and Guides No. 29, (UNESCO, 1994). 63. Parsons, T. R., Maita, Y., Lalli, C. M., A manual of chemical and biological methods for seawater analysis (Pergamon Press, Oxford, UK, 1984). 64. Smith, D. C. & Azam, F. A simple, economical method for measuring bacterial protein synthesis rates in seawater using 3H-leucine 1. Mar. Microb. Food Webs. 6, 107–114 (1992). 65. Kirchman, D. L. In Handbook of methods in aquatic microbial ecology (Lewis Publishers Boca Raton, FL, 1993), 58, 509–512. 37. Gutt, J. et al. Biodiversity change after climate-induced ice-shelf collapse in the Antarctic. Deep Sea Res. Part II Top. Stud. Oceanogr. 58, 74–83 (2011). 38. Nowald, N. et al, in Oceans 2009-Europe (IEEE, 2009). 39. Kirchman, D. L., Morán, X. A. G. & Ducklow, H. Microbial growth in the polar oceans - role of temperature and potential impact of climate change. Nat. Rev. Microbiol. 7, 451–459 (2009). 66. Simon, M. & Azam, F. Protein content and protein synthesis rates of planktonic marine bacteria. Mar. Ecol. Prog. Ser. 51, 201–213 (1989). 67. Lønborg, C. et al. Depth Dependent Relationships between Temperature and Ocean Heterotrophic Prokaryotic Production. Front. Mar. Sci. 3, 90 (2016). 68. Bankevich, A. et al. SPAdes: a new genome assembly algorithm and its applications to single-cell sequencing. J. Comput. Biol. 19, 455–477 (2012). NATURE COMMUNICATIONS | (2022) 13:117 | https://doi.org/10.1038/s41467-021-27769-5 | www.nature.com/naturecommunications 13 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-27769-5 69. Stepanauskas, R. et al.Improved genome recovery and integrated cell-size analyses of individual uncultured microbial cells and viral particles. Nat. Commun.8, 84 (2017). 101. Strous, M., Kraft, B., Bisdorf, R. & Tegetmeyer, H. E. The binning of metagenomic contigs for microbial physiology of mixed cultures. Front. Microbiol. 3, 410 (2012). 70. Parks, D. H., Imelfort, M., Skennerton, C. T., Hugenholtz, P. & Tyson, G. W. CheckM: assessing the quality of microbial genomes recovered from isolates, single cells, and metagenomes. Genome Res. 25, 1043–1055 (2015). 71. Woyke, T. et al. Assembling the marine metagenome, one cell at a time. PLoS 102. Wu, Y.-W., Simmons, B. A. & Singer, S. W. MaxBin 2.0: an automated binning algorithm to recover genomes from multiple metagenomic datasets. Bioinformatics 32, 605–607 (2016). 103. Kang, D. D., Froula, J., Egan, R. & Wang, Z. MetaBAT, an efficient tool for One 4, e5299 (2009). 72. Seemann, T. Prokka: rapid prokaryotic genome annotation. Bioinformatics 30, 2068–2069 (2014). 73. Poux, S. et al. On expert curation and scalability: UniProtKB/Swiss-Prot as a case study. Bioinformatics 33, 3454–3460 (2017). 74. Thompson, L. R. et al. A communal catalogue reveals Earth’s multiscale microbial diversity. Nat 551, 457–463 (2017). 2017 5517681. 75. Caporaso, J. G. et al. Ultra-high-throughput microbial community analysis on the Illumina HiSeq and MiSeq platforms. ISME J. 6, 1621–1624 (2012). 76. Parada, A. E., Needham, D. M. & Fuhrman, J. A. Every base matters: assessing small subunit rRNA primers for marine microbiomes with mock communities, time series and global field samples. Environ. Microbiol. 18, 1403–1414 (2016). 77. Apprill, A., McNally, S., Parsons, R. & Weber, L. Minor revision to V4 region SSU rRNA 806R gene primer greatly increases detection of SAR11 bacterioplankton. Aquat. Microb. Ecol. 75, 129–137 (2015). 78. Callahan, B. J. et al. DADA2: High-resolution sample inference from Illumina amplicon data. Nat. Methods 13, 581–583 (2016). 79. Bolyen, E. et al. Reproducible, interactive, scalable and extensible microbiome data science using QIIME 2. Nat. Biotechnol. 37, 852–857 (2019). 80. Ewels, P., Magnusson, M., Lundin, S. & Käller, M. MultiQC: summarize analysis results for multiple tools and samples in a single report. Bioinformatics 32, 3047–3048 (2016). 81. Kopylova, E., Noé, L. & Touzet, H. SortMeRNA: fast and accurate filtering of ribosomal RNAs in metatranscriptomic data. Bioinformatics 28, 3211–3217 (2012). 82. Quast, C. et al. The SILVA ribosomal RNA gene database project: improved data processing and web-based tools. Nucleic Acids Res 41, D590–D596 (2012). 83. Gruber-Vodicka, H. R., Seah, B. K. B. & Pruesse, E. phyloFlash: Rapid small- subunit rRNA profiling and targeted assembly from metagenomes. Msystems 5, e00920–e00920 (2020). 84. Roux, S., Enault, F., Hurwitz, B. L. & Sullivan, M. B. VirSorter: mining viral signal from microbial genomic data. PeerJ 3, e985 (2015). 85. Martinez-Hernandez, F. et al. Single-virus genomics reveals hidden cosmopolitan and abundant viruses. Nat. Commun. 8, 15892 (2017). 86. Hyatt, D. et al. Prodigal: prokaryotic gene recognition and translation initiation site identification. BMC Bioinforma. 11, 119 (2010). 87. Camacho, C. et al. BLAST+: architecture and applications. BMC Bioinforma. 10, 421 (2009). 88. Eren, A. M. et al. Anvi’o: an advanced analysis and visualization platform for ‘omics data. PeerJ 3, e1319 (2015). 89. Langmead, B. & Salzberg, S. L. Fast gapped-read alignment with Bowtie 2. Nat. Methods 9, 357–359 (2012). 90. McMurdie, P. J. & Holmes, S. phyloseq: an R package for reproducible interactive analysis and graphics of microbiome census data. PLoS One 8, e61217 (2013). 91. Oksanen, J. et al, vegan: Community Ecology Package. 2019. R package version 2.5-6 (2019). 92. Dufrêne, M. & Legendre, P. Species assemblages and indicator species: the need for a flexible asymmetrycal approach. Ecol. Monogr. 67, 345–366 (1997). 93. De Cáceres, M., Legendre, P., Wiser, S. K. & Brotons, L. Using species combinations in indicator value analyses. Methods Ecol. Evol. 3, 973–982 (2012). 94. De Cáceres, M., Legendre, P. & Moretti, M. Improving indicator species analysis by combining groups of sites. Oikos 119, 1674–1684 (2010). 95. Warton, D. I., Wright, S. T. & Wang, Y. Distance-based multivariate analyses confound location and dispersion effects. Methods Ecol. Evol. 3, 89–101 (2012). 96. Wang, Y., Naumann, U., Wright, S. T. & Warton, D. I. mvabund– an R package for model-based analysis of multivariate abundance data. Methods Ecol. Evol. 3, 471–474 (2012). 97. Clarke, K. R. Non‐parametric multivariate analyses of changes in community structure. Aust. J. Ecol. 18, 117–143 (1993). 98. Bushnell, B., Rood, J. & Singer, E. BBMerge–accurate paired shotgun read accurately reconstructing single genomes from complex microbial communities. PeerJ 3, e1165 (2015). 104. Sieber, C. M. K. et al. Recovery of genomes from metagenomes via a dereplication, aggregation and scoring strategy. Nat. Microbiol. 3, 836–843 (2018). 105. Albertsen, M. et al. Genome sequences of rare, uncultured bacteria obtained by differential coverage binning of multiple metagenomes. Nat. Biotechnol. 31, 533–538 (2013). 106. Olm, M. R., Brown, C. T., Brooks, B. & Banfield, J. F. dRep: a tool for fast and accurate genomic comparisons that enables improved genome recovery from metagenomes through de-replication. ISME J. 11, 2864–2868 (2017). 107. Parks, D. H. et al. A standardized bacterial taxonomy based on genome phylogeny substantially revises the tree of life. Nat. Biotechnol. 36, 996 (2018). 108. Katoh, K. & Standley, D. M. MAFFT multiple sequence alignment software version 7: improvements in performance and usability. Mol. Biol. Evol. 30, 772–780 (2013). 109. Price, M. N., Dehal, P. S. & Arkin, A. P. FastTree 2–approximately maximum- likelihood trees for large alignments. PLoS One 5, e9490 (2010). 110. Anantharaman, K. et al. Thousands of microbial genomes shed light on interconnected biogeochemical processes in an aquifer system. Nat. Commun. 7, 13219 (2016). 111. Aziz, R. K. et al. The RAST Server: Rapid Annotations using Subsystems Technology. BMC Genomics 9, 75 (2008). 112. El-Gebali, S. et al. The Pfam protein families database in 2019. Nucleic Acids Res 47, D427–D432 (2019). 113. Haft, D. H., Selengut, J. D. & White, O. The TIGRFAMs database of protein families. Nucleic Acids Res 31, 371–373 (2003). 114. Jones, P. et al. InterProScan 5: genome-scale protein function classification. Bioinformatics 30, 1236–1240 (2014). 115. Kumar, S., Stecher, G. & Tamura, K. MEGA7: molecular evolutionary genetics analysis version 7.0 for bigger datasets. Mol. Biol. Evol. 33, 1870–1874 (2016). 116. Zhang, H. et al. dbCAN2: a meta server for automated carbohydrate-active enzyme annotation. Nucleic Acids Res 46, 95–101 (2018). 117. Liao, Y., Smyth, G. K. & Shi, W. featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics 30, 923–930 (2013). 118. Ortiz, M. et al, bioRxiv, in press. 119. Scambos, T. A., Haran, T. M., Fahnestock, M. A., Painter, T. H. & Bohlander, J. MODIS-based Mosaic of Antarctica (MOA) data sets: Continent-wide surface morphology and snow grain size. Remote Sens. Environ. 111, 242–257 (2007). Acknowledgements We are grateful to D. V. Meier and P. M. Leung for insightful discussions. We thank the Victoria University of Wellington Hot Water Drilling Team led by A. Pyne and D. Mendeno for fieldwork support. Staff of the Bigelow Laboratory Single Cell Genomics Center are acknowledged for generating SAG data. We also thank L. Montiel and V. Balagué from the Institut de Ciències del Mar (ICM, CSIC) for extracting RNA, and the CNAG staff for RNAseq library preparation and sequencing. Bioinformatics analyses were performed at the LiSC Cluster (University of Vienna), MARBITS platform (ICM, CSIC) and the MonARCH HPC Cluster (Monash University). This research was facilitated by the New Zealand Antarctic Research Institute (NZARI) funded Aotearoa New Zealand Ross Ice Shelf Programme, the New Zealand Antarctic Science Platform ANTA1801, the Austrian science fond (FWF) project AP3430411/21 (FB) and a Rutherford Discovery Fellowship from the Royal Society of New Zealand (FB), the US National Science Foundation grants DEB-1441717 (RS) and OCE 1335810 (RS), the Simons Foundation Grant 827839 (RS), the Austrian Science Fund project P28781-B21 (GJH), the Spanish Ministry of Science and Innovation (Spanish State Research Agency, https://doi.org/10.13039/501100011033) fellowship RYC-2013-12554 (RL) and projects CTM2015-69936-P (RL) and PID2019-110011RB-C32 (JMG), the NHMRC EL2 Fel- lowship APP1178715 (CG) and Discovery Project grant DP180101762 (CG), the ARC SRIEAS Grant SR200100005 Securing Antarctica’s Environmental Future (SKB), and the H2020 MSCA Individual Fellowship 886198 (CMP). merging via overlap. PLoS One 12, e0185056 (2017). 99. Li, D. et al. MEGAHIT v1.0: A fast and scalable metagenome assembler driven by advanced methodologies and community practices. Methods 102, 3–11 (2016). 100. Li, H. et al. The Sequence Alignment/Map format and SAMtools. Bioinformatics 25, 2078–2079 (2009). Author contributions F.B., C.H., S.E.M., and C.O. designed field experiments. F.B., S.E.M., C.H., C.O., C.S., and B.T. performed field sampling and measurements. S.E.M. and R.L. performed nucleic acid extraction and library preparation for metagenomics and metatranscriptomics, respectively. R.S. provided single-cell amplified genome sequencing. C.M.P., Z.Z., R.J.L., 14 NATURE COMMUNICATIONS | (2022) 13:117 | https://doi.org/10.1038/s41467-021-27769-5 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-27769-5 ARTICLE S.K.B,. D.D.C., B.T., J.M.G., F.B., and C.G. analyzed the data. C.M.P., C.G., and F.B. wrote the manuscript with assistance from all coauthors. Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Competing interests The authors declare no competing interests. Additional information Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41467-021-27769-5. Correspondence and requests for materials should be addressed to Sergio E. Morales or Federico Baltar. Peer review information Nature Communications thanks Jeff Bowman and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/ licenses/by/4.0/. Reprints and permission information is available at http://www.nature.com/reprints © The Author(s) 2022 NATURE COMMUNICATIONS | (2022) 13:117 | https://doi.org/10.1038/s41467-021-27769-5 | www.nature.com/naturecommunications 15
null
10.1371_journal.pone.0240176.pdf
Data Availability Statement: All relevant data are within the manuscript and its Supporting Information files.
All relevant data are within the manuscript and its Supporting Information files.
RESEARCH ARTICLE Behavioral and corticosterone responses to carbon dioxide exposure in reptiles Daniel J. D. NatuschID Ain Isa5, Che Ku Zamzuri5, Andre GanswindtID 6,7, Dale F. DeNardo8 1,2☯*, Patrick W. Aust3,4☯, Syarifah Khadiejah5, Hartini Ithnin5, a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Natusch DJD, Aust PW, Khadiejah S, Ithnin H, Isa A, Zamzuri CK, et al. (2020) Behavioral and corticosterone responses to carbon dioxide exposure in reptiles. PLoS ONE 15(10): e0240176. https://doi.org/10.1371/journal.pone.0240176 Editor: Todd Adam Castoe, University of Texas at Arlington, UNITED STATES Received: July 2, 2020 Accepted: September 21, 2020 Published: October 6, 2020 Copyright: © 2020 Natusch et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the manuscript and its Supporting Information files. Funding: Daniel Natusch recieved funding from the Southeast Asian Reptile Conservation Alliance and the Swiss Federal Veterinary Office. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. 1 Department of Biological Sciences, Macquarie University, North Ryde, NSW, Australia, 2 EPIC Biodiversity, Frogs Hollow, NSW, Australia, 3 Department of Zoology, University of Oxford, Oxford, United Kingdom, 4 Bushtick Environmental Services, Grantham, Lincolnshire, United Kingdom, 5 Department of Wildlife and National Parks, Peninsular Malaysia, Kuala Lumpur, Malaysia, 6 Endocrine Research Laboratory, Mammal Research Institute, Department of Zoology and Entomology, Faculty of Natural and Agricultural Sciences, University of Pretoria, Pretoria, South Africa, 7 Centre of Veterinary Wildlife Studies, Faculty of Veterinary Science, University of Pretoria, Pretoria, Onderstepoort, South Africa, 8 School of Life Sciences, Arizona State University, Tempe, Arizona, United States of America ☯ These authors contributed equally to this work. * d.natusch@epicbiodiversity.com Abstract The use of carbon dioxide (CO2) exposure as a means of animal euthanasia has received considerable attention in mammals and birds but remains virtually untested in reptiles. We measured the behavioral responses of four squamate reptile species (Homalopsis buccata, Malayopython reticulatus, Python bivitattus, and Varanus salvator) to exposure to 99.5% CO2 for durations of 15, 30, or 90 minutes. We also examined alterations in plasma cortico- sterone levels of M. reticulatus and V. salvator before and after 15 minutes of CO2 exposure relative to control individuals. The four reptile taxa showed consistent behavioral responses to CO2 exposure characterized by gaping and minor movements. The time taken to lose responsiveness to stimuli and cessation of movements varied between 240–4260 seconds (4–71 minutes), with considerable intra- and inter-specific variation. Duration of CO2 expo- sure influenced the likelihood of recovery, which also varied among species (e.g., from 0–100% recovery after 30-min exposure). Plasma corticosterone concentrations increased after CO2 exposure in both V. salvator (18%) and M. reticulatus (14%), but only significantly in the former species. Based on our results, CO2 appears to be a mild stressor for reptiles, but the relatively minor responses to CO2 suggest it may not cause considerable distress or pain. However, our results are preliminary, and further testing is required to understand opti- mal CO2 delivery mechanisms and interspecific responses to CO2 exposure before endors- ing this method for reptile euthanasia. Introduction Ensuring the humane euthanasia of animals used by humans is critically important to fulfil our ethical obligation for compassion towards other species. In addition, a painless and dis- tress-free death can, in some contexts, result in a higher quality meat product for human PLOS ONE | https://doi.org/10.1371/journal.pone.0240176 October 6, 2020 1 / 14 PLOS ONE CO2 exposure in reptiles consumption [1]. In pursuit of these goals, methodologies, guidelines, and regulations for humane euthanasia have been developed and implemented for animal use ranging from meat production to scientific research [2]. However, a severe taxonomic bias currently exists. Although humane treatment protocols are well established for mammals and birds, the welfare needs of reptiles and the methodolo- gies considered humane and acceptable for euthanasia, especially in instances where human consumption of part of the carcass occurs, remain in their infancy [2]. For example, debate continues about the appropriateness of hypothermia (freezing) as an euthanasia method [3–5], and humane killing methods for reptiles used in the meat and skin industries were only adopted by the World Organization for Animal Health (OIE) in 2019 [see 6, 7]. Chemical agents offer an effective and humane way to euthanize reptiles, but their useful- ness is sometimes limited. Access and use restrictions, and situations where large numbers of animals are slaughtered for human consumption in short periods, often prohibit their use. With the possible exception of hypothermia, all recommended non-chemical methods of rep- tile euthanasia involve destruction of the brain (e.g., captive bolt, pithing). However, the effec- tiveness of brain destruction is vulnerable to operator error and may be impractical in situations where large numbers of animals need to be killed at one time. Carbon dioxide (CO2) is widely used as a euthanizing agent in the livestock industry and for scientific research [2, 8–10]. The guidelines of the American Veterinary Medical Associa- tion cite 86 studies on the effectiveness and suitability of CO2 as a humane means of euthanasia for mammals and birds [2]. Mammalian and avian responses to CO2 exposure vary consider- ably by species, and are dependent on CO2 concentration and delivery method [2, 8–10]. Mice, rats, cats, dogs, pigs, rabbits, chickens, and turkeys lose consciousness after 20–120 sec- onds of CO2 exposure, but may require exposures of 5–50 minutes to ensure death [2, 9, 10]. Exposure to CO2 has been shown to increase plasma corticosterone levels in rats and dogs and results in mouth gaping in mice, rats, and chickens [2, 9]. Rats and mink will actively avoid CO2 exposure if given the opportunity, but goats and chickens will not (despite the latter gap- ing when exposed; [2, 8]). The use of CO2 to euthanize reptiles has generally been discouraged by veterinary guidance, animals ethics committees, and by the OIE based on physiological considerations [2, 6, 11, 12]. The rationale implies that because reptiles have a variable metabolic rate and can potentially tolerate long periods without breathing or oxygen, they are vulnerable to the distressful effects of suffocation. However, to the best of our knowledge the argumentation against using CO2 to euthanize reptiles lacks empirical data and rests instead upon untested hypotheses and theoret- ical inference. Here, we examine the efficacy of CO2 to humanely euthanize squamate reptiles (lizards and snakes). Specifically, we tested the potential value of CO2 in (1) creating a low-stress, tempo- rary unconscious state to make physical methods of euthanasia safer and more efficient and (2) killing squamates outright. We used both behavioral responses and blood corticosterone concentrations (the primary glucocorticoid associated with stress in reptiles) to determine whether CO2 exposure provides a humane transition to unconsciousness and examined how duration of CO2 exposure influences the post-exposure duration of unconsciousness and like- lihood of death. Materials and methods Study species and locations Behavioral responses to CO2 exposure were examined in four species of reptile: reticulated pythons (Malayopython reticulatus); Burmese pythons (Python bivittatus); masked water PLOS ONE | https://doi.org/10.1371/journal.pone.0240176 October 6, 2020 2 / 14 PLOS ONE CO2 exposure in reptiles snakes (Homalopsis buccata); and Asian water monitors (Varanus salvator). These species are semi-aquatic to varying degrees and wide-ranging in Southeast Asia. The two python species grow to be large (> 5 m), while masked water snakes are relatively small (< 1.2 m). Asian water monitors are the world’s second largest lizard, growing to 3 metres in length and weigh- ing as much as 25 kg. In many instances, these species are commensal with humans and are regularly harvested and traded for their meat, skin, and medicinal value. In May 2019, we examined responses to CO2 in these reptiles in Malaysia (2˚14’N, 103˚ 03’E) and Thailand (17˚38’N, 100˚07’E) at two commercial facilities producing meat for human consumption and skins for the exotic leather trade. In Malaysia, free-roaming M. reti- culatus and V. salvator are legally collected from the wild by licensed hunters and brought to abattoirs for processing [13, 14]. Animals are kept alive at the facility for up to a week before being killed using a strong blow to the head followed by decapitation. No individual-based his- tory was available for the animals used in our study, and animals were held according to stan- dard commercial protocols (i.e., maintained individually in mesh bags with water provided intermittently). In Thailand, we examined specimens of M. reticulatus, P. bivittatus, and H. buccata. The two python species were captive-bred for commercial purposes following proto- cols described in Natusch and Lyons [15]. The H. buccata were wild-caught and temporarily held in large outdoor ponds with food provided. This research was undertaken with approval from the Animal Institutional Care and Use Committee of Arizona State University (protocol # 10-1689R). Experimental design—behavioral monitoring To assess behavioral responses of reptiles to CO2 exposure, we placed study animals individu- ally into 100 micron 375 mm x 500 mm clear plastic bags. Very large animals were double- bagged as a precaution. CO2 was supplied via 47 litre steel cylinders containing 99.5% CO2 and fitted with single-stage CO2 regulators. A 5 mm inside diameter CO2 supply hose was placed in the bag through the opening at the top, and the bag was sealed with an elastic band to limit but not eliminate the escape of gas. Bags were gently compressed around the body of the animal prior to CO2 admission to minimize residual air pockets. This design enabled CO2 to rapidly displace the limited amount of air present in the bag and thus minimized gas equili- bration time [16]. By using plastic bags instead of a rigid container, we were able to closely evaluate the animal during its exposure to CO2 (e.g., examine the animal’s righting response and its response to touch stimulation). CO2 flow was set to rapidly replace any existing air and then reduced to maintain positive CO2 pressure in the bag. For the longer exposure times, once the animal was unconscious, the flow of CO2 was stopped and the bagged was completely sealed. The process was similar for water monitors except that the bag was secured over their head rather than placing the entire body inside the bag (to minimize damage to the plastic bag by the lizard’s claws). We prevented monitors from perforating the bag during movements by gently placing a hand around the animal’s neck and preventing the forelimbs from contacting the bag. For some individuals this was not necessary and did not prevent observation of gen- eral body movements in response to CO2 exposure. For all individuals, the response of the ani- mal to CO2 exposure was recorded via direct visual examination until the animal was removed from the bag after the duration of CO2 exposure dictated by its assigned treatment group. For each animal, we recorded signs of consciousness and all behavioral responses to CO2, including movement, tongue flicking, and gaping. The animal’s behavior and body move- ments at the time of removal were recorded, as were changes in behavior over time and the eventual outcome (i.e., recovery or confirmed death). It was difficult to determine conscious- ness in many specimens. Although several individuals continued to respond to deep-touch PLOS ONE | https://doi.org/10.1371/journal.pone.0240176 October 6, 2020 3 / 14 PLOS ONE CO2 exposure in reptiles stimuli (e.g., a deep pinch of the tail), a lack of righting reflex (failure to turnover when placed upside down), corneal reflex in lizards, and cessation of breathing, strongly indicated that indi- viduals were unconscious despite exhibiting a muscular response to deep stimuli. Animals that reached a state indicative of imminent recovery of consciousness (i.e., voluntary movement often associated with tongue flicking) were euthanized using standard commercial practices (i.e., forceful blunt trauma to the dorsal surface of the head at the location of the brain case). Animals were deemed dead if no heartbeat and/or movements were detected (visually or via palpation) or by a lack of response to all stimuli (most notably a deep tail pinch) for up to one hour after removal from CO2 exposure. To test the effect of CO2 exposure duration on reptile responses, we first conducted a pre- liminary assessment using different exposure durations on five M. reticulatus (30 min, 60 min, 90 min, 120 min, or 180 min; n = 1 per duration). Based on related observations, we selected three CO2 exposure durations (15 min, 30 min, and 90 min) for the primary study. We used the results from the reticulated pythons to select exposure durations for the other species. As our results from M. reticulatus showed that 15 min was an insufficient duration, we began studies of other species with the 30 min exposure duration to minimise the number of animals used and to streamline efforts. If all specimens of the species failed to recover at this exposure duration, we assumed longer durations would achieve the same result, so did not conduct lon- ger duration trials. This was not true for H. buccata for which we did not complete the 90 min exposure treatment due to specimen availability and logistic constraints. We measured snout- vent length (SVL; using a steel tape measure) and body mass (using a digital scale) of each specimen while unconscious or dead, and then determined sex via direct inspection of the gonads upon dissection. Sample sizes for each species and their CO2 exposure times are pre- sented in Table 1. Air temperature was recorded to confirm constant temperatures throughout the course of study. Experimental design–sample collection for hormone monitoring We measured the effect of the CO2 euthanasia process on circulating corticosterone by collect- ing blood from seven M. reticulatus and seven V. salvator before and after CO2 exposure. Spec- imens were brought to the National Wildlife Forensic Laboratory, Department of Wildlife and National Parks Peninsular Malaysia. Sexes and body sizes are reported in Table 2. Each animal was kept individually within a mesh bag and secured within a plastic crate at ambient tempera- ture for two days before trials began. We collected 2 ml of blood from each individual within Table 1. Means, standard errors and ranges for snout-vent length (SVL) and body mass for reptile specimens used to examine behavioral responses to CO2 exposure. Species Thailand Malayopython reticulatus Python bivittatus Homalopsis buccata Malaysia Malayopython reticulatus Varanus salvator Sex M F M M M F M F N 1 3 18 11 12 14 5 5 https://doi.org/10.1371/journal.pone.0240176.t001 SVL (cm) Mass (g) N per exposure duration Mean Range Mean Range 15 min 30 min 90 min 273 265.3 ± 8.9 241.5 ± 2.7 104 ± 2.2 272.8 ± 8.6 297.4 ±8.3 63 ± 3.3 59 ± 3.8 - 255–283 220–263 93–116 238–331 255–374 50–68 52–71 8200 7200 ± 1790 6941 ± 545 686 ± 36 7335 ± 728 7878 ± 608 4990 ± 708 4000 ± 714 - 4200–10400 3900–11800 530–850 4550–13450 4050–12850 2250–6350 2550–6000 0 0 0 0 3 5 0 0 1 3 9 8 4 6 5 5 0 0 9 0 4 4 0 0 PLOS ONE | https://doi.org/10.1371/journal.pone.0240176 October 6, 2020 4 / 14 PLOS ONE CO2 exposure in reptiles Table 2. Means, standard errors and ranges for snout-vent length (SVL) and body mass for reptile specimens used to examine plasma corticosterone responses to CO2 exposure. Species Treatment Sex Malayopython reticulatus CO2 Varanus salvator Control CO2 Control https://doi.org/10.1371/journal.pone.0240176.t002 M F M F M F M F N 3 4 2 2 2 5 2 1 SVL (cm) Mass (g) Mean 246 ± 5.6 253.5 ± 4.6 295 ± 55 375 ± 25 53.7 ± 1.8 56.2 ± 2.9 79 ± 10 69 Range 235–255 240–260 240–350 350–400 51–57 47–63 69–89 - Mean 4720 ± 204 5280 ± 225 8500 ± 3500 35000 ± 0 2830 ± 233 2900 ± 370 7850 ± 2350 6500 Range 4400–5100 4720–5800 5000–12000 35000 2600–3300 1500–3750 5500–10200 - 90 seconds of removal from the mesh bag using a 22 gauge needle and 5 ml syringe inserted into the caudal vein at the base of the tail. The blood sample was then placed in a tube contain- ing lithium heparin (Vacuette #454084, Greiner Bio-One, Kremsmu¨nster, Austria). After blood collection, the same specimens were immediately exposed to CO2. A second blood sam- ple was collected from the same specimen after 15 minutes of CO2 exposure when the animal was unconscious. We did this by amputating the lower third of the tail and collecting the blood directly into a heparinized tube. The animal was then immediately euthanized following standard methods as described above. Blood samples were placed on ice until centrifugation to separate the plasma. We stored the isolated plasma samples at -20˚C until they were assayed. As confinement in the mesh bag may in itself result in elevated levels of corticosterone, we col- lected blood samples from several ‘control’ animals for comparison. The control water moni- tors (n = 3) were freshly killed wild animals harvested during a government sanctioned control program in Ladang Eng Tai, Malaysia (4˚57’N 100˚27’E). Animals were harvested using a 12-gauge shotgun at close range, with head shots resulting in near-instantaneous death. We collected blood from the severed tail of each animal within 90 seconds using the same method described above. Control reticulated python (n = 4) samples were obtained from captive-bred animals at a commercial reptile breeding facility outside Kuala Lumpur, Malaysia (2˚56’N 101˚53’E). The farm breeds high-value pythons for the pet trade, and general husbandry and welfare standards are high. Animals were selected based on size and relative docility (i.e., ease of handling), and blood samples were collected from the caudal vein within 90 seconds of removal from their enclosures using the same method described above. We recorded tempera- tures (27–30˚C) and kept all animals at approximately the same temperature both before and after exposure to CO2. This was not possible for control specimens sampled in the wild, but plasma corticosterone levels are not highly sensitive to body temperature in reptiles [17]. We obtained all blood samples over several hours on the same day to avoid diel and seasonal varia- tion in plasma hormone levels. Hormone analysis Immunoreactive plasma corticosterone concentrations were determined via an enzyme-linked immunosorbent assay (ELISA; ADI-900-097, Enzo Life Sciences, Farmingdale, NY) following the manufacturer’s instructions. This kit has been used in previous studies assessing plasma corticosterone concentrations in a variety of animal species, including alligators [18], birds [19], lizards [20] and turtles [21], but had not been previously documented for pythons or monitor lizards. Based on results from other species, we used a dilution ratio of 40:1. All sam- ples were run in duplicate format on a single assay plate. Results confirmed an average PLOS ONE | https://doi.org/10.1371/journal.pone.0240176 October 6, 2020 5 / 14 PLOS ONE CO2 exposure in reptiles difference between duplicates of less than 1.8% (mean: 1.73 ± 1.18%), and duplicate means were thus used in the analysis. Data analysis Our behavioral analysis measured the binary dependent variable of whether reptiles recovered after CO2 exposure or not. This metric was evaluated after different CO2 exposure durations for each species. For our corticosterone study we used a paired sample t-test to test for signifi- cant differences in plasma corticosterone concentrations before and after CO2 exposure. We used a one-way analysis of variance to test for differences in corticosterone level between the control animals and the pre-CO2 exposure samples from the study animals. Data were ln- transformed where needed to meet the normality and homogeneity of variance assumptions required for our parametric tests. All analyses were conducted in JMP Pro 14 (SAS Institute, Cary, NC). Results Behavioral observations Reticulated pythons (Malayopython reticulatus). After exposure to CO2, reticulated pythons remained still for 60–300 secs (1–5 mins) before tongue flicking and gaping (Fig 1). These responses eventually proceeded to slow and controlled whole-body movements; at this time snakes were responsive to touch through the bag. It was difficult to determine the point at which snakes lost full consciousness. However, we suspect that snakes lost consciousness, but continued to undergo unconscious movements including a response to touch stimuli. Between 240–1380 secs (4–23 mins) after CO2 exposure the snakes ceased all movements and lost responsiveness to stimuli (Fig 1). After the cessation of movement, but sometimes before, 18 of the 30 snakes exhibited mild muscle twitching of parts of their body. This twitching was unique to the reticulated pythons. All Malaysian reticulated pythons that were exposed to CO2 for 15 and 30 min eventually recovered (Fig 2). At the time of removal from the bag, none of these snakes had voluntary movements, but 7 of 8 snakes in the 15-min exposure group and 1 of 10 snakes in the 30-min group responded to a deep tail pinch with local movement. First voluntary movements occurred 4.9 ± 0.9 (mean ± SE) and 23.8 ± 4.7 min after removal from CO2 for the 15 min and 30-min exposure groups, respectively. In contrast, all reticulated pythons exposed to 90-min of CO2 did not recover, never having any reflex or voluntary movements (Fig 2). Reticulated pythons tested in Thailand that were exposed to CO2 for 30 min responded similarly to those in Malaysia, but one of the four snakes did not recover and, for those that did, recovery took 13.7 ± 3.7 min (42% faster than the 30-min exposure snakes in Malaysia). Burmese pythons (Python bivittatus). Burmese pythons showed similar behavioral responses to reticulated pythons, but took slighter longer to gape and lose responsiveness to stimuli (Fig 1). Burmese pythons also did not undergo muscle twitching and late-stage non- responsive (likely unconscious) movements were greater. All 8 snakes in the 30-min group responded to a deep tail pinch upon removal from the CO2, while none of the 90-min snakes responded. Two of the 8 snakes exposed to CO2 for 30 min and all of the snakes exposed to CO2 for 90 min did not recover (Fig 2). For the six 30-min snakes that did recover, it took 17.4 ± 2.5 min until they showed their first voluntary movements. Masked water snakes (Homalopsis buccata). The water snakes exposed to CO2 for 30 min showed behavioral responses that were very similar to those of the Burmese pythons, with no twitching but a considerable amount of unconscious movements. Mean time of first gape was about 120 secs (range: 60–420 secs, 1–7 min) and complete loss of consciousness was 300– PLOS ONE | https://doi.org/10.1371/journal.pone.0240176 October 6, 2020 6 / 14 PLOS ONE CO2 exposure in reptiles Fig 1. Variation in timing (in minutes) of key behavioural changes in (a) Malayopython reticulatus, (b) Python bivittatus, (c) Homalopsis buccata, and (d) Varanus salvator subject to carbon dioxide (CO2) exposure. Gaping: the time at which the mouth of the specimen opened. Unresponsive: the time the specimen had ceased movement and became unresponsive to stimuli. Thicker parts of the violin plots represent CO2 exposure times where the behaviour was most often observed. Note the different time scales represented on the x-axes of each panel. https://doi.org/10.1371/journal.pone.0240176.g001 840 secs (5–14 mins) after the onset of exposure (Fig 1). While all eight water snakes had a tail pinch reflex upon removal from the CO2, only two of the eight snakes recovered after 10 and 20 min, respectively. Water monitors (Varanus salvator). The water monitors showed the least behavioral response to exposure to CO2. The lizards exhibited no tongue flicking and no muscle twitching during the 30 min exposure. All monitors gaped within 240 secs (4 mins) of the onset of CO2 exposure (Fig 1) Both conscious and unconscious movements were limited in number and intensity with the last detected movements occurring 930 ± 66 secs (range: 720–1560 seconds) after the onset of exposure (Fig 1). All monitors lacked a tail pinch reflex when removed from the CO2, and they all failed to recover (Fig 2). Plasma corticosterone concentrations Corticosterone concentrations for the animals that did not go through the capture and con- finement associated with the trade prior to killing (i.e., ‘controls’) were significantly lower than those of the CO2-euthanized animals prior to CO2 exposure (pythons: 7.2 ± 1.3 ng/ml; F1,10 = 9.01, P = 0.015; monitors: 3.1 ± 0.7 ng/ml; F1,10 = 24.4, P < 0.001; Fig 3). Reticulated python PLOS ONE | https://doi.org/10.1371/journal.pone.0240176 October 6, 2020 7 / 14 PLOS ONE CO2 exposure in reptiles Fig 2. Percentage of Malayopython reticulatus, Python bivittatus, Homalopsis buccata, and Varanus salvator that recovered from different durations of CO2 exposure. X denotes treatments where no individuals recovered from CO2 exposure. Sample sizes appear above each column. https://doi.org/10.1371/journal.pone.0240176.g002 plasma corticosterone concentrations increased by 14% after CO2 exposure, (t0 = 11.8 ± 0.9 ng/ml vs t15 = 13.2 ± 0.4 ng/ml). However, this increasing trend was not statistically significant (matched pairs t-test: t6 = 2.23, P = 0.065; Fig 3). In contrast, CO2 exposure significantly increased plasma corticosterone concentrations in water monitors (by 18%; t0 = 9.6 ± 0.9 ng/ ml; t15 = 11.7 ± 0.8 ng/ml; t6 = 5.03, P = 0.02; Fig 3). Individual immunoreactive plasma corti- costerone concentrations before and after CO2 exposure were significantly correlated (pythons: n = 7; r2 = 0.61; P = 0.037; lizards: n = 8; r2 = 0.77; P = 0.009). Discussion Although available euthanasia methods for commercial reptile processing (e.g., brain destruc- tion) are humane, they can be vulnerable to operator error, are aesthetically displeasing, and are inefficient for rapidly processing numerous individuals. Our study provides initial results supporting the potential for carbon dioxide asphyxiation as an effective option for euthanizing PLOS ONE | https://doi.org/10.1371/journal.pone.0240176 October 6, 2020 8 / 14 PLOS ONE CO2 exposure in reptiles Fig 3. Mean plasma corticosterone concentrations (ng/ml) before and after 15 minutes of CO2 exposure and in control specimens (free-ranging or farmed; see text) of (a) Malayopython reticulatus and (b) Varanus salvator. Differences between corticosterone concentrations before and after CO2 exposure were not statistically significant for M. reticulatus, but were for V. salvator. Corticosterone concentrations between control specimens not subject to capture and handling are significantly lower than those captured from the wild for trade (although sample sizes were low; see text for details). Sample sizes for each group are reported directly above the x-axis. https://doi.org/10.1371/journal.pone.0240176.g003 reptiles in a variety of settings. Exposure to CO2 was effective for creating a temporary uncon- scious state at all exposure durations that was sufficient to safely and humanely employ a phys- ical method of euthanasia. Longer but still logistically practical exposures to CO2 were able to kill reptiles. The different taxa in our study varied subtly in their responses to CO2 exposure, both while conscious and after losing consciousness. For example, despite the similar body size of the two python species, the CO2 exposure duration required to induce unconsciousness in P. bivittatus was greater than M. reticulatus (Fig 1). The only lizard species in our study was rapidly ren- dered unconscious and did not recover from CO2 exposure durations that were unable to kill most of the snakes (Fig 2). Taxonomic differences and variation in metabolic rates may both be responsible for this difference [22–24]. The species we studied also differed in the effects PLOS ONE | https://doi.org/10.1371/journal.pone.0240176 October 6, 2020 9 / 14 PLOS ONE CO2 exposure in reptiles that a given duration of CO2 exposure had once the animal was removed from CO2, including the extent of involuntary/reflex muscle activity and the likelihood of death. Unfortunately, we did not have a sufficient sample size to examine sexual differences in species’ responses to CO2 exposure. Plausibly, CO2 may affect males and females differently, especially in those species with strong sexual dimorphism. Related to this, our study was undertaken on several of the world’s largest reptiles, all of which are semi-aquatic and can remain submerged under water for considerable periods. Application of CO2 exposure to the myriad of smaller-bodied rep- tiles, and to strictly terrestrial species, may yield different results. We made the assumption that because the density of CO2 is greater than air, complete (100%) CO2 saturation would occur as air was expelled from the small opening positioned at the top of the bag [25]. However, we did not directly measure the concentration of CO2 within the bag and whether the concentration was homogenous. Layering of CO2 could enable speci- mens to avoid exposure [2]. The variation in responses to CO2 exposure in our study may be related to minor but functionally significant difference in CO2 distribution [see 26]. In order to more broadly apply CO2 as a euthanasia method in reptiles, there needs to be a better under- standing of interspecific difference among taxa as well as a delivery system with established displacement parameters and sufficient holding capacity. Regardless of species, our behavioral observations suggest the reptiles used in our study do not suffer significant distress from CO2 exposure. Although our observational assessments were subjective, the body movements made by conscious reptiles were minor and appeared considerably less vigorous than the escape behavior displayed by these same animals when first removed from their holding bags. In the case of V. salvator, some specimens went unconscious without showing any signs of movement. Nevertheless, it is challenging to accurately deter- mine if reptiles are indeed dead, let alone feeling pain, based solely on behavioral responses [27, 28]. For example, an active heartbeat, involuntary movements, and response to touch sti- muli can continue for hours after complete destruction, pithing, and removal of the brain [Natusch unpubl. data 2020, 2]. Similarly, our data on the time reptiles take to lose responsive- ness are difficult to interpret. It was often unknown if specimens were consciously responsive, or unconscious and merely exhibiting involuntary muscular reflex. Importantly, the difficulty of assuring death, and the high but less than 100% effectiveness at killing at some CO2 expo- sure durations, may warrant the use of a secondary method to ensure death as is commonly used for chemical-induced euthanasia of research animals [see 2]. The most consistent behavioral response to CO2 exposure was the non-violent gaping dis- played by most (90%) individuals. Gaping is common in mammals and birds subject to CO2 exposure, and in birds does not appear to be a sign of distress when exposed to CO2 [29]. It is unknown whether gaping is a sign of significant distress in reptiles. Gaping occurred within 30 seconds to 16 minutes of initiating CO2 exposure and the timing varied among taxa (Fig 1). The short duration between initial exposure and gaping, and then unconsciousness, suggests that suffocation may not be the cause of death in reptiles exposed to CO2. All species used in our study are semi-aquatic, and capable of spending significant time underwater (>20 minutes), suggesting another physiological response is taking place. Despite the lack of behavioral indica- tors for stress and pain, reptiles take considerably longer to lose consciousness than mammals and birds [30–32]. Some consider a gentle death that takes longer is preferable to a rapid but more distressing death [26, 33]. In the context of CO2 and reptiles, further research is needed. Our additional approach to investigate the impact of CO2 exposure in our study species, by monitoring plasma corticosterone concentrations, also suggests that reptiles experience rela- tively minor distress from CO2 exposure. Comparison to our control (wild or farmed) speci- mens suggests the relative increase in stress involved in restraint and transportation of specimens to the laboratory was greater than the distress induced by CO2 exposure [2, 34]. PLOS ONE | https://doi.org/10.1371/journal.pone.0240176 October 6, 2020 10 / 14 PLOS ONE CO2 exposure in reptiles Brown tree snakes (Boiga irregularis) and red-sided garter snakes (Thamnophis sirtalis) cap- tured and placed in bags for 2–4 hours increased plasma corticosterone levels by 280–1200% [35, 36], but resulted in no appreciable increase in corticosterone concentrations in bearded dragons (Pogona barbata) [37]. Several studies reveal a lack of adverse impacts of corticoste- rone increase on survival, feeding behavior, and reproduction [38–40]. Other studies docu- ment invasive procedures (e.g., toe clipping, microchipping) inducing smaller corticosterone increases than did natural stresses experienced in the wild [27]. The relatively small increases in plasma corticosterone concentrations observed in pythons (14%) and lizards (18%) in our study may suggest that the functional relevance (distress or pain) of CO2 exposure-induced increases in corticosterone may be negligible. It is possible that the small increases in cortico- sterone levels we observed were related mostly to the stress caused by restraining and collect- ing an initial (T0) blood sample from each specimen, rather than by exposure to the CO2 itself. Alternatively, a post-CO2 exposure increase in corticosterone may have been suppressed because the recent capture, confinement, and handling had already maximized the hypotha- lamic-pituitary-adrenal (HPA) axis response. Intriguingly, exposure to CO2 may have additional benefits beyond the possibility of a pain- less death. After death, animals can have spinal cord induced muscle activity, and this can last for an extended duration in reptiles due to their tissue’s high tolerance of hypoxia. This phe- nomenon can lead to the impression that the animals is still alive [2], and thus has been capi- talized on by activists who oppose the consumption of animals, claiming they are being processed while still alive. In addition to being aesthetically displeasing, continued muscle movements after death force staff in commercial facilities to delay the harvesting of tissues for up to two hours after death [41]. When killed via CO2 exposure, we recorded no involuntary muscle movements after the presumed point of death, including during the processing of the reptiles. The physiological cause of this lack of muscle tone is unknown but, given its func- tional and cosmetic advantages, warrants further investigation. In conclusion, our study presents some of the first results on the effects of CO2 exposure in reptiles. We stress that our results are preliminary and therefore are reluctant to recommend CO2 as a humane method of reptile euthanasia at this time. Despite our results being generally positive, we identified some interspecific differences and methodological variables that may influence the effectiveness of CO2 exposure. Future studies could usefully disentangle the influence of these variables and employ alternative methods for assessing stress, pain, and death in reptiles (e.g., electroencephalography). Supporting information S1 Data. CO analyses. (XLSX) Acknowledgments We thank Yuan Wai Lek reptile trading company, Sisatchanalai python farm, and Lim Maju Jaya Trading for providing the animals used in this study. We also thank the Malaysian Department of Wildlife and National Parks Peninsular Malaysia for providing access to their forensic laboratory and equipment. We thank anonymous reviewers for comments that improved an earlier draft of this manuscript. Author Contributions Conceptualization: Daniel J. D. Natusch, Patrick W. Aust, Dale F. DeNardo. PLOS ONE | https://doi.org/10.1371/journal.pone.0240176 October 6, 2020 11 / 14 PLOS ONE CO2 exposure in reptiles Data curation: Daniel J. D. Natusch, Dale F. DeNardo. Formal analysis: Daniel J. D. Natusch, Andre Ganswindt, Dale F. DeNardo. Funding acquisition: Daniel J. D. Natusch. Investigation: Daniel J. D. Natusch, Patrick W. Aust, Syarifah Khadiejah, Hartini Ithnin, Ain Isa, Che Ku Zamzuri, Dale F. DeNardo. Methodology: Daniel J. D. Natusch, Patrick W. Aust, Syarifah Khadiejah, Hartini Ithnin, Andre Ganswindt, Dale F. DeNardo. Project administration: Daniel J. D. Natusch, Patrick W. Aust. Resources: Daniel J. D. Natusch, Hartini Ithnin, Ain Isa, Che Ku Zamzuri. Software: Daniel J. D. Natusch. Supervision: Daniel J. D. Natusch, Patrick W. Aust, Dale F. DeNardo. Validation: Daniel J. D. Natusch. Visualization: Daniel J. D. Natusch. Writing – original draft: Daniel J. D. Natusch, Patrick W. Aust, Andre Ganswindt, Dale F. DeNardo. Writing – review & editing: Daniel J. D. Natusch, Patrick W. Aust, Syarifah Khadiejah, Har- tini Ithnin, Ain Isa, Che Ku Zamzuri, Andre Ganswindt, Dale F. DeNardo. References 1. Chambers PG, Grandin T. Guidelines for humane handling, transport and slaughter of livestock. Heinz G, Srisuvan T, editors. Humane Society and International 2001. [accessed on 28 May 2020]. pp. 1–17. Available from: http://www.fao.org/3/a-x6909e.pdf. 2. Leary SL, Underwood W, Anthony R, Cartner S, Corey D, Grandin T, et al. AVMA Guidelines for the Euthanasia of Animals: 2020 Edition. American Veterinary Medical Association\. Schaumburg, IL. [(accessed on 30 May 2020)]. Available from: https://www.avma.org/sites/default/files/2020-01/2020- Euthanasia-Final-1-17-20.pdf. 3. Shine R, Amiel J, Munn AJ, Stewart M, Vyssotski AL, Lesku JA. Is “cooling then freezing” a humane way to kill amphibians and reptiles? Biol Open 2015; 4: 760–763. https://doi.org/10.1242/bio.012179 PMID: 26015533 4. Shine R, Lesku JA, Lillywhite HB. Assessment of the cooling-then-freezing method for euthanasia of amphibians and reptiles. J Am Vet Med Assoc 2019; 255: 48–50. https://doi.org/10.2460/javma.255.1. 48 PMID: 31194656. 5. Lillywhite HB, Shine R, Jacobson E, DeNardo DF, Gordon MS, Navas CA, et al. Anesthesia and eutha- nasia of amphibians and reptiles used in scientific research: should hypothermia and freezing be prohib- ited? Bioscience 2017; 67:53–61. https://doi.org/10.1093/biosci/biw143 6. Swiss Expert Panel. Analysis of humane killing methods for reptiles in the skin trade. Swiss Federal Vet- erinary Office. 2013. [(accessed on 30 May 2020)]. Available from: https://recht.pogona.ch/data/_ uploaded/file/3.0%20F%C3%BCtterung/5.2_BVET_Analysis%20on%20humane%20killing% 20methods%20for%20reptiles%20in%20the%20skin%20trade%20frm%5B1%5D.pdf. 7. World Organisation for Animal Health. Terrestrial Animal Health Code, Chapter 7.14. 2019 [accessed on 29 June 2020)]. Available from: https://www.oie.int/international-standard-setting/terrestrial-code/. 8. Withrock IC. The use of carbon dioxide (CO2) as an alternative euthanasia method for goat kids. M.Sc. Thesis, Iowa State University. 2015. Available from: https://lib.dr.iastate.edu/etd/14718. 9. Boivin GP, Hickman DL, Creamer-Hente MA, Pritchett-Corning KR, Bratcher NA. Review of CO2 as a euthanasia agent for laboratory rats and mice. J Am Assoc Lab Anim Sci. 2017; 56: 491–499. PMID: 28903819. 10. C¸ avuşoğlu E, Rault JL, Gates R, Lay DC. Behavioral Response of Weaned Pigs during Gas Euthanasia with CO2, CO2 with Butorphanol, or Nitrous Oxide. Animals 2020; 10: 787. https://doi.org/10.3390/ ani10050787 PMID: 32370086 PLOS ONE | https://doi.org/10.1371/journal.pone.0240176 October 6, 2020 12 / 14 PLOS ONE CO2 exposure in reptiles 11. Close B, Banister K, Baumans V, Bernoth EM, Bromage N, Bunyan J, et al. Recommendations for euthanasia of experimental animals: Part 2. Lab anim 1997; 31: 1–32. https://doi.org/10.1258/ 002367797780600297 PMID: 9121105. 12. Warren K. Reptile Euthanasia—No Easy Solution? Pac Conserv Biol 2014; 20: 25–27. https://doi.org/ 10.1071/PC140025 13. Khadiejah S, Razak N, Ward-Fear G, Shine R, Natusch DJD. Asian water monitors (Varanus salvator) remain common in Peninsular Malaysia, despite intense harvesting. Wildl Res 2019; 46: 265–275. https://doi.org/10.1071/WR18166 14. Natusch DJD, Lyons JA, Riyanto A, Mumpuni, Khadiejah S, Shine R. Detailed biological data are infor- mative, but robust trends are needed for informing sustainability of wildlife harvesting: A case study of reptile offtake in Southeast Asia. Biol Conserv 2019; 233: 83–92. https://doi.org/10.1016/j.biocon.2019. 02.016 15. Natusch DJD, Lyons JA. Assessment of python breeding farms supplying the international high-end leather industry. A report under the ‘Python Conservation Partnership’ programme of research. Occa- sional Paper of the IUCN Species Survival Commission 2014; 50. [(accessed on 15 May 2020)]. Avail- able from: https://portals.iucn.org/library/sites/library/files/documents/SSC-OP-050.PDF. 16. Lasiewski RC, Acosta AL, Bern-stein MH. Evaporative water loss in birds. 1. Characteristics of the open flow method of determination, and their relation to estimates of thermoregulatory ability. Comp Biochem Physiol 1966; 19:445–457. 17. Sykes KL, Klukowski M. Effects of acute temperature change, confinement and housing on plasma cor- ticosterone in water snakes, Nerodia sipedon (Colubridae: Natricinae). J Exp Zool 2009; 311A: 172– 181. https://doi.org/10.1002/jez.515 PMID: 19051318. 18. Finger JW Jr, Hamilton MT, Kelley MD, Stacy NI, Glenn TC, Tuberville TD. Examining the Effects of Chronic Selenium Exposure on Traditionally Used Stress Parameters in Juvenile American Alligators, Alligator mississippiensis. Arch Environ Contam Toxicol. 2019; 77(1):14–21. https://doi.org/10.1007/ s00244-019-00626-9 PMID: 30976886 19. Xie S, Romero LM, Htut ZW, McWhorter TJ. Stress responses to heat exposure in three species of Aus- tralian desert birds. Physiol Biochem Zool. 2017; 90(3):348–358. https://doi.org/10.1086/690484 PMID: 28384428 20. Seddon RJ, Hews DK. Populations of the Lizard, Sceloporus occidentalis, that differ in melanization have different rates of wound healing. J Exp Zool A Ecol Genet Physiol. 2016; 325(8):491–500. https:// doi.org/10.1002/jez.2033 PMID: 27597293 21. West JM, Klukowski M. Seasonal changes in baseline corticosterone, association with innate immunity, and effects of confinement in free-ranging Eastern Box Turtles, Terrapene carolina carolina. Gen Comp Endocrinol. 2018; 262:71–80. https://doi.org/10.1016/j.ygcen.2018.03.016 PMID: 29548757 22. Chappell MA, Ellis TM. Resting metabolic rates in boid snakes: allometric relationships and temperature effects. J Comp Physiol B 1987; 157: 227–235. https://doi.org/10.1007/BF00692367 PMID: 3106432. 23. Thompson G, Heger N, Heger T, Withers P. Standard metabolic rate of the largest Australian lizard, Varanus giganteus. Comp Biochem Physiol A Mol Integr Physiol 1995; 111: 603–608. 24. Hopkins W, Roe J, Philippi T, Congdon J. Standard and digestive metabolism in the banded water snake, Nerodia fasciata fasciata. Comp Biochem Physiol A Mol Integr Physiol 2004; 137: 141–149. https://doi.org/10.1016/j.cbpb.2003.09.017 PMID: 14720599. 25. Murray D. (2009). CO2 euthanasia methods for neonatal piglets. Allen D. Lemon Swine Conference; 2009 University of Minnesota: Minneapolis, MN United States of America. p. 114–116. [(accessed on 15 May 2020)]. Available from: http://hdl.handle.net/11299/139762. 26. Coenen AM, Drinkenburg WH, Hoenderken R, van Luijtelaar EL. Carbon dioxide euthanasia in rats: oxygen supplementation minimizes signs of agitation and asphyxia. Lab Anim 1995; 29: 262–268. https://doi.org/10.1258/002367795781088289 PMID: 7564209. 27. Langkilde T, Shine R. How much stress do researchers inflict on their study animals? A case study using a scincid lizard, Eulamprus heatwolei. J Exp Biol 2006; 209:1035– 1043. https://doi.org/10.1242/ jeb.02112 PMID: 16513929. 28. Eatwell K. Options for analgesia and anaesthesia in reptiles. In Pract 2010; 32: 306–311. https://doi.org/ 10.1136/inp.c3917 29. McKeegan DEF, McIntyre J, Demmers TGM, Wathes CM, Bryan Jones R. Behavioural responses of broiler chickens during acute exposure to gaseous stimulation. Appl Anim Behav Sci 2006; 99: 271– 286. https://doi.org/10.1016/j.applanim.2005.11.002 30. Hewett TA, Kovacs MS, Artwohl JE, Bennett BT. A comparison of euthanasia methods in rats, using carbon dioxide in prefilled and fixed flow-rate filled chambers. Lab Anim Sci 1993; 43: 579–582. PMID: 8158983. PLOS ONE | https://doi.org/10.1371/journal.pone.0240176 October 6, 2020 13 / 14 PLOS ONE CO2 exposure in reptiles 31. Shea SA, Harty HR, Banzett RB. Self-control of level of mechanical ventilation to minimize CO2- induced air hunger. Respir Physiol 1996; 103: 113–125. https://doi.org/10.1016/0034-5687(95)00086-0 PMID: 8833543. 32. Reed B, Varon J, Chait BT, Kreek MJ. Carbon dioxide-induced anaesthesia result in a rapid increase in plasma levels of vasopressin. Endocrinology 2009; 150: 2934–2939. https://doi.org/10.1210/en.2008- 1408 PMID: 19213839 33. Hawkins P, Playle L, Golledge H, Leach M, Banzett R, Coenen A, et al. Newcastle Consensus Meeting on Carbon Dioxide Euthanasia of Laboratory Animals; Newcastle upon Tyne, UK. 27–28 February 2006; [(accessed on 28 May 2020)]. pp. 1–17. Available from: www.nc3rs.org.uk/downloaddoc.asp?id= 416&page=292&skin=0. 34. Sharp J, Azar T, Lawson D. Comparison of carbon dioxide, argon, and nitrogen for inducing uncon- sciousness or euthanasia of rats. J Am Assoc Lab Anim Sci 2006; 45: 21–25. PMID: 16542038. 35. Moore IT, Lemaster MP, Mason RT. Behavioural and hormonal responses to capture stress in the male red-sided garter snake, Thamnophis sirtalis parietalis. Anim Behav 2000; 59: 529e534. https://doi.org/ 10.1006/anbe.1999.1344 PMID: 10715174. 36. Mathies T, Felix TA, Lance VA. Effects of trapping and subsequent short-term confinement stress on plasma corticosterone in the brown treesnake, Boiga irregularis on Guam. Gen Comp Endocrinol 2001; 124(1):106–114. https://doi.org/10.1006/gcen.2001.7694 PMID: 11703076. 37. Cree A, Amey AP, Whittier JM. Lack of consistent hormonal responses to capture during the breeding season of the bearded dragon, Pogona barbata. Comp Biochem Physiol 2000; 126: 275–285. https:// doi.org/10.1016/j.yhbeh.2005.08.004 PMID: 16153645. 38. Cote J, Clobert J, Meylan S, Fitze P. Experimental enhancement of corticosterone levels positively affects subsequent male survival. Horm Behav 2006; 49: 320–327. https://doi.org/10.1016/j.yhbeh. 2005.08.004 PMID: 16153645. 39. Thaker M, Vanak AT, Lima SL, Hews DK. Stress and aversive learning in a wild vertebrate: the role of corticosterone in mediating escape from a novel stressor. Am Nat 2010; 175: 50–60. https://doi.org/10. 1086/648558 PMID: 19922261. 40. Dupoue´ A, Angelier F, Brischoux F, DeNardo DF, Trouve´ C, Parenteau C, et al. Water deprivation increases maternal corticosterone levels and enhances offspring growth in the snake, Vipera aspis. J Exp Biol 2016; 219: 658–667. https://doi.org/10.1242/jeb.132639 PMID: 26747902. 41. Aust PW, Webb GJW, DeNardo DF, Natusch DJD. Welfare Principles for Snakes and Monitor Lizards in the Southeast Asian Skin Trade–a guide for stakeholders. Swiss Federal Veterinary Office, Switzer- land; Southeast Asian Reptile Conservation Alliance, France 2019. PLOS ONE | https://doi.org/10.1371/journal.pone.0240176 October 6, 2020 14 / 14 PLOS ONE
null
10.1073_pnas.2220576120.pdf
Data, Materials, and Software Availability. Sequencing data is available at National Center for Biotechnology Information Gene Expression Omnibus under accession GSE214456 (52). All other data are included in the manuscript and/ or SI Appendix.
null
RESEARCH ARTICLE | GENETICS OPEN ACCESS Derepression of Y-linked multicopy protamine-like genes interferes with sperm nuclear compaction in D. melanogaster Jun I. Parka,b,c , and Yukiko M. Yamashitad,e,f,1 , George W. Belld Edited by Mariana Wolfner, Cornell University, Ithaca, NY; received December 3, 2022; accepted March 17, 2023 Across species, sperm maturation involves the dramatic reconfiguration of chromatin into highly compact nuclei that enhance hydrodynamic ability and ensure paternal genomic integrity. This process is mediated by the replacement of histones by sperm nuclear basic proteins, also referred to as protamines. In humans, a carefully bal- anced dosage between two known protamine genes is required for optimal fertility. However, it remains unknown how their proper balance is regulated and how defects in balance may lead to compromised fertility. Here, we show that a nucleolar pro- tein, modulo, a homolog of nucleolin, mediates the histone-to-protamine transition during Drosophila spermatogenesis. We find that modulo mutants display nuclear compaction defects during late spermatogenesis due to decreased expression of auto- somal protamine genes (including Mst77F) and derepression of Y-linked multicopy Mst77F homologs (Mst77Y), leading to the mutant’s known sterility. Overexpression of Mst77Y in a wild-type background is sufficient to cause nuclear compaction defects, similar to modulo mutant, indicating that Mst77Y is a dominant-negative variant interfering with the process of histone-to-protamine transition. Interestingly, ectopic overexpression of Mst77Y caused decompaction of X-bearing spermatids nuclei more frequently than Y-bearing spermatid nuclei, although this did not greatly affect the sex ratio of offspring. We further show that modulo regulates these protamine genes at the step of transcript polyadenylation. We conclude that the regulation of protamines mediated by modulo, ensuring the expression of functional ones while repressing dominant-negative ones, is critical for male fertility. protamine | spermatogenesis | Drosophila In many species, spermatids undergo the process of nuclear compaction, an essential process to produce sperm that are capable of fertilization (1–3). Nuclear compaction is critical for the sperm’s hydrodynamic performance and protecting the paternal genome against mutagens (4–6). Nuclear compaction involves the dramatic chromatin reorgan- ization mediated by the histone-to-protamine transition (1–5, 7, 8). Sperm nuclear basic proteins, also referred to as protamines, are small, positively charged proteins that replace histone-based nucleosomes to achieve the extreme degree of DNA compaction often seen in sperm (2). As such, these protamines are required for fertility across many different species (4). Although protamines are essential for fertility, they are rapidly evolving across species (4, 9, 10), where the primary sequence, the number, and the functionality of protamine genes are not well conserved. For example, human and mouse protamine genes, PRM1 and PRM2, are required for fertility (4, 6, 7), while PRM2 has become nonfunctional in bulls and boars (4, 11). Closely related Drosophila species have independently evolved many different protamine-like genes (10): Drosophila melanogaster has Mst35Ba and Mst35Bb (also known as ProtA and ProtB), which are the most similar to mammalian PRM1 and PRM2 (3, 12, 13), as well as Mst77F, Prtl99C, and Y-linked multicopy Mst77Y, with evidence that several more uncharacterized genes may also be involved (14). In contrast, in Drosophila simulans, there is just one orthologous copy of the ProtA/B gene (Prot) as well as one ortholog each for Mst77F (GD12157) and Prtl99c (GD21472). D. simulans lacks Mst77Y (10, 14), but have evolved their own clade-specific genes that contain large regions of protamine sequences (Dox family genes), which are not present in D. melanogaster (15, 16). Surprisingly, while ProtA and ProtB are most similar to their mammalian counterparts, they are not required for fertility in D. melanogaster (12); instead, more divergent genes Mst77F and Prtl99C are essential (17–19). The potential function of the D. melanogaster-specific multicopy locus of Mst77F homologs (the Mst77Y genes) is unknown (20, 21). Interestingly, it has been observed that mammals appear to feature their own species-specific ratios of protamine dosage (2, 11, 22, 23), and in humans, even small alterations in the ratio of PRM1 and PRM2 are associated with infertility (2, 23–26), suggesting that a specific balance of protamines is important for sperm DNA packaging. However, it remains unknown Significance Protamines are small, highly positively charged proteins that are required for packaging DNA to produce mature sperm with highly condensed nuclei capable of fertilization. Even small changes in the dosage of protamines in humans is associated with infertility. Our work reveals the presence of dominant-negative protamine genes on the Y chromosome of Drosophila melanogaster and shows that the precise expression of functional protamines and repression of dominant-negative protamines is a critical process to ensure male fertility. Author affiliations: aLife Sciences Institute, University of Michigan, Ann Arbor, MI 48109; bDepartment of Cell and Developmental Biology, University of Michigan cMedical Medical School, Ann Arbor, MI 48109; Scientist Training Program, University of Michigan Medical School, Ann Arbor, MI 48109; dWhitehead Institute for Biomedical Research, Cambridge, MA 02142; eDepartment  of  Biology,  School of Science, Massachusetts  Institute  of Technology, Cambridge, MA 02142; and fHHMI, Cambridge, MA 02142 Author contributions: J.I.P. and Y.M.Y. designed research; J.I.P. performed J.I.P. contributed new reagents/analytic tools; J.I.P., G.W.B., and Y.M.Y. analyzed data; and J.I.P. and Y.M.Y. wrote the paper. research; The authors declare no competing interest. This article is a PNAS Direct Submission. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1To whom correspondence may be addressed. Email: yukikomy@wi.mit.edu. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2220576120/-/DCSupplemental. Published April 10, 2023. PNAS  2023  Vol. 120  No. 16  e2220576120 https://doi.org/10.1073/pnas.2220576120   1 of 8 why carefully balanced protamine expression is important and how it is achieved to support fertility. While studying D. melanogaster modulo mutants, we discovered that modulo transheterozygotic mutant causes misregulation of protamine genes. modulo mutant spermatids display decreased nuclear incorporation of protamine-like protein Mst77F and increased incorporation of its Y-linked homolog, Mst77Y, which is barely incorporated in the wild type, leading to a DNA com- paction defect that explains the reported sterility of modulo mutant. Our data indicate that Mst77Y likely acts as a dominant-negative form of Mst77F, interfering with the process of histone-to-protamine transition. Interestingly, Mst77Y has disproportionate effects on spermatids carrying an X chromosome, leading to biased decom- paction of X-bearing spermatid nuclei, although it does not lead to large effects on the sex ratio of offspring. We further find that modulo is involved in safeguarding polyadenylation of Mst77F transcript over that of the Y-linked Mst77Y. Our study reveals a mechanism of protamine gene expression mediated by modulo, balancing the correct ratio of protamine gene expression to ensure male fertility. Results modulo Mutant Is Defective in Sperm Nuclear Compaction. Modulo is the Drosophila homolog of Nucleolin, a nucleolar protein implicated in RNA processing (27, 28). Although modulo-mutant males have been known to be sterile (27, 29), the cytological defects that lead to their sterility have not been characterized. We find that the modulo transheterozygote mutant (mod L8/mod 07570) exhibits defects in nuclear morphology transformation during late spermiogenesis. In wild-type males, postmeiotic spermatid nuclei undergo well-documented morphological changes (1), from round spermatid stage, to “leaf ” stage, to “canoe” stage, resulting in highly condensed “needle-” stage nuclei, which is accompanied by the histone-to-protamine transition (Fig. 1A). Although modulo- mutant germ cells proceeded through spermatogenesis normally, including early nuclear compaction (Fig. 1 B and C), the modulo mutant exhibited striking “decompaction” of the nuclei after reaching the canoe stage, coinciding with the individualization of spermatids (Fig. 1 D and E). Immunofluorescence (IF) staining using anti-dsDNA, which has been previously used to assess the compaction status of spermatid nuclei (30), revealed that defective spermatid nuclei of modulo mutant are indeed decompacted (Fig. 1 F and G). Decompacting nuclei are initially negative via Terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL), a method used to identify DNA breaks that occur during apoptosis (Fig. 1 H and I), then become TUNEL positive (Fig. 1J), suggesting that decompaction is not the result of cell death, but may rather be a cause. Overall, 100% of the modulo-mutant testes exhibited a nuclear decompaction phenotype (Fig. 1K), and it appeared that all nuclei eventually become decompacted and die, filling the distal end of the testis with cellular debris (SI Appendix, Fig. S1 A and B). The eventual death of all sperm nuclei likely results in the entire lack of mature sperm in the seminal vesicles (SI Appendix, Fig. S1 C and D) and the modulo mutant’s known sterility. modulo Mutant Fails in Histone-to-Protamine Transition. Because nuclear decompaction in the modulo mutant occurs at stages when sperm chromatin is known to undergo reorganization through the histone-to-protamine transition, we explored whether the modulo mutant is defective in this process. Histone-to- protamine transition occurs step wise: 1) histone modification and removal, 2) transition protein incorporation then removal, and 3)  protamine incorporation (1). IF staining revealed that modulo-mutant spermatids undergo proper histone removal and transient transition protein incorporation (SI Appendix, Fig. S2 A– F), but fail to properly accumulate ProtA/B and Mst77F (Fig. 2 A and B). Moreover, using a specific antibody (SI Appendix, Fig. S3 A and B), we found that Mst77Y, Y-linked multicopy homologs of Mst77F (20, 21) (SI Appendix, Fig. S4A), strongly accumulated in modulo-mutant spermatid nuclei, whereas it was barely detectable in control (Fig.  2 C–G), suggesting that Mst77Y is aberrantly expressed in the modulo mutant. As the deletion of Mst77F and ProtA/B does not cause nuclear decompaction as severe as that of the modulo mutant (18), we infer that the incorporation of Mst77Y (in addition to the depletion of Mst77F and ProtA/B) causes the observed catastrophic nuclear decompaction seen in the modulo- mutant spermatids. Ectopic Expression of Mst77Y Alone Is Sufficient to Cause Nuclear Decompaction. The Mst77Y genes have several interesting features. First, the gene locus contains 18 copies of Mst77F homolog (SI Appendix, Fig. S4 A and B), which originated from a single event of Mst77F translocation to the Y chromosome, followed by gene amplification (20, 21). Second, many of the Mst77Y genes have mutations, which have resulted in changes in the position and number of critical arginine, lysine, and cysteine residues believed to be important for protamine function (4, 31). Other mutations have resulted in premature truncations (SI Appendix, Fig. S4B) (21). Note that anti-Mst77Y antibody was generated by using multiple peptides from Mst77Y that are distinct from Mst77F to increase specificity. The antibodies were also designed to be able to identify all copies of Mst77Y, which feature similar mutations and were tested to be able to identify both full-length Mst77Y (Mst77Y-12) and normally truncated Mst77Y (Mst77Y-3) (SI Appendix, Figs. S4 B and S5 A–C). Because Mst77Y’s mutations likely alter Mst77F’s normal function, we hypothesized that Mst77Y genes may function as a dominant-negative form of Mst77F. Accordingly, Mst77Y’s aberrantly high expression in the modulo mutant may interfere with the process of normal histone- to-protamine transition. To test the possibility that the expression of Mst77Y causes the nuclear decompaction phenotype, we generated transgenic lines that express Mst77Y under a male germline-specific tubulin pro- moter (β2-tubulin promoter) (32–34). From the 18 copies of Mst77Y homologs present on the Y chromosome (20, 21) we generated two lines expressing either Mst77Y-12 (a full-length version) or Mst77Y-3 (a truncated version due to premature stop codon) (SI Appendix, Figs. S4B and S6), as the transcripts of these two genes have been previously detected by qRT-PCR (21). Strikingly, expression of either Mst77Y-3 or Mst77Y-12 recapitu- lated a nuclear decompaction phenotype similar to that seen in modulo mutant (Fig. 3 A–D): 45.7% and 43.2% of testes exam- ined exhibited nuclear decompaction upon expression of Mst77Y-3 or Mst77Y-12, respectively (Fig. 3E), suggesting that high Mst77Y expression is sufficient to cause nuclear decompaction in a subset of spermatids. Notably, in contrast to the eventual decompaction of all spermatids seen in the modulo mutant, Mst77Y overexpres- sion alone does not cause sterility. We speculate that this might be due to the added effect of the decreased incorporation of Mst77F and ProtA/B, in addition to high Mst77Y incorporation, seen in the modulo mutant. Given that Barckmann et al. utilized the same promoter to overexpress autosomal Mst77F and did not observe such nuclear compaction defects during spermiogenesis (32) as we observed with Mst77Y overexpression, we infer that Mst77Y may act as a dominant-negative form, perhaps interfering with the function of Mst77F (Discussion). This notion is further supported by the 2 of 8   https://doi.org/10.1073/pnas.2220576120 pnas.org A Histone : Protamine exchange round spermatid Elongation leaf stage spermatocyte growth/ maturation mitoses GSCs meiotic divisions Seminal Vesicle canoe stage needle stage Nuclear Shaping, nuclear compaction, and individualization ) + / 0 7 5 7 0 d o m ( l o r t n o C ) 0 7 5 7 0 d o m / 8 L d o m ( t n a t u M B DAPI D DAPI phalloidin F N DAPI dsDNA H C DAPI TUNEL mod 07570/+ mod 07570/+ mod 07570/+ mod 07570/+ f o e g a t n e c r e P C E G I DAPI TUNEL J K g n i y a l p s i d s e t s e t e p y t o n e h p n o i t c a p m o c e d 100 50 0 *** 100% (105/105) 0% (0/67) control mod 07570/+ mutant modL8/mod 07570 DAPI TUNEL mod L8/mod 07570 mod L8/mod 07570 mod L8/mod 07570 mod L8/mod 07570 mod L8/mod 07570 Fig. 1. Sterility of modulo mutant is accompanied by defective spermatid chromatin compaction. (A) Schematic of spermatogenesis in Drosophila proceeding from germline stem cells to mature sperm. Proceeding from meiotic divisions onward, only nuclei are depicted. (B and C) Representative images of canoe-stage nuclei stained with DAPI (gray) in control males (mod07570/+) (B), and modulo-mutant males (modL8/mod07570) (C). (D and E) Representative images at the stage shortly before individualization stained with DAPI (gray) and phalloidin (cyan, to visualize the individualization complex) in control males (mod07570/+) (D) and modulo- mutant males (modL8/mod07570) (E). Although all nuclei eventually become decompacted in modulo-mutant males, individualization complex (marked by phalloidin staining) appears to be normally formed. Yellow arrowheads indicating decompacted nuclei. (F and G) Representative images of late canoe to needle stages stained with anti-dsDNA (red) and DAPI (gray) in control (mod07570/+) (F) and modulo-mutant males (modL8/mod07570) (G). N: needle-stage spermatids that do not stain for dsDNA due to advanced DNA compaction, C: canoe-stage spermatids that are less compact and positive for anti-dsDNA staining. (H–J) Staining via Terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL) (magenta) of needle-shaped spermatid cysts in control (mod07570/+) (H) and mutant (modL8/mod07570) males without (I) or with (J) TUNEL signal. (K) Percentage of decompaction phenotype in modulo-mutant vs. wild-type males. *** indicates P < 0.001 (unpaired Student’s t test assuming unequal variances in five independent experiments). n (total number of testes counted per genotype) is presented on the bar graph. fact that a truncated version (Mst77Y-3) also causes the decom- paction phenotype. Indeed, spermatid cysts of transgenic males expressing Mst77Y-3 exhibited uneven Mst77F staining, sug- gesting that some nuclei fail to accomplish proper Mst77F incor- poration (SI Appendix, Fig. S5 D and E). It is important to note that the nuclear decompaction was most prominently observed when males were raised in 25 °C after their parents were raised at 18 °C (Methods). Interestingly, using DNA Fluorescence in situ hybridization (FISH) to distinguish X- vs. Y-bearing sper- matids, we found that overexpression of Mst77Y results in biased demise of X-bearing spermatids, where 61.8% of decompacting nuclei were X-bearing, compared to only 38.2% being Y-bearing (Fig. 3 F and G). It is important to note that this bias is not due to differential efficiency of hybridization of X chromosome vs. Y chromosome DNA FISH probes: Leaf to canoe stage sperma- tids of control males (SI Appendix, Fig. S7 A and B), as well as leaf to canoe stage spermatids of modulo-mutant males (before they exhibit decompaction defects), exhibited ~50:50 ratio of X:Y signal (SI Appendix, Fig. S7 C and D), further suggesting that decompaction is biased toward X-bearing spermatids. However, a fertility assay revealed only a minor increase in the male progeny compared to sex chromosome–matched controls (51.8% vs. 47.8%, P = 0.0005) (SI Appendix, Fig. S8A). Likewise, only a small degree of sex ratio distortion was observed in modulo heterozygous mutant, compared to sex chromosome– matched control (SI Appendix, Fig. S8B) (Discussion). Together, these results suggest that Mst77Y acts as a dominant-negative form of Mst77F, interfering with the incorpo- ration of normal protamines and leading to spermatid nuclear decompaction. Modulo Promotes Polyadenylation of Autosomal Mst77F Transcript. How does modulo regulate the expression of Mst77F and Mst77Y? Modulo protein is expressed in the nucleolus of spermatogonia and spermatocytes, but is down-regulated prior to the meiotic divisions (Fig. 4 A and B), days earlier than the stages in which its mutant phenotype manifests. Protamine genes are known to be transcribed many days prior to the sperm nuclear compaction process in both flies and mice (3, 32, 35). Interestingly, we found that Mst77F transcripts colocalize with Modulo in the spermatocyte nucleolus, while Mst77Y transcripts localize adjacent to the nucleolus (Fig. 4C). These results prompted us to examine whether Mst77F and/or Mst77Y transcripts may be deregulated in modulo mutant. Indeed, we found that Mst77F messenger RNA (mRNA) is dramatically reduced in modulo mutant, whereas Mst77Y mRNA was increased approximately threefold using qRT-PCR of polyA-selected RNA (Fig.  4D). However, when total RNA was used for qRT-PCR or total RNA sequencing, we found that both Mst77F and Mst77Y transcripts were increased in modulo mutant (Fig. 4D and SI Appendix, Fig. S9A). RNA FISH, which visualizes total RNA, also confirmed the increase of both Mst77F and Mst77Y transcripts in modulo mutant (SI Appendix, PNAS  2023  Vol. 120  No. 16  e2220576120 https://doi.org/10.1073/pnas.2220576120   3 of 8 DAPI Prot A/B Mst77F DAPI Mst77F Prot A/B A B A′ A″ A‴ B′ B″ B‴ Control (mod 07570/+) Mutant (mod L8/mod 07570) G DAPI Mst77Y Basal End E Basal End ) + / 0 7 5 7 0 d o m ( l o r t n o C ) 0 7 5 7 0 d o m / 8 L d o m ( t n a t u M C 1 0.5 0 e g a t s e o n a c - e t a l f o n o i t r o p o r P Y 7 7 t s M r o f g n n i a t s i s t s y c d e p a h s - e l d e e n r o control mod 07570/+ mutant modL8/mod 07570 D D′ F F′ Fig. 2. Nuclear decompaction in modulo-mutant spermatids is associated with decreased incorporation of Mst77F and increased incorporation of Mst77Y. (A  and B) Representative images of late canoe-stage nuclei stained with DAPI (gray), anti-Prot A/B (cyan), and anti-Mst77F (magenta) in control (mod07570/+) (A) and mutant (modL8/mod07570) (B) males. Split channel view of DAPI (A' and B'), anti-Mst77F (A'' and B''), and anti-Prot A/B (A''' and B''') in control (A) and mutant (B) males. (C–F) Representative images of canoe-stage and needle-stage spermatids at the basal end of testis stained with DAPI (gray) and anti-Mst77Y (green) in control (mod07570/+) (C and D) and mutant (modL8/mod07570) (E and F) males. Split channel view of anti-Mst77Y in control (D') and mutant (F') males. Dotted lines outline the testis. (G) Proportion of canoe-stage cysts with nuclei positive for Mst77Y staining in mutant (modL8/mod07570) vs. control (mod07570/+) males. *** indicates P ≤ 0.001 (unpaired Student’s t test assuming unequal variances) with n=10 testes in control and n=9 testes in mutant males from 2 independent experiments. Exact P-values are listed SI Appendix, Table S1. Fig. S9B). Furthermore, total RNA-Seq and qRT-PCR did not detect any splicing defects of Mst77F or Mst77Y in modulo mutant (SI  Appendix, Fig.  S10). Collectively, these results suggest that Modulo specifically regulates transcripts of Mst77F and Mst77Y at the step of polyadenylation. Given that Modulo protein and Mst77F transcript colocalize in the nucleolus, we speculate that Mst77F is directly regulated by Modulo, whereas increased mRNA level of Mst77Y may be an indi- rect consequence of reduced functional Mst77F mRNA. Interestingly, RNA FISH using poly(T) probes revealed that poly(A) signal encir- cles the nucleolus in wild-type spermatocytes, whereas markedly less poly(A) was detected around the nucleolus in the modulo mutant (Fig. 4 E and F), further suggesting that modulo may function to facilitate polyadenylation of transcripts localized to the nucleolus. Our findings are consistent with the known importance of polyade- nylation to sperm-specific transcripts, such as protamines, which must be translationally repressed for long periods(36–39). Taken together, these results suggest that modulo plays an essential role in sperm nuclear compaction by facilitating maturation of canonical Mst77F transcript over that of Y-linked Mst77Y (Fig. 4G). Discussion The present study reveals a regulatory mechanism mediated by a nucleolar protein Modulo that balances the expression of protamine subtypes in D. melanogaster. This finding may represent a similar theme to what is seen in the fragile balance of PRM1 and PRM2 in mammalian fertility (2, 7, 24, 25). In the case of Mst77Y, Y-linked 4 of 8   https://doi.org/10.1073/pnas.2220576120 pnas.org A y w DAPI C 2-Tub-Mst77Y3/+ DAPI E l i g n y a p s d s i t s e t i B mod L8/mod 07570 DAPI D 2-Tub-Mst77Y12/+ DAPI f o n o i t r o p o r P F G X *** 95 (38.2%) Y 154 (61.8%) e p y t o n e h p n o i t c a p m o c e d 1.0 0.5 ns *** *** 0 0 UAS- Mst77Y-12/+ (no driver) 2-Tub- Mst77Y-3/+ 2-Tub- Mst77Y-12/+ n = 56 61 40 DAPI X Y X (TAGA) Y (AATAC) F′ F″ Fig.  3. Mst77Y overexpression is sufficient to cause nuclear decompaction and causes biased decompaction of X chromosome-bearing spermatids. (A–D) Representative images of needle-stage nuclei stained with DAPI (gray) showing normal morphology in control (y w) (A), decompaction phenotype in modulo- mutant males (modL8/mod07570) (B), transgenic males expressing Mst77Y-3 (truncated copy) (C) or Mst77Y-12 (full-length copy) (D) driven by β2-tubulin promoter. IF confirming overexpression shown in SI Appendix, Fig. S5 A–C. (E) Proportion of testes displaying decompaction phenotype in transgenic Mst77Y males. Control (UAS–Mst77Y-12/+) does not express Mst77Y-12 due to the absence of driver. *** indicates P ≤ 0.001 (unpaired Student’s t test assuming unequal variance), ns indicates P > 0.05, n = 56 testes in control, n = 61 testes in β2-tub–Mst77Y-3/+ condition, n = 40 in β2-tub–Mst77Y-12/+ condition from three independent experiments. (F) Representative images of DNA Fluorescence in  situ hybridization of decompacted spermatids in Mst77Y-3-expressing males using TAGA-Cy3 (magenta, X-specific probe) (F') and AATAC-Cy5 (cyan, Y-specific probe) (F''). (G) Percentage of decompacted haploid nuclei containing X chromosome vs. Y chromosome in Mst77Y-3-expressing males. *** indicates P(X ≥ 154) < 0.001 (exact binomial distribution) assuming P = 0.5 with n = 249 nuclei counted from three independent experiments. Exact P-values listed in Table S1. multicopy Mst77F homologs, our study suggests that they have the ability to act as dominant-negative protamines and thus must be carefully regulated/repressed. The present study also confirmed that Mst77Y genes are expressed as proteins as suggested previously by the finding that several of the copies contain complete open-reading frames (21) and is also consistent with small RNA sequencing reveal- ing that the Mst77Y locus is not a source of small RNAs (40). We showed that overexpression of Mst77Y dominantly inter- feres with Mst77F incorporation, leading to decompaction of sperm nuclei and their demise. Mst77Y genes feature differences from their autosomal homolog that further support the idea that they are dominant-negative versions of Mst77F and interfere with sperm chromatin compaction. Mst77Y-12, which retains the full ORF of Mst77F (SI Appendix, Fig. S4), exhibits 87% overall sequence homology to autosomal Mst77F. At the domain/motif level, the MST-HMG-box domain, suggested to be important for DNA binding (14), maintains 100% homology, while the coiled-coil motif and C-terminal domain maintain only ~79.5% and ~85% homology, respectively (SI Appendix, Fig. S6B). It has been shown that the N-terminal domain of Mst77F, which con- tains the coiled-coil motif, interacts with the C-terminal domain to induce multimerization to mediate DNA compaction (41). The changes to Mst77Y at important regions may thus influence the multimerization of protamines and the formation of proper sperm chromatin structure, by interfering with the ability of the canon- ical version to multimerize. The notion that Mst77Y behaves as a dominant-negative version of Mst77F is further supported by the fact that overexpression of Mst77Y-3, a truncated version which does not contain the C-terminal domain (SI Appendix, Fig. S6B), is still sufficient to cause defects in nuclear compaction (Fig. 3C). What is the potential “function” of dominant-negative pro- tamines? We propose a few nonmutually exclusive possibilities. First, dominant-negative protamines may participate in meiotic drive, as suggested by recent works in D. simulans (15, 16) as well as D. melanogaster (10). Indeed, our data suggest that Mst77Y has the ability to disproportionally affect X-bearing spermatids. While this did not result in a large sex ratio distortion in offspring (SI Appendix, Fig. S8), this ability to harm a subset of developing spermatids during postmeiotic development may indicate the pos- sibility that these protamine variants could be exploited by a meiotic drive system to unleash its own selfish purpose. Intriguingly, studies on the Winters sex-ratio meiotic drive system in D. simulans revealed that the driver, Dox, contains a large portion of the Protamine gene (15, 16). While it has not been confirmed whether this protamine-like region is essential for sex ratio distortion, the derepression of Dox does seem to cause nuclear defects during sper- miogenesis (42). We propose that a drive system that would be able to localize a dominant-negative protamine such as Mst77Y to a subset of spermatids containing one homolog over another could be quite successful at achieving bias. Alternatively, the dominant-negative version of a protamine may be utilized when spermatogenesis needs to be aborted (similar to the concept of “programmed cell death”), for example under stress conditions. In such a case, dominant-negative protamines (such as Mst77Y) can be up-regulated to lead to abortive spermatogenesis. In such a sce- nario, a dominant-negative protamine may have a beneficial func- tion for the organism. Yet another possibility that may contribute toward the rapid divergence of protamines is that protamine genes evolve to optimally package the genome, which may be greatly influenced by the composition of the most abundant sequences in a given genome, i.e., repetitive DNA such as satellite DNA. As these repetitive sequences are known to rapidly diverge across species (43), protamine genes may have to adapt to accommodate diverged repetitive DNA sequences, leading to rapid divergence and/or emer- gence of multiple protamine genes to optimally package different repetitive DNA with distinct structure/sequence. In such a scenario, PNAS  2023  Vol. 120  No. 16  e2220576120 https://doi.org/10.1073/pnas.2220576120   5 of 8 A A′ n i r a l l i r b i f p f g - d o M I P A D p f g - d o M B B′ B″ e g r e M p f g - d o M n i r a l l i r b i f DAPI Mod-gfp Mst77F Mst77Y Mod-gfp C C′ Mst77F Mst77Y C″ C‴ D n i e g n a h C d o F d e z i l l a m r o N l o r t n o c o t . l e r t n a t u m o u d o m l 5 4 3 2 1 0 RT-qPCR method: F 7 7 t s M A y o P l I P A D A y o p l F 7 7 t s M G p = 0.0166 p = 0.0166 p = 0.0248 p = 0.0248 p = 0.0021 p = 0.0021 p < 0.001 p < 0.001 0 0 Mst77F Mst77Y Mst77F Mst77Y Poly(A) selection Total RNA y w mod L8/mod 07570 E E′ F F′ E″ F″ Modulo AAAAA Mst77F Nucleolus Mst77Y AAAAA Fig. 4. Modulo localizes to the nucleolus and functions to promote polyadenylation of Mst77F. (A and B) Localization of Modulo to the nucleolus in the apical tip of the testis (A) and in the spermatocyte nuclei (B). Males expressing Modulotagged with Green fluorescent protein (gfp) at the C-terminus (yellow) stained with anti-fibrillarin (magenta), a nucleolar marker, and DAPI (gray). Dotted lines outline the testis (A) and nucleus (B). (C) RNA FISH for Mst77F and Mst77Y transcripts in wild-type spermatocyte nucleus. DAPI (gray), Modulo–gfp (yellow), Mst77F (magenta), and Mst77Y (cyan). Split channel view of Modulo-gfp (C'), Mst77F transcript (C''), and Mst77Y transcript (C''') in spermatocytes. Dotted lines outline the nucleus. (D) qRT-PCR following polyA selection (dark gray) or using total RNA qRT-PCR (light gray) in modulo-mutant males (modL8/mod07570) vs. sibling control males (mod07570/+) assessing levels of Mst77F (magenta) and Mst77Y (cyan). Data were normalized to Rp49 and control. Mean ± SD from three technical replicates is shown. P-values are listed (unpaired Student’s t test assuming unequal variance on untransformed ddct values). Similar results were obtained from two biological replicates. Primer locations are shown in SI Appendix, Fig. S11A. (E and F) RNA FISH for polyA (magenta) and Mst77F transcript (cyan) in control (y w) (F) vs. modulo-mutant males (modL8/mod07570) (E), counterstained with DAPI (gray). Split channel view of polyA-containing transcripts (E' and F') and Mst77F transcripts (E'' and F'') in control (E) and mutant males (F). Mst77F probe was used to identify nucleolus. Yellow arrowhead indicates polyA-containing RNA-encircling nucleolus. (G) Model for Modulo function in the nucleolus. fine-tuning the expression of different protamine genes may be critical. Additionally, if any protamine genes have evolved to opti- mally package certain satellite DNA, conversion of such protamine into a dominant-negative version can immediately target the chro- mosome that harbors the given satellite DNA, leading to meiotic drive that selectively harms the specific chromosome. This possibility 6 of 8   https://doi.org/10.1073/pnas.2220576120 pnas.org is intriguing as the Segregation Distorter (SD) meiotic drive system in D. melanogaster is known to target Responder satellite DNA repeats (44–46) and exhibits sperm nuclei decompaction similar to what is observed in this study (30). The possibility that dominant- negative protamines are involved in the decompaction of spermatid nuclei in the SD drive system remains to be studied. Taken together, our study identified a mechanism by which var- ious protamine variants are coordinately regulated at the posttran- scriptional level, possibly to achieve balanced expression of multiple protamine variants. A similar mechanism may be at play to fine-tune the expression levels of protamine variants in human and mouse, disruption of which is associated with compromised fertility. Methods Fly Husbandry and Strains. All fly stocks were raised on standard Bloomington medium at 25 °C, and young flies (1- to 3-d-old adults) were used for all exper- iments unless otherwise specified. Flies used for wild-type experiments were the standard laboratory wild-type strain y w (y1w1). The following fly stocks were used: modulo07570/TM3 [Bloomington Drosophila Stock Center (BDSC): 11795], moduloL8/TM3 (BDSC: 38432), and C(1)RM/C(1;Y)6, y1w1f1/0 (BDSC: 9460). The β2-tubulin promoter sequence used for producing Mst77Y overexpression was generously provided by Peiwei Chen and Alexei Aravin. The Mst77Y transgenic flies were generated by phiC31 site–directed integra- tion into the Drosophila genome. For UAS–Mst77Y-12, β2-tubulin–Mst77Y-3, and β2-tubulin–Mst77Y-12 transgenic lines, the Mst77Y overexpression sequences in D. melanogaster were synthesized by gene synthesis service from Thermo Fisher Scientific (GeneArt Gene Synthesis) and were cloned into pattB vector to insert into specific integration site on second chromosome (attP40) (SI Appendix, Fig. S5C and Table S2). All injection and selection of flies containing integrated transgene were performed by BestGene Inc. Because UAS–Mst77Y-12 transgene was injected to the same host fly strain as β2-tubulin–Mst77Y-3, and β2-tubulin–Mst77Y-12 trans- genic lines, we used this (without gal4 driver) as a “background-matched control.” Modulo–gfp strain was constructed using CRISPR-mediated knock-in of a Green fluorescent protein (gfp)-tag at the C terminus of Modulo (Beijing Fungene Biotechnology Co.) (SI Appendix, Table S3). Sex Ratio Assay. Individual 1-d-old males raised for at least one generation at 18 °C were crossed with 3× 1- to 3-d-old virgin y w females at 25 °C. After 1 d, males were removed. This was done to maximize the proportion of males exhib- iting decompaction phenotype described in Fig. 3. Females were left to produce embryos for a total of 5 d before cleared. Following the onset of eclosion, sex of offspring was scored for 10 consecutive days. RNA Fluorescent In Situ Hybridization. All solutions used were Rnase free. Testes from 1- to 3-d-old flies were dissected in 1X phosphate buffered saline (PBS) and fixed in 4% formaldehyde in 1X PBS for 30 min. Then, the testes were washed briefly in PBS and permeabilized in 70% ethanol overnight at 4 °C. For strains expressing gfp (i.e., Modulo–gfp), the overnight permeabilization in 70% ethanol was omitted. The testes were briefly rinsed with wash buffer (2X saline-so- dium citrate (SSC), 10% formamide) and then hybridized overnight at 37 °C with fluorescently labeled probes in hybridization buffer [2X SSC, 10% dextran sulfate (sigma, D8906), 1 mg/mL E. coli transfer RNA (sigma, R8759), 2 mM vanadyl ribonucleoside complex (NEB S142), 0.5% Bovine serum albumin (BSA) (Ambion, AM2618), 10% formamide]. Following hybridization, samples were washed two times in wash buffer for 30 min each at 37 °C and mounted in VECTASHIELD with DAPI (Vector Labs). Fluorescently labeled probes were added to the hybridization buffer to a final concentration of 100 nM. Poly(T) probes for recognizing Poly(A) sequence were from Integrated DNA Technologies. Probes against Mst77F and Mst77Y were designed using the Stellaris1 RNA FISH Probe Designer (Biosearch Technologies, Inc.) available online at www.biosearchtech.com/stellarisdesigner. Each set of custom Stellaris1 RNA FISH probes was labeled with Quasar 670 or Quasar 570 (SI Appendix, Table S4). Images were acquired using an upright Leica TCS SP8 confocal microscope with a 63× oil immersion objective lens (NA = 1.4) and processed using Adobe Photoshop and ImageJ software. DNA Fluorescence In Situ Hybridization. Testes from 1- to 3-d-old flies were rapidly dissected in 4% formaldehyde and 1mM Ethylenediaminetetraacetic acid (EDTA) in 1X PBS and then nutated for 30 min. Then, the testes were washed three times in 1X PBS containing 0.1% Triton-X (PBST) +1 mM EDTA for 30 min each. The testes were briefly rinsed with 1X PBST and then incubated at 37 °C for 10 min with 2 mg/mL Rnase A in PBST. Following Rnase treatment, samples were washed once in 1X PBST + 1 mM EDTA for 10 min. The samples were then briefly rinsed with 2X SSC + 1 mM EDTA + 0.1% Tween-20, and then washed three times in 2X SSC + 0.1% Tween-20 + formamide (20% for first wash, 40% for second, 50% for third) for 15 min each. The samples were then washed with 2X SSC + 0.1% Tween-20 + 50% formamide for 30 min. The samples were then incubated for 5 min at 95 °C with fluorescently labeled probes in hybridization buffer (2X SSC, 10% dextran sulfate, 50% formamide, 1 mM EDTA) and then transferred to 37 °C overnight. Following hybridization, the samples were washed three times in 2X SSC + 1 mM EDTA + 0.1% Tween-20 for 20 min each and then mounted in VECTASHIELD with DAPI (Vector Labs). Fluorescently labeled probes were added to the hybridization buffer to a final concentration of 500 nM. Satellite DNA probes distinguishing X and Y chromosomes (AATAC)6-Cy5 for Y and (TAGA)8-Cy3 were from Integrated DNA Technologies. IF Staining. Testes were dissected in 1X PBS, transferred to 4% formaldehyde in 1X PBS, and fixed for 30 min. The testes were then washed three times in 1X PBST for 20 min each followed by incubation with primary antibodies diluted in 1X PBST with 3% BSA at 4 °C overnight. Samples were washed three times in 1X PBST for 20 min each and incubated with secondary antibody in 1X PBST with 3% BSA for 2 h at room temperature. The samples were then washed three times in 1X PBST for 20 min each and mounted in VECTASHIELD with DAPI (Vector Labs). The following primary antibodies were used: anti-fibrillarin (1:200, mouse; Abcam; ab5812), anti-Modulo (1:1,000, guinea pig; this study), anti-protamine A/B [1:200, guinea pig, gift of Elaine Dunleavy, Centre for Chromosome Biology, National University of Ireland, Galway, Ireland (47), anti-dsDNA (1:500; mouse,; Abcam; ab27156), anti-histone H3 (1:200, rabbit; Abcam; ab1791), anti-Mst77F (1:1,000; guinea pig, this study), anti-Mst77Y (1:500; rabbit, this study), anti-Tpl94D (1:500; rabbit, this study), and phalloidin-Alexa Fluor 546 or 488 (1:200; Thermo Fisher Scientific; A22283 or A12379). The Modulo antibody was generated by injecting a peptide sequence CRKQPVKEVPQFSEED[48-62] targeting the N-terminal end of Modulo in guinea pigs (Covance). The Tpl94D antibody was generated by injecting a peptide DKGSAYKPLTLNRSYVIRKC[96-114] in rabbits (Covance). The Mst77F antibody was generated by injecting multiple peptides (SKPEVAVTC[9-16], YKKSIEYVNC[22-30], CRSSEGEHR[112-119], LQRSSEGEHRMHSEC[110-123], RSSGKPKPKGARPRKC[169-183]) targeting sites in Mst77F, as indicated, differen- tiating it from Mst77Y in guinea pigs (Covance). The Mst77Y antibody was gen- erated by injecting multiple peptides (IKPDVAVSC[9-16], SRKAIEYVKC[22-30], CRSIEAELR[112-119], KTSRKAIEYVKSD[20-32], CVSSLQRSIEAELR[107-119]) target- ing sites of varying aa length in Mst77Y, differentiating it from Mst77F in rabbits (Covance). Alexa Fluor–conjugated secondary antibodies (Life Technologies) were used at a dilution of 1:200. qRT-PCR. Total RNA was purified from D. melanogaster adult testes (75 pairs/sam- ple) by Direct-zol RNA Miniprep (Zymo Research), with DNase treatment according to manufacturer’s protocol. One microgram total RNA was reverse transcribed following priming with either random hexamers or polyT using SuperScript III® Reverse Transcriptase (Invitrogen) followed by qPCR using Power SYBR Green rea- gent (Applied Biosystems). Primers for qPCR were designed to amplify mRNA or intron-containing transcript as indicated. Relative expression levels were normal- ized to Rp49 and control siblings. All reactions were done in technical triplicates with at least two biological replicates. Graphical representation was inclusive of all replicates and P-values were calculated using a t test performed on untransformed average ddct values. Primers used are described in SI Appendix, Fig. S11 A and B. Total RNA-Seq. Total RNA was purified from D. melanogaster adult testes by Direct-zol RNA Miniprep (Zymo Research), with Dnase treatment. Quality of the indexed libraries was confirmed using an Agilent Fragment Analyzer and qPCR. Sequencing libraries were prepared with the KAPA RNA HyperPrep Kit with RiboErase. Samples were sequenced on a HiSeq 2500, producing 100 × 100 nt paired-end reads. The read pairs were mapped to the canonical chromosomes of the D. melanogaster genome (assembly BDGP6/dm6) using STAR 2.7.1a (48); PNAS  2023  Vol. 120  No. 16  e2220576120 https://doi.org/10.1073/pnas.2220576120   7 of 8 default parameters, except “—alignIntronMax 25000,” indexed with all FlyBase genes (FB2020_06 Dmel Release 6.37) and the option “—sjdbOverhang 100.” Gene counts were obtained using featureCounts (49); v 2.0.1, with “-M –fraction -p -s 2.” After summing gene counts for technical replicates, differential expres- sion was assayed using DESeq2 v1.26.0 (50), with lfcShrink(type=”ashr”)). RNA coverage across genes at nucleotide resolution was quantified with “bedtools coverage” (51) and scaled by the total number of reads mapped to genes. All the statistical details of the experiments are provided in the main text and leg- ends. P-values are listed either in figure, figure legends, or SI Appendix, Table S1. Data, Materials, and Software Availability. Sequencing data is available at National Center for Biotechnology Information Gene Expression Omnibus under accession GSE214456 (52). All other data are included in the manuscript and/ or SI Appendix. Statistics and Reproducibility. Data are presented as mean ± SD unless oth- erwise indicated. The sample number (n) indicates the number of testes, nuclei, or male flies in each experiment as specified in the figure legends. We utilized two-sided Student’s t test to compare paired or independent samples, as applica- ble and is specified in the figure legends. We calculated probability using exact binomial distribution with parameters specified in Fig. 3G legend. No statistical methods were used to predetermine sample sizes. The experimenters were not blinded to the experimental conditions, and no randomization was performed. ACKNOWLEDGMENTS. We thank the Bloomington Drosophila Stock Center and Dr. Elaine Dunleavy for reagents. We thank the Data Science, Bioinformatics, and Informatics Core at the University of Michigan for consulting and Dr. Bing Ye for advice and support. We thank the Yamashita, Lehmann, and Ye lab members, Drs. Daven Presgraves and Eric Lai for discussions, and Yamashita Lab members for comments on the manuscript. The research was supported by the Eunice Kennedy Shriver Institute of Child Health and Development of the NIH (to J.I.P., F30HD105324), HHMI (to Y.M.Y.), and Whitehead Institute for Biological Research. 1. 2. 3. 4. 5. 6. 7. 8. 9. C. Rathke et al., Transition from a nucleosome-based to a protamine-based chromatin configuration during spermiogenesis in Drosophila. J. Cell Sci. 120, 1689–1700 (2007). D. T. Carrell, B. R. Emery, S. Hammoud, Altered protamine expression and diminished spermatogenesis: What is the link? Hum. Reprod. Update 13, 313–327 (2007). C. Rathke, W. M. Baarends, S. Awe, R. Renkawitz-Pohl, Chromatin dynamics during spermiogenesis. Biochim. Biophys. Acta (BBA) - Gene Regul. Mechanisms 1839, 155–168 (2014). R. Balhorn, The protamine family of sperm nuclear proteins. Genome Biol. 8, 227 (2007). R. Oliva, Protamines and male infertility. Hum. Reprod. Update 12, 417–435 (2006). C. Cho et al., Protamine 2 deficiency leads to sperm DNA damage and embryo death in Mice1. Biol. Reprod. 69, 211–217 (2003). C. Cho et al., Haploinsufficiency of protamine-1 or -2 causes infertility in mice. Nat. Genet. 28, 82–86 (2001). D. Miller, M. Brinkworth, D. Iles, Paternal DNA packaging in spermatozoa: More than the sum of its parts? DNA, histones, protamines and epigenetics. Reproduction 139, 287–301 (2010). L. Lüke, A. Vicens, M. Tourmente, E. R. S. Roldan, Evolution of protamine genes and changes in sperm head phenotype in Rodents1. Biol. Reprod. 90, 67 (2014). 28. K. Abdelmohsen, M. Gorospe, RNA-binding protein nucleolin in disease. RNA Biol. 9, 799–808 (2012). 29. D. H. Castrillon et al., Toward a molecular genetic analysis of spermatogenesis in Drosophila melanogaster: Characterization of male-sterile mutants generated by single P element mutagenesis. Genetics 135, 489–505 (1993). 30. M. Herbette et al., Distinct spermiogenic phenotypes underlie sperm elimination in the Segregation Distorter meiotic drive system. PLoS Genet. 17, e1009662 (2021). 31. J. D. Lewis et al., Histone H1 and the origin of protamines. Proc. Natl. Acad. Sci. U.S.A. 101, 4148–4152 (2004). 32. B. Barckmann et al., Three levels of regulation lead to protamine and Mst77F expression in Drosophila. Dev. Biol. 377, 33–45 (2013). 33. F. Michiels, D. Buttgereit, R. Renkawitz-Pohl, An 18-bp element in the 5’ untranslated region of the Drosophila beta 2 tubulin mRNA regulates the mRNA level during postmeiotic stages of spermatogenesis. Eur. J. Cell Biol. 62, 66–74 (1993). 34. F. Michiels, A. Gasch, B. Kaltschmidt, R. Renkawitz-Pohl, A 14 bp promoter element directs the testis specificity of the Drosophila beta 2 tubulin gene. EMBO J. 8, 1559–1565 (1989). 10. C.-H. Chang, I. Mejia Natividad, H. S. Malik, Expansion and loss of sperm nuclear basic protein genes in Drosophila correspond with genetic conflicts between sex chromosomes. Elife 12, e85249 (2023). 11. W.-M. Maier, G. Nussbaum, L. Domenjoud, U. Klemm, W. Engel, The lack of protamine 2 (P2) in boar and bull spermatozoa is due to mutations within the P2 gene. Nucleic Acids Res. 18, 1249–1254 (1990). 35. R. E. Braun, J. J. Peschon, R. R. Behringer, R. L. Brinster, R. D. Palmiter, Protamine 3’-untranslated sequences regulate temporal translational control and subcellular localization of growth hormone in spermatids of transgenic mice. Genes Dev. 3, 793–802 (1989). 36. D. Elliott, Pathways of post-transcriptional gene regulation in mammalian germ cell development. 12. S. Tirmarche et al., Drosophila protamine-like Mst35ba and Mst35bb are required for proper Cytogenet. Genome Res. 103, 210–216 (2003). sperm nuclear morphology but are dispensable for male fertility. G3: Genes, Genomes, Genetics 4, 2241–2245 (2014). 37. G. Dreyf, Structure and function of nuclear and cytoplasmic ribonucleoprotein particles. Annu. Rev. Cell Biol. 2, 459–498 (1986). 13. S. Jayaramaiah Raja, R. Renkawitz-Pohl, Replacement by Drosophila melanogaster Protamines and 38. S. Ozturk, F. Uysal, Potential roles of the poly(A)-binding proteins in translational regulation during Mst77F of histones during chromatin condensation in late spermatids and role of sesame in the removal of these proteins from the male pronucleus. Mol. Cell Biol. 25, 6165–6177 (2005). spermatogenesis. J. Reprod. Dev. 64, 289–296 (2018). 39. H. K. Kini, M. R. Vishnu, S. A. Liebhaber, Too much PABP, too little translation. J. Clin. Invest. 120, 14. C. M. Doyen et al., A testis-specific chaperone and the chromatin remodeler ISWI mediate 3090–3093 (2010). repackaging of the paternal genome. Cell Rep. 13, 1310–1318 (2015). 40. P. Chen et al., piRNA-mediated gene regulation and adaptation to sex-specific transposon 15. C. A. Muirhead, D. C. Presgraves, Satellite DNA-mediated diversification of a sex-ratio meiotic drive expression in D. melanogaster male germline. Genes. Dev. 35, 914–935 (2021). gene family in Drosophila. Nat. Ecol. Evol. 5, 1604–1612 (2021). 41. N. Kost et al., Multimerization of Drosophila sperm protein Mst77F causes a unique condensed 16. J. Vedanayagam, C. J. Lin, E. C. Lai, Rapid evolutionary dynamics of an expanding family of meiotic chromatin structure. Nucleic Acids Res. 43, 3033–3045 (2015). drive factors and their hpRNA suppressors. Nat. Ecol. Evol. 5, 1613–1623 (2021). 42. Y. Tao, J. P. Masly, L. Araripe, Y. Ke, D. L. Hartl, A sex-ratio meiotic drive System in Drosophila simulans 17. C. Rathke et al., Distinct functions of Mst77F and protamines in nuclear shaping and chromatin condensation during Drosophila spermiogenesis. Eur. J. Cell Biol. 89, 326–338 (2010). 18. S. Kimura, B. Loppin, The Drosophila chromosomal protein Mst77F is processed to generate an I: An Autosomal Suppressor. PLoS Biol. 5, e292 (2007). 43. M. Jagannathan, Y. M. Yamashita, Function of Junk: Pericentromeric satellite DNA in chromosome maintenance. Cold Spring Harb. Symp. Quant. Biol. 82, 319–327 (2017). essential component of mature sperm chromatin. Open Biol. 6, 160207 (2016). 44. A. M. Larracuente, The organization and evolution of the Responder satellite in species of the 19. Z. Eren-Ghiani, C. Rathke, I. Theofel, R. Renkawitz-Pohl, Prtl99C acts together with protamines and safeguards male fertility in drosophila. Cell Rep. 13, 2327–2335 (2015). 20. F. J. Krsticevic, C. G. Schrago, A. B. Carvalho, Long-read single molecule sequencing to resolve tandem gene copies: The Mst77Y region on the drosophila melanogaster Y chromosome. G3: Genes, Genomes, Genetics 5, 1145–1150 (2015). Drosophila melanogaster group: dynamic evolution of a target of meiotic drive. BMC Evol. Biol. 14, 233 (2014). 45. K. Houtchens, T. W. Lyttle, Responder (Rsp) alleles in the segregation distorter (SD) system of meiotic drive in Drosophila may represent a complex family of satellite repeat sequences. Genetica 117, 291–302 (2003). 21. F. J. Krsticevic, H. L. Santos, S. Januário, C. G. Schrago, A. B. Carvalho, Functional copies of the 46. Y. Hiraizumi, L. Sandler, J. F. Crow, Meiotic drive in natural populations of Drosophila Mst77F gene on the Y chromosome of Drosophila melanogaster. Genetics 184, 295–307 (2010). 22. M. Corzett, J. Mazrimas, R. Balhorn, Protamine 1: Protamine 2 stoichiometry in the sperm of eutherian mammals. Mol. Reprod. Dev. 61, 519–527 (2002). 23. S. Hammoud, L. Liu, D. T. Carrell, Protamine ratio and the level of histone retention in sperm selected from a density gradient preparation. Andrologia 41, 88–94 (2009). 24. K. Steger et al., Both protamine-1 to protamine-2 mRNA ratio and Bcl2 mRNA content in testicular spermatids and ejaculated spermatozoa discriminate between fertile and infertile men. Hum. Reprod. 23, 11–16 (2007). melanogaster. III. Populational implications of the Segregation-Distorter Locus. Evolution (N Y) 14, 433 (1960). 47. W. K. Mills, Y. C. G. Lee, A. M. Kochendoerfer, E. M. Dunleavy, G. H. Karpen, RNA from a simple- tandem repeat is required for sperm maturation and male fertility in Drosophila melanogaster. Elife 8, e48940 (2019). 48. A. Dobin et al., STAR: Ultrafast universal RNA-seq aligner. Bioinformatics 29, 15–21 (2013). 49. Y. Liao, G. K. Smyth, W. Shi, featureCounts: An efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics 30, 923–930 (2014). 25. S. Amjad et al., Protamine 1/Protamine 2 mRNA ratio in nonobstructive azoospermic patients. 50. M. I. Love, W. Huber, S. Anders, Moderated estimation of fold change and dispersion for RNA-seq Andrologia 53, e13936 (2021). data with DESeq2. Genome Biol. 15, 550 (2014). 26. V. W. Aoki, L. Liu, D. T. Carrell, Identification and evaluation of a novel sperm protamine abnormality 51. A. R. Quinlan, I. M. Hall, BEDTools: A flexible suite of utilities for comparing genomic features. in a population of infertile males. Hum. Reprod. 20, 1298–1306 (2005). Bioinformatics 26, 841–842 (2010). 27. L. M. Mikhaylova, A. M. Boutanaev, D. I. Nurminsky, Transcriptional regulation by Modulo integrates 52. J. I. Park, G. W. Bell, Y. M. Yamashita, Total RNA-seq in the modulo mutant reveals broad changes meiosis and spermatid differentiation in male germ line. Proc. Natl. Acad. Sci. U.S.A. 103, 11975–11980 (2006). to the transcriptome. Gene Expression Omnibus. https://www.ncbi.nlm.nih.gov/geo/query/acc. cgi?acc=GSE214456. Deposited 29 September 2022. 8 of 8   https://doi.org/10.1073/pnas.2220576120 pnas.org
null
10.1088_1748-3190_ad00a2.pdf
Data availability statement All data that support the findings of this study are included within the article (and any supplementary files).
Data availability statement All data that support the findings of this study are included within the article (and any supplementary files).
RECEIVED 15 May 2023 REVISED 21 September 2023 ACCEPTED FOR PUBLICATION 5 October 2023 PUBLISHED 30 October 2023 Bioinspir. Biomim. 18 (2023) 066016 https://doi.org/10.1088/1748-3190/ad00a2 PAPER Exploring storm petrel pattering and sea-anchoring using deep reinforcement learning Jiaqi Xue1,2,3,6, Fei Han1,2,6, Brett Klaassen van Oorschot4, Glenna Clifton5 and Dixia Fan1,2,∗ 1 Key Laboratory of Coastal Environment and Resources of Zhejiang Province, School of Engineering, Westlake University, Hangzhou, Zhejiang 310030, People’s Republic of China 2 School of Engineering, Westlake University, Hangzhou, Zhejiang 310024, People’s Republic of China 3 Walker Department of Mechanical Engineering, University of Texas at Austin, Austin, TX, United States of America 4 Biomimetics Lab, Experimental Zoology Group, Wageningen University, Wageningen, The Netherlands 5 Department of Biology, University of Portland, Portland, OR, United States of America 6 Equally contributed first authors. ∗ Author to whom any correspondence should be addressed. E-mail: fandixia@westlake.edu.cn Keywords: storm petrel, locomotory behaviors, deep reinforcement learning Supplementary material for this article is available online Abstract Developing hybrid aerial-aquatic vehicles that can interact with water surfaces while remaining aloft is valuable for various tasks, including ecological monitoring, water quality sampling, and search and rescue operations. Storm petrels are a group of pelagic seabirds that exhibit a unique locomotion pattern known as ‘pattering’ or ‘sea-anchoring,’ which is hypothesized to support forward locomotion and/or stationary posture at the water surface. In this study, we use morphological measurements of three storm petrel species and aero/hydrodynamic models to develop a computational storm petrel model and interact it with a hybrid fluid environment. Using deep reinforcement learning algorithms, we find that the storm petrel model exhibits high maneuverability and stability under a wide range of constant wind velocities after training. We also verify in the simulation that the storm petrel can use its ‘pattering’ or ‘sea-anchoring’ behavior to achieve different biomechanical sub-tasks (e.g. weight support, forward locomotion, stabilization) and adapt it under different wind speeds and optimization objectives. Specifically, we observe an adjustment in storm petrel’s movement patterns as wind velocity increases and quantitively analyze its biomechanics underneath. Our results provide new insights into how storm petrels achieve efficient locomotion and dynamic stability at the air–water interface and adapt their behaviors to different wind velocities and tasks in open environments. Ultimately, our study will guide the design of next-generation biomimetic petrel-inspired robots for tasks requiring proximity to the water interface and efficiency. 1. Introduction Increasing attention from both academia and industry has been put on water quality sampling [1], ecological monitoring [2], and water search and res- cue operations [3], especially in open and unexplored environments. Aerial-aquatic vehicles that can inter- act with the water’s surface while remaining aloft can be useful in these tasks [1, 3, 4]. However, thus far, no promising aerial-aquatic vehicle has been developed due to the challenge of maintaining equilibrium on yielding fluid surfaces and the unpredictable disturb- ances of the open ocean. Animals, on the other hand, have evolved a few ways to achieve aerial-aquatic locomotion that over- come these problems [5–9]. Some very small anim- als, such as water striders (Gerridae spp.), benefit from a low weight-to-volume ratio and can use surface tension to remain above water [10]. However, lar- ger animals must either wholly jump between the two media (such as flying fish [11]) or paddle an appendage in the water to prevent the body from © 2023 IOP Publishing Ltd Bioinspir. Biomim. 18 (2023) 066016 J Xue et al submerging (such as basilisk lizards, grebes, and dolphins [6]). A lesser-known example of air–water interface locomotion occurs when some species of storm petrels repetitively slap their feet against the surface of the water to either jump around or stay stationary. These behaviors are called ‘pattering’ or ‘sea-anchoring’, respectively [12]. The functions of these behaviors have been disputed and hypothesized to either ‘anchor’ the bird in place against incom- ing winds [13, 14], exploit ground effect and hydro- dynamic forces to support its weight, or provide for- ward thrust. Not all species of storm petrel have been observed pattering/sea-anchoring, with larger species and those with relatively longer tarsi exhibiting more pattering/sea-anchoring behavior [12]. Pattering and sea-anchoring are challenging to study in situ due to the unpredictable nature of these behaviors and their occurrence within harsh pelagic environments. Therefore, most descriptions result from personal accounts or recordings that cannot quantify three-dimensional movements [14]. Pattering/sea-anchoring also cannot be studied in a lab setting as storm petrels normally travel over large oceanic areas and require natural environ- ments to induce these foraging behaviors. The first theoretical attempt at understanding pattering/sea- anchoring behavior was from a mathematical per- spective and used static models. It suggested that the drag forces produced by the feet may be feas- ible for an ‘anchoring’ function of the behavior [13]. The lack of further research on these behaviors could be attributed to the high dimensionality and non- linearity of fluid and storm petrel dynamics, which is difficult to capture with steady or static mod- els. To account for the complexity and variability in this system, we build a reduced-order computa- tional model based on the anatomy and biomechan- ics of storm petrels and the fundamental principles of hydrodynamics and aerodynamics under quasi- steady assumptions. Built upon that, we use artifi- cial intelligence (AI), specifically deep reinforcement learning (DRL) algorithms, to help the storm pet- rel learn the pattering/sea-anchoring behavior under generic reward and environment conditions. DRL algorithms, such as the deep deterministic policy gradient (DDPG) algorithm [15], are a type of model-free reinforcement learning method [16]. These algorithms combine deep neural networks with reinforcement learning techniques to enable agents to learn from their interactions with an environment and make decisions based on their observations [17, 18]. DRL algorithms have emerged as powerful tools for tackling complex systems and problems [19, 20] and have demonstrated superior performance com- pared to traditional approaches, mainly when dealing with high-dimensional or uncertain environments [15]. In particular, the DDPG algorithm has been successful in tasks involving dynamic tracking for soft robots [21], or faster and efficient locomotion in 2 quadrupedal systems [22]. These examples demon- strate the potential of model-free DRL, specifically the DDPG algorithm, to tackle complex problems in vari- ous applications and highlight its ability to outper- form traditional approaches in specific scenarios. Accordingly, the goal of this paper is to develop an integrated simulation framework, including a com- putational storm petrel model, a quasi-steady fluid environment, and a model-free DRL agent, that enables the storm petrel model to learn how to inter- act with the dynamic environment spontaneously and produce biologically similar locomotory behavi- ors to storm petrels in nature (i.e. pattering and/or sea-anchoring). This paper will also investigate the biomechanical implications of these behaviors and explore how storm petrels adapt their behaviors and performance under varying objectives and environ- mental conditions, such as wind speeds. 2. Materials and methods To systematically model the interaction between the biological system (i.e. the storm petrel) and the air/water fluid environment, we require some basic information: (1) the storm petrel’s morphological and kinematic constraints (see tables 1 and 2, respect- ively), and (2) the aerodynamic and hydrodynamic forces acting on the petrel. We then use a model- free reinforcement learning algorithm to recreate the locomotory behavior of storm petrels. 2.1. Simplified model of storm petrels 2.1.1. Anatomy The bird’s anatomy (see figure 1) is approximated using parametric geometries based on first-hand ana- tomical measurements of three storm petrel speci- mens (Oceanites oceanicus, Oceanodroma furcata, and Oceanodroma leucorhoa) available at the University of Washington Burke Museum of Natural History (Seattle, WA, USA). These measurements can be found in appendix A. The trunk and head of the storm petrel are approximated as a single ellipsoid with a uniform prolate spheroidal cross-section, which is referred as ‘body’ in the following paper. The semi-major axis length, a, relates to the body length of the storm pet- rel, while the semi-minor axis length, b, corresponds to the body width, shown in figures 1(A) and (B) as a dashed ellipse. The projected body areas along the x and y directions, Sx and Sy, are calculated as Sx = π b2 and Sy = π ab. The tail and the wings are modeled as thin plates with trapezoidal and rect- angular shapes, respectively. The legs of the storm petrel are comprised of the femur, tibiotarsus, tar- sometatarsus, and foot, which are demonstrated in figure 1(B). The femur is neglected in this simpli- fied model since the hip demonstrates only minor flexion/extension movements in running [23] and hopping [24] birds. The leg segments distal to the Bioinspir. Biomim. 18 (2023) 066016 J Xue et al Figure 1. Model extraction of storm petrels. Presuming a symmetrical body structure of the storm petrel, we define the model’s anatomical and kinematic parameters based on a single side. (A) Bottom view: simplification to parametric geometries from the frozen specimen of storm petrels, and the definition of the flapping and pitching axis of the wings and the pitching axis of the tail; (B) Side view: the anatomical definitions on storm petrel’s skeleton and degree of freedom (DoF) definitions of all controllable joints except for the wing flapping joint, which is isolated and demonstrated in (C). The black circle-dot symbol denotes the location of each joint and its rotation axis. The light-grey arrow line indicates the reference axes (zero rotation) for each DoF. (C) Front view of the extracted model: Illustration of the flapping DoF of the wings. Table 1. Dimensions of the simplified storm petrel model. Table 2. Kinematic constraints at each joint. Body (ellipsoid) Semi-major axis (a) Semi-minor axis (b) Aspect ratio Wings (rectangle) Span (s) Chord (c) Maximum camber Aspect ratio Legs (Columns) Tibiotarsus (L1) Tarsometatarsus (L2) Leg radius Webs (Sectors) Length (L3) Web angle (ε) Tail (Trapezoid) Root width (w1) Tip width (w2) Length (t) Value 72 mm 25 mm 2.6 Value 190 mm 52.6 mm 5 mm 3.04 Value 40 mm 35 mm 2 mm Value 25 mm 50◦ Value 30 mm 70 mm 65 mm femur are modeled as solid columns. The metatars- als (i.e. digits) were combined with the foot’s web and were thus modeled as a single thin plate with a sector shape. The sector angle is estimated to be 50◦ 3 Joint Wing Pitching Wing Flapping Tail Pitching Knee Ankle Metatarsal Range of motion (degree) Max.velocity (ms−1) [−10, 50] [−60, 30] [−30, 30] [−75, −30] [0, 135] [−45, 45] 15 20 15 25 70 20 based on the maximum intersection angle of meta- tarsal bones measurement. The dimensions of these simplified parametric geometries are summarized in table 1 and illustrated in figure 1(A). 2.1.2. Kinematics Flapping and hopping in birds are complex kin- ematic behaviors comprising many degrees of free- dom (DoF) at the base, within the wings and tail, and at each hindlimb (leg) joint. Here, we gener- ate a nine DoF kinematic model based on frame-by- frame analysis of open-source videos [25] and ana- tomical analysis of specimens. Specifically, we model wing pitching and flapping, tail pitching, and flexion/ extension at the knee, intertarsal ‘ankle’, and metatar- sophalangeal (or metatarsal) joints, shown in figure 1. To estimate the leg joints’ maximum angular excur- sions and velocities, we analyze the approximately two-dimensional joint motion for a storm petrel over Bioinspir. Biomim. 18 (2023) 066016 J Xue et al Figure 2. Estimation of joint kinematics from video of a pattering storm petrel. (A-1)–(A-3) show the initial pose (landing) of the storm petrel before it immerses its legs and strikes backward, while (B-1)–(B-3) exhibit the finishing pose (takeoff) after each leg 2, and αf 1, αf strike. αs stand for these joint angles at the takeoff phase. Still frame images courtesy of the Macaulay Library at the Cornell Lab of Ornithology, ML201429031. Reproduced with permission from [25]. [© Josep del Hoyo]. 3 represent the knee, ankle, and metatarsal joint angle at the landing phase, respectively, while αf 2, and αs 1, αs 3 several strides and wing beats, shown in figure 2 cap- turing the initial (landing) and finishing (takeoff) phase of the pattering behavior. 2.2. Calculation of fluid forces The storm petrel model interacts with a complex fluid environment, which we quantify through aero- dynamic and hydrodynamic forces exerted on the bird’s body components. The models are based on quasi-steady assumptions using coefficients and cor- recting factors from experimental data from previ- ous literature [26–31]. These quasi-steady assump- tions are a potential source of error in our models, but since our focus here is on the integration of the bird, hybrid fluid environment, and the RL framework, a fully resolved unsteady fluid dynamic model is bey- ond the scope of this study. As demonstrated in figure 3, we define the free- stream (absolute) wind velocity, U∞, as constant and always in the horizontal direction (negative x-axis). With non-zero local wind velocities, aerodynamics will exert drag or/and lift on the storm petrel’s body, wings, and tail. Lift, L, is defined herein as the force perpendicular to the local wind velocity, while drag, D, is the force parallel to the local velocity. Although we will distinguish the pattering and sea-anchoring behavior in the next section (section 2.3) due to their different presumed goals, 4 we intend not to model the fluid dynamics separ- ately for these two behaviors in this initial study until more kinematics data are available, where these two behaviors are currently assumed the same kinematic constraints in the modeling. 2.2.1. Aerodynamic forces acting on the wings In a generalized form, lift, LW, and drag, DW, can be written in the following forms: LW = DW = 1 2 1 2 ρaSl ˆU2 wCLw ρaSd ˆU2 wCDw, (1) (2) where ρa, Sl, Sd, CL, and CD denote the air density, the projected area along the lift force direction, pro- jected area along the local velocity, lift coefficient, and drag coefficient. ˆUw refers to the local wind velo- city at the equivalent motion center of the wings, ˆUw = U∞ − Uw. The observed pattering or sea-anchoring loco- motion of storm petrels is near the water surface, where the ground effect plays a significant role in aug- menting lift [28]. Under this condition, we assume the direction of lift is always perpendicular to the local wind velocity, with no induced drag, as demonstrated in figure 3 [28]. We calculate the wing’s lift coefficient, CLw, through the thin-plate model [26] as: CLw = 2π sin (|α|) . (3) Bioinspir. Biomim. 18 (2023) 066016 J Xue et al Figure 3. Body-fluid interaction. (A) Environmental forces present in the multi-body dynamic modeling: R represents the hydrodynamic resistance on the feet; DB represents the aerodynamic drag on the body; LW and DW represent the aerodynamic lift and drag on the wings, respectively; LT and DT represent the aerodynamic lift and drag on the tail, respectively. ˆUb, ˆUt, ˆUw and ˆUf represent the local wind velocity at the storm petrel’s body, tail, wings, and feet in the global coordinates. ˆUf is the equivalent velocity of water relative to the feet, which is the velocity component of feet velocity, Uf, in the direction normal to the feet surface. (B) Demonstration of the body drag calculation, which is based on transforming the local wind velocity and reaction forces between the fixed global coordinates (X), (Y) and local coordinates (u), (v) attached to the body. The blue dash lines are velocity components of the local velocity at the body center discomposed in the local body coordinate (ˆUl b); on the other hand, the orange and red dash lines are force components represented in the local body coordinate (Dl B) and global coordinate (DB), respectively. α is the angle of attack of the airfoil of the body, while θ is the body pitch angle relative to the positive x-direction. (C) Demonstration of the aerodynamic forces calculation of the wings. Local wind velocity at the wings can be decomposed to free stream velocity, U∞, and the wing’s absolute velocity, Uw, which is equal to the body velocity, Ub, plus the relative wing’s velocity to the body, Ub w. The positive direction of global coordinates (X), (Y) are represented by two black dashed arrows in (A) and (C). Figure 4. Drag coefficients for various bird wings across attack angles. Colored solid lines represent experimentally measured drag. The black dashed line indicates the averaged data over all 13 bird species. This data is subsequently fitted by a bi-harmonic function regression to be used in the actual calculation: CDw = 16.4sin(0.03|α| + 1.4) + 16.2sin(0.03|α| + 4.5). Reproduced with permission from [32]. [© 2016. Published by The Company of Biologists Ltd]. While the drag coefficient, CDw, of a storm pet- rel’s wing has not been experimentally measured, a comparative analysis of 13 species shows reasonable consistency across birds [32], shown in figure 4. It is worth noticing that these experimental data are collected from −5◦ to 50◦ angle of attack, α. To enable a larger motion space for the simulated wings, we mirror the data to the negative domain of α, which leads to a symmetric data set with an α range of [−50◦, 50◦]. 2.2.2. Aerodynamic forces acting on the tail We use the slender lifting surface theory to model the aerodynamic force on the tail [33], which formulates the lift force generated on the tail as: LT = Kg π 4 ρα ˆU2 t bmax 2 (4) where Kg = 1.3 is the ground effect correction factor [27, 28], ρ is the air density, α is the angle of attack, 5 Bioinspir. Biomim. 18 (2023) 066016 J Xue et al and ˆUt is the local wind velocity at the tail’s geomet- ric center, ˆUt = U∞ − Ut. bmax is defined as the tail’s maximum width and equals the tail’s tip width, w2, in our model. The total drag on the tail consists of the induced drag and the profile drag. The induced drag, Di, the pressure drag, Dp, and profile drag skin friction drag, Dr, are formulated as follows: Di = Dp = Dr = 1 2 1 2 1 2 LTα ρ ˆU2 t SpCdp ρ ˆU2 t SfCdf, (5) where Sp and Sf are the frontal projected area of the tail and the wetted surface area, respectively. Cdf = 1.328√ is the skin friction coefficient for Re < 106 [29], Re where Re = ρUc µ is the Reynolds number and c is the chord length. Cdp is the pressure drag coefficient, which can be reasonably ignored over a realistic range of steady angles of attack for bird tails of −10◦ to 25◦ [34]. Therefore, the drag of the tail, DT, can be formu- lated as: DT = 1 2 LTα + 1 2 ρ ˆU2 t SfCdf. (6) 2.2.3. Aerodynamic forces acting on the body The drag on the body ellipsoid is first calculated in the body coordinates and then transformed to the global coordinates, demonstrated in figure 3(B). The local velocity at the body’s centroid in the spatial (global) coordinates is denoted as Ub (or Ug b) and is calculated by ˆUb = U∞ − Ub, while ˆUl b is that same velocity in the storm petrel’s body (local) coordinates. ˆUl b = Rl g ˆUb (7) where Rl g is the rotation matrix mapping from local to global coordinates. Then we calculated the aerody- namic drag on the body in the local coordinate, Dl B, as follows: Dl B = 1 2 ρ ˆUl2 b SpCdp (8) where the pressure drag coefficient Cdp of the ellips- oidal body is selected to be 0.167 and 0.7 along the major and minor axis, respectively [30]. After calculating the aerodynamic forces along the major and minor axis of the elliptical cross-section in the local coordinate, Dl B, we transform these forces back to the global coordinate, denoted as DB. DB = Rg l Dl B, (9) where Rg from the global to the local frame. l = (Rl g)−1 is the rotation matrix mapping 6 2.2.4. Hydrodynamic forces acting on the feet In general, small water walkers, such as insects, rely primarily on surface tension to support weight and stay statically on the surface, while large creatures dynamically generate hydrodynamic forces to support body weight (y-direction) or/and provide propul- sion or thrust (x-direction) with form drag, added mass, and buoyancy playing significant roles [6]. Nevertheless, some large creatures, such as storm pet- rels and basilisk lizards, do not submerge their bod- ies during locomotion, so we can ignore the buoy- ancy force as well considering the minimized volume of feet. Although basilisk lizards are found to reduce downward drag by retracting their feet from the water through a generated air cavity, grebes retract their feet laterally and do not use an air cavity, suggesting that the utilization of air cavity varies across water-walking organisms [9]. Although it is possible, there is no con- crete evidence of air cavities forming in pattering or sea-anchoring in storm petrels. Therefore, we do not model air cavities in this initial study. Additionally, although we did consider added mass force in our early- hydrodynamics modeling, we find in our pre- liminary testing that the magnitude of the added mass force is not comparable to the form drag under the current acceleration of the legs during the strike phase. Therefore, the final modeling of the hydro- dynamic resistance force acting on the feet R includes only a form drag term, which is formulated as: R = 1 2 ρwAUf 2CR, (10) where ρw, A, and CR denote the seawater density, foot web area, and the hydrodynamic drag coefficient. Here, we define Uf as the velocity of the feet at the web’s geometric center along its surface’s normal dir- ection. CR is estimated based on the experimental res- ult of a Basilisk Lizard study [31]. 2.3. DRL Beginning from this section, we will distinguish the pattering and sea-anchoring as two different loco- motion patterns to validate different assumptions of observed behavior’s biomechanical purposes (stabil- ity, weight support, thrust), as well as to explore the behavior or/and performance changes of storm pet- rels under different goals and environmental condi- tions (i.e. wind speed). ‘Pattering’ is defined in the following sections as a persistent forward locomotive behavior in which storm petrels strike their feet on the water while flap- ping their wings to ‘jump’ on the water while foraging for prey. This behavior is mostly observed on videos with relatively still water surfaces, suggesting a low free-stream wind velocity. In addition to forward locomotion, storm petrels show instances when an individual will drag their feet in the water while keeping their wings outstretched to stay stationary and/or against the incoming gust. Bioinspir. Biomim. 18 (2023) 066016 J Xue et al Figure 5. The relationship of agent-body-fluid interaction and the framework of the DDPG algorithm. K1–k7 are parameters or weights of different terms in the reward functions tuned depending on specific cases. Here, we use k1 = 2.8, k2 = −0.65, k3 = −0.02, k4 = 0.15, k5 = −3, k6 = 1, and k7 = −7.5. The red-inked terms in the reward function for ‘Locomotion pattern 2: Sea anchoring’ highlight the modifications made based on the reward function for ‘Locomotion pattern 1: Pattering’, which filters the forward velocity reward and adds an absolute x-position deviation penalty. Although the biomechanics of this behavior is not fully understood, it is hypothesized to help the storm petrel maintain balance and avoid being blown away by using its feet to generate hydrodynamic resist- ance, like an anchor, against the aerodynamic drag on the body, wings, and tail [13, 14]. This beha- vior is defined in this paper as ‘sea-anchoring.’ In videos that show this behavior, we observe that there are often significant waves while storm petrels sea- anchor on the water. Therefore, we assume that this behavior is associated with relatively high wind velocities. and Without validated quantitative kin- ematic data for these behaviors, hard-coded con- trol algorithms are not practically implemented. Therefore, we use a model-free DRL strategy for sen- sorimotor learning and motion planning to explore these behaviors. As graphically demonstrated in figure 5, we implement DDPG algorithm using the Reinforcement Learning Toolbox in Matlab (Matlab R2021b, MathWorks, Inc.), which we integrate with the storm petrel model and fluid environment mod- els (force models) constructed in the Simscape Multibody simulation environment (Matlab R2021b, MathWorks, Inc.). The relationship of this agent- body-fluid interaction and the framework of the DDPG algorithm is demonstrated in figure 5. 2.3.1. Observation For both training scenarios, the DRL agent will the environment at each observe the states of sampling point, defined and summarized in table 3, 7 and take the optimal action based on these states and the defined reward functions. In training scen- arios that encourage forward motion (e.g. pattering in section 3.1), the absolute x-position of the storm pet- rel model makes no difference regarding control and is therefore not one of the observed states. However, the absolute x-position is a meaningful indicator when the training scenario involves sea-anchoring or destination-specified locomotion, as described in section 3.2. Nevertheless, we include this x-position state only in the reward function, not in the obser- vation, of sea-anchoring training to have a uniform input space (observation) for both behaviors and improve the convergence of the model. 2.3.2. Terminating conditions We set multiple conditions that terminate the cur- rent training episode to prevent the storm petrel from drowning, to minimize unstable body orientations, and to accelerate the training process. The termin- ating conditions are summarized in table 3. These boundary conditions restrict the model’s range of motion and guide it to behave naturally and remain within the scope of this study. 2.3.3. Reward formulation Although many generalized reward formulations for achieving efficient and stable locomotion patterns have been validated on various terrains, designing reward functions for hybrid locomotion between air and water is still intricate due to the innate instability and complexity of the fluid environment. Bioinspir. Biomim. 18 (2023) 066016 J Xue et al Table 3. Variables definition and usage in the DRL framework. The double stars, ‘∗∗’, denotes that this variable is used in both locomotion training scenarios (pattering and sea-anchoring), while the single star, ‘∗’, indicates that it is only used in the sea-anchoring training. The bold font indicates variables that are in a vector form. Variable Definition Observation Reward Termination X Vx Y Vy Ywt Ytt Yfc θ ˙θ γ ϕ β α1 α2 α3 AOAt P a preAct The x-direction position of the body center The x-direction velocity of the body center The y-direction position of the body center The y-direction velocity of the body center The y-direction position of the wings’ tip The y-direction position of the tail’s tip The y-direction position of the feet’ center The pitch angle of the body (degree) The rotational velocity of the body The pitching angle of the wing The flapping angle of the wing The pitching angle of the tail The angle of the knee joint The angle of the ankle joint The angle of the metatarsal bones joint Angle of attack of the tail The vector of the instantaneous output power of joints The vector of the normalized instantaneous acceleration of joints The vector of the actions taken at the last sampling time ∗∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗ ∗∗ <= −2.5 m & >= 6.75 m <= −0.06 m & >= 0.4 m <= 0 <= 0 ∗∗ <= −50◦ & >= 50◦ <= −50◦ & >= 50◦ ∗∗ ∗∗ Specifically, in our case, except for stability and efficiency, either pattering or sea-anchoring has its own presumed biological purposes, which need more heuristics to guide and restrict the model. Based on early-stage exploratory testing, we sum- marize several empirical but vital rules that affect the storm petrel model’s locomotion or its capability to patter or sea anchor: (i) A reward for forward velocity encourages the emergence of locomotion, while a penalty for backward velocity can improve the gust- resistance ability. (ii) A constant survival reward helps to extend the surviving time under the unstable scenario of each episode. (iii) A penalty for large body pitch rotations improves the stability of the model. (iv) A penalty for the system’s large instantaneous net power output and acceleration at the joints limits and reduces the energy consumption (i.e. metabolic cost). (v) A penalty for x-position deviation from the ref- erenced location helps the model to stay around the reference. (vi) A reward for upward velocity encourages loco- motion at low wind speeds but causes instability at high wind speeds (e.g. tendency to be blown away by a gust). It is, therefore, not included in the final reward function. Based on the previous definition of pattering and sea-anchoring, pattering exhibits efficient forward locomotion while close to the water surface, while sea-anchoring focuses on staying around a certain 8 position. Thus, we differentiate the reward function for sea-anchoring from pattering, because of their dif- ferent focuses and purposes by filtering the forward velocity reward and including a penalty for the x- position offset from the initial position. The explicit mathematical formulations of reward functions are demonstrated in figure 5. 3. Results and discussion 3.1. Pattering the After approximately 3000 training episodes, DRL model enables the storm petrel model to exhibit persistent forward locomotion in behavior that resembles storm petrels in nature. The qualitat- ive comparison of video captures in nature and sim- ulation is presented in appendix B. The quantitative analysis of the pattering storm petrel in simulation is shown in figure 6. In general, this ‘pattering’ behavior can be successfully trained over ambient wind speeds ranging from 1 to 5 ms−1). Here, 1.5 m s−1 is a typical wind condition in which storm petrels are most likely to emerge pattering. Based on the definition in section 2.3 and simula- tion results, we characterize the pattering behavior of storm petrels as the following phases: (i) Landing: The storm petrel exerts negative torque (extension) all lower limb joints (knee, ankle, and metatarsal) during gliding to swing the legs forward and moves the wings dorsally before the legs drop into the water. (ii) Strike: The storm petrel applies a rapid burst leg and of positive torque (flexion) on all Bioinspir. Biomim. 18 (2023) 066016 J Xue et al Figure 6. (A-1): Emergence of the pattering behavior in the simulation after a DRL training under 1.5 ms−1 incoming wind speed (The simulation videos can be found in electronic supplemental material (ESM)): the blue dash line indicates the ocean surface, and ten motion captures are evenly extracted from the data set. (A-2): trajectories of foot web tip and leg joints: the light blue, red, and green trajectories represent the time-varying position of the web tip, Metatarsal joint, and ankle joint, respectively. Filled-in green squares, red rhombuses, and blue triangles denote the joints’ position at thirty sampled points evenly extracted from the data set. Black lines connect digitized joint positions in the same frame. (B) and (C): force profiles of the body, wings, tail, and webbed feet during the pattering in the forward (B-1) and vertical (B-2) global directions. (C-1) and (C-2) exhibit the net forces of all body parts in the forward and vertical directions, respectively. The forces are all normalized by the averaged weight of the storm petrel specimens (∼0.41 (N), F∗ = F/0.41. wing-flapping joints to generate both hydro- dynamic and aerodynamic thrust and positive vertical forces. (iii) Takeoff: The foot webs leave the water surface due to the momentum generated from the last phase. The storm petrel then starts to glide before recommencing the landing phase and repeating this process to jump forward. The dynamic contribution of each body part is shown in figures 6(B-1) and (B-2) for the X and Y dir- ections, respectively. Its Y-coordinate is dimension- less and normalized by the average weight of the pet- rel (0.41 N). It shows that the forward force is mainly provided by the feet, with an average impulsive force of up to four times body weight. Both the wings and feet provide upward force. The feet generate part of the upward force through impact with the water, and the wing’s angle of attack modulates lift production while maintaining appropriate body pitch. Once in the air, the petrel mainly glides, maintaining the for- ward relative flight speed out of the water. Table 4 shows the maximum torque, velocity, and power at each joint. The ankle produces most of the power associated with pattering, partially due to high angular velocities. These results are similar to those found for ducks extending their legs during aquatic take-offs, which demonstrate high ankle angular velo- cities and produce over three times as much muscle power in the ankle-extending lateral gastrocnemius muscle compared to terrestrial take-offs [35]. The simulation values we list may provide a preliminary 9 Bioinspir. Biomim. 18 (2023) 066016 J Xue et al Table 4. Maximum torque, velocity, and instantaneous power of each controllable joint. Joints Knee Ankle Metatarsal Wing Pitch Wing Flap Tail Pitch Max. torque (Nm) Max. velo- city (ms−1) Max. power (W) 0.13 0.17 0.05 0.028 0.07 0.022 32 100 30 5.2 20 5 4 15.2 1.5 1.4 1.2 0.09 reference for future robotic designs, especially actu- ation system design or motor selections. 3.2. Sea-anchoring Although sea-anchoring behavior has been math- ematically justified under a static scenario, previous research has not considered or analyzed the behavior from a dynamic perspective. In this work, we first evaluate the sea-anchoring dynamics of the storm petrel under a medium wind speed, 3 ms−1. It can be observed from figure 7(A) that the storm petrel keeps itself aloft and balanced by rotating its wings, tail, and body dynamically while dabbling its feet in the water. Figure 7(B) shows that the storm petrel stays around the initial position, although the centroid of the model oscillates slightly caused by the periodic leg strikes. The trained storm petrel model exhibits a high capability of maintaining the body balance even under an abrupt and consider- able body posture change (e.g. during the 1.5 s–2 s period). Though the peak value of aerodynamic drag is small, its persistent effect on the petrel requires rel- atively high positive hydrodynamic impulses released in a short time, shown in figures 7(C-1) and (D-1), to compensate for the induced momentum and main- tain the absolute body position in the horizontal direction. This sea-anchoring behavior should be regarded as dynamic rather than statically stable. This stabil- ity is achieved by the storm petrel model persist- ently coordinating, adjusting, or manipulating all the effective body components to balance the trade-off between the weight support and drag. The fact that the increase in the lift of either the wings or the tail will generally accompany an increase in the aerody- namic drag under a low angle of attack causes a trade- off or a control challenge. Additionally, we observe and analyze the change in sea-anchoring perform- ance and behavioral pattern as wind speed increases from 1 ms−1 to 8 ms−1 through separate simulations. It is unsurprising that the sea-anchoring perform- ance is limited and will be affected by the increasing wind speed, which can be seen as a gradually back- ward average x-position, see figure 8(A). It should be acknowledged that the actual environment condition is more complicated than a constant wind speed and chances that the highest wind speed in nature under which storm petrels exhibit sea-anchoring is much higher than in the simulation. Despite that, there has been a hypothesis that storm petrels will seek shelter in the wave troughs, suggesting a much lower local wind speed compared to the free stream [14]. The contributions of each body component (tail, wings, feet, and body) to sub-tasks (e.g. weight sup- port, balance) during sea-anchoring differ from those observed during pattering and vary with wind speed. Specifically, we observe that simulated sea-anchoring storm petrels alter wing pitch instead of flapping (as observed in pattering simulations), which helps them maintain balance as wind speed increases. The lack of flapping, especially at high wind speeds, could reduce the risk of generating excess lift and elevating storm petrel’s feet out of the water. This inference is suppor- ted by early-stage exploratory testing where flapping causes the petrel to be blown backward or flipped over after the foot leaves the water. Besides, as shown in figures 8(B) and (C), the drag acting on the wings is almost constant across wind speeds through the change of angle of attack modu- lated by the pronation or supination (pitching) of the wings. This is supported by the concomitant reduc- tion of aerodynamic lift generated by the wings across increasing wind speed. The tail plays an increas- ingly significant role in generating lift as the wind grows stronger, which could provide weight support and counteract the rotational momentum induced by the wings so as to maintain an appropriate body angle. In addition, the simulated sea-anchoring storm petrel submerges its feet more frequently and for longer durations under high wind conditions. At low wind speeds, hydrodynamic forces acting on the feet provide weight support (vertical) but limited propulsion (horizontal) (figures 8(B) and (C)). At higher wind speeds, these hydrodynamic forces shift to primarily generating forward propulsion or thrust. This aligns with intuition—low wind speeds gener- ate less aerodynamic lift, requiring stronger contribu- tions from the feet, but stronger winds induce larger backward drag on the wings, demanding counteract- ing propulsion from the feet. The force modulation by the feet could be achieved by varying the trajectory or angle of attack of the feet. It is interesting to note that the total vertical force produced by all body components is less than the storm petrel’s weight (∼0.41 N). The result that the averaged vertical force is less than the body weight could be explained by the discrete striking pattern of the legs, which leads to a less-than-one value after the time average. The increase of this magnitude could, therefore, be explained by an increase in the striking frequency or duty cycle. 10 Bioinspir. Biomim. 18 (2023) 066016 J Xue et al Figure 7. Emergence of the sea-anchoring behavior in the simulation. (A) Motion captures (2.2 Hz) of a sea-anchoring petrel with an incoming wind speed of 3 ms−1. The blue dashed line denotes the ocean surface. (B) Profile of the x-position of the body centroid across time (y-axis). (C-1) and (C-2) show the net aerodynamic forces acting on all body parts in the x- and y-directions, respectively. (D-1) and (D-2) show the net hydrodynamic forces along the x- and y-respectively. All the forces are normalized here by the storm petrel’s weight (∼0.41 N). 11 Bioinspir. Biomim. 18 (2023) 066016 J Xue et al Figure 8. Displacement and force metrics of sea-anchoring storm petrels simulated under different constant wind velocities. All values are calculated as averages across the training episodes (all more than 3 s) with the corresponding wind speed. (A) Mean forward (x) displacement of the foot relative to the global starting position. (B) Mean forward (x-direction) forces acting on different body parts (foot, wings, tail, trunk). (C) Mean vertical (y-direction) forces acting on different body parts. The x- and y-forces are normalized using the averaged weight of measured storm petrel specimens (∼0.41 (N)), F∗ = F/0.41. 4. Conclusion In this work, we simulate two storm petrel for- aging behaviors at the air–water interface using meas- ured anatomical data of storm petrel specimens and quasi-steady fluid dynamics models. The complex locomotive motions of the storm petrel are planned using DRL techniques, successfully generating stable behaviors that qualitatively resemble observations of pattering (forward locomotion) and sea-anchoring (stationary) behaviors in nature at different wind speeds. However, we have to acknowledge that due to the scope of this initial study and the shortage of resources, the simplified storm petrel model and quasi-steady fluid dynamic models used in the sim- ulation have limitations and might, to some extent, affect the results of our simulations. While the quasi-steady model may not fully cap- ture the complexities of the fluid dynamics of storm petrel behavior, the DRL models generated in this study provide novel insight into why birds choose pat- tering or sea-anchoring and provide a foundation to further explore reinforcement learning as a way to test behavioral hypotheses. The model presented here performs pattering when forward motion is rewar- ded and sea-anchoring when maintaining a station- ary position is rewarded. These results support is lift, while the provided mainly by the wings’ forward propulsion is primarily generated by the show that weight 12 hydrodynamic drag produced by the webbed feet. To maintain a stable body posture, the pattering simu- lations coordinate different body components (e.g. wings and tail) to reduce net torque. During sea- anchoring simulations that prioritize a stationary global position, the RL agent submerges the feet to induce hydrodynamic resistance against the aero- dynamic drag acting on the wings and tail. When increasing constant wind speeds, sea-anchoring sim- ulations (1) adjust the pitch of the wings and reduce flapping to reduce drag and lift, and (2) increasingly use the tail to counteract pitch moments generated by the wings. Additionally, we present the integration approach used in this manuscript as a strategy for inform- ing our understanding of otherwise elusive animal behaviors, and we hope our methodological frame- work inspires future investigations into storm pet- rel behavior and/or other multidisciplinary investig- ations. For example, additional analyses could exam- ine how variation in wing loading and foot loading in storm petrels [12] contributes to pattering suc- cess, and therefore the evolution of this unique beha- vior within Procellariiformes. Observations of storm petrel foraging under varying environmental con- ditions may confirm behavioral plasticity correlated with wind speeds, and therefore predict the influence of climatic shifts and anthropogenic influences, such as ocean wind turbines [36]. Nevertheless, learning algorithms used in this the reduced-order models and study have Bioinspir. Biomim. 18 (2023) 066016 J Xue et al improvements that can be made in potential including (1) collecting more future research, quantitative data on storm petrel pattering/sea- anchoring to validate modeling approaches, (2) exploring new learning techniques, such as reward to enhance the interoperabil- shaping [37, 38], ity of the AI ‘black box’, and (3) using empir- ical data to measure the drag coefficient of objects that replicate a storm petrel’s anatomy by fluid dynamic experiments to build a higher fidelity force model [39–41]. Data availability statement All data that support the findings of this study are included within the article (and any supplementary files). Acknowledgments This work is partially supported by grants from the research initiation grant provided by Westlake University (No. 103110556022101). The authors are grateful to Dr Yuqing Chen at Xi’an Jiaotong- Liverpool University for the useful discussion on dynamic modeling and nonlinear control and PhD candidate Bing Luo at Westlake University for the guidance on 3D flapping wing kinematics. The authors also want to acknowledge the University of Washington Burke Museum of Natural History for access to petrel specimens and the Macaulay Library at the Cornell Lab of Ornithology for access to videos of storm petrels. Conflict of interest We declare we have no competing interest. Author contributions J X: conceptualization, primary modeling of both the biological system and fluid environment, DRL implementation and tuning, data collection, formal analysis, methodology, software, validation, visualiz- ation, writing—original draft, writing—review and editing; F H: modeling, DRL tuning, data collec- tion and analysis, Graph design and plot, writing— literature review and editing; B K v O: Anatomical measurements and photos of storm petrel speci- mens, modeling of the biological system, biological hypothesis justification, writing—review and editing; experiment design; G C: Biological and biomech- anics hypothesis and assumption justification, mod- eling of the biological system, experiment design; D F: Conceptualization, modeling of both the fluid environment, funding acquisition, supervision, pro- ject administration, writing—review and editing. All authors give final approval for publication and agree to be held accountable for the work performed therein. Appendix A. Anatomical measurements on storm petrel specimens We take photographs of specimens available at the University of Washington Burke Museum of Natural History (Seattle, WA, USA). The measurements are calibrated by a ruler put in the scene and analyzed using ImageJ (ImageJ 1.53o, National Institutes of Health (NIH)). Web angle is defined as the maximum acute intersection angle of the foot webs. Total length is measured from the specimens’ head tip to the tail tip. Trunk length is measured from the head tip to the tail root of the specimens, while tail length is meas- ured from the tail root to the tail tip. 13 Bioinspir. Biomim. 18 (2023) 066016 J Xue et al Table A1. Skeleton measurements (mm). Species Femur (mm) Tibiotarsus (mm) Tarsometatarsus (mm) Oceanites oceanicus Oceanodroma furcata Oceanodroma leucorhoa Average 14.9 18.5 15.1 16.2 46.1 38.3 34.9 39.8 33.5 23.6 21.4 26.2 Table A2. Skin Measurements (mm). Species Web Angle (degree) Total Length (mm) Trunk Length (mm) Tail Length (mm) Tarsometatarsus (mm) Metatarsals (mm) Oceanites oceanicus Oceanodroma furcata Oceanodroma leucorhoa N/A 51.0 Average 42 60 187 224 194.5 201.8 129.7 160.6 130 140.1 57.3 63.4 64.5 61.73 34.9 N/A N/A 34.9 26 21.9 25.1 24.3 Figure B1. (A) Video footage of the storm petrel in nature (Reproduced with permission from [25]. [© Josep del Hoyo]). (B) Captures of different pattering cycles of the storm petrel in simulation. The landing phase for all cycles recorded here is timestamped as 0 ms, while the striking and takeoff phases are timestamped relative to the landing phase capture respectively for each pattering cycle. Appendix B. Video footage comparison of pattering storm petrel in simulation and nature We capture and timestamp the gestures of patter- ing storm petrels in the simulation and subsequently compare these with video footage of storm petrels in nature. Overall, we observe that the timing of the striking and takeoff phases is of a similar order of magnitude. Although figure B1 suggests that the sim- ulation’s timestamps can vary between different pat- tering cycles and differ from those seen in real storm petrels, we find this variation acceptable. It can be attributed to factors such as the low frame rate of the video source, the limited availability of capture data, and the challenges of manually aligning phases (land- ing, striking, takeoff) across different cycles. ORCID iDs Jiaqi Xue  https://orcid.org/0000-0002-8590-3764 Fei Han  https://orcid.org/0009-0007-7982-1870 Brett Klaassen van Oorschot  https://orcid.org/0000-0003-4347-5391 Glenna Clifton  https://orcid.org/0000-0002-5806- 7254 Dixia Fan  https://orcid.org/0000-0002-6201-5860 References [1] Horricks R A, Bannister C, Lewis-McCrea L M, Hicks J, Watson K and Reid G K 2022 Comparison of drone and vessel-based collection of microbiological water samples in marine environments Environ. Monit. Assess. 194 1–9 14 Bioinspir. Biomim. 18 (2023) 066016 J Xue et al [2] Wang J, Li G and Chen F 2022 Eco-environmental effect evaluation of tamarix chinesis forest on coastal saline-alkali land based on RSEI model Sensors 22 5052 [3] Gonçalves L and Damas B 2022 Automatic detection of rescue targets in maritime search and rescue missions using UAVs 2022 Int. Conf. on Unmanned Aircraft Systems (ICUAS) (IEEE) pp 1638–43 [4] Winfield A F T et al 2016 euRathlon 2015: a multi-domain multi-robot grand challenge for search and rescue robots Annual Conf. Towards Autonomous Robotic Systems (Springer) pp 351–63 [5] Hsieh S T and Lauder G V 2004 Running on water: deep reinforcement learning IEEE Robot. Autom. Lett. 6 7193–200 [23] Kambic R E, Roberts T J and Gatesy S M 2014 Long-axis rotation: a missing degree of freedom in avian bipedal locomotion J. Exp. Biol. 217 2770–82 [24] Verstappen M, Aerts P and Van Damme R 2000 Terrestrial locomotion in the black-billed magpie: kinematic analysis of walking, running and out-of-phase hopping J. Exp. Biol. 203 2159–70 [25] Hoyo J del 2013 White-faced Storm-Petrel (ML201429031). The macaulay library at the cornell lab of ornithology (available at: https://macaulaylibrary.org/asset/201429031) three-dimensional force generation by basilisk lizards Proc. Natl Acad. Sci. 101 16784–8 [26] Munson B R, Okiishi T H, Huebsch W W and Rothmayer A P 2013 Fluid Mechanics (Wiley) [6] Bush J W M and Hu D L 2006 Walking on water: biolocomotion at the interface Ann. Rev. Fluid Mech. 38 339–69 [7] Ma J, Lu H, Li X and Tian Y 2020 Interfacial phenomena of water striders on water surfaces: a review from biology to biomechanics Zool. Res. 41 231 [8] Dornburg A and Near T J 2021 The emerging phylogenetic perspective on the evolution of actinopterygian fishes Annu. Rev. Ecol. Evol. Syst. 52 427–52 [27] Mivehchi A, Zhong Q, Kurt M, Quinn D B and Moored K W 2021 Scaling laws for the propulsive performance of a purely pitching foil in ground effect J. Fluid Mech. 919 R1 [28] Zhong Q, Han T, Moored K W and Quinn D B 2021 Aspect ratio affects the equilibrium altitude of near-ground swimmers J. Fluid Mech. 917 A36 [29] Yang C, Huang F and Noblesse F 2013 Practical evaluation of the drag of a ship for design and optimization J. Hydrodyn. 25 645–54 [9] Clifton G T, Hedrick T L and Biewener A A 2015 Western [30] Hoerner S F 1965 Fluid-dynamic drag. Theoretical, and clark’s grebes use novel strategies for running on water J. Exp. Biol. 218 1235–43 [10] Denny M W 2004 Paradox lost: answers and questions about experimental and statistical information. Copyright by: SF Hoerner Fluid Dynamics, Vancouver, Printed in the USA, Card Number 64-19666 walking on water J. Exp. Biol. 207 1601–6 [11] Park H and Choi H 2010 Aerodynamic characteristics of flying fish in gliding flight J. Exp. Biol. 213 3269–79 [12] Sausner J, Torres-Mura J C, Robertson J and Hertel F 2016 Ecomorphological differences in foraging and pattering behavior among storm-petrels in the eastern Pacific Ocean Auk 133 397–414 [31] Wei X, Xu L, Zhao J and Cao K 2012 Study and simulation of control method based biped robot walking on water Proc. 31st Chinese Control Conf. (IEEE) pp 4900–3 [32] Klaassen van Oorschot B, Mistick E A and Tobalske B W 2016 Aerodynamic consequences of wing morphing during emulated take-off and gliding in birds J. Exp. Biol. 219 3146–54 [13] Sugimoto T 1998 A theoretical analysis of sea-anchor soaring [33] Thomas A L R 1993 On the aerodynamics of birds’ tails Phil. J. Theor. Biol. 192 393–402 Trans. R. Soc. B 340 361–80 [14] Withers P C 1979 Aerodynamics and hydrodynamics of the ‘hovering’ flight of Wilson’s storm petrel J. Exp. Biol. 80 83–91 [34] Tucker V A 1992 Pitching equilibrium, wing span and tail span in a gliding harris’ hawk, parabuteo unicinctus J. Exp. Biol. 165 21–41 [15] Lillicrap T P, Hunt J J, Pritzel A, Heess N M O, Erez T, Tassa Y, Silver D and Wierstra D 2015 Continuous control with deep reinforcement learning CoRR (arXiv:1509. 02971) [16] Sutton R S and Barto A G 2018 Reinforcement Learning: An Introduction (MIT Press) [35] Taylor-Burt K R and Biewener A A 2020 Aquatic and terrestrial takeoffs require different hindlimb kinematics and muscle function in mallard ducks J. Exp. Biol. 223 jeb223743 [36] Furness R W, Wade H M and Masden E A 2013 Assessing vulnerability of marine bird populations to offshore wind farms J. Environ. Manag. 119 56–66 [17] Mnih V et al 2015 Human-level control through deep [37] Memarian F, Goo W, Lioutikov R, Niekum S and Topcu U reinforcement learning Nature 518 529–33 [18] Silver D et al 2016 Mastering the game of go with deep neural networks and tree search Nature 529 484–9 2021 Self-supervised online reward shaping in sparse-reward environments 2021 IEEE/RSJ Int. Conf. on Intelligent Robots and Systems (IROS) pp 2369–75 [19] Wang X, Wang S, Liang X, Zhao D, Huang J, Xu X, Dai B and Miao Q 2022 Deep reinforcement learning: a survey IEEE Trans. Neural Netw Learn. Syst. [38] Okudo T and Yamada S 2021 Reward shaping with dynamic trajectory aggregation 2021 Int. Joint Conf. on Neural Networks (IJCNN) pp 1–9 [20] Zhu K and Zhang T 2021 Deep reinforcement learning based mobile robot navigation: a review Tsinghua Sci. Technol. 26 674–91 [21] Centurelli A, Arleo L, Rizzo A, Tolu S, Laschi C and Falotico E 2022 Closed-loop dynamic control of a soft manipulator using deep reinforcement learning IEEE Robot. Autom. Lett. 7 4741–8 [39] Leung Chan W and Kang T 2011 Simultaneous determination of drag coefficient and added mass IEEE J. Ocean. Eng. 36 422–30 [40] Zhang C, Huang H and Lu Xi-Y 2020 Effect of trailing-edge shape on the self-propulsive performance of heaving flexible plates J. Fluid Mech. 887 A7 [41] Han P, Lauder G V and Dong H 2020 Hydrodynamics of [22] Wang J, Hu C and Zhu Y 2021 CPG-based hierarchical locomotion control for modular quadrupedal robots using median-fin interactions in fish-like locomotion: effects of fin shape and movement Phys. Fluids 32 011902 15
null
10.1093_nar_gkad633.pdf
Analyses and data acquisition codes are upload on lab GitHub account and archi v ed in Zenodo with the following doi. Additionally, raw data that support our findings have Nucleic Acids Research, 2023, Vol. 51, No. 17 8967 been uploaded and archi v ed in Zenodo, corresponding to each individual figure. GitHub: https://github.com/Ha-SingleMoleculeLab Analyses , data acquisition codes , and raw data are archi v ed in Zenodo: Raw data analysis DOI: 10.5281 / zenodo.4925617 Data acquisition DOI: 10.5281 / zenodo.4925630
Data acquisition DOI: 10.5281 / zenodo.4925630 Raw data DOI: 10.5281 / zenodo.8088172
Published online 31 July 2023 Nucleic Acids Research, 2023, Vol. 51, No. 17 8957–8969 https://doi.org/10.1093/nar/gkad633 Linking folding dynamics and function of SAM / SAH riboswitches at the single molecule level Ting-Wei Liao 1 , Lin Huang 2 , Timothy J. Wilson 3 , Laura R. Ganser 1 , David M.J. Lilley 3 and Taekjip Ha 1 , 4 , 5 , * 1 Department of Biophysics, Johns Hopkins University, Baltimore, MD 21218, USA, 2 Guangdong Provincial Key Laboratory of Malignant Tumor Epigenetics and Gene Regulation, Guangdong-Hong Kong Joint Laboratory for RNA Medicine, Sun Yat-Sen Memorial Hospital, Sun Yat-Sen University, Guangzhou 510120, China, 3 Nucleic Acid Structure Research Group, MSI / WTB Complex, The University of Dundee, Dundee, Dow Street, Dundee DD1 5EH, UK, 4 Department of Biophysics and Biophysical Chemistry, Johns Hopkins University School of Medicine, Baltimore, MD 21205, USA and 5 Ho w ard Hughes Medical Institute, Baltimore, MD, USA Received April 19, 2023; Revised June 27, 2023; Editorial Decision July 12, 2023; Accepted July 18, 2023 ABSTRACT GRAPHICAL ABSTRACT Riboswitches are regulatory elements found in bacterial mRNAs that control downstream gene expression through ligand-induced conformational chang es. Here , we used single-molecule FRET to map the conformational landscape of the translational SAM / SAH riboswitch and probe how ligand-induced conformational co-transcriptional c hanges aff ect its translation regulation function. Ri- boswitch folding is highly heterog eneous, sugg est- ing a rugged conformational landscape that allows for sampling of the ligand-bound conformation even in the absence of ligand. The addition of ligand shifts the landscape, favoring the ligand-bound con- formation. Mutation studies identified a key struc- tural element, the pseudoknot helix, that is crucial for determining ligand-free conformations and their ligand responsiveness. We also investigated ribo- somal binding site accessibility under tw o scenar - ios: pre-folding and co-transcriptional folding. The regulatory function of the SAM / SAH riboswitch in- volves kinetically favoring ligand binding, but co- transcriptional folding reduces this preference with a less compact initial conformation that exposes the Shine–Dalgarno sequence and takes min to redis- tribute to more compact conformations of the pre- folded riboswitc h. Suc h slow equilibration decreases the effective ligand affinity. Overall, our study pro- vides a deeper understanding of the complex folding process and how the riboswitch adapts its folding pattern in response to ligand, modulates ribosome accessibility and the role of co-transcriptional fold- ing in these processes. INTRODUCTION Riboswitches are regulatory units of RNA that mediate gene expression in response to binding of specific metabo- lites. They are widely found in bacteria ( 1–3 ) but also exist in archaea ( 4 ), plants ( 5 ) and fungi ( 6 , 7 ). To date, > 40 classes of riboswitches have been discovered, and they bind chem- ically di v erse ligands and contribute up to 4% to the bacte- ria genetic control, especially in gram positi v e bacteria. Ri- (cid:2) -untransla ted regions boswitches are mostly loca ted a t the 5 (cid:2) -UTR), upstream of the regulated genes, and include an (5 a ptamer domain ca pable of binding a particular metabolite with exceptionally high specificity. The riboswitch adopts a specific fold on binding the ligand, leading to up- or down-regulation of the gene either by altering transcrip- tion or translation. Since the riboswitch folds and acts as a regulatory unit during transcription, the timing of ligand * To whom correspondence should be addressed. Tel: +1 217 398 0865; Email: tjha@jhu.edu C (cid:3) The Author(s) 2023. Published by Oxford University Press on behalf of Nucleic Acids Research. This is an Open Access article distributed under the terms of the Creati v e Commons Attribution License (http: // creati v ecommons.org / licenses / by / 4.0 / ), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. 8958 Nucleic Acids Research, 2023, Vol. 51, No. 17 binding and conformational change is critical, necessitating investigation into its folding kinetics. Substantial r esear ch has been devoted to the origins of specificity, correlation of sequence and structure ( 8 , 9 ), folding kinetics ( 10–12 ), and identification of candidates that can be adapted for drug- deli v ery ( 13 ) and in vivo imaging ( 14 ). (SAM)-binding S-adenosylmethionine riboswitches comprise one of the largest classes of riboswitches ( 1 ). SAM is synthesized from methionine and ATP by SAM synthetase, encoded by the metK gene. SAM is an es- sential co-substrate of methyltransferases, supplying the methyl group for methyl transfer. Once the methyl group is donated, the resulting S -adenosyl- L -homocysteine (SAH) is degraded due to its toxicity ( 15 , 16 ). To main- tain SAM concentration, SAM acts as an inhibitor of MetK synthesis ( 17–20 ) . This regulation is achie v ed by the SAM-riboswitches, which bind SAM and acts as negati v e feedback unit for genes in methionine or cysteine biosynthesis. W hen SAM concentra tion goes up, expres- sion of genes in methionine or cysteine biosynthesis is (cid:2) -UTR adopting transla tion-of f reduced by its upstream 5 conformation. Six sub-classes of SAM riboswitches (SAM-I to SAM- VI) have been identified, classified into three families ac- cording to their structural features ( 17–25 ). In general, SAM riboswitches exhibit strong discrimination between SAM and SAH by electrostatically interacting with the positi v e-charged sulfonium cation of the SAM molecule ( 26–33 ), as previously shown by X-ray crystallography and single-molecule methods ( 22 , 34 ). By contrast, the SAM / SAH riboswitch does not discriminate between SAM and SAH ( 35 , 36 ). The ligand binding interactions of this particular SAM / SAH riboswitch have been pr eviously r evealed by NMR and X-ray crystallo gra phy ( 35 , 36 ). Binding of SAM or SAH is accompanied by the formation of three base pairs that extend the helix of the stem-loop P1, and formation of a pseudoknot helix PK (Supplementary Figure S1A). The ligand binds in the major groove of the extended he- lix, with the methionyl nitrogen and the adenine moiety hy- drogen bonded to a specific cytosine nucleobase. The me- thionine side chain containing the sulfonium of SAM or the thioether of SAH does not make any direct contact with the RN A, w hich explains the inability to distinguish between the two ligands. Single-molecule FRET (fluorescence res- onance energy transfer) ( 37 ) was utilized to compare the binding of SAM and SAH and their kinetic characteristics ( 36 ), and no significant differences were observed between the two ligands. Although the ligand-bound state and basic kinetics have been characterized, important features such as ligand-free conformations, binding, folding kinetics, and its role in modula ting transla tion initia tion activity are still unknown. Here, we used single-molecule FRET to investigate the ligand-free and ligand-bound conformations of the SAM / SAH riboswitch and map the energy landscape of folding dynamics and altered ribosome accessibility. Fold- ing of the riboswitch is highly heterogeneous, suggesting a rugged conformational landscape that allows for sam- pling of the ligand-bound conformation e v en in the ab- sence of ligand. The addition of ligand shifts the land- scape, favoring the ligand-bound conformation. Site spe- cific mutations showed that the PK helix is crucial for de- ter mining ligand-free confor mations and their ligand re- sponsi v eness. In addition, we investigated the accessibility of the ribosomal binding site under two scenarios: ( i ) pre- folding of the riboswitch: folding equilibrium is reached in advance and ( i i) vectorial release of the RNA by mim- icking co-transcriptional folding. Vectorial folding initially favors an open conformation that exposes the ribosome binding site, and it takes min bef ore conf ormational redis- tribution to that of the pre-folded riboswitch. Such slow equilibration decreases the effecti v e ligand affinity. Over- all, our studies offer a deeper understanding of the com- plexity of the folding process, revealing the mechanism by which the riboswitch adapts its folding pattern in response to ligand and modulates ribosome accessibility, and how co- transcriptional folding influences these processes. MATERIALS AND METHODS Riboswitch ligands SAM (A7007), SAH (A9384) were all obtained from Sigma. The RNAs (wide-typed and mutants) are synthesized as described in the following sections. DNA oligonucleotides for mimicking ribosome binding were pur- chased from Integrated DNA Technologies (Coralville). RNA synthesis for single-molecule experiments The wild-type and mutated SAM / SAH riboswitches for single molecule measurements contain a Cy3 flu- orophore attached to the O2’ of U20 generated by Cu 2+ -catalyzed reaction of alkyne-modified RNA with an azide-attached fluorophore (Lumiprobe Corp). (cid:2) DNA exten- The wild-typed RNA had an 18 nt 3 sion for base-pairing to the anchor DNA, and the complete sequence was (DNA starts with d and under- scored): GAUACCUGUCACAACGGCU(U-Cy3)CCU GGCGUGA CGAGGUGA CCUCAGUGGAGCAA d( ACCGCTGCCGT CGCT CCG ), and all the other mu- tated sequences were showed in Supplementary Table S1. (cid:2) -biotin and 3 The anchor DNA had a 5 (cid:2) Cy5 flu- orophore and was complementary to the 18-nt ex- tension of the SAMSAH riboswitch strand. Its se- quence was: biotin-CGGA GCGACGGCA GCGGT-Cy5. RNA oligonucleotides were synthesized using t-BDMS phosphoramidite chemistry ( 38 ) as described in Wil- son et al. ( 39 ), implemented on an Applied Biosys- tems 394 DN A / RN A synthesizer. RN A was synthesized using ribonucleotide phosphoramidites with 2 -Otter- butyldimethyl-silyl (t-BDMS) protection (Link Technolo- gies) ( 40 , 41 ). Oligonucleotides containing 5-bromocytidine (ChemGenes) were deprotected in a 25% ethanol / ammonia ◦C. All oligoribonucleotides were re- solution for 36 h at 20 dissolved in 100 (cid:2)l of anhydrous DMSO and 125 (cid:2)l tri- eth ylamine trih ydrofluoride (Sigma-Aldrich) to remove t- ◦C in the dark for 2.5 h. BDMS groups, and agitated at 65 After cooling on ice for 10 min, the RNA was precipitated with 1 ml of butanol, washed once with 70% ethanol and suspended in double-distilled w ater. RNA w as further puri- fied by gel electrophoresis in polyacrylamide under denatur- ing conditions in the presence of 7 M urea. The full-length RNA product was visualized by UV shadowing. The band was excised and electroeluted using an Elutrap Electroelu- tion System (GE Healthcare) into 45 mM Tris-borate (pH ◦C . The 8.5), 5 mM EDTA buf fer for 12 h. a t 150 V a t 4 RNA was precipitated with isopropanol, washed once with 70% ethanol and suspended in water or ITC buffer (40 mM HEPES-K (pH 7.0), 100 mM KCl, 10 mM MgCl 2 ). Ligand titration of pre-folded riboswitches: wild-typed and mutated riboswitches 40 pM of the pre-annealed SAM / SAH riboswitch immobilized on a neutravidin- molecules were functionalized, polymer-passivated surface and free molecules were washed out with T50 b uffer. Ima ge b uffer containing an oxygen-scavenging system was freshly mixed befor e measur ements , comprising 1% (w / v) dextrose , 2 mM Trolo x, glucose o xidase (1 mg / ml; Sigma-Aldrich), and catalase (500 U / ml; Sigma-Aldrich)] in buffer containing 40 mM HEPES (pH 7.5), 100 mM KCl, 2 mM MgCl 2 . All the ligands were diluted with image buffer immediately prior to measurements. The ligands were incubated for 5 min before imaging. Short movies (duration of 1.5 s: 20 frames) were collected for 30 field of view for generating distribution of FRET efficiencies ( E FRET ). The distribution is then fitted by two individual Gaussian function, and the high E FRET ratio is estimated accordingly. Single-molecule imaging and data acquisition Single-molecule FRET data were obtained using a prism- based total internal reflection fluorescence (TIRF) micro- scope. The Cy3 and Cy5 fluorophores were excited by a 532- nm laser (Coherent Compass 315M) and a 638-nm laser (Cobolt 06-MLD) respecti v ely. The fluorescence emission was collected by a water immersion objecti v e (Olympus NA 1.2, 60 ×) and recorded by a back-illuminated electron- m ultipl ying charge-coupled device camera (iXON, Andor Technology) with a dual-vie w setup. The dual-vie w setup used a long-pass emission filter (Semrock BLP02-561R- 25) for eliminating the 532-nm laser, and a notch filter (Chroma ZET633TopNotch) for eliminating the 638-nm laser. The fluorescence emission was separated into donor and acceptor emission by a long-pass dichr oic mirr or (Sem- rock FF640-FDi01-25X36). The passivated PEG quartz slides and coverslips were purchased from Johns Hopkins Slides Core and were assembled into a reaction cham- ber. ( 42 ) Spots detection, background subtraction, donor leakage and acceptor direct-excitation correction followed our previous protocol ( 42 ). Custom codes are available on GitHub ( https://github.com/Ha-SingleMoleculeLab ) and archi v ed in Zenodo with the following doi. Data acqui- sition DOI: 10.5281 / zenodo.4925630; Raw data analysis DOI: 10.5281 / zenodo.4925617. Nucleic Acids Research, 2023, Vol. 51, No. 17 8959 E FRET fluctuated between the middle and high values. To characterize further the dynamic species, the regions of dy- namics were collected and analyzed by ebFRET ( 43 ) and the two-state dwell time was plotted into log-scale scatter plot. Single-molecule data analysis of vectorially folded riboswitch Single-molecule traces showing the immobilized heterodu- plex was unwound were categorized into four types of be- havior. We classified the riboswitch folding behavior into four types (Supplementary Figure S7A): (i) molecules tran- sitioned from the heteroduplex state to the closed confor- mation without any detectable intermediate, then remain- ing there, (ii) molecules transitioned from the heteroduplex state to the open conformation, then remaining there, (iii) molecules transitioned from the heteroduplex state to one undergoing fluctuations between the open and closed con- formations, (iv) molecules transitioned from the heterodu- plex to fluctuating states after which they became locked in the closed conformation. Pseudo-functional readout of ribosome accessibility of pre- folded riboswitch 40 pM of the pre-annealed SAM / SAH riboswitch immobilized on a neutravidin- molecules were functionalized, polymer-passivated surface and free molecules were washed out with T50 buffer. All the ligand (SAM) and DNA oligonucleotides with a designated con- centration was freshly mixed with ima ge b uffer containing an oxygen-scavenging system. Short movies (duration of 1.5 sec: 20 frames) were collected for 30 field of view immedia tely or a t designa ted time (5-min or 1 h) after injection for generating distribution of FRET efficiencies ( E FRET ). Sample pr epar ation f or the single-molecule FRET measur e- ments For preparation of the pre-folded assay, 10 (cid:2)M of the SAM / SAH riboswitch molecule or the riboswitch mutants with internal Cy3 labeled was annealed with 15 (cid:2)M an- chored DNA with Cy5 and biotin label under 1 × T50 [10 mM Tris (pH 8.0), 50 mM NaCl) buffer follo wed by slo w ◦C to room temperature. cooling from 95 For preparing of the vectorial folding assays, 10 (cid:2)M an- chored DNA with Cy5 and biotin label was annealed with 20 (cid:2)M of the Cy3-labeled SAM / SAH riboswitch and 40 (cid:2)M complementary DNA oligos (cDNA) with dT30 over- hang in 10 (cid:2)l of 1 × T50 [10 mM Tris (pH 8.0), 50 mM NaCl] ◦C for 5 min, by incubating the mixture at 95 ◦C for 15 min and finally equilibrating at room tempera- 37 ture for 5 min ( 44 , 45 ). ◦C for 1 min, 75 Single-molecule data analysis of pr e-f olded riboswitch Vectorial folding as a mimic of riboswitch folding and ligand binding Single-molecule traces showing E FRET as a function of time were categorized into three types of behavior. (i) E FRET re- mained middle for the duration of observation, up to 1 min, (ii) E FRET remained high for the duration of observation (iii) Labeled and biotinylated heteroduplex es wer e immobilized on a neutravidin-functionalized surface and free heterodu- plex es wer e washed out. 50 nM of Rep-X ( 46 ) was incu- bated for 2 min with heteroduplexes in the imaging buffer 8960 Nucleic Acids Research, 2023, Vol. 51, No. 17 (40 mM HEPES (pH 7.5), 100 mM KCl, 2 mM MgCl 2 ) containing an oxygen-scavenging system, and images (du- ration of 1.5 s: 20 frames) were collected for 30 field of view for confirming the heteroduplex conformation. Unwind- ing was initiated by mixing unwinding buffer with / without ligands at designated concentration. Unless specified oth- erwise, the unwinding buffer contained 40 mM HEPES (pH 7.5), 100 mM KCl, 2 mM MgCl 2 , 2 mM ATP with an oxygen-scavenging system. Buffer with ATP then trig- gered the pre-bound RepX into unwinding the anchored heteroduplex. For the real-time observation (‘flow-in’ experiments) of riboswitch released from heteroduplex, imaging was started 12 s before the addition of the unwinding buffer. For char- acterization of VFA products after helicase unwinding, im- ages were taken after the addition of the unwinding buffer a t designa ted time. The loading and unwinding buf fers used during imaging contained additional 1% (w / v) dextrose, 2 mM Trolo x, glucose o xidase (1 mg / ml; Sigma-Aldrich), and catalase (500 U / ml; Sigma-Aldrich). Vectorial folding with ligand and ribosome mimic addition si- multaneously Labeled and biotinylated heteroduplex es wer e incubated with 50 nM Rep-X as previously described, images were taken before addition to confirm the heteroduplex confor- mations. Ligands, ribosome mimics (oligonucleotides with 9-nt or 15-nt complementary to ribosome binding site), and additional Rep-X (50 nM for 9-nt; 100 nM for 15-nt) wer e mix ed with unwinding buffer. The additional Rep-X is added in order to reduce the competition of free oligonu- cleotides to the Rep-X pre-incubated before unwinding mixture. This optimized condition is tested with a nega- ti v e control, where additional Rep-X with dT9 or dT15 were added sim ultaneousl y into the heteroduplex, no ob- servable loss of unwinding efficiency in this condition. The negati v e control experiments were shown in Supplementary Figure S10. RESULTS Heterogeneous folding energy landscape of ligand-free ri- boswitch First, we determined the conforma tional d ynamics of the SAM / SAH riboswitch in the absence of ligand. Single ri- boswitch molecules were tethered to the quartz slide by hy- (cid:2) extension to an oligonucleotide carry- bridization of a 3 (cid:2) termin us. The ribos witch con- ing a Cy5 acceptor at its 3 struct has a Cy3 donor attached internally within the loop region such that FRET efficiency, E FRET , between the two fluorophores can be used to distinguish between conforma- tions. We anticipated two major conformations: the open state with a stem-loop structure previously determined by in-line probing and the closed state with a H-type pseudo- knot (Figure 1 A, ( 31 )). The closed conformation likely has a global conformation similar to the crystal structure of the liganded riboswitch that showed the 8-bp extended P1 he- lix and 5-bp pseudoknot (PK) helix coaxially stacked with each other ( 36 ). We previously showed that Cy3 labeling in the loop region does not perturb folding ( 36 ). Single-molecule histograms of E FRET showed two major peaks, likely corresponding to the open conformation (mid- E FRET = 0.4, Figure 1 B) and the closed conformation (high- E FRET = 0.84, Figure 1 B), suggesting the ligand-bound con- formation is adopted e v en without ligand. Lowering mag- nesium concentration reduced the high- E FRET population but a significant percentage ( > 30%) of high- E FRET popula- tion remained e v en in the absence of Mg 2+ (Supplementary Figure S1B), suggesting divalent cations promote the closed conformation, but are not required. Single-molecule time traces of E FRET displayed three types of behavior: (i) constant mid-FRET ( E FRET = 0.4) (ii) constant high-FRET ( E FRET = 0.84) and (iii) dynamic be- havior showing transitions between mid- and high-FRET values (Figure 1 C). The majority (55%, Figure 1 D) of traces showed dynamic behavior, further indicating that e v en in the ligand-free state the closed conformation is sampled. The interconversion kinetics of the d ynamic popula tion was quantified by calculating the average dwell times for high and mid- E FRET states for each molecule and visualized as a log-scale scatter plot. The average dwell times covered a wide range, spanning up to 3 orders of magnitude (Fig- ure 1 E and F). In most cases, the open conformation was longer-li v ed than the closed conformation (Figure 1 G). The dynamic transitioning was a long-lasting characteristic with no clear population interconversions to or from constant mid- or high-FRET states within our experimental window, up to 50 min long (a typical trace shown in Supplemen- tary Figure S2A with zoom-in traces in Supplementary Fig- ure S2B-D, with intermittent 30 s e xposure e v ery 5 min). We attribute this ‘static heterogeneity’ to deep energy wells, and our preliminary investigation at higher temperature still showed static heterogeneity. Ligand binding reshapes the folding energy landscape Next, we measured riboswitch folding in the presence of the cognate ligand SAM. The high-FRET state indeed r epr e- sents the closed conformation because SAM increased the high-FRET population (Figure 2 A). SAM concentrations we used in our study are similar to the physiological con- centration in E. coli, ranging from 28 (cid:2)M to 228 (cid:2)M ( 20 ). The fraction of molecules in the high-FRET population vs ligand concentration could be fitted using a simple two- state binding isotherm, yielding a dissociation constant ( K d ) of 10 (cid:2)M, similar to those measured in bulk solution us- ing isothermal calorimetry ( 36 ). The fraction of molecules in the constant high-FRET species increased from 0.16 to 0.43, and this increase appeared to occur at the expense of the dynamic species while the population of the constant mid-FRET species remained unchanged upon ligand bind- ing (Figure 2 B). This suggests that the molecules already in dynamic exchange with the closed conformation were mor e r eadily locked into the closed conformation via lig- and binding, while the constant mid-FRET population may be trapped in a misfolded state that is not easily rescued by ligand binding. Indeed, flow experiments demonstrated that ∼43% of dynamic species (37 of 86) showed clear lock- ing into the closed conformation after addition of 1 mM SAM (Figure 2 C). Notably, a significant fraction (38%) of molecules still exhibited dynamic transitioning e v en when Nucleic Acids Research, 2023, Vol. 51, No. 17 8961 Figure 1. Studies of ligand-free conformations of the SAM / SAH riboswitch by single-molecule FRET. ( A ) A scheme showing the probable folding of SAM / SAH riboswitch RNA. An 18 nt DNA molecule with a 3 (cid:2) Cy5 acceptor (red circle) was attached via its biotinylated 5 (cid:2) terminus to a quartz slide. Cy3 donor (green circle) was attached to the bulged nucleotide in the PK helix of the riboswitch, and an 18 nt 3 (cid:2) DNA extension complementary to the surface-attached DNA allowed the riboswitch to be tethered to the slide. If the pseudoknot helix is not formed the fluorophores should be separated (the open conformation with mid FRET efficiency) whereas in the folded structure the fluorophores should be much closer (the closed conformation with high FRET efficiency). ( B ) Distribution of FRET efficiencies ( E FRET ) for SAM / SAH riboswitch molecules corresponding to the open and closed conformations. ( C ) Char acteristic tr aces of E FRET as a function of time r ecorded. Thr ee r epr esentati v e tr aces are shown, illustr ati v e of constant high FRET (top), constant mid FRET (middle) and dynamic molecules (bottom) undergoing transitions between states of high and middle E FRET . ( D ) Histograms showing the relati v e fraction of constant high FRET (black), constant middle FRET (dark gray) and dynamic molecules (light grey with line). ( E ) Among traces showing dynamic transitioning, two characteristic traces are shown, indicating the kinetics of transitioning is di v erse and heterogenous. (F, G) Such di v erse tr ansitioning kinetics is then fitted into two states tr ansitioning by ebFRET. Indi vidual molecule av erage dwell time is plotted into log- scale scatter plot ( F ), indicating transitioning heterogeneity. And the high FRET and middle FRET probability within the d ynamic popula tion is plotted individually ( G ). excess ligand was added. However, the average high-FRET dwell time became significantly longer upon ligand addition (Figure 2 D). A schematic model is presented in Figure 2 F, which illustra tes tha t in the absence of ligand, the riboswitch is in a dynamic equilibrium between folded and unfolded sta tes, and tha t the addition of a ligand results in a shift to- wards the closed conformation. Structural perturbations provide insights into the folding en- ergy landscape We ne xt e xamined alterations in the conformational land- scape caused by mutations designed to impact the local structural stability. As shown in Figure 1 A and Figure 3 A, the ligand-bound riboswitch adopts H-type pseudoknot structure with three stabilizing features: ( i ) P1x: the exten- sion to helix P1, comprising one W-C base pair and two non-W–C pairs, ( i i) PK: the pseudoknot helix, involving the Shine–Dalgarno sequence and ( i ii) T: a triple base interac- tion (G47:C16–G16) that is part of the PK helix (Figure 3 A). These structural features are abbreviated here as P1x, PK and T, respecti v ely. To perturb the closed conformation, we designed four dif- ferent mutants, named according to the location of muta- tion: P1x C26Z, P1x A14P, PK C18A / G49U / C50U, and T G16P. For P1x mutants, the original base pairing was al- tered by introducing a modified nucleotide: zebularine (Z: cytosine with N4 removed) or purine (P: adenine with N6 r emoved) (Figur e 3 A). For mutation of the PK helix, two original CG base pairings were replaced with weaker pair- ings: AU and GU. For mutation of the base triple T G16P, G16 was replaced by purine, disrupting the interaction with the Hoogsteen edge of G47. In choosing the muta- tion sites, we avoided altering nucleotides that interact di- rectly with the ligand to minimize disruption of the binding site. Additionally, the number of hydrogen bonds removed was kept to a minimum. The positions of these sequence variations can be found in Supplementary Figure S3, and the sequences of the mutants are listed in Supplementary Table S1. All mutants exhibited an increase in the high FRET population with increasing ligand concentrations, show- ing that the mutations did not eliminate the ligand’s abil- ity to stimulate riboswitch folding (Supplementary Figure S4A–D). The fraction of the high FRET state versus lig- and concentration could be fitted using a two-state bind- ing isotherm, yielding K d values (Table 1 ). Mutants that af- fect the PK helix stability (PK and T m utants) greatl y re- duced binding affinity: K d (PK C18A / G49U / C50U) > 1 mM; K d (T G16P) = 607 (cid:2)M. In contrast, muta tions a t the 8962 Nucleic Acids Research, 2023, Vol. 51, No. 17 Figure 2. Studies of ligand-induced conformations of the SAM / SAH riboswitch by single-molecule FRET. ( A ) Distribution of FRET efficiencies ( E FRET ) as a function of SAM and ( B ) the histograms showing the relati v e fraction of constant high FRET (black), constant middle FRET (dark gray) and dynamic molecules (light grey with line) in the presence of 1 mM SAM. ( C ) Two typical trajectories of riboswitches showed populations converted from dynamics to constant high FRET while SAM was flowed into the reaction chamber at 20 s, corresponding to the change of the relati v e populations in the presence of SAM. ( D, E ) Among molecules remained tr ansitioning, tr ansitioning kinetics was then fitted into two states by ebFRET. Individual molecule average dwell time is plotted into log-scale scatter plot ( D ). And the high FRET and middle FRET probability within the dynamic population is plotted individually ( E ). ( F ) A scheme showing in the presence of ligand, conformations are shifted toward the closed conformations. P1x region had milder effects: K d (P1x C26Z: 41 (cid:2)M; P1x A14P: 61 (cid:2)M). Ne xt, we e xamined the impact of mutations on ligand- free folding dynamics and ligand responsi v eness. In the absence of ligand, all mutants displayed peaks at similar FRET values as the wild type, indica ting tha t open and closed conformations themselves were not significantly al- tered, but their relati v e populations changed (Figure 3 B and C). For example, the destabilization of the closed con- formation in the PK C18A / G49U / C50U mutant led to a near-complete depletion of the closed conformations (Fig- ure 3 D middle panel). The other three mutants showed a more modest decrease in high FRET peak, by < 3% for P1x C26Z, 20% for P1x A14P and 22% for T G16P (Figure 3 B and C). Considering both the binding affinity and the ligand-free conformations, our findings provide evidence for the notion that greater disruptions to the original ligand- free conformations result in greater reduction in binding affinity. Upon ligand introduction, the general behavior observed in the wild-type riboswitch was observed for all mutants, but the relati v e populations and their ligand-induced changes wer e mutation-dependent (Figur e 3 D and Supplementary Figur e S5). Compar ed to mutants targeting the extended P1 stem, mutants that were specifically designed to re- duce the PK helix stability (PK C18A / G49U / C50U and T G16P) exhibited larger changes in conformation, cor- r esponding to r educed binding affinities. As an example, the PK C18A / G49U / C50U mutant had the constant high FRET population almost depleted, replaced by the domi- nant constant mid FRET populations (Figure 3 E). Addi- tionally, in the presence of the ligand, the dynamic popula- tion became more prevalent at the expense of the constant mid FRET population, while the constant high FRET pop- ulation still remained nearly depleted (Figure 3 E). We also tested another mutant P1x A14C / C26U, which stabilizes the extended P1 helix by introducing extra hy- drogen bond (P1x A14C, C26U) by replacing two original non-WC base pairs ( cis sugar-Hoogsteen A13:C26, trans Hoogsteen-sugar A14:G25) with WC base pairs (A13:U26, C14:G25). Despite the introduction of extra hydrogen bonds, the mutants showed a reduction in closed confor- mation (Figure 3 C bottom panel). In addition, the dynamic species increased in population accompanied by a loss of the constant high FRET species (Figure 3 D). The substitution of the WC base pairs may have disturbed the original stack- ing geometry of the three extended P1 base pairs, resulting in less stable PK conformation and affecting the base pair C15:G24 that interacts with the ligand, and thus leading to a reduced binding affinity ( K d = 593 (cid:2)M). These findings highlight the complex nature of riboswitch folding and how small, localized changes can alter the overall folding equi- librium and responsi v eness to ligands, ultimately impacting binding affinity. Ribosome accessibility assay in the absence of ligand We next explored the riboswitch function of blocking trans- la tional initia tion in a ligand dependent manner by mim- (cid:2) region of the riboswitch icking ribosome binding. The 3 Nucleic Acids Research, 2023, Vol. 51, No. 17 8963 Figure 3. Studies of muta tions a t local structures affect conformations and ligand responsi v eness. ( A ) The ligand-bound riboswitch previously re v ealed by X-r ay crystallogr aphy ( 36 ) adopts H-type pseudoknot structure with three stabilizing features: ( i ) P1x: the extension to helix P1, comprising one W-C base pair and two non-W-C pairs, ( i i) PK: the pseudoknot helix, involving the Shine–Dalgarno sequence and ( i ii) a triple base interaction (G47:C16-G16). (B, C) Distribution of FRET efficiencies ( E FRET ) of the ligand-free conformations of, ( B ) wild-type, P1x C26Z and P1x A14P mutants; ( C ) T G16P, PK C18A / G49U / C50U and P1x A14C / C26U m utants. All m utations shared similar folding behaviors of constant high FRET, constant middle FRET, and dynamics. ( D , E ) The histograms showing the relati v e fraction of constant high FRET (black), constant middle FRET (dark gray) and dynamic molecules (light grey) of wild-type and all mutants (D). Among all mutants, PK C18A / G49U / C50U showed the most different populations both in the absence or in the presence of ligands (E). Table 1. Binding affinity of SAM Mutation Wild-typed P1x C26Z P1x A14P T G16P PK C18A, G49U, C50U P1xA14C, C26U Binding affinity ( K d ) in (cid:2)M 11 42 61 607 > 1000 593 contains Shine–Dalgarno (S–D) site to which the ribosome binds to initiate translation. To fully cover the S–D sequence and form stable binding, we used a 9 nt oligonucleotide complementary to the S–D region in assessing the accessi- bility of the transla tion initia tion site (Figure 4 A). Binding of this oligonucleotide was easily monitored using our sin- gle molecule experiment. In the absence of ligands, adding the oligonucleotides de- creased the two major FRET populations ( E FRET = 0.4 & 0.84) and created a new population with an E FRET of a pproximatel y 0.22 (Figure 4 B). To identify a condition that the oligonucleotides are saturated to the pre-exposed S–D region, we conducted experiments with varying con- centrations of oligonucleotides, and no discernible changes in conformation were observed above 500 nM (Supple- mentary Figure S6A). Consequently, all subsequent exper- iments were performed with the sa tura ting oligonucleotide concentration of 500 nM. We attribute the low-FRET state to the lengthening of 9 nt S–D region upon ribos witch-oligon ucleotide complex formation. A similar experiment using a 15 nt oligonu- cleotide showed the low-FRET peak at a slightly lower value ( ∼0.18), consistent with the longer helix that would be formed (Figure 4 C and Supplementary Figure S6B). Duplex formation was very stable and persisted even after w ashing aw ay fr ee oligonucleotides (Supplementary Figur e S6C), in contrast to transient duplex formation observed us- ing a shorter 7 nt long ribosome mimic for 7-aminomethyl- 7-deazaguanine (preQ 1 )-sensing riboswitch ( 47 ). We further examined the oligonucleotide binding reac- tion in real time by flowing in the oligonucleotide dur- ing observation. Most molecules remained unchanged or showed photobleaching because oligo binding generally took longer than our observation window of ∼180 s. Among molecules showing evidence of oligonucleotide binding, d ynamic fluctua tions between 0.4 and 0.84 FRET states were observed before they were locked into the low- FRET state (Figure 4 D). In addition, most low-FRET states (75% and 91% for the 9-nt oligo and 15-nt oligo, respecti v ely) were reached from the mid-FRET, open con- formation (Figure 4 D), indicating that the S–D region be- comes accessible in the open conformation prior to oligonu- cleotide binding. 8964 Nucleic Acids Research, 2023, Vol. 51, No. 17 Figure 4. Pseudo-functional studies for assessing the accessibility of the translation initiation site. ( A ) A scheme showing 9 or 15-nt complementary oligonu- cleotides bind to the open conformation of the SAM / SAH riboswitch. ( B ) Distribution of E FRET before and after addition of sa tura ted concentra tion ( = 500 nM) with various time points. ( C ) Distribution of E FRET comparing the oligo-free (top), 9-nt (middle) and 15-nt (bottom) oligonucleotide-bound conformations. ( D ) Typical trajectories of riboswitches showed population converted from the mid-FRET to low FRET while oligonucleotides were flowed into the reaction chamber at 12 s, corresponding to the generated conformations observed in distribution of E FRET . ( E ) Distribution of E FRET after si- multaneous addition of 9-nt oligonucleotides ( = 500 nM) and SAM ( = 1 (cid:2)M) at various time points, both additions are a t sa tura ted concentra tion. ( F ) Distribution of E FRET showing the ligand responsi v eness after the riboswitches were pre-bound by oligonucleotides. Ligand-induced conformational change outcompetes ribo- some mimic binding We ne xt e xamined the effect of the SAM ligand on the S– D region accessibility and riboswitch folding by simulta- neously adding the ligand SAM and oligonucleotide ribo- some mimics. Within min, the high-FRET population in- creased at the expense of the mid-FRET population. Only a small fraction of molecules ( < 20%) went to the low-FRET conformation corresponding to the ribosome-mimic bound state (Figure 4 E). Therefore, under our experimental con- ditions (1 mM SAM and 500 nM oligonucleotide), ligand binding outcompetes 9-nt oligonucleotide binding at early timepoints. The riboswitch remained in the high-FRET state for the whole observation window for the 9-nt oligonucleotide, up to 60 min. Howe v er, for the 15-nt oligonucleotides, ∼40% of high-FRET conformation was lost by 10 min and by 1 h, the low-FRET conformation became dominant (Supple- mentary Figure S6D), suggesting that the ligand bound ri- boswitch can still undergo occasional visits to the open con- formation. The additional foothold of the 15-nt oligo en- ables capturing the transiently exposed S–D site. At least for the longer ribosome-mimic, the equilibrium favors the oligo-bound ‘gene-on’ state w hereas earl y in the process, ligand binding can trap the molecule in the closed con- formation, momentarily blocking access to the ribosome. Howe v er, it is also possible that the extra f oothold f or the long oligo may facilitate pseudoknot unwinding by par- tially hybridizing to an unpaired region followed by strand displacement. To test if the oligo-bound riboswitch remains responsi v e to ligand, we added the ligand after pre-incubation with oligonucleotide. Only ∼25% of 9-nt oligo-bound structures converted to the ligand-bound ( E FRET = 0.84 state), and a negligible fraction of the 15-nt oligo-bound riboswitch was responsi v e to the ligand (Figure 4 F and Supplemen- tary Figure S6E). Therefore, to function as a transla- tional riboswitch, the decision must be made before the ar- rival of the ribosome, assuming ribosome binding happens only once. In the more realistic case of multiple ribosome molecules arriving and initiating translation in succession, the riboswitch activity can be gradational. Vectorial folding assay disfavors ligand binding compared to the pr e-f olded riboswitch Because RNA folds co-transcriptionally, riboswitch func- tion should be examined in the context of ongoing tran- scription ( 48 ). Se v eral different methods of mimicking co- transcriptional riboswitch folding and function are avail- able ( 44 , 45 , 49–55 ). Here we used the vectorial folding assay w here a DN A helicase is used to mimic co-transcriptional RNA folding ( 44 , 45 ). The riboswitch was hybridized with a complementary DNA oligonucleotide to form an RNA- (cid:2) overhang at the DNA ter- DNA heteroduplex with a 3 minus. The same Cy3–Cy5 FRET pair as in our pre- vious experiments was used to determine the riboswitch Nucleic Acids Research, 2023, Vol. 51, No. 17 8965 Figur e 5. Vectoriall y f olded assa ys f or mimicking co-transcriptional f olding. ( A ) A scheme showing the engineered superhelicase Rep-X was preincubated and initia ted a t designa ted time for unwinding the heteroduplex. The riboswitch was hybridized with a complementary DNA oligonucleotide to form an RN A-DN A heteroduplex with a 3 (cid:2) overhang at the DNA terminus. The same Cy3–Cy5 FRET pair was used to determine the riboswitch conformations. ( B ) Distribution of E FRET before introducing ATP, corresponding to the heteroduplex conformation. ( C ) Distribution of E FRET after vectorially folding, cor- responding to the conformations released from the heteroduplex. ( D ) A typical trajectory showing a heteroduplex is unwound and folded into the constant high FRET conformation, where the ATP is flowed in at 12 s. ( E ) Histograms of relati v e populations in the presence of various SAM concentrations. conformation (Figure 5 A). A highly processi v e, engineered DNA helicase, Rep-X ( 46 ), was used to unwind the het- eroduplex unidirectionally by translocating on the DNA (cid:2) direction, to release the RNA strand (cid:2) to 5 strand in the 3 (cid:2) direction of transcription and at (cid:2) to 3 progressi v ely in the 5 the speed of transcription, about ∼60 nt per second ( 44 , 45 ). Upon Rep-X addition without ATP, we observed low FRET efficiency ( E FRET = 0.2) because the fluorophores remain separated by the heteroduplex (Figure 5 B). After ATP addition, two new populations centered at 0.4 and 0.84 emerged, corresponding to the open and closed con- formations, respecti v ely (Figure 5 C). A representati v e v ec- torial folding trace shows two features (Figure 5 D). First, Cy3 intensity shows a transient increase due to protein- induced fluorescence enhancement ( 56 , 57 ), signifying Rep- X approaching the Cy3 fluorophore on the RNA strand. Second, the heteroduplex unwinds and riboswitch fold- ing begins. We classified the riboswitch folding behavior into four types (Supplementary Figure S7A) : (i) molecules that transitioned from the heteroduplex state to the stable closed conformation without any detectable intermediate, (ii) molecules that transitioned from the heteroduplex state to the stable open conformation, (iii) molecules that transi- tioned from the heteroduplex state to one undergoing fluc- tuations between the open and closed conformations and (iv) molecules that transitioned from the heteroduplex to the stable closed conformation after first fluctuating be- tween open and closed states. These distinct populations are in agreement with the observed conformations for pre- folded riboswitches except for the type (iv), likely because this behavior is observable only on the path to reach fold- ing equilibrium. Addition of ligand during vectorial folding changed the relati v e populations of the four types of folding behavior. Type I, direct transition to stable high-FRET state, became more popula ted a t higher ligand concentra tions (Figure 5 E) and its fraction vs ligand concentration could be well fit- ted using a two-state binding isotherm (Supplementary Fig- ure S7B), yielding an apparent K d value of 108 (cid:2)M. This apparent K d is an order of magnitude higher than the K d value of 10 (cid:2)M we observed for pre-folded RNA, suggest- ing ligand-responsi v e conforma tion is not immedia tely ob- tained during co-transcriptional folding. We hypothesized that the increase in K d is due to insufficient time for the nascent riboswitch to reach the stead y-sta te conforma tions. Indeed, as shown in Supplementary Figure S7C, the con- formational analyses of time points after vectorial folding e xhibited noticeab le differences. Specifically, the conforma- tions observed at 5 seconds post-folding displayed a lower fraction of high-FRET population and decreased respon- si v eness to ligand. These results imply that the nascent ri- boswitch necessitates more than a few seconds to attain a 8966 Nucleic Acids Research, 2023, Vol. 51, No. 17 Figure 6. Pseudo-functional studies for assessing the accessibility of the translation initiation site during vectorially folding. ( A ) A scheme showing SAM, oligonucleotides, and ATP are flowed in sim ultaneousl y for simulating competition over mutually e xclusi v e conformations during co-transcriptional fold- ing. The oligonucleotide-bound state is termed ‘ON’ sta te, indica ting transla tion can be initia ted, whereas the ligand-bound sta te is termed ‘OFF’ state, indicating the Shine–Dalgarno site is blocked. ( B ) Distribution of E FRET for pre-folded riboswitches under simultaneous addition of 9-nt oligonucleotides and various concentrations of SAM. Similar competitions between oligonucleotides and ligands were carried out while the riboswitches are vectorially folded and distribution of E FRET with various SAM concentrations is shown in ( D ). Similar competition experiments were carried out with longer 15-nt oligonucleotides, f or pre-f olded competition, ( C ) distribution of E FRET with various SAM concentrations; for vectorially folded competition ( E ) distribu- tion of E FRET with various SAM concentrations. stead y sta te conforma tional distribution, ther eby r educing its ligand-binding affinity during transcription. Vectorial folding favors ribosome mimic binding In the ribosome accessibility assay on pre-folded ri- boswitches, we found that ligand binding is kinetically fa- vored over ribosome mimic binding. To test if this result holds for vectorial folding, we included sa tura ting concen- tration of-9 nt or 15-nt oligonucleotides during vectorial folding (Figure 6 A). In the absence of ligand, the closed conformation was rarely observed, likely because the S–D r egion r e v ealed thr ough heter oduplex unwinding is bound by the ribosome mimic before the aptamer can fold into the closed form (top histogram of Figure 6 D and E). This find- ing is also in line with our observa tion tha t it takes min for the nascent riboswitch to attain a stead y-sta te confor- mational distribution (as demonstrated in Supplementary Figure S7C). In the presence of ligand, the closed conformation was obtained in a ligand-concentration dependent manner. The efficacy of the ligand in converting the riboswitch to its closed conforma tion sta te was diminished in the vectorial folding condition compared to the pre-folded condition for both ribosome mimics (Figure 6 B–E and Supplementary Figure S9). The nascent riboswitch likely first adopts the open conformation, which facilitates ribosome mimic bind- ing, and reduces the m utuall y e xclusi v e ligand bound con- formation. DISCUSSION We propose a model describing the folding scheme and its energy landscape based on our findings of multiple populations of static folds, open or closed and dynamic switching, and the highly heterogenous switching rates. The FRET values (0.4 and 0.84) of the switching molecules wer e indiffer entiable from those with sta tic conforma tions, suggesting there is significant structural resemblance. We were surprised that the majority ( ∼55%) of ligand-free ri- boswitches showed dynamic switching between the closed and open conformations. Most of the dynamically switch- ing molecules are responsi v e to ligand, either by population conversion to the static closed conformation or rate alter- ations. The observed heterogeneity is likely to be a prop- erty inherent to the riboswitch because our constructs with their modifications for surface tethering and fluorescence imaging showed comparable binding affinity to what was determined from unmodified RNA in bulk solution. Fur- thermore, all fiv e single-site mutants we tested show simi- larl y hetero geneous behavior with onl y their relati v e popu- lations and kinetics changed. Mutants examined in this study showed that not only the populations of static and dynamic populations were strongly affected, but the rates of switching between con- formations changed (Supplementary Figure S8). We specu- la te tha t an y incomplete base pair f ormation of the extended P1 stem (E-P1) or PK helix may introduce metastable conformations, leading to heterogeneous folding / unfolding rates for this riboswitch, and potentially for other func- tional RNAs that also contain the H-type pseudoknot ( 58–61 ). Such heterogeneity, if present in vivo , may buffer the riboswitch activity against a wide range of ligand concentrations. Our findings are most consistent with the previously pro- posed hybrid model combining conformation selection & induced-fit ( 10 ): whereas all conformations are sampled in the absence of ligand (conformation selection), ligand addi- tion repopulates the population ensemble by imparting fur- ther stability to the ligand-bound state (induced-fit). A pre- vious SAM-II riboswitch study reported that transient con- formational excursions occur in the absence of ligand, sug- gesting conformational sampling ( 10 ). Howe v er, they could not determine if those transient conformations were respon- si v e to ligands or how folding and ligand binding are pro- moted through specific structural motifs. Relevant to our evaluation of the riboswitch’s accessibil- ity for ribosome mimics, a previous study probed the folding of the 7-aminomethyl-7-deazaguanine-sensing riboswitch using a 7 nt long fluorescently labeled oligonucleotides as transla tional initia tion mimic ( 46 ). They observed bursts of probe binding and showed that ligand addition reduces burst duration and extends the intervals between bursts. Howe v er, the use of fluorescently labeled probes limited their analysis to sub- K d concentrations. By employing un- labeled oligonucleotides, we were able to mimic translation initiation under conditions of saturating ribosome mimic so that the exposure of the binding site is rate-limiting and show that the nascent folds adopted have yet to reach an equilibrium, thus leading to a reduced ligand binding affinity. In the vectorial f olding assa y, we observed a decrease in ligand binding affinity (Figure 6 and Supplementary Fig- ur e S9), r esulting in a r eduction in the effecti v eness of lig- and binding when competing with a ribosome mimic. These differ ences between pr e-f olded and vectorially f olded ri- boswitches suggest that the timing of regulatory decision is critical to the effecti v eness of the riboswitch and may ex- plain the r equir ement f or higher ligand concentration f or ef- fecti v e regulatory control in vivo ( 62 ) . It is possible that there ar e differ ent modes of r egulating accessibility, and the tim- ing of transcription and translation coupling. For tighter regulation, riboswitch needs to reach equilibrium first, thus transcription needs to be carried out in advance of trans- lation. Howe v er, when regulation needs not to be tight, the transcription and translation can happen simultaneously. In conclusion, our studies on this small SAM / SAH ri- boswitch provide valuable insights into the complexities of the folding landscape, including individual folding hetero- geneity and the role of RNA folding kinetics. Furthermore, our findings have implications for the translational control governed by the riboswitch, highlighting the critical influ- ence of folding equilibrium on the efficacy of regulatory decisions. DA T A A V AILABILITY Analyses and data acquisition codes are upload on lab GitHub account and archi v ed in Zenodo with the following doi. Additionally, raw data that support our findings have Nucleic Acids Research, 2023, Vol. 51, No. 17 8967 been uploaded and archi v ed in Zenodo, corresponding to each individual figure. GitHub: https://github.com/Ha-SingleMoleculeLab Analyses , data acquisition codes , and raw data are archi v ed in Zenodo: Raw data analysis DOI: 10.5281 / zenodo.4925617 Data acquisition DOI: 10.5281 / zenodo.4925630 Raw data DOI: 10.5281 / zenodo.8088172 SUPPLEMENT ARY DA T A Supplementary Data are available at NAR Online. ACKNOWLEDGEMENTS We thank Prof. Hui-Ting Lee, Dr Olivia Yang, Dr Boyang Hua, and the members of the Ha laboratory and Sua My- ong laboratory for their input and support. All the authors w ould lik e to expr ess their gratitude to the funding sour ce for their generous support. FUNDING US National Institutes of Health [R35 GM 122569 to T.H. and F32 GM 139268 to L.R.G.]; Cancer Research UK [Progr am gr ant A18604]; EPSRC [EP / X01567X / 1 to D.M.J.L.]; National Natural Science Foundation of China [32171191 to L.H.]; Guangdong Science and Technol- ogy Department [2022A1515010328, 2020B1212060018, 2020B1212030004 to L.H.]; T.H. is an investigator of the Howard Hughes Medical Institute. Funding for open ac- cess charge: U.S. Department of Health and Human Ser- vices [R35 GM 122569]. Conflict of interest statement. None declared. REFERENCES 1. Mccown,P.J., Corbino,K.A., Stav,S., Sherlock,M.E. and Breaker,R.R. (2017) Riboswitch di v ersity and distribution. RNA , 23 , 995–1011. 2. Serganov,A. and Nudler,E. (2013) A decade of riboswitches. Cell , 152 , 17–24. 3. Sherwood,A.V. and Henkin,T.M. (2016) Riboswitch-mediated gene regulation: novel RNA architectures dictate gene expression responses. Annu. Rev. Microbiol. , 70 , 361–374. 4. Speed,M.C., Burkhart,B.W., Picking,J.W. and Santangelo,J. (2018) An Archaeal Fluoride-Responsi v e Riboswitch Provides an Inducible Expression System for Hyperthermophiles. Appl. Environ. Microbiol. , 84 , e02306-17. 5. Wachter,A., Tunc-Ozdemir,M., Grove,B.C., Green,P.J., Shintani,D.K. and Breaker,R.R. (2007) Riboswitch control of gene expression in plants by splicing and alternative 3 (cid:2) end processing of mRNAs. Plant Cell , 19 , 3437–3450. 6. Moldo van,M.A., Petro va,S.A. and Gelfand,M.S. (2018) Comparati v e genomic analysis of fungal TPP-riboswitches. Fungal Genet. Biol. , 114 , 34–41. 7. Li,S. and Breaker,R.R. (2013) Eukaryotic TPP riboswitch regulation of alternati v e splicing involving long-distance base pairing. Nucleic Acids Res. , 41 , 3022–3031. 8. Andreasson,J.O.L., Savinov,A., Block,S.M. and Greenleaf,W.J. (2020) Comprehensi v e sequence-to-function mapping of cofactor-dependent RN A catal ysis in the glmS ribozyme. Nat. Commun. , 11 , 1663. 9. Valeri,J.A., Collins,K.M., Ramesh,P., Alcantar,M.A., Lepe,B.A., Lu,T.K. and Camacho,D.M. (2020) Sequence-to-function deep learning frame wor ks for engineer ed ribor egulators. Nat. Commun. , 11 , 5058. 8968 Nucleic Acids Research, 2023, Vol. 51, No. 17 10. Haller,A., Rieder,U., Aigner,M., Blanchard,S.C. and Micura,R. (2011) Conformational capture of the SAM-II riboswitch. Nat. Chem. Biol. , 7 , 393–400. 11. Savinov,A., Perez,C.F. and Block,S.M. (2014) Single-molecule studies of riboswitch folding. Biochim. Biophys. Acta - Gene Regul. Mech. , 1839 , 1030–1045. 12. Holmstrom,E.D., Polaski,J.T., Batey,R.T. and Nesbitt,D.J. (2014) Single-molecule conformational dynamics of a biolo gicall y functional hydro x ocobalamin riboswitch. J. Am. Chem. Soc. , 136 , 16832–16843. 13. Panchal,V. and Brenk,R. (2021) Riboswitches as drug targets for antibiotics. Antibiotics , 10 , 45. 14. Hallberg,Z.F., Su,Y., Kitto,R.Z. and Hammond,M.C. (2017) Engineering and in vivo applications of riboswitches. Annu. Rev. Biochem. , 86 , 515–539. 15. Loenen,W.A.M. (2006) S-Adenosylmethionine: jack of all trades and master of e v erything? Biochem. Soc. T r ans . , 34 , 330–333. 16. Kredich,N.M. and Hershfield,M.S. (1979) S-adenosylhomocysteine toxicity in normal and adenosine kinase-deficient lymphoblasts of human origin. Proc. Natl. Acad. Sci. U.S. A. , 76 , 2450–2454. 17. Schubert,H.L., Blumenthal,R.M. and Cheng,X. (2003) Many paths to methyltr ansfer : a chronicle of convergence. T r ends Biochem. Sci. , 28 , 329–335. 18. Roje,S. (2006) S-Adenosyl-l-methionine: beyond the uni v ersal methyl group donor. Phytoc hemistr y , 67 , 1686–1698. 19. Cantoni,G.L. (1975) Biological methylation: selected aspects. Annu. Rev. Biochem. , 44 , 435–451. 20. Posnick,L.M. and Samson,L.D. (1999) Influence of S-adenosylmethionine pool size on spontaneous mutation, dam methylation, and cell growth of Escherichia coli. J. Bacteriol. , 181 , 6756–6762. 21. Price,I.R., Grigg,J.C. and Ke,A. (2014) Common themes and differences in SAM recognition among SAM riboswitches. Biochim. Biophys. Acta - Gene Regul. Mech. , 1839 , 931–938. 34. Gilbert,S.D., Rambo,R.P., Van Tyne,D. and Batey,R.T. (2008) Structure of the SAM-II riboswitch bound to S-adenosylmethionine. Nat. Struct. Mol. Biol. , 15 , 177–182. 35. Weickhmann,A.K., Keller,H., Wurm,J.P., Strebitzer,E., Juen,M.A., Kremser,J., Weinberg,Z., Kreutz,C., Duchardt-Ferner,E. and W ¨ohnert,J. (2019) The structure of the SAM / SAH-binding riboswitch. Nucleic Acids Res. , 47 , 2654–2665. 36. Huang,L., Liao,T.W., Wang,J., Ha,T. and Lilley,D.M.J. (2020) Crystal structure and ligand-induced folding of the SAM / SAH riboswitch. Nucleic Acids Res. , 48 , 7545–7556. 37. Ha,T ., Enderle,T ., Ogletree,D .F., Chemla,D .S., Selvin,P.R. and Weiss,S. (1996) Probing the interaction between two single molecules: fluor escence r esonance energy transfer between a single donor and a single acceptor. Proc. Natl. Acad. Sci. U.S.A. , 93 , 6264–6268. 38. Beaucage,S.L. and Caruthers,M.H. (1981) Deoxynucleoside phosphoramidites-A new class of key intermediates for deoxypolynucleotide synthesis. Tetr ahedr on Lett. , 22 , 1859–1862. 39. Wilson,T.J., Zhao,Z.Y., Maxwell,K., Kontogiannis,L. and Lilley,D.M.J. (2001) Importance of specific nucleotides in the folding of the natural form of the hairpin ribozyme. Bioc hemistr y , 40 , 2291–2302. 40. Hakimelahi,G.H., Proba,Z.A. and Ogilvie,K.K. (1981) High yield selecti v e 3 (cid:2) -silylation of ribonucleosides. Tetrahedron Lett. , 22 , 5243–5246. 41. Perreault,J., Wutt,T., Cousineau,B., Ogilviett,K.K. and Cedergren,R. (1990) Mixed deoxyribo- and ribo-oligonucleotides with catalytic activity. Nature , 344 , 1988–1990. 42. Roy,R., Hohng,S. and Ha,T. (2008) A practical guide to single-molecule FRET. Nat. Methods , 5 , 507–516. 43. Van De Meent,J.W., Bronson,J.E., Wiggins,C.H. and Gonzalez,R.L. (2014) Empirical bayes methods enable advanced population-level analyses of single-molecule FRET experiments. Biophys. J. , 106 , 1327–1337. 22. Huang,L. and Lilley,D.M.J. (2018) Structure and ligand binding of 44. Hua,B., Panja,S., Wang,Y., Woodson,S.A. and Ha,T. (2018) the SAM-V riboswitch. Nucleic Acids Res. , 46 , 6869–6879. 23. Fuchs,R.T., Grundy,F.J. and Henkin,T.M. (2006) The SMK box is a Mimicking co-transcriptional RNA folding using a superhelicase. J. Am. Chem. Soc. , 140 , 10067–10070. new SAM-binding RNA for translational regulation of SAM synthetase. Nat. Struct. Mol. Biol. , 13 , 226–233. 24. Lu,C., Smith,A.M., Fuchs,R.T., Ding,F., Rajashankar,K., Henkin,T.M. and Ke,A. (2008) Crystal structures of the SAM-III / SMK riboswitch re v eal the SAM-dependent translation inhibition mechanism. Nat. Struct. Mol. Biol. , 15 , 1076–1083. 25. Sun,A., Gasser,C., Li,F., Chen,H., Mair,S., Krasheninina,O., Micura,R. and Ren,A. (2019) SAM-VI riboswitch structure and signature for ligand discrimination. Nat. Commun. , 10 , 5728. 26. Wang,J.X., Lee,E.R., Morales,D.R., Lim,J. and Breaker,R.R. (2008) Riboswitches that sense S-adenosylhomocysteine and activate genes involved in coenzyme recycling. Mol. Cell , 29 , 691–702. 27. Montange,R.K. and Batey,R.T. (2006) Structure of the S-adenosylmethionine riboswitch regulatory mRNA element. Nature , 441 , 1172–1175. 28. Mirihana Arachchilage,G., Sherlock,M.E., Weinberg,Z. and Breaker,R.R. (2018) SAM-VI RNAs selecti v ely bind S-adenosylmethionine and exhibit similarities to SAM-III riboswitches. RNA Biol , 15 , 371–378. 29. Winkler,W.C., Nahvi,A., Sudarsan,N., Barrick,J.E. and Breaker,R.R. (2003) An mRNA structure that controls gene expression by binding S-adenosylmethionine. Nat. Struct. Biol. , 10 , 701–707. 30. Weinberg,Z., Regulski,E.E., Hammond,M.C., Barrick,J.E., Yao,Z., Ruzzo,W.L. and Breaker,R.R. (2008) The aptamer core of SAM-IV riboswitches mimics the ligand-binding site of SAM-I riboswitches. RNA , 14 , 822–828. 31. Weinberg,Z., Wang,J.X., Bogue,J., Yang,J., Corbino,K., Moy,R.H. and Breaker,R.R. (2010) Comparati v e genomics re v eals 104 candidate structured RNAs from bacteria, archaea, and their metagenomes. Genome Biol. , 11 , R31. 32. Corbino,K.A., Barrick,J.E., Lim,J., Welz,R., Tucker,B.J., Puskarz,I., Mandal,M., Rudnick,N.D. and Breaker,R.R. (2005) Evidence for a second class of S-adenosylmethionine riboswitches and other regulatory RNA motifs in alpha-proteobacteria. Genome Biol. , 6 , R70. 45. Hua,B., Jones,C.P., Mitra,J., Murray,P.J., Rosenthal,R., Ferr ´e-D’Amar ´e,A.R. and Ha,T. (2020) Real-time monitoring of single ZTP riboswitches re v eals a complex and kinetically controlled decision landscape. Nat. Commun. , 11 , 4531. 46. Arslan,S., Khafizov,R., Thomas,C.D., Chemla,Y.R. and Ha,T. (2015) Engineering of a superhelicase through conformational control. Science , 348 , 344–347. 47. Rinaldi,A.J., Lund,P.E., Blanco,M.R. and Walter,N.G. (2016) The Shine–Dalgarno sequence of riboswitch-regulated single mRNAs shows ligand-dependent accessibility bursts. Nat. Commun. , 7 , 8976. 48. Gong,S., Wang,Y., Wang,Z. and Zhang,W. (2017) Co-transcriptional folding and regulation mechanisms of riboswitches. Molecules , 22 , 1–14. 49. Duss,O., Stepanyuk,G.A., Puglisi,J.D. and Williamson,J.R. (2019) Transient protein-RNA interactions guide nascent ribosomal RNA folding. Cell , 179 , 1357–1369.e16. 50. Rodgers,M.L. and Woodson,S.A. (2019) Transcription increases the cooperativity of ribonucleoprotein assembly. Cell , 179 , 1370–1381. 51. Juette,M.F., Terry,D.S., Wasserman,M.R., Altman,R.B., Zhou,Z., Zhao,H. and Blanchard,S.C. (2016) Single-molecule imaging of non-equilibrium molecular ensembles on the millisecond timescale. Nat. Methods , 13 , 341–344. 52. Uhm,H., Kang,W., Ha,K.S., Kang,C. and Hohng,S. (2017) Single-molecule FRET studies on the cotranscriptional folding of a thiamine pyrophosphate riboswitch. Proc. Natl. Acad. Sci. U.S.A. , 115 , 331–336. 53. Yu,A.M., Gasper,P.M., Cheng,L., Lai,L.B., Kaur,S., Gopalan,V., Chen,A.A. and Lucks,J.B. (2021) Computationally reconstructing cotranscriptional RNA folding from experimental data reveals rearrangement of non-nati v e folding intermediates. Mol. Cell , 81 , 870–883. 54. Strobel,E.J., Watters,K.E., Nedialkov,Y., Artsimovitch,I. and Lucks,J.B. (2017) Distributed biotin-streptavidin transcription roadblocks for mapping cotranscriptional RNA folding. Nucleic Acids Res. , 45 , e109. 33. Poiata,E., Meyer,M.M., Ames,T.D. and Breaker,R.R. (2009) A 55. Chauvier,A., St-Pierre,P., Nadon,J.F., Hien,E.D.M., variant riboswitch aptamer class for S-adenosylmethionine common in marine bacteria. RNA , 15 , 2046–2056. Perez-Gonzalez,C., Eschbach,S.H., Lamontagne,A.M., Carlos Penedo,J. and Lafontaine,D.A. (2021) Monitoring RNA dynamics in Nucleic Acids Research, 2023, Vol. 51, No. 17 8969 nati v e transcriptional complexes. Proc. Natl. Acad. Sci. U.S.A. , 118 , e2106564118. 59. Staple,D.W. and Butcher,S.E. (2005) Pseudoknots: RNA structures with di v erse functions. PLoS Biol. , 3 , 0956–0959. 56. Pettigrew,N.R., Thomas,A.C., Mcmanus,M.A., Paduan,J.D., 60. Shen,L.X. and Tinoco,I. (1995) The structure of an RNA Chavez,F.P., Nielsen,T.G., Desiderio,R.A., Carr,M., Osborn,T.R., Abe,T. et al. (2009) Cytosolic viral sensor RIG-I is a 5 (cid:2) -triphosphate-dependent translocase on double-stranded RNA. Science , 323 , 1070–1074. 57. Hwang,H., Kim,H. and Myong,S. (2011) Protein induced fluorescence enhancement as a single molecule assay with short distance sensitivity. Proc. Natl. Acad. Sci. U.S.A. , 108 , 7414–7418. 58. Egli,M., Minasov,G., Su,L. and Rich,A. (2002) Metal ions and flexibility in a viral RNA pseudoknot at atomic resolution. Proc. Natl. Acad. Sci. U.S.A. , 99 , 4302–4307. pseudoknot that causes efficient frameshifting in mouse mammary tumor virus. J. Mol. Biol. , 247 , 963–978. 61. Wang,Y., Yesselman,J.D., Zhang,Q., Kang,M. and Feigon,J. (2016) Structural conservation in the template / pseudoknot domain of vertebr ate telomer ase RNA from teleost fish to human. Proc. Natl. Acad. Sci. U.S.A. , 113 , E5125–E5134. 62. Neuner,E., Frener,M., Lusser,A. and Micura,R. (2018) Superior cellular activities of azido- over amino-functionalized ligands for engineer ed pr eQ1 riboswitches in E.coli. RNA Biol. , 15 , 1376–1383. C (cid:3) The Author(s) 2023. Published by Oxford University Press on behalf of Nucleic Acids Research. This is an Open Access article distributed under the terms of the Creati v e Commons Attribution License (http: // creati v ecommons.org / licenses / by / 4.0 / ), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.
null
10.1186_s12872-022-02488-x.pdf
ial, or not-for-profit sectors. Availability of data and materials The data that support the findings of this study are available from the National Health Insurance Service in the Republic of Korea, but restrictions apply to the availability of these data, which were used under license for the current study and so are not publicly available. Data are, however, available from the authors Son et al. BMC
Availability of data and materials The data that support the findings of this study are available from the National Health Insurance Service in the Republic of Korea, but restrictions apply to the availability of these data, which were used under license for the current study and so are not publicly available. Data are, however, available from the authors
Son et al. BMC Cardiovascular Disorders (2022) 22:44 https://doi.org/10.1186/s12872-022-02488-x ORIGINAL RESEARCH Open Access Risk of aortic aneurysm and aortic dissection with the use of fluoroquinolones in Korea: a nested case–control study Nayeong Son†, Eunmi Choi†, Soo Youn Chung, Soon Young Han and Bonggi Kim* Abstract Background: Recent studies have raised concern about the association of fluoroquinolones with an increased risk of aortic aneurysm and aortic dissection. We aimed to evaluate such risk in a Korean population. Methods: We conducted a nested case–control study using data from the National Health Insurance Service col- lected from 2013 to 2017 in Korea. The study cohort included patients older than 40 years and excluded patients who had used fluoroquinolones or been diagnosed with aortic aneurysm, aortic dissection, or related diseases 1 year prior to the cohort entry date. We randomly matched four controls in the risk set with each case of aortic aneurysm and aortic dissection (same sex, age, and cohort entry date). We assessed the risk of aortic aneurysm and aortic dissection from fluoroquinolones and adjusted for potential confounders using a conditional logistic regression model. Results: A total of 29,638 aortic aneurysm and aortic dissection patients were identified between 2014 and 2017. The use of fluoroquinolones within a year was associated with a 10% increased risk of aortic aneurysm and aortic dissec- tion (adjusted odds ratio: 1.10, 95% CI 1.07–1.14, p < 0.05) compared with nonusers. The risk was higher in patients who had used fluoroquinolones within 60 days (adjusted odds ratio: 1.53, 95% CI 1.46–1.62, p < 0.05). The risk of aortic aneurysm and aortic dissection positively correlated with the cumulative dose and duration of fluoroquinolone therapy (p < 0.001). Conclusions: Our study provides real-world evidence of the risk of aortic aneurysm and aortic dissection from fluo- roquinolones in Korea. Patients and medical professionals should be aware that fluoroquinolones can increase the risk of aortic aneurysm and aortic dissection, which may be acerbated by high dosage and duration of use. Keywords: Fluoroquinolone, Aortic aneurysm, Aortic dissection, Drug safety, Pharmacovigilance, Adverse effect Background Fluoroquinolones (FQs) are among the most widely used antibiotics in Korea, and their use has consistently increased to account for 9 to 11% of all antibiotic use [1]. Although FQs are powerful antibiotics with a wide antibacterial spectrum [2], they induce degradation of collagen and other structural components of the extra- cellular matrix by stimulating matrix metalloproteinases [3]. The possibility of excessive tissue breakdown by this mechanism has raised concern about the risk of adverse *Correspondence: bgkim@drugsafe.or.kr †Nayeong Son and Eunmi Choi are co-first authors and contributed equally. All the authors take responsibility for all aspects of the reliability and freedom from bias of the data presented and their discussed interpretation Korea Institute of Drug Safety and Risk Management, 6th FL, 30, Burim-ro 169beon-gil, Dongan-gu, Anyang-si, Gyeonggi-do, Republic of Korea © The Author(s) 2022. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/. The Creative Commons Public Domain Dedication waiver (http:// creat iveco mmons. org/ publi cdoma in/ zero/1. 0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data. Son et al. BMC Cardiovascular Disorders (2022) 22:44 Page 2 of 9 reactions, such as aortic aneurysm (AA) and aortic dis- section (AD). In December 2018, the U.S. Food and Drug Adminis- tration warned that FQs can increase the occurrence of rare but serious ruptures or tears in the aorta. The warn- ing included special caution for patients with a history of aneurysms, blockages, or hardening of the arteries, high blood pressure, or genetic conditions such as Mar- fan or Ehlers–Danlos syndrome and instructed patients to inform their health-care professional before starting a fluoroquinolone prescription [4]. Following that warning, the Ministry of Food and Drug Safety in Korea issued a safety letter warning about the potential association of fluoroquinolone use and the risk of AA/AD [5]. Many observational studies have suggested that fluo- roquinolone use could be significantly associated with an increased risk of AA/AD [3, 6–9]. Recently, a systemic review and meta-analysis showed that fluoroquinolone use incurs a risk of developing three collagen-associated diseases, including AA/AD [10]. However, it has not yet been established whether fluoroquinolone use increases the risk of AA/AD in the Korean population. This study aims to evaluate the association between fluoroquinolone use and the risk of AA/AD in the Korean population. Methods Data source Insurance Service We conducted a nested case-control study using National (NHIS)-customized data Health (NHIS-2019-1-024). The NHIS database covers almost 98% of the total population in Korea. It contains patient demographic information such as sex, date of birth, date of death, and medical treatment records, including details of disease and prescriptions [11]. The authors declare no conflicts of interest with NHIS. Study population The study population comprised all patients aged 40 to 99 years 2014–2017 in the NHIS database. The date of 1 January 2014 was defined as the cohort entry date for patients aged 40 years or older in 2014. For patients aged less than 40 years in 2014, we established the cohort entry date as the first day of the year that the patient became 40 years old. We excluded 510,805 patients who: • Had taken FQs more than once during the year prior to the cohort entry date • Were diagnosed with AA/AD during the year prior to the cohort entry date • Were diagnosed with underlying related diseases (atherosclerosis of the aorta, arteritis, aortitis, Lerche’s syndrome, coarctation of the aorta, Marfan’s syndrome, valve diseases, endocarditis, congenital malformations of valves, heart failure) (Additional file  1: Table  S1) during the year prior to the cohort entry date. Case selection From the cohort, we identified 29,638 patients aged 40 years or older who had experienced AA/AD from 2014 to 2017 according to the definition of health outcomes of interest “Statistical analysis” section . Patients in the case group were observed from the cohort entry date to the index date, which was defined as the first date of diagno- sis of AA/AD. Control selection After we stratified the case group based on age and sex, we created a risk set for each case using patients who were of the same sex and age as those with AA/AD and did not have a history of an AA/AD diagnosis. The size of the risk set was 20 times the sample size of each stra- tum. We randomly matched four controls in the risk set. Patients in the control group were observed from the cohort entry date to the index date of matched cases. Health outcomes of interest The outcome of the main analysis was defined as a diag- nosis of AA/AD after entry to the cohort. Incident cases were defined as those who had received an ICD 10 code I71 (ICD 10 I71.0–I71.9) for all kinds of AA/AD. The outcome for the sensitivity analysis was redefined as a diagnosis of AA/AD in addition to having received a laboratory test specific for AA/AD (abdominal/thoracic aortography, computed tomography (CT), magnetic res- onance imaging, ultrasonography, Doppler echocardiog- raphy, transesophageal/transthoracic echocardiography, abdominal vascular ultrasonography, or aorta Doppler ultrasonography) within 28 days prior to the diagnosis. The diagnosis and treatment of AA/AD were based on the ACCF/AHA/AATS/ACR/ASA/SCA/SCAI/SIR/STS/ SVM guideline in the general Korean hospitals [12]. The first date of diagnosis was defined as the index date for cases and matched controls. Exposure The exposure of interest was the use of a fluoroquinolone (balofloxacin, ciprofloxacin, enoxacin, gatifloxacin, gemi- floxacin, levofloxacin, lomefloxacin, moxifloxacin, nor- floxacin, ofloxacin, tosufloxacin, and zabofloxacin) in the year prior to the index date. We categorized fluoroquinolone users as current, recent, or past users according to the time from the end of supply of the fluoroquinolone prescription to the index date. In this definition, ‘termination of fluoroquinolone Son et al. BMC Cardiovascular Disorders (2022) 22:44 Page 3 of 9 exposure’ means the end of supply of the fluoroquinolone prescription. Current users were defined as patients who had terminated fluoroquinolone exposure within the 60 days prior to the index date. Recent users were defined as patients who had terminated fluoroquinolone expo- sure 61–120 days prior to the index date. Past users were defined as patients who had terminated fluoroquinolone exposure 121–365 days prior to the index date. To investigate the effects of cumulative dose and dura- tion of FQ exposure on the prevalence of AA/AD, we categorized fluoroquinolone users into three groups according to the quantiles of duration and into four groups according to the quantiles of cumulative dose. The duration was calculated as the sum of the total days of supply for each prescription in the year prior to the index date. The first and third quantiles of the cumulative days of supply were found to be 2 and 14 days, respectively. We represented the cumulative dose in the year prior to the index date in terms of the defined daily dose (DDD), as defined by the anatomical therapeutic chemical clas- sification system. The first, second, and third quantiles of cumulative dose were found to be 4 DDD, 7.5 DDD, and 15 DDD, respectively. The NHIS dataset included the Korean ingredient code of the drug, the date the prescription was written, the number of days of supply, and the quantity. We used these data to identify prescriptions for FQs and any con- comitantly used drugs. Statistical analysis Pearson Chi-square tests and Fisher’s exact test were used for the analysis of categorical variables. The odds ratios of the association between FQ use and AA/AD were calculated using multivariate conditional logistic regression analysis. We considered covariates known to be related to AA/AD or fluoroquinolone use from pre- vious studies and included them as confounders in the model [3, 6–9]. The covariates are listed in Table  1. We also tested the tendency of AA/AD to occur with changes in timing, cumulative dose, and duration of FQ use using the Cochran-Armitage trend test. All data processing and statistical analyses were performed using SAS 9.4 and R 5.3.1 using two-sided tests, and a p value of <  0.05 was considered significant. Results Demographic and clinical characteristics The final study population was composed of 148,190 patients, including 29,638 cases and 118,552 controls. Table  1 shows the baseline characteristics of the study population. This cohort comprised 92,645 male patients (62.5%) and 55,545 female patients (37.5%). More than half of the study population was 60–69 years old (23.7%) or 70–79  years old (32.0%). Patients in the AA/AD case group had a higher prevalence of cerebrovascular dis- ease and cardiovascular disorders such as arterial disease and ischemic heart disease. In the year prior to the index date. Patients in the AA/AD case group were more often users of angiotensin-converting enzyme inhibitors, anti- arrhythmics, anticonvulsants, etc. from the cohort entry date to the index date and experienced more cardiac or aortic procedures and surgeries in the previous year. Association between AA/AD and FQ use During the observation period (1 year before the index date), 8562 cases (28.9%) and 25,387 controls (21.4%) received at least one prescription for FQs. Table  2 and Figure  1 show the results of the conditional logistic regression analysis. The adjusted odds ratio was 1.10 (95% CI 1.07–1.14, p < 0.05) during the 1-year observa- tion period. However, the risk was substantially higher in current users (adjusted OR 1.53, 95% CI 1.46–1.62, p  <  0.05). FQ use did not have a significant association with AA/AD in recent users (adjusted OR 1.00, 95% CI 0.93–1.07, p < 0.05). The risk was even lower in past users (adjusted OR 0.92, 95% CI 0.87–0.96, p < 0.05). The risk of AA/AD was studied according to the dura- tion of exposure and the cumulative dose of FQs. In this study, 25% of FQ users were exposed to FQs for 2 days or less. On the other hand, 25% of the FQ users were exposed to FQs for more than 14 days. Among them, 50% of the FQ users were exposed to FQs for between 3 days and 13 days. We used the same covariates as those adopted for the primary analysis. Patients who used FQs for less than three days had a lower risk of AA/AD than nonusers (adjusted OR 0.87, 95% CI 0.82–0.92, p < 0.05). However, the risk was significantly higher in patients who had used FQs for between three days and 13 days (adjusted OR 1.14, 95% CI 1.09–1.19, p  <  0.05) and was highest in patients who used FQs for more than 14 days (adjusted OR 1.33, 95% CI 1.26–1.40, p < 0.05). FQ users were categorized into four groups with regard to dose (low, mid-low, mid-high, or high) according to the quantiles of cumulative dose. Patients in the low- dose group had used FQs less than 4 DDDs during the observation period. Patients in the mid-low dose group and mid-high dose group had used 4 DDDs to 7.5 DDDs and 7.5 DDDs to 15 DDDs, respectively. Patients in the high-dose group had used more than 15 DDDs during the observation period. Compared with nonusers, the risk of AA/AD in the low-dose group (<4 DDDs) was not significantly higher (adjusted OR 0.97, 95% CI 0.89–1.04, p > 0.05). However, the risk was significantly higher in patients who had used FQs at more than 4 DDDs. Spe- cifically, the adjusted odds ratio of AA/AD was 1.25 (95% CI 1.16–1.34, p  <  0.05) in the mid-low dose group, 1.29 Son et al. BMC Cardiovascular Disorders (2022) 22:44 Page 4 of 9 Table 1 Demographics and clinical characteristics of the study population Case (29,638) Control (118,552) p value* Sex Male Female Age (year) 40–49 50–59 60–69 70–79 80–89 90 + Underlying disease Cerebrovascular disease Arterial disease Ischemic heart disease Cardiac valve disease Conduction disorder Heart failure or cardiomyopathy Chronic obstructive pulmonary disease Pneumonia Cancer Liver disease Renal disease Rheumatism Psychiatric disorder Diabetes Hypertension Lipid disorder Trauma Obstructive sleep apnea Asthma Obesity Seizure disorder Decubitus ulcer Infectious disease Hypothyrodism Inflammatory bowel disease Urinary tract infection Ehlers–Danlos syndrome Charlson comorbidity Index Mean(SD) 0 1 2 3 + Myocardial infarction Congestive heart failure Peripheral vascular disease Cerebrovascular disease N 18,529 11,109 2080 4422 7031 9479 5748 878 3124 8227 7845 884 115 2723 10,626 3579 3422 10,520 1773 2005 13,498 9518 19,716 16,673 15,191 73 6747 41 1225 9 13,949 1729 3876 2287 1 (%) 62.5 37.5 7 14.9 23.7 32 19.4 3 10.5 27.8 26.5 3 0.4 9.2 35.9 12.1 11.5 35.5 6 6.8 45.5 32.1 66.5 56.3 51.3 0.2 22.8 0.1 4.1 0 47.1 5.8 13.1 7.7 0 2.67(2.41) 1.86(2.07) 5371 5781 5425 13,061 1377 3731 7376 6913 18.1 19.5 18.3 44.1 4.6 12.6 24.9 23.3 N 74,116 44,436 8320 17,688 28,124 37,916 22,992 3512 6418 22,080 14,715 444 272 4108 31,726 9047 8462 32,012 2902 5720 41,650 33,700 60,520 50,268 53,280 190 19,825 116 3015 26 46,881 5087 11,773 5442 0 37,377 27,124 19,785 34,266 1924 6315 20,782 16,487 (%) 62.5 37.5 7 14.9 23.7 32 19.4 3 5.4 18.6 12.4 0.4 0.2 3.5 26.8 7.6 7.1 27 2.4 4.8 35.1 28.4 51 42.4 44.9 0.2 16.7 0.1 2.5 0 39.5 4.3 9.9 4.6 0 31.5 22.9 16.7 28.9 1.6 5.3 17.5 13.9 1 1 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 0.002 < 0.001 0.069 < 0.001 0.526 < 0.001 < 0.001 < 0.001 < 0.001 0.2 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 Son et al. BMC Cardiovascular Disorders (2022) 22:44 Page 5 of 9 Table 1 (continued) Dementia Chronic pulmonary diseases Connective tissue disease Peptic ulcer Mild liver diseases Uncomplicated diabetes Diabetes complicated with retinopathy, neuropathy, renal disease Hemiplegia Moderate or severe renal diseases Nonmetastatic solid cancer, leukemia, lymphoma, multiple myeloma Moderate or severe liver diseases Metastatic solid cancer AIDS/HIV Medication use** Angiotensin-converting enzyme inhibitors Antiarrhythmic Anticonvulsant Antidepressant Immunodepressant Anticoagulant β-blocker Oral hypoglycemic agent Benzodiazepine Calcium Channel Blockers corticosteroid Disease-modifying antirheumatic drugs Insulin Loop diuretics Nonsteroidal anti-inflammatory drugs Antipsychotic Peripheral vasodilators Lipid-lowering agent Parkinson medication Hydroxyzine Cardiac or aortic procedure/surgery *The p values are results from Chi-square or Fisher’s exact tests Case (29,638) Control (118,552) p value* N 3995 13,066 1827 10,817 9697 106 2868 936 1773 3288 180 337 7 1119 13,410 3375 5542 8901 20,898 7664 3935 10,934 13,194 16,632 618 1136 3483 23,354 7588 3425 11,079 1990 2616 287 (%) 13.5 44.1 6.2 36.5 32.7 0.4 9.7 3.2 6 11.1 0.6 1.1 0 3.8 45.2 11.4 18.7 30 70.5 25.9 13.3 36.9 44.5 56.1 2.1 3.8 11.8 78.8 25.6 11.6 37.4 6.7 8.8 1 N 11,955 40,316 5172 33,712 29,509 427 10,911 2016 2902 8082 545 765 25 2174 34,164 8889 14,066 27,220 68,727 15,373 18,862 30,683 38,216 56,174 1559 2876 5350 82,504 23,261 4115 31,002 6220 8049 251 (%) 10.1 34 4.4 28.4 24.9 0.4 9.2 1.7 2.4 6.8 0.5 0.6 0 1.8 28.8 7.5 11.9 23 58 13 15.9 25.9 32.2 47.4 1.3 2.4 4.5 69.6 19.6 3.5 26.2 5.2 6.8 0.2 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 0.991 0.012 < 0.001 < 0.001 < 0.001 0.001 < 0.001 0.965 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 **Information on underlying disease were derived from data recorded prior to the index date and after cohort entry. Information on medication use were derived from data recorded in 1 year prior to the index date Table 2 Results of conditional logistic regression analysis of the association between AA/AD and FQ use Case N 21,076 8562 Main analysis Nonusers Users Control Crude OR Adjusted OR* % N % OR 95% CI OR 95% CI 71.1 28.9 93,165 25,387 78.6 21.4 1 1.51** – 1.47–1.56 1 1.10** – 1.07–1.14 *Adjusted for covariates presented in Table 1 (sex, age, underlying disease, Charlson comorbidity index, medication use, history of procedure/surgery) **p < 0.05 Son et al. BMC Cardiovascular Disorders (2022) 22:44 Page 6 of 9 Fig. 1 Results of conditional logistic regression analysis of the association between AA/AD and FQ use (95% CI 1.20–1.38, p < 0.05) in the mid-high dose group, and 1.36 (95% CI 1.26–1.45) in the high dose group. Subgroup analysis and sensitivity analysis From the subgroup analysis by sex (Fig. 1), we found that the association between AA/AD and FQ use remained statistically significant in both the male and female sub- groups. In particular, the risk was high in female patients (adjusted OR 1.15, 95% CI 1.09–1.21, p < 0.05) compared with female nonusers. When ages were grouped into 10-year bands, the association between AA/AD and FQ use remained statistically significant in every age group. To verify the consistency of the results, we performed sensitivity analysis (Table 3) by changing the definition of AA/AD occurrence. The AA/AD occurrence in the primary analysis was identified using the ICD 10 code for AA/AD. For the sensitivity analysis, we changed the definition of an AA/AD case to a diagnosis of AA/AD in addition to having received a laboratory test specific for AA/AD within the 28 days prior to the initial diag- nosis of AA/AD. Among 29,648 AA/AD cases, 21,528 (72.6%) received the laboratory test specific for AA/ AD within the 28 days prior to the initial diagnosis of AA/AD. Among those 21,528 patients, 17,875 (83.0%) were diagnosed with AA/AD the day they took the tests. Abdominal/thoracic CT, aortography, and tran- sthoracic echocardiography were found to have been the commonly performed procedures. The results remained consistent with the primary results under the new definition. The risk of AA/AD by FQs was substantially higher in current users. The risk increased as the duration of exposure and cumulative dose increased. The associa- tion remained statistically significant in every subgroup Son et al. BMC Cardiovascular Disorders (2022) 22:44 Page 7 of 9 Table 3 Results of sensitivity analysis of the association between AA/AD and FQ use Case N 15,294 6234 Sensitivity analysis Nonusers Users Controls Crude OR Adjusted OR* % N % OR 95% CI OR 95% CI 71.0 29.0 67,570 18,542 78.5 21.5 1 1.51** – 1.46–1.56 1 1.10** – 1.06–1.14 Cases that received laboratory tests specific for AA/AD within 28 days prior to the initial diagnosis of AA/AD and their matched controls were included *Adjusted for covariates presented in Table 1 (sex, age, underlying disease, Charlson comorbidity index, medication use, history of procedure/surgery) **p < 0.05 by sex and age. See Additional file  1: Table  S2 for numeric results. Discussion In this study, FQ use showed a trend to be associated with an increased risk of AA/AD during the 1-year observation period, but the effect size was not remark- able. However, the risk of AA/AD in current users of FQs was relatively considerable. This result is in line with preceding research in many ways. In an in  vitro study that assessed the effect of FQs on MMP activi- ties in human aortic smooth muscle cells, 48 hours of treatment with ciprofloxacin significantly increased total MMP activity. Observational studies using Tai- wanese and Swedish databases also showed that the risk of AA/AD within 60 days after FQ use was signifi- cantly higher than that of nonusers [3, 7, 8]. In addition, a cohort study in Ontario, Canada and a signal analysis using U.S. FAERS data also indicated significant asso- ciations between FQ use and AA/AD [6, 9]. This trend is consistent with the results of a systematic literature review and meta-analysis conducted in 2019 [10]. In particular, Pasternak et al. [3] showed that the cumula- tive incidence of AA/AD increased significantly during the first 10 days after FQ use. Given these findings, fur- ther studies are needed to evaluate the risk in the early period of FQ use. Studies that utilized the Taiwanese database [7, 8] reported that the risk of AA/AD increased as the dura- tion of drug use increased. In this study, the adjusted odds ratio of AA/AD also increased as the cumula- tive duration of FQ use increased. In addition, while no prior study has determined the effect of the cumu- lative dose of FQs on the risk for AA/AD, this study showed that an increased cumulative dose of FQs could increase the risk for AA/AD. The dose–response rela- tionship and duration-response relationship can be interpreted as considerable evidence of the causal rela- tionship between FQ use and the occurrence of AA/ AD. Therefore, the patient’s condition should be care- fully monitored, keeping in mind that the risk of AA/ AD may increase as the cumulative dose or duration of FQ use increases. Our study suggests some different results from the general understanding of AA/AD. In general, AA/AD progresses slowly over several years, and men and old age are known as risk factors. However, we found that the risk of AA/AD from FQ use was significant (1) in the early period of FQ use, (2) in female patients and male patients, and (3) in all age groups. In this research, the risk of AA/AD was 8% higher in male FQ users and 15% higher in female FQ users than in nonusers of each sex. Although the risk difference between female patients and male patients was not statistically significant, it gives us a reasonable inference that female patients may have a higher risk of FQ-induced AA/AD, contrary to general knowledge that the incidence of AA/AD is higher in male patients. A previous study also showed that the risk was higher in female patients [7]. For age, the risk was signifi- cant in all age groups, but the differences between sub- groups were not statistically significant. Given that the risk was higher in patients aged 70 or older in a previous study [7], we recommend that further research be under- taken to understand the risk factors for FQ-induced AA/ AD. The sensitivity analysis supported the robustness of the results, as they were very similar to the results before the definition of the study population was changed. Strengths and limitations As an indication of the strength of this research, it was conducted using the national health insurance claim data of all adults aged 40 or older in Korea during the five years from 2013 to 2017. The NHIS-customized data are well accumulated in the form of detailed medi- cal activities and drugs, making it easy to generalize the research results, as nearly all domestic AA/AD patients were included in the study population. Additionally, we comprehensively considered various confounding fac- tors, such as underlying diseases, medication use, and Son et al. BMC Cardiovascular Disorders (2022) 22:44 Page 8 of 9 procedures and surgeries related to AA/AD. Moreover, we performed a sensitivity analysis by changing the def- inition of health outcomes of interest to minimize the effect of classification errors on the results. The preced- ing research results showed a 92% positive prediction for the identification of AA/AD cases when defining a group of cases considering both examination and diag- nosis [7]. Our work clearly has some limitations. First, the results may have been affected by confounding indica- tions. To reduce the bias from confounding indications, we excluded patients who had taken FQs during the year prior to the cohort entry date and included major indications of FQs as covariates in the adjusted model. However, the results may still have been affected by unmeasured underlying indications or the severity of the indication. The result must be carefully interpreted considering that patients who take FQs are possibly at higher risk of AA/AD due to unmeasured underlying conditions, indications for the drug, and important risk factors such as smoking. We emphasize that this result should not be interpreted as explicit evidence for causal effects. It is clear that more studies would be necessary to determine whether there is a causal relationship. Second, due to the nature of the claim data, it is dif- ficult to pinpoint the exact timing of treatment and drug use, and it is not possible to analyze drug use, procedures, or surgeries that are not covered by NHIS. Third, socioeconomic and clinical confounding fac- tors that could not be measured or predicted may have affected the results. For example, the difference in base- line characteristics of cases and controls can affect the results. To minimize the effect of known risk factors for AA/AD, we excluded patients with a history of AA/AD or related diseases during the year prior to the cohort entry date. However, some risk factors generally known to affect AA/AD, such as blood pressure, smoking sta- tus, and family history, were not considered in this study. In this sense, further studies are needed to evalu- ate the risk by patients’ baseline health status and par- ticular medical conditions, such as known risk factors for AA/AD. Finally, the various clinical types and characteristics of AA/AD were not analyzed because clinical information such as severity and detailed disease symptoms could not be fully determined by the diagnosis code alone. Thus, the results of this study are not appropriate for direct application to individuals, as patients may present with a variety of clinical characteristics. To overcome the potential bias introduced by confounding factors and the definition of exposure and health outcome of interest, we performed subgroup analysis, sensitivity analysis, and examined the dose–response relationship. Conclusion In this nested case–control study, we found that the use of FQs within a year was associated with a 10% increased risk of AA/AD in the Korean population. AA/ AD is a life-threatening disease accompanied by severe complications such as low blood pressure, shock, myo- cardial infarction, stroke, lower limb paralysis, and acute renal failure, which can lead to sudden death. In particular, early diagnosis and prompt treatment of abdominal AA/AD are critical, as 65% of patients die from cases of rupture [13]. Therefore, if patients feel symptoms such as chest pain in the early period of FQ use, even if the patient is not in the previously known risk group, medical professionals should suspect acute FQ-induced AA/AD, make a close diagnosis and con- sider changing or stopping the prescription. Moreover, if FQs are used in patients with already identified AA/ AD, medical professionals should review the patient’s history and carefully monitor them after drug admin- istration, keeping in mind that FQs could increase the risk of AA/AD and that the cumulative dose or dura- tion of FQ use may affect the risk. Abbreviations AA: Aortic aneurysm; AD: Aortic dissection; DDD: Defined daily dose; FQs: Fluoroquinolones; NHIS: National Health Insurance Service. Supplementary Information The online version contains supplementary material available at https:// doi. org/ 10. 1186/ s12872- 022- 02488-x. Additional file 1: Table S1. ICD 10 code of AA/AD–related dis- ease. Table S2. Results of conditional logistic regression analysis of the association between AA/AD and FQ use. Table S3. Frequency of underlying disease and mediation use in exposed and unexposed con- trols. Table S4. Association between AA/AD and FQ use in patients with cardiovascular diseases or indications of FQs. Acknowledgements We would like to thank the Benefits Strategy Department of the National Health Insurance Service for support. Authors’ contributions NYS and EMC are co-first authors and contributed equally. NYS contributed to the conceptualization, methodology, software, statistical analysis, and writing of the original draft. EMC contributed to conceptualization, methodology, and writing—review & editing. SYH and SYC contributed to supervision. BGK contributed to the supervision, project administration, writing, reviewing, and editing of the manuscript. All authors read and approved the final manuscript. Funding This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Availability of data and materials The data that support the findings of this study are available from the National Health Insurance Service in the Republic of Korea, but restrictions apply to the availability of these data, which were used under license for the current study and so are not publicly available. Data are, however, available from the authors Son et al. BMC Cardiovascular Disorders (2022) 22:44 Page 9 of 9 upon reasonable request and with permission from the National Health Insur- ance Service. Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in pub- lished maps and institutional affiliations. Declarations Ethics approval and consent to participate Ethics approval for this study was obtained from the institutional review board of Korea Institute of Drug Safety and Risk Management, which waived informed consent (IRB approval number 2019-4). The study protocol conforms to the ethical guidelines of the 1975 Declaration of Helsinki. Consent for publication Not applicable. Competing interests The authors declare that they have no competing interests. Received: 4 March 2021 Accepted: 31 January 2022 References 1. Kim Y, Park Y, Youk T, Lee S, Son Y. A study on the use of antibiotics in Korea and the resistance of major pathogens to antibiotics. NHIS Ilsan hospital Report. 2016:20-001. 2. Redgrave LS, Sutton SB, Webber MA, Piddock LJ. Fluoroquinolone resist- ance: mechanisms, impact on bacteria, and role in evolutionary success. Trends Microbiol. 2014;22(8):438–45. 3. Pasternak B, Inghammar M, Svanström H. Fluoroquinolone use and 4. risk of aortic aneurysm and dissection: nationwide cohort study. BMJ. 2018;360:k678. Food and Drug Administration (FDA). Drug Safety Communication: FDA warns about increased risk of ruptures or tears in the aorta blood vessel with fluoroquinolone antibiotics in certain patients. 2018. https:// www. fda. gov/ Drugs/ DrugS afety/ ucm62 8753. htm. Accessed 14 Jan 2022. 5. Ministry of Food and Drug Safety (MFDS). A notification on the distribu- tion of safety letters on fluoroquinolone antibiotics. https:// www. mfds. go. kr/ brd/m_ 545/ view. do? seq % ED% 94% 8C% EB% A3% A8% EC% 98% A4% EB% A1% 9C% ED% 80% B4% EB% 86% 80% EB% A1% A0& srchTp 0& itm_ seq_1 = itm_ seq & compa ny_ nm & page 0& itm_ seq_2 & Data_ stts_ gubun = 1. Accessed 14 Jan 2022. 0& compa ny_ cd 286& srchFr & srchW ord 0& multi_ & srchTo = = = = = = = = = C9999 6. Daneman N, Lu H, Redelmeier DA. Fluoroquinolones and collagen = 7. 8. associated severe adverse events: a longitudinal cohort study. BMJ Open. 2015;5(11):e010077. Lee C-C, Lee MG, Chen Y-S, Lee S-H, Chen Y-S, Chen S-C, et al. Risk of aortic dissection and aortic aneurysm in patients taking oral fluoroqui- nolone. JAMA Intern Med. 2015;175(11):1839–47. Lee C-C, Lee MG, Hsieh R, Porta L, Lee W-C, Lee S-H, et al. Oral fluo- roquinolone and the risk of aortic dissection. J Am Coll Cardiol. 2018;72(12):1369–78. 9. Meng L, Huang J, Jia Y, Huang H, Qiu F, Sun S. Assessing fluoroquinolone- associated aortic aneurysm and dissection: data mining of the public version of the FDA adverse event reporting system. Int J Clin Pract. 2019;73(5):e13331. 10. Singh S, Nautiyal A. Aortic dissection and aortic aneurysms associated with fluoroquinolones: a systematic review and meta-analysis. Am J Med. 2017;130(12):1449-57.e9. 11. Lee J, Lee JS, Park S-H, Shin SA, Kim K. Cohort Profile: The National Health Insurance Service-National Sample Cohort (NHIS-NSC), South Korea. Int J Epidemiol. 2016;46(2):e15. 12. Foundation ACoC, Guidelines AHATFoP, Surgery AAfT, Radiology ACo, Association AS, Anesthesiologists SoC, et al. 2010 ACCF/AHA/AATS/ ACR/ASA/SCA/SCAI/SIR/STS/SVM guidelines for the diagnosis and management of patients with thoracic aortic disease. J Am Coll Cardiol. 2010;55(14):e27–129. 13. Sakalihasan N, Limet R, Defawe OD. Abdominal aortic aneurysm. Lancet. 2005;365(9470):1577–89. • fast, convenient online submission • thorough peer review by experienced researchers in your field• rapid publication on acceptance• support for research data, including large and complex data types• gold Open Access which fosters wider collaboration and increased citations maximum visibility for your research: over 100M website views per year • At BMC, research is always in progress.Learn more biomedcentral.com/submissionsReady to submit your researchReady to submit your research ? Choose BMC and benefit from: ? Choose BMC and benefit from:
null
10.1088_1402-4896_ad075b.pdf
Data availability statement All data that support the findings of this study are included within the article (and any supplementary files).
Data availability statement All data that support the findings of this study are included within the article (and any supplementary files).
Phys. Scr. 98 (2023) 125223 https://doi.org/10.1088/1402-4896/ad075b PAPER RECEIVED 25 August 2023 REVISED 17 October 2023 ACCEPTED FOR PUBLICATION 26 October 2023 PUBLISHED 10 November 2023 Analysis of a non-integer order compartmental model for cholera and COVID-19 incorporating human and environmental transmissions Muhammad Usman1 , Mujahid Abbas2,3 and Andrew Omame1,4,∗ 1 Abdus Salam School of Mathematical Sciences, Government College University Katchery Road, Lahore 54000, Pakistan 2 Department of Mathematics, Government College University Katchery Road, Lahore 54000, Pakistan 3 Department of Medical Research, China Medical University Hospital, China Medical University, Taichung 40402, Taiwan 4 Department of Mathematics, Federal University of Technology, Owerri, Nigeria ∗ Author to whom any correspondence should be addressed. E-mail: m.usman_20@sms.edu.pk, abbas.mujahid@gcu.edu.pk and andrew.omame@futo.edu.ng Keywords: COVID-19, Cholera, Co-dynamics, Existence and uniqueness, stability, fixed point Abstract Fractional differential operators have increasingly gained wider applications in epidemiological modelling due to their ability to capture memory effect in their definitions; an attribute which lacks in the concept of classical integer derivatives. In this paper, employing the Caputo fractional operator with singular kernel, the co-dynamical model for cholera and COVID-19 diseases is proposed and analyzed, incorporating both direct and indirect transmission routes for cholera. The necessary conditions for existence of the unique solution of the proposed model are studied. Using the results from fixed point theory, Ulam-Hyers stability analysis of the system is performed. The model is fitted to real data from Pakistan and the optimized order of the fractional derivative for which the system fits well to data is obtained. Other numerical assessments of the model are also executed. Phase portraits of the infected classes with different initial conditions and various order of the fractional derivative are obtained in the cases when the reproduction number 0 < trajectories for the infected compartments tend towards the infection-free steady state when and the endemic steady state when irrespective of the initial conditions and the order of the fractional derivative. Increment in the COVID-19 vaccine efficacy, keeping the vaccination rate fixed at , resulted in a decline in the COVID-19 disease. Also, increasing the COVID-19 vaccination rate, keeping the vaccine efficacy for COVID-19 fixed at , led to a decline in the COVID-19 associated reproduction number. The simulations also pointed out the impact of COVID- 19 and cholera vaccinations, direct and indirect transmissions of cholera. 1 . It is observed that the cf =w 1d =w 0 > 0 < 0 > 0.85 and 0.8 1 1 1 1. Introduction The ‘coronavirus disease 2019’ (COVID-19) is a respiratory illness which is caused by the ‘severe acute respiratory syndrome coronavirus 2’ (SARS-CoV-2) [1]. It has symptoms such as a flu-like ailment, fever, muscle pains, loss of taste/smell, fatigue, inability to breath well, cough as well as sore throat [2]. It could be transmitted between humans via direct contact with objects or surfaces that are contaminated [3]. To slow down its transmission, several vaccines have been developed in addition to the multiple non-clinical intervention mechanisms [4, 5]. On the other hand, an acute diarrheal disease known as Cholera has become an endemic infection in Pakistan [6] with different outbreaks reported mainly in most populous regions such as Karachi [7, 8], Swat [9] and other areas of the country [10]. Diarrhea and vomiting are the initial symptoms of this disease. The main cause of this disease is the transmission of a bacterium ‘Vibrio Cholerae (V Cholerae)’ through ingestion of contaminated food and water [11]. Risk factors that tends to increase the susceptibility of indvidual to the infection include lack of proper sanitation, potable water, ecological factors such as heavy downpour and air temperature [12, 13]. Although it originated in Asia © 2023 IOP Publishing Ltd Phys. Scr. 98 (2023) 125223 M Usman et al but its outbreaks have also been reported in many other parts of the world [14]. In 2020, both cholera and COVID-19 pandemics coincided. Although, lower cases of cholera infections were reported globally in that year, the pandemic of COVID-19 greatly affected worldwide surveillance/reporting of cholera. 1 b SI a+ I Numerous integer-order mathematical models have been created to explore the dynamics of cholera transmission b SI based on various transmission routes and incidence rates (see, for example, [15–19]. The saturated incidence rate a+ I was first used by Capasso and Serio [15, 16]. The expression measures the inhibition effect from the behavioural change of susceptible individuals when their number increases or from the crowd, and βI measures the infection force when the disease is entering a fully susceptible population. Because it takes into account the behavioural changes and crowding effects of the infected people and avoids the unboundedness of the contact rate by selecting appropriate parameters, this incidence rate is more desirable than the bilinear incidence rate. In order to model cholera transmission, Codeco [17] presented an incidence form of exclusively) in the year 2001. This was the first time the pathogen concentration was explicitly incorporated. Both transmission paths were assumed by Mukandavire et al [18] in the form of (with environment-to-human transmission SB e + ) K B have since used the aforementioned incidence types. We will combine the incidence rates: respect to cholera transmission in our proposed COVID-19 and cholera co-infection model. and 1 1 a+ I with . Many epidemic models + b ( aSB )+ K B aSB )+ K B SIh b 1 ( ( Mathematical models of the classical integer-order derivative have been adopted in studying the dynamics of infectious diseases [20–26]. These models, due to the integer nature of the derivative constitute limitations. Different fractional operators relying on power-law [27], exponential [28], generalized Mittag-Leffler [29] and other forms of kernels have emerged and their applications to modelling biological processes have gained much interest in recent times [30–39]. Few models have attempted to study the interactions between COVID-19 and cholera in the literature. Hezam [26] proposed and analyzed an optimal control model for COVID-19 and cholera using the integer order derivative. Also, the authors [40] have studied and analyzed a model for SARS- CoV-2 and cholera using real data from Congo. In this paper, a comprehensive non-integer order compartmental model for COVID-19 and cholera incorporating human and environmental transmissions of cholera is proposed, and validated using data from Pakistan (a country with frequent cholera outbreaks and high COVID-19 reported cases). The proposed model also assumes separate vaccination groups for COVID-19 and cholera given the fact that both vaccines have different effectiveness. We have also assumed two co-infection compartments (one involving asymptomatic stage of COVID-19 and the other involving symptomatic stage of COVID-19) which are not available in comparable existing models. We not only have established the conditions for existence, uniqueness, stability but also assessed the impact of COVID-19 and cholera vaccinations as well as direct and indirect transmissions on the dynamics of their co-infection. To the best of our knowledge, the proposed model for this research is novel and appropriate to study the co-circulation of COVID-19 and cholera using fractional calculus tools. 1.1. Preliminaries Definition 1.1. [27] The ‘Caputo fractional derivative’ of a function f of order w Î ) ( 0, 1 is defined by w C D f t ( ) t = 1 G - ( n w ) t ò 0 ( t - à ) n - - w ( ) n 1 f à à ( ) d , ( ) 1.1 where, n [ ]a= + and Γ stands for the Gamma function. 1 Definition 1.2. [27] The Riemann-Liouville fractional integral of a function f of order w Î ) ( 0, 1 is defined by w C I t f ( ) t = t 1 w ( G ) ò 0 ( t - à ) w - 1 f à à ( ) d , t > 0, Lemma 1.1. [27] The ‘Laplace transform of Caputo fractional derivative’ is defined as  { w C D f t ( )} t = w s  { ( )} t f - s w- 1 ( ) f 0 , 0 < < w 1, where  is the ‘Laplace transform operator’. ( ) 1.2 ( ) 1.3 2. Model formulation t( ) , individuals vaccinated against COVID-19 c , individuals vaccinated against cholera t( ) at a given time t is subdivided into: vulnerable or tc( ) , infected with COVID-19 (symptomatic stage) , individuals infected with COVID-19 (in asymptomatic stage) and To formulate the model, the human population uninfected persons k , infected with COVID-19 (in asymptomatic stage) tc( ) infected with cholera  ck( ) cholera t , individuals infected with cholera ck( ) t  tk( ) , individuals infected with COVID-19 (in symptomatic stage) and infected with and recovered from COVID-19, cholera or both t( ) . The cholera population is denoted by . In 2 Phys. Scr. 98 (2023) 125223 M Usman et al Figure 1. Schematic diagram of the model, where, l c = w w b r ( c   c + + c  w r  ck +  ) ck , a  ( ) = c w k w  +  + 1 + w b k w  1 ( +   w  2 k +  k ck  +  + ck ) ck w  3 .  ck Table 1. Model (2.1) parameters’ description. Parameter Description Value References cb w kb w cx w Λω kx w ckx w 1d w 2d w cfw kfw chw khw ckhw μ ω , , ω w 2 w 2 zw z ,1 1aw r ρ ω w w    1 3 χ ω κ ω Ψω w q , 1 Bmw 1Jw 2Jw w 3 w 2 q q , COVID-19 transmission rate Cholera direct transmission rate COVID-19 recovery rate Human recruitment rate Cholera recovery rate Co-infection recovery rate Vaccination rate for COVID-19 Vaccination rate for Cholera Vaccine efficacy against COVID-19 Vaccine efficacy against Cholera COVID-19 induced mortality rate Cholera induced mortality rate co-infection induced mortality rate natural death rate modification parameter progression rate to symptomatic stage Per capita pathogen reproduction rate modification parameter saturated incidence rates Cholera indirect transmission rate bacteria concentration Pathogen carrying capacity Pathogen shed rates Bacteria removal rate −1 −6 day −1 0.0101 day 7.0232 × 10 −1 1 ,1 ] day [ 30 3 225, 000, 000 365´ 67.27 day 0.1 day −1 −1 0.15 day −1 −1 −1 1.701694 day 1.818662 day −1 0.82 day −1 0.60-0.85 day −1 0.0364 day −1 0.024 07 day −1 0.05 day 1 365´ 67.27 −1 day 1.0 −1 [ −1 ,1 14 ] day 1 3 0.3-14.3 day 0.5 0.005 0.07 5000 105 − 107 day 0–100 cell litre day - 1 0.0333 day −1 −1 - 1 Rate of becoming susceptible to COVID-19 after recovery Rate of becoming susceptible to Cholera after recovery 0.003 0.003 3 Fitted Estimated [50] [47] [51] Assumed [40] [40] [52] [53] Fitted [54] Assumed [47] Assumed [50] [51, 55] Assumed Assumed Assumed [56] [55] [51] [26] [26] [26] Phys. Scr. 98 (2023) 125223 M Usman et al this study, the saturated form of incidence is adopted for cholera. Based on the dynamics of COVID-19 and cholera, the following assumptions are taken into consideration: Vulnerable persons acquire COVID-19 and   + ( (via direct transmission from humans) or k cholera at the rates w   + k 2   + + c c  ) ck w  3 w b k w  1 and w b r ( c  + ck   + + +   r 1 ck ck ck ck ) w w w  (via indirect transmission from bacteria). The terms 1z w and 2z w are modification parameters for  c w k + susceptibility to different infection. Diseases related death rates are are Natural death rate is assumed to be μ ω described in table 1 and figure 1, respectively, while the system’s equations are presented in (2.1). w and ckhw , respectively. Recovery rates w and ckx w , for symptomatic infected with COVID-19, cholera and co-infected individuals, respectively. for all compartments. The model’s parameters and flowchart are hw ,c xw ,c h x k k C w   t ( ) t w = L - w b r ( c w   c + + c  w r  +  ) ck ck  - c w k w  +   w b k w +   1 k 1 w m + ( +   ( ck w  2 + k +  w d 1 + ck d  ) ck w +  3 w  ) 2 - -  ck +  w J  + 1 w w b r ( ) c c f w J  , 2 w + +   c c  w r  +  ) ck ck  c C w   c t ( ) t = d w 1  - - ( 1 - c w k w  +   c - 1 + C w   t ( ) t k = d w 2  - w b k w  1 +  k ck  k + +   ( w 2   c w b r ( c  w ´ c w k w  +   k - - ( 1 f w k ) 1 +  w ck ck  r  + ck  ) w 3 + + c  +   ( w 2 w b k w  1 k +  k   c - w m  c +  ) ck ck  k - - ( 1 f w k ) ck  +  + ck ck  ) w 3  k - w m  , k  ck C w   t c ( ) t = w b r ( c w - ( w a 1 + w m )  c   c r + + c  w c 1 w k z - w  +   c w  +  ) ck ck [  + - ( 1 f w c )  c +  ] k - z w 1 1 + w b k w  1 +   ( w 2 k +   k ck  +  + ck ck  ) w 3  , c  ck C w   t c ( ) t = w a 1  c - ( x w c + h w c + w m )  c - z w c 1 w k w  +   c - z w 1 ´ 1 + w b k w  1 +   ( w 2 k +   k ck  +  + ck ck  ) w 3  c ,  ck C w   t k ( ) t = - z w w w b r ( c 2 C w   t ( ) t ck = ( c w k w  +    c + + c  w c 1 w k z + 1 + w b k w  1 +   ( w 2 k +   k ck  w r  +  ) ck ck  , k +  + ck ck  ) w 3 ) [   ck + + - 1  c ( f w k )  ] k - ( x w k + h w k + w m )  k w  +   c + z w 1 1 + w b k w  1 +   ( w 2 k +   k ck  +  + ck ck  ) w 3  c + z w 2  ck w b r ( c w ´   c + + c  w r  +  ) ck ck  k - ( a w 2 + w m )  , ck C w   t ( ) t ck = a w 2  ck + z w c 1 w k w  +   c + z w 1 ´ 1 + w b k w  1 +   ( w 2 k +   k ck  +  + ck ck  ) w 3 C w   t ( ) t = w r  -  ck  )  k  w Y w k - ( x c w ck + h w ck + w m )  , ck + + q w 1  k + q w 2  ck + q w 3  ck - m w B  , x w ck  ck - w ( m + w J 1 + J w 2 )  . ( 1  c + x C w   t ( ) t = x w c where, C C C C C C C C C C ⎧ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎨ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎩ c c c c c c c c c c k k k k k , , , , , , , ck ck ck ck ck ck w   ( ) t t w   ( ) t c t w   ( ) t k t w   ( ) t c t w   ( ) t t w   ( ) t k t w   ( ) t ck t w   ( ) t ck t w   ( ) t t w   ( ) t t , , , c , , , , , c , , , , , k , ,           = Q ) ( t , , , 1 k           = Q ) ( t , , , , 2           = Q ) ( t , , , , 3           = Q ) ( t , , , , 4 c k           = Q ) ( t , , , , 5           = Q ) ( t , , , , 6 k           = Q ( t , 7 c           = Q ( t , , 8 c           = Q ) ( t , , , 9 k           = Q ( t , , , , ck , , , , , ck , , , , , , , ck , ck , ck , ck , k , k , c , c , k , k , c , c , 10 ck ck ck ck ck ck ck ck , , , , , , , , , , , , , , , , , , , , , k k k k k k c c c c c c c c c c c c ) ) ) 4 ( ) 2.1 ( ) 2.2 Phys. Scr. 98 (2023) 125223 M Usman et al The system (2.1) can be represented in the compact form given as: = ( ) where, K t ( ) t ( Î = ] b0,[ . That is, K J: t ( ) t  c   ( , t K t ( )) , C 0 K w ( ) D K t t = ( ) 0 K ⎧ ⎨⎩   ( ) t k 10 is a function. Also, = , ( ) t ( ) t  c  0 k c  ( ) t   :  ( ) ( ) t t ck ck 10 ´     10 T  ( )) ( ) t t defines a function. 10 Î ( ) 2.3 , for each  ( 1 t K t , ( )) w = L - w b r ( c w   c + + c  w r  +  ) ck ck  - c w k w  +   w b k w +   1 k 1 w m + ( +   ( ck w  2 + k +  w d 1 + ck d  ) ck w +  3 w  ) 2 - -  ck +   2 ( , t K t ( )) = d w 1  - - ( 1 w J 1  J + w w b r ( ) c c f w 2 w  , + +   c c  w r  +  ) ck ck  c - c w k w  +  - 1 + w b k w  1 +   ( w 2 k +   k ck  +  + ck ck  ) w 3  3 ( , t K t ( )) = d w 2  -  c - w m  c  ck w b r ( c w   c + + c  w r  +  ) ck ck  k - - ( 1 f w k ) c w k w  +   c  k - - ( 1 f w k )  k - w m  , k ´ 1 + w b k w  1  k + ck  k + +   ( w 2 w b r ( c  ck  + w ck  ) w 3   c  ck  ( 4 , t K t ( )) = - ( w a 1 + w m )  c - + + c  w c 1 w k z w  +   c w r  +  ) ck ck [  + - ( 1 f w c )  c +  ] k - z w 1 1 + w b k w  1 +   ( w 2 k +   k ck  +  + ck ck  ) w 3  , c  ck  5 ( , t K t ( )) = w a 1  c - ( x w c + h w c + w m )  c - z w c 1 w k w  +   c - z w 1 ´ 1 + w b k w  1 +   ( w 2 k +   k ck  +  + ck ck  ) w 3  6 ( , t K t ( )) = ( c w k w  +   ck +  c , 1 + [   ´ + + - c w 1 (   c f w w k  ck r - z w w b r ( c 2 )  ] k +  ) ck  , k + + c  w c 1 w k z w b k w  1 - ck  k + +   ( w 2 +  k ( x  w k  +  ck h + w k ck  ) w 3 + ck m ) w )  k  7 ( t K t , ( )) = w  +   c + z w 1 1 + w b k w  1 +   ( w 2 k +   k ck  +  + ck ck  ) w 3  c + z w 2  ck w b r ( c w ´   c + + c  w r  +  ) ck ck  k - ( a 2 + w m )  , ck  8 ( t K t , ( )) = a 2  ck + z w c 1 w k w  +   c + z w 1 ´ 1 + w b k w  1 +   ( w 2 k +   k ck   9 ( t K t , ( )) = w r  10 ( t K t , ( )) = x w c +  + ck ck  ) w 3   ck c  ( 1  c - +  w Y w x k - )  k ( x w ck + h w ck + w m )  , ck + q w 1  k + q w 2  ck + q w 3  ck - m  , + x w ck  ck - w ( m + w J 1 + J w 2 )  w B . System (2.4) can be written in form of the Volterra integral equation given by ( ) K t = K ( ) 0 + t 1 w ( G ) ò 0 ( t - à ) w -  1 ( à , K à à ( )) d ( ) 2.5 2.1. Basic properties of the model 2.2. Invariant domain Theorem 2.1. The closed set   = h ´  , where b  h =  {( ( ) t ,  c ( ) t ,  k ( ) t ,  c ( ) t ,  c ( ) t ,  k ( ) t ,  ( ) t , ck  ck ( ) t ,  ( )) t Î R 9 + :  ( ) t +  c ( ) t +  k ( ) t +  ( ) t ck +  ( ) t ck +  ( ) t  w L w m } , { is ‘positively invariant’ in relation to the system (2.1). - w w w ( r R  =  Î   + w B Y m : b r ) } . 5 Phys. Scr. 98 (2023) 125223 M Usman et al Proof. Adding all the human components of the system (2.1) gives C 0 w D t  w = L - w m  ( ) t - [ h w c  c + h w k  k + h w ck  ] . ck From (2.6), we have Applying Laplace transform on (2.7), we obtain that C 0 w D t  w < L - w m  , w s   { ( )} t - s w - 1  ( ) 0  w L s - w m   { ( )} t , which further implies that   { ( )} t  w L + w ( s s w m ) +  ( ) 0 s w s w - 1 + m w . By partial fraction, the above expression reduces to   { ( )} t w L  ⎛ w m ⎝ 1 s ⎞ ⎠ - ( w L w m -  ( ) 0 ) s w s w - 1 + m w . The inverse Laplace transform gives  ( ) t  w L w m - ( w L w m -  ( ) 0 ) E w (( - m w w ) ) t . ( ) 2.6 ( ) 2.7 ( ) 2.8 ( ) 2.9 ( ) 2.10 Since the ‘Mittag-Leffler function’ has asymptotic behaviour, we have (2.4) has solutions in  and hence is ‘positively invariant’.  t ( ) Lw w m as t  ¥. Therefore, system 3. Existence and Uniqueness of the solution 3.1. Existence In this section, necessary conditions for existence of solution of the proposed model shall be studied. Consider a Banach space sup ∣ FÎ  F = t  , where, ( )∣ t =   [ ,  10 ] equipped with the norm: |Φ(t)| = |Φ1(t)| + |Φ2(t)| + |Φ3(t)| + |Φ4(t)| + |Φ5(t)| + |Φ6(t)| + |Φ7(t)| + |Φ8(t)| + |Φ9(t)| + |Φ10(t)|. The norms on   ([ , 10 ]) or   ([ ,  ) will be evident from the context of the framework. Theorem 3.1. [41] Let M be a non-empty closed, bounded and convex subset in a Banach Space , operators P P M E 1 2  satisfy the following conditions: :  =   ([ , 10  ]) . If (i) P 1 F + F Î , whenever M P 2 1 2 F F Î , 1 2 M ; (ii) P2 is a contraction. (iii) P1 is compact and continuous. Then there exists MF Î such that F = F + F. P 2 P 1 Theorem 3.2. If  F   ( ))∣ ∣ t Then the proposed model (2.1) has at the least one solution. is continuous and satisfies  Î ´ ( t , 10 ´   , for all   : Y ( )∣ ∣ t ( )) t  F ( t 10 10 , and Y Î   ( ,  ) + with  Y = supt YÎ ∣ ( )∣ t .   F + W Y , F   Proof. Consider B Bh is closed convex and bounded subset of E. Define operators P P B : , 2 1 = F Î { } , where 0 ∣ :  h h ∣ h F Î 0 h 10 by and W = w b wG + ( 1 ) . Obviously t ( t - à ) w -  1 ( à F à à " Î d ( )) t ,  ( P 1 F )( ) t = G 1 w ( 0 ) ò ( P 2 )( )F t = F 0 , " Î  t . Respectively. According to the given assumptions   : 10 ´   10  is continuous and satisfies the condition, for each t Î  and we have ( )F t Î 10 . That is,   : 10 ´   10  is point-wise bounded. Now, for any  ( t , F ( )) t  Y ∣ ( )∣ t 6 F F Î h, 2 ,1 B Phys. Scr. 98 (2023) 125223 M Usman et al  ( P 1 F )( ) t 1 + ( P 2 F 2 )( ) t  = sup Î  t ∣ P 1 F 1 ( ) t + F P 2 ( )∣ t = sup Î  t  sup Î  t ⎡ ⎣ ⎡ ∣ ⎣ F + 0 F + ∣ 0 1 w ( ) G 1 w ( ) G t ( t - à ) w - 1  ( ò 0 à F à à 2 ( )) d , ⎤ ⎦ t ( t - à ) w - 1 ∣ Y à ( )∣ à d ⎤ ⎦ ò 0 t  ∣ F + ∣ 0  G Y w (  ) sup Î  t ò 0 ( t - à ) w - 1 à d  ∣ F + ∣ 0 w b G w w ( w Y   ) ∣ ∣ = F + 0 b G + w ( ) 1 = F + W Y   ∣ ∣ 0 Y    h Hence, P 1 F + F Î h. P 2 B 1 2 It is clear that P2 is a contraction since it is a constant operator. As the function  is continuous, so the operator P1 is also continuous. Now, for any Φ ä Bη, we have  ( P 1 F )( ) t  = sup Î  t ∣ P 1 F ( )∣ t = sup Î  t 1 w ( ) G t ( t - à ) w - 1  ( à F à à ( )) d , ò 0  sup Î G  t 1 w ( ) t ( t - à ) w - 1 ∣ Y à ( )∣ à d ò 0   G Y w (  ) sup Î  t t ( t - à ) w - 1 à d ò 0 w  b G + w ( ) 1   = W Y   Y  h )h is bounded and closed. In order to apply the ‘Arzela Ascoli theorem’, it now Therefore, P1(Bη) ⊂ Bη. As P B1( remains to show that P B1( Now for any Φ ä Bη, consider )h is equicontinuous. ∣( P 1 F )( t ) 2 - ( P 1 F )( )∣ t 1 = 1 w ( ) G t 2 ò 0 ( t 2 - à ) w - 1  ( à F à à ( )) d , - 1 w ( ) G t 1 ò 0 ( t 1 - à ) w - 1  ( à F à à ( )) d , = 1 w ( ) G ⎡ ⎣⎢  Y   G + w ( 1 ) t 1 ò 0 [( t 2 - à ) w - 1 - - à t 1 ( ) w - 1 ]  ( à F à Ã+ ( )) d , [( t w 2 - w t 1 )] t 2 ò t 1 ( t 2 - à ) w - 1 K ( à F à à ( )) d , ⎤ ⎦⎥ It is clear to see that the right hand side of the inequality above tends to zero as t2 → t1. Therefore, P1Bη is equicontinuous and thus, P B1( that implies P1 is a compact operator. Hence, all the conditions of theorem 3.1 are fulfilled. Therefore, there exists Φin  such that Φ(t) = P1Φ(t) + P2Φ(t). That is, )h is closed, bounded and equicontinuous, it is compact and )h . Thus, since P B1( F = F + ( ) t 0 t 1 w ( G ) ò 0 ( t - à ) w -  1 ( à F à à ( )) d , 3.2. Uniqueness Theorem 3.3. If Î   ([ , 10 ]) satisfies the Lipschitz condition for all tÎ  , W < that F F Î ,  ,1 2 . 1 ∣  ( t , F 1 ( )) t -  ( t , F 2 ( ))∣ t   ∣ F 1 ( ) t - F 2 ( )∣ t , ( ) 3.1 0> . Then system (2.5), or its equivalent form (2.1) has unique solution provided 7 Phys. Scr. 98 (2023) 125223 M Usman et al Proof. Define P:   by by ( P F )( ) t = F + 0 t 1 w ( G ) ò 0  ( à F , ( ))( t t - à ) w - 1 à d . For any F F Î , we have  ,1 2  ( P F )( ) t 1 - F ( P 2 )( ) t   sup  Î t ⎡ ⎣ F + 0 1 w ( ) G t ( t - à ) w - 1  ( ò 0 à F à à 1 ( )) d , - F + 0 ⎛ ⎝ 1 w ( ) G t ( t - à ) w - 1  ( ò 0 à F à à 2 ( ) d , ) ⎞ ⎠ ⎤ ⎦⎥ t ) 1 w ( t G (  sup  Î t  ( ò 0 - à F à ( ))∣ , t 2  sup  Î t   w G ( ) ò 0 ( t - à ) w - 1 ∣  ( à F à 1 ( )) , à d - à ) w - 1 ∣ F à - F à ( ) ( )∣ 2 1 à d      2 F - F 1 w G ) ( sup  Î t t ( t - à ) w - 1 à d ò 0 w  b w G + (  =W  ) 1  F 1   F - F 2 1   ( ) t - F 2 ( ) t  , This implies that P is a contraction mapping. Since P(Φ(t)) = P1(Φ(t)) + P2(Φ(t)), PBη ⊂ Bη and the set Bη is closed and convex, the proposed model possess a unique solution as a consequence of Banach contraction principle. 3.3. The Basic reproduction number of the model The ‘disease free equilibrium’ (DFE) of the model (2.1) is: y = 0 (      * , * , c * , k    , * , k * ck * *   ) , * , c w L w + + d 1 = ( w m * , c w w L d 1 w + + d 1 , d w 2 w m m ( w , d w 2 ) w m m ( w * , ck w w L d 2 w + + d 1 , 0, 0, 0, 0, 0, 0, 0, 0 . ) d w 2 ) Following the approach from [42], the transfer matrices for the model are, respe given by w A 1 c 0 0 w w b r c 0 0 A 1 A 1 b b b b b F = ⎛ ⎜ ⎜ ⎜ ⎜ ⎜ ⎝ 0 0 0 + w m 0 0 0 x w k 0 0 w A 2 k 0 0 0 w w b r c 0 w A 2 k 0 0 0 w A 1 c 0 w A 2 k 0 0 0 + + w m 0 0 w h k 0 0 q - w 1 0 0 0 m + w a 2 w q 2 w 2 - - w a x w ck 0 0 w c A 2 w k 0 0 0 ⎞ ⎟ ⎟ ⎟ ⎟ ⎟ ⎠ 0 0 0 0 w h + ck w q - 3 + 0 0 0 0 0 - ⎞ ⎟ ⎟ ⎟ ⎟ ⎟⎟ ⎠ w r w m m w B w w a 1 - m + w a 1 x w c + 0 0 0 0 0 w h c 0 0 0 0 V = ⎛ ⎜ ⎜ ⎜ ⎜ ⎜⎜ ⎝ ( ) 3.2 ( ) 3.3 The ‘basic reproduction number’ of the model (2.1) is given by  r= =- 1 ) ,W COVID-19 and cholera are given by max ( FV 0   , where { 0 D } 0 C0 and K0 are the associated ‘reproduction numbers’ for  0 C = w b a ( c w 1 w a ( 1 + + x ( w c m x )( w + + h c w w + + h c c w m r ) w m ) w ) A 1  0 K = w k b A 2 w + + h k x w k w m + w w c q A 2 1 w w w + + h x )( r k k w m ) w k m ( w B - respectively, where A 1 = [  * + - ( 1 w f c *  )  * c +  * ] k A 2 =  [ * + - ( 1 f w k )  * k +  * ] . c 3.4. Local asymptotic stability of the disease free equilibrium (DFE) of the model Theorem 3.4. The system’s DFE, 0 , is ‘locally asymptotically stable’ (LAS) if 0 < 1 , and unstable if 0 > 1 . 8 Phys. Scr. 98 (2023) 125223 M Usman et al Proof. The stability of system (2.1) in the neighborhood of the DFE is analyzed by Jacobian of system (2.1) evaluated at DFE, 0 , which is given by:  1 0 0 0 - m ( 1 - ( m - w d w w w b r m c w w w d + + 2 1 w w w w f b r d ) c c 1 w w d + + ) 2 1 w w w c 2 w d + + 1 b r d w 2 d d ) w - w m - ( m 0 0 0 0 0 0 0 F - 1  1 w a 1 0 0 0 0 0 - w m 0 0 0 0 0 0 0 0 d d w 1 w 2 ⎛ - ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎝ 0 0 0 0 0 0 0 - - - w w b m [ c ( m ( ( d d w m m - 1 w w w b m c w w d + + 2 1 w w w f b d ) c c 1 w w d + + 1 2 w w b d c 2 w w d d + + 2 1 w w f d + - ( ) 1 c 1 w w w d + + 1 2 -  m w d ) ( 2 ) ) + - w m w 2 w w b L k w d d + + 1 w w w b d L k 1 w w w w d m m + + ) ( 2 1 w w w w L k k 2 w w d + + 2 1 - 1 w w m m ( f b d ) d d ( ) - - d w 2 ] 0 0 F - 2  3 0 0 w q 1 w k x 0 0 0 0 0 0 0 0 0 0 0 0 c - ) d w 2 w w k m ( c d w w w ( w w c L w d + + 1 w w w L 1 w w d d + + 2 1 w w w w f c d L - ) ( k w w w w d + + ( 1 w 2 1 d 2 ) ) k m m k m m - - 0 0 w w w w d m + + - [ ( 1 1 w w w w d + + [ 1 k m m L ] w w f d ) k 2 w ] 2 d -  4 w a 2 w q 2 0 5 0 -  w q 3 w ck x 0 0 - 0 w r w m B - ( w m + w J 1 + J w 2 ⎞ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎠ 0 0 0 0 0 0 0 0 w J 1 + J w 2 ) 0 0 0 0 w x c w x c  1 w 2 d ] where,  1  5 = = f 1 = w a + 1 w x + ck w w b r m c [ = w , + m w h ck w + - ( 1 w w + + d 1  2 w m , w f d ) c w d 2 w 1 ) m ( + The ‘characteristic polynomial’ is given by: ( l + w m 2 ) ( l + +   1 2 ( 1 - w m w b a ( 1 c d w w 1  + + r   1 2 ) 2 A 1 +  ( m 3 w B - w r )( 1 - b w k  A 2 3 This can be written as: + = w c m h ( w + w m d + , F = 2 b , w 1 w w L k + w k w k ) x h +  = 3 w d , 2 w w w + + - f d ) 1 k w w + + d 2 1 ( 1 d d ) w m [ w m m ( + w m ,  4 = a w 2 + w m , w 2 ] . + d w 2 )( l +  )( 4 l +  )( 5 l + w m + w J 1 + J w 2 )[ 2 l + l (  1 +  2 - w w b c c A 1 ) ) ⎤ ⎦ - [ 2 l + l (  3 + m w B w - - r b w k A 2 ) q c 1 w w k m ( B w A 2 - ) r = 0. ) ⎤ ⎦  3 ( ) 3.4 ( ) 3.5 ( ) 3.6 ( ) 3.7 ( l w + + + 2 m ) (   ( 1 1 2 w  m ( B 3 l - - + w m  w C 0 )( 1 w + d 1 2 l )][  - r + + 0 K w d )( 2  l ( 3 = )] 0. l + + m  + l )( 4 w w - - r B  )( 5 w b k l A 2 w m + w J 1 + J w 2 )[ 2 l + l (  1 +  2 - w b f c A 1 ) + ) The eigenvalues are given by: w l 1 l 4 = - = - m x ( w ck ( ) with multiplicity of two , w + m ) = - + m l h ( w 5 w ck w = - l m ( 2 w w + J J ) 2 1 , + + d w 1 + d w 2 ) , l 3 = - ( a w 2 + w m ) , and the solution of the equations given by 2 l + l (  1 +  2 - w b f c A 1 ) +   1 2 ( 1 -  ) 0 C and 2 l + l (  3 + m w B w - - r b w k A 2 ) +  ( m 3 w B - w r )( 1 -  ) 0 K From the Routh-Hurwitz criterion, the equation (3.6) and equation (3.7) has roots with negative real parts if and only if max . Hence, the DFE is locally asymptotically stable if 1 max } <  = < . 1 }   { , OC   { , C OK K 0 0 0 4. Ulam-Hyers stability The Ulam-Hyers (UH) stability and generalized UH stability [43, 44] for the fractional system is discussed in this section. =   (  , Let sup FÎ  F = ( )∣ ∣ , where t t be space of ‘continuous functions’from  to 10 , coupled with the norm b0,[ = 10 ) ] . 9 Phys. Scr. 98 (2023) 125223 M Usman et al Figure 2. Fitting the model to data. Definition 4.1. The model (2.1) or its transformed version given by C w D t ( ) 0 F = ( ) t F = F , 0 ⎧ ⎨⎩  ( t , F ( )) t , is UH stable if ∃k 0 ¯ ( ) C  F e t F Î , of system (4.1) in such a way that, ∃unique solution 0> , such that ∀ ¯ ( ) F - t   D  Î  e ( w t t , , , e > and a given solution of (4.1) satisfying the following inequality, = max ( e i T ) , i = 1, 2 ,... 10. ( ) 4.1 ( ) 4.2 ¯ ( ) F - F  t ( ) t   k e , t Î  , k = max ( k j T ) , j = 1, 2 ,... 10. Definition 4.2. System (4.1) is ‘generalized UH stable’ if ∃a continous function :  such that for any other solution f ¯F Î of the inequality (4.2), ∃ unique solution  +  + with ( )f 0 = 0 F Î satisfying the following: ¯ ( ) F - F  t ( ) t   f e ( ) , t Î  , f = max ( f j T ) , j = 1, 2 ,... 10. Remark 4.1. ‘A function the following properties:’ ¯F Î satisfies the inequality (4.2) if and only if there exists a function h Î , having  i.  ( ) h t  e , t Î  . C ii. D w ¯ ( ) F = t  ( t , ¯ ( ) F + t ( ) h t , t Î  . Lemma 4.1. If ¯F Î holds for system (4.2), then ¯F also holds for the following:  ¯ ( ) F - F + t ¯ 0 ⎛ ⎝ t 1 w ( G ) ò 0 ( t - à ) w -  1 ( ¯ ( )) à F à à d ,  W e ⎞ ⎠ ( ) 4.3 Proof. Using (ii.) of the remark 4.1, we have D C w ¯ ( ) F = t  ( t , ¯ ( )) F t + ( ) h t , t Î  . Apply Caputo integral, so that this is re-written as, ¯ ( ) F = F + t ¯ 0 1 w ( ) G t ò 0 ( t - à ) w - 1  ( ¯ ( )) à F à à + d , 1 w ( ) G t ò 0 ( t - à ) w - 1 h à à ( ) d ( ) 4.4 Re-arranging, and then taking the norm on the both sides and applying the item (i.) of remark 4.1, we obtain that ¯ ( ) F - F + t ¯ 0 1 w ( ) G t ( t - à ) w - 1  ( ¯ ( )) à F à à d , ò 0 ⎞ ⎠  1 w ( ) G t ( t - à ) w - 1 à ∣ ( )∣ h à d ⎛ ⎝ ò 0  ( w b G + w ( 1 ) ) e  e W Theorem 4.1. For all , where 1 - W > 0 F Î and the Lipschitz mapping W = w b wG + ( 1 ) , the model (4.1) is generalized UH stable.   : 10 ´   10  with Lipschitz constant 0> ) with 10 Phys. Scr. 98 (2023) 125223 M Usman et al Figure 3. Simulations of the various classes for different fractional order when 0 < 1 . 11 Phys. Scr. 98 (2023) 125223 M Usman et al Figure 4. Simulations of the Infected classes at different initial conditions when 0 < 1. Proof. If e" >  ¯F Î satisfies the inequality given by (4.2) and Î  , together with lemma 4.1, we have, t0, F Î is a unique solution of (4.1). Then  ¯ ( ) F - F t ( ) t  = sup Î  t ¯ F + 0 1 w ( ) G t ( t - à ) w - 1  ( ¯ ( )) à F à à + d , ò 0 1 w ( ) G t ( t - à ) w - 1 h à à ( ) d ò 0 - F + 0 ⎛ ⎝ 1 w ( ) G t ( t - à ) w - 1  ( à F à à ( )) d , ò 0 ⎞ ⎠  sup Î  t ∣ ¯ F - F + ∣ 0 0 sup Î  t ∣ ( )∣ h t ⎡ ⎣ 1 w ( ) G ⎛ ⎝ t ( t - à ) w - 1 à d ò 0 ⎞ ⎠ ⎤ ⎦ + sup Î G  t 1 w ( ) t ( t - à ) w - 1 ∣  ( t , ¯ ( )) F t -  ( t , F ( ))∣ t à d ò 0  e W +    ¯ F - F  G w ( ) sup Î  t t ( t - à ) w - 1 à d ò 0  e W + ( w b G + w ( e = W + W       ) ) 1 ¯ ( ) F - F t ¯ F - F  ( ) t  . Thus, we have = W 1 where, k - W (UH) and generalized UH stable. . Thus, if we take ¯ F - F    e k , ( ) 4.5 ( )f e e= , then k ( )f 0 = and hence the system (4.1) is both Ulam Hyers 0 12 Phys. Scr. 98 (2023) 125223 5. Numerical scheme M Usman et al The fractional Euler Method shall be adopted to approximate the solution of the model designed in this study. Applying the fundamental theorem of fractional calculus on (2.3), we have ( ) K t = K 0 + t 1 w ( G ) ò 0 ( t - à ) w -  1 ( à , K à à ( )) d , At a given point t = tς+1 = (ς + 1)h , where h = tς+1 − tς is the time step size and ς = 0, 1, 2..., the above equation discretizes to ( K t ) 1 = V + K ( ) 0 + ( K t ) 1 = V + K ( ) 0 + 1 w ( ) G 1 w ( ) G t V + 1 ò 0 V òå j = 0 t ( t V + 1 - à ) w - 1  ( à , K à à ( )) d t j + 1 j ( t V + 1 - à ) w - 1  ( à , K à à ( )) d . ( ) 5.1 With the help of the product rectangle rule [45], we get t j + 1 j ò t ( t V + 1 - à ) w - 1  ( à , K à à = ( )) d w h w [( V - + j w ) 1 - - V ( j w ) ]  ( ( t K t j , j )) Thus, ( K t ) 1 V + = K ( ) 0 + w h G + w ( 1 V ) å V [( j = 0 - + j w ) 1 - - V ( j w ) ]  ( ( t K t j , j )) ( ) 5.2 Adopting the numerical scheme (5.2) into the fractional system (2.5) yields the following numerical solution;  ( t ) 1 V + = S ( ) 0 +  c ( t V + ) 1 = V c ( ) 0 + w h w G + ( 1 ) w h w G + ( 1 V - + j w ) 1 - - V ( j w ) ]  ( 1 ( t K t j , j )) [( V - + j w ) 1 - - V ( j w ) ]  ( ( t K t j , j )) 2  ( t k V + ) 1 = V k ( ) 0 + w h w G + ( 1 [( V - + j w ) 1 - - V ( j w ) ]  ( ( t K t j , j )) 3 1  ( t c V + ) 1 = A c ( ) 0 +  c ( t V + ) 1 = I c ( ) 0 +  k ( t V + ) 1 = I k ( ) 0 + w h w G + ( w h w G + ( 1 w h w G + ( 1 ) ) [( V - + j w ) 1 - - V ( j w ) ]  ( ( t K t j , j )) 4 [( [( V - + j w ) 1 - - V ( j w ) ]  ( ( t K t j , j )) 5 V - + j w ) 1 - - V ( j w ) ]  ( ( t K t j , j )) 6  ( t ck V + ) 1 = A ck ( ) 0 + [( V - + j w ) 1 - - V ( j w ) ]  ( ( t K t j , j )) 7 w h w G + ( ) ) ) [( V å j = 0 V å j = 0 V å j = 0 V å j = 0 V å j = 0 V å j = 0 V å j = 0 V å j = 0 V å j = 0 V å j = 0 1 ) )  ck ( t V + ) 1 = I ck ( ) 0 +  ( t ) 1 V + = B ( ) 0 + w h w G + ( w h w G + ( 1 1 )  ( t ) 1 = V + R ( ) 0 + w h w G + ( 1 ) [( V - + j w ) 1 - - V ( j w ) ]  ( ( t K t j , j )) 8 [( V - + j w ) 1 - - V ( j w ) ]  ( ( t K t j , j )) 9 [( V - + j w ) 1 - - V ( j w ) ]  10 ( ( t K t j , j )) ( ) 5.3 The error estimates, stability analysis and th high accuracy of this numerical scheme have been well explored in [46] 6. Model Fitting and numerical assessment 180, 000, 000  , 3000 Demographic data related to Pakistan have been used for the simulations. The initial conditions are set as: ( ) =  c( ) = k( ) = c( ) = 0 0 15, 000 0 0 . For the fitting of model to data, available ( ) = k( ) =  , 0 0 0 records for reported COVID-19 cases in Pakistan [47] between Jan 1, 2022 to Apr 10, 2022, the fractional model is fitted to real data.  c( ) = 0 ck( ) = 0 5, 000 20, 000, 000  ck( ) = , 0  , 200, 000 ( ) = , 0 1, 296, 527  , 1000 0 , , , 13 Phys. Scr. 98 (2023) 125223 M Usman et al Figure 5. Simulations of the various classes for different fractional order when 0 > 1 . 14 Phys. Scr. 98 (2023) 125223 M Usman et al Figure 6. Simulations of the Infection classes at different initial conditions when 0 > 1. Figure 7. Simulations to assess the impact of COVID-19 vaccination rates on infected classes. 15 Phys. Scr. 98 (2023) 125223 M Usman et al Figure 8. Impact of COVID-19 vaccine efficacy while keeping the vaccination rate same. Figure 9. Impact of COVID-19 vaccination rate while keeping the vaccine efficacy same. The fitting of the model to the cumulative COVID-19 cases [47], was done using the fmincon function in the Optimization Toolbox of MATLAB [48]. The fmincon’s optimization routine syntax: = ( , , , , , x @ , 0, fmincon modelfun x A b Aeq beq lb ub nonlcon options , starts at x0 (the initial guesses) and finds an optimized x to the function described in @modelfun that fits the model to a given data set, subject to the nonlinear inequalities c(x) or equalities ceq(x) defined in nonlcon, and also subject to the linear inequalities A. x „ b and linear equalities Aeq. x = beq, defined in A, b, Aeq, beq, respectively. x0 can be a scalar, vector, or matrix. lb and ub are the bounds on the parameters to be estimated. The optimization parameters and error tolerance are specified in options. The method utilizes the least squares method, which is very efficient and reliable [49]. The method seeks to fit the observed data sets, Yi, with the estimated values, Xi, such that; the sum of squares of errors between the observed and fitted curve is minimal [49]. The sum of squares error, SSE, is illustrated mathematically as: , , ) k å= = 1 i The fitting which is shown in figure 2 reveals that our model behaves very well with the Pakistan real data. The estimated parameters are given in table 1. The optimized value of the Caputo derivative for which the model fits well to data is ω = 0.97. 2 ) X . i SSE - Y i ( The flow chart describing the fitting process is also given in the Appendix (figure A1). In figures 3(a)–(j), simulations of all the epidemiological compartments at different orders of the derivatives are presented. It is observed that the fractional order greatly impact the dynamics of the disease in when each compartment. This is due to the memory effect which is an important feature of definition of the non- integer derivative. 0 < 1 The trajectory diagram for the infected classes (AC, IK, ACK) at different initial conditions when and different order of the derivative are presented in figures 4(a)–(e), respectively. It can be observed that trajectories for each infected compartment tends towards the infection-free steady state when initial conditions and order of the derivative. irrespective of the 0 < 0 < 1 1 Also, as can be observed in figures 5(a)–(j), simulations of all the model compartments at different orders of are presented. It can be observed that the fractional order greatly impact 1 the fractional derivatives when the dynamics of the disease in each compartment. 0 > It is worthy of mention, to point out that, the Caputo fractional operator endowed with a singular kernel provides more advantages in modeling epidemic disease transmissions by assuming a more flexible framework that captures 16 Phys. Scr. 98 (2023) 125223 M Usman et al Figure 10. Simulations of the cholera and co-infected classes to assess the impact of direct and indirect transmissions. memory effects, non-local behavior, as well as complex dynamics. This memory effects suggests that the history of the system is captured. In a co-dynamical transmission model for COVID-19 and cholera, this might be particularly useful for capturing the effect of past infections, immunity, or interventions on the current state of the population. Unlike classical integer derivative, fractional derivative shows that the behavior of the system at a particular point in time depends on its history over a range of time, which could be crucial for modeling the spread of disease co- dynamics since past interactions can greatly influence future outcomes. The phase portraits of the infected classes AC, IK and ACK at different initial conditions when and for different order of the derivative are presented in figures 6(a)–(e), respectively. It can be observed that the solution paths for all the infected classes tend towards the endemic equilibrium when conditions and order of the derivative. 1 , irrespective of the initial 0 > 0 > 1 Numerical assessment to observe the epidemiological impact of COVID-19 vaccination on the dynamics of infected compartments are presented in figures 7(a)–(d). It is observed that increasing vaccination rates for COVID-19 lead to positive population level impact on infected compartments. Thus, for the reduction of the co-spread of both diseases, more effort should be harnessed to increase vaccination rates for COVID-19. It is observed from the figures 8(a)–(b) that increasing the vaccine efficacy while keeping the vaccination rate same yield decline in the infected individuals. It has been estimated that increment in the vaccine efficacy from fc = 0.40 to fc = 0.60 yield decline of 34.21% and 33.32% for asymptomatic and symptomatic infected individuals with COVID-19, respectively. It has also been observed that increment from fc = 0.40 to fc = 0.80 yield decline of 67.13% and 65.87% for the above mentioned compartments. It is observed from the figures 9(a)–(b) that increasing the vaccination rate while keeping the vaccine efficacy same yield the same behaviour described in the previous simulations. Specifically, it was observed that increment in the vaccination rate from fc = 0.40 to fc = 0.60 yield decline of 31.80% and 30.78%, 17 Phys. Scr. 98 (2023) 125223 M Usman et al Figure 11. Effect of some important model parameters on COVID-19 associated reproduction number. respectively for asymptomatic infected individuals and symptomatic infected individuals with COVID-19. It was also observed that an increment in the vaccination rate, from fc = 0.40 to fc = 0.80 per day yields decrease of 48.36% and 46.98% asymptomatic and symptomatic infected individuals, respectively. Simulations to assess the impact of direct transmission of cholera disease on the dynamics of single infections and co-infections are presented in figures 10(a)–(c), respectively. It is observed that the direct transmission rates greatly impact the dynamics of the disease in these compartments. Similarly, simulations to investigate the impact of indirect transmission of cholera on the infected components of the model are presented in figures 10(d)–(f). Also, a noticeable impact of this transmission route is observed on the dynamics in the infected compartments. Figures 11(a)–(g) show the contour plots of the COVID-19 associated reproduction number as a function of the transmission rate and some other important parameters. It can be noticed from the figures that an increase in the transmission rates for COVID-19 resulted in a corresponding increase in the reproduction number (as expected). Similarly, increasing the transition rate 1aw also lead to an increase in the reproduction number. Increment in the COVID-19 vaccine efficacy (while keeping the vaccination rate fixed at COVID-19 vaccination rate (while fixing the vaccine efficacy for COVID-19 at cf =w COVID-19 associated reproduction number. The impacts of other parameters such as ) resulted in a decline in the COVID-19 disease burden. Also, increasing the ) led to a decline in the w and ρ ω x on the c 0.85 w 2d , 1d =w 0.8 w ,c h 18 Phys. Scr. 98 (2023) 125223 M Usman et al Figure 12. Effect of different parameters on reproduction number for cholera reproduction number are also pointed out. Figures 12(a)–(f) show the contour plots of the cholera associated reproduction number as a function of the transmission rate and some other important parameters. It is observed that, increasing the cholera vaccine efficacy ( ) led to a decline in the cholera disease burden. Similarly, reduction in cholera associated reproduction number is observed when we increase the vaccination rate( influencing the cholera dynamics are 2d w) for cholera while fixing the cholera vaccine efficacy ( w and 1qw. k kfw) while fixing the cholera vaccination rate ( ). Other parameters kf =w 2d =w hw ,k 0.22 0.80 x 7. Conclusion In this paper, a co-dynamical cholera and COVID-19 model, incorporating both direct and indirect transmissions for cholera, is developed and analyzed using the concepts from fractional calculus. The definition of Caputo operator is used and the necessary conditions for the existence of unique solution of the system are derived. Using the results from fixed point theory, the system’s stability analysis is discussed in the sense of Ulam-Hyers. The model is related with real data from Pakistan and fractional order which gave the best fit was also investigated. Numerical experiments on the impact of vaccination were also performed. Through simulations, it was also pointed out the impact of COVID-19 and cholera vaccinations, direct and indirect transmissions of cholera on both diseases dynamics. Highlights of the simulation include: (i) The model relates well with data when the order of fractional derivative is taken as ω = 0.97. (ii) Varying fractional order greatly impact the dynamics of diseases in each compartment. (iii) Phase portraits of the infected classes at different initial conditions revealed that the trajectories of the infected compartments tend towards the infection-free steady state when state when 1 , irrespective of the initial conditions and order of the fractional derivative. 0 < 1 and the endemic steady 0 > 19 Phys. Scr. 98 (2023) 125223 M Usman et al (iv) Increment in the COVID-19 vaccine efficacy (while keeping the vaccination rate fixed at ) resulted in a decline in the COVID-19 disease burden. Also, increasing the COVID-19 vaccination rate (while fixing the vaccine efficacy for COVID-19 at ) led to a decline in the COVID-19 associated reproduction number. The simulations also pointed out the impact of COVID-19 and cholera vaccinations, direct and indirect transmissions of cholera. cf =w 1d =w 0.85 0.8 (v) Increasing vaccination rates for COVID-19 or cholera also resulted in some positive impact on the co- infected compartments. (vi) The indirect transmission rate had more impact on the dynamics of cholera in single and co-infected compartments. Thus, to reduce the co-spread of both diseases, more effort should be harnessed to increase vaccination rates for both diseases, and also curtail the direct and indirect transmission of cholera infection. The research in this paper can be extended in the following ways: One could consider stochastic equivalence as well as fractal fractional of the current model for a possible research problem. Approximate solution of the model using some other novel numerical schemes that can yield the better results can also be considered. Moreover, one could also establish the existence, uniqueness and stability results using some novel fixed point theorems. Acknowledgments The second author was supported by the Higher Education Commission of Pakistan (NRPU project No. 9340). Data availability statement All data that support the findings of this study are included within the article (and any supplementary files). Conflict of interest The authors declare that they have no conflict of interests. Appendix Figure A1. Flow chart for data fitting of the model. 20 Phys. Scr. 98 (2023) 125223 ORCID iDs M Usman et al Muhammad Usman Mujahid Abbas Andrew Omame https://orcid.org/0000-0001-6818-086X https://orcid.org/0000-0001-5528-1207 https://orcid.org/0000-0002-1252-1650 References [1] Phelan A L, Katz R and Gostin L O 2020 The novel coronavirus originating in Wuhan, China: challenges for global health governance JAMA 323 709–10 [2] Woelfel R et al 2020 Virological assessment of hospitalized patients with COVID-2019 Nature 581 465–9 [3] Bai Y, Yao L, Wei T, Tian F, Jin D Y, Chen L and Wang M 2020 Presumed asymptomatic carrier transmission of COVID-19 JAMA 323 1406–7 [4] United States Food and Drug Administration (2020). FDA Takes key action in fight against covid-19 by issuing emergency use authorization for first covid-19 vaccine. Accessed June 17, 2021. 2021 https://fda.gov/news-events/press-announcements/ [5] Interim Clinical Considerations for Use of COVID-19 Vaccines Currently Authorized in the United States. Accessed July 14, 2021. 2021 https://cdc.gov/vaccines/covid-19/clinical-considerations/covid-19-vaccines-us.html [6] Maliha Naseer T J 2022 Epidemiology, determinants and dynamics of cholera in Pakistan: gaps and prospects for future research— PubMed J. Coll. Physicians Surg .Pak. 24 855–60. https://pubmed.ncbi.nlm.nih.gov/25404447/ [7] Dickson W P 1882 Report on an Outbreak of Cholera in the Lahore Central Jail in August 1881 Ind Med Gaz 17 204–8 [8] Siddiqui F J et al 2006 Consecutive outbreaks of Vibrio cholerae O139 and V. cholerae O1 cholera in a fishing village near Karachi, Pakistan Trans R Soc. Trop. Med. Hyg. 100 476–82 [9] Khan H A, Masood W, Siddiqui A, Ahmad S, Salman Y and Essar M Y 2022 The Cholera outbreak in Karachi, Pakistan: Challenges, efforts and recommendations Annals of Medicine and Surgery 78 103873 [10] 129 confirmed cholera cases in Karachi in 2022 but no outbreak, say health officials. https://www.thenews.com.pk/print/953609- 129-confirmed-cholera-cases-in-karachi-in-2022-but-no-outbreak-say-health-officials. (Accessed 6 May 2022). 2022 [11] Cholera. https://www.who.int/news-room/fact-sheets/detail/cholera. (Accessed 6 2022). 2022 [12] Jutla A et al 2013 Environmental factors influencing epidemic cholera The American Journal of Tropical Medicine and Hygiene 89 597 [13] Cholera—symptoms and causes—Mayo clinic. https://www.mayoclinic.org/diseases-conditions/cholera/symptoms-causes/syc- 20355287. (Accessed 6 May 2022). 2022 [14] Cholera, 2015. Releve epidemiologique hebdomadaire/Section d-hygiene du Secretariat de la Societe des Nations=Weekly epidemiological record/Health Section of the Secretariat of the League of Nations. 2016;91:433-40. [15] Capasso V and Serio G 1978 A generalization of the Kermack-McKendrick deterministic epidemic model Math. Biosci. 42 43–61 [16] Capasso V and Paveri-Fontana S L 1979 Mathematical model for the 1973 cholera epidemic in the European Mediterranean region, Revue d’Epidemiologie et de Sante Publique 27 121–32 [17] Codeco C T 2001 Endemic and epidemic dynamics of cholera: the role of the aquatic reservoir BMC Infectious Diseases 1 1 [18] Mukandavire Z, Liao S, Wang J, Gaff H, Smith D L and Morris J G Jr. 2011 Estimating the reproductive numbers for the 2008-2009 cholera outbreaks in Zimbabwe PNAS 108 8767–72 [19] Liao S and Wang J 2012 Global stability analysis of epidemiological models based on Volterra-Lyapunov stable matrices Chaos Solitons Fractals 45 966–77 [20] Omame A and Abbas M 2023 The stability analysis of a co-circulation model for COVID-19, dengue, and zika with nonlinear incidence rates and vaccination strategies Healthcare Analytics 3 100151 [21] Olaniyi S 2018 Dynamics of zika virus model with nonlinear incidence and optimal control strategies Appl. Math. Inform. Sci. 12 969–82 [22] Opara C Z, Uche-Iwe N, Inyama S C and Omame A 2020 A mathematical model and analysis of an sveir model for streptococcus pneumonia with saturated incidence force of infection Mathematical Modelling and Applications. 5 16–38 [23] Asamoah J K K, Okyere E, Abidemi A, Moore S E, Sun G-Q, Jin Z, Acheampong E and Gordon J F 2022 Optimal control and comprehensive cost-effectiveness analysis for COVID-19 Results in Physics 33 105177 [24] Addai E, Zhang L, Preko A K and Asamoah J K K 2022 Fractional order epidemiological model of SARS-CoV-2 dynamism involving Alzheimer’s disease Healthcare Analytics 2 100114 [25] Omame A and Abbas M 2023 Modeling SARS-CoV-2 and HBV co-dynamics with optimal control Physica A 615 128607 [26] Hezam I M, Foul A and Alrasheedi A 2021 A dynamic optimal control model for COVID-19 and cholera co-infection in Yemen Adv Differ Equ. 108 2021 [27] Caputo M 1966 Linear models of dissipation whose Q is almost frequency independent Annals of Geophysics 19 383–93 [28] Caputo M and Fabrizio M 2015 A new definition of fractional derivative without singular kernel Progress in Fractional Differentiation and Applications 1 1–3 [29] Atangana A and Baleanu D 2016 New fractional derivatives with nonlocal and non-singular kernel: theory and applications to heat transfer model Therm Sci. 20 763–9 [30] Peter O J, Qureshi S, Yusuf A, Al-Shomrani M and Abioye Idowu A 2021 A new mathematical model of COVID-19 using real data from Pakistan Results in Physics 24 104098 [31] Naik P A et al 2020 Modeling and analysis of COVID-19 epidemics with treatment in fractional derivatives using real data from Pakistan Eur. Phys. J. Plus 135 795 [32] Musa S S, Qureshi S, Zhao S, Yusuf A, Mustapha U T and He D 2021 Mathematical modeling of COVID-19 epidemic with effect of awareness programs Infectious Disease Modelling 6 448–60 [33] Memon Z, Qureshi S and Memon B R 2021 Assessing the role of quarantine and isolation as control strategies for COVID-19 outbreak: A case study Chaos, Solitons Fractals 144 110655 [34] Ibrahim A, Humphries U W, Ngiamsunthorn P S, Baba I A, Qureshi S and Khan A 2023 Modeling the dynamics of COVID-19 with real data from Thailand Sci. Rep. 13 13082 [35] Din A, Li Y, Khan F M, Khan Z U and Liu P 2021 On Analysis of fractional order mathematical model of Hepatitis B using Atangana- Baleanu Caputo (ABC) derivative Fractals 30 2240017 [36] Din A, Li Y, Yusuf A and Ali A I 2022 Caputo type fractional operator applied to Hepatitis B system Fractals 30 2240023 21 Phys. Scr. 98 (2023) 125223 M Usman et al [37] Liu P, Din A and Zarin R 2022 Numerical dynamics and fractional modeling of hepatitis B virus model with non-singular and non-local kernels Results in Physics 39 105757 [38] Liu P, Huang X, Zarin R, Cui T and Din A 2023 Modeling and numerical analysis of a fractional order model for dual variants of SARS- CoV-2 Alexandria Engineering Journal 65 427–42 [39] Usman M, Abbas M and Omame A 2023 Analysis of the solution of a model of sars-cov-2 variants and its approximation using two-step lagrange polynomial and euler techniques Axioms 12 480 [40] Ozkose F, Habbireeh R and Senel M T 2023 A novel fractional order model of SARS-CoV-2 and Cholera disease with real data J. Comput Appl Math 423 114969 [41] Yong Z, Jinrong W and Lu Z 2016 Basic Theory of Fractional Differential Equations (World Scientific) [42] van den Driessche P and Watmough J 2002 Reproduction numbers and sub-threshold endemic equilibria for compartmental models of disease transmission Math. Biosci. 180 29–48 [43] Ulam S M 1960 A Collection of Mathematical Problems (New York, NY, USA: Interscience Publ.) [44] Ulam S M 2004 Problem in Modern Mathematics (Mineola, NY, USA: Dover Publications) [45] Diethelm K, Ford N J and Freed A D 2004 Detailed error analysis for a fractional Adams method Numer. Algorithms 36 31–52 [46] Baleanu D, Jajarmi A and Hajipour M 2018 On the nonlinear dynamical systems within the generalized fractional derivatives with Mittag-Leffler kernel Nonlinear Dyn 94 397–414 [47] Pakistan: Coronavirus Pandemic Country Profile. https://ourworldindata.org/coronavirus/country/pakistan. Accessed 11 Novem- ber, 2022 2022 [48] McCall J 2005 Genetic algorithms for modelling and optimisation J. Comput. Appl. Math. 184 205–22 [49] Martcheva M 2015 An Introduction to Mathematical Epidemiology vol 61 (Springer) [50] Okuonghae D and Omame A 2020 Analysis of a mathematical model for COVID-19 population dynamics in Lagos Nigeria, Chaos, Solitons Fractals 139 [51] Cash R A, Music S I, Libonati J P, Snyder M J, Wenzel R P and Hornick R B 1974 Response of man to infection with Vibrio cholerae. I. Clinical, serologic, and bacteriologic responses to a known inoculum J. Infect .Dis. 129 45–52 [52] Yang Z R et al 2023 Efficacy of SARS-CoV-2 vaccines and the dose-response relationship with three major antibodies: a systematic review and meta-analysis of randomised controlled trials Lancet Microbe. 4 e236–46 [53] Song K R, Lim J K, Park S E, Saluja T, Cho S I, Wartel T A and Lynch J 2021 Oral Cholera Vaccine Efficacy and Effectiveness (Basel) 9 1482 [54] Shuai Z, Tien J H and van den Driessche P 2012 Cholera models with hyper infectivity and temporary immunity Bull. Math. Biol. 74 2423–45 [55] Jensen M A, Faruque S M, Mekalanos J J and Levin B R 2006 Modeling the role of bacteriophage in the control of cholera outbreaks PNAS 103 4652–7 [56] Sanches R P, Ferreira C P and Kraenkel R A 2011 The role of immunity and seasonality in cholera epidemics Bull. Math. Biol. 73 2916–31 22
null
10.1103_physrevd.106.012002.pdf
null
null
PHYSICAL REVIEW D 106, 012002 (2022) Search for resonances decaying to three W bosons in the hadronic p ffiffi s = 13 TeV final state in proton-proton collisions at A. Tumasyan et al.* (CMS Collaboration) (Received 24 December 2021; accepted 15 June 2022; published 6 July 2022) p ffiffiffi s A search for Kaluza-Klein excited vector boson resonances, WKK, decaying in cascade to three W bosons via a scalar radion R, WKK → WR → WWW, in a final state containing two or three massive jets is ¼ 13 TeV proton-proton collision data collected by the CMS presented. The search is performed with experiment at the CERN LHC during 2016–2018, corresponding to an integrated luminosity of 138 fb−1. Two final states are simultaneously probed, one where the two W bosons produced by the R decay are reconstructed as separate, large-radius, massive jets, and one where they are merged into a single large- radius jet. The observed data are in agreement with the standard model expectations. Limits are set on the product of the WKK resonance cross section and branching fraction to three W bosons in an extended warped extra-dimensional model and are the first of their kind at the LHC. DOI: 10.1103/PhysRevD.106.012002 I. INTRODUCTION The search for physics beyond the standard model (SM) is one of the most important elements of the research program at the CERN LHC. Direct searches performed at the LHC have not yet found any compelling evidence for such new physics. However, novel ideas and recently developed techniques expand the potential for discovery. For example, in the CMS Collaboration, deep machine learning techniques for tagging Lorentz-boosted resonan- ces decaying hadronically [1] have been developed and exploited extensively for both searches beyond the SM and measurements of the properties of the Higgs boson (H) [2]. New physics scenarios involving yet-unprobed signatures of resonant triboson final states through a two-step cascade decay of heavy resonances in extended warped extra- dimensional models [3–8] have recently been proposed. These models provide an attractive extension of the SM, which addresses the Planck-electroweak scale difference and flavor hierarchy problems simultaneously. The theory model probed assumes a Randall-Sundrum scenario with an extended bulk consisting of two extra branes other than the one on which the SM resides [3]. Only the electroweak gauge fields can propagate into the extended bulk. The size *Full author list given at the end of the article. Published by the American Physical Society under the terms of license. the Creative Commons Attribution 4.0 International Further distribution of this work must maintain attribution to the author(s) and the published article’s title, journal citation, and DOI. Funded by SCOAP3. of the extra dimension is stabilized with a mechanism introducing a potential with a modulus field [9], resulting in a bulk scalar boson, the radion, for each additional brane. Such extended models can also incorporate heavy reso- nances that have enhanced decays into triboson final states as compared with direct decays into dibosons and top quark-antiquark pairs. Thus, a set of new final states emerges with a discovery potential within LHC reach. In this paper, we report on a search for massive resonances decaying in a cascade into three W bosons, → WR and R → WW, where WKK is a through WKK Kaluza-Klein (KK) massive excited gauge boson and R is a scalar radion. The analysis is based on proton-proton ¼ 13 TeV collected by the CMS (pp) collision data at experiment at the LHC during 2016–2018, corresponding to an integrated luminosity of 138 fb−1. Since the WKK excitation has a mass of the order of several TeV, the W bosons typically have transverse momenta (pT) of several hundred GeV. p ffiffiffi s In a large fraction of the parameter space (mR ≲ 0.8mWKK), the W boson not originating from the radion decay is highly boosted and its decay products are contained in a single large- radius jet. However, depending on the relative masses of the WKK and R resonances, the two W bosons from the R decay can either produce two large-radius jets (“resolved” case), or one single large-radius jet containing both W bosons (“merged” case). These two possibilities are illustrated in Fig. 1; the merged case is predominant when mR ≤ 0.2mWKK, where mR and mWKK are the masses of the R and WKK bosons, respectively. As a result, the final states considered here require two or three massive jets, predominantly targeting 2470-0010=2022=106(1)=012002(33) 012002-1 © 2022 CERN, for the CMS Collaboration A. TUMASYAN et al. PHYS. REV. D 106, 012002 (2022) using jet substructure techniques, can be detected through this search. This paper is organized as follows: Section II provides a description of the CMS detector. Section III describes the datasets and simulation samples used in the analysis. The triggers used for data collection and the event reconstruction are discussed in Sec. IV. The massive jet tagging is described in Sec. V. The event selection and event categorization are presented in Sec. VI. The jet tagger calibration is described in Sec. VII. Section VIII describes the estimation of the SM background. Systematic uncertainties are discussed in Sec. IX. The results and their interpretation are given in Sec. X. A summary is presented in Sec. XI. Tabulated results are provided in the HEPData record for this analysis [13]. II. THE CMS DETECTOR The central feature of the CMS detector is a super- conducting solenoid of 6 m internal diameter, providing a magnetic field of 3.8 T. A silicon pixel and strip tracker, a lead tungstate crystal electromagnetic calorimeter (ECAL), and a brass and scintillator hadron calorimeter (HCAL), each composed of a barrel and two end cap sections resides within the solenoid volume. Forward calorimeters extend the coverage provided by the barrel and end cap detectors up to pseudorapidities of jηj ¼ 5. Muons are measured in gas-ionization detectors embedded in the steel flux-return yoke outside the solenoid. Events of interest are selected using a two-tiered trigger system. The first level, composed of custom hardware processors, uses information from the calorimeters and muon detectors to select events at a rate of around 100 kHz, making a decision within the fixed period of 4 μs following the beam crossing, allowed by the latency implemented in the readout path [14]. The second level, known as the high- level trigger (HLT), consists of a farm of processors running a version of the full event reconstruction software optimized for fast processing, and reduces the event rate to around 1 kHz before data storage [15]. A more detailed description of the CMS detector, together with a definition of the coordinate system and kinematic variables, can be found in Ref. [16]. III. DATA SAMPLES AND EVENT SIMULATION The data samples analyzed in this search correspond to a total integrated luminosity of 138 fb−1. They were recorded ¼ 13 TeV in the years 2016, 2017, in pp collisions at and 2018, comprising 36.3, 41.5, and 59.7 fb−1, respec- tively [17–19]. p ffiffiffi s The signal is simulated at leading order (LO) using the MadGraph5_aMC@NLO 2.4.2 generator [20], covering a wide range of WKK and R masses (mWKK from 1.5 to 5.0 TeV, and mR from 6 up to 90% of mWKK), together with the parameters recommended by the authors of Refs. [3–6], ¼ 6, i.e., a KK coupling to the radion and a W boson ggrav FIG. 1. Schematic diagrams of the decay of a KK excitation of a W boson (WKK) to the final states considered in this analysis. Additional jets are allowed in the analysis but not considered explicitly. Left: three individually reconstructed W bosons; right: one individually reconstructed W boson and two W bosons reconstructed as a single large-radius jet, which is predominant for mR ≤ 0.2mWKK. merged and resolved R decay topologies, respectively, and no isolated charged leptons. However, nonisolated leptons are allowed to be present inside the jets formed by merged radion decay products R → WW → lνqq. It is also possible to have additional jets in the “compressed mass” scenario, mR ≳ 0.8mWKK (depending on the specific value of mWKK), which can feature at least one W boson with a low boost, whose decay is resolved as two individual small-radius jets. Such events are not explicitly targeted by this analysis as their pro- duction rate is much smaller than the ones of the standard scenarios described above. This is the first resonance search of this kind in the all-hadronic final state. In the nonreso- nant form, as predicted by the SM, the WWW process has recently been observed in final states with at least two charged leptons [10,11]. In both cases, merged and resolved, dedicated techniques are applied to exploit the substructure of the W boson jets. For the merged case, apart from the case in which a nonisolated charged lepton overlaps with the hadronically decayed W boson, it is also possible that the hadronization products of one or more quarks from the fully hadronic decay R → WW → qqqq are not clustered into the same jet. Events identified as hadronically decaying W bosons can also include cases where the decay W → τν is followed by a hadronic decay of the tau lepton. These effects lead to a complicated jet mass spectrum from the merged radion that requires the design of a hybrid discriminant (“tagger”). Events with a single isolated charged lepton in the final state are considered in a similar, separate analysis with nonoverlapping event selection, described in Ref. [12]. While the search is by design optimized for a WWW signal, it is also partly sensitive to signals with similar decay topologies. In particular, heavy resonances decaying into WW, WZ, ZZ, WWZ, WZZ, ZZZ, Wt, Zt, WH, ZH, WX, or ZX, where X denotes an unknown particle with mass above 70 GeV whose decay products can be identified 012002-2 SEARCH FOR RESONANCES DECAYING TO THREE W … PHYS. REV. D 106, 012002 (2022) ¼ 6.708, and a KK gauge couplings gWKK confinement parameter ϵ ¼ 0.5. The branching fraction → RW → WWW can reach values for the decay WKK above 50%. ¼ 3 and gZKK Top quark pair and single top quark production are modeled at next-to-LO (NLO) using the POWHEG 2.0 generator [21–26]. Events composed uniquely of jets produced through the strong interaction are referred to as quantum chromodynamics (QCD) multijet events. These processes, along with background from W þ jets and Z þ jets production, are simulated at LO with MadGraph5_aMC@NLO, and matched to parton showers with less important the MLM [27] algorithm. The other, backgrounds, three vector bosons V ¼ W, Z (diboson and triboson produc- tion, respectively), are simulated at NLO with either POWHEG (WW production) or MadGraph5_aMC@NLO (all others). The simulation of t¯tW=Z events is performed at LO using MadGraph5_aMC@NLO. including processes with two or All background and signal samples for the 2016 data- taking conditions are generated with the NNPDF3.0 NLO or LO parton distribution functions (PDFs) [28], with the order matching that in the matrix element calculations. To the model processes in the 2017 and 2018 data sets, NNPDF3.1 next-to-next-to-LO PDFs [29] are used for all samples. Parton showering, fragmentation, and hadro- nization for all samples are performed using PYTHIA 8.230 [30] with the underlying event tune CUETP8M1 [31] for the 2016 analysis, and CP5 [32] for the 2017 and 2018 analyses. The CMS detector response is modeled using the GEANT4 package [33,34]. A tag-and-probe procedure [35] for data-to-simulation is used to derive corrections differences in reconstruction and selection efficiencies. The simulated events include additional pp interactions in the same and neighboring bunch crossings, referred to as pileup (PU). The simulated events are weighted so the PU vertex distribution matches the one from the data. IV. EVENT RECONSTRUCTION The candidate vertex with the largest value of summed physics object p2 T is taken to be the primary pp interaction vertex. The physics objects used for this determination are the jets, clustered using the anti-kT jet finding algorithm [36,37] with the tracks assigned to candidate vertices as inputs, and the associated missing transverse momentum ( ⃗pmiss ), taken as the negative vector sum of the pT of T those jets. A particle-flow (PF) algorithm [38] aims to reconstruct and identify each interacting particle in an event, with an optimized combination of information from the various elements of the CMS detector. The energy of electrons is determined from a combination of the track momentum at the primary interaction vertex, the corresponding ECAL cluster energy, and the energy sum of all bremsstrahlung photons attached to the track. The energy of muons is obtained from the curvature of the corresponding track. The energy of charged hadrons is determined from a combi- nation of their momentum measured in the tracker and the matching ECAL and HCAL energy deposits, corrected for the response function of the calorimeters to hadronic showers. Finally, the energy of neutral hadrons is obtained from the corresponding corrected ECAL and HCAL energies. For each event, hadronic jets are clustered from these reconstructed particles using the infrared and collinear safe anti-kT algorithm [36,37]. The clustering algorithm is run twice over the same inputs, once with a distance parameter of 0.4 (AK4 jets) and once with 0.8 (AK8 jets). Jet momentum is determined as the vectorial sum of all particle momenta in the jet, and is found from simulation to be, on average, within 5% to 10% of the true momentum over the entire pT spectrum and detector acceptance. Pileup interactions can contribute additional tracks and calorimetric energy depositions to the jet momentum. The pileup per particle identification algorithm (PUPPI) [39,40] is used to mitigate the effect of PU at the reconstructed particle level. Using this algorithm, the momenta of charged and neutral particles are rescaled. Jet energy corrections are derived from simulation to bring the measured response of jets to that of particle-level jets on average. In situ measurements of the momentum balance in dijet, photon þ jet, Z þ jet, and multijet events are used to account for any residual differences in the jet energy scale between data and simulation [41]. The jet energy resolution amounts typically to 15%–20% at 30 GeV, 10% at 100 GeV, and 5% at 1 TeV [41]. Additional selection criteria are applied to each jet to remove jets potentially dominated by anomalous contributions from various sub- detector components or reconstruction failures [42]. takes as input: Jets originating from the hadronization of b quarks are identified using a deep neural network algorithm (DeepCSV) that tracks displaced from the primary interaction vertex, identified secondary vertices, jet kin- ematic variables, and information related to the presence of soft leptons in the jet [43]. Working points (WPs) are used that yield either a 1% (medium WP) or a 10% (loose WP) probability of misidentifying a light flavor quark or a gluon (udsg) AK4 jet with pT > 30 GeV as a b quark jet. The corresponding average efficiencies for the identification of the hadronization products of a bottom quark as a b quark jet are about 70% and 85%, respectively. The vector ⃗pmiss is computed as the negative vector sum of the transverse momenta of all the PF candidates in an [44]. The ⃗pmiss event, and its magnitude is denoted as pmiss is modified to account for corrections to the energy scale of the reconstructed jets in the event. Anomalous high-pmiss events can be due to a variety of reconstruction failures, detector malfunctions, or noncollision backgrounds. Such events are rejected by event filters that are designed to T T T T 012002-3 A. TUMASYAN et al. PHYS. REV. D 106, 012002 (2022) identify more than 85%–90% of the spurious high-pmiss events with a mistagging rate of less than 0.1% [44]. T Hadronic decays of W=Z bosons are identified with the groomed jet mass (mj) and a novel deep learning algorithm with the PF candidates and secondary vertices as inputs [1]. The groomed jet mass is calculated after applying a modified mass-drop algorithm [45,46] to AK8 jets, with parameters β ¼ 0, zcut ¼ 0.1, and R0 ¼ 0.8. This algorithm is also known as the soft-drop algorithm [47]. The variables are calibrated in a top quark-antiquark sample enriched in hadronically decaying W bosons [48]. Further details on the calibration method used for this analysis are given in Sec. VII. Muon (μ) and electron (e) candidates are reconstructed in order to veto events containing such energetic leptons. Muon candidates are required to be within the geometrical acceptance of the muon detectors (jηj < 2.4) and are reconstructed by combining the information from the silicon tracker and the muon chambers [49]. These candi- dates are required to satisfy a set of quality criteria based on the number of hits measured in the silicon tracker and in the muon system, the properties of the fitted muon track, and the transverse and longitudinal impact parameters of the to the primary vertex of the event. track with respect Electron candidates within jηj < 2.5 are reconstructed using an algorithm that associates fitted tracks in the silicon tracker with electromagnetic energy clusters in the ECAL [50]. To reduce the misidentification rate, these candidates are required to satisfy identification criteria based on the shower shape of the energy deposit, the matching of the electron track to the ECAL energy cluster, the relative amount of energy deposited in the HCAL detector, and the consistency of the electron track with the primary vertex. Because of nonoptimal reconstruction performance, elec- tron candidates in the transition region between the ECAL barrel and end caps, 1.44 < jηj < 1.57, are discarded. Electron candidates identified as coming from photon conversions in the detector are also rejected. Identified muons and electrons are required to be isolated from hadronic activity in the event. The isolation sum is defined by summing the pT of all the PF candidates in a cone of radius ΔR ¼ ¼ 0.4ð0.3Þ around the muon (electron) track, and is corrected for the contribution of neutral particles from PU interactions [49,50]. ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðΔηÞ2 þ ðΔϕÞ2 p (ii) R3q, similar to the former but with one quark leaking outside of the jet cone, producing a three-prong jet (iii) Rlqq, where one of the two daughter W bosons decays leptonically (R → WW → lνqq), resulting in a jet containing an energetic, charged, nonisolated lepton All these types of R candidate jets are reconstructed as AK8 jets. Both W and R boson candidates are tagged using the DEEPAK8 jet classification framework [1]. This modular tagging framework has been designed by the CMS Collaboration to identify hadronically decaying top quarks as well as W, Z, and Higgs bosons. The algorithm uses machine learning techniques based on PF candidates, secondary vertices, and other inputs to classify the AK8 jets into 17 categories. These categories include jets arising from W → qq, Z → qq, t → bqq, H → 4q, and gluon or light-quark decay. To remove a potential mass dependence from the classifier output, a generative adversarial neural network is used to create “mass-decorrelated” outputs. The final output is a set of 17 “raw scores” per jet, where each one gives the likelihood of the jet originating from a particular decay. Discriminants have been developed by summing these raw scores and taking appropriate ratios to select particular types of jets, while rejecting others. Two particular discriminants are used for this analysis. The first, “deep-W,” aims to identify W boson candidates through the W → qq and QCD multijet raw scores, selecting and rejecting compatible jets, respectively. The second, “deep-WH,” is used to identify merged R boson candidates of types Rlqq, R3q, and R4q. This is achieved by making use of the W → qq and H → 4q raw scores, which select radionlike jet types while rejecting QCD multijet candidates. The corresponding formulas are as follows: deep-W ¼ raw scoreðW → qqÞ raw scoreðW → qqÞ þ raw scoreðQCDÞ ; ð1Þ used for tagging W boson candidates with mass mj in the range 60–100 GeV, and deep-WH ¼ r:s:ðW → qqÞ þ r:s:ðH → 4qÞ r:s:ðW → qqÞ þ r:s:ðH → 4qÞ þ r:s:ðQCDÞ ; ð2Þ V. MASSIVE JET TAGGING The signal event signatures include two types of massive jets (mj > 60 GeV) originating from the merged decay products of either W or R bosons. We consider three main cases for the merged R boson decay, designated and defined as follows: (i) R4q, where the two daughter W bosons decay hadronically (R → WW → qqqq) with all four fi- nal-state quarks contained in the reconstructed jet where “r.s.” denotes the raw score, used for tagging radion candidates with mass mj > 100 GeV. For both taggers, the mass-decorrelated version is used to avoid distorting the mass distribution (mass sculpting) and to retain the sensitivity to radions with mass greater than those of the W and Higgs bosons. The tagger discriminant distributions are presented in Fig. 2 (lower row) using a loose selection that will be defined in Sec. VI B. 012002-4 SEARCH FOR RESONANCES DECAYING TO THREE W … PHYS. REV. D 106, 012002 (2022) 138 fb (13 TeV) 1− , single t Multijet W+jets tt Other (VV, Z+jets) Wm Rm Wm = 0.2 TeV Wm = 1 TeV Wm Rm = 1.5 TeV, = 2 TeV, = 2 TeV, = 2.5 TeV, = 1 TeV Rm Rm = 0.3 TeV 310× CMS Simulation Preselection N = 2 j 1400 1200 1000 800 600 400 200 V e T 1 . 0 / s t n e v E 310× CMS Simulation Preselection N = 3 j 450 400 350 300 250 200 150 100 50 V e T 1 . 0 / s t n e v E 138 fb (13 TeV) 1− , single t Multijet W+jets tt Other (VV, Z+jets) Wm Rm Wm = 0.2 TeV Wm = 1 TeV Wm Rm = 1.5 TeV, = 2 TeV, = 2 TeV, = 2.5 TeV, = 1 TeV Rm Rm = 0.3 TeV 0 1 1.5 2 2.5 jjm 3 (TeV) 3.5 4 0 1 1.5 2 2.5 jjjm 3 3.5 4 (TeV) 138 fb (13 TeV) 1− , single t Multijet W+jets tt Other (VV, Z+jets) Wm Rm Wm = 0.2 TeV Wm = 1 TeV Wm Rm = 1.5 TeV, = 2 TeV, = 2 TeV, = 2.5 TeV, = 1 TeV Rm Rm = 0.3 TeV 138 fb (13 TeV) 1− , single t Multijet W+jets tt Other (VV, Z+jets) Wm Rm Wm = 0.2 TeV Wm = 1 TeV Wm Rm = 1.5 TeV, = 2 TeV, = 2 TeV, = 2.5 TeV, = 1 TeV Rm Rm = 0.3 TeV 310× CMS Simulation Preselection = 2 N j 700 600 500 400 300 200 100 V e G 0 1 / s t n e v E 0 50 100 150 maxm j 200 (GeV) 250 300 310× CMS Simulation Preselection = 3 N j 350 300 250 200 150 100 50 0 50 100 150 138 fb (13 TeV) 1− , single t Multijet W+jets tt Other (VV, Z+jets) Wm Rm Wm = 0.2 TeV Wm = 1 TeV Wm Rm = 1.5 TeV, = 2 TeV, = 2 TeV, = 2.5 TeV, = 1 TeV Rm Rm = 0.3 TeV 310× CMS Simulation Preselection = 2 N j maxm j > 100 GeV 1000 800 600 400 200 5 0 0 / . s t n e v E 0 0 0.2 0.4 200 (GeV) maxm j 250 300 138 fb (13 TeV) 1− , single t Multijet W+jets tt Other (VV, Z+jets) Wm Rm Wm = 0.2 TeV Wm = 1 TeV Wm Rm = 1.5 TeV, = 2 TeV, = 2 TeV, = 2.5 TeV, = 1 TeV Rm Rm = 0.3 TeV 310× CMS Simulation Preselection = 3 N j maxm j : 60-100 GeV 140 120 100 80 60 40 20 V e G 0 1 / s t n e v E 5 0 0 / . s t n e v E 0.6 max deep-WH 0.8 1 0 0 0.2 0.4 deep-W 0.6 max 0.8 1 , and deep-WH (for highest mass jet with mmax FIG. 2. Variables discriminating between signal and background in simulation. Left column, upper to lower rows: the distributions of j > 100 GeV) for preselected events with Nj ¼ 2. Right column, upper to lower mjj, mmax j < 100 GeV) for preselected events with Nj ¼ 3. rows: the distributions of mjjj, mmax The signal processes are scaled to 500 times their theoretical cross sections for visibility. , and deep-W (for highest mass jet with 60 < mmax j j 012002-5 A. TUMASYAN et al. PHYS. REV. D 106, 012002 (2022) VI. EVENT SELECTION A. Trigger The analysis uses events that are selected by a range of different HLT paths. One set of paths requires HT, the scalar sum of the pT of all AK4 jets in the event, to be greater than 800, 900, or 1050 GeV, depending on the data collection year. In addition, events with HT > 650 GeV and a pair of jets with invariant mass above 900 GeV and a pseudorapidity separation jΔηj < 1.5 are also selected for the 2016 dataset. A different set of paths selects events where the pT of the leading AK8 jet is greater than 500 GeV, or the pT is greater than 360 GeV and the “trimmed mass” of an AK8 jet is above 30 GeV. The jet trimmed mass is obtained after removing remnants of soft radiation with the jet trimming technique [51], using a subjet size parameter of 0.3 and a subjet-to-AK8 jet pT fraction of 0.1. The trigger selection efficiency is measured to be greater than 99% for events with HT > 1.1 TeV, using an independent sample of data events collected with a single-muon trigger. B. Preselection and signal region Events are selected in two stages; the first, “preselec- tion,” is initially applied to explore kinematic features of the signal compared to the SM background. A tighter selection, the signal region (SR) selection, is then applied to further improve the background rejection. The final analysis uses the SR events, while the preselected events are used to calibrate and validate the DEEPAK8 discriminants deep-W and deep-WH. In the following, we simply use the term “jets” to indicate massive AK8 jets if not stated differently. the conditions kinematic following define The preselection: (i) jet pj T > 200 GeV (ii) number of jets, Nj, exactly 2 or 3 (iii) highest pT jet pj1 T > 400 GeV (iv) mass of the two highest pT jets mj1;j2 > 40 GeV (v) no isolated lepton (Nl ¼ 0) with pl and jηlj < 2.4ð2.5Þ for μ (e). T > 20ð35Þ GeV The triboson signal is expected to show a peak in the distribution of the invariant mass of the jets, mjj in dijet events and mjjj in trijet events. These distribution are used for the statistical analysis. Figure 2 (upper row) shows the mjj (mjjj) spectra for signal and background after prese- lection. The signal processes are scaled to 500 times their theoretical cross sections. To define the SR selection, we add the following ¼ conditions to the preselection criteria. In the case of Nj 2 events, the higher and lower jet masses are designated as mmax , respectively. The higher-mass jet is taken to j be the radion candidate, and the lower-mass jet to be the W j > 70 GeV boson candidate. Therefore, we require mmax and 70 < mmin j < 100 GeV. In the case of events with and mmin j j j j ¼ 3, mmax ; mmax j and mmin are defined as above, with mmid Nj j designating the jet intermediate in mass. The two highest mass jets are considered as W boson candidates, Þ < 100 GeV. The low- and we demand 70 < ðmmid j < 100 GeV. est mass jet is required to have mass mmin This jet can correspond to either a merged W boson (60 < mmin j < 100 GeV) or to a single quark originating j < 60 GeV). Therefore, we from a W boson decay (mmin allow at most one of the three W bosons to be resolved into a pair of low-mass jets (mj < 60 GeV) with exactly one of the two daughter-quark jets required to have pT above the 200 GeV threshold. Figure 2 (middle row) shows the mmax distributions. further candidates, Jets in the mass range 60–100 (> 100) GeV, as W boson (radion) selected using the deep-Wðdeep-WHÞ discriminant. Figure 2 (lower row) presents the deep-Wðdeep-WHÞ distributions for the high- est mass jets after preselection. The conditions deep-W > 0.8 and deep-WH > 0.8 are required for events with two massive jets, while the less stringent requirement of at least two massive jets with deep-W > 0.6 is imposed for events with three jets. are j T T T In order to select Lorentz-boosted final states, we addi- tionally require that ST, the scalar sum of the transverse momenta of the selected jets and the pmiss , is greater than 1.3 TeV. The pmiss in the ST sum enhances signal separation for the cases where a hadronic τ lepton decay is present, or where the decay products from a merged radion decay include a nonisolated lepton, since in these cases the pmiss arises from the undetected neutrino(s). To suppress t¯t background, events are vetoed that contain a b-tagged AK4 jet not overlapping with any AK8 jet (ΔR > 0.8). The DeepCSV discriminant at the medium working point [43] is used for this veto. As the signal region explored corre- ≥ 1.5 TeV, we also impose the condition sponds to mWKK Þ > 1.1 TeV to probe only the high-mass region, mjj although this condition has minimal impact on top of the ST and HT constraints. While the selection requirements do not explicitly target the case where the lowest pT W boson is resolved into two single-quark jets, some of these events are accepted if only one of the two single-quark jets has pT > 200 GeV. ðmjjj The SR selection criteria, which are applied on top of preselection, can be summarized as (i) Number of additional b-tagged jets (nonoverlapping with the AK8 jets) Nb ¼ 0 (medium WP) the and the pT of (ii) Sum of pmiss T selected jets: ST > 1.3 TeV (iii) Dijet (trijet) invariant mass mjj ðmjjj Þ > 1.1 TeV for ¼ 2 (3) Nj (iv) For Nj ¼ 2: mmax j < 100 GeV, j > 70 GeV, 70 < mmin with deep-WðWHÞ > 0.8 for 70 < mj < 100 GeV (mj > 100 GeV) 012002-6 SEARCH FOR RESONANCES DECAYING TO THREE W … PHYS. REV. D 106, 012002 (2022) FIG. 3. Schematic of the 2D jet mass regions for two-jet events (left) and 3D jet mass regions for three-jet events (right), indicating the location of the six independent signal regions SR1–6, indicated by the colored areas. The SR4 and SR5 differ by the requirement of exactly three and two W-tagged jets, respectively. The jet tagging discriminants used in the event selection are also shown for each of the mass-ordered jets. The values in parentheses indicate that, depending on the SR, different selection requirements are employed. ¼ 3: 70 < ðmmax Þ < 100 GeV and j < 100 GeV, with deep-W > 0.6 (0.8) for three ; mmid j (v) For Nj mmin (two) massive jets. j C. Signal region definition j Six different SRs are defined in the following and are summarized in Table I. In addition, Fig. 3 illustrates these SRs in two-dimensional (2D) and 3D diagrams of the jet mass. ¼ 2 are split into three samples The SR events with Nj based on the value of mmax : SR1, SR2, and SR3 correspond values of 70–100, 100–200, and > 200 GeV, to mmax j respectively. This categorization serves as a binning over the unknown radion mass. As Fig. 2 (middle row) illus- trates, the merged radion jet mass has a broad distribution populating the mmax range of 70 GeV to mR. Signal events j j > 100 GeV) generally in SR2 and SR3 (i.e., with mmax contain a merged radion jet (Rlqq, R3q, R4q), and the deep-WH discriminant separates these jets from the SM background. Events in SR1 have both jets in the 70–100 GeV mass window. The merged radion jet lies in SR1 either for cases where the higher-mass jet is in the Rlqq category and the neutrino acquires most of the parent W boson momentum, or when the higher-mass jet is a W boson jet (when the decay products of R → WW receive imbalanced Lorentz boosts and the softer W boson is not merged). Resolved- radion events, i.e., events where the radion is reconstructed as two W boson jets, can lie in SR1 if the softest hadronically decaying W boson (typically the one produced promptly from the WKK decay) is not reconstructed as a single jet and therefore not selected as a candidate jet. In addition, SR1 is sensitive to any diboson resonant signal that might be present. Any jet of SR1–3 with a mass in the range 70–100 GeV is required to satisfy the deep-W > 0.8 requirement to be tagged as a W (or Rlqq) boson candidate. ¼ 3 are split into three samples SR4–6 as follows. In the case of mmin j > 60 GeV, we three jets to be W-tagged satisfying the demand all condition deep-W > 0.6, which defines the SR4 region; events with exactly two W-tagged jets are placed in SR5. j < 60 GeV and the other two massive jets Events with mmin satisfying 70 < ðmmid Þ < 100 GeV and deep-W > 0.8 constitute SR6. These three regions are sensitive only to the resolved-radion signal, with SR4 being the most sensitive among them, as it demands three W-tagged jets. The SR events with Nj ; mmax j j TABLE I. Summary of the selection requirements for each of the signal regions. Region SR1 SR2 SR3 SR4 SR5 SR6 Nj 2 2 2 3 3 3 mmax j (GeV) 70–100 100–200 > 200 70–100 70–100 70–100 mmid j (GeV) (cid:2) (cid:2) (cid:2) (cid:2) (cid:2) (cid:2) (cid:2) (cid:2) (cid:2) 70–100 70–100 70–100 (GeV) mmin j 70–100 70–100 70–100 60–100 60–100 0–60 Jet tagging conditions Both with deep-W > 0.8 Higher with deep-WH > 0.8, lower with deep-W > 0.8 Higher with deep-WH > 0.8, lower with deep-W > 0.8 All three with deep-W > 0.6 Exactly two with deep-W > 0.6 Two highest with deep-W > 0.8 012002-7 A. TUMASYAN et al. PHYS. REV. D 106, 012002 (2022) The six different regions provide complementary sensi- tivity to different regions of the mWKK-mR plane. Signal scenarios with radion masses producing jets in the mass range 100–200 GeV are predominantly probed in SR2; for mR in the range 200–300 GeV, the signal events predomi- nantly lie in SR2 and SR3; while for mR > 300 GeV (if the the radion remains merged), SR3 provides most of sensitivity. VII. CALIBRATION OF THE DEEPAK8 TAGGER The deep-Wðdeep-WHÞ discriminants are not fully reproduced in simulation, especially at low and high scores. In this section we describes the calibration procedure followed to correct the deep-Wðdeep-WHÞ spectra for each type of jet in two bins of pj T and mj. All types of jets involved in this procedure are illustrated in Table II. The correction is quantified using scale factors (SFs), which are applied to all simulated events (signal and background). Events in the preselected sample are dominated by QCD multijet background (99%). In SR1–6, QCD multijet events make up 50%–75% of the expected background. The rest of the events are from t¯t and single t quark processes (10%– 25%), W þ jets processes (10%–20%), and other processes (e.g., WW, WZ, ttW=Z, or tribosons, making up less than 15%). Therefore, massive jets (mj > 60 GeV) selected in jet the SRs are predominantly a mixture of different categories that we define as follows: (i) hadronically decaying W bosons producing merged W boson jets (ii) light quarks or gluons (q=g), with radiation or fragmentation, which are reconstructed as massive q=g jets (iii) three types of jets from hadronically decaying t quarks, t → bW → bqq: jets including the b quark and only one of the quarks from the W boson decay, designated “t2” jets including the b quark and both of the quarks from the W boson decay, designated “t3” same as “t3,” but requiring an additional energetic quark or gluon inside the jet cone to define a four-prong category, designated “t4” For the t4 category, the additional q=g inside the jet cone needs to have pT > 50 GeV. By considering t3 and t4 jets separately, they can be compared directly to signal jets of similar jet substructure (as discussed in Sec. VII C) and systematic uncertainties can be derived as discussed in Sec. IX C. For the calibration in data, these categories are difficult to distinguish experimentally and their tagger response is similar. Thus, t3 and t4 jets are treated together and designated t3;4 in the following. In simulation, jets are placed into these categories, as well as signal categories, by matching the reconstructed jets to the generator-level partons in ΔR. The matching criteria are summarized in Table II. The proportion of jets not matched to any of these categories, is less than 6 (5)% of the SR (preselection) events, and they have a negligible impact on the analysis. The calibration of the W, t2, and t3;4 jets requires samples enriched in those jets. Therefore, dedicated calibration samples are defined, and the calibration for these jets is summarized in Sec. VII A. The q=g jets are calibrated using preselection jets, and this procedure is described in Sec. VII B. The calibration of signal jets is presented in Sec. VII C. A. Calibration of W boson and top quark jets with a matrix method For the calibration of the taggers, a control sample similar to the preselected one, but enriched in W boson and top quark jets, is used. We refer to this sample as the TABLE II. Matching criteria used to place a jet in one of the SM jet categories (left four columns) or merged radion jet categories (right two columns). Each column lists the ΔR conditions demanded between the reconstructed jet (j) and the generator-level parton in order to match a jet with a particular jet substructure. Lower indexes enumerate partons and indicate the particle from whose decay they originate (e.g., t → btq1Wq2W). Schematic diagrams for each jet type are shown below each column. q=g ðq=g; jÞ < 0.6 (cid:2) (cid:2) (cid:2) (cid:2) (cid:2) (cid:2) (cid:2) (cid:2) (cid:2) (cid:2) (cid:2) (cid:2) W ðW; jÞ < 0.6 ðq1W; jÞ < 0.8 ðq2W; jÞ < 0.8 ðbt; jÞ > 0.8 (cid:2) (cid:2) (cid:2) t2 ðt; jÞ < 0.6 ðbt; jÞ < 0.8 ðq1W; jÞ < 0.8 ðq2W; jÞ > 0.8 (cid:2) (cid:2) (cid:2) t3;4 ðt; jÞ < 0.6 ðbt; jÞ < 0.8 ðq1W; jÞ < 0.8 ðq2W; jÞ < 0.8 For t4 (t3): ðq=g; jÞ < ð>Þ0.8 R3;4q ðR; jÞ < 0.6 ðq1; jÞ < 0.8 ðq2; jÞ < 0.8 ðq3; jÞ < 0.8 For R4q (R3q) : ðq4; jÞ < ð>Þ0.8 Rlqq ðR; jÞ < 0.6 ðq1; jÞ < 0.8 ðq2; jÞ < 0.8 ðl; jÞ < 0.8 (cid:2) (cid:2) (cid:2) 012002-8 SEARCH FOR RESONANCES DECAYING TO THREE W … PHYS. REV. D 106, 012002 (2022) “sideband.” The sideband is defined by requiring one isolated lepton (μ or e), pmiss T > 40ð80Þ GeV for μ (e), and one or two massive jets. The neutrino pz is recon- structed under the assumptions that the invariant mass of the lν system is equal to the W boson mass mW ¼ 80 GeV (as described in Ref. [52]) and the transverse momentum of lν > 200 GeV. This the lν system is required to satisfy pT means that for the sideband one of the massive jets used for the preselection is effectively replaced by a leptonically decaying W boson candidate. The highest mass jets in these sideband events with mj > 60 GeV are used for the calibration. These jets are categorized by the matching to the W, t2, t3;4, and q=g categories described previously. We split the events into two mj bins, one with 60 < mj < 120 GeV (low mass) and the other one with mj > 120 GeV (high mass). In addi- tion, we split the sideband events further into two bins, based on the jet pj T. For low-mass jets, the bins used are 200–400 and > 400 GeV, while for high-mass jets the bins used are 200–500 and > 500 GeV. The resulting four samples are designated LL, LH, HL, and HH, where the first letter indicates low or high mj and the second letter low or high pj T. The two SM processes (W þ jets and top quark production) are normalized in each of these four categories by scaling them to match the data separately for events with zero or one b-tagged jet(s). This corrects the simulation for an Oð10%Þ mismodeling of the cross section, and the residual data-to-simulation differences in the deep-Wðdeep-WHÞ distribution can be attributed to the mismodeling of the discriminant. The LL and LH samples contain primarily events of the W, t2, and q=g jet categories; the HL and HH samples primarily events of the t2, t3;4, and q=g jet categories. Any other jet contribution or unmatched jets (collectively < 5%) can be ignored. For each of these jet types, we apply a set of kinematic conditions to split into three subsamples so that each sample is highly pure in a single jet type. The splitting conditions include kinematic varia- bles such as N-subjettiness [53], Nj, Nb, and mj, as well as DEEPAK8 discriminants other than the calibrated one. them further The deep-Wðdeep-WHÞ distributions are formed for each of the three pure subsamples for LL. One equation is written for each pure subsample by equating the data yields Di;k for a jet type i in a deep-Wðdeep-WHÞ bin k with the simulated jet yields for W, t2, and q=g (which we write as W, t, and g here), scaled by the scale factors SFW k , kgi;k þ di;k. SFt The di;k term accounts for the other types of jet yields; their contribution is small (amounting to < 5% for most of the bins), and these jet types are treated as not contributing to the mismodeling. A similar equation can be written for each of the three (i ¼ 1, 2, 3) subsamples W, t, and g to form a system of three equations: k: Di;k ¼ SFW k Wi;k þ SFt k, and SFg kti;k þ SFg 0 B @ D1;k − d1;k D2;k − d2;k D3;k − d3;k 1 0 C A ¼ B @ W1;k W2;k W3;k t1;k t2;k t3;k g1;k g2;k g3;k 1 0 C A B @ 1 C A; SFW k SFt k SFg k ð3Þ k type, k , SFt k, SFg k ; SFt2 in which the jet yields and the data are known, while the three SFs (SFW k) are unknown. We solve this 3 × 3 system per deep-Wðdeep-WHÞ bin k to derive the SFs for each type of jet. The scale factors obtained with this k and SFt3;4 matrix method SFW are shown in Fig. 4 for LL, LH, HL, and HH. As the three subsamples are highly enriched in exactly one jet the matrix is nearly diagonal, and the derivation of the SFs is dominated by the data vs simulation modeling in the corresponding pure subsamples. For example, the data/simulated yields in the W-pure subsample dominate the determination of SFW k . The method yields reliable SFs in the regime where subsamples are highly pure. Both deep-W and deep-WH are calibrated with this procedure for each of the LL, LH, HL, and HH bins separately. While the SFs are quite large and vary from about 0.5 to 3, the integral over the tagger score yields an effective SF close to 1; for example, W-boson jets with deep-W > 0.6 ð0.8Þ have effective SFs of 0.89 (0.78) and 0.80 (0.74) for the LL and LH samples, respectively. All simulated events, based on the types of the selected jets they contain and their pj T and mj, are corrected by the SFs for the respective deep-W (deep-WH) bins. The discriminant distributions before and after corrections are shown in Fig. 5. Various validation tests show good agreement between data and simulation. As the extracted SFs are found to depend on the choice of splitting conditions defining the pure subsamples, systematic uncer- tainties resulting from the selection criteria are assigned, as described in Sec. IX. B. Calibration of quark and gluon jets The quark and gluon jets are treated collectively as a single type of jet, q=g. Their calibration is performed using the preselected sample where SR events and events with b- tagged AK4 jets are vetoed. This sample consists of more than 13 million events, of which more than 97% are QCD multijet events. Similarly to the single-lepton sideband sample, we consider only the highest mass jet with mj > 60 GeV in each event, and define the same four LL, LH, HL, and HH bins in mj and pj T. The QCD events in each bin are normalized to the data. The contribution from W, t2, and t3;4 jets, amounting to less than 2%, is estimated using simulation and subtracted from the data. The result is divided by the q=g yields to define SFq=g in each deep-Wðdeep-WHÞ discriminant value bin k. The resulting values of SFq=g are presented together with SFW, SFt2 , and SFt3;4 in Fig. 4. The relative fraction of quarks and gluons is the same for the preselection region where the SFs are k 012002-9 A. TUMASYAN et al. CMS 3 138 fb (13 TeV) 1− PHYS. REV. D 106, 012002 (2022) CMS 3 138 fb (13 TeV) 1− 2.5 2 F S 1.5 1 0.5 0 tSF +Stat. unc. Stat. + PS unc. tSF +Stat. unc. Stat. + PS unc. WSF +Stat. unc. Stat. + PS unc. 2.5 WSF +Stat. unc. Stat. + PS unc. q/g SF +Stat. unc. Stat. + PS unc. q/g SF +Stat. unc. Stat. + PS unc. jm60 < j 200 < p T < 120 GeV < 400 GeV 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 deep-W 1 2 F S 1.5 1 0.5 0 < 120 GeV jm60 < j p T > 400 GeV 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 deep-W 1 CMS 3 138 fb (13 TeV) 1− CMS 3 138 fb (13 TeV) 1− tSF +Stat. unc. Stat. + PS unc. tSF +Stat. unc. Stat. + PS unc. 2.5 tSF +Stat. unc. Stat. + PS unc. 2.5 tSF +Stat. unc. Stat. + PS unc. q/g SF +Stat. unc. Stat. + PS unc. q/g SF +Stat. unc. Stat. + PS unc. 2 F S 1.5 1 0.5 0 > 120 GeV jm 200 < p < 500 GeV j T 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 deep-WH 1 2 F S 1.5 1 0.5 0 jm j p > 120 GeV > 500 GeV T 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 deep-WH 1 FIG. 4. Measured scale factors (SFs) for the deep-W and deep-WH discriminants. Upper row: SFs for W (dark blue), t2 (light blue), and q=g (yellow) matched jets in the low-mj bins, LL (left) and LH (right), as functions of the deep-W discriminant value. Lower row: SFs for t2 (light blue), t3;4 (green), and q=g (yellow) matched jets in the high-mj bins, HL (left) and HH (right), as functions of the deep-WH discriminant value. For each discriminant value bin, the sum of the SF-corrected jet yields is required to be equal to the observed data. The statistical and parton shower (PS) uncertainties are shown by the shaded bands. defined, and the SRs and control regions (CRs). The only difference between the jets is therefore their pT spectra. Validation tests have shown a good post-correction per- formance, where the ratio of data to simulation is consistent with unity over the entire deep-Wðdeep-WHÞ range. To perform these tests, we define CRs by using the SR1–6 least one of the deep-Wðdeep-WHÞ selections with at conditions inverted. For SR4 and SR5, this inversion leads to the same sample with zero or one W-tagged jet. The five resulting CRs, associated with the SRs, are named CR1, CR2, CR3, CR45, and CR6. Figure 6 shows the deep-Wðdeep-WHÞ distributions for the highest mass jet in each CR after the SFs are applied. A similar, almost flat performance is exhibited by the middle and minimum mass jets. Validation tests in samples using other CR definitions lead to similar post-correction performance. 012002-10 SEARCH FOR RESONANCES DECAYING TO THREE W … PHYS. REV. D 106, 012002 (2022) 138 fb (13 TeV) 1− CMS 138 fb (13 TeV) 1− CMS 60 < jm < 120 GeV jp 200 < T < 400 GeV Data Total uncorrected Total SF-corrected 3,4t 2t W q/g Rest Uncorrected SF-corrected SF uncertainty 0.1 0.2 0.3 0.4 0.6 0.5 deep-W 0.7 0.8 0.9 1 CMS jm 200 < > 120 GeV jp T < 500 GeV 138 fb (13 TeV) 1− Data Total uncorrected Total SF-corrected 3,4t 2t W q/g Rest Uncorrected SF-corrected SF uncertainty 0.1 0.2 0.3 0.4 0.5 0.6 deep-WH 0.7 0.8 0.9 1 5 0 . 0 / s t n e v E 50000 40000 30000 20000 10000 m s i / a t a D 0 1.5 1 0.5 0 8000 7000 6000 5000 4000 3000 2000 1000 5 0 . 0 / s t n e v E m s i / a t a D 0 1.5 1 0.5 0 5 0 . 0 / s t n e v E 9000 8000 7000 6000 5000 4000 3000 2000 1000 m s i / a t a D 0 1.5 1 0.5 0 3000 2500 2000 1500 1000 500 5 0 . 0 / s t n e v E m s i / a t a D 0 1.5 1 0.5 0 60 < jp T jm < 120 GeV > 400 GeV Data Total uncorrected Total SF-corrected 3,4t 2t W q/g Rest Uncorrected SF-corrected SF uncertainty 0.1 0.2 0.3 0.4 0.6 0.5 deep-W 0.7 0.8 0.9 1 CMS 138 fb (13 TeV) 1− > 120 GeV > 500 GeV jm jp T Data Total uncorrected Total SF-corrected 3,4t 2t W q/g Rest Uncorrected SF-corrected SF uncertainty 0.1 0.2 0.3 0.4 0.5 0.6 deep-WH 0.7 0.8 0.9 1 FIG. 5. DEEPAK8 discriminants of the jet with highest mass in the single-lepton sideband. The deep-W spectra in the LL (upper left) and LH (upper right) samples are presented together with the deep-WH spectra in the HL (lower left) and HH (lower right) samples. The W boson jets are shown in dark blue, t2 in light blue, t3;4 in green, q=g in yellow, and the “Rest” jet types (jets not matching any of the categories) in gray. Before corrections (red), discrepancies between the prediction and the data can be observed, in particular at low and high discriminant values. The corrected distributions after application of the scale factors (SFs) are shown in dark green. The lower panels show the data-to-simulation ratios before and after corrections. The SF uncertainties are indicated by the shaded bands. C. Calibration of signal jets with SM proxy jets The deep-Wðdeep-WHÞ discriminant distributions for simulated signal events are also corrected using SFs. For ¼ 3, SR4–6), the W- resolved-radion signal events (Nj boson-matched jets are scaled by SFW according to the pj T, mj, and deep-W values for each jet. Merged radion signal events (Nj ¼ 2, SR1–3) contain jets of the form W, Rlqq, R3q, and R4q. Figure 7 (left) shows the relative contributions of each of these categories to the total as a function of mmax . There are very few SM jets with the same substructure and flavor compositions as Rlqq, R3q, and R4q jets that can be directly used for calibration j 012002-11 A. TUMASYAN et al. PHYS. REV. D 106, 012002 (2022) 14000 12000 10000 8000 6000 4000 2000 5 0 . 0 / s t n e v E 0 1.5 1 0.5 0 m s i / a t a D 60000 50000 40000 30000 20000 10000 5 0 . 0 / s t n e v E 0 1.5 1 0.5 0 m s i / a t a D 35000 30000 25000 20000 15000 10000 5000 5 0 . 0 / s t n e v E 0 1.5 1 0.5 0 m s i / a t a D 138 fb (13 TeV) 1− 138 fb (13 TeV) 1− CMS CR1 multijet scaled by 0.60 Data Multijet W+jets tt , single t Other (VV, Z+jets) 0.1 0.2 0.3 0.5 0.4 deep-W 0.6 max 0.7 0.8 0.9 1 2500 2000 1500 1000 500 5 0 . 0 / s t n e v E 0 1.5 1 0.5 0 m s i / a t a D CMS CR45 multijet scaled by 0.98 Data Multijet W+jets tt , single t Other (VV, Z+jets) 0.1 0.2 0.3 0.5 0.4 deep-W 0.6 max 0.7 0.8 0.9 1 138 fb (13 TeV) 1− 138 fb (13 TeV) 1− CMS CR2 multijet scaled by 0.61 Data Multijet W+jets tt , single t Other (VV, Z+jets) 0.1 0.2 0.3 0.5 0.4 0.6 max deep-WH 0.7 0.8 0.9 1 10000 CMS CR6 multijet scaled by 0.91 5 0 . 0 / s t n e v E 8000 6000 4000 2000 0 1.5 1 0.5 0 m s i / a t a D 0.1 0.2 0.3 0.5 0.4 deep-W 0.6 max Data Multijet W+jets tt , single t Other (VV, Z+jets) 0.7 0.8 0.9 1 138 fb (13 TeV) 1− CMS CR3 multijet scaled by 0.70 Data Multijet W+jets tt , single t Other (VV, Z+jets) 0.1 0.2 0.3 0.4 0.5 0.7 0.8 0.9 1 deep-WH 0.6 max FIG. 6. Comparison between data (black markers) and simulated background events (histograms) of the deep-Wðdeep-WHÞ distributions for the highest mass jet after SF application. The control regions CR1, CR2, CR3 are shown in the left column, upper to lower rows, while CR45 and CR6 are presented in the right column, upper and middle rows, respectively. The lower panels show the data-to-simulation ratio. 012002-12 SEARCH FOR RESONANCES DECAYING TO THREE W … PHYS. REV. D 106, 012002 (2022) 138 fb (13 TeV) 1− = 2.5 TeV, Rm = 0.2 TeV Total 4q R 3q R lqq R W Rest CMS Simulation SR1+SR2+SR3 Wm 60 50 40 30 20 10 V e G 0 1 / s t n e v E 0 50 100 150 200 max jm 250 (GeV) 300 350 400 138 fb (13 TeV) 1− CMS Simulation SR1+SR2+SR3 Wm = 2.5 TeV, Rm = 0.2 TeV . u . a . u . a 4q R 3q R lqq R W 3,4t 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 deep-W 1 138 fb (13 TeV) 1− CMS Simulation SR1+SR2+SR3 Wm = 2.5 TeV, Rm = 0.2 TeV 4q R 3q R lqq R W 3,4t 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 deep-WH 1 j the mmax distributions FIG. 7. Shape comparison for different jet types in simulation. for SR1–3 events without Upper: deep-Wðdeep-WHÞ constraints. Middle and lower: the deep-W and deep-WH distributions normalized to unity for the shown components, respectively. The t3;4 jets from the preselected sample, normalized to unity, are superimposed to compare shapes with the R3q and R4q distributions. (considering that boosted Higgs bosons are not abundant). Instead, we calibrate these using SM jets that have similar prong substructure and deep-Wðdeep-WHÞ response, which we call “proxy jets.” The W boson jets exhibit highly similar deep-W and deep-WH distributions to Rlqq jets. Thus, we use W boson jets as proxy jets for the Rlqq calibration. The similarity of the two spectra can be seen in Fig. 7 for both the deep-W (used in SR1) and deep-WH (used in SR2–3) discrimi- nants. This similarity results from the discriminant design, as the raw scores in both the numerator and denominator have not been derived for events with leptons inside jets, and so the deep-W and WH discriminants are largely blind to the presence of a lepton. The closest abundant SM jets with substructure similar to R3q and R4q are fully merged top quark jets t3;4. As Fig. 7 (lower) shows, the deep-WH distributions of those three jet types are similar and thus the t3;4 jets are used as proxy jets to calibrate signal R3q and R4q jets. Accordingly, the corresponding SFt3;4 values derived in Sec. VII A are used to calibrate R3q and R4q jets. We find that the individual t3 and t4 components have an even better shape agreement with their corresponding signal jets R3q and R4q, respec- tively. This consistency suggests that despite their differences (in quark flavor, kinematics, and color recom- bination), the t3;4 and R3q;4q jets have a largely similar response to the deep-WH discriminant. Systematic uncer- tainties are assigned to account for differences in this response and also to account for residual shape differences as discussed in Sec. IX. VIII. BACKGROUND ESTIMATION The dominant background in all SRs consists of QCD multijet events, making up 60%–80% of the total. As the DEEPAK8 tagger rejects the majority of these events, only a few of them remain in the SRs according to simulation. Thus, we estimate this background contribution directly from the data using CRs. The five CRs are defined by inverting at least one tagger condition, as described in Sec. VII B. The selected jets in these regions possess similar kinematic properties to the ones in the correspond- ing SRs. The mjj (mjjj) distributions in CRs 1–3 (4–6) are shown in Fig. 8, where the SF-corrected simulation is normalized to the data. After subtracting the other back- ground processes estimated from simulated samples from the data, the resulting mjj (mjjj) distributions are used to predict the shape of the QCD multijet background in the corresponding SRs. This shape compatibility has been validated in simulation in multiple selections, and the mjj (mjjj) distributions agree within the statistical uncer- tainties over the entire spectra. The a priori normalization in the SRs is taken from the SF-corrected simulation. All other smaller background contributions such as W þ jets, 012002-13 A. TUMASYAN et al. PHYS. REV. D 106, 012002 (2022) V e T 1 . 0 / s t n e v E 18000 16000 14000 12000 10000 8000 6000 4000 2000 0 2 1.5 1 0.5 0 m s i / a t a D V e T 1 . 0 / s t n e v E 90000 80000 70000 60000 50000 40000 30000 20000 10000 0 2 1.5 1 0.5 0 m s i / a t a D V e T 1 . 0 / s t n e v E 35000 30000 25000 20000 15000 10000 5000 0 2 1.5 1 0.5 0 m s i / a t a D 138 fb (13 TeV) 1− CMS CR1 multijet scaled by 0.60 Data Multijet W+jets tt , single t Other (VV, Z+jets) 1.5 2 2.5 3 jj (TeV) m 3.5 4 138 fb (13 TeV) 1− CMS CR2 multijet scaled by 0.61 Data Multijet W+jets tt , single t Other (VV, Z+jets) V e T 1 . 0 / s t n e v E m s i / a t a D 2500 2000 1500 1000 500 0 2 1.5 1 0.5 0 138 fb (13 TeV) 1− CMS CR45 multijet scaled by 0.98 Data Multijet W+jets tt , single t Other (VV, Z+jets) 1.5 2 2.5 3 jjjm (GeV) 3.5 4 138 fb (13 TeV) 1− Data Multijet W+jets tt , single t Other (VV, Z+jets) V e T 1 . 0 / s t n e v E 10000 CMS CR6 multijet scaled by 0.91 8000 6000 4000 2000 1.5 2 2.5 3 (TeV) jjm 3.5 4 0 2 1.5 1 0.5 0 m s i / a t a D 1.5 2 2.5 3 jjjm (GeV) 3.5 4 138 fb (13 TeV) 1− CMS CR3 multijet scaled by 0.70 Data Multijet W+jets tt , single t Other (VV, Z+jets) 1.5 2 2.5 3 (TeV) jjm 3.5 4 FIG. 8. Invariant mass distributions of the reconstructed triboson systems for control regions in data (black markers) and simulated events (histograms). The mjj distributions for CR1, CR2, CR3 are presented in the left column, upper to lower rows, respectively; the mjjj distributions for control regions CR45 and CR6 are presented in the right column, upper and middle rows, respectively. The simulation is corrected by SFs, and the QCD multijet background is scaled to the data yields. 012002-14 SEARCH FOR RESONANCES DECAYING TO THREE W … PHYS. REV. D 106, 012002 (2022) t¯t, single t quark, diboson, t¯tV, and triboson production are taken from simulation. IX. SYSTEMATIC UNCERTAINTIES Systematic uncertainties are taken into account in the background estimation and the signal prediction. For each source of uncertainty, a nuisance parameter is assigned, which is constrained by the data in the six SRs. These are summarized in Table III. A. Systematic uncertainties in the scale factor estimation Systematic uncertainties in the signal and background rate and shape arise from the DEEPAK8 SF derivation. Two uncertainty sources common to signal and background jets are considered, and an additional two only for signal, which are described in Sec. IX C. The common uncertainties are from parton shower variations, and the SF dependence on the jet subsample selection, referred to as the “selection bias” uncertainty in Table III. The SFs are derived using three different t¯t simulation samples: the nominal sample is generated using POWHEG with PYTHIA8, a second one using POWHEG with HERWIG7 [54], and a third one using MadGraph5_aMC@NLO with PYTHIA8. The maximum difference of the three resulting SFs is symmetrized and assigned as the parton shower uncertainty for the W, t2, and t3;4 SFs. For the q=g SFs, the parameters controlling the parton shower behavior in the QCD multijet PYTHIA sample are varied to derive an uncertainty. The resulting uncertainty bands are shown in Fig. 4, combined with the significantly smaller statistical uncertainty. The bias in the SF calculation due to the selection conditions defining the jet subsample is estimated by performing closure tests in several validation regions such as jet mass sidebands. The maximum nonclosure observed amounts to 10% for W, t3;4, and q=g jets. Because of the limited numbers of events in the validation regions for t2 jets and for jets not matching any of these categories, a 100% uncertainty is assigned to those. Uncertainties in the parton shower modeling and those arising from the selec- tion bias are added in quadrature, and are assigned a single nuisance parameter for each matched jet in each LL, LH, HL, or HH bin. The per-jet variation is treated as fully correlated. Effects on both rate and shape of the mjj (mjjj) distributions are considered. The overall rate uncertainties due to this variation amount to about 35% for SRs 1–3 and SR6, 52% for SR4, and 45% for SR5. These values are TABLE III. Sources of systematic uncertainties accounted for in the analysis. The first three sets of uncertainty sources originate from the tagger calibration. It is also indicated whether the uncertainties are evaluated for background (B) and/or signal (S), whether the uncertainty affects shape and/or rate, and the total number of nuisance parameters used per source. Sources Parton shower þ selection bias Parton shower þ selection bias for W, Rlqq for t2 Parton shower þ selection bias for t3;4, R3q;4q Parton shower þ selection bias for q=g Proxy uncertainty for Rlqq Proxy uncertainty for R3q;4q Proxy uncertainty for unmatched High-pT extrapolation for W High-pT extrapolation for Rlqq High-pT extrapolation for R3q High-pT extrapolation for R4q QCD multijet normalization t¯t normalization Other background normalization mjj, mjjj tail shape t¯t shape Pileup and integrated luminosity PDFs, renormalization and factorization scales Jet energy scale and resolution Jet mass scale B or S B þ S Effect on Shape þ rate Magnitude (cid:2) (cid:2) (cid:2) Nuisance parameters 4 for deep-Wðdeep-WHÞ × LL, LH (cid:2) (cid:2) (cid:2) (cid:2) (cid:2) (cid:2) (cid:2) (cid:2) (cid:2) 10%–35% 12%–43% 100% 100% 23%–30% 16%–34% 24%–33% 5%–40% 15%–30% 30% (cid:2) (cid:2) (cid:2) (cid:2) (cid:2) (cid:2) 1.7% 1.4% (cid:2) (cid:2) (cid:2) (cid:2) (cid:2) (cid:2) B Shape þ rate B þ S Shape þ rate B S S S S S S S B B B B B S S S S Shape þ rate Rate Rate Rate Rate Rate Rate Rate Rate Rate Rate Shape Shape Rate Rate Shape Shape 012002-15 2ðþ4Þ for deep-Wðdeep-WHÞ LL, LH (LL, …, HH) 4 for deep-Wðdeep-WHÞ × HL, HH 2ðþ4Þ for deep-Wðdeep-WHÞ LL, LH (LL, …, HH) 2, for deep-Wðdeep-WHÞ 2, for deep-Wðdeep-WHÞ 2, for deep-Wðdeep-WHÞ 2, for deep-Wðdeep-WHÞ 2, for deep-Wðdeep-WHÞ 2, for deep-Wðdeep-WHÞ 2, for deep-Wðdeep-WHÞ 5, common for SR4,5 5, common for SR4,5 5, common for SR4,5 6, one for each SR 6, one for each SR 1, common for all SRs 1, common for all SRs 2, common for all SRs 1, common for all SRs A. TUMASYAN et al. PHYS. REV. D 106, 012002 (2022) driven by the SF uncertainty on q=g jets, which constitute 75%–90% of the highest mass jets in the SRs. B. Systematic uncertainties in the background estimation For the shape of the dominant QCD multijet background, we account for an additional uncertainty in the tail shape of the mjj (mjjj) distributions. This uncertainty is derived in the CRs by comparing the QCD multijet prediction in simu- lation to the data. A linear fit is performed to the ratio of the data and the simulation. The resulting (cid:3)2 standard deviation bands are used as shape variations of the mjj (mjjj) distributions in the SRs. A single nuisance parameter with a Gaussian prior is used for each SR. This shape uncertainty allows the tails of the distributions to be adjusted by the data, accounting for effects that could lead to differences between CRs and SRs, e.g., a potential residual mass correlation of the taggers. The uncertainty in the normalization of the QCD multijet background is taken as the normalization difference between data and SF-corrected simulation in the corre- sponding CRs. These differences range from 9% to 40% for SRs 1–3 and SR6, and 5% for SR4–5. For the top quark production rate, uncertainties in the normalization to NNLO and NLO predictions and missing higher orders are accounted for and are in the range 15%–30%. In addition, uncertainties in the t¯t shape are derived by varying the top quark pT spectrum based on the measurements in Refs. [55,56]. For the other background processes, which are treated collectively, a 30% normalization rate uncer- tainty is assigned for all SRs. Because of their similarity, the same normalization nuisance parameters are used for SRs 4 and 5. All rate uncertainties are estimated using a log-normal prior. C. Signal systematic uncertainties The integrated luminosities of the 2016, 2017, and 2018 data-taking periods are individually known with uncertain- ties in the 1.2%–2.5% range [17–19], while the total 2016– 2018 integrated luminosity has an uncertainty of 1.6%. The simulated PU distribution is scaled to match data using an effective total inelastic cross section of 69.2 mb. The uncertainty in this procedure is evaluated by varying the total inelastic cross section by (cid:3)4.6% [57]. This results in a 0.5% uncertainty in the signal normalization in the SRs, which is combined with the integrated luminosity uncer- tainty for a total uncertainty of 1.7%, implemented with a log-normal prior. Renormalization μ R and factorization μ F scales and PDF uncertainties affecting the signal selection efficiency are evaluated per SR and mass point. The scale uncertainties are obtained by varying μ R and μ F independently by factors of 1=2 and 2 (without considering the extreme cases of the opposite-direction variations). The maximum value of these variations is taken as the prefit uncertainty. For the overall scale uncertainty, a single nuisance parameter is used. Its typical magnitude is up to 1.4% for signal with mWKK ≤ 4 TeV. The jet energy scale is varied by its uncertainty and the impact on the mjj (mjjj) distributions is taken to be the associated shape uncertainty. Similarly, for the uncertain- ties in the jet energy and jet mass resolution, shape uncertainties are considered by varying the jets selected three uncertainty by the respective uncertainties. All sources are implemented as nuisance parameters using Gaussian priors. All of the above signal uncertainties only have a small impact on the final result. The largest signal uncertainty originates from the DEEPAK8 tagger SF correction procedure. Four different uncertainty sources are considered for the SFs applied to the signal jets. The first two uncertainty sources are the parton shower and selection bias, and have only a small impact. They are evaluated together with those of the background processes (described in Sec. IX A), using common nuisance parameters for signal and background jets. Signal jets categorized as W, Rlqq (R3q, R4q) are assigned the same nuisance parameters as their correspond- ing proxy jets W (t3;4) and are constrained using the data in the SRs. The other two sources of SF uncertainty, described below, are due to the differences between signal and proxy jets (proxy uncertainty), and due to the significantly higher pT that signal jets have compared to the SM jets (high-pT extrapolation uncertainty). Varying the SFs within these uncertainties has a major effect on the signal rates. Although the signal jets share similar substructures with the corresponding SM proxies and also have similar deep-Wðdeep-WHÞ distributions, they are, with the excep- tion of W boson jets, not the same objects. For example, the flavor of the most energetic quarks might differ, the color flow structure might not be the same, and overall jet substructure kinematic properties could be different. To account for all these differences, the shape difference of proxy and signal jets in six deep-Wðdeep-WHÞ bins above the 0.7 discriminant selection value (0.7–1.0 in 0.05 bins) is evaluated in simulation. For each of these six bins, the relative difference between the proxy and the signal jets is jets categorized as taken as an uncertainty. For signal R3q;4q, for which the corresponding proxy jet category is t3;4, an additional uncertainty due to the difference observed between t3 and t4 is assigned. It amounts to 5% and 10% for the deep-WH and deep-W discriminants, respectively. The total resulting proxy uncertainties for Rlqq, R3q, and R4q signal jets lie in the ranges 10%–35%, 13%–34%, and 12%–43%, respectively. This source of uncertainty has the largest effect on the rate for the merged signal. Signal jets not matching any of these categories are assigned a 100% proxy uncertainty. The proxy uncertainty 012002-16 SEARCH FOR RESONANCES DECAYING TO THREE W … PHYS. REV. D 106, 012002 (2022) 138 fb 1(cid:2) (13 TeV) 138 fb 1(cid:2) (13 TeV) CMS SR1 Data Multijet tt , single t Other (VV, V+jets) Systematic uncertainty Wm = 2.5 TeV, Wm = 2.5 TeV, Rm Rm = 0.2 TeV = 1.25 TeV (Data-Prediction)/ (cid:3) STAT 1.5 2 3 2.5 (TeV) jj m (cid:3)(cid:4) (cid:3)/SYS 3.5 STAT CMS SR2 138 fb 1(cid:2) (13 TeV) Data Multijet tt , single t Other (VV, V+jets) Systematic uncertainty Wm = 2.5 TeV, Wm = 2.5 TeV, Rm Rm = 0.2 TeV = 1.25 TeV (Data-Prediction)/ (cid:3) STAT 1.5 2 2.5 jjm (TeV) 3 (cid:3)(cid:4) (cid:3)/SYS 3.5 STAT 4 CMS SR3 138 fb 1(cid:2) (13 TeV) Data Multijet tt , single t Other (VV, V+jets) Systematic uncertainty Wm = 2.5 TeV, Wm = 2.5 TeV, Rm Rm = 0.2 TeV = 1.25 TeV (Data-Prediction)/ (cid:3) STAT 160 140 120 100 V e T 1 . 0 / s t n e v E l l u P 80 60 40 20 0 4 2 0 2(cid:2) 4(cid:2) 1000 V e T 1 . 0 / s t n e v E 800 600 400 200 0 4 2 0 2(cid:2) 4(cid:2) l l u P 400 350 300 250 200 150 100 V e T 1 . 0 / s t n e v E 50 0 4 2 0 2(cid:2) 4(cid:2) l l u P 4 1.5 2 2.5 m 3 (TeV) jjj (cid:3)(cid:4) (cid:3)/SYS 3.5 STAT 4 CMS SR4 Data Multijet tt , single t Other (VV, V+jets) Systematic uncertainty Wm = 2.5 TeV, Wm = 2.5 TeV, Rm Rm = 0.2 TeV = 1.25 TeV (Data-Prediction)/ (cid:3) STAT CMS SR5 138 fb 1(cid:2) (13 TeV) Data Multijet tt , single t Other (VV, V+jets) Systematic uncertainty Wm = 2.5 TeV, Wm = 2.5 TeV, Rm Rm = 0.2 TeV = 1.25 TeV (Data-Prediction)/ (cid:3) STAT 1.5 2 2.5 jjjm 3 (TeV) (cid:3)(cid:4) (cid:3)/SYS 3.5 STAT 4 CMS SR6 138 fb 1(cid:2) (13 TeV) Data Multijet tt , single t Other (VV, V+jets) Systematic uncertainty Wm = 2.5 TeV, Wm = 2.5 TeV, Rm Rm = 0.2 TeV = 1.25 TeV (Data-Prediction)/ (cid:3) STAT V e T 1 . 0 / s t n e v E l l u P 35 30 25 20 15 10 5 0 4 2 0 2(cid:2) 4(cid:2) 350 300 250 200 150 100 V e T 1 . 0 / s t n e v E 50 0 4 2 0 2(cid:2) 4(cid:2) 70 60 50 40 30 20 10 0 4 2 0 2(cid:2) 4(cid:2) l l u P V e T 1 . 0 / s t n e v E l l u P 1.5 2 2.5 jjm (TeV) 3 (cid:3)(cid:4) (cid:3)/SYS 3.5 STAT 4 1.5 2 2.5 jjjm 3 (TeV) (cid:3)(cid:4) (cid:3)/SYS 3.5 STAT 4 FIG. 9. Post-fit distributions of the invariant mass of the reconstructed triboson system (mjj, mjjj) in data (black markers) and simulation (histograms) for all SRs (SRs 1–3 in the left column and SRs 4–6 in the right column). Systematic uncertainties are indicated by the shaded bands. Signal examples are superimposed, normalized to the theoretical prediction for the production cross section of ¼ 2.5 TeV with mR ¼ 0.2 TeV (solid light blue line) and 1.25 TeV (dashed purple line). The bottom panels show the pull mWKK distributions, indicating the difference between the data and background prediction, divided by the statistical uncertainty in the background, with error bars representing the statistical uncertainty and shaded bands showing the one standard deviation systematic uncertainty, normalized by the statistical uncertainty. 012002-17 A. TUMASYAN et al. PHYS. REV. D 106, 012002 (2022) is evaluated separately for the deep-W (used in SR1) and deep-WH (used in SR2–3) distributions, and is different for each signal mass scenario. The high-pT extrapolation uncertainty accounts for the fact that the SFs are derived in events containing jets with transverse momenta of a few hundred GeV, while the signal ≥ 1 TeV. To account for this effect, the jets often have pT difference in the signal selection efficiency when using HERWIG++ [58] to perform the parton shower is evaluated with respect to the default PYTHIA8 parton shower. The uncertainty is evaluated separately for each of the four types of signal jets (W, Rlqq, R3q, and R4q) for the deep-W and deep-WH discriminants. It lies in the ranges 20%–30% and 5%–40% for the merged and resolved signal, respectively. The four DEEPAK8 tagger SF uncertainties (PS, selection bias, proxy, and high-pT extrapolation) are considered as uncorrelated and result in a total uncertainty in the range 53%–63% and 10%–45% for merged and resolved signals, respectively. X. STATISTICAL ANALYSIS AND RESULTS The final mjj (mjjj) distributions for the SRs after performing a binned maximum likelihood fit in all six SRs simultaneously are shown in Fig. 9. No signal-like excess over the background expectation is observed in the data. Upper limits at 95% confidence level (C.L.) are set on CMS W→pp KK Expected Expected σ 1 ± σ 1 ± Observed limit → WR → WWW experiment experiment ) V e T ( R m 3.5 3 2.5 2 1.5 1 0.5 138 fb (13 TeV) 1− ) b p ( n o i t c e s s s o r c n o t i m i l r e p p u L C % 5 9 2− 10 3− 10 4− 10 Resolved R, merged W boson decays Merged R and W boson decays 3.5 4 4.5 5 0 1.5 2 2.5 3 KKWm (TeV) FIG. 10. Expected (red dashed lines) and observed (solid black line) lower limits at 95% C.L. on the WKK and R resonance masses for the particular parameters of the explored model. The colored area indicates the observed upper limit on the product of the signal cross section and the branching fraction to three W bosons. The blue dashed line indicates the border between the merged and resolved decay topologies probed. A signal with mR lower than 180 GeV is not considered in this search to maintain on-shell W bosons, while for mWKK > 3 TeV, we only consider mR > 0.06mWKK . the production cross section of a potential resonance signal as functions of the WKK and R resonance masses. The limits are set following the modified frequentist approach as described in Refs. [59,60] and the definition of the profile likelihood test statistic as in Ref. [61] using an asymptotic approximation [62]. Figure 10 shows the limits on the product of the WKK production cross section and the branching fraction to three W bosons. We exclude WKK resonances decaying in cascade via a scalar radion R to three W bosons at 95% C.L. with mWKK up to 3 TeV for the lowest mR of 200 GeV probed using the model provided in Refs. [3–6]. The highest mR value excluded is 1.5 TeV for mWKK ¼ 2.3 TeV. The lower limits set on the production cross sections range from 70 fb at ¼ 5 TeV. The mWKK observed limits set in the mWKK-mR plane are weaker than the expected ones because of a mild excess of data events ¼ 3 (cid:3) 0.3 TeV, which, how- observed in SR4 around mjjj ever, exhibits no resonant structure. ¼ 1.5 TeV down to 0.5 fb at mWKK For the resolved case, most of the sensitivity originates from SR4, complemented by SR5. For the merged case, SR2 and SR3 dominate the sensitivity and contribute roughly equally. The SR1 and SR6 recover sensitivity to events where one W boson has relatively low pT or mass. XI. SUMMARY A search for resonances decaying in cascade via a radion → WR → WWW, in the all- R to three W bosons, WKK hadronic final state has been presented. The search is performed in proton-proton collision data at a center-of- mass energy of 13 TeV, corresponding to a total integrated luminosity of 138 fb−1. The final states include two or three massive, large-radius jets containing the decay products of the hadronically decaying W bosons. The two-jet case corresponds to events where the radion decay products are reconstructed as a single merged jet. The three-jet case corresponds to events where each W boson from the radion decay is reconstructed as a single merged jet. In this analysis and the analysis in the single-lepton channel reported in Ref. [12], previously unexplored signatures are probed, using novel jet substructure techniques. In particular, a dedicated radion tagger based on a neural network, targeting simultaneously three different radion decay topologies, has been developed. This tagger has been calibrated with a novel “matrix method.” These techniques are also applicable to the identification of H → 4q and H → qqlν decays of Lorentz-boosted Higgs bosons. Exclusion limits are set on the product of the production cross section and the branching fraction to three W bosons in an extended warped extra-dimensional model. This result and the analysis in the single-lepton channel [12] are the first of their kind, and constrain the parameters of this model for the first time. 012002-18 SEARCH FOR RESONANCES DECAYING TO THREE W … PHYS. REV. D 106, 012002 (2022) ACKNOWLEDGMENTS We congratulate our colleagues in the CERN accelerator departments for the excellent performance of the LHC and thank the technical and administrative staffs at CERN and at other CMS institutes for their contributions to the success of the CMS effort. In addition, we gratefully acknowledge the computing centers and personnel of the Worldwide LHC Computing Grid and other centers for delivering so effectively the computing infrastructure essential to our analyses. Finally, we acknowledge the enduring support for the CMS the construction and operation of the LHC, detector, and the supporting computing infrastructure provided by the following funding agencies: BMBWF and FWF (Austria); FNRS and FWO (Belgium); CNPq, CAPES, FAPERJ, FAPERGS, and FAPESP (Brazil); MES and BNSF (Bulgaria); CERN; CAS, MoST, and NSFC (China); MINCIENCIAS (Colombia); MSES and CSF (Croatia); RIF (Cyprus); SENESCYT (Ecuador); MoER, ERC PUT, and ERDF (Estonia); Academy of Finland, MEC, and HIP (Finland); CEA and CNRS/IN2P3 (France); BMBF, DFG, and HGF (Germany); GSRI (Greece); NKFIA (Hungary); DAE and DST (India); IPM (Iran); SFI (Ireland); INFN (Italy); MSIP and NRF (Republic of Korea); MES (Latvia); LAS (Lithuania); MOE and UM (Malaysia); BUAP, CINVESTAV, CONACYT, LNS, SEP, and UASLP-FAI (Mexico); MOS (Montenegro); MBIE (New Zealand); PAEC (Pakistan); MSHE and NSC JINR (Dubna); MON, (Poland); FCT (Portugal); RosAtom, RAS, RFBR, and NRC KI (Russia); MESTD (Serbia); MCIN/AEI and PCTI (Spain); MOSTR (Sri Lanka); Swiss Funding Agencies (Switzerland); MST and NSTDA (Taipei); ThEPCenter, (Thailand); TUBITAK and TAEK (Turkey); NASU (Ukraine); STFC (United Kingdom); DOE and NSF (USA). from the the European Research Marie-Curie Council and Horizon 2020 Grant, Contracts No. 675440, No. 724704, No. 752730, No. 758316, No. 765710, No. 824093, No. 884104, and COST Action CA16108 (European Union); the Leventis Foundation; the Alfred P. von Humboldt Sloan Foundation; Individuals have received support the Alexander IPST, STAR, program and of Science—EOS”—be.h Foundation; the Belgian Federal Science Policy Office; the Fonds pour la Formation `a la Recherche dans l’Industrie et dans l’Agriculture (FRIA-Belgium); the Agentschap voor Innovatie door Wetenschap en Technologie (IWT- Belgium); the F. R. S.-FNRS and FWO (Belgium) under “Excellence the Project the Beijing Municipal Science & No. 30820817; Technology Commission, No. Z191100007219010; the Ministry of Education, Youth and Sports (MEYS) of the Czech Republic; the Deutsche Forschungsgemeinschaft (DFG), under Germany’s Excellence Strategy—EXC 2121 “Quantum Universe”—390833306, and under 400140256—GRK2497; Project No. the Lendület (“Momentum”) Program and the János Bolyai Research Scholarship of the Hungarian Academy of Sciences, the the NKFIA New National Excellence Program ÚNKP, research grants 123842, 123959, 124845, 124850, 125105, 128713, 128786, and 129058 (Hungary); the Council of Science and Industrial Research, India; the Latvian Council of Science; the Ministry of Science and Higher Education and the National Science Center, Contracts Opus No. 2014/ 15/B/ST2/03998 and No. 2015/19/B/ST2/02861 (Poland); the Fundação para a Ciência e a Tecnologia, Grant the National No. CEECIND/01334/2018 (Portugal); Priorities Research Program by Qatar National Research Fund; the Ministry of Science and Higher Education, Projects No. 14.W03.31.0026 and No. FSWW-2020- 0008, and the Russian Foundation for Basic Research, Project No. 19-42-703014 (Russia); MCIN/AEI/10.13039/ 501100011033, ERDF “a way of making Europe,” and the Programa Estatal de Fomento de la Investigación Científica y T´ecnica de Excelencia María de Maeztu, Grant No. MDM-2017-0765 and Programa Severo Ochoa del Principado de Asturias (Spain); the Stavros Niarchos Foundation (Greece); the Rachadapisek Sompot Fund for Postdoctoral Fellowship, Chulalongkorn University and the Chulalongkorn Academic into Its 2nd Century Project Advancement Project (Thailand); the Kavli Foundation; the Nvidia Corporation; the SuperMicro Corporation; the Welch Foundation, Contract No. C-1845; and the Weston Havens Foundation (USA). [1] CMS Collaboration, Identification of heavy, energetic, hadronically decaying particles using machine-learning techniques, J. Instrum. 15, P06005 (2020). [2] CMS Collaboration, A search for the standard model Higgs boson decaying to charm quarks, J. High Energy Phys. 03 (2020) 131. [3] K. Agashe, P. Du, S. Hong, and R. Sundrum, Flavor universal resonances and warped gravity, J. High Energy Phys. 01 (2017) 016. [4] K. S. Agashe, J. Collins, P. Du, S. Hong, D. Kim, and R. K. Mishra, LHC signals from cascade decays of warped vector resonances, J. High Energy Phys. 05 (2017) 078. [5] K. Agashe, J. H. Collins, P. Du, S. Hong, D. Kim, and R. K. Mishra, Dedicated strategies for triboson signals from cascade decays of vector resonances, Phys. Rev. D 99, 075016 (2019). [6] K. Agashe, J. H. Collins, P. Du, S. Hong, D. Kim, and R. K. Mishra, Detecting a boosted diboson resonance, J. High Energy Phys. 11 (2018) 027. 012002-19 A. TUMASYAN et al. PHYS. REV. D 106, 012002 (2022) [7] Y.-P. Kuang, H.-Y. Ren, and L.-H. Xia, Further investigation of the model-independent probe of heavy neutral Higgs bosons at LHC Run 2, Chin. Phys. C 40, 023101 (2016). [8] Y.-P. Kuang, H.-Y. Ren, and L.-H. Xia, Model-independent probe of anomalous heavy neutral Higgs bosons at the LHC, Phys. Rev. D 90, 115002 (2014). [9] W. D. Goldberger and M. B. Wise, Modulus Stabilization with Bulk Fields, Phys. Rev. Lett. 83, 4922 (1999). [10] ATLAS Collaboration, Observation of WWW Production in ¼ 13 TeV with the ATLAS Detector ffiffiffi s p pp Collisions at (to be published). [11] CMS Collaboration, Observation of the Production of Three ¼ 13 TeV, Phys. Rev. Lett. ffiffiffi s p Massive Gauge Bosons at 125, 151802 (2020). [12] CMS Collaboration, companion Letter, Search for Resonan- ces Decaying to Three W Bosons in Proton-Proton Collisions ¼ 13 TeV, Phys. Rev. Lett. 129, 021802 (2022). at ffiffiffi s p [13] HEPData record for this analysis (2021), 10.17182/hep- data.115182. [14] CMS Collaboration, Performance of the CMS Level-1 ¼ 13 TeV, in proton-proton collisions at ffiffiffi s p trigger J. Instrum. 15, P10017 (2020). [15] CMS Collaboration, The CMS trigger system, J. Instrum. 12, P01020 (2017). [16] CMS Collaboration, The CMS experiment at the CERN LHC, J. Instrum. 3, S08004 (2008). [17] CMS Collaboration, Precision luminosity measurement in p ¼ 13 TeV in 2015 and 2016 proton-proton collisions at at CMS, Eur. Phys. J. C 81, 800 (2021). ffiffiffi s [18] CMS Collaboration, CMS luminosity measurement for the ¼ 13 TeV, CMS Physics 2017 data-taking period at Analysis Summary Report No. CMS-PAS-LUM-17-004, 2018, https://cds.cern.ch/record/2621960. ffiffiffi s p [19] CMS Collaboration, CMS luminosity measurement for the ¼ 13 TeV, CMS Physics 2018 data-taking period at Analysis Summary Report No. CMS-PAS-LUM-18-002, 2019, https://cds.cern.ch/record/2676164. ffiffiffi s p [20] J. Alwall, R. Frederix, S. Frixione, V. Hirschi, F. Maltoni, O. Mattelaer, H.-S. Shao, T. Stelzer, P. Torielli, and M. Zaro, The automated computation of tree-level and next-to-lead- ing order differential cross sections, and their matching to parton shower simulations, J. High Energy Phys. 07 (2014) 079. [21] P. Nason, A new method for combining NLO QCD with shower Monte Carlo algorithms, J. High Energy Phys. 11 (2004) 040. [22] S. Frixione, P. Nason, and C. Oleari, Matching NLO QCD computations with parton shower simulations: The POWHEG method, J. High Energy Phys. 11 (2007) 070. [23] S. Alioli, P. Nason, C. Oleari, and E. Re, A general framework for implementing NLO calculations in shower Monte Carlo programs: The POWHEG BOX, J. High Energy Phys. 06 (2010) 043. [24] S. Alioli, S.-O. Moch, and P. Uwer, Hadronic top-quark pair-production with one jet and parton showering, J. High Energy Phys. 01 (2012) 137. [25] S. Alioli, P. Nason, C. Oleari, and E. Re, NLO single-top in POWHEG: s- and production matched with shower t-channel contributions, J. High Energy Phys. 09 (2009) 111; 02 (2010) 011(E). [26] R. Frederix, E. Re, and P. Torrielli, Single-top t-channel hadroproduction in the four-flavour scheme with POWHEG and AMC@NLO, J. High Energy Phys. 09 (2012) 130. [27] J. Alwall, S. Höche, F. Krauss, N. Lavesson, L. Lonnblad, F. Maltoni, M. L. Mangano, M. Moretti, C. G. Papadopoulos, F. Piccinini, S. Schumann, M. Treccani, J. Winter, and M. Worek, Comparative study of various algorithms for the merging of parton showers and matrix elements in hadronic collisions, Eur. Phys. J. C 53, 473 (2008). [28] R. D. Ball et al. (NNPDF Collaboration), Parton distribu- tions for the LHC Run II, J. High Energy Phys. 04 (2015) 040. [29] R. D. Ball et al. (NNPDF Collaboration), Parton distribu- tions from high-precision collider data, Eur. Phys. J. C 77, 663 (2017). [30] T. Sjöstrand, S. Ask, J. R. Christiansen, R. Corke, N. Desai, P. Ilten, S. Mrenna, S. Prestel, C. O. Rasmussen, and P. Z. Skands, An introduction to PYTHIA 8.2, Comput. Phys. Commun. 191, 159 (2015). [31] CMS Collaboration, Event generator tunes obtained from underlying event and multiparton scattering measurements, Eur. Phys. J. C 76, 155 (2016). [32] CMS Collaboration, Extraction and validation of a new set of CMS PYTHIA8 tunes from underlying-event measure- ments, Eur. Phys. J. C 80, 4 (2020). [33] S. Agostinelli et al. (GEANT4 Collaboration), GEANT4—A simulation toolkit, Nucl. Instrum. Methods Phys. Res., Sect. A 506, 250 (2003). [34] J. Allison et al., GEANT4 developments and applications, IEEE Trans. Nucl. Sci. 53, 270 (2006). [35] CMS Collaboration, Measurement of the inclusive W and Z ¼ 7 TeV production cross sections in pp collisions at with the CMS experiment, J. High Energy Phys. 10 (2011) 132. ffiffiffi s p [36] M. Cacciari, G. P. Salam, and G. Soyez, The anti-kT J. High Energy Phys. 04 jet clustering algorithm, (2008) 063. [37] M. Cacciari, G. P. Salam, and G. Soyez, FastJet user manual, Eur. Phys. J. C 72, 1896 (2012). [38] CMS Collaboration, Particle-flow reconstruction and global event description with the CMS detector, J. Instrum. 12, P10003 (2017). [39] CMS Collaboration, Pileup mitigation at CMS in 13 TeV data, J. Instrum. 15, P09018 (2020). [40] D. Bertolini, P. Harris, M. Low, and N. Tran, Pileup per particle identification, J. High Energy Phys. 10 (2014) 059. [41] CMS Collaboration, Jet energy scale and resolution in the CMS experiment in pp collisions at 8 TeV, J. Instrum. 12, P02014 (2017). [42] CMS Collaboration, in 13 TeV data, CMS Physics Analysis Summary Report No. CMS-PAS-JME-16-003, 2017, https://cds.cern.ch/ record/2256875. algorithms performance Jet [43] CMS Collaboration, Identification of heavy-flavour jets in pp collisions at 13 TeV, with the CMS detector J. Instrum. 13, P05011 (2018). [44] CMS Collaboration, Performance of missing transverse momentum reconstruction in proton-proton collisions at 012002-20 SEARCH FOR RESONANCES DECAYING TO THREE W … PHYS. REV. D 106, 012002 (2022) p ffiffiffi s ¼ 13 TeV using the CMS detector, J. Instrum. 14, P07004 (2019). [45] M. Dasgupta, A. Fregoso, S. Marzani, and G. P. Salam, Towards an understanding of jet substructure, J. High Energy Phys. 09 (2013) 029. [46] J. M. Butterworth, A. R. Davison, M. Rubin, and G. P. Salam, Jet Substructure as a New Higgs Search Channel at the LHC, Phys. Rev. Lett. 100, 242001 (2008). [47] A. J. Larkoski, S. Marzani, G. Soyez, and J. Thaler, Soft drop, J. High Energy Phys. 05 (2014) 146. [48] CMS Collaboration, Identification techniques for highly boosted W bosons that decay into hadrons, J. High Energy Phys. 12 (2014) 017. [49] CMS Collaboration, Performance of the CMS muon de- tector and muon reconstruction with proton-proton colli- sions at ¼ 13 TeV, J. Instrum. 13, P06015 (2018). ffiffiffi s p [50] CMS Collaboration, Performance of electron reconstruction in proton- ¼ 8 TeV, J. Instrum. 10, P06005 and selection with the CMS detector p ffiffiffi s proton collisions at (2015). [51] D. Krohn, J. Thaler, and L. Wang, Jet trimming, J. High Energy Phys. 02 (2010) 084. [52] CMS Collaboration, Search for heavy resonances decaying to WW, WZ, or WH boson pairs in a final state consisting of a lepton and a large-radius jet in proton-proton collisions at p ¼ 13 TeV, Phys. Rev. D 105, 032008 (2022). [53] J. Thaler and K. Van Tilburg, Maximizing boosted top identification by minimizing N-subjettiness, J. High Energy Phys. 02 (2012) 093. ffiffiffi s [54] J. Bellm et al., HERWIG7.0/HERWIG++ 3.0 release note, Eur. Phys. J. C 76, 196 (2016). [55] CMS Collaboration, Measurement of differential cross sections for top quark pair production using the lepton þ jets final state in proton-proton collisions at 13 TeV, Phys. Rev. D 95, 092001 (2017). [56] CMS Collaboration, Measurement of t¯t normalised multi- ¼ 13 TeV, differential cross sections in pp collisions at and simultaneous determination of the strong coupling strength, top quark pole mass, and parton distribution functions, Eur. Phys. J. C 80, 658 (2020). ffiffiffi s p [57] CMS Collaboration, Measurement of the inelastic proton- ¼ 13 TeV, J. High Energy Phys. ffiffiffi s p proton cross section at 07 (2018) 161. [58] M. Bähr, S. Gieseke, M. A. Gigg, D. Grellscheid, K. Hamilton, O. Latunde-Dada, S. Plätzer, P. Richardson, M. H. Seymour, A. Sherstnev, J. Tully, and B. R. Webber, HERWIG++ physics and manual, Eur. Phys. J. C 58, 639 (2008). [59] T. Junk, Confidence level computation for combining searches with small statistics, Nucl. Instrum. Methods Phys. Res., Sect. A 434, 435 (1999). [60] A. L. Read, Presentation of search results: The CLs tech- nique, J. Phys. G 28, 2693 (2002). [61] ATLAS and CMS Collaborations, and the LHC Higgs Combination Group, Procedure for the LHC Higgs boson search combination in Summer 2011, Reports No. CMS- NOTE-2011-005 and No. ATL-PHYS-PUB-2011-11, 2011, https://cds.cern.ch/record/1379837. [62] G. Cowan, K. Cranmer, E. Gross, and O. Vitells, Asymp- totic formulae for likelihood-based tests of new physics, Eur. Phys. J. C 71, 1554 (2011); 73, 2501(E) (2013). A. Tumasyan,1 W. Adam,2 J. W. Andrejkovic,2 T. Bergauer,2 S. Chatterjee,2 K. Damanakis,2 M. Dragicevic,2 A. Escalante Del Valle,2 R. Frühwirth,2,b M. Jeitler,2,b N. Krammer,2 L. Lechner,2 D. Liko,2 I. Mikulec,2 P. Paulitsch,2 F. M. Pitters,2 J. Schieck,2,b R. Schöfbeck,2 D. Schwarz,2 S. Templ,2 W. Waltenberger,2 C.-E. Wulz,2,b V. Chekhovsky,3 A. Litomin,3 V. Makarenko,3 M. R. Darwish,4,c E. A. De Wolf,4 T. Janssen,4 T. Kello,4,d A. Lelek,4 H. Rejeb Sfar,4 P. Van Mechelen,4 S. Van Putte,4 N. Van Remortel,4 E. S. Bols,5 J. D’Hondt,5 M. Delcourt,5 H. El Faham,5 S. Lowette,5 S. Moortgat,5 A. Morton,5 D. Müller,5 A. R. Sahasransu,5 S. Tavernier,5 W. Van Doninck,5 D. Vannerom,5 D. Beghin,6 B. Bilin,6 B. Clerbaux,6 G. De Lentdecker,6 L. Favart,6 A. K. Kalsi,6 K. Lee,6 M. Mahdavikhorrami,6 I. Makarenko,6 L. Moureaux,6 S. Paredes,6 L. P´etr´e,6 A. Popov,6 N. Postiau,6 E. Starling,6 L. Thomas,6 M. Vanden Bemden,6 C. Vander Velde,6 P. Vanlaer,6 T. Cornelis,7 D. Dobur,7 J. Knolle,7 L. Lambrecht,7 G. Mestdach,7 M. Niedziela,7 C. Rendón,7 C. Roskas,7 A. Samalan,7 K. Skovpen,7 M. Tytgat,7 B. Vermassen,7 L. Wezenbeek,7 A. Benecke,8 A. Bethani,8 G. Bruno,8 F. Bury,8 C. Caputo,8 P. David,8 C. Delaere,8 I. S. Donertas,8 A. Giammanco,8 K. Jaffel,8 Sa. Jain,8 V. Lemaitre,8 K. Mondal,8 J. Prisciandaro,8 A. Taliercio,8 M. Teklishyn,8 T. T. Tran,8 P. Vischia,8 S. Wertz,8 G. A. Alves,9 C. Hensel,9 A. Moraes,9 P. Rebello Teles,9 W. L. Aldá Júnior,10 M. Alves Gallo Pereira,10 M. Barroso Ferreira Filho,10 H. Brandao Malbouisson,10 W. Carvalho,10 J. Chinellato,10,e E. M. Da Costa,10 G. G. Da Silveira,10,f D. De Jesus Damiao,10 V. Dos Santos Sousa,10 S. Fonseca De Souza,10 C. Mora Herrera,10 K. Mota Amarilo,10 L. Mundim,10 H. Nogima,10 A. Santoro,10 S. M. Silva Do Amaral,10 A. Sznajder,10 M. Thiel,10 F. Torres Da Silva De Araujo,10,g A. Vilela Pereira,10 C. A. Bernardes,11,f L. Calligaris,11 T. R. Fernandez Perez Tomei,11 E. M. Gregores,11 D. S. Lemos,11 P. G. Mercadante,11 S. F. Novaes,11 Sandra S. Padula,11 A. Aleksandrov,12 G. Antchev,12 R. Hadjiiska,12 P. Iaydjiev,12 M. Misheva,12 M. Rodozov,12 M. Shopova,12 G. Sultanov,12 A. Dimitrov,13 T. Ivanov,13 L. Litov,13 B. Pavlov,13 P. Petkov,13 A. Petrov,13 T. Cheng,14 T. Javaid,14,h M. Mittal,14 L. Yuan,14 M. Ahmad,15 G. Bauer,15 C. Dozen,15,i Z. Hu,15 J. Martins,15,j Y. Wang,15 012002-21 A. TUMASYAN et al. PHYS. REV. D 106, 012002 (2022) K. Yi,15,k,l E. Chapon,16 G. M. Chen,16,h H. S. Chen,16,h M. Chen,16 F. Iemmi,16 A. Kapoor,16 D. Leggat,16 H. Liao,16 Z.-A. Liu,16,m V. Milosevic,16 F. Monti,16 R. Sharma,16 J. Tao,16 J. Thomas-Wilsker,16 J. Wang,16 H. Zhang,16 J. Zhao,16 A. Agapitos,17 Y. An,17 Y. Ban,17 C. Chen,17 A. Levin,17 Q. Li,17 X. Lyu,17 Y. Mao,17 S. J. Qian,17 D. Wang,17 J. Xiao,17 H. Yang,17 M. Lu,18 Z. You,18 X. Gao,19,d H. Okawa,19 Y. Zhang,19 Z. Lin,20 M. Xiao,20 C. Avila,21 A. Cabrera,21 C. Florez,21 J. Fraga,21 J. Mejia Guisao,22 F. Ramirez,22 J. D. Ruiz Alvarez,22 D. Giljanovic,23 N. Godinovic,23 D. Lelas,23 I. Puljak,23 Z. Antunovic,24 M. Kovac,24 T. Sculac,24 V. Brigljevic,25 D. Ferencek,25 D. Majumder,25 M. Roguljic,25 A. Starodumov,25,n T. Susa,25 A. Attikis,26 K. Christoforou,26 A. Ioannou,26 G. Kole,26 M. Kolosova,26 S. Konstantinou,26 J. Mousa,26 C. Nicolaou,26 F. Ptochos,26 P. A. Razis,26 H. Rykaczewski,26 H. Saka,26 M. Finger,27,o M. Finger Jr.,27,o A. Kveton,27 E. Ayala,28 E. Carrera Jarrin,29 H. Abdalla,30,p A. A. Abdelalim,30,q,r M. A. Mahmoud,31 Y. Mohammed,31 S. Bhowmik,32 R. K. Dewanjee,32 K. Ehataht,32 M. Kadastik,32 S. Nandan,32 C. Nielsen,32 J. Pata,32 M. Raidal,32 L. Tani,32 C. Veelken,32 P. Eerola,33 H. Kirschenmann,33 K. Osterberg,33 M. Voutilainen,33 S. Bharthuar,34 E. Brücken,34 F. Garcia,34 J. Havukainen,34 M. S. Kim,34 R. Kinnunen,34 T. Lamp´en,34 K. Lassila-Perini,34 S. Lehti,34 T. Lind´en,34 M. Lotti,34 L. Martikainen,34 M. Myllymäki,34 J. Ott,34 H. Siikonen,34 E. Tuominen,34 J. Tuominiemi,34 P. Luukka,35 H. Petrow,35 T. Tuuva,35 C. Amendola,36 M. Besancon,36 F. Couderc,36 M. Dejardin,36 D. Denegri,36 J. L. Faure,36 F. Ferri,36 S. Ganjour,36 P. Gras,36 G. Hamel de Monchenault,36 P. Jarry,36 B. Lenzi,36 E. Locci,36 J. Malcles,36 J. Rander,36 A. Rosowsky,36 M. Ö. Sahin,36 A. Savoy-Navarro,36,s M. Titov,36 G. B. Yu,36 S. Ahuja,37 F. Beaudette,37 M. Bonanomi,37 A. Buchot Perraguin,37 P. Busson,37 A. Cappati,37 C. Charlot,37 O. Davignon,37 B. Diab,37 G. Falmagne,37 S. Ghosh,37 R. Granier de Cassagnac,37 A. Hakimi,37 I. Kucher,37 J. Motta,37 M. Nguyen,37 C. Ochando,37 P. Paganini,37 J. Rembser,37 R. Salerno,37 U. Sarkar,37 J. B. Sauvan,37 Y. Sirois,37 A. Tarabini,37 A. Zabi,37 A. Zghiche,37 J.-L. Agram,38,t J. Andrea,38 D. Apparu,38 D. Bloch,38 G. Bourgatte,38 J.-M. Brom,38 E. C. Chabert,38 C. Collard,38 D. Darej,38 J.-C. Fontaine,38,t U. Goerlach,38 C. Grimault,38 A.-C. Le Bihan,38 E. Nibigira,38 P. Van Hove,38 E. Asilar,39 S. Beauceron,39 C. Bernet,39 G. Boudoul,39 C. Camen,39 A. Carle,39 N. Chanon,39 D. Contardo,39 P. Depasse,39 H. El Mamouni,39 J. Fay,39 S. Gascon,39 M. Gouzevitch,39 B. Ille,39 I. B. Laktineh,39 H. Lattaud,39 A. Lesauvage,39 M. Lethuillier,39 L. Mirabito,39 S. Perries,39 K. Shchablo,39 V. Sordini,39 L. Torterotot,39 G. Touquet,39 M. Vander Donckt,39 S. Viret,39 I. Lomidze,40 T. Toriashvili,40,u Z. Tsamalaidze,40,o V. Botta,41 L. Feld,41 K. Klein,41 M. Lipinski,41 D. Meuser,41 A. Pauls,41 N. Röwert,41 J. Schulz,41 M. Teroerde,41 A. Dodonova,42 D. Eliseev,42 M. Erdmann,42 P. Fackeldey,42 B. Fischer,42 T. Hebbeker,42 K. Hoepfner,42 F. Ivone,42 L. Mastrolorenzo,42 M. Merschmeyer,42 A. Meyer,42 G. Mocellin,42 S. Mondal,42 S. Mukherjee,42 D. Noll,42 A. Novak,42 A. Pozdnyakov,42 Y. Rath,42 H. Reithler,42 A. Schmidt,42 S. C. Schuler,42 A. Sharma,42 L. Vigilante,42 S. Wiedenbeck,42 S. Zaleski,42 C. Dziwok,43 G. Flügge,43 W. Haj Ahmad,43,v O. Hlushchenko,43 T. Kress,43 A. Nowack,43 O. Pooth,43 D. Roy,43 A. Stahl,43,w T. Ziemons,43 A. Zotz,43 H. Aarup Petersen,44 M. Aldaya Martin,44 P. Asmuss,44 S. Baxter,44 M. Bayatmakou,44 O. Behnke,44 A. Bermúdez Martínez,44 S. Bhattacharya,44 A. A. Bin Anuar,44 F. Blekman,44 K. Borras,44,x D. Brunner,44 A. Campbell,44 A. Cardini,44 C. Cheng,44 F. Colombina,44 S. Consuegra Rodríguez,44 G. Correia Silva,44 V. Danilov,44 M. De Silva,44 L. Didukh,44 G. Eckerlin,44 D. Eckstein,44 L. I. Estevez Banos,44 O. Filatov,44 E. Gallo,44,y A. Geiser,44 A. Giraldi,44 A. Grohsjean,44 M. Guthoff,44 A. Jafari,44,z N. Z. Jomhari,44 H. Jung,44 A. Kasem,44,x M. Kasemann,44 H. Kaveh,44 C. Kleinwort,44 R. Kogler,44 D. Krücker,44 W. Lange,44 K. Lipka,44 W. Lohmann,44,aa R. Mankel,44 I.-A. Melzer-Pellmann,44 M. Mendizabal Morentin,44 J. Metwally,44 A. B. Meyer,44 M. Meyer,44 J. Mnich,44 A. Mussgiller,44 A. Nürnberg,44 Y. Otarid,44 D. P´erez Adán,44 D. Pitzl,44 A. Raspereza,44 B. Ribeiro Lopes,44 J. Rübenach,44 A. Saggio,44 A. Saibel,44 M. Savitskyi,44 M. Scham,44,bb V. Scheurer,44 S. Schnake,44 P. Schütze,44 C. Schwanenberger,44,y M. Shchedrolosiev,44 R. E. Sosa Ricardo,44 D. Stafford,44 N. Tonon,44 M. Van De Klundert,44 F. Vazzoler,44 R. Walsh,44 D. Walter,44 Q. Wang,44 Y. Wen,44 K. Wichmann,44 L. Wiens,44 C. Wissing,44 S. Wuchterl,44 R. Aggleton,45 S. Albrecht,45 S. Bein,45 L. Benato,45 P. Connor,45 K. De Leo,45 M. Eich,45 F. Feindt,45 A. Fröhlich,45 C. Garbers,45 E. Garutti,45 P. Gunnellini,45 M. Hajheidari,45 J. Haller,45 A. Hinzmann,45 G. Kasieczka,45 R. Klanner,45 T. Kramer,45 V. Kutzner,45 J. Lange,45 T. Lange,45 A. Lobanov,45 A. Malara,45 A. Mehta,45 A. Nigamova,45 K. J. Pena Rodriguez,45 M. Rieger,45 O. Rieger,45 P. Schleper,45 M. Schröder,45 J. Schwandt,45 J. Sonneveld,45 H. Stadie,45 G. Steinbrück,45 A. Tews,45 I. Zoi,45 J. Bechtel,46 S. Brommer,46 M. Burkart,46 E. Butz,46 R. Caspart,46 T. Chwalek,46 W. De Boer,46,a A. Dierlamm,46 A. Droll,46 K. El Morabit,46 N. Faltermann,46 M. Giffels,46 J. O. Gosewisch,46 A. Gottmann,46 F. Hartmann,46,w C. Heidecker,46 U. Husemann,46 P. Keicher,46 R. Koppenhöfer,46 S. Maier,46 M. Metzler,46 S. Mitra,46 Th. Müller,46 M. Neukum,46 G. Quast,46 K. Rabbertz,46 J. Rauser,46 D. Savoiu,46 M. Schnepf,46 D. Seith,46 I. Shvetsov,46 H. J. Simonis,46 R. Ulrich,46 J. Van Der Linden,46 R. F. Von Cube,46 M. Wassmer,46 012002-22 SEARCH FOR RESONANCES DECAYING TO THREE W … PHYS. REV. D 106, 012002 (2022) M. Weber,46 S. Wieland,46 R. Wolf,46 S. Wozniewski,46 S. Wunsch,46 G. Anagnostou,47 G. Daskalakis,47 A. Kyriakis,47 D. Loukas,47 A. Stakia,47 M. Diamantopoulou,48 D. Karasavvas,48 P. Kontaxakis,48 C. K. Koraka,48 A. Manousakis-Katsikakis,48 A. Panagiotou,48 I. Papavergou,48 N. Saoulidou,48 K. Theofilatos,48 E. Tziaferi,48 K. Vellidis,48 E. Vourliotis,48 G. Bakas,49 K. Kousouris,49 I. Papakrivopoulos,49 G. Tsipolitis,49 A. Zacharopoulou,49 K. Adamidis,50 I. Bestintzanos,50 I. Evangelou,50 C. Foudas,50 P. Gianneios,50 P. Katsoulis,50 P. Kokkas,50 N. Manthos,50 I. Papadopoulos,50 J. Strologas,50 M. Csanad,51 K. Farkas,51 M. M. A. Gadallah,51,cc S. Lökös,51,dd P. Major,51 K. Mandal,51 G. Pasztor,51 A. J. Rádl,51 O. Surányi,51 G. I. Veres,51 M. Bartók,52,ee G. Bencze,52 C. Hajdu,52 D. Horvath,52,ff,gg F. Sikler,52 V. Veszpremi,52 S. Czellar,53 D. Fasanella,53 F. Fienga,53 J. Karancsi,53,ee J. Molnar,53 Z. Szillasi,53 D. Teyssier,53 P. Raics,54 Z. L. Trocsanyi,54,hh B. Ujvari,54 T. Csorgo,55,ii F. Nemes,55,ii T. Novak,55 S. Bahinipati,56,jj C. Kar,56 P. Mal,56 T. Mishra,56 V. K. Muraleedharan Nair Bindhu,56,kk A. Nayak,56,kk P. Saha,56 N. Sur,56 S. K. Swain,56 D. Vats,56,kk S. Bansal,57 S. B. Beri,57 V. Bhatnagar,57 G. Chaudhary,57 S. Chauhan,57 N. Dhingra,57,ll R. Gupta,57 A. Kaur,57 H. Kaur,57 M. Kaur,57 P. Kumari,57 M. Meena,57 K. Sandeep,57 J. B. Singh,57 A. K. Virdi,57 A. Ahmed,58 A. Bhardwaj,58 B. C. Choudhary,58 M. Gola,58 S. Keshri,58 A. Kumar,58 M. Naimuddin,58 P. Priyanka,58 K. Ranjan,58 A. Shah,58 M. Bharti,59,mm R. Bhattacharya,59 S. Bhattacharya,59 D. Bhowmik,59 S. Dutta,59 S. Dutta,59 B. Gomber,59,nn M. Maity,59,oo P. Palit,59 P. K. Rout,59 G. Saha,59 B. Sahu,59 S. Sarkar,59 M. Sharan,59 P. K. Behera,60 S. C. Behera,60 P. Kalbhor,60 J. R. Komaragiri,60,pp D. Kumar,60,pp A. Muhammad,60 L. Panwar,60,pp R. Pradhan,60 P. R. Pujahari,60 A. Sharma,60 A. K. Sikdar,60 P. C. Tiwari,60,pp K. Naskar,61,qq T. Aziz,62 S. Dugad,62 M. Kumar,62 S. Banerjee,63 R. Chudasama,63 M. Guchait,63 S. Karmakar,63 S. Kumar,63 G. Majumder,63 K. Mazumdar,63 S. Mukherjee,63 A. Alpana,64 S. Dube,64 B. Kansal,64 A. Laha,64 S. Pandey,64 A. Rastogi,64 S. Sharma,64 H. Bakhshiansohi,65,rr,ss E. Khazaie,65,ss M. Sedghi,65,tt S. Chenarani,66,uu S. M. Etesami,66 M. Khakzad,66 M. Mohammadi Najafabadi,66 M. Grunewald,67 M. Abbrescia,68a,68b R. Aly,68a,68b,vv C. Aruta,68a,68b A. Colaleo,68a D. Creanza,68a,68c N. De Filippis,68a,68c M. De Palma,68a,68b A. Di Florio,68a,68b A. Di Pilato,68a,68b W. Elmetenawee,68a,68b F. Errico,68a,68b L. Fiore,68a A. Gelmi,68a,68b M. Gul,68a G. Iaselli,68a,68c M. Ince,68a,68b S. Lezki,68a,68b G. Maggi,68a,68c M. Maggi,68a I. Margjeka,68a,68b V. Mastrapasqua,68a,68b S. My,68a,68b S. Nuzzo,68a,68b A. Pellecchia,68a,68b A. Pompili,68a,68b G. Pugliese,68a,68c D. Ramos,68a A. Ranieri,68a G. Selvaggi,68a,68b L. Silvestris,68a F. M. Simone,68a,68b Ü. Sözbilir,68a R. Venditti,68a P. Verwilligen,68a G. Abbiendi,69a C. Battilana,69a,69b D. Bonacorsi,69a,69b L. Borgonovi,69a L. Brigliadori,69a R. Campanini,69a,69b P. Capiluppi,69a,69b A. Castro,69a,69b F. R. Cavallo,69a C. Ciocca,69a M. Cuffiani,69a,69b G. M. Dallavalle,69a T. Diotalevi,69a,69b F. Fabbri,69a A. Fanfani,69a,69b P. Giacomelli,69a L. Giommi,69a,69b C. Grandi,69a L. Guiducci,69a,69b S. Lo Meo,69a,ww L. Lunerti,69a,69b S. Marcellini,69a G. Masetti,69a F. L. Navarria,69a,69b A. Perrotta,69a F. Primavera,69a,69b A. M. Rossi,69a,69b T. Rovelli,69a,69b G. P. Siroli,69a,69b S. Albergo,70a,70b,xx S. Costa,70a,70b,xx A. Di Mattia,70a R. Potenza,70a,70b A. Tricomi,70a,70b,xx C. Tuve,70a,70b G. Barbagli,71a A. Cassese,71a R. Ceccarelli,71a,71b V. Ciulli,71a,71b C. Civinini,71a R. D’Alessandro,71a,71b E. Focardi,71a,71b G. Latino,71a,71b P. Lenzi,71a,71b M. Lizzo,71a,71b M. Meschini,71a S. Paoletti,71a R. Seidita,71a,71b G. Sguazzoni,71a L. Viliani,71a L. Benussi,72 S. Bianco,72 D. Piccolo,72 M. Bozzo,73a,73b F. Ferro,73a R. Mulargia,73a E. Robutti,73a S. Tosi,73a,73b A. Benaglia,74a G. Boldrini,74a F. Brivio,74a,74b F. Cetorelli,74a,74b F. De Guio,74a,74b M. E. Dinardo,74a,74b P. Dini,74a S. Gennai,74a A. Ghezzi,74a,74b P. Govoni,74a,74b L. Guzzi,74a,74b M. T. Lucchini,74a,74b M. Malberti,74a S. Malvezzi,74a A. Massironi,74a D. Menasce,74a L. Moroni,74a M. Paganoni,74a,74b D. Pedrini,74a B. S. Pinolini,74a S. Ragazzi,74a,74b N. Redaelli,74a T. Tabarelli de Fatis,74a,74b D. Valsecchi,74a,74b,w D. Zuolo,74a,74b S. Buontempo,75a F. Carnevali,75a,75b N. Cavallo,75a,75c A. De Iorio,75a,75b F. Fabozzi,75a,75c A. O. M. Iorio,75a,75b L. Lista,75a,75b,yy S. Meola,75a,75d,w P. Paolucci,75a,w B. Rossi,75a C. Sciacca,75a,75b P. Azzi,76a N. Bacchetta,76a D. Bisello,76a,76b P. Bortignon,76a A. Bragagnolo,76a,76b R. Carlin,76a,76b P. Checchia,76a T. Dorigo,76a U. Dosselli,76a F. Gasparini,76a,76b U. Gasparini,76a,76b G. Grosso,76a L. Layer,76a,zz E. Lusiani,76a M. Margoni,76a,76b A. T. Meneguzzo,76a,76b J. Pazzini,76a,76b P. Ronchese,76a,76b R. Rossin,76a,76b F. Simonetto,76a,76b G. Strong,76a M. Tosi,76a,76b H. Yarar,76a,76b M. Zanetti,76a,76b P. Zotto,76a,76b A. Zucchetta,76a,76b G. Zumerle,76a,76b C. Aim`e,77a,77b A. Braghieri,77a S. Calzaferri,77a,77b D. Fiorina,77a,77b P. Montagna,77a,77b S. P. Ratti,77a,77b V. Re,77a C. Riccardi,77a,77b P. Salvini,77a I. Vai,77a P. Vitulo,77a,77b P. Asenov,78a,aaa G. M. Bilei,78a D. Ciangottini,78a,78b L. Fanò,78a,78b M. Magherini,78a,78b G. Mantovani,78a,78b V. Mariani,78a,78b M. Menichelli,78a F. Moscatelli,78a,aaa A. Piccinelli,78a,78b M. Presilla,78a,78b A. Rossi,78a,78b A. Santocchia,78a,78b D. Spiga,78a T. Tedeschi,78a,78b P. Azzurri,79a G. Bagliesi,79a V. Bertacchi,79a,79c L. Bianchini,79a T. Boccali,79a E. Bossini,79a,79b R. Castaldi,79a M. A. Ciocci,79a,79b V. D’Amante,79a,79d R. Dell’Orso,79a M. R. Di Domenico,79a,79d S. Donato,79a A. Giassi,79a F. Ligabue,79a,79c E. Manca,79a,79c G. Mandorli,79a,79c D. Matos Figueiredo,79a A. Messineo,79a,79b M. Musich,79a F. Palla,79a S. Parolia,79a,79b G. Ramirez-Sanchez,79a,79c 012002-23 A. TUMASYAN et al. PHYS. REV. D 106, 012002 (2022) A. Rizzi,79a,79b G. Rolandi,79a,79c S. Roy Chowdhury,79a,79c A. Scribano,79a N. Shafiei,79a,79b P. Spagnolo,79a R. Tenchini,79a G. Tonelli,79a,79b N. Turini,79a,79d A. Venturi,79a P. G. Verdini,79a P. Barria,80a M. Campana,80a,80b F. Cavallari,80a D. Del Re,80a,80b E. Di Marco,80a M. Diemoz,80a E. Longo,80a,80b P. Meridiani,80a G. Organtini,80a,80b F. Pandolfi,80a R. Paramatti,80a,80b C. Quaranta,80a,80b S. Rahatlou,80a,80b C. Rovelli,80a F. Santanastasio,80a,80b L. Soffi,80a R. Tramontano,80a,80b N. Amapane,81a,81b R. Arcidiacono,81a,81c S. Argiro,81a,81b M. Arneodo,81a,81c N. Bartosik,81a R. Bellan,81a,81b A. Bellora,81a,81b J. Berenguer Antequera,81a,81b C. Biino,81a N. Cartiglia,81a M. Costa,81a,81b R. Covarelli,81a,81b N. Demaria,81a B. Kiani,81a,81b F. Legger,81a C. Mariotti,81a S. Maselli,81a E. Migliore,81a,81b E. Monteil,81a,81b M. Monteno,81a M. M. Obertino,81a,81b G. Ortona,81a L. Pacher,81a,81b N. Pastrone,81a M. Pelliccioni,81a M. Ruspa,81a,81c K. Shchelina,81a F. Siviero,81a,81b V. Sola,81a A. Solano,81a,81b D. Soldi,81a,81b A. Staiano,81a M. Tornago,81a,81b D. Trocino,81a A. Vagnerini,81a,81b S. Belforte,82a V. Candelise,82a,82b M. Casarsa,82a F. Cossutti,82a A. Da Rold,82a,82b G. Della Ricca,82a,82b G. Sorrentino,82a,82b S. Dogra,83 C. Huh,83 B. Kim,83 D. H. Kim,83 G. N. Kim,83 J. Kim,83 J. Lee,83 S. W. Lee,83 C. S. Moon,83 Y. D. Oh,83 S. I. Pak,83 S. Sekmen,83 Y. C. Yang,83 H. Kim,84 D. H. Moon,84 B. Francois,85 T. J. Kim,85 J. Park,85 S. Cho,86 S. Choi,86 B. Hong,86 K. Lee,86 K. S. Lee,86 J. Lim,86 J. Park,86 S. K. Park,86 J. Yoo,86 J. Goh,87 A. Gurtu,87 H. S. Kim,88 Y. Kim,88 J. Almond,89 J. H. Bhyun,89 J. Choi,89 S. Jeon,89 J. Kim,89 J. S. Kim,89 S. Ko,89 H. Kwon,89 H. Lee,89 S. Lee,89 B. H. Oh,89 M. Oh,89 S. B. Oh,89 H. Seo,89 U. K. Yang,89 I. Yoon,89 W. Jang,90 D. Y. Kang,90 Y. Kang,90 S. Kim,90 B. Ko,90 J. S. H. Lee,90 Y. Lee,90 J. A. Merlin,90 I. C. Park,90 Y. Roh,90 M. S. Ryu,90 D. Song,90 I. J. Watson,90 S. Yang,90 S. Ha,91 H. D. Yoo,91 M. Choi,92 H. Lee,92 Y. Lee,92 I. Yu,92 T. Beyrouthy,93 Y. Maghrbi,93 K. Dreimanis,94 V. Veckalns,94,bbb M. Ambrozas,95 A. Carvalho Antunes De Oliveira,95 A. Juodagalvis,95 A. Rinkevicius,95 G. Tamulaitis,95 N. Bin Norjoharuddeen,96 Z. Zolkapli,96 J. F. Benitez,97 A. Castaneda Hernandez,97 L. G. Gallegos Maríñez,97 M. León Coello,97 J. A. Murillo Quijada,97 A. Sehrawat,97 L. Valencia Palomo,97 G. Ayala,98 H. Castilla-Valdez,98 E. De La Cruz-Burelo,98 I. Heredia-De La Cruz,98,ccc R. Lopez-Fernandez,98 C. A. Mondragon Herrera,98 D. A. Perez Navarro,98 R. Reyes-Almanza,98 A. Sánchez Hernández,98 S. Carrillo Moreno,99 C. Oropeza Barrera,99 F. Vazquez Valencia,99 I. Pedraza,100 H. A. Salazar Ibarguen,100 C. Uribe Estrada,100 J. Mijuskovic,101,ddd N. Raicevic,101 D. Krofcheck,102 P. H. Butler,103 A. Ahmad,104 M. I. Asghar,104 A. Awais,104 M. I. M. Awan,104 H. R. Hoorani,104 W. A. Khan,104 M. A. Shah,104 M. Shoaib,104 M. Waqas,104 V. Avati,105 L. Grzanka,105 M. Malawski,105 H. Bialkowska,106 M. Bluj,106 B. Boimska,106 M. Górski,106 M. Kazana,106 M. Szleper,106 P. Zalewski,106 K. Bunkowski,107 K. Doroba,107 A. Kalinowski,107 M. Konecki,107 J. Krolikowski,107 M. Araujo,108 P. Bargassa,108 D. Bastos,108 A. Boletti,108 P. Faccioli,108 M. Gallinaro,108 J. Hollar,108 N. Leonardo,108 T. Niknejad,108 M. Pisano,108 J. Seixas,108 O. Toldaiev,108 J. Varela,108 S. Afanasiev,109 D. Budkouski,109 I. Golutvin,109 I. Gorbunov,109 V. Karjavine,109 V. Korenkov,109 A. Lanev,109 A. Malakhov,109 V. Matveev,109,eee,fff V. Palichik,109 V. Perelygin,109 M. Savina,109 V. Shalaev,109 S. Shmatov,109 S. Shulha,109 V. Smirnov,109 O. Teryaev,109 N. Voytishin,109 B. S. Yuldashev,109,ggg A. Zarubin,109 I. Zhizhin,109 G. Gavrilov,110 V. Golovtcov,110 Y. Ivanov,110 V. Kim,110,hhh E. Kuznetsova,110,iii V. Murzin,110 V. Oreshkin,110 I. Smirnov,110 D. Sosnov,110 V. Sulimov,110 L. Uvarov,110 S. Volkov,110 A. Vorobyev,110 Yu. Andreev,111 A. Dermenev,111 S. Gninenko,111 N. Golubev,111 A. Karneyeu,111 D. Kirpichnikov,111 M. Kirsanov,111 N. Krasnikov,111 A. Pashenkov,111 G. Pivovarov,111 A. Toropin,111 V. Epshteyn,112 V. Gavrilov,112 N. Lychkovskaya,112 A. Nikitenko,112,jjj V. Popov,112 A. Stepennov,112 M. Toms,112 E. Vlasov,112 A. Zhokin,112 T. Aushev,113 O. Bychkova,114 M. Chadeeva,114,kkk P. Parygin,114 E. Popova,114 V. Rusinov,114 D. Selivanova,114 V. Andreev,115 M. Azarkin,115 I. Dremin,115 M. Kirakosyan,115 A. Terkulov,115 A. Belyaev,116 E. Boos,116 V. Bunichev,116 M. Dubinin,116,lll L. Dudko,116 A. Ershov,116 V. Klyukhin,116 O. Kodolova,116 I. Lokhtin,116 S. Obraztsov,116 M. Perfilov,116 S. Petrushanko,116 V. Savrin,116 V. Blinov,117,mmm T. Dimova,117,mmm L. Kardapoltsev,117,mmm A. Kozyrev,117,mmm I. Ovtin,117,mmm O. Radchenko,117,mmm Y. Skovpen,117,mmm I. Azhgirey,118 I. Bayshev,118 D. Elumakhov,118 V. Kachanov,118 D. Konstantinov,118 P. Mandrik,118 V. Petrov,118 R. Ryutin,118 S. Slabospitskii,118 A. Sobol,118 S. Troshin,118 N. Tyurin,118 A. Uzunian,118 A. Volkov,118 A. Babaev,119 V. Okhotnikov,119 V. Borshch,120 V. Ivanchenko,120 E. Tcherniaev,120 P. Adzic,121,nnn M. Dordevic,121 P. Milenovic,121 J. Milosevic,121 M. Aguilar-Benitez,122 J. Alcaraz Maestre,122 A. Álvarez Fernández,122 I. Bachiller,122 M. Barrio Luna,122 Cristina F. Bedoya,122 C. A. Carrillo Montoya,122 M. Cepeda,122 M. Cerrada,122 N. Colino,122 B. De La Cruz,122 A. Delgado Peris,122 J. P. Fernández Ramos,122 J. Flix,122 M. C. Fouz,122 O. Gonzalez Lopez,122 S. Goy Lopez,122 J. M. Hernandez,122 M. I. Josa,122 J. León Holgado,122 D. Moran,122 Á. Navarro Tobar,122 C. Perez Dengra,122 A. P´erez-Calero Yzquierdo,122 J. Puerta Pelayo,122 I. Redondo,122 L. Romero,122 S. Sánchez Navas,122 L. Urda Gómez,122 C. Willmott,122 J. F. de Trocóniz,123 B. Alvarez Gonzalez,124 J. Cuevas,124 C. Erice,124 J. Fernandez Menendez,124 012002-24 SEARCH FOR RESONANCES DECAYING TO THREE W … PHYS. REV. D 106, 012002 (2022) S. Folgueras,124 I. Gonzalez Caballero,124 J. R. González Fernández,124 E. Palencia Cortezon,124 C. Ramón Álvarez,124 V. Rodríguez Bouza,124 A. Soto Rodríguez,124 A. Trapote,124 N. Trevisani,124 C. Vico Villalba,124 J. A. Brochero Cifuentes,125 I. J. Cabrillo,125 A. Calderon,125 J. Duarte Campderros,125 M. Fernandez,125 C. Fernandez Madrazo,125 P. J. Fernández Manteca,125 A. García Alonso,125 G. Gomez,125 C. Martinez Rivero,125 P. Martinez Ruiz del Arbol,125 F. Matorras,125 P. Matorras Cuevas,125 J. Piedra Gomez,125 C. Prieels,125 A. Ruiz-Jimeno,125 L. Scodellaro,125 I. Vila,125 J. M. Vizan Garcia,125 M. K. Jayananda,126 B. Kailasapathy,126,ooo D. U. J. Sonnadara,126 D. D. C. Wickramarathna,126 W. G. D. Dharmaratna,127 K. Liyanage,127 N. Perera,127 N. Wickramage,127 T. K. Aarrestad,128 D. Abbaneo,128 J. Alimena,128 E. Auffray,128 G. Auzinger,128 J. Baechler,128 P. Baillon,128,a D. Barney,128 J. Bendavid,128 M. Bianco,128 A. Bocci,128 C. Caillol,128 T. Camporesi,128 M. Capeans Garrido,128 G. Cerminara,128 N. Chernyavskaya,128 S. S. Chhibra,128 S. Choudhury,128 M. Cipriani,128 L. Cristella,128 D. d’Enterria,128 A. Dabrowski,128 A. David,128 A. De Roeck,128 M. M. Defranchis,128 M. Deile,128 M. Dobson,128 M. Dünser,128 N. Dupont,128 A. Elliott-Peisert,128 F. Fallavollita,128,ppp A. Florent,128 L. Forthomme,128 G. Franzoni,128 W. Funk,128 S. Ghosh,128 S. Giani,128 D. Gigi,128 K. Gill,128 F. Glege,128 L. Gouskos,128 M. Haranko,128 J. Hegeman,128 V. Innocente,128 T. James,128 P. Janot,128 J. Kaspar,128 J. Kieseler,128 M. Komm,128 N. Kratochwil,128 C. Lange,128 S. Laurila,128 P. Lecoq,128 A. Lintuluoto,128 K. Long,128 C. Lourenço,128 B. Maier,128 L. Malgeri,128 S. Mallios,128 M. Mannelli,128 A. C. Marini,128 F. Meijers,128 S. Mersi,128 E. Meschi,128 F. Moortgat,128 M. Mulders,128 S. Orfanelli,128 L. Orsini,128 F. Pantaleo,128 E. Perez,128 M. Peruzzi,128 A. Petrilli,128 G. Petrucciani,128 A. Pfeiffer,128 M. Pierini,128 D. Piparo,128 M. Pitt,128 H. Qu,128 T. Quast,128 D. Rabady,128 A. Racz,128 G. Reales Guti´errez,128 M. Rovere,128 H. Sakulin,128 J. Salfeld-Nebgen,128 S. Scarfi,128 C. Schäfer,128 C. Schwick,128 M. Selvaggi,128 A. Sharma,128 P. Silva,128 W. Snoeys,128 P. Sphicas,128,qqq S. Summers,128 K. Tatar,128 V. R. Tavolaro,128 D. Treille,128 P. Tropea,128 A. Tsirou,128 J. Wanczyk,128,rrr K. A. Wozniak,128 W. D. Zeuner,128 L. Caminada,129,sss A. Ebrahimi,129 W. Erdmann,129 R. Horisberger,129 Q. Ingram,129 H. C. Kaestli,129 D. Kotlinski,129 U. Langenegger,129 M. Missiroli,129,sss L. Noehte,129,sss T. Rohe,129 K. Androsov,130,rrr M. Backhaus,130 P. Berger,130 A. Calandri,130 A. De Cosa,130 G. Dissertori,130 M. Dittmar,130 M. Doneg`a,130 C. Dorfer,130 F. Eble,130 K. Gedia,130 F. Glessgen,130 T. A. Gómez Espinosa,130 C. Grab,130 D. Hits,130 W. Lustermann,130 A.-M. Lyon,130 R. A. Manzoni,130 L. Marchese,130 C. Martin Perez,130 M. T. Meinhard,130 F. Nessi-Tedaldi,130 J. Niedziela,130 F. Pauss,130 V. Perovic,130 S. Pigazzini,130 M. G. Ratti,130 M. Reichmann,130 C. Reissel,130 T. Reitenspiess,130 B. Ristic,130 D. Ruini,130 D. A. Sanz Becerra,130 V. Stampf,130 J. Steggemann,130,rrr R. Wallny,130 C. Amsler,131,ttt P. Bärtschi,131 C. Botta,131 D. Brzhechko,131 M. F. Canelli,131 K. Cormier,131 A. De Wit,131 R. Del Burgo,131 J. K. Heikkilä,131 M. Huwiler,131 W. Jin,131 A. Jofrehei,131 B. Kilminster,131 S. Leontsinis,131 S. P. Liechti,131 A. Macchiolo,131 P. Meiring,131 V. M. Mikuni,131 U. Molinatti,131 I. Neutelings,131 A. Reimers,131 P. Robmann,131 S. Sanchez Cruz,131 K. Schweiger,131 M. Senger,131 Y. Takahashi,131 C. Adloff,132,uuu C. M. Kuo,132 W. Lin,132 A. Roy,132 T. Sarkar,132,oo S. S. Yu,132 L. Ceard,133 Y. Chao,133 K. F. Chen,133 P. H. Chen,133 P. s. Chen,133 H. Cheng,133 W.-S. Hou,133 Y. y. Li,133 R.-S. Lu,133 E. Paganis,133 A. Psallidas,133 A. Steen,133 H. y. Wu,133 E. Yazgan,133 P. r. Yu,133 B. Asavapibhop,134 C. Asawatangtrakuldee,134 N. Srimanobhas,134 F. Boran,135 S. Damarseckin,135,vvv Z. S. Demiroglu,135 F. Dolek,135 I. Dumanoglu,135,www E. Eskut,135 Y. Guler,135,xxx E. Gurpinar Guler,135,xxx C. Isik,135 O. Kara,135 A. Kayis Topaksu,135 U. Kiminsu,135 G. Onengut,135 K. Ozdemir,135,yyy A. Polatoz,135 A. E. Simsek,135 B. Tali,135,zzz U. G. Tok,135 S. Turkcapar,135 I. S. Zorbakir,135 G. Karapinar,136 K. Ocalan,136,aaaa M. Yalvac,136,bbbb B. Akgun,137 I. O. Atakisi,137 E. Gulmez,137 M. Kaya,137,cccc O. Kaya,137,dddd Ö. Özçelik,137 S. Tekten,137,eeee E. A. Yetkin,137,ffff A. Cakir,138 K. Cankocak,138,www Y. Komurcu,138 S. Sen,138,gggg S. Cerci,139,zzz I. Hos,139,hhhh B. Isildak,139,iiii B. Kaynak,139 S. Ozkorucuklu,139 H. Sert,139 D. Sunar Cerci,139,zzz C. Zorbilmez,139 B. Grynyov,140 L. Levchuk,141 D. Anthony,142 E. Bhal,142 S. Bologna,142 J. J. Brooke,142 A. Bundock,142 E. Clement,142 D. Cussans,142 H. Flacher,142 J. Goldstein,142 G. P. Heath,142 H. F. Heath,142 L. Kreczko,142 B. Krikler,142 S. Paramesvaran,142 S. Seif El Nasr-Storey,142 V. J. Smith,142 N. Stylianou,142,jjjj K. Walkingshaw Pass,142 R. White,142 K. W. Bell,143 A. Belyaev,143,kkkk C. Brew,143 R. M. Brown,143 D. J. A. Cockerill,143 C. Cooke,143 K. V. Ellis,143 K. Harder,143 S. Harper,143 M.-L. Holmberg,143,llll J. Linacre,143 K. Manolopoulos,143 D. M. Newbold,143 E. Olaiya,143 D. Petyt,143 T. Reis,143 T. Schuh,143 C. H. Shepherd-Themistocleous,143 I. R. Tomalin,143 T. Williams,143 R. Bainbridge,144 P. Bloch,144 S. Bonomally,144 J. Borg,144 S. Breeze,144 O. Buchmuller,144 V. Cepaitis,144 G. S. Chahal,144,mmmm D. Colling,144 P. Dauncey,144 G. Davies,144 M. Della Negra,144 S. Fayer,144 G. Fedi,144 G. Hall,144 M. H. Hassanshahi,144 G. Iles,144 J. Langford,144 L. Lyons,144 A.-M. Magnan,144 S. Malik,144 A. Martelli,144 D. G. Monk,144 J. Nash,144,nnnn M. Pesaresi,144 B. C. Radburn-Smith,144 D. M. Raymond,144 A. Richards,144 A. Rose,144 E. Scott,144 C. Seez,144 A. Shtipliyski,144 012002-25 A. TUMASYAN et al. PHYS. REV. D 106, 012002 (2022) A. Tapper,144 K. Uchida,144 T. Virdee,144,w M. Vojinovic,144 N. Wardle,144 S. N. Webb,144 D. Winterbottom,144 K. Coldham,145 J. E. Cole,145 A. Khan,145 P. Kyberd,145 I. D. Reid,145 L. Teodorescu,145 S. Zahid,145 S. Abdullin,146 A. Brinkerhoff,146 B. Caraway,146 J. Dittmann,146 K. Hatakeyama,146 A. R. Kanuganti,146 B. McMaster,146 N. Pastika,146 M. Saunders,146 S. Sawant,146 C. Sutantawibul,146 J. Wilson,146 R. Bartek,147 A. Dominguez,147 R. Uniyal,147 A. M. Vargas Hernandez,147 A. Buccilli,148 S. I. Cooper,148 D. Di Croce,148 S. V. Gleyzer,148 C. Henderson,148 C. U. Perez,148 P. Rumerio,148,oooo C. West,148 A. Akpinar,149 A. Albert,149 D. Arcaro,149 C. Cosby,149 Z. Demiragli,149 E. Fontanesi,149 D. Gastler,149 S. May,149 J. Rohlf,149 K. Salyer,149 D. Sperka,149 D. Spitzbart,149 I. Suarez,149 A. Tsatsos,149 S. Yuan,149 D. Zou,149 G. Benelli,150 B. Burkle,150 X. Coubez,150,x D. Cutts,150 M. Hadley,150 U. Heintz,150 J. M. Hogan,150,pppp T. Kwon,150 G. Landsberg,150 K. T. Lau,150 D. Li,150 M. Lukasik,150 J. Luo,150 M. Narain,150 N. Pervan,150 S. Sagir,150,qqqq F. Simpson,150 E. Usai,150 W. Y. Wong,150 X. Yan,150 D. Yu,150 W. Zhang,150 J. Bonilla,151 C. Brainerd,151 R. Breedon,151 M. Calderon De La Barca Sanchez,151 M. Chertok,151 J. Conway,151 P. T. Cox,151 R. Erbacher,151 G. Haza,151 F. Jensen,151 O. Kukral,151 R. Lander,151 M. Mulhearn,151 D. Pellett,151 B. Regnery,151 D. Taylor,151 Y. Yao,151 F. Zhang,151 M. Bachtis,152 R. Cousins,152 A. Datta,152 D. Hamilton,152 J. Hauser,152 M. Ignatenko,152 M. A. Iqbal,152 T. Lam,152 W. A. Nash,152 S. Regnard,152 D. Saltzberg,152 B. Stone,152 V. Valuev,152 Y. Chen,153 R. Clare,153 J. W. Gary,153 M. Gordon,153 G. Hanson,153 G. Karapostoli,153 O. R. Long,153 N. Manganelli,153 W. Si,153 S. Wimpenny,153 Y. Zhang,153 J. G. Branson,154 P. Chang,154 S. Cittolin,154 S. Cooperstein,154 N. Deelen,154 D. Diaz,154 J. Duarte,154 R. Gerosa,154 L. Giannini,154 J. Guiang,154 R. Kansal,154 V. Krutelyov,154 R. Lee,154 J. Letts,154 M. Masciovecchio,154 F. Mokhtar,154 M. Pieri,154 B. V. Sathia Narayanan,154 V. Sharma,154 M. Tadel,154 F. Würthwein,154 Y. Xiang,154 A. Yagil,154 N. Amin,155 C. Campagnari,155 M. Citron,155 G. Collura,155 A. Dorsett,155 V. Dutta,155 J. Incandela,155 M. Kilpatrick,155 J. Kim,155 B. Marsh,155 H. Mei,155 M. Oshiro,155 M. Quinnan,155 J. Richman,155 U. Sarica,155 F. Setti,155 J. Sheplock,155 P. Siddireddy,155 D. Stuart,155 S. Wang,155 A. Bornheim,156 O. Cerri,156 I. Dutta,156 J. M. Lawhorn,156 N. Lu,156 J. Mao,156 H. B. Newman,156 T. Q. Nguyen,156 M. Spiropulu,156 J. R. Vlimant,156 C. Wang,156 S. Xie,156 Z. Zhang,156 R. Y. Zhu,156 J. Alison,157 S. An,157 M. B. Andrews,157 P. Bryant,157 T. Ferguson,157 A. Harilal,157 C. Liu,157 T. Mudholkar,157 M. Paulini,157 A. Sanchez,157 W. Terrill,157 J. P. Cumalat,158 W. T. Ford,158 A. Hassani,158 G. Karathanasis,158 E. MacDonald,158 R. Patel,158 A. Perloff,158 C. Savard,158 N. Schonbeck,158 K. Stenson,158 K. A. Ulmer,158 S. R. Wagner,158 N. Zipper,158 J. Alexander,159 S. Bright-Thonney,159 X. Chen,159 Y. Cheng,159 D. J. Cranshaw,159 S. Hogan,159 J. Monroy,159 J. R. Patterson,159 D. Quach,159 J. Reichert,159 M. Reid,159 A. Ryd,159 W. Sun,159 J. Thom,159 P. Wittich,159 R. Zou,159 M. Albrow,160 M. Alyari,160 G. Apollinari,160 A. Apresyan,160 A. Apyan,160 L. A. T. Bauerdick,160 D. Berry,160 J. Berryhill,160 P. C. Bhat,160 K. Burkett,160 J. N. Butler,160 A. Canepa,160 G. B. Cerati,160 H. W. K. Cheung,160 F. Chlebana,160 K. F. Di Petrillo,160 J. Dickinson,160 V. D. Elvira,160 Y. Feng,160 J. Freeman,160 Z. Gecse,160 L. Gray,160 D. Green,160 S. Grünendahl,160 O. Gutsche,160 R. M. Harris,160 R. Heller,160 T. C. Herwig,160 J. Hirschauer,160 B. Jayatilaka,160 S. Jindariani,160 M. Johnson,160 U. Joshi,160 T. Klijnsma,160 B. Klima,160 K. H. M. Kwok,160 S. Lammel,160 D. Lincoln,160 R. Lipton,160 T. Liu,160 C. Madrid,160 K. Maeshima,160 C. Mantilla,160 D. Mason,160 P. McBride,160 P. Merkel,160 S. Mrenna,160 S. Nahn,160 J. Ngadiuba,160 V. Papadimitriou,160 K. Pedro,160 C. Pena,160,lll F. Ravera,160 A. Reinsvold Hall,160,rrrr L. Ristori,160 E. Sexton-Kennedy,160 N. Smith,160 A. Soha,160 L. Spiegel,160 S. Stoynev,160 J. Strait,160 L. Taylor,160 S. Tkaczyk,160 N. V. Tran,160 L. Uplegger,160 E. W. Vaandering,160 H. A. Weber,160 P. Avery,161 D. Bourilkov,161 L. Cadamuro,161 V. Cherepanov,161 R. D. Field,161 D. Guerrero,161 B. M. Joshi,161 M. Kim,161 E. Koenig,161 J. Konigsberg,161 A. Korytov,161 K. H. Lo,161 K. Matchev,161 N. Menendez,161 G. Mitselmakher,161 A. Muthirakalayil Madhu,161 N. Rawal,161 D. Rosenzweig,161 S. Rosenzweig,161 K. Shi,161 J. Wang,161 Z. Wu,161 E. Yigitbasi,161 X. Zuo,161 T. Adams,162 A. Askew,162 R. Habibullah,162 V. Hagopian,162 K. F. Johnson,162 R. Khurana,162 T. Kolberg,162 G. Martinez,162 H. Prosper,162 C. Schiber,162 O. Viazlo,162 R. Yohay,162 J. Zhang,162 M. M. Baarmand,163 S. Butalla,163 T. Elkafrawy,163,ssss M. Hohlmann,163 R. Kumar Verma,163 D. Noonan,163 M. Rahmani,163 F. Yumiceva,163 M. R. Adams,164 H. Becerril Gonzalez,164 R. Cavanaugh,164 S. Dittmer,164 O. Evdokimov,164 C. E. Gerber,164 D. J. Hofman,164 A. H. Merrit,164 C. Mills,164 G. Oh,164 T. Roy,164 S. Rudrabhatla,164 M. B. Tonjes,164 N. Varelas,164 J. Viinikainen,164 X. Wang,164 Z. Ye,164 M. Alhusseini,165 K. Dilsiz,165,tttt L. Emediato,165 R. P. Gandrajula,165 O. K. Köseyan,165 J.-P. Merlo,165 A. Mestvirishvili,165,uuuu J. Nachtman,165 H. Ogul,165,vvvv Y. Onel,165 A. Penzo,165 C. Snyder,165 E. Tiras,165,wwww O. Amram,166 B. Blumenfeld,166 L. Corcodilos,166 J. Davis,166 A. V. Gritsan,166 S. Kyriacou,166 P. Maksimovic,166 J. Roskes,166 M. Swartz,166 T. Á. Vámi,166 A. Abreu,167 J. Anguiano,167 C. Baldenegro Barrera,167 P. Baringer,167 A. Bean,167 Z. Flowers,167 T. Isidori,167 S. Khalil,167 J. King,167 G. Krintiras,167 012002-26 SEARCH FOR RESONANCES DECAYING TO THREE W … PHYS. REV. D 106, 012002 (2022) A. Kropivnitskaya,167 M. Lazarovits,167 C. Le Mahieu,167 C. Lindsey,167 J. Marquez,167 N. Minafra,167 M. Murray,167 M. Nickel,167 C. Rogan,167 C. Royon,167 R. Salvatico,167 S. Sanders,167 E. Schmitz,167 C. Smith,167 Q. Wang,167 Z. Warner,167 J. Williams,167 G. Wilson,167 S. Duric,168 A. Ivanov,168 K. Kaadze,168 D. Kim,168 Y. Maravin,168 T. Mitchell,168 A. Modak,168 K. Nam,168 F. Rebassoo,169 D. Wright,169 E. Adams,170 A. Baden,170 O. Baron,170 A. Belloni,170 S. C. Eno,170 N. J. Hadley,170 S. Jabeen,170 R. G. Kellogg,170 T. Koeth,170 Y. Lai,170 S. Lascio,170 A. C. Mignerey,170 S. Nabili,170 C. Palmer,170 M. Seidel,170 A. Skuja,170 L. Wang,170 K. Wong,170 D. Abercrombie,171 G. Andreassi,171 R. Bi,171 W. Busza,171 I. A. Cali,171 Y. Chen,171 M. D’Alfonso,171 J. Eysermans,171 C. Freer,171 G. Gomez Ceballos,171 M. Goncharov,171 P. Harris,171 M. Hu,171 M. Klute,171 D. Kovalskyi,171 J. Krupa,171 Y.-J. Lee,171 C. Mironov,171 C. Paus,171 D. Rankin,171 C. Roland,171 G. Roland,171 Z. Shi,171 G. S. F. Stephans,171 J. Wang,171 Z. Wang,171 B. Wyslouch,171 R. M. Chatterjee,172 A. Evans,172 J. Hiltbrand,172 Sh. Jain,172 M. Krohn,172 Y. Kubota,172 J. Mans,172 M. Revering,172 R. Rusack,172 R. Saradhy,172 N. Schroeder,172 N. Strobbe,172 M. A. Wadud,172 K. Bloom,173 M. Bryson,173 S. Chauhan,173 D. R. Claes,173 C. Fangmeier,173 L. Finco,173 F. Golf,173 C. Joo,173 I. Kravchenko,173 I. Reed,173 J. E. Siado,173 G. R. Snow,173,a W. Tabb,173 A. Wightman,173 F. Yan,173 A. G. Zecchinelli,173 G. Agarwal,174 H. Bandyopadhyay,174 L. Hay,174 I. Iashvili,174 A. Kharchilava,174 C. McLean,174 D. Nguyen,174 J. Pekkanen,174 S. Rappoccio,174 A. Williams,174 G. Alverson,175 E. Barberis,175 Y. Haddad,175 Y. Han,175 A. Hortiangtham,175 A. Krishna,175 J. Li,175 J. Lidrych,175 G. Madigan,175 B. Marzocchi,175 D. M. Morse,175 V. Nguyen,175 T. Orimoto,175 A. Parker,175 L. Skinnari,175 A. Tishelman-Charny,175 T. Wamorkar,175 B. Wang,175 A. Wisecarver,175 D. Wood,175 S. Bhattacharya,176 J. Bueghly,176 Z. Chen,176 A. Gilbert,176 T. Gunter,176 K. A. Hahn,176 Y. Liu,176 N. Odell,176 M. H. Schmitt,176 M. Velasco,176 R. Band,177 R. Bucci,177 M. Cremonesi,177 A. Das,177 N. Dev,177 R. Goldouzian,177 M. Hildreth,177 K. Hurtado Anampa,177 C. Jessop,177 K. Lannon,177 J. Lawrence,177 N. Loukas,177 D. Lutton,177 J. Mariano,177 N. Marinelli,177 I. Mcalister,177 T. McCauley,177 C. Mcgrady,177 K. Mohrman,177 C. Moore,177 Y. Musienko,177,eee R. Ruchti,177 A. Townsend,177 M. Wayne,177 M. Zarucki,177 L. Zygala,177 B. Bylsma,178 L. S. Durkin,178 B. Francis,178 C. Hill,178 M. Nunez Ornelas,178 K. Wei,178 B. L. Winer,178 B. R. Yates,178 F. M. Addesa,179 B. Bonham,179 P. Das,179 G. Dezoort,179 P. Elmer,179 A. Frankenthal,179 B. Greenberg,179 N. Haubrich,179 S. Higginbotham,179 A. Kalogeropoulos,179 G. Kopp,179 S. Kwan,179 D. Lange,179 D. Marlow,179 K. Mei,179 I. Ojalvo,179 J. Olsen,179 D. Stickland,179 C. Tully,179 S. Malik,180 S. Norberg,180 A. S. Bakshi,181 V. E. Barnes,181 R. Chawla,181 S. Das,181 L. Gutay,181 M. Jones,181 A. W. Jung,181 D. Kondratyev,181 A. M. Koshy,181 M. Liu,181 G. Negro,181 N. Neumeister,181 G. Paspalaki,181 S. Piperov,181 A. Purohit,181 J. F. Schulte,181 M. Stojanovic,181,s J. Thieman,181 F. Wang,181 R. Xiao,181 W. Xie,181 J. Dolen,182 N. Parashar,182 D. Acosta,183 A. Baty,183 T. Carnahan,183 M. Decaro,183 S. Dildick,183 K. M. Ecklund,183 S. Freed,183 P. Gardner,183 F. J. M. Geurts,183 A. Kumar,183 W. Li,183 B. P. Padley,183 R. Redjimi,183 J. Rotter,183 W. Shi,183 A. G. Stahl Leiton,183 S. Yang,183 L. Zhang,183,xxxx Y. Zhang,183 A. Bodek,184 P. de Barbaro,184 R. Demina,184 J. L. Dulemba,184 C. Fallon,184 T. Ferbel,184 M. Galanti,184 A. Garcia-Bellido,184 O. Hindrichs,184 A. Khukhunaishvili,184 E. Ranken,184 R. Taus,184 G. P. Van Onsem,184 B. Chiarito,185 J. P. Chou,185 A. Gandrakota,185 Y. Gershtein,185 E. Halkiadakis,185 A. Hart,185 M. Heindl,185 O. Karacheban,185,aa I. Laflotte,185 A. Lath,185 R. Montalvo,185 K. Nash,185 M. Osherson,185 S. Salur,185 S. Schnetzer,185 S. Somalwar,185 R. Stone,185 S. A. Thayil,185 S. Thomas,185 H. Wang,185 H. Acharya,186 A. G. Delannoy,186 S. Fiorendi,186 S. Spanier,186 O. Bouhali,187,yyyy M. Dalchenko,187 A. Delgado,187 R. Eusebi,187 J. Gilmore,187 T. Huang,187 T. Kamon,187,zzzz H. Kim,187 S. Luo,187 S. Malhotra,187 R. Mueller,187 D. Overton,187 D. Rathjens,187 A. Safonov,187 N. Akchurin,188 J. Damgov,188 V. Hegde,188 S. Kunori,188 K. Lamichhane,188 S. W. Lee,188 T. Mengke,188 S. Muthumuni,188 T. Peltola,188 I. Volobouev,188 Z. Wang,188 A. Whitbeck,188 E. Appelt,189 S. Greene,189 A. Gurrola,189 W. Johns,189 A. Melo,189 K. Padeken,189 F. Romeo,189 P. Sheldon,189 S. Tuo,189 J. Velkovska,189 M. W. Arenton,190 B. Cardwell,190 B. Cox,190 G. Cummings,190 J. Hakala,190 R. Hirosky,190 M. Joyce,190 A. Ledovskoy,190 A. Li,190 C. Neu,190 C. E. Perez Lara,190 B. Tannenwald,190 S. White,190 N. Poudyal,191 S. Banerjee,192 K. Black,192 T. Bose,192 S. Dasu,192 I. De Bruyn,192 P. Everaerts,192 C. Galloni,192 H. He,192 M. Herndon,192 A. Herve,192 U. Hussain,192 A. Lanaro,192 A. Loeliger,192 R. Loveless,192 J. Madhusudanan Sreekala,192 A. Mallampalli,192 A. Mohammadi,192 D. Pinna,192 A. Savin,192 V. Shang,192 V. Sharma,192 W. H. Smith,192 D. Teague,192 S. Trembath-Reichert,192 and W. Vetens192 (CMS Collaboration) 012002-27 A. TUMASYAN et al. PHYS. REV. D 106, 012002 (2022) 1Yerevan Physics Institute, Yerevan, Armenia 2Institut für Hochenergiephysik, Vienna, Austria 3Institute for Nuclear Problems, Minsk, Belarus 4Universiteit Antwerpen, Antwerpen, Belgium 5Vrije Universiteit Brussel, Brussel, Belgium 6Universit´e Libre de Bruxelles, Bruxelles, Belgium 7Ghent University, Ghent, Belgium 8Universit´e Catholique de Louvain, Louvain-la-Neuve, Belgium 9Centro Brasileiro de Pesquisas Fisicas, Rio de Janeiro, Brazil 10Universidade do Estado do Rio de Janeiro, Rio de Janeiro, Brazil 11Universidade Estadual Paulista, Universidade Federal do ABC, São Paulo, Brazil 12Institute for Nuclear Research and Nuclear Energy, Bulgarian Academy of Sciences, Sofia, Bulgaria 13University of Sofia, Sofia, Bulgaria 14Beihang University, Beijing, China 15Department of Physics, Tsinghua University, Beijing, China 16Institute of High Energy Physics, Beijing, China 17State Key Laboratory of Nuclear Physics and Technology, Peking University, Beijing, China 18Sun Yat-Sen University, Guangzhou, China 19Institute of Modern Physics and Key Laboratory of Nuclear Physics and Ion-beam Application (MOE)—Fudan University, Shanghai, China 20Zhejiang University, Hangzhou, China, Zhejiang, China 21Universidad de Los Andes, Bogota, Colombia 22Universidad de Antioquia, Medellin, Colombia 23University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, Split, Croatia 24University of Split, Faculty of Science, Split, Croatia 25Institute Rudjer Boskovic, Zagreb, Croatia 26University of Cyprus, Nicosia, Cyprus 27Charles University, Prague, Czech Republic 28Escuela Politecnica Nacional, Quito, Ecuador 29Universidad San Francisco de Quito, Quito, Ecuador 30Academy of Scientific Research and Technology of the Arab Republic of Egypt, Egyptian Network of High Energy Physics, Cairo, Egypt 31Center for High Energy Physics (CHEP-FU), Fayoum University, El-Fayoum, Egypt 32National Institute of Chemical Physics and Biophysics, Tallinn, Estonia 33Department of Physics, University of Helsinki, Helsinki, Finland 34Helsinki Institute of Physics, Helsinki, Finland 35Lappeenranta University of Technology, Lappeenranta, Finland 36IRFU, CEA, Universit´e Paris-Saclay, Gif-sur-Yvette, France 37Laboratoire Leprince-Ringuet, CNRS/IN2P3, Ecole Polytechnique, Institut Polytechnique de Paris, Palaiseau, France 38Universit´e de Strasbourg, CNRS, IPHC UMR 7178, Strasbourg, France 39Institut de Physique des 2 Infinis de Lyon (IP2I), Villeurbanne, France 40Georgian Technical University, Tbilisi, Georgia 41RWTH Aachen University, I. Physikalisches Institut, Aachen, Germany 42RWTH Aachen University, III. Physikalisches Institut A, Aachen, Germany 43RWTH Aachen University, III. Physikalisches Institut B, Aachen, Germany 44Deutsches Elektronen-Synchrotron, Hamburg, Germany 45University of Hamburg, Hamburg, Germany 46Karlsruher Institut fuer Technologie, Karlsruhe, Germany 47Institute of Nuclear and Particle Physics (INPP), NCSR Demokritos, Aghia Paraskevi, Greece 48National and Kapodistrian University of Athens, Athens, Greece 49National Technical University of Athens, Athens, Greece 50University of Ioánnina, Ioánnina, Greece 51MTA-ELTE Lendület CMS Particle and Nuclear Physics Group, Eötvös Loránd University, Budapest, Hungary 52Wigner Research Centre for Physics, Budapest, Hungary 53Institute of Nuclear Research ATOMKI, Debrecen, Hungary 54Institute of Physics, University of Debrecen, Debrecen, Hungary 012002-28 SEARCH FOR RESONANCES DECAYING TO THREE W … PHYS. REV. D 106, 012002 (2022) 55Karoly Robert Campus, MATE Institute of Technology, Gyongyos, Hungary 56National Institute of Science Education and Research, HBNI, Bhubaneswar, India 57Panjab University, Chandigarh, India 58University of Delhi, Delhi, India 59Saha Institute of Nuclear Physics, HBNI, Kolkata, India 60Indian Institute of Technology Madras, Madras, India 61Bhabha Atomic Research Centre, Mumbai, India 62Tata Institute of Fundamental Research-A, Mumbai, India 63Tata Institute of Fundamental Research-B, Mumbai, India 64Indian Institute of Science Education and Research (IISER), Pune, India 65Isfahan University of Technology, Isfahan, Iran 66Institute for Research in Fundamental Sciences (IPM), Tehran, Iran 67University College Dublin, Dublin, Ireland 68aINFN Sezione di Bari, Bari, Italy 68bUniversit`a di Bari, Bari, Italy 68cPolitecnico di Bari, Bari, Italy 69aINFN Sezione di Bologna, Universit`a di Bologna, Bologna, Italy 69bINFN Sezione di Bologna, Bologna, Italy 69cUniversit`a di Bologna, Bologna, Italy 70aINFN Sezione di Catania, Universit`a di Catania, Catania, Italy 70bINFN Sezione di Catania, Catania, Italy 70cUniversit`a di Catania, Catania, Italy 71aINFN Sezione di Firenze, Universit`a di Firenze, Firenze, Italy 71bINFN Sezione di Firenze, Firenze, Italy 71cUniversit`a di Firenze, Firenze, Italy 72INFN Laboratori Nazionali di Frascati, Frascati, Italy 73aINFN Sezione di Genova, Universit`a di Genova, Genova, Italy 73bINFN Sezione di Genova, Genova, Italy 73cUniversit`a di Genova, Genova, Italy 74aINFN Sezione di Milano-Bicocca, Universit`a di Milano-Bicocca, Milano, Italy 74bINFN Sezione di Milano-Bicocca, Milano, Italy 74cUniversit`a di Milano-Bicocca, Milano, Italy 75aINFN Sezione di Napoli, Universit`a di Napoli ‘Federico II’, Napoli, Italy, Universit`a della Basilicata, Potenza, Italy, Universit`a G. Marconi, Roma, Italy, Napoli, Italy 75bINFN Sezione di Napoli, Napoli, Italy 75cUniversit`a di Napoli ’Federico II’, Napoli, Italy 75dUniversit`a della Basilicata, Potenza, Italy 75eUniversit`a G. Marconi, Roma, Italy 76aINFN Sezione di Padova, Universit`a di Padova, Padova, Italy, Universit`a di Trento, Trento, Italy, Padova, Italy 76abINFN Sezione di Padova, Padova, Italy 76cUniversit`a di Padova, Padova, Italy 76dUniversit`a di Trento, Trento, Italy 77aINFN Sezione di Pavia, Pavia, Italy 77bUniversit`a di Pavia, Pavia, Italy 78aINFN Sezione di Perugia, Universit`a di Perugia, Perugia, Italy 78bINFN Sezione di Perugia, Perugia, Italy 78cUniversit`a di Perugia, Perugia, Italy 79aINFN Sezione di Pisa, Universit`a di Pisa, Scuola Normale Superiore di Pisa, Pisa Italy, Universit`a di Siena, Siena, Italy, Pisa, Italy 79bINFN Sezione di Pisa, Pisa, Italy 79cUniversit`a di Pisa, Pisa, Italy 79dScuola Normale Superiore di Pisa, Pisa, Italy 79eUniversit`a di Siena, Siena, Italy 80aINFN Sezione di Roma, Sapienza Universit`a di Roma, Rome, Italy, Rome, Italy 80bINFN Sezione di Roma, Rome, Italy 80cSapienza Universit`a di Roma, Rome, Italy 81aINFN Sezione di Torino, Universit`a di Torino, Torino, Italy, Universit`a del Piemonte Orientale, Novara, Italy, Torino, Italy 81bINFN Sezione di Torino, Torino, Italy 012002-29 A. TUMASYAN et al. PHYS. REV. D 106, 012002 (2022) 81cUniversit`a di Torino, Torino, Italy 81dUniversit`a del Piemonte Orientale, Novara, Italy 82aINFN Sezione di Trieste, Universit`a di Trieste, Trieste, Italy 82bINFN Sezione di Trieste, Trieste, Italy 82cUniversit`a di Trieste, Trieste, Italy 83Kyungpook National University, Daegu, Korea 84Chonnam National University, Institute for Universe and Elementary Particles, Kwangju, Korea 85Hanyang University, Seoul, Korea 86Korea University, Seoul, Korea 87Kyung Hee University, Department of Physics, Seoul, Republic of Korea, Seoul, Korea 88Sejong University, Seoul, Korea 89Seoul National University, Seoul, Korea 90University of Seoul, Seoul, Korea 91Yonsei University, Department of Physics, Seoul, Korea 92Sungkyunkwan University, Suwon, Korea 93College of Engineering and Technology, American University of the Middle East (AUM), Egaila, Kuwait, Dasman, Kuwait 94Riga Technical University, Riga, Latvia 95Vilnius University, Vilnius, Lithuania 96National Centre for Particle Physics, Universiti Malaya, Kuala Lumpur, Malaysia 97Universidad de Sonora (UNISON), Hermosillo, Mexico 98Centro de Investigacion y de Estudios Avanzados del IPN, Mexico City, Mexico 99Universidad Iberoamericana, Mexico City, Mexico 100Benemerita Universidad Autonoma de Puebla, Puebla, Mexico 101University of Montenegro, Podgorica, Montenegro 102University of Auckland, Auckland, New Zealand 103University of Canterbury, Christchurch, New Zealand 104National Centre for Physics, Quaid-I-Azam University, Islamabad, Pakistan 105AGH University of Science and Technology Faculty of Computer Science, Electronics and Telecommunications, Krakow, Poland 106National Centre for Nuclear Research, Swierk, Poland 107Institute of Experimental Physics, Faculty of Physics, University of Warsaw, Warsaw, Poland 108Laboratório de Instrumentação e Física Experimental de Partículas, Lisboa, Portugal 109Joint Institute for Nuclear Research, Dubna, Russia 110Petersburg Nuclear Physics Institute, Gatchina (St. Petersburg), Russia 111Institute for Nuclear Research, Moscow, Russia 112Institute for Theoretical and Experimental Physics named by A.I. Alikhanov of NRC ‘Kurchatov Institute’, Moscow, Russia 113Moscow Institute of Physics and Technology, Moscow, Russia 114National Research Nuclear University ’Moscow Engineering Physics Institute’ (MEPhI), Moscow, Russia 115P.N. Lebedev Physical Institute, Moscow, Russia 116Skobeltsyn Institute of Nuclear Physics, Lomonosov Moscow State University, Moscow, Russia 117Novosibirsk State University (NSU), Novosibirsk, Russia 118Institute for High Energy Physics of National Research Centre ‘Kurchatov Institute’, Protvino, Russia 119National Research Tomsk Polytechnic University, Tomsk, Russia 120Tomsk State University, Tomsk, Russia 121University of Belgrade: Faculty of Physics and VINCA Institute of Nuclear Sciences, Belgrade, Serbia 122Centro de Investigaciones Energ´eticas Medioambientales y Tecnológicas (CIEMAT), Madrid, Spain 123Universidad Autónoma de Madrid, Madrid, Spain 124Universidad de Oviedo, Instituto Universitario de Ciencias y Tecnologías Espaciales de Asturias (ICTEA), Oviedo, Spain 125Instituto de Física de Cantabria (IFCA), CSIC-Universidad de Cantabria, Santander, Spain 126University of Colombo, Colombo, Sri Lanka 127University of Ruhuna, Department of Physics, Matara, Sri Lanka 128CERN, European Organization for Nuclear Research, Geneva, Switzerland 129Paul Scherrer Institut, Villigen, Switzerland 130ETH Zurich—Institute for Particle Physics and Astrophysics (IPA), Zurich, Switzerland 131Universität Zürich, Zurich, Switzerland 132National Central University, Chung-Li, Taiwan 012002-30 SEARCH FOR RESONANCES DECAYING TO THREE W … PHYS. REV. D 106, 012002 (2022) 133National Taiwan University (NTU), Taipei, Taiwan 134Chulalongkorn University, Faculty of Science, Department of Physics, Bangkok, Thailand 135Çukurova University, Physics Department, Science and Art Faculty, Adana, Turkey 136Middle East Technical University, Physics Department, Ankara, Turkey 137Bogazici University, Istanbul, Turkey 138Istanbul Technical University, Istanbul, Turkey 139Istanbul University, Istanbul, Turkey 140Institute for Scintillation Materials of National Academy of Science of Ukraine, Kharkov, Ukraine 141National Scientific Center, Kharkov Institute of Physics and Technology, Kharkov, Ukraine 142University of Bristol, Bristol, United Kingdom 143Rutherford Appleton Laboratory, Didcot, United Kingdom 144Imperial College, London, United Kingdom 145Brunel University, Uxbridge, United Kingdom 146Baylor University, Waco, Texas, USA 147Catholic University of America, Washington, DC, USA 148The University of Alabama, Tuscaloosa, Alabama, USA 149Boston University, Boston, Massachusetts, USA 150Brown University, Providence, Rhode Island, USA 151University of California, Davis, Davis, California, USA 152University of California, Los Angeles, California, USA 153University of California, Riverside, Riverside, California, USA 154University of California, San Diego, La Jolla, California, USA 155University of California, Santa Barbara—Department of Physics, Santa Barbara, California, USA 156California Institute of Technology, Pasadena, California, USA 157Carnegie Mellon University, Pittsburgh, Pennsylvania, USA 158University of Colorado Boulder, Boulder, Colorado, USA 159Cornell University, Ithaca, New York, USA 160Fermi National Accelerator Laboratory, Batavia, Illinois, USA 161University of Florida, Gainesville, Florida, USA 162Florida State University, Tallahassee, Florida, USA 163Florida Institute of Technology, Melbourne, Florida, USA 164University of Illinois at Chicago (UIC), Chicago, Illinois, USA 165The University of Iowa, Iowa City, Iowa, USA 166Johns Hopkins University, Baltimore, Maryland, USA 167The University of Kansas, Lawrence, Kansas, USA 168Kansas State University, Manhattan, Kansas, USA 169Lawrence Livermore National Laboratory, Livermore, California, USA 170University of Maryland, College Park, Maryland, USA 171Massachusetts Institute of Technology, Cambridge, Massachusetts, USA 172University of Minnesota, Minneapolis, Minnesota, USA 173University of Nebraska-Lincoln, Lincoln, Nebraska, USA 174State University of New York at Buffalo, Buffalo, New York, USA 175Northeastern University, Boston, Massachusetts, USA 176Northwestern University, Evanston, Illinois, USA 177University of Notre Dame, Notre Dame, Indiana, USA 178The Ohio State University, Columbus, Ohio, USA 179Princeton University, Princeton, New Jersey, USA 180University of Puerto Rico, Mayaguez, Puerto Rico, USA 181Purdue University, West Lafayette, Indiana, USA 182Purdue University Northwest, Hammond, Indiana, USA 183Rice University, Houston, Texas, USA 184University of Rochester, Rochester, New York, USA 185Rutgers, The State University of New Jersey, Piscataway, New Jersey, USA 186University of Tennessee, Knoxville, Tennessee, USA 187Texas A&M University, College Station, Texas, USA 188Texas Tech University, Lubbock, Texas, USA 189Vanderbilt University, Nashville, Tennessee, USA 190University of Virginia, Charlottesville, Virginia, USA 191Wayne State University, Detroit, Michigan, USA 192University of Wisconsin—Madison, Madison, Wisconsin, USA 012002-31 A. TUMASYAN et al. PHYS. REV. D 106, 012002 (2022) aDeceased. bAlso at TU Wien, Wien, Austria. cAlso at Institute of Basic and Applied Sciences, Faculty of Engineering, Arab Academy for Science, Technology and Maritime Transport, Alexandria, Egypt. dAlso at Universit´e Libre de Bruxelles, Bruxelles, Belgium. eAlso at Universidade Estadual de Campinas, Campinas, Brazil. fAlso at Federal University of Rio Grande do Sul, Porto Alegre, Brazil. gAlso at The University of the State of Amazonas, Manaus, Brazil. hAlso at University of Chinese Academy of Sciences, Beijing, China. iAlso at Department of Physics, Tsinghua University, Beijing, China. jAlso at UFMS, Nova Andradina, Brazil. kAlso at The University of Iowa, Iowa City, Iowa, USA. lAlso at Nanjing Normal University Department of Physics, Nanjing, China. mAlso at University of Chinese Academy of Sciences, Beijing, China. nAlso at Institute for Theoretical and Experimental Physics named by A.I. Alikhanov of NRC ‘Kurchatov Institute’, Moscow, Russia. oAlso at Joint Institute for Nuclear Research, Dubna, Russia. pAlso at Cairo University, Cairo, Egypt. qAlso at Helwan University, Cairo, Egypt. rAlso at Zewail City of Science and Technology, Zewail, Egypt. sAlso at Purdue University, West Lafayette, Indiana, USA. tAlso at Universit´e de Haute Alsace, Mulhouse, France. uAlso at Tbilisi State University, Tbilisi, Georgia. vAlso at Erzincan Binali Yildirim University, Erzincan, Turkey. wAlso at CERN, European Organization for Nuclear Research, Geneva, Switzerland. xAlso at RWTH Aachen University, III. Physikalisches Institut A, Aachen, Germany. yAlso at University of Hamburg, Hamburg, Germany. zAlso at Isfahan University of Technology, Isfahan, Iran. aaAlso at Brandenburg University of Technology, Cottbus, Germany. bbAlso at Forschungszentrum Jülich, Juelich, Germany. ccAlso at Physics Department, Faculty of Science, Assiut University, Assiut, Egypt. ddAlso at Karoly Robert Campus, MATE Institute of Technology, Gyongyos, Hungary. eeAlso at Institute of Physics, University of Debrecen, Debrecen, Hungary. ffAlso at Institute of Nuclear Research ATOMKI, Debrecen, Hungary. ggAlso at Universitatea Babes-Bolyai—Facultatea de Fizica, Cluj-Napoca, Romania. hhAlso at MTA-ELTE Lendület CMS Particle and Nuclear Physics Group, Eötvös Loránd University, Budapest, Hungary. iiAlso at Wigner Research Centre for Physics, Budapest, Hungary. jjAlso at IIT Bhubaneswar, Bhubaneswar, India. kkAlso at Institute of Physics, Bhubaneswar, India. llAlso at Punjab Agricultural University, Ludhiana, India. mmAlso at Shoolini University, Solan, India. nnAlso at University of Hyderabad, Hyderabad, India. ooAlso at University of Visva-Bharati, Santiniketan, India. ppAlso at Indian Institute of Science (IISc), Bangalore, India. qqAlso at Indian Institute of Technology (IIT), Mumbai, India. rrAlso at Deutsches Elektronen-Synchrotron, Hamburg, Germany. ssAlso at Department of Physics, Isfahan University of Technology, Isfahan, Iran. ttAlso at Department of Electrical and Computer Engineering, Isfahan University of Technology, Isfahan, Iran. uuAlso at Department of Physics, University of Science and Technology of Mazandaran, Behshahr, Iran. vvAlso at INFN Sezione di Bari, Universit`a di Bari, Politecnico di Bari, Bari, Italy. wwAlso at Italian National Agency for New Technologies, Energy and Sustainable Economic Development, Bologna, Italy. xxAlso at Centro Siciliano di Fisica Nucleare e di Struttura Della Materia, Catania, Italy. yyAlso at Scuola Superiore Meridionale, Universit`a di Napoli Federico II, Napoli, Italy. zzAlso at Universit`a di Napoli ’Federico II’, Napoli, Italy. aaaAlso at Consiglio Nazionale delle Ricerche—Istituto Officina dei Materiali, Perugia, Italy. bbbAlso at Riga Technical University, Riga, Latvia. cccAlso at Consejo Nacional de Ciencia y Tecnología, Mexico City, Mexico. dddAlso at IRFU, CEA, Universit´e Paris-Saclay, Gif-sur-Yvette, France. eeeAlso at Institute for Nuclear Research, Moscow, Russia. 012002-32 SEARCH FOR RESONANCES DECAYING TO THREE W … PHYS. REV. D 106, 012002 (2022) fffAlso at National Research Nuclear University ’Moscow Engineering Physics Institute’ (MEPhI), Moscow, Russia. gggAlso at Institute of Nuclear Physics of the Uzbekistan Academy of Sciences, Tashkent, Uzbekistan. hhhAlso at St. Petersburg Polytechnic University, St. Petersburg, Russia. iiiAlso at University of Florida, Gainesville, Florida, USA. jjjAlso at Imperial College, London, United Kingdom. kkkAlso at P.N. Lebedev Physical Institute, Moscow, Russia. lllAlso at California Institute of Technology, Pasadena, California, USA. mmmAlso at Budker Institute of Nuclear Physics, Novosibirsk, Russia. nnnAlso at Faculty of Physics, University of Belgrade, Belgrade, Serbia. oooAlso at Trincomalee Campus, Eastern University, Sri Lanka, Nilaveli, Sri Lanka. pppAlso at INFN Sezione di Pavia, Universit`a di Pavia, Pavia, Italy. qqqAlso at National and Kapodistrian University of Athens, Athens, Greece. rrrAlso at Ecole Polytechnique F´ed´erale Lausanne, Lausanne, Switzerland. sssAlso at Universität Zürich, Zurich, Switzerland. tttAlso at Stefan Meyer Institute for Subatomic Physics, Vienna, Austria. uuuAlso at Laboratoire d’Annecy-le-Vieux de Physique des Particules, IN2P3-CNRS, Annecy-le-Vieux, France. vvvAlso at Şırnak University, Sirnak, Turkey. wwwAlso at Near East University, Research Center of Experimental Health Science, Nicosia, Turkey. xxxAlso at Konya Technical University, Konya, Turkey. yyyAlso at Piri Reis University, Istanbul, Turkey. zzzAlso at Adiyaman University, Adiyaman, Turkey. aaaaAlso at Necmettin Erbakan University, Konya, Turkey. bbbbAlso at Bozok Universitetesi Rektörlügü, Yozgat, Turkey. ccccAlso at Marmara University, Istanbul, Turkey. ddddAlso at Milli Savunma University, Istanbul, Turkey. eeeeAlso at Kafkas University, Kars, Turkey. ffffAlso at Istanbul Bilgi University, Istanbul, Turkey. ggggAlso at Hacettepe University, Ankara, Turkey. hhhhAlso at Istanbul University—Cerrahpasa, Faculty of Engineering, Istanbul, Turkey. iiiiAlso at Ozyegin University, Istanbul, Turkey. jjjjAlso at Vrije Universiteit Brussel, Brussel, Belgium. kkkkAlso at School of Physics and Astronomy, University of Southampton, Southampton, United Kingdom. uuuuAlso at Georgian Technical University, Tbilisi, Georgia. vvvvAlso at Sinop University, Sinop, Turkey. wwwwAlso at Erciyes University, Kayseri, Turkey. xxxxAlso at Institute of Modern Physics and Key Laboratory of Nuclear Physics and Ion-beam Application (MOE)—Fudan University, llllAlso at Rutherford Appleton Laboratory, Didcot, United Kingdom. mmmmAlso at IPPP Durham University, Durham, United Kingdom. nnnnAlso at Monash University, Faculty of Science, Clayton, Australia. ooooAlso at Universit`a di Torino, Torino, Italy. ppppAlso at Bethel University, St. Paul, Minneapolis, USA. qqqqAlso at Karamanoğlu Mehmetbey University, Karaman, Turkey. rrrrAlso at United States Naval Academy, Annapolis, Maryland, USA. ssssAlso at Ain Shams University, Cairo, Egypt. ttttAlso at Bingol University, Bingol, Turkey. Shanghai, China. yyyyAlso at Texas A&M University at Qatar, Doha, Qatar. zzzzAlso at Kyungpook National University, Daegu, Korea. 012002-33
null
10.1057_s41287-023-00576-y.pdf
null
null
The European Journal of Development Research (2023) 35:323–350 https://doi.org/10.1057/s41287-023-00576-y SPECIAL ISSUE ARTICLE Revealing the Relational Mechanisms of Research for Development Through Social Network Analysis  · Guillaume Fournie2 · Barbara Haesler2 · Grace Lyn Higdon1 · Marina Apgar1 Leah Kenny3 · Annalena Oppel3 · Evelyn Pauls3 · Matthew Smith4 · Mieke Snijder1 · Daan Vink2 · Mazeda Hossain3 Accepted: 6 December 2022 / Published online: 25 January 2023 © The Author(s) 2023 Abstract Achieving impact through research for development programmes (R4D) requires engagement with diverse stakeholders across the research, development and policy divides. Understanding how such programmes support the emergence of outcomes, therefore, requires a focus on the relational aspects of engagement and collaboration. Increasingly, evaluation of large research collaborations is employing social network analysis (SNA), making use of its relational view of causation. In this paper, we use three applications of SNA within similar large R4D programmes, through our work within evaluation of three Interidsiplinary Hubs of the Global Challenges Research Fund, to explore its potential as an evaluation method. Our comparative analysis shows that SNA can uncover the structural dimensions of interactions within R4D programmes and enable learning about how networks evolve through time. We reflect on common challenges across the cases including navigating different forms of bias that result from incomplete network data, multiple interpretations across scales, and the challenges of making causal inference and related ethical dilemmas. We conclude with lessons on the methodological and operational dimensions of using SNA within monitoring, evaluation and learning (MEL) systems that aim to support both learning and accountability. Keywords Social network analysis · Collaboration · Relational · Evaluation · Learning · Research for Development * Marina Apgar m.apgar@ids.ac.uk 1 Institute of Development Studies, University of Sussex, Library Road, Falmer, Brighton BN1 9RE, East Sussex, UK 2 Royal Veterinary College, 4 Royal College St, London NW1 0TU, UK 3 London School of Economics, Houghton St, London WC2A 2AE, UK 4 Edinburgh Napier University, Sighthill Campus, Sighthill Court, Edinburgh EH11 4BN, UK Vol.:(0123456789) 324 M. Apgar et al. Résumé Pour que les programmes de recherche pour le développement (R4D ou Research for Developmement en anglais) aient un impact, il faut un engagement entre diverses parties prenantes dans les domaines de la recherche, du développement et des poli- tiques. Il est nécessaire de se concentrer sur les aspects relationnels de l’engagement et de la collaboration si l’on souhaite comprendre la façon dont ce type de programme permet l’émergence de résultats. L’évaluation des grands consortia de recherche utilise de plus en plus fréquemment l’analyse des réseaux sociaux (SNA ou social network analysis en anglais) en appliquant sa vision relationnelle de la causalité. Dans cet article, en vue d’explorer son potentiel en tant que méthode d’évaluation, nous utilisons trois applications d’analyse des réseaux sociaux au sein de grands pro- grammes R4D similaires dans le cadre de notre travail d’évaluation de trois pôles interdisciplinaires du Fonds de recherche sur les défis mondiaux. Notre analyse comparative montre que l’analyse des réseaux sociaux peut révéler les dimensions structurelles des interactions au sein de ces programmes et permettre d’apprendre comment les réseaux évoluent dans le temps. Nous menons une réflexion quant aux défis communs qui émanent de ces cas, y compris la gestion de différentes formes de biais qui résultent de données de réseau incomplètes, de multiples interprétations sur des échelles différentes et les défis liés au fait d’établir une inférence causale et les dilemmes éthiques connexes. Nous concluons par des leçons sur les dimensions méthodologiques et opérationnelles de l’utilisation de l’analyse des réseaux sociaux dans les systèmes de suivi, d’évaluation et d’apprentissage (SEA) qui visent à soute- nir à la fois l’apprentissage et la redevabilité. Introduction Research for development programmes (R4D) aim to put research at the service of solving intractable development challenges, and often have a focus on improving livelihoods of marginalised or excluded populations. Achieving development out- comes for these populations requires engagement with diverse stakeholders across the research, development and policy divides. Relationships between partners within R4D networks are central mechanisms for shaping activities as well as engagement and impact strategies through collaboration (Temple et al. 2018). Outcomes emerge from these interactions, leading to uncertainty in their pathways to impact (Jacobi et al. 2020; Maru et al. 2018; Thornton et al. 2017). Understanding if and how R4D programmes support the emergence of outcomes, therefore, requires a focus on the relational aspects of engagement and collaboration. In the case of the Global Challenges Research Fund of the UK the funder, UKRI, set the ambition for the portfolio of R4D programmes well beyond the delivery of world class research, and included partnerships between the UK and the Global South as a desired outcome, requiring strong networks to be built between diverse stakeholders (Barr et al. 2019).1 The scale of the GCRF portfolio (initially proposed 1 https:// www. ukri. org/ our- work/ colla borat ing- inter natio nally/ global- chall enges- resea rch- fund/. 325 at £1.5billion) embracing R4D programme as network building initiatives, empha- sising collaboration and learning across the research and development sectors, cre- ated an unprecedented opportunity to deepen understanding of the relational mecha- nisms that contribute to achieving outcomes and impact. In this paper, we explore methodologies from within GCRF programmes used in evaluating them as network building initiatives. We focus on the use of social network analysis (SNA) as one tool in an R4D methodological repertoire. Although SNA alone is not sufficient to fully understand the contribution of these complex programmes to development outcomes and impact, our comparative analysis shows that it can uncover the structural dimensions of interactions within large R4D pro- grammes and enable learning about how networks evolve through time. We reflect on common challenges across the cases including navigating different forms of bias that result from incomplete network data, multiple interpretations across scales, the challenges of making causal inference and related ethical dilemmas. We con- clude with lessons on the methodological and operational dimensions of using SNA within monitoring, evaluation and learning (MEL) systems with dual aims of sup- porting both learning and accountability. SNA Within Complexity‑Aware Evaluation The uncertainty of impact pathways in R4D programmes, and the need to centre the network of social actors and their interactions throughout implementation call for evaluation designs that focus on explaining how change is unfolding in real time, often referred to as complexity-aware (Bamberger et al. 2015; Douthwaite and Hoffecker 2017; Gates and Fils‐Aime 2021; Patton 2010). These designs respond to understanding development programmes, policies and interventions as operat- ing under conditions of complexity, requiring multiple strategies and engagement with diverse actors within systems. Programme outcomes emerge from interac- tions between the parts (relationships between actors) rather than from what indi- vidual parts achieve alone (Hargreaves 2021; Walton 2016). This is even more evi- dent when programmes are working in conflict-affected contexts which are highly dynamic. Such programming requires non-linear evaluation designs to capture emergent outcomes through the interactions, as well as understanding achievement of intended outcomes, and emphasize iterative learning as change happens (Apgar et  al. 2020). These new approaches to evaluation offer opportunities for focussing on the interactions between actors in an R4D network. Within these broad designs, there is a need to zoom into the structural dimensions of collaboration in order to then explore causal relationships between networking and intended outcomes, along impact pathways. SNA is a recognised interdisciplinary methodological field within social sci- ence research, building on its sociological and mathematical (graph theory) roots (Freeman 2000). One of the central offerings of SNA is its relational view of causation, as Marin and Wellman (2011, p. 13) describe it “social net- work analysts argue that causation is not located in the individual, but in the social structure”. Using SNA as a method allows for intuitive visualisations of Revealing the Relational Mechanisms of Research for Development… 326 M. Apgar et al. relationships as well as tangible measures of “network quality” (Davies 2009). Analytical approaches for SNA are diversifying (including quantitative, qualita- tive, and mixed strategies), and combining structural and relational approaches to causation is leading to greater exploration of its use for evaluation. As a recent scoping review of the use of SNA in evaluation shows, there is a steady increase in its application since the turn of the century (Popelier 2018) increas- ing its potential to support evaluation of complex systems. A number of appli- cations are relevant to the R4D programming context (e.g. Aboelela et al. n.d.; Drew et  al. 2011; Haines et  al. 2011; Honeycutt and Strong 2012) and high- light both opportunities and challenges. In this paper, we add to this nascent field through comparative analysis of three experiences of SNA in the context of large R4D programmes. Methodology We use a case study methodology (Yin 1989) to learn within and across three applications of SNA in similar large interdisciplinary collaborations funded as Interdisciplinary Hubs by UKRI under the GCRF—we will refer to these R4D programme as ‘Hubs’. They have sufficient similarity in scale and approach to evaluation to support cross-case analysis, while each application is necessarily bespoke to its programme context and needs. Table  1 summarises each of the cases, showing that evaluation and research questions that drove the use of SNA in each Hub differ slightly, and consequently, the design of the data collection tools and analytical strategies also differ (justification for each analytical strat- egy can be found in the Online Technical Appendix). The Hubs experienced two major disruptions in the early phases of imple- mentation that influenced both the network formation processes and relatedly the application of the SNA method; (i) the COVID-19 pandemic required all Hubs to adapt to online collaboration and many network forming activities were no longer possible, and (ii) an unexpected and significant reduction in fund- ing (due to a reduction in overall UK government funding for ODA) led to loss of staff and reduced scope of monitoring activities for a 12-month period. Our focus in this paper, therefore, is necessarily on the initial phases of work. All three cases include a baseline application of SNA with the shared goal of assess- ing the way in which collaborations were shaped through the early phases of implementation, and where possible, how this was influenced by the disruptions experienced. We, the co-authors, are the designers and implementers of the SNA within the Hubs, involved as researchers, MEL specialists, data analysts and pro- gramme managers. The within-case analysis was carried out by each programme team independently, following its own strategy, and focussed on what the SNA revealed about the particular evaluation and learning goals. We were not exter- nal researchers using SNA to understand the programmes but active users of the method as a mechanism for programmatic learning through our positions within s e g n a h c g n i w o h s s i k r o w t e n g n i - g r e m e e h t r e h t e h w e t a g i t s e v n i d n a n o i t a s i l a r t n e c f o s m r e t n i n o i t a u l a v e f o t r a P . y h c r a r e i h y e k w o h e t a g i t s e v n i o t s m i a t a h t g n i n r a e l , n o i t a r o b a l l o c y l e m a n , g n i d l i u b y t i l i b a p a c f o s r a l l i p n i g n i v l o v e e r a , g n i r a h s d n a d n a s r e b m e m b u H g n o m a b u H e h t ) n o i t a r t s i n i m d a r e t f a e r e h t y l l a u n n a , n o i t s n o i t s e u q g n i n r a e l d n a n o i t a u l a v E s e t u b i r t t a l a n o i t a l e R s e t u b i r t t a e d o N y c n e u q e r f / e m a r f e m T i s b u H h c r a e s e r y r a n i l p i c s i d r e t n i d e d n u f F R C G n i A N S f o e s u f o s e s a c e e r h T 1 e l b a T b u H d e p a h s g n i e b e r a s n o i t a r o b a l , h c a e r t u o , h c r a e s e r . g . e ( s n o i t e n i l p i c s i d - p e c n i e c n i s s h t n o m 2 1 , ) e v i t - l o c h c i h w n i y a w e h t s s e s s a o T - c a r e t n i f o s e i r o g e t a c c fi i c e p S , y r t n u o c , e g a t s r e e r a c , r e d n e G - c e p s o r t e r ( n o i t p e c n i b u h e r o f e B b u H y r t l u o P h t l a e H e n O t c e j o r p h c r a e s e r r e d a o r b a f o t r a P n i g i r o , s n o i t c e n n o c f o h t g n e r t S , e g a t s r e e r a c , y h p a r g o e G - o r t e r ( n o i t p e c n i b u H e r o f e B y t i r u c e S d n a e c i t s u J , r e d n e G 327 p i h s n o i t a l e r e h t d n a t s r e d n u o t r e g r a l a d n a b u H e h t n e e w t e b d n a e c a e P , n e m o W f o k r o w t e n f o t r a P . s r e n o i t i t c a r p y t i r u c e S : 1 t c a p m I b u H f o n o i t a u l a v e y c a c o v d a d n a e g d e l w o n k w e N “ s e c i o v e h t y f i l p m a s k r o w t e n d e s i l a n i g r a m d n a n e m o w f o e g n a h c e s y l a t a c o t s p u o r g ” s e t i s d n a s e u s s i s s o r c a ) y t i v i t c a b u H c fi i c e p s , b u H , b u H - n o n ( s n o i t c e n n o c f o f o e p y t , ) e m e h t ( m a e r t s e c n i s s h t n o m 6 1 , ) e v i t c e p s n o i t u t i t s n i ) 0 2 0 2 e n u J ( n o i t p e c n i b u H Revealing the Relational Mechanisms of Research for Development… 328 M. Apgar et al. w o H : n o i t s e u q n o i t a u l a v e e h t o t k r o w t e n a g n i d l i u b b u H e h t s i d e g a g n e d n a y r a n i l p i c s i d r e t n i f o r e t a l n I ? n o i t c u d e r k s i r r e t s a s i d - u b i r t n o c e h t g n i t a u l a v e s e g a t s n a b r u n o d e s s u c o f s r e h c r a e s e r n i a m ( e c n a n r e v o g k s i r n a b r u r o f s a e r a t c a p m i d n a e m o c t u o n i s t f i h s o t b u H e h t f o n o i t ) b u H e h t e g a t s r e e r a c , n o i t u t i t s n i f o t r a p s a d n a , s t n e v e b u u H s w e i v e r l a u n n a s n o i t s e u q g n i n r a e l d n a n o i t a u l a v E s e t u b i r t t a l a n o i t a l e R s e t u b i r t t a e d o N y c n e u q e r f / e m a r f e m T i b u H g n i d n o p s e r s e g a t s l a i t i n I . n o i t a u ) l a m r o f n i d n a l a m r o f h t o b g n i f o e p y t r e d n e g , ) n o i t a i l ffi a r o j a m r e t f a y l t n e u q e s b u s d n a - l a v e d e s a b - y r o e h t f o t n e n o p m o C - r u t p a c ( s n o i t c a r e t n i f o h t g n e r t S y t i c ( y h p a r g o e g , e n i l p i c s i D n o i t a t n e m e l p m i f o s r a e y 2 r e t f A b u H s e i t i C s ’ w o r r o m o T ) d e u n i t n o c ( 1 e l b a T 329 each Hub. The diversity of roles we played has enabled analysis across methodo- logical, operational and strategic layers of use of SNA as an evaluation method. Learning from Use of SNA In this section, we summarise the application of SNA in each case and present the findings from within-case analysis. Full technical details of the SNA applications in each Hub are presented in the Online Technical Appendix and illustrate that analyti- cal strategies were specific to each case. In all cases, we reflect on whether the SNA findings primarily displayed aspects of project design (controlled) or social collabo- ration that occurs within the project (uncontrolled) in the early phases of programme implementation. One Health Poultry Hub The One Health Poultry Hub (OHPH) addresses zoonotic disease risks associated with poultry intensification, with a geographic focus on Bangladesh, India, Sri Lanka and Viet Nam. To address this challenge and ensure the safe and sustain- able production of poultry, it aims to promote interdisciplinary and cross-sectoral dialogue within a One Health environment. Indeed, given the cross-cutting nature of these issues, strengthening interdisciplinary research capability and competencies, collaboration and knowledge exchange are core activities. Assessment and monitor- ing of such attributes over the OHPH’s lifetime form part of the Hub’s MEL frame- work. Given the programme design we expected that (i) connections between study countries would be mainly mediated by a small number of UK partners in the early network, and (ii) the network structure would then become less centralised in the later study periods, with more direct connections between study country partners. A key principle driving the evaluation and so the demand by the UK management team was to produce learning to feed adaptive programme management and encour- age decentralised network growth. In this context, we applied SNA methods to investigate the evolution of the OHPH network, a dynamic partnership network consisting of approximately 120 named researchers from 27 institutions in 10 countries. Specific objectives were (i) to assess the way in which collaborations were being shaped among its members during the course of the project; (ii) to characterise the extent to which the emerg- ing network is dynamically changing across countries and research areas; and (iii) to investigate characteristics in the development of the OHPH network associated with factors such as career stage, scientific discipline and gender. Methods The SNA was conducted using data from two bespoke online surveys (see full tech- nical details in the Online Technical Appendix). The first was carried out in March Revealing the Relational Mechanisms of Research for Development… 330 M. Apgar et al. Fig. 1 Network diagrams showing the OHPH cohort networks (network of those who responded to all three time periods). Nodes are coloured according to the country in which they were based 2020, one year after the Hub’s launch, and the second in February 2021. All co- investigators and researchers engaged with the Hub, contracted research staff, post- graduate students and managerial staff were invited to respond (120 named research- ers). Respondents were asked to consider their collaborations and activities with all other Hub members over three periods: P0 (before the Hub’s inception), P1 (dur- ing the first year of the Hub) and P2 (during the second year of the Hub). In addi- tion, respondents were asked to indicate their primary scientific discipline or area of expertise, their primary role in the Hub, gender, and age category. Findings While some respondents filled the survey for all three periods, others provided information for only one or two. For each period, we, thus, considered two sets of nodes: all respondents who responded within each period (period-specific net- works), and respondents who completed all three questionnaires (cohort networks) for which changes in connection patterns over time among the same set of nodes can be assessed. The comparison of cohort and period-specific networks allows us to assess whether analytical results are affected by the composition of our sampled net- works (i.e. selection bias). All networks were undirected: if at least one respondent reported a collaboration with another respondent, an edge was constructed between them. The size of the period-specific networks ranged from 58 to 81 nodes (35 to 45% of Hub partners). The cohort networks had 37 nodes (see Online Technical Appendix for details). About two thirds of respondents were from the study coun- tries, and almost all others were based in the UK. Most respondents were male, bio- logical scientists, and at mid to late career stage. Each period-specific and cohort network showed a high small-world index which is indicative of high clustering and short path lengths between nodes (Humphries and Gurney 2008) See Fig. 1.  From P0 to P1, the proportion of connected dyads increased as the OHPH’s project activities started. This increase in connectedness was distributed among partners, reducing the extent to which a small number of 331 actors acted as mediators between most others. This small-world structure, and the evolution towards a reduction in centralisation would be expected to promote the diffusion of information and knowledge, and their equitable access by Hub mem- bers. Several face-to-face meetings were organised during P0 and P1, including a whole-Hub conference. Such meetings enabled partners from all disciplines and partner countries to meet and collaborate directly. This was likely a major driver in increasing the network connectedness and reducing its centralisation when com- pared to the pre-inception period. However, the COVID -19 pandemic effectively eliminated all such opportunities from the OHPH’s second year (which commenced in March 2020). Similar to other GCRF interdisciplinary Hubs, all OHPH-wide events, regular project coordination meetings, meetings of working groups leading the design and implementation of research, impact and learning activities, ad-hoc workshops, conferences, early career researcher group meetings and other opportu- nities for interaction were migrated to online platforms. Possibly as a result of this, we found that the network’s connectedness decreased from P1 to P2 as well as it becoming more centralised—that is, more connections were mediated by a small number of highly-connected nodes. This pattern was observed in the period-specific as well as cohort networks. We assessed whether the centrality of a node was associated with its attributes (country, discipline, career stage, gender) using multivariable permutation-based lin- ear models (as in Delabouglise et al. 2017). We considered two centrality measures: degree (the number of other nodes with which a node was connected) and between- ness (the extent to which a node lay on the shortest path between two others). For the period-specific network at P2, there was weak evidence that the average degree could be lower for study countries than for UK partners, and for women than for men. Active participation in P2 online events varied depending on internet access, bandwidth and quality. Not all participants had access to the required IT hardware when working from home, and it was apparent that online formats made participa- tion more challenging for partners for whom English is a second language. These associations were not, however, observed on the cohort network. Betweenness was not associated with any abovementioned node-level attributes. We assessed the possible influence of individual respondent factors (country, dis- cipline, career stage, gender) on the occurrence of edges between any two nodes. We found that, for all periods and network types, the likelihood of a connection increased if two partners were from the same country, but decreased if they were both from different study countries. By the second year of the Hub’s operation (P2), all UK and 82% of study country partners were engaged in connections between the UK and study countries, whereas connections across study countries only involved 47% of study country partners. The connectedness was higher among social sci- entists, mid and late career stage and male partners in the period-specific network. These associations were not, however, found in the cohort-specific networks. The design of the OHPH’s research programme, which was initiated in the second year, was to an extent replicated across the countries in which it was implemented (to enable standardisation and comparability of outcomes), but required modifications for each of the study sites to enable specific research questions to be addressed, as well as incorporating local differences. This required central coordination by the Revealing the Relational Mechanisms of Research for Development… 332 M. Apgar et al. core research and management team (based primarily in the UK) as well as intensive collaboration within the site teams. It is likely that this further contributed to cen- tralisation and compartmentalisation of the network over the P2 period. Moreover, online meetings might have made it more difficult for participants who may per- ceive themselves as being lower in the hierarchy of their institution (e.g. early career researchers) to express opinions or share knowledge. Reflections on Contributions of SNA for MEL While SNA has been useful to visualise and assess changes in connectivity and centrality of the OHPH’s network, care should be taken not to overinterpret these results. A major limitation in this analysis is the likely selection bias. Only a small proportion of the total hub partners (between 35 and 45%) took part in the study. The composition of the respondent groups was likely to be non-representative in the two surveys, as well as varying between the two surveys. Hence, depending on the factors which affected participation, some of the results may only be valid for the group of respondents and not the entire collaborative network, as suggested by the discrepancies in some analytical results between the cohort and period-specific networks. Nevertheless, the implementation of SNA using a repeated annual survey was found to be helpful to characterise the changing nature of the OHPH network over its lifetime, and provide insights into the dynamic processes and factors (some of which are external) which influence this. It allowed the Hub to assess whether the network structure was evolving towards the emergence of desirable characteris- tics, such as a reduced UK-focus centralisation, increased interactions across study countries and disciplines and reduced influence of one’s career stage and gender on node centrality. Although the limitations mentioned previously imply that the scope for SNA alone to explore or quantify such nuanced or complex issues is limited, it should be considered as a tool that can be applied alongside output-based indicator measurement approaches. Gender Justice and Security Hub Gendered political, economic and social injustices shape the outbreak and dynamics of conflict; war itself involves violence against women and girls as well as violations of other human rights; and redress from the gendered harms of war is intimately connected to the establishment of a lasting and just peace. In short, gender is a fun- damental web of social relations through which justice and security are mediated. The Gender, Justice and Security Hub (GJSH) responds to this challenge through bringing together 121 members across 42 partner organisations from a variety of 333 disciplinary perspectives, skill sets, career stages and geographic locations—with Afghanistan, Colombia, Kurdistan-Iraq, Lebanon, Sierra Leone, Sri Lanka and Uganda as focus countries and 17 more in which project work takes place. It aims to achieve the creation and growth of a network of academics, activists, practitioners and policymakers to advance progress towards gender justice and inclusive security in conflict-affected societies. A central component of the Hub’s MEL plan was tracking whether and how network collaborations and connections develop and change over time.. A quan- titative social network analysis was used to understand how collaborations were being shaped during the course of the project so that they could produce learning to inform programme adaptation. The study was to be applied at every year of the Hub’s lifetime (2020–2024); however, due to the ODA funding cuts, we only report on the first round of the survey in this paper. The objectives of the SNA were (i) to map the Hub network in order to visualise its overall structure, including density and strength of connections between members; (ii) to document how network con- nections change in number and strength over time; (iii) to identify patterns of con- nections within the network by attributes (by stream, career stage, geography, etc.) and (iv) to facilitate introductions among Hub members to improve the number and depth of collaborative relationships within the Hub. The expectation was that the Hub would be a fairly centralised network in its early stages given that the setting up of the Hub-level structure was run by a central Management Impact Commu- nications Administration (MICA) team based in the UK and the Executive Group, consisting of 12 Co-Directors leading six research streams.2 The Hub activities were designed to strengthen and expand connections over time between Hub members who are not in the core group during the initial setting up period, leading to a less centralised network structure. Methods Data were collected between June and December 2020. Our sampling framework included all individuals associated with the GJSH at the time of the study (121 individuals). This included individuals directly or indirectly involved with the Hub research and advocacy-related activities (e.g. administrative staff, research partners, management, activists and Hub Champions). Nearly half (45%) of GJSH members completed the full survey. (See further methodological details in the Online Techni- cal Appendix). Findings The analysis shows that one year after the establishment of the Hub, it was a mildly centralised network whereby a small number of actors at the centre of the network 2 The six research streams are thematic areas of work (Law and Policy Frameworks; Livelihood Land and Rights; Masculinities and Sexualities; Methodological Innovation; Migration and Displacement; Transformation and Empowerment). Revealing the Relational Mechanisms of Research for Development… 334 M. Apgar et al. Fig. 2 Network diagram showing all GJS Hub country networks Fig. 3 Network diagram showing all GJS Hub stream networks have a large number of connections across the Hub, and a majority have a small number of connections across the network. We further looked at whether actors with similar levels of connectivity are more likely to connect with each other or if it var- ies depending on their centralisation (e.g. the core group has a lot of ties to less well connected actors but they do not connect well with each other). We found that the central group is very well connected across the entire Hub but the individuals outside the core group are not well connected to each other. This matches the expectation 335 of how the setting up of the Hub would develop with the UK-based MICA team (see red dots in Fig. 2 and blue dots in Fig. 3) at the core. This small group of peo- ple (high degree actors), which also includes most members of the Executive Group (half of which are based in the UK), have connections to almost every individual in the network. We can see this confirmed in Fig. 2, which segments the network by country affiliation, clearly showing a core (red) group of UK-based Hub members. These findings confirm our initial design. Regarding connections within the different thematic streams of the Hub, mem- bers who responded to the survey had more connections external to their stream than within (see Fig. 3). This is most likely the result of streams having been established by the core group of the Hub, rather than being pre-existing research networks who joined the Hub as a whole. At the time of the survey, streams had only met in person once, at the Hub Convention in January 2020. Although we have only conducted one round of the survey so far, respondents could indicate if they had a connection with another member prior to the inception of the Hub and if so, how their relationship had changed over the last year of their joint Hub membership and what was driving this change. We found that 72% of tie changes were driven by the Hub (31% through the Convention and 41% through other Hub-related interactions). Roughly 60% of Hub members attended the in-per- son Convention in Sri Lanka in January 2020 and we would expect this to be the major event that introduced a lot of people to each other. In fact, 66% of respondents first met through a Hub-related activity. Apart from the Convention, members of the Hub mostly interact through their project work on the Hub, which most likely accounts for most of the ‘other Hub-related’ tie changes, but also, in the case of members of the management team and executive group, through regular meetings. Many members of the executive group were instrumental in putting together the ini- tial application for the Hub and setting up its operational structures so we would expect their relationships to have strengthened during the first year of the project. Reflections on Contribution of SNA for MEL The interpretation of SNA results is based on several assumptions from network theory. The most important one being that it is a good thing if individuals in a net- work have many ties and are well connected to other members of the network. This assumes that everyone wants or should want to connect, network and collaborate. It does, in the first instance, not take into account whether there might be disciplinary, gendered, or geographical differences, which might, for example, encourage work in small, closely knit teams—maybe because extremely sensitive work is taking place in conflict contexts—rather than across the entire Hub. Some of the projects e.g. in Colombia, Iraqi Kurdistan, Sierra Leone, or Uganda rely on extensive networks across the Hub for comparative work and policy influence nationally and with inter- national organisations—we would expect researchers and partners on these projects to be as connected as possible across the network. Other projects, e.g. in Afghani- stan or Sri Lanka, due to the sensitivity of the work on gender in Afghanistan (even before the Taliban takeover in August 2021), or on human rights violations and post- conflict issues in Sri Lanka, might look to connect extensively with communities Revealing the Relational Mechanisms of Research for Development… 336 M. Apgar et al. they are embedded and limit their interaction with foreign researchers and institu- tions as it might increase their exposure. There are two stages of growing this network over the course of the Hub. First, the Hub itself as a network had to be established by facilitating connections and collaborations between different members of the Hub who are from different geographies, disciplines and professions. In the early years of the Hub, we would likely expect a fairly centralised network structure with the Principal Investiga- tor (PI), Executive Group and the Management, Impact, Communication and Administration (MICA) team at its core. Those are the actors who were involved in the application stage and set up the governance structure of the Hub. The PI, MICA team and half of the Executive Group are Global North based and one aim of developing the Hub as a network is to decentralise it and facilitate more South-to-South connections. Hub members come from a variety of disciplinary perspectives, skill sets, career stages and geographic locations. The Hub model was designed to encompass feminist principles, which also includes an empha- sis on collaborative working and ensuring that opportunities, including network- ing opportunities, are equitably distributed among all Hub members. Since the inception of the Hub and especially in preparation for and during the in-person Convention in January 2020, almost all communication happened over Microsoft Teams, a cloud-based team collaboration software. It was used as a collabora- tive platform for whole-Hub interaction, including to communicate information from the central team, but also for stream- and project-level interaction with some projects using it as their main platform for data storage and virtual meet- ings. It also allowed for the quick creation of new communication channels when members expressed an interest, e.g. in arts-based approaches. Since Teams was already established as a platform for communication and all Hub members had access to it prior to the COVID-19 pandemic, the move to online-only interac- tion was fairly smooth. During the pandemic, Conventions were moved online and included cross-project collaborations and presentations. In the later stages of the Hub and after the SNA was completed, we have moved to co-creating outputs with Hub members from different projects, streams and countries such as books, papers, documentaries, and trainings. The SNA results allowed for the mapping of emerging relationships across the network and were designed to trace the development of relationships between and within groups including Global North and Global South partners, Early Career Researchers (ECRs) and more senior members, and practitioners and academics. Importantly, the SNA study aimed to provide tangible evidence on how networks like the GJS Hub can generate new insights through relationship development. The SNA would have been an important addition to surveys and anecdotal evidence in deciding where to target efforts of partnership building (as per objective iv). How- ever, in the absence of this longitudinal evidence due to the interrupted funding mid-way, the ability to do this in an adaptive and tailored way was significantly hampered. The second phase is to grow the Hub’s connections externally and link them to existing and emerging international networks and communities of practice on Women, Peace and Security, peacebuilding, International Law, and development. 337 It would have been possible to identify which Hub members would benefit from a closer connection with external networks—that could then be fostered—and which networks to tap into at a Hub level to create the most impact, especially on a policy level. The connections built with these other networks are ultimately what will facilitate local and global policy change and institutional reform to advance gender justice and sustainable peace, building on new knowledge and advocacy networks, which amplify marginalised voices across different conflict contexts. Tomorrow’s Cities Hub In a rapidly urbanising world, 60% of the area expected to be urban by 2030 remains to be built, opening a huge opportunity to build risk out of tomorrow’s cities. An initial assessment of the challenge showed that the disaster risk reduc- tion community (including scientists, policymakers, development practitioners and business leaders) is currently operating in disconnected communities of prac- tice and despite existing policy frameworks (e.g. Sendai Framework) approaches to disaster risk reduction are focussed on crisis management and not integrated into urban planning. The Tomorrow’s Cities Hub responds to this opportunity by working through research and development partnerships in four cities (Kath- mandu, Nairobi, Istanbul and Quito). Its aim is to co-produce interdisciplinary research on multi-hazard risk working with stakeholders in order to influence dis- aster risk reduction policy and practice. The co-produced research is implemented through a network consisting of 174 individuals from 54 partners including academic institutions, research centres, government departments and NGOs focussed in the four core cities. Given the aim of the Hub, strengthening capacity for interdisciplinary working as well as facilitating collaboration between researchers and stakeholders are critical to suc- cess. Monitoring how the hub network evolves through time, therefore, is a core component of the hubs MEL strategy. The drive for use of SNA as a tool for monitoring evolution of the hub network came from the evaluation and learning team, as part of a theory-based evaluation design. The intended users of the find- ings were the managers at central Hub level as well as in the city teams who were responsible for building an enabling environment for collaboration. SNA was used as a baseline to understand the extent to which cross-collabora- tion was occurring across different attributes of individuals including their gen- der, career level and location in the initial phases of implementation. The expec- tation based on the Hub design was for an initial centralised network given the central UK-based leadership team had built the proposal through their own rela- tionships with partners based in cities. Given the Hub’s commitment to equitable partnerships and to learn across the city contexts, we expected the network to evolve towards a less centralised structure through time. Intentionality in equita- ble partnerships and the potential of power asymmetries that funding structures across partners of the global North and South could reproduce the management Revealing the Relational Mechanisms of Research for Development… 338 M. Apgar et al. Fig. 4 Network diagram showing all TC Hub geographic location networks and research structures were built to enable work across UK and cities. Collabo- ration was encouraged through formation of cross-hub thematic research groups, as well as building and supporting a network of ECRs to co-produce research outputs. Regular all hub meetings were convened online and plans for the first in- person all hub conference in 2021 was curtailed by the pandemic. Methods Data were collected between February and March 2021, through two databases. An administrative database with generic information about collaborators in the hub and an online survey through the SumApp platform generated a specialised sur- vey to capture and visualise connections with other collaborators in the network. Our sampling framework included all individuals associated with the TC project at the time of the study. The total sample was 174 individuals and 53% made at least one connection (47% only show incoming ties and so we assume did not complete the survey) (for full details see Online Technical appendix). Self-identified female workers are slightly over-represented in the respondents (41% in respondents versus 23% non-respondents) which could account for more ties for women across the net- work, while ECRs are also slightly over-represented (47% respondents versus 37% non-respondents) which could also explain higher observed connections for ECRs overall. 339 Findings The analysis shows that after the establishment of the TC project, the network is fairly centralized. UK-based collaborators tend to be at the centre of the network and have a large number of connections across different locations—they are high degree actors. Yet, UK-based collaborators are also connected amongst themselves (purple nodes in Fig. 4). For example, connections realised (expressed as density) is twice as high among UK-based collaborators compared to those that work in LMIC coun- tries. This shows that overall collaborations are initiated by the UK as the central hub whereby the four city networks tend to be connected through UK-based and cross-city affiliated individuals (shown in Fig. 4). We then analysed collaborations between key attributes of location, career stage, gender and disciplines. We found that on average, individuals have 13 connections with those based at different locations. Yet only 9% (or an average of 1.2 connec- tions) of those occur between LMIC-based collaborators (South–South collabora- tion). Collaborations are higher between UK-based members than between LMIC located members (an average of 9.5 versus 4.2 connections, respectively). It is more likely for those located in the same location to collaborate if they have a common contact. These findings were expected given the project design. The UK represents the biggest segment with 80 collaborators who are linked to one or more cities and organized around disciplinary/thematically focussed groups that co-designed research within and across disciplines. A greater extent of shared contacts within members in similar locations may further reflect historical collaborations and con- textual knowledge held prior to establishing the new TC network. Collaboration also occurs among and across career stages. Peer-to-peer col- laboration is more frequent overall among those at mid and senior career stages (an average of 17) that among those at early career stage (an average of 12). This could be explained in part by the fact that most of the 77 ECRs were independently recruited new hires in the TC project and so did not have existing connections to each other. Yet when looking at location, this pattern is different. For those based in the UK, peer-to-peer connection is higher among mid and senior career mem- bers (16 connections on average) than among ECRs (6 connections on average). And cross-career level collaborations are 8 connections on average. In LMIC locations, however, peer-to-peer collaboration among ECRs is more common (7 connections on average), while peer-to-peer collaboration among mid and senior researchers is lower (average of 4). Cross-level collaborations are 7 connections on average. What this shows is that peer-to-peer collaboration of ECRs is driven largely by their loca- tion as is illustrated in Fig. 5. Lastly, regarding collaboration across genders, there are more connections among male collaborators overall (17 versus 12 among women). Yet, there is also col- laboration between men and women (an average of 13 connections) overall. Again we found the pattern differed depending on location. Men located in the UK are more likely to collaborate with other men as compared to men located in LMIC. For instance, collaboration between men located in the UK is higher (on average by 6 connections) than between women located in the UK. In LMIC, this is overall Revealing the Relational Mechanisms of Research for Development… 340 M. Apgar et al. Fig. 5 Network diagram showing collaboration among Early Career Researchers based outside of the UK for TC Hub more balanced. This suggests that gender influences how much members collaborate within their peer group. Reflections on Contribution of SNA for MEL This application of SNA was part of a staged and modular theory-based evalua- tion design that aimed to explore a core assumption of the hub’s theory of change around interdisciplinary working and equitable partnerships. The baseline appli- cation described the network at the outset in order: (i) to identify opportunities to enhance interdisciplinary and equitable partnerships as part of adaptive programme management; and, (ii) to inform impact evaluation design to assess the contribution of network collaborations to achieving intended shifts in urban planning (evaluating relationships beyond the hub). Visualization of the network overall and specific visual patterns in collaboration did enhance understanding of how the initial project design is reflected in the social fabric of collaboration. The fairly centralised initial network built confidence in the initial design with a large UK-based central team, and an explicit intent to build city- focussed research partnerships. While we did not have an ‘ideal network’ structure against which we intended to monitor the evolution of the network, we did anticipate that through time we would see greater collaboration between the non-UK-based members, in line with the intention to build equity in the partnership. 341 The analysis also enabled observation of unanticipated dynamics—such as gen- dered dynamics of collaboration between UK and non-UK-based individuals, and the marked difference between the way ECRs and more senior individuals are con- nected across the network. These structural patterns revealed  would need to be deepened through focus groups with hub collaborators to explore the drivers behind them and identify potential programme adaptations in line with the goal of moving towards a network structure with greater connections across more members. Visual- ising these patterns could also support existing conversations within the Hub around undertaking gender bias and power training as signalled by at least one of the city leadership teams as a priority. In this way, the application of SNA has shown poten- tial to produce learning that could influence the next phases of work to support col- laboration in the Hub. The application of these findings for decision-making processes, however, depends in large part on the quality of data and resulting findings. Data limitations are a common challenge in SNA (e.g. Wasserman and Faust 1994; Newman 2003) due to its exponential growth of observations and complexity. Further, SumApp required respondents to scroll through a list of all 174 people in the Hub which could result in biases, e.g. individuals mentioned towards the end could be less fre- quently selected due to response fatigue. This and other sources of response biases (which are common in SNA) influence the extent to which these findings constituted actionable learning for the Hub. Findings from Cross‑Case Analysis As R4D programmes, all three Hubs set out to intentionally build collaboration across disciplines, geographies and hierarchies. All cases share the dual (and inter- connected) objectives of (i) using SNA to monitor progress of their intended designs through describing and tracking how collaborative relationships change over time across significant attributes, and (ii) using the visualisations and resulting appre- ciation of the structure of the network to influence its development in intentional ways—to support ‘network weaving’ (Vance-Borland and Holley 2011). Our experi- ences are from the early phases of implementation. While we do not discuss result- ing adaptations, we do reflect on the opportunities for responding to learning that emerged when using SNA as a learning and an evaluation tool. Data Challenges and Respondent Bias The practical challenges of data collection and analysis in SNA are well described in the literature and relate, among other factors, to the extensive time required to respond to lengthy surveys leading to incomplete data sets with consequent implica- tions for rigorous understandings of whole networks (e.g. (Newman 2003; Penuel et  al. 2006; Popelier 2018). Particularly relevant to applications within evaluation are threats to construct validity that result from ambiguity in how relational attrib- utes are collected (how the type and strength of collaboration is described) and Revealing the Relational Mechanisms of Research for Development… 342 M. Apgar et al. relatedly, how these are interpreted (Popelier 2018). Incomplete datasets can lead to weak ties being under reported, influencing the overall validity of findings. Consequently, perhaps the most important step in the design of a SNA study in the context of evaluation is defining the ties, or connections, at the outset. This requires clarity of the aspects of collaboration that are of interest to the study, as well as knowledge of how these will be interpreted by respondents. All our cases are large networks (with over 100 members), and given the novelty of using SNA to explore relational aspects of research projects, choosing where to focus had to align with key evaluation and learning demands. As shown in Table  1, the relational attributes used in each case were driven by the specific and different evaluation objectives of each—OHPH mapped Hub-related collab- orations (including research, outreach, administrative activities), GJSH mapped both the strength (light, good, strong, none) and origin of connections (non-Hub, Hub, specific Hub activity), while TCH mapped both formal and informal inter- actions within the hub through strengths (in four categories). OHPH members were asked with whom they had worked on a range of activi- ties over pre-defined periods of time. However, the interpretation of the level of engagement which qualifies as “working together”, as well as the definition of a specific task is likely to vary among respondents. In the TCH case, four catego- ries or strengths of collaboration were used, but interpretations of each cannot be assumed to have been uniform. Given the size of the partnerships and the period of time covered by each survey, especially in the case of OHPH, recall bias cannot be excluded. Attempts were made to minimise them by including the list of all members in the questionnaire (and TCH and GJS included photos), so respondents were less likely to forget collaborators. In spite of these efforts, unsurprisingly, missing data was a challenge in all three cases. This was dealt with in different ways. OHPH avoided under estimat- ing connections by examining only partial networks, TCH used the reconstruc- tion method (Liu et al. 2019; Stork and Richards 1992) and assumed incoming ties are reciprocal for non-respondents in order to examine the full network, and GJSH examined both the full network (including non-respondents) and the par- tial network using the listwise approach (Pepinsky 2018). Complex model-based approaches, such as Baysean models, are gaining popularity, but are often dif- ficult to implement (requiring a complex model to be specified and estimated), and can result in introducing other forms of bias by imputing edges that over generalise the tendencies observed in other parts (i.e. information rich areas) of the network (Smith et al. 2022). In the context of SNA for learning, these more complex modelling strategies were not considered worth the additional time and effort. As we discuss later, there are inherent limits to what SNA can reveal on its own, and as part of broader MEL strategies, triangulation with other methods we posit is a better approach to mitigating the challenge of missing data. An obvious yet not insignificant response to overcoming the challenge of incomplete network data is to invest early in strategies to increase the proportion of Hub members participating in the survey. The TCH chose to use the SumApp survey in order to turn the SNA process into an explicit network weaving exer- cise, with the assumption that this would motivate hub members to respond. 343 SumApp creates a personal profile for each hub member with a unique URL and users can visualise the network real time as respondents update their con- nections (while the application is ‘live’). Feedback from ECR members sug- gests that this was indeed motivating for many of them because it aligned with their motivation to network within a large hub. Yet this did not necessarily hold true for other members of the Hub. Understanding what might motivate greater response, therefore, is an important step in planning SNA as a learning and net- work weaving tool. Challenges of Interpretation The challenge of confirmation bias in interpreting SNA findings in the context of programme evaluation is well described in the literature (e.g. (Popelier 2018)). Critics argue that limited ability to objectively interpret the results may lead to an interpretation which aligns with the investigators’ (and/or programme managers’) preconceived ideas rather than taking the data at face value, or indeed, pretend- ing that an ‘objective’ interpretation exists. This follows a gold standard view in evaluation that confirmation bias is to be avoided at all costs. In contrast, employ- ing complexity-aware evaluation designs, we were working within programmes implementing SNA as a participatory and learning-oriented evaluation method, working with (rather than controlling for) the experiences and aspirations of those involved in programme design (Apgar and Allen 2021). Interpretation of the find- ings, therefore, required the situated experiences of programme implementers and we embraced their interpretations (which are inherently biased) as an important explanatory device. As (Durland and Fredericks 2005) note the importance of specific network information should be seen as relative to programme needs at that particular time, embracing internal interpretation as the principle goal. In all three cases, the baseline application of SNA served as a useful empirical check on how the initial programme design was reflected in the social fabric of collaboration. The network structures that became visible at the initial phase were interpreted based on our expectations given the intentional designs. Table 2 provides a comparative view across the three cases. In all three cases, the network structures revealed in the early stages of the Hubs matched the expectation of centralised structures, based on their set up driven by the parameters of the funding set by UKRI. In all cases UK-based research leaders built the Hub networks initially through contracting partners and researchers based in countries of focus or operation. As others have shown, it is the repeated applications of SNA that enables a picture of evolution and change through time and brings it to life as a monitor- ing tool (Aboelela et al. n.d.; Provan et al. 2005). Yet as social networks are liv- ing and constantly evolving systems, we expect that they will organically shift in time and some of their dynamics will be unpredictable. This leads us to ask— how should we interpret the changes as part of monitoring the network? Some argue that lack of an ideal network structure means there cannot be standard Revealing the Relational Mechanisms of Research for Development… 344 M. Apgar et al. n a h t r e h t a r n i h t i w e v i t c a e r o m e r e w s n o i t a r o b a l l o c t a h t t n a e m y r t - n u o c y d u t s h c a e n i h t i w s e i t i v i t c a t c e j o r p f o n o i t a t n e m e l p m i e h t s e i r t n u o c y d u t s n e e w t e b . y r t n u o c e m a s e h t n i , t l u c ffi i d e r o m s u h t e r e w m u i t r o s n o c e h t s s o r c a s n o i t c a r e t n i e l i h W s r o t a r o b a l l o c n e e w t e b y l e k i l e r o m e b o t d n u o f s a w n o i t a r o b a l l o C s n o i t c a r e t n i . d e t c e p x e t o n s a w s d o i r e p e m i t d r i h t d n a d n o c e s e c a f - o t - e c a f d e t a t i l i c a f n i e g a g n e o t s r e b m e m k r o w t e n f o y t i l i b a n i e h t n e e w t e b s s e n d e t c e n n o c n i e s a e r c e d a e l i h w , d e t c e p x e s a w e h t o t d e l c i m e d n a p e h t h g u o r h t n o i t a r o b a l l o c o t n o i t p u r s i D s d o i r e p e m i t o w t t s r fi e h t s s o r c a s s e n d e t c e n n o c n i e s a e r c n i n A b u H y r t l u o P h t l a e H e n O s n o i t a r o b a l l o c g n i t s i x e - e r p d n u o r a d e m r o f g n i e b n a h t r e h t a r e m o s h t i w s r o t c a l l a o t d e t c e n n o c l l e w e r e w s r e b m e m k r o w t e n s r e b m e m d e s a b - K U y b s n o i t a r o b a l l o c w e n s a p u t e s e r e w s m a e r t S d e s i l a r t n e c , d e t c e p x e s a , n o i t a t n e m e l p m i f o r a e y e n o r e t f A b u H y t i r u c e S d n a e c i t s u J , r e d n e G s n o i t a n a l p x E e r u t c u r t s k r o w t e n d e t c e p x e s u s r e v e r u t c u r t s k r o w t e n d e v r e s b O s g n i d n fi A N S f o n o s i r a p m o C 2 e l b a T b u H h c a e f o e d i s t u o s r e b m e m h t i w d e h s i l b a t s e e r e w s n o i t a r o b a l l o c w e n , d e n o i s i v n e s A . s r e b m e m b u H l l a o t s n o i t c e n n o c g n i v a h . s e i t i v i t c a b u H o t e u d ) m a e r t s r o ( p u o r g c i t a m e h t f o e d i s t u o d n a n i h t i w s r e b m e m b u H h t i w s n o i t a r o b a l l o c r e t a e r G . d e t c e p x e n e e b e v a h d l u o w ) m a e r t s r o ( p u o r g c i t a m e h t h c a e - a r o b a l l o c f o s c i m a n y d d e r e d n e g e h t , b u H e h t n i s s e n e r a w a r e w o p c i t a m e h t r o y r a n i l p i c s i d n a h t r e h t a r d e t a c o l e r a y e h t e r e h w n e k a t r e d n e g o t s e h c a o r p p a t n e r e ff i d e h t y b d e n i a l p x e e b n a c n o i t - l o c h c u m w o h d e c n e u fl n i r e d n e g , r e h t r u F . s g n i p u o r g h c r a e s e r d n a r e d n e g n o n o i t c e fl e r h g u o r h t t l i u b g n i d n a t s r e d n u n o d e s a B y b y l e g r a l n e v i r d e b o t d n u o f s a w s R C E n e e w t e b n o i t a r o b a l l o C s n o i t a c o l d n a s e m e h t h c r a e s e r s s o r c a s r e d a e l y b t n e r e ff i d n i s m a e t d n a s p u o r g r e e p n i h t i w s n e p p a h n o i t a r o b a l g n i n n a l p t c a p m i . n g i s e d . s n o i t a c o l d n a n o i t a t n e m e l p m i d e t a r g e t n i r o f n o i t a r o b a l l o c n o i t a c o l n i h t i w l a i t i n i e h t g n i m r fi n o c , e r t n e c e h t t a s r o t a r o b a l l o c d e s a b - K U d e r i u q e r y l i r a s s e c e n b u H e h t f o n g i s e d h c r a e s e r d e s s u c o f - y t i c e h T h t i w , d e s i l a r t n e c s a w b u H e h t n o i t a t n e m e l p m i f o r a e y e n o r e t f A b u H s e i t i C s ’ w o r r o m o T 345 benchmarks for judging performance through use of SNA and this weakens its evaluation potential (e.g. (Haines et al. 2011)). Our intention was not to monitor progress against an ideal structure, but rather, to iteratively learn with and as the structure evolves. But assumptions from net- work theory are, often implicitly, applied. For example, assuming that a network is ‘stronger’ when individuals have many ties and are well connected to other members of the network. In large networks, such as R4D collaborations, however, it may in fact be that specialisation, or particular geographic clustering within the network is optimal for achievement of desired goals. Interpretation of the struc- ture, then, must follow the intention of the network, which in turn is always situ- ated in a particular moment and context. Our cases illustrate such a situated con- textual approach to using SNA. In all cases, we held assumptions about a desirable evolution of the hub struc- ture in line with the stated goals of equity across specific hierarchies of power (such as gender, Global North-Global South, career level) (see Table  2 for observed versus expected network structures). These assumptions are aligned with the theory of change of the GCRF funding mechanism overall and the pro- grammes specifically. The OHPH interpreted their findings based on an assumed evolution along a spectrum—a hypothetical star network (with the PI being connected to everyone and no link between others) at one end, and a complete network (with everyone connected to each other) at the other end. Indeed the SNA across time periods revealed an unexpected surge in centralisation in the last study period which suggested movement in the opposite direction. These observed trends in network development could be related to the move from face- to-face to online interactions due to COVID-19, adding some confidence that they are reflective of ‘actual’ processes. This finding can help to identify individuals between which collaborations should be fostered, and support thinking further about how the hub could recover post-pandemic to continue to build collabora- tion in the ways it intended. The GJSH captured connections prior to the inception of the Hub (retrospec- tively) and then 16 months into the project (time of the survey). This showed how the Hub was set up and how members started to connect. The importance of in- person all-Hub Conventions was underlined as it was both a key event where peo- ple first met each other as well as a key driver of strengthening connections. Over the course of the pandemic, similar to the OHPH, Conventions moved online and while this continued the opportunity to meet officially, it greatly reduced the chances for spontaneous interactions which are most likely to strengthen relationships. What our experiences suggest is that whether implicit or explicit, the programme designs were manifest through the baseline and subsequent application in a way that made them visible. This offered the opportunity to challenge earlier assumptions and to identify hidden or unexpected dimensions that warrant further exploration, as well as mechanisms to nudge the network in desirable directions. In the case of the TCH baseline, we saw that even in the early stages some unexpected dynam- ics were revealed—such as collaborations between female members based outside of the UK being greater than for female members based in the UK. The TCH had Revealing the Relational Mechanisms of Research for Development… 346 M. Apgar et al. procedures for reflecting on the ways in which power imbalances influenced internal team dynamics and how equitable decision making was. This was part of a broader intention to work in equitable ways (Snijder et al. this issue). Revealing the gendered dynamics of collaboration could add further weight to the requests of some city management teams to provide gender sensitivity training to all network members. It also allowed questioning an underpinning (positivist) premise of ‘more is better’. Starting with a deliberate project design as we did, network studies as an M&E tool offer the opportunity to re-evaluate and -classify measures from an aspect of appli- cability for the ‘ideal function’ rather than ‘ideal structure’ of a network. Strengthening Causal Inference The ways in which SNA can support causal inference are still debated within the evaluation literature, and many applications of SNA still struggle to determine causal relationships between internal network structure and external network out- comes. In the context of evaluating R4D programmes, being able to make this link is important to add weight to SNA as an evaluation method. The structural paradigm of SNA suggests that structural mechanisms influence how changes unfold, yet the relational aspects are so many, indeed potentially infinite, that establishing clear causal links is challenging (Doreian 2001; VanderWeele and An 2013). A response to this dilemma is to situate the structural analysis within a contextualized theory about how the causal inference is hypothesized (a causal theory of change). As discussed above, all three of our SNA cases were part of broader evalua- tion designs. While all three are focussed on visualising and describing the ‘inter- nal’ networks, in TCH and GJSH, the intention (prior to funding cuts) was to use understanding of how the internal network is working (and evolving through time), alongside other evaluation research on equitable partnerships to build con- tribution claims around how the network structure (and ways collaborations are taking shape across hierarchies and power structures) contributes to intended pro- gramme outcomes. In the case of TCH, the focus of external interactions is on influencing the co-production of risk informed urban planning, while in the case of GJSH, the focus is on the end goal of shifting patriarchal modes of knowledge production on sustainable peace. In this way, using SNA as a monitoring tool can not only build a picture of how the internal structure is evolving but also can offer data points to support theory-based evaluation. Additional methods are required to enhance interpretation and reveal the mean- ing given to relations identified and so to crystalise key causal  mechanisms for evaluation  to investigate further. Often quantitative SNA is complemented with qualitative approaches (Hopkins 2017; Kolleck 2016). All three Hubs have mul- tiple other sources of monitoring data that could be used to supplement the SNA. TCH, for example, developed outcome case studies for monitoring outcomes, and implemented a survey on interdisciplinary working to shed light on some of the patterns revealed. For example, an outcome case study on interdisciplinary work- ing in Quito evidences how intentional reflection on ways of working and building 347 of interdisciplinary capacities within the team has opened up opportunities for greater engagement with local partners for multi-hazard risk research. This quali- tative analysis can support causal claims around how collaboration between team members from different disciplines (visualised through SNA) which produce internal outcomes (such as openness to participatory methods) relates to move- ment along a desired pathway towards equitable outcomes. While SNA is not a causal methodology, it can provide evidence of collaboration and thus can con- firm or disconfirm achievement of early and internal desired outcomes in the way the programme (network) is set up to deliver impact. Ethical Dilemmas in SNA for Evaluation Making visible how individuals are interacting with colleagues or partners comes with ethical challenges and risks if the results are interpreted as judgements of indi- vidual performance (Kadushin 2005; Penuel et  al. 2006). The GCRF Hubs were funded to intentionally support collaboration across established hierarchies of power, turning collaborative behaviours into expectations. Anonymization, which is the standard approach to managing research ethics and minimizing risk to partici- pants, is often not possible and usually not desirable when using SNA to intention- ally support network weaving. The GJS Hub experienced a concern that members might alter how they reported a connection based on a perception of how socially acceptable that connection might be. Of course any reported connection is a subjective measure of that individual’s perception of a relationship but there is still likely to be systematic under- or over- reporting. For example, it is likely that an ECR might underreport connections with more senior scholars so as not to assume a ‘strong’ relationship when that percep- tion might not be reciprocal. Similarly, there might be cultural differences in percep- tions of relationship strength, where, for example, a US-based member of the Hub might consistently rate relationships as stronger than UK-based Hub members. As discussed above, the TCH used the SumApp online tool to support intentional network weaving which was experienced as motivating for ECRs. On the other hand, accessibility of all connections to everyone in the Hub might have deterred some from reporting for similar reasons described for GJSH. Visualising your own posi- tion in a network that aspires to be collaborative could, therefore, be experienced as positive or negative depending on how connected you are and how much you value connection. One advantage is that individual learning, and first person reflection, is made possible through seeing oneself as part of the network. But this requires net- work members to value this capability for self-reflections above any concerns they have about being judged based on their position in a network. Revealing the Relational Mechanisms of Research for Development… 348 M. Apgar et al. Lessons for Future Use of SNA in Evaluation of R4D Programmes Across the three cases of SNA in evaluation of large R4D programmes, we have illustrated that in spite of the challenges with data and interpretation (which are common to SNA), it is useful as a monitoring tool when used to reflect on under- lying assumptions about collaboration and resulting network structures. From our analysis we conclude with three lessons for future use of SNA within evaluation of R4D programmes. 1. The more explicit assumptions about collaboration are at the outset, the more useful the empirical view of collaboration revealed is to programme learning. A contextualized theory of collaboration could be created at the outset to guide the SNA study. This is in line with Davies’ (2009) call for a theory-based and deduc- tive approach to SNA in evaluation. 2. Combining SNA with other methods can enhance interpretation and reveal the meaning given to structural views. This can strengthen causal inference about relational causal mechanisms making SNA a necessary, but not sufficient method to evaluate R4D programmes. 3. Navigating the challenges of interpretation and ethical dilemmas requires careful consideration as well as an enabling institutional and political environment for use of SNA to support learning. Embedding the interpretation of SNA findings within participatory learning moments (such as after-action reviews) would strengthen the use of SNA findings in learning-oriented evaluation design, as suggested by others (Drew et al. 2011; Durland and Fredericks 2005) Supplementary Information The online version contains supplementary material available at https:// doi. org/ 10. 1057/ s41287- 023- 00576-y. Funding Funding was provided by Natural Environment Research Council. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Com- mons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/. References Aboelela, S.W., J.A.M. Rn, K.M. Carley, and E.L. Rn. 2007. Social Network Analysis to Evaluate an Interdisciplinary Research Center. Journal of Research Administration 38: 15. Apgar, M., and W. Allen. 2021. Participatory Monitoring, Evaluation and Learning: Taking Stock and Breaking New Ground. In SAGE Handbook of Participatory Research and Inquiry, ed. D. Burns, J. Howard, and S. Ospina. London: Sage Publications. 349 Apgar, Marina, K. Hernandez, and G. Ton. 2020. Contribution Analysis for Adaptive Management. 14. Bamberger, M., J. Vaessen, and E. Raimondo. 2015. Dealing with Complexity in Development Evalua- tion: A Practical Approach. London: Sage Publications. Barr, J., P. Simmonds, B. Bryan, and I. Vogel. 2019. Inception report—Global Challenge Research Fund (GCRF) Evaluation—Foundation Stage.https:// doi. org/ 10. 13140/ RG.2. 2. 31983. 89762 Davies, R. 2009. The Use of Social Network Analysis Tools in the Evaluation of Social Change Com- munications C.pdf. https:// mande. co. uk/ wp- conte nt/ uploa ds/ 2018/ 08/ The% 20Use% 20of% 20Soc ial% 20Net work% 20Ana lysis% 20Too ls% 20in% 20the% 20Eva luati on% 20of% 20Soc ial% 20Cha nge% 20Com munic ations% 20C. pdf Delabouglise, A., N. Antoine‐Moussiaux, D. Tatong, A. Chumkaeo, A. Binot, G. Fournié, E. Pilot, W. Phimpraphi, S. Kasemsuwan, M.C. Paul, and R. Duboz. 2017. Cultural practices shaping zoonotic diseases surveillance: the case of highly pathogenic avian influenza and Thailand native chicken farmers. Transboundary and emerging diseases 64 (4): 1294–1305. Doreian, P. 2001. Causality in Social Network Analysis. Sociological Methods & Research 30 (1): 81–114. https:// doi. org/ 10. 1177/ 00491 24101 03000 1005. Douthwaite, B., and E. Hoffecker. 2017. Towards a Complexity-Aware Theory of Change for Partici- patory Research Programs Working Within Agricultural Innovation Systems. Agricultural Systems 155: 88–102. https:// doi. org/ 10. 1016/j. agsy. 2017. 04. 002. Drew, R., P. Aggleton, P. Boyce, H. Chalmers, C. Maxwell, S. Pachauri, F. Thomas, I. Warwick, and K. Wood. 2011. Social Network Analysis to Evaluate Organisational Networks on Sexual Health and Rights. Development in Practice 21 (8): 1062–1079. https:// doi. org/ 10. 1080/ 09614 524. 2011. 590884. Durland, M.M., and K.A. Fredericks. 2005. An Introduction to Social Network Analysis. New Directions for Evaluation 2005 (107): 5–13. https:// doi. org/ 10. 1002/ ev. 157. Freeman, L.C. 2000. Social Network Analysis: Definition and History. New York: Oxford University Press. Gates, E.F., and F. Fils-Aime. 2021. System Change Evaluation: Insights from The Rippel Foundation and its ReThink Health Initiative. New Directions for Evaluation 2021 (170): 125–138. https:// doi. org/ 10. 1002/ ev. 20462. Haines, V.A., J. Godley, and P. Hawe. 2011. Understanding Interdisciplinary Collaborations as Social Networks. American Journal of Community Psychology 47 (1–2): 1–11. https:// doi. org/ 10. 1007/ s10464- 010- 9374-1. Hargreaves, M. 2021. Bricolage: A Pluralistic Approach to Evaluating Human Ecosystem Initiatives. New Directions for Evaluation 2021 (170): 113–124. https:// doi. org/ 10. 1002/ ev. 20460. Honeycutt, T.C., and D.A. Strong. 2012. Using Social Network Analysis to Predict Early Collaboration Within Health Advocacy Coalitions. American Journal of Evaluation 33 (2): 221–239. https:// doi. org/ 10. 1177/ 10982 14011 424201. Hopkins, M. 2017. A Review of Social Network Analysis and Education: Theory, Methods, and Applica- tions. Journal of Educational and Behavioral Statistics 42 (5): 639–646. https:// doi. org/ 10. 3102/ 10769 98617 698111. Humphries, M.D., and K. Gurney. 2008. Network ‘Small-World-Ness’: A Quantitative Method for Deter- mining Canonical Network Equivalence. PLoS ONE 3 (4): e0002051. https:// doi. org/ 10. 1371/ journ al. pone. 00020 51. Jacobi, J., A. Llanque, S. Bieri, E. Birachi, R. Cochard, N.D. Chauvin, C. Diebold, R. Eschen, E. Frossard, T. Guillaume, S. Jaquet, F. Kämpfen, M. Kenis, D.I. Kiba, H. Komarudin, J. Madrazo, G. Manoli, S.M. Mukhovi, V.T.H. Nguyen, and C. Robledo-Abad. 2020. Utilization of Research Knowledge in Sustainable Development Pathways: Insights from a Transdisciplinary Research-for- Development Programme. Environmental Science & Policy 103: 21–29. https:// doi. org/ 10. 1016/j. envsci. 2019. 10. 003. Kadushin, C. 2005. Who Benefits from Network Analysis: Ethics of Social Network Research. Social Networks 2 (27): 139–153. https:// doi. org/ 10. 1016/j. socnet. 2005. 01. 005. Kolleck, N. 2016. Uncovering Influence Through Social Network Analysis: The Role of Schools in Edu- cation for Sustainable Development. Journal of Education Policy 31 (3): 308–329. https:// doi. org/ 10. 1080/ 02680 939. 2015. 11193 15. Liu, X., L. Li, S. Wang, Z.-J. Zha, D. Meng, and Q. Huang. 2019. Adaptive Reconstruction Network for Weakly Supervised Referring Expression Grounding. IEEE/CVF International Conference on Com- puter Vision (ICCV) 2019: 2611–2620. https:// doi. org/ 10. 1109/ ICCV. 2019. 00270. Marin, A., and B. Wellman. 2011. SNA: An Introduction. Jhon Scott and Peter J. Carrington. Revealing the Relational Mechanisms of Research for Development… 350 M. Apgar et al. Maru, Y.T., A. Sparrow, J.R.A. Butler, O. Banerjee, R. Ison, A. Hall, and P. Carberry. 2018. Towards Appropriate Mainstreaming of “Theory of Change” Approaches into Agricultural Research for Development: Challenges and Opportunities. Agricultural Systems 165: 344–353. https:// doi. org/ 10. 1016/j. agsy. 2018. 04. 010. Newman, M.E.J. 2003. The Structure and Function of Complex Networks. SIAM Review 45 (2): 167– 256. https:// doi. org/ 10. 1137/ S0036 14450 342480. Patton, M.Q. 2010. Developmental Evaluation: Applying Complexity Concepts to Enhance Innovation and Use. New York: Guilford press. Penuel, W.R., W. Sussex, C. Korbak, and C. Hoadley. 2006. Investigating the Potential of Using Social Network Analysis in Educational Evaluation. American Journal of Evaluation 27 (4): 437–451. https:// doi. org/ 10. 1177/ 10982 14006 294307. Pepinsky, T.B. 2018. A Note on Listwise Deletion versus Multiple Imputation. Political Analysis 26 (4): 480–488. https:// doi. org/ 10. 1017/ pan. 2018. 18. Popelier, L. 2018. A Scoping Review on the Current and Potential Use of Social Network Analysis for Evaluation Purposes. Evaluation 24 (3): 325–352. https:// doi. org/ 10. 1177/ 13563 89018 782219. Provan, K.G., M.A. Veazie, L.K. Staten, and N.I. Teufel-Shone. 2005. The Use of Network Analysis to Strengthen Community Partnerships. Public Administration Review 65 (5): 603–613. https:// doi. org/ 10. 1111/j. 1540- 6210. 2005. 00487.x. Smith, J.A., J.H. Morgan, and J. Moody. 2022. Network Sampling Coverage III: Imputation of Missing Network Data Under Different Network and Missing Data Conditions. Social Networks 68: 148– 178. https:// doi. org/ 10. 1016/j. socnet. 2021. 05. 002. Stork, D., and W.D. Richards. 1992. Nonrespondents in Communication Network Studies: Problems and Possibilities. Group & Organization Management 17 (2): 193–209. https:// doi. org/ 10. 1177/ 10596 01192 172006. Temple, L., D. Barret, G. Blundo Canto, M.-H. Dabat, A. Devaux-Spatarakis, G. Faure, E. Hainzelin, S. Mathé, A. Toillier, and B. Triomphe. 2018. Assessing Impacts of Agricultural Research for Devel- opment: A Systemic Model Focusing on Outcomes. Research Evaluation 27 (2): 157–170. https:// doi. org/ 10. 1093/ resev al/ rvy005. Thornton, P., T. Schuetz, W. Förch, L. Cramer, D. Abreu, S. Vermeulen, and B. Campbell. 2017. Responding to Global Change: A Theory of Change Approach to Making Agricultural Research for Development Outcome-Based. Agricultural Systems 152: 145–153. https:// doi. org/ 10. 1016/j. agsy. 2017. 01. 005. Vance-Borland, K., and J. Holley. 2011. Conservation Stakeholder Network Mapping, Analysis, and Weaving: Conservation Stakeholder Networks. Conservation Letters 4 (4): 278–288. https:// doi. org/ 10. 1111/j. 1755- 263X. 2011. 00176.x. VanderWeele, T.J., and W. An. 2013. Social Networks and Causal Inference. In Handbook of Causal Analysis for Social Research, ed. S.L. Morgan, 353–374. Amsterdam: Springer. Walton, M. 2016. Expert Views on Applying Complexity Theory in Evaluation: Opportunities and Barri- ers. Evaluation 22 (4): 410–423. https:// doi. org/ 10. 1177/ 13563 89016 667890. Wasserman, S., and K. Faust. 1994. Social Network Analysis: Methods and Applications. Cambridge, U.K.: Cambridge University Press. Yin, R. K. 1989. Case Study Research: Design and Methods. https:// books. google. co. uk/ books? hl= en& lr= & id= FzawI AdilH kC& oi= fnd& pg= PR1& dq= Case+ study+ resea rch:+ Design+ and+ metho ds& ots=l_ 1X8ci W3x& sig= oQbz6 MCByW Q0zM0 kBHxq amutE n4#v= onepa ge&q= Case% 20stu dy% 20res earch% 3A% 20Des ign% 20and% 20met hods&f= false Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
null
10.1038_s41467-022-35112-9.pdf
Data availability Data from this study is available at https://zenodo.org/badge/ latestdoi/566835035. Code availability The version of the model code used in this study is tagged as release v1.0 and is available at https://zenodo.org/badge/latestdoi/566835035. Necessary boundary condition files and observational data are inclu- ded as part of the code release.
Data availability Data from this study is available at https://zenodo.org/badge/ latestdoi/566835035 . Code availability The version of the model code used in this study is tagged as release v1.0 and is available at https://zenodo.org/badge/latestdoi/566835035 . Necessary boundary condition files and observational data are included as part of the code release.
Article https://doi.org/10.1038/s41467-022-35112-9 Transfer efficiency of organic carbon in marine sediments Received: 29 April 2022 Accepted: 18 November 2022 Check for updates ; , : ) ( 0 9 8 7 6 5 4 3 2 1 ; , : ) ( 0 9 8 7 6 5 4 3 2 1 James A. Bradley 1,2,7 Sandra Arndt6 , Dominik Hülse 3,4,7, Douglas E. LaRowe5 & Quantifying the organic carbon (OC) sink in marine sediments is crucial for assessing how the marine carbon cycle regulates Earth’s climate. However, burial efficiency (BE) – the commonly-used metric reporting the percentage of OC deposited on the seafloor that becomes buried (beyond an arbitrary and often unspecified reference depth) – is loosely defined, misleading, and inconsistent. Here, we use a global diagenetic model to highlight orders-of- magnitude differences in sediment ages at fixed sub-seafloor depths (and vice- versa), and vastly different BE’s depending on sediment depth or age horizons used to calculate BE. We propose using transfer efficiencies (Teff’s) for quan- tifying sediment OC burial: Teff is numerically equivalent to BE but requires precise specification of spatial or temporal references, and emphasizes that OC degradation continues beyond these horizons. Ultimately, quantifying OC burial with precise sediment-depth and sediment-age-resolved metrics will enable a more consistent and transferable assessment of OC fluxes through the Earth system. Quantifying feedbacks between the carbon cycle and climate requires knowledge of organic carbon (OC) fluxes between Earth’s main reservoirs. The ocean’s biological carbon pump (BCP) delivers OC from the sunlit ocean to the deep sea, where it can be buried and sequestered in sediments over geological timescales. Variations in the long-term OC burial rate have played an important role in regulating atmospheric O2 and CO2 over Earth’s history1,2, and potentially con- tributed to glacial-interglacial cycles3. Geologic sequestration of OC relies ultimately on removal of OC from the active carbon cycle by burial in marine sediments and incorporation into the solid Earth. Burial efficiency (BE) is a commonly-used metric to assess the burial versus degradation of OC in marine sediments. It thus serves as an important link in quantifying the flux of OC between fast-cycling sur- ficial reservoirs (i.e., the ocean, atmosphere, biosphere, soils, upper sediments) and geological reservoirs (i.e., deeper sediments, crustal rocks) that cycle slowly over timescales of thousands to mil- lions of years. BE is loosely defined as the percentage of the OC deposited on the seafloor that becomes buried. Similar to assessing the BCP in the ocean4, the benthic BE metric requires that a particular reference depth beneath the seafloor (zref) is prescribed, beyond which OC is considered ‘buried’ and ostensibly ‘preserved’. However, OC continues to be degraded beyond these horizons, which are often unspecified. Furthermore, different depth horizons can represent vastly different timescales of burial (largely due to differences in local sedimentation rates). The lack of clearly defined reference horizons for the calculation of BE renders this idealized notion of OC burial and preservation imprecise, inconsistent, misleading, and vague. It thus hinders the comparability of benthic OC fluxes between studies, sites, and reservoirs. Specifically, BE at a certain depth (zref) beneath the seafloor (hereafter BEdepth) is the percentage of the OC flux through the sediment-water interface (SWI) (FSWI) that is transferred to depth zref (Fz) (Fig. 1). Assuming steady-state conditions (i.e., that the sum of OC degradation (during its transit from the SWI to zref) and burial (i.e., the 1Queen Mary University of London, London, UK. 2GFZ German Research Center for Geosciences, Potsdam, Germany. 3University of California, Riverside, Riverside, CA, USA. 4Max-Planck-Institute for Meteorology, Hamburg, Germany. 5University of Southern California, Los Angeles, CA, USA. 6Université Libre de Bruxelles, Brussels, Belgium. 7These authors contributed equally: James A. Bradley, Dominik Hülse. e-mail: jbradley.earth@gmail.com Nature Communications | (2022) 13:7297 1 Article https://doi.org/10.1038/s41467-022-35112-9 Fig. 1 | Schematic of the deposition and burial of organic carbon (OC) in idea- lized marine sediments in shelf and abyssal zones. The dashed black lines represent illustrative OC concentrations ([OC]) for shelf and abyssal sediments at certain depths and their equivalent (exemplar) ages, and the dark shading repre- sents possible variability in OC concentration between sites. The red arrows indi- cate the flux of OC through the sediment water interface (FSWI), as well as through specific depth or age layers (Fdepth (e.g., F0.5 m), and Fage (e.g., F0.1 ka), respectively). The widths of the red arrows represent the magnitude of the OC fluxes through those layers. In shelf sediments, OC is rapidly degraded near the sediment-water interface, where shallow sediment depths correspond to short burial times. Con- versely, in abyssal sediments, low concentrations of OC persist over long time- scales. In the deep ocean, sediments buried at shallow depths (beneath the SWI) have much longer burial times than sediments of an equivalent depth (beneath the SWI) in shallow water. This is due to low sedimentation rates in abyssal zones. flux of OC through zref) balances the OC flux through the SWI), BEdepth is calculated according to: BEdepth = 100 × F z F SW I ð1Þ typical coastal sediments, and millions of years of burial in some abyssal regions (Fig. 2a). In addition, post-depositional reworking of sediments (e.g., due to bioturbation, erosion, tectonic events, and turbidity currents) may alter their position relative to sediments of other ages. The burial depth-horizon (zref) is intended to be the lower limit of the zone within which early diagenesis occurs—which, under steady- state conditions, is represented by the point at which the change in OC concentration (OC) with sediment depth (z) reaches zero (i.e., δOC/ δz = 0)5. We note the following issues with the BEdepth metric: I. OC is never irreversibly ‘buried’ or ‘preserved’ in the sediment. Empirical evidence and numerical modeling affirm that OC con- tinues to be utilized by microbes even in very deep and ancient sediment6–9. Thus, the theoretical point at which OC degradation stops (zref) (under steady-state conditions, where δOC/δz = 0) does not exist. The continual nature of OC degradation becomes particularly apparent when OC degradation processes are framed over longer timescales. II. Specified reference depths (beneath the SWI) are highly variable between studies and can be from as little as 15 cm to tens of meters, sometimes pragmatically chosen to be the maximum depth of the sampled sediment core, and sometimes not reported5,10,11. III. Sediment depth can be an inadequate reference frame since biogeochemists and modelers are often concerned with under- standing the fate of elements over particular timescales, rather than depth horizons. IV. There is limited comparability of BEdepth between sites. A depth- based reference horizon ignores vastly different sedimentation rates between sites and thus the differing amounts of time that OC has been subject to degradation processes and other diagenetic alterations. For example, a sediment depth of 10 meters below the seafloor (mbsf) represents several thousand years of burial in We argue that it is crucial to quantify OC burial by its depositional history and not simply by considering its depth beneath the seafloor. Studies should therefore consider using both explicitly-stated refer- ence depth and age-horizons for quantifying carbon transfer through the ocean-sediment system. BE can also be calculated on a temporal (rather than spatial) basis according to the flux of OC through a specified sediment age horizon (BEage): BEage = 100 × F age F SW I ð2Þ Here, Fage represents the OC flux through a specific sediment age horizon (defined by the transit time t since deposition on the seafloor, e.g., t = 100 ka). BEage may be adjusted depending on the timescale of interest. The comparison of equivalent BEage’s between different benthic settings may offer more consistency than using BEdepth’s—since the timescales of diagenetic alterations can be standardized using BEage. However, we are aware of only one study 12. The limitation of this metric is that the age of a that uses BEage particular sediment horizon must be known or estimated (e.g., by using knowledge of past sedimentation rates, and chemical and biological age markers, whilst accounting for any post-depositional disturbances and sediment reworking). We propose a new terminology, transfer efficiency (Teff), for describing the fate of OC through clearly defined depth (Teff,depth) or time (Teff,age) horizons in marine sediments. The calculation of Teff is numerically equivalent to the calculation of BE, but it requires a precise definition of spatial or temporal reference horizons. Nature Communications | (2022) 13:7297 2 Article https://doi.org/10.1038/s41467-022-35112-9 Teff,depth|age is calculated according to: T ef f ,depth∣age SW I ! depth∣age ð Þ = 100 × F depth∣age F SW I ð3Þ Where depth|age represents the depth or age of the sediment horizon of interest, and F depth∣age refers to the flux of organic carbon through that depth or age horizon. For example, Teff,age (SWI→100 ka) denotes the percentage of OC that has survived 100 ka of burial since its deposition at the SWI. The Teff terminology emphasizes that OC is not irreversibly buried but simply transits through a specified horizon. In addition, the precise specification of reference horizons enables comparability and upscaling between sites and studies. We have carried out a series of calculations to illustrate how inconsistencies in BE metrics translate across different timescales, spatial scales, and depositional settings, using a spatially-resolved reaction transport model (RTM) for global sediments13,14. Results and discussion We estimate that the global OC burial rate at 0.11 mbsf (approxi- mately equivalent to the bottom of the bioturbated zone) is between 0.114 and 0.202 Pg C yr−1 (Fig. 3c, Supplementary Table 1). Our cal- culated OC burial rate is at the lower end of previous estimates (0.15–0.31 Pg C yr−1)15,16. However, these previous estimates reported OC burial at unspecified depths beneath the seafloor. We estimate that the majority of OC is buried on the shelves (~0.105 Pg C yr−1 at 0.11 mbsf). This is also the area with the highest uncertainty in estimated burial rates (between 0.079 and 0.135 Pg C yr−1, Supple- mentary Table 1). Calculated Teff’s are highest in abyssal sediments (Fig. 3a, b). However, the total OC burial flux in abyssal zones is low (between 0.024 and 0.048 Pg C yr−1 at 0.11 mbsf, Fig. 3c, Supple- mentary Fig. 1, and Supplementary Table 1) since the OC con- centrations in sediments in these regions (at the SWI and throughout the sediment depth profile) are generally much lower than in shelf and margin sediments17,18. The transit time (t) of sedi- ment from deposition at the seafloor to 0.11 mbsf is also con- siderably longer in abyssal zones than in margin settings (Fig. 2). Our results show that reference depths and ages (used to calculate Teff,depth and Teff,age, respectively) greatly influence the total amount of carbon assumed to be buried in different depositional settings and across the entire seafloor (Fig. 3). Values of Teff,depth and Teff,age, as well as the rates of OC burial, are most sensitive to reference depths and ages in shallower (<100 cm) and younger (<10 ka) sediments (Fig. 3b). These upper-most zones of sediments correspond to areas where OC degradation is fastest, due to the greater availability and preferential degradation of more reactive OC compounds (refs. 19, 20 and refer- ences therein). Therefore, precise specification and reporting of Teff,depth or Teff,age is particularly important for studies focusing on early diagenesis. The clear specification of reference horizons used in the calcu- lation of Teff,depth’s or Teff,age’s allows for adjustments to be made to these metrics based on the characteristic (temporal or spatial) scales of the problem considered. For example, to quantify the near- instantaneous interactions between the sediment and the ocean over annual timescales, the mixed-layer depth could be specified as a depth-horizon. Alternatively, a reference depth of meters to tens of meters below the seafloor could be specified to make estimates of OC budgets on millennial to million-year timescales. What deter- mines a suitable reference depth or age depends on the specific application and problem to be addressed. However, studies report- ing BE using a reference depth that is too shallow or a reference age that is too young may convey the impression that an unrealistically high amount of OC is buried (and presumed sequestered) in sedi- ments. This is because OC continues to be degraded beyond these horizons (in deeper and older layers) (Fig. 3c). Fig. 2 | Sediment ages and depths at specific horizons. a Estimated sediment age (i.e., the time elapsed since its deposition at the SWI) at 10 mbsf. The estimated age of sediment at 10 mbsf varies by over three orders of magnitude globally. b Estimated sediment depth (mbsf) at horizons of equal sediment ages: 0.1 ka, 10 ka, and 100 ka. For a fixed sediment age, sediment depth beneath the seafloor varies globally by over three orders of magnitude. Nature Communications | (2022) 13:7297 3 Article https://doi.org/10.1038/s41467-022-35112-9 Fig. 3 | Transfer and burial of organic carbon (OC) in marine sediments according to sediment depth and sediment age. a Global maps of transfer effi- ciency from the sediment-water interface (SWI) to 1 mbsf (Teff,depth (SWI→1 mbsf), %) and from the SWI to 0.1 ka (Teff,age (SWI→0.1 ka), %). b Transfer (or burial) efficiency according to changes in the specified reference depth horizons and reference age horizons. c Total OC buried beyond specified sediment depths and ages. Gray shading in (b) and (c) represent uncertainty envelopes (±10% in φ and ω, see Supplementary Discussion). Nature Communications | (2022) 13:7297 4 Article https://doi.org/10.1038/s41467-022-35112-9 Transfer efficiency (Teff) (Eq. 3) is a more consistent and precise terminology for describing the fate of OC in marine sediments as it requires the specification of clearly defined depth (Teff,depth) or age (Teff,age) horizons within sediments, and is thus similar to how parti- culate OC fluxes are reported in the ocean4,21,22. This explicit descrip- tion of OC burial according to a common pelagic-benthic framework (e.g., using Teff (SWI→100 ka) to describe the proportion of OC deposited that has survived 100,000 years of burial) ensures com- parability of mass balance and flux calculations, as well as facilitating upscaling efforts between studies. In addition, the Teff notation emphasizes that OC is not irreversibly buried but simply transits through a given horizon and is thus removed from the OC pool at the particular spatial or temporal scale that is defined by Teff. The Teff notation is thus more precise and consistent than BE in evaluating the transport and continual degradation of OC from the surface ocean to specific sediment depths and ages. We propose that if Teff’s are to be compared across settings, both depth and age should be considered. This is owing to (i) the enormous spatial heterogeneities in the age of sediment layers at fixed depths below the seafloor (Fig. 2a), and similarly variable sediment depths at fixed sediment age horizons (Fig. 2b), as well as (ii) the effect of changing reference depths and ages on Teff (Fig. 3). Studies should ideally consider both depth and time, i.e., when specifying a reference depth, time should be discussed (and vice-versa). A complete mechanistic and quantitative understanding of the flux of OC through the sunlit ocean, its sinking and degradation in the water column, and its burial and degradation within sediments is necessary to understand global elemental cycling and its various roles on climate and the biosphere. The numerous biological, che- mical, and physical processes controlling OC degradation and sequestration in sediments are highly heterogeneous over a wide range of spatial and temporal scales19,20. Moreover, the varying characteristics of diverse depositional settings (e.g., burial velocities, porosities, geochemistry) directly affect the timescales over which OC is degraded, and these must be considered when labeling OC as ‘buried’ or ‘sequestered’. Reporting benthic OC fluxes according to a common spatially and temporally defined framework, Teff, will ensure comparability between sites and studies, enable the integration between new measurements and existing data, and facilitate knowl- edge transfer and upscaling efforts. Ultimately, quantifying marine OC fluxes using consistent and robust metrics will enable an improved understanding of benthic-pelagic coupling and the role of marine carbon cycling in the Earth system. Methods We use a one-dimensional RTM to calculate the burial and degrada- tion rate of OC in sub-seafloor sediments13,14, following the approach described in refs. 12, 23, and 24. The model is implemented on a 0.25° × 0.25° resolution global grid. The geographical delineation of shelf, margin, and abyssal zones are adopted from ref. 12 (Supple- mentary Fig. 2). Teff,depth and Teff,age are calculated according to Eq. 3. The OC flux through a specific depth (Fz) is calculated according to: (e.g., refs. 25, 26): ÞOC ∂ 1 (cid:2) φ ð ∂t = ∂ ∂z (cid:1) ð Db 1 (cid:2) φ Þ (cid:3) (cid:2) ∂OC ∂z ÞωOC ∂ 1 (cid:2) φ ð ∂z + 1 (cid:2) φ ð ÞROC ð5Þ Where Db (cm2 yr−1) denotes the bioturbation coefficient, and ROC (g C cm−3 yr−1) stands for the rate of organic carbon degradation. We use a multi-G approximation of a reactive continuum model (RCM) to simulate organic carbon degradation kinetics (building on previous approaches14,27). The initial OC distribution of the RCM is constrained using the Gamma-distribution (Γ) and parameters a, ν, and k: f k, 0ð Þ = aν (cid:3) k ν(cid:2)1 (cid:3) exp (cid:2)a (cid:3) k ð Þ Γ νð Þ ð6Þ Where f(k, 0) determines the fraction of OC having a reactivity of k at time zero. In Eq. 6, a is the average lifetime (years) of the more reactive components of the OC mixture and ν is a dimensionless parameter determining the shape of the distribution near k = 0. The adjustable parameters a and ν completely determine the shape of the initial distribution of OC compounds over the reactivity range and thus its overall reactivity. High ν and low a values define an OC mixture dominated by compounds that are more rapidly degraded, and vice-versa. The Gamma distribution is defined (for any random variable, x) as: Γ νð Þ = Z 1 0 xν(cid:2)1 (cid:3) exp (cid:2)xð Þdx ð7Þ The corresponding cumulative distribution function (CDF) which gives the fraction of total OC having a reactivity of ≤ k at time zero is defined as: F k, 0ð Þ = ð Γ ν, 0, a (cid:3) k Γ νð Þ Þ = R a(cid:3)k 0 xν(cid:2)1 (cid:3) exp (cid:2)xð R 1 0 xν(cid:2)1 (cid:3) exp (cid:2)xð Þdx Þdx ð8Þ Bulk OC, as constrained by the RCM above, is then approxi- mated by 100 finite fractions each with their own first-order degradation rate constant, ki. The reactivity range, here chosen to be k = [10−15, 10emax], with emax = − log(a) + 2 (ref. 12), is divided into i = 100 equal reactivity bins. The fraction of OC within the least reactive fraction i = 1 (i.e., with a degradation rate constant k ≤ 10−15 yr−1) is calculated based on the lower incomplete Gamma function: (cid:4) F 1 10 (cid:5) (cid:2)15, 0 = Z (cid:2)15 a(cid:3)10 0 xν(cid:2)1 (cid:3) exp (cid:2)xð Þdx ð9Þ The fraction, i = 100, of OC characterized by the highest reactivity is calculated based on the upper incomplete Gamma function: ð F z = (cid:2) 1 (cid:2) φ (cid:1) Þ (cid:3) (cid:2)Db (cid:3) ∂OC zð Þ ∂z (cid:3) + ω (cid:3) OC zð Þ ð4Þ (cid:6) (cid:7) F 100 10emax , 0 = R 1 a(cid:3)emax 0 xν(cid:2)1 (cid:3) exp (cid:2)xð R 1 0 0 xν(cid:2)1 (cid:3) exp (cid:2)xð Þdx (cid:2) R xν(cid:2)1 (cid:3) exp (cid:2)xð Þdx Þdx ð10Þ Where OC is the concentration of organic carbon (g C cm−3 dry sedi- ment), z is depth below the seafloor (cm), φ represents sediment porosity and ω is the sedimentation rate (cm yr−1). Organic carbon degradation dynamics The one-dimensional conservation equation describing the transport and transformation of organic carbon (OC) in porous media is given by Nature Communications | (2022) 13:7297 The fractions of total OC within intermediate reactivity bins, i ∈ [2, 99], are calculated with the CDF: (cid:7) (cid:6) F i ki, 0 = (cid:6) Γ ν, 0, a (cid:3) ki + 1 (cid:7) (cid:6) (cid:2) Γ ν, 0, a (cid:3) ki (cid:7) R a(cid:3)ki + 1 0 = Γ νð Þ R xν(cid:2)1 (cid:3) exp (cid:2)xð a(cid:3)ki Þdx (cid:2) R 1 0 0 xν(cid:2)1 (cid:3) exp (cid:2)xð xν(cid:2)1 (cid:3) exp (cid:2)xð Þdx Þdx ð11Þ 5 Article https://doi.org/10.1038/s41467-022-35112-9 All fractions Fi add up to unity. The degradation rate of bulk OC can thus be calculated as: ROC = X100 i = 1 ki (cid:3) OCi zð Þ ð12Þ Where OCi(0) = Fi · OC0 assuming a known OC content at the SWI, OC0. The derived degradation rate of OC, ROC, was then used in Eq. 5 (i.e., the general conservation equation) to calculate OC concentrations, degradation and burial rates for the different sediment layers. For this purpose, the general conservation equation (Eq. 5) was solved analy- tically. Assuming steady-state conditions (i.e., ∂OC ∂t = 0), and Db = 0 for z > zbio (where Db represents the bioturbation coefficient (cm2 year−1), and zbio is the maximum depth of the bioturbated zone (cm)), the general solution of Eq. 5 for each organic carbon fraction i in the bioturbated zone (z ≤ zbio) is given by: OCi zð Þ = A1ie a1iz ð Þ ð + B1ie b1iz Þ And in the non-bioturbated zone (z > zbio) by: With: OCi zð Þ = A2ie a2iz ð Þ ω (cid:2) a1i = q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (cid:7) (cid:6) ω2 + 4Dbki 2Db ω + b1i = q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (cid:7) (cid:6) ω2 + 4Dbki 2Db a2i = (cid:2)ki ω ð13Þ ð14Þ ð15Þ ð16Þ ð17Þ The bulk OC concentration as a function of depth is then calculated as: OC zð Þ = X100 i = 1 OCi zð Þ ð18Þ The integration constants A1i, B1i, and A2i are defined by chosen boundary conditions. Here, we apply a known OC concentration at the SWI and we assume continuity (in concentration and flux) across the bottom of the bioturbated zone, zbio. The integration constants are thus calculated as: B1i = OC0 (cid:2) A1i ð19Þ (cid:1) (cid:3) exp a1i A1i (cid:3) (cid:7) (cid:6) (cid:3) lim h!0 (cid:2) h zbio (cid:1) (cid:1) (cid:3) exp b1i (cid:3) (cid:7) (cid:3) (cid:7) (cid:6) (cid:3) lim h!0 zbio (cid:2) h + B1i (cid:6) exp a2i (cid:3) lim h!0 zbio + h A2i = (cid:2)B1ib1i A1i = a1i (cid:1) (cid:3) exp b1i (cid:1) (cid:3) exp a1i (cid:6) zbio (cid:7) (cid:2) h (cid:3) (cid:7) (cid:3) lim h!0 (cid:6) (cid:3) lim h!0 zbio (cid:2) h (cid:3) ð20Þ ð21Þ For h > 0 (see e.g., ref. 13 for details). Nature Communications | (2022) 13:7297 Parameters and boundary conditions For every grid cell we prescribe a particular concentration of organic carbon at the SWI, OC0, and a set of parameter values (i.e., ω, Db, φ, and a) (Supplementary Fig. 3). Values of OC0 are taken from ref. 18. Sedimentation rates, ω, were calculated using an algorithm that cor- relates water depth and sedimentation rate according to a double logistic equation28. The bioturbation coefficient, Db, also depends on water depth and follows the empirical relationship of ref. 29. The porosities of sediments at the SWI were taken from ref. 30. We neglect sediment compaction and porosity changes (approximately 1/ 600 m−1, ref. 31) in the upper 10 m of the sediment in order to find an analytical solution to Eq. 5. A comparison of the analytical solution with a numerical early diagenetic model with depth dependent porosity shows that porosity changes do not meaningfully affect our results13. A global parameter compilation20 and inversely calculated RCM parameters32,33 indicate that ν does not vary much between sites, while parameter a can vary over orders of magnitude. Based on these results, we assume a constant ν value of ν = 0.125 (characteristic of fresh organic matter). The values of parameter a (i.e., shelf a = 0.1 yr, margin a = 1.0 yr, abyss a = 20.0 yr) were chosen to produce a realistic global OC burial rate that reflects the range observed in ref. 20. In order to account for lower OM reactivities and minimal bioturbation in low oxygen environments (e.g., refs. 29, 34, 35) we reduce the OM reac- tivity by an order of magnitude and set zbio equal to 1 cm in hypoxic seafloor zones (i.e., [O2] < 60 μM, according to bottom-water marine oxygen concentrations from the World Ocean Atlas 201836). Model evaluation and sensitivity analysis A detailed evaluation of the diagenetic model is provided in refs. 13, 14. We also compared our model output to five organic carbon (OC) profiles measured in sediment cores collected from different ocean (Supplementary Table 2, Supplementary depths and regions Discussion). We performed a global sensitivity analysis to generate a ranking of the most important unknown model parameters (besides the reactivity of OM, i.e., φ, ω, zbio, and Db) according to their relative contribution to the variability in model output (SI Fig. 4, Supplementary Discussion). The sensitivity analysis was used to generate uncertainty envelopes for our estimates of Teff and OC burial (Fig. 3) using a variability of ±10% of the two most influential parameters (i.e., ω and φ). We used the method of ref. 37, also called the ‘Elementary Effect Test’ (EET38), which takes the mean of r finite differences (also called the ‘Elementary Effects’ or EEs) as a measure of global sensitivity of input parameter i: Si = 1 r Xr j = 1 EEj = 1 r P r (cid:4) j = 1 g xj 1,:::,xj i + Δj i,:::,xj M Δj i (cid:5) (cid:4) (cid:2) g xj 1,:::,xj i,:::,xj M (cid:5) ð22Þ Where g() is our diagenetic model, OMEN-SED, that maps the j) into the output space—here j, …, xM vector of the input factors xj = (x1 the simulated OC burial rates at 1 mbsf. Δi j represents the variation of the input parameter i. We compute the standard deviation of the EEs, which measures the degree of interaction of input parameter i with the other input parameters. Both sensitivity indices are relative measures, hence their values do not have a specific meaning and can only be used to rank the influence of the input parameters. As a strategy to select the parameter vectors xj (j = 1, …,r) and the input variations Δi for the investigated model parameters (M = 4), we used the Latin hypercube sampling approach as implemented in the Sensitivity Analysis for Everyone (SAFE) MATLAB toolbox39. For zbio we explored a range between 1 and 15 cm. For φ, ω, and Db we varied the nominal values in each grid cell by up to 20%. The calculations of the mean and standard deviation of the EEs of M input parameters requires N = r·(M + 1) model evaluations. To assess 6 Article https://doi.org/10.1038/s41467-022-35112-9 the robustness of our sensitivity indices, i.e., to analyze if they are independent of the specific input–output sample, we calculated bootstrapping-based confidence limits of the indices. Following recommendations in the literature (e.g., ref. 40), we calculated r = 30 finite differences, which is sufficient to differentiate between influen- tial and non-influential parameters, to calculate reasonable confidence bounds of the sensitivity indices. In total we ran N = r·(M + 1) = 150 global model simulations with different input parameter values. Data availability Data from this study is available at https://zenodo.org/badge/ latestdoi/566835035. Code availability The version of the model code used in this study is tagged as release v1.0 and is available at https://zenodo.org/badge/latestdoi/566835035. Necessary boundary condition files and observational data are inclu- ded as part of the code release. References 1. Berner, R. A. Burial of organic carbon and pyrite sulfur in the modern ocean: Its geochemical and environmental significance. Am. J. Sci. 282, 451–473 (1982). Berner, R. A. A model for atmospheric CO2 over Phanerozoic time. Am. J. Sci. 291, 339–376 (1991). 2. 3. Cartapanis, O., Bianchi, D., Jaccard, S. L. & Galbraith, E. D. Global pulses of organic carbon burial in deep-sea sediments during gla- cial maxima. Nat. Commun. 7, 1–7 (2016). Buesseler, K. O., Boyd, P. W., Black, E. E. & Siegel, D. A. Metrics that matter for assessing the ocean biological carbon pump. Proc. Natl Acad. Sci. USA 117, 9679 LP–9687 (2020). 4. 5. Canfield, D. E. Factors influencing organic carbon preservation in marine sediments. Chem. Geol. 114, 315–329 (1994). 6. Arndt, S., Brumsack, H.-J. & Wirtz, K. W. Cretaceous black shales as active bioreactors: A biogeochemical model for the deep biosphere encountered during ODP Leg 207 (Demerara Rise). Geochim. Cos- mochim. Acta 70, 408–425 (2006). Inagaki, F. et al. Exploring deep microbial life in coal-bearing sedi- ment down to ~2.5 km below the ocean floor. Science 349, 420 LP–420424 (2015). 7. 8. Westrich, J. T. & Berner, R. A. The role of sedimentary organic- 9. matter in bacterial sulfate reduction—the G model tested. Limnol. Oceanogr. 29, 236–249 (1984). Jørgensen, B. B. A comparison of methods for the quantification of bacterial sulfate reduction in coastal marine sediments: I. Mea- surement with radiotracer techniques. Geomicrobiol. J. 1, 11–27 (1978). 10. Stein, R. Surface-water paleo-productivity as inferred from sedi- ments deposited in oxic and anoxic deepwater environments of the mesozoic Atlantic Ocean. Biogeochem. Black Shales 60, 55–70 (1986). 11. Hartnett, H. E., Keil, R. G., Hedges, J. I. & Devol, A. H. Influence of oxygen exposure time on organic carbon preservation in con- tinental margin sediments. Nature 391, 572–575 (1998). 12. LaRowe, D. E. et al. Organic carbon and microbial activity in marine sediments on a global scale throughout the Quaternary. Geochim. Cosmochim. Acta 286, 227–247 (2020). 13. Hülse, D., Arndt, S., Daines, S., Regnier, P. & Ridgwell, A. OMEN-SED 1.0: a novel, numerically efficient organic matter sediment diag- enesis module for coupling to Earth system models. Geosci. Model Dev. 11, 2649–2689 (2018). 14. Pika, P., Hülse, D. & Arndt, S. OMEN-SED(-RCM) (v1.1): a pseudo- 15. Muller-Karger, F. E. et al. The importance of continental margins in the global carbon cycle. Geophys. Res. Lett. 32, 1–4 (2005). 16. Dunne, J. P., Sarmiento, J. L. & Gnanadesikan, A. A synthesis of global particle export from the surface ocean and cycling through the ocean interior and on the seafloor. Global Biogeochem. Cycles 21, 1–16 (2007). 17. Seiter, K., Hensen, C., Schröter, J. & Zabel, M. Organic carbon content in surface sediments—defining regional provinces. Deep Sea Res. Part I Oceanogr. Res. Pap. 51, 2001–2026 (2004). 18. Lee, T. R., Wood, W. T. & Phrampus, B. J. A machine learning (kNN) approach to predicting global seafloor total organic carbon. Global Biogeochem. Cycles 33, 37–46 (2019). 19. LaRowe, D. E. et al. The fate of organic carbon in marine sediments —new insights from recent data and analysis. Earth-Sci. Rev. 204, 103146 (2020). 20. Arndt, S. et al. Quantifying the degradation of organic matter in marine sediments: a review and synthesis. Earth-Sci. Rev. 123, 53–86 (2013). 21. Weber, T., Cram, J. A., Leung, S. W., DeVries, T. & Deutsch, C. Deep ocean nutrients imply large latitudinal variation in particle transfer efficiency. Proc. Natl Acad. Sci. 113, 8606 LP–8608611 (2016). 22. Henson, S. A., Sanders, R. & Madsen, E. Global patterns in efficiency of particulate organic carbon export and transfer to the deep ocean. Global Biogeochem. Cycles 26, 1–14 (2012). 23. Bradley, J. A. et al. Widespread energy limitation to life in global subseafloor sediments. Sci. Adv. 6, eaba0697 (2020). 24. Bradley, J. A., Arndt, S., Amend, J. P., Burwicz-Galerne, E. & LaRowe, D. E. Sources and Fluxes of Organic Carbon and Energy to Micro- organisms in Global Marine Sediments. Front. Microbiol. 13, (2022). 25. Berner, R. A. Early Diagenesis: A Theoretical Approach. Princeton Series in Geochemistry (1980). 26. Boudreau, B. P. Diagenetic Models and Their Implementation. Modelling Transport and Reactions in Aquatic Sediments Vol. 171 (Springer, 1997). 27. Dale, A. W. et al. A revised global estimate of dissolved iron fluxes from marine sediments. Glob. Biogeochem. Cycles 29, 691–707 (2015). 28. Burwicz, E. B., Rüpke, L. H. & Wallmann, K. Estimation of the global amount of submarine gas hydrates formed via microbial methane formation based on numerical reaction-transport modeling and a novel parameterization of Holocene sedimentation. Geochim. Cosmochim. Acta 75, 4562–4576 (2011). 29. Middelburg, J. J., Soetaert, K. & Herman, P. M. J. Empirical rela- tionships for use in global diagenetic models. Deep Sea Res. Part I Oceanogr. Res. Pap. 44, 327–344 (1997). 30. Martin, K. M., Wood, W. T. & Becker, J. J. A global prediction of seafloor sediment porosity using machine learning. Geophys. Res. Lett. 42, 10640–10646 (2015). 31. Einsele, G. Sedimentary Basins: Evolution, Facies, and Sediment Budget (Springer, 2000). 32. Boudreau, B. P. & Ruddick, B. R. On a reactive continuum repre- sentation of organic matter diagenesis. Am. J. Sci. 291, 507–538 (1991). 33. Freitas, F. S. et al. New insights into large-scale trends of apparent organic matter reactivity in marine sediments and patterns of benthic carbon transformation. Biogeosciences 18, 4651–4679 (2021). 34. LaRowe, D. E. & Van Cappellen, P. Degradation of natural organic matter: a thermodynamic analysis. Geochim. Cosmochim. Acta 75, 2030–2042 (2011). 35. Aller, R. C. in Treatise on Geochemistry: Second Edition Vol. 8 (2013). 36. Garcia, H. et al. World Ocean Atlas 2018, Volume 3: Dissolved reactive continuum representation of organic matter degradation dynamics for OMEN-SED. Geosci. Model Dev. 14, 7155–7174 (2021). Oxygen, Apparent Oxygen Utilization, and Dissolved Oxygen Saturation. NOAA Atlas NESDIS Vol. 83 (2019). Nature Communications | (2022) 13:7297 7 Article https://doi.org/10.1038/s41467-022-35112-9 37. Morris, M. D. Factorial sampling plans for preliminary computa- tional experiments. Technometrics 33, 161–174 (1991). Correspondence and requests for materials should be addressed to James A. Bradley. 38. Saltelli, A. et al. Global sensitivity analysis: The primer. Global Sen- sitivity Analysis: The Primer. https://doi.org/10.1002/ 9780470725184 (2008). 39. Pianosi, F., Sarrazin, F. & Wagener, T. A Matlab toolbox for global sensitivity analysis. Environ. Model. Softw. 70, 80–85 (2015). 40. Pianosi, F. et al. Sensitivity analysis of environmental models: a Peer review information Nature Communications thanks Virginia Edg- comb, Jamie Wilson and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available. systematic review with practical workflow. Environ. Model. Softw. 79, 214–232 (2016). Reprints and permissions information is available at http://www.nature.com/reprints Acknowledgements We acknowledge funding from NERC (NE/T010967/1) (J.A.B.), the Alex- ander von Humboldt Foundation (J.A.B.), the Human Frontier Science Program (J.A.B.), the Simons Foundation (653829) (D.H.), C-DEBI (NSF OCE0939564) (D.E.L.), NASA-NSF Origins of Life Ideas Lab (NNN13D466T) (D.E.L.), NASA Habitable Worlds (80NSSC20K0228) (D.E.L.), and BELSPO FedtWin program RECAP (S.A.). Author contributions J.A.B. and D.H. contributed equally to this work. J.A.B. conceived the study. J.A.B., D.H., and S.A. designed the research. D.H. conducted the simulations. J.A.B., D.H., D.E.L., and S.A. analyzed model output. J.A.B. and D.H. wrote the manuscript with input from D.E.L. and S.A. Competing interests The authors declare no competing interests. Additional information Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41467-022-35112-9. Publisher’s note Springer Nature remains neutral with regard to jur- isdictional claims in published maps and institutional affiliations. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/ licenses/by/4.0/. © The Author(s) 2022 Nature Communications | (2022) 13:7297 8
null
10.1016_j.celrep.2023.112408.pdf
Data and code availability—Original small-RNA sequencing datasets are publicly available in NCBI under the accession number BioProject: PRJNA874806.
null
A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t HHS Public Access Author manuscript Cell Rep. Author manuscript; available in PMC 2023 August 22. Published in final edited form as: Cell Rep. 2023 May 30; 42(5): 112408. doi:10.1016/j.celrep.2023.112408. The nuclear Argonaute HRDE-1 directs target gene re- localization and shuttles to nuage to promote small RNA- mediated inherited silencing Yue-He Ding1, Humberto J. Ochoa1, Takao Ishidate1, Masaki Shirayama1, Craig C. Mello1,2,3,* 1RNA Therapeutics Institute, University of Massachusetts Medical School, Worcester, MA 01605, USA 2Howard Hughes Medical Institute, Worcester, MA 01605, USA 3Lead contact SUMMARY Argonaute/small RNA pathways and heterochromatin work together to propagate transgenerational gene silencing, but the mechanisms behind their interaction are not well understood. Here, we show that induction of heterochromatin silencing in C. elegans by RNAi or by artificially tethering pathway components to target RNA causes co-localization of target alleles in pachytene nuclei. Tethering the nuclear Argonaute WAGO-9/HRDE-1 induces heterochromatin formation and independently induces small RNA amplification. Consistent with this finding, HRDE-1, while predominantly nuclear, also localizes to peri-nuclear nuage domains, where amplification is thought to occur. Tethering a heterochromatin-silencing factor, NRDE-2, induces heterochromatin formation, which subsequently causes de novo synthesis of HRDE-1 guide RNAs. HRDE-1 then acts to further amplify small RNAs that load on downstream Argonautes. These findings suggest that HRDE-1 plays a dual role, acting upstream to initiate heterochromatin silencing and downstream to stimulate a new cycle of small RNA amplification, thus establishing a self-enforcing mechanism that propagates gene silencing to future generations. In brief Ding and colleagues investigate inherited silencing in C. elegans. They demonstrate that the nuclear Argonaute HRDE-1 induces subnuclear-co-localization of target genes in heterochromatin. Heterochromatin formation subsequently triggers de novo HRDE-1 guide RNA loading. Finally, This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). *Correspondence: craig.mello@umassmed.edu. AUTHOR CONTRIBUTIONS Conceptualization, Y.-H.D. and C.C.M.; investigation, Y.-H.D. and H.J.O.; methodology, Y.-H.D., H.J.O., T.I., and C.C.M.; data analysis, Y.-H.D.; writing – review & editing, Y.-H.D. and C.C.M.; supervision, C.C.M. SUPPLEMENTAL INFORMATION Supplemental information can be found online at https://doi.org/10.1016/j.celrep.2023.112408. DECLARATION OF INTERESTS The authors declare no competing interests. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 2 HRDE-1 enters nuage and activates small RNA amplification. Thus, HRDE-1 effects multiple steps of a self-enforcing transgenerational silencing process. Graphical Abstract INTRODUCTION In many animal germlines, small RNA/Argonaute pathways function transgenerationally to install and re-inforce chromatin silencing essential for fertility. For example, in flies, worms, and mammals, members of the PIWI Argonaute family engage genomically encoded small RNAs termed PIWI-interacting RNAs (piRNAs) that silence transposons to maintain genome integrity.1–6 Although the details differ, all transgenerational small RNA silencing pathways studied to date require amplification and engagement of secondary Argonautes.7 Many of the components of the amplification machinery localize prominently in peri-nuclear non-membranous organelles called nuage. However, how the amplification system in nuage communicates with and drives the nuclear events during the initiation and maintenance of transgenerational silencing is not well understood. In C. elegans, transgenerational silencing can be initiated by the PIWI pathway, by the canonical double-stranded RNA (dsRNA)-induced RNAi pathway, or by intronless mRNA.8–11 Inherited silencing is maintained by a family of related downstream worm- specific Argonautes (WAGO Argonautes) guided by small RNAs (22G-RNAs) produced Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 3 by cellular RNA-dependent RNA polymerase. Once established, inherited silencing can be propagated independently of the initiating cues via continuous cycles of WAGO 22G-RNA amplification and transmission of the WAGO Argonautes and their small RNA co-factors to progeny.8,12–14 The nuclear WAGO Argonaute, HRDE-1/WAGO-9, plays a central role in transgenerational silencing in C. elegans.15,16 HRDE-1 is thought to engage nascent transcripts at target loci to induce heterochromatin and transcriptional silencing through the nuclear RNAi pathway.15,17 HRDE-1 promotes the transgenerational silencing of many genes18 and is thought to do so by recruiting chromatin remodeling factors, including the nucleosome remodeling and deacetylase complex (NuRD) and histone methyltransferases (e.g., MET-2, SET-25, SET-32).9,18,19 The nuclear RNAi pathway is also required for the spreading of secondary small RNAs from piRNA target sites.14,20 Transgenerational silencing requires a series of events that are thought to occur in the nuage, nucleus, and cytoplasm. Because all of these events are essential for the cycle of inherited silencing, their order has been difficult to determine. For example, it is not known whether the nuclear Argonaute HRDE-1 directly triggers RdRP recruitment and amplification of small RNAs or whether it must first induce heterochromatin at its targets to elicit small RNA amplification. Here, we use the phage lambda N (λN)-boxB tethering system21–25 to recruit—i.e., tether—HRDE-1 or the nuclear silencing factor NRDE-2 to a reporter mRNA. In principle, tethering enables initiation of silencing in the absence of upstream initiators such as piRNAs or dsRNA and, with appropriate genetic tests, can be used to order events in the pathway. We show that tethering either HRDE-1 or NRDE-2 can induce a complete silencing response, including small RNA amplification and transgenerational silencing that persists even after the λN-fusion protein is crossed from the strain. Tethering NRDE-2 initiates chromatin silencing through nrde-4 and independently of hrde-1 but requires hrde-1 for small RNA amplification. By contrast, tethering HRDE-1 stimulates chromatin silencing through NRDE-2 and NRDE-4 but can elicit small RNA amplification independently of both these chromatin-silencing factors. Mutations that block HRDE-1 from binding small RNA disarm silencing and cause HRDE-1 to become cytoplasmic, but tethering HRDE-1 in these mutants nevertheless initiates a strong silencing response that requires small RNA amplification proximal to the tether site. The small RNA amplification machinery is recruited to the tether site by sequences in the N-terminal half of HRDE-1 (the N-terminal domain [NTD]). Like full-length HRDE-1 protein, HRDE-1 NTD co-localizes with MUT-16 in Mutator foci, subdomains of cytoplasmic nuage where the small RNA amplification machinery resides.26 Our findings suggest that HRDE-1 lies at a nexus in the silencing pathway, shuttling from the nucleus to the nuage and back, to coordinate the nuclear and cytoplasmic events of transgenerational silencing. RESULTS HRDE-1 and NRDE-2 tethering induce transgenerational silencing To order events in inherited silencing, we sought to uncouple initiation and maintenance of silencing. To do this, we used the phage λN-boxB tethering system to recruit nuclear silencing factors HRDE-1 or NRDE-2 to a target reporter that is robustly expressed in the Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 4 germline (Figure 1A). We hypothesized that if artificial recruitment of a silencing factor mimics a physiological event, then it should elicit a silencing response that is independent of upstream factors but depends on known downstream factors. For example, directly tethering a chromatin factor should, in principle, induce silencing without requiring machinery necessary to amplify the small RNAs that would normally guide the chromatin silencing machinery to the appropriate targets. Using CRISPR, we inserted an in-frame λN coding sequence at the 5′ end of the endogenous hrde-1 or nrde-2 loci (see STAR Methods). Both fusion genes were fully functional, based on their ability to mediate piRNA silencing (Figure S1A). Moreover, both strains exhibited wild-type patterns and distributions of endogenous small RNA species (Figure S1B).We then tested whether the λN fusions could induce heritable silencing of a reporter gene whose 3′ UTR contains λN-binding sites (i.e., boxB elements) (Figures 1A–1D). Both λN::HRDE-1 and λN::NRDE-2 induced silencing of the reporter beginning at the initial heterozygous generation (Figures 1C and 1D). Notably, silencing of the reporter persisted in subsequent generations after genetically segregating away the λN-fusion alleles (Figures 1E and S1C; data not shown). As expected, inherited silencing (after segregating the λN-fusion alleles) required known components of the transgenerational RNA silencing pathway, including HRDE-1, the small RNA amplification factors RDE-3/MUT-2 and MUT-16,6,27,28 and the nuclear silencing factors NRDE-2 and NRDE-49,29 (Figures 1E– 1G, S1C, and S1H; data not shown). Moreover, λN::HRDE-1 and λN::NRDE-2 tethering induced trimethylation of histone H3 lysine 9 (H3K9me3; Figures 2A and 2B) and reduced both reporter mRNA and pre-mRNA levels (Figures 2C and 2D), consistent with the role of H3K9me3 in transcriptional silencing.15 Thus, artificially recruiting HRDE-1 or NRDE-2 to a target locus was sufficient to initiate the full cycle of events required for inherited silencing, including small RNA amplification and heterochromatin formation. Having established that tethering induces inherited silencing that depends genetically on known components of the RNA silencing pathway, we asked which factors were required for silencing when the tethered protein was continuously present. For example, because the λN-boxB interaction recruits HRDE-1 and NRDE-2 independently of a guide RNA, we reasoned that the small RNA amplification machinery should be unnecessary when nuclear silencing factors are tethered to the reporter. Consistent with this idea, we found that λN::NRDE-2 silenced the reporter in the absence of rde-3, mut-16, and hrde-1 (Figures 2E, S2A–S2C) but failed to silence it in the absence of nrde-4 (Figures 2E and S2D). These results suggest that NRDE-2 acts downstream of HRDE-1 and upstream of NRDE-4 in nuclear silencing. In wild-type animals without tethering, inherited silencing requires nuclear chromatin silencing factors (e.g., nrde-2 and nrde-4) and nuage-localized factors (e.g., rde-3 and mut-16; Figure 1G), indicating that these pathways function together, possibly sequentially, to propagate inherited silencing. In contrast, when λN::HRDE-1 was tethered to the reporter, we found that leaving either pathway intact was sufficient to maintain silencing, as monitored by GFP epifluorescence. For example, silencing of the reporter GFP was maintained independently of nrde-2, nrde-4, or rde-3 and only partly required mut-16 activity (Figures 2F and S2E–S2H). To completely prevent silencing, it was necessary to Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 5 simultaneously mutate components of both the small RNA amplification machinery (rde-3 or mut-16) and components of the chromatin nuclear silencing machinery (nrde-2 or nrde-4) (Figures 2F and S2E–S2K). HRDE-1 tethering in wild-type worms reduced the unspliced pre-mRNA reporter level by 2-fold and the spliced RNA level by 100-fold, as measured by quantitative PCR (qPCR) (Figure 2C). For unknown reasons, nrde-2 mutants exhibited a 4-fold increase in reporter pre-mRNA both with and without HRDE-1 tethering (Figure 2C) but exhibited discordant effects on spliced reporter RNA levels. Removing nrde-2 activity in animals without tethering had little effect on spliced reporter mRNA levels (a slight 1.2-fold increase) compared with wild type, but removing nrde-2 activity in the context of tethering caused spliced RNA levels to increase (compared with levels in wild-type HRDE-1-tethered animals), reaching levels of approximately 40% of wild-type mRNA levels. It is important to note that the qPCR assay cannot distinguish mRNA from template RNA being silenced, as template RNAs derive from spliced RNAs. Moreover, the high levels of spliced RNA in λN::HRDE-1;nrde-2 worms correlate with a marked accumulation of reporter RNA localized in nuage (via RNA fluorescence in situ hybridization [FISH], shown below). Thus, the accumulated spliced RNA likely reflects template RNA engaged in amplifying the small RNA silencing signal, perhaps to compensate for the loss of heterochromatin silencing. Further study is needed to understand the effects of nrde-2 mutants on pre-mRNA levels, such as whether increased pre-mRNA levels in nrde-2 mutants reflect processing defects.30 Nevertheless, in the nrde-2 background, HRDE-1 tethering reduces mRNA and pre-mRNA levels by 2-to 3-fold, suggesting that tethered HRDE-1 can exert effects on both mRNA and pre-mRNA levels independently of NRDE-2. Taken together, our findings suggest that HRDE-1 functions twice during inherited silencing—upstream of nuclear silencing to recruit NRDE-2 and NRDE-4 and again downstream of these factors to induce small RNA amplification and post-transcriptional clearance of mRNA. While these events likely occur sequentially and thus depend on each other during the normal course of inherited silencing,31 tethering HRDE-1 initiates both modes of silencing independently, either of which is sufficient to prevent reporter GFP expression. HRDE-1 acts downstream of NRDE-2 to promote small RNA amplification The above findings indicate that HRDE-1 can initiate inherited silencing independently of nrde-2 and nrde-4, while NRDE-2 requires both nrde-4 and hrde-1. A likely explanation for these findings is that heterochromatin silencing directed by NRDE-2 and NRDE-4 induces the de novo synthesis of small RNAs that engage HRDE-1 and that HRDE-1 can further amplify these small RNAs to propagate silencing to offspring. Indeed, whereas we detected very few small RNAs targeting the reporter in the absence of tethering (Figure 3A), λN::NRDE-2 induced small RNA accumulation that required nrde-4, rde-3, and hrde-1 (Figures 3B–3E). These findings suggest that NRDE-2 tethering induces silencing and heterochromatin formation through NRDE-4 (Figures 2B and 2D) and that downstream events (e.g., heterochromatin formation itself or other NRDE-4-dependent events) act through RDE-3 and HRDE-1 to induce small RNA amplification. Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 6 λN::HRDE-1 tethering induced abundant small RNA accumulation that was independent of nrde-2 and nrde-4 (Figures 3F, 3G, and S3A). However, interestingly, both the distribution of small RNAs and their levels of accumulation along the target mRNA were dramatically altered in the nrde mutants. Small RNA levels were markedly increased adjacent to the boxB sites and were diminished on the gfp coding sequences (Figures 3F, 3G, S3A, and S3B). Small RNAs targeting the reporter were greatly reduced by mutations in rde-3 and mut-16, as expected, (Figures 3H and 3I). Interestingly, however, a low level of small RNAs persisted directly adjacent to the boxB sites when λN::HRDE-1 was tethered in the absence of rde-3 but not in the absence of mut-16 (Figure 3H). This result is consistent with the observation that tethering of λN::HRDE-1 can bypass rde-3 but cannot fully bypass mut-16 (Figure 2F). When outcrossed to a hrde-1(+) background to segregate away λN::HRDE-1, the reporter remained silent for at least 13 generations, with no change in penetrance. Moreover, we observed only a slight reduction in small RNA levels primarily in regions juxtaposed to the boxB hairpins (Figure 3J). In contrast, when outcrossed to a hrde-1 null background, the reporter was fully de-silenced, and small RNAs were absent (Figure 3K). As expected, the maintenance of silencing, and of small RNA levels, also required rde-3(+) and mut-16(+) (Figures 3L and 3M). Taken together, these findings suggest that heterochromatin formation at the target locus induces de novo transcription and loading of small RNAs onto the nuclear Argonaute HRDE-1. HRDE-1, in turn, further promotes small RNA amplification and then functions again, perhaps in the next life cycle, to reinitiate heterochromatin silencing (see discussion). HRDE-1 guide RNA loading is not required for small RNA amplification The finding that λN::HRDE-1 can direct chromatin silencing in rde-3 and mut-16 mutants, which are defective in small RNA amplification, suggests that the unloaded Argonaute can direct chromatin silencing when tethered. To further test this idea, we monitored silencing (1) by λN::HRDE-1 in an hrde-2 mutant, which is defective in HRDE-1 small RNA loading13 and (2) by a λN::HRDE-1(Y669E) mutant, predicted by structural work to be defective in guide RNA binding (Figure S5B).32 In both cases, tethering completely silenced the boxB reporter as monitored by GFP fluorescence (Figure 4C and S4D) and by quantitative reverse transcription PCR (qRT-PCR) of the mRNA (Figure 4D). For unknown reasons, compromising nuclear silencing by hrde1-(Y669E) caused elevated pre-mRNA levels as measured by qRT-PCR (Figure 4D), similar to nrde-2 mutants. As expected, the hrde-1(Y669E) mutant was defective in silencing a piRNA reporter (Figure S4A) and showed a collapse of small RNAs resembling that in hrde-1(null) (Figures S4B and S4C). However, in these mutant contexts, loss of rde-3 alone was sufficient to completely de-silence the reporter (Figures S4E and 4C), suggesting that in the absence of guide RNA loading, HRDE-1 fails to engage the NRDE heterochromatin machinery. Deep sequencing revealed an abundant accumulation of rde-3-dependent small RNAs targeting the boxB reporter in λN::HRDE-1(Y669E) animals (Figures 4E and 4F). Notably, the pattern and levels of small RNA accumulation induced by λN::HRDE-1(Y669E) resembled those observed when wild-type λN::HRDE-1 is tethered in a nrde-2 mutant (compare Figures 4E–3G)—i.e., resulting in increased levels of small RNAs targeting sequences adjacent to Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 7 the boxB sites and reduced levels targeting GFP sequences. Taken together, these results suggest that tethering of unloaded HRDE-1 can induce local small RNA amplification and silencing but that tethered HRDE-1 must be loaded with small RNAs to induce chromatin silencing, which is in turn required for small RNA targeting to spread into the 5′ sequences of the target mRNA. HRDE-1 promotes small RNA amplification through its NTD We next attempted to dissect functional domains of HRDE-1 required for small RNA amplification. We used CRISPR to make a series of λN::hrde-1 truncation mutants (Figure 5A). These studies identified the N-terminal half (herein the NTD) as the minimal fragment of HRDE-1 that could fully silence the reporter. The NTD and the remaining C-terminal domain (CTD) truncations of HRDE-1 are predicted by I-TASSER33 to fold into self-contained globular structures, with subdomains similar to those identified in atomic resolution studies on humanAgo234 (Figures 5B, S5A, and S5B). As expected, in the absence of tethering, hrde-1(NTD) and hrde-1(CTD) alleles failed to silence a piRNA sensor (Figure S4A). Silencing by λN::NTD required rde-3 but not nrde-2 (Figures 5C and S5C), and deep sequencing revealed that λN::NTD induces abundant rde-3-dependent small RNAs targeting the boxB reporter (Figures 5D and 5E). Truncations that failed to silence the reporter did not trigger small RNA generation (Figure S5D). The small RNA pattern induced by λN::NTD resembled the patterns caused by λN::HRDE-1 in nrde-2 mutants or by λN::HRDE-1(Y669E)—i.e., dramatically increased levels of small RNAs proximal to the boxB sites and reduced levels of small RNAs targeting GFP sequences. Interestingly, the magnitude of small RNA accumulation induced by λN::NTD at the boxB sites was ~4-fold greater than that induced by either λN::HRDE-1 in nrde-2 mutants or by λN::HRDE-1(Y669E) (compare Figure 5D with Figures 3G and 4E). These results suggest that the NTD of HRDE-1 robustly recruits the small-RNA amplification machinery to the target and promotes silencing that is independent of the NRDE-2 nuclear silencing pathway. HRDE-1 tethering promotes accumulation of poly-UG-modified target fragments During RNA silencing in worms, truncated target RNAs are converted into templates for small RNA production via the RDE-3-dependent addition of poly-UG tails.27 We therefore used a qPCR assay27 to detect poly-UG additions to reporter RNA in the absence of a λN fusion or in worms expressing λN::HRDE-1, λN::NTD, or λN::HRDE-1(Y669E) (Figures 5G and 5H). Priming from an endogenous UGUG motif in the reporter 3′ UTR serves as a control for the presence of full-length mRNA. This analysis revealed that faster-migrating, poly-UG-modified RNAs accumulated in strains where silencing was active. In wild-type λN::HRDE-1 worms, poly-UG-modified RNAs were most robustly detected at truncations within the GFP sequences (Figures 5G and 5H). As expected, only full-length mRNA was detected in rde-3 mutants, confirming that RDE-3 is absolutely required for poly-UG RNA accumulation. Notably, mutation of nrde-2 or tethering the NTD or Y669E mutants shifted poly-UG addition toward the 3′ end of the reporter, close to the boxB elements (Figures 5G and 5H). These results suggest that HRDE-1 tethering induces RDE-3-dependent poly- UG modification of truncation products that are generated near the tethering sites and Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 8 that nuclear silencing promotes the induction of additional truncations far away from the tethering sites that likely support the 5′ spread of small RNA amplification. To further analyze changes in target RNA caused by tethering, we used qRT-PCR. Surprisingly, whereas tethering wild-type λN::HRDE-1 reduced the reporter pre-mRNA by 50% and mRNA by 99% (Figure 2C), λN::NTD increased the reporter pre-mRNA by ~2.5-fold and reduced the mRNA by ~40% (Figure 5F). This result was surprising given that GFP fluorescence was undetectable in λN::NTD worms (Figures 5C and S5C) and suggested that the accumulating species in λN::NTD animals might reflect the accumulation of nearly full-length pUG RNA. Functional HRDE-1 RNA-induced silencing complex (RISC) is not required parentally for transmission of silencing to offspring We next asked if λN::NTD can initiate inherited silencing. To do this, we first established reporter silencing by tethering λN::NTD in otherwise wild-type worms. We then crossed to a reporter strain homozygous for a hrde-1 null allele to generate animals heterozygous for the tethering construct. Finally, we crossed these λN::NTD/null heterozygotes (either as males or hermaphrodites) to a hrde-1(+) reporter strain, resulting in two types of cross progeny—λN::NTD/+ or null/+ heterozygotes. Remarkably, although the λN::NTD/null parents lacked a functional HRDE-1 RISC, they nevertheless robustly transmitted silencing to the next generation (Figures S7A and S7B). As expected, HRDE-1(+) was required in the inheriting generation for silencing to occur (Buckley et al.15 and Figure 1F). Since the NTD fails to establish heterochromatin upon tethering and cannot directly form a RISC complex, these findings suggest that parentally established heterochromatin and HRDE-1 RISC are not required in gametes for inheritance, a finding consistent with previous work in which hrde-1 homozygous mutant hermaphrodites were shown to transmit silencing to their heterozygous progeny.15 Rather, in the parental generation, the tethered NTD can stimulate amplification of small RNAs that likely engage with other Argonautes to propagate silencing to offspring (see discussion). HRDE-1 localizes to Mutator foci HRDE-1 localization is primarily nuclear15; however, template formation and small RNA amplification are thought to occur in domains of peri-nuclear nuage termed Mutator foci, where several components of the small RNA amplification machinery localize.26–28 To examine whether HRDE-1 localizes in Mutator foci, we expressed GFP::HRDE-1 (without tethering) in worms that also express either mCherry::GLH-1, which localizes broadly within nuage, or MUT-16::mCherry, which localizes prominently in Mutator foci. GFP::HRDE-1 co-localized to a subset of peri-nuclear mCherry::GLH-1 foci, especially in association with late pachytene germ nuclei (Figures 6A and S6A). Moreover, the GFP::HRDE-1 foci only partially overlapped with mCherry::GLH-1 foci, suggesting that the HRDE-1+ foci occupy subdomains of larger GLH-1+ nuage, reminiscent of Mutator foci. Indeed, GFP::HRDE-1 foci coincided almost perfectly with MUT-16::mCherry foci (Figure 6B). Similarly, GFP::HRDE-1(NTD) co-localized with GLH-1::mCherry and mCherry::MUT-16 foci (Figures 6C, 6D, and S6B). Taken together, these findings suggest Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 9 that HRDE-1 localizes via its NTD to Mutator foci, where it functions to promote small RNA amplification. Silencing by dsRNA or tethering causes target genes to co-localize To understand how HRDE-1 and nuclear silencing regulate their target genes and RNAs, we performed RNA and DNA FISH studies to visualize the boxB reporter mRNA and DNA. In the absence of silencing, reporter RNA foci were detected throughout the germline cytoplasm (Figures 6E and S6C). In addition, we observed prominent RNA signals in the majority (~70%) of pachytene nuclei (most nuclei, 57%, exhibited at least two closely paired nuclear dots, while the remainder exhibited a single dot; Figures 6E and 6I). The positions of these nuclear signals adjacent to DAPI-stained chromosomes suggests that they correspond to sites of transcription on the paired sister chromatids within the axial loops of synapsed meiotic homologs. Silencing, induced either by exposure to dsRNA targeting the reporter or by tethering λN::HRDE-1, eliminated cytoplasmic reporter RNA signal and greatly reduced the nuclear signal (Figures 6F, 6L, and S6C). More than 80% of the pachytene nuclei with visible RNA signal exhibited a single nuclear focus (Figures 6F, 6L, 6I, and 6O). The changes in nuclear RNA signal induced by silencing correlated with changes in the reporter DNA FISH signal. In the absence of silencing, we observed a pair of nuclear DNA FISH signals in approximately 50% of pachytene nuclei that have visible DNA signal (Figures 6P and 6T), while in the presence of silencing, we observed a single focus of DNA FISH signal in approximately 90% of pachytene nuclei with visible DNA signal (Figures 6Q, 6J, 6T, and S6E). These results suggest that nuclear silencing mediated by HRDE-1 causes the target alleles to become merged from predominantly paired DNA FISH signals into a single focus containing all 4 silenced alleles. Mutations that disarm nuclear silencing cause target RNA to accumulate in nuage subdomains that resemble Mutator foci We next examined how mutations that disarm only the nuclear silencing pathway impact RNA and DNA localization after RNAi or tethering. To do this, we performed RNA and DNA FISH on λN::NTD worms and on nrde-2 mutants. In these mutants, where nuclear silencing is disarmed, we found that nuclear RNA and DNA FISH signals resembled the nuclear signals observed in wild-type animals in the absence of silencing: predominantly two foci of RNA and DNA FISH signals detected in each background (Figures 6M, 6N, 6I, 6J, 6O, 6T, and S6D). In contrast, however, the cytoplasmic RNA FISH signals were dramatically altered. While RNA signal was absent from the bulk cytoplasm throughout the gonad, consistent with cytoplasmic post-transcriptional silencing, we noticed pronounced accumulation of reporter RNA signals in multiple peri-nuclear foci surrounding pachytene nuclei. Co-staining experiments with GFP::GLH-1 or MUT-16::GFP revealed that these RNA foci coincide with most of the nuage subdomains that express MUT-16::GFP (Figures 6G, 6H, 6M, and 6N). The accumulation of target RNA in the MUT-16 foci required RDE-3(+) activity (Figure S6F), suggesting that these RNA signals may correspond to RdRP templates engaged in small RNA amplification. Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 10 MUT-16 promotes the nuclear localization of GFP::HRDE-1 but not its nuage localization MUT-16 is required for the co-localization of small RNA amplification factors within Mutator foci.26,28,35 We therefore wondered if MUT-16 is also required for the co- localization of HRDE-1 in Mutator foci. To answer this question, we introduced a null allele of mut-16 into worms expressing both GFP::HRDE-1 and mCherry::GLH-1. As shown previously,24 we found that MUT-16 activity is required for the nuclear localization of HRDE-1 (Figures 7A and 7B). MUT-16 was not, however, required for the localization of GFP::HRDE-1 to nuage (Figures 7A and 7B). The localization of GFP::HRDE-1 in nuage appeared more obvious in mut-16 mutants, but the levels of GFP::HRDE-1 within nuage and the approximate numbers of foci appeared similar with or without mut-16 activity (Figures 7A and 7B). Finally, the localization of MUT-16 itself to nuage was not disrupted in hrde-1 mutants (data not shown), thus HRDE-1 and MUT-16 localize within a nuage subdomain (or domains) independently of each other. DISCUSSION In many eukaryotes, the installation and maintenance of chromatin silencing is coupled to Argonaute small RNA pathways that promote transmission to offspring. Here, we have explored the role of a nuclear Argonaute HRDE-1 in coordinating transgenerational silencing in the C. elegans germline. In addition to its known role in directing heterochromatin silencing downstream of RNAi13,15 and Piwi Argonaute silencing,8,9,14 our tethering studies have shown that HRDE-1 is also de novo loaded with small RNA, downstream of heterochromatin silencing, enabling it to prime a new round of small RNA amplification within nuage (Figure 7C, model). The nuclear silencing events that depend on HRDE-1 cause the target alleles to co-localize into a single focus of DNA FISH signal (Figures 6P–6S and S6E). Presumably, the heterochromatinized alleles within this focus are transcribed at low levels to produce template RNA that feeds transgenerational silencing; indeed, the continued expression of the target locus after heterochromatin induction is a conserved feature of co-transcriptional small RNA silencing.36 Consistent with this idea, the inactivation of heterochromatin silencing caused target alleles to remain separated and increased the levels of the nuclear- and nuage-localized RNA signals as measured by RNA FISH. The failure to engage nuclear silencing did not de-silence protein expression in the context of our tethering studies nor indeed in previously published studies on nuclear-silencing mutants when an RNAi trigger is present.13,15 Instead, our RNA FISH studies suggest that unabated transcription of the target gene feeds increased levels of target RNA localization in nuage (also noted in a recent study by Ouyang et al.37) and that small RNA levels also increase dramatically to compensate and silence mRNA expression. Taken together, our findings suggest that when the nuclear heterochromatin pathways are inactive, the target mRNA is silenced by a combination of cytoplasmic clearance or trapping in the P granule. In the yeast S. pombe, the RNAi-induced transcriptional silencing complex (RITS), which includes an RdRP and a nuclear Argonaute AGO1p, resides in heterochromatin. A previous study showed that tethering of AGO1p to RNA via a boxB reporter system, similar to the Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 11 one used here, was sufficient to recruit the RITS complex, induce small RNA amplification, and drive reporter silencing25. HRDE-1 associates with NRDE-2 and components of the nucleosome re-modeling and deacetylase NuRD complex to establish heterochromatin silencing.15,18,38 How heterochromatin leads to de novo programming of HRDE-1 is nevertheless unknown. In C. elegans, the RdRP EGO-1 has been shown to associate with germline chromatin,39,40 and several of our findings would be consistent with a cycle of nuclear small RNA transcription and de novo HRDE-1 loading within heterochromatin. Such a mechanism could explain why tethering NRDE-2 in the absence of HRDE-1 initiates heterochromatin silencing but not small RNA amplification (Figures 2E and 3E). Perhaps after a nuclear cycle of HRDE-1 loading, the protein exits the nucleus along with nascent target/template RNA to further amplify small RNA production. Consistent with this idea, we have shown that the N-terminal half of HRDE-1 is sufficient to stimulate small RNA amplification and loading and that both the NTD and full-length HRDE-1 (as well as target RNA) localize within a specialized nuage domain known as Mutator foci. Mutator foci accumulate poly-UG-modified templates derived from target RNA27 and are thought to serve in the amplification of small RNA signals that are propagated to offspring. Thus, our findings suggest that HRDE-1 shuttles out of the nucleus to nuage to promote small RNA amplification. A mutant HRDE-1 protein incapable of binding guide RNA was sufficient (when tethered) to induce silencing that transmits to offspring via either the sperm or the egg (Figures S7A and S7B). Thus, as previously reported,15 a functional HRDE-1 RISC is not required in gametes for transgenerational silencing but is required in offspring to renew silencing for another generation (Buckley et al.15 and Figure 1F). In the parental germline, Mutator foci likely serve as locations where HRDE-1 and other upstream Argonautes trigger the expansion of small RNAs that are loaded onto downstream WAGO Argonautes, including the two prominent nuage-localized Argonautes WAGO-18 and WAGO-4.41 Consistent with this idea, silencing induced by λN::HRDE-1(Y669E) was partially dependent on wago-1 (75% de-silenced, N = 32, and Figure S4G). Taken together, our findings suggest that heterochromatin renews small RNA silencing (and vice versa) during each germline life cycle. For example, small RNAs guide heterochromatin formation in the zygote, and heterochromatin then propagates silencing before feeding back into the de novo synthesis of guide RNAs that load onto HRDE-1. HRDE-1 promotes expansion of small RNAs that are then transmitted to offspring through HRDE-1 and other WAGOs to re-establish heterochromatin. Heterochromatin then, in turn, transcribes RNA that forms templates for RdRP-dependent amplification, renewing the cycle. Consistent with these ideas, neither pathway, small RNA or heterochromatin alone, is sufficient to stably transmit silencing signals for multiple generations8,9,13,15 (Figures S7C–S7F). Given the similarities between the worm and yeast mechanisms—and by extension, the intriguing relationships between long non-coding RNAs and chromatin modifiers in flies and mammals7—feedforward RNA-chromatin circuits that amplify and maintain silencing across cell divisions or generations will likely be a common feature of gene regulation in eukaryotes. Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Limitations of the study Page 12 In this study, we use an artificial mechanism to recruit RNA silencing factors to their targets. Recruiting, factors via the λN/boxB system may elicit non-physiological mechanisms that block gene expression. For example, tethering factors to the reporter UTR could prevent proper recruitment of translation-initiation machinery or 3′ end processing factors. Transcripts that are not processed properly (for example, unspliced mRNA11) could trigger default recruitment of the same RNA silencing factors that mediate physiological silencing in response to bona fide Argonaute-guided silencing. To control for such possibilities, we used genetics to dissect the nature of the silencing pathways induced by tethering and found that tethering different factors elicited different genetic dependencies for silencing. For example, λN::NRDE-2 required nrde-4(+) activity for silencing but λN::HRDE-1 tethering did not. We have controlled for possible artifacts by initiating parallel studies on untethered factors and by using a combination of genetics, microscopy, and RNA-expression profiling. Together, these studies give us high confidence that tethering, in these instances, has faithfully replicated actual physiological steps in silencing. STAR★METHODS RESOURCE AVAILABILITY Lead contact—Further information and requests for resources and materials should be directed to and will be fulfilled by the lead contact, Craig Mello (Craig.Mello@umassmed.edu). Materials availability—All materials generated in this study are available from the lead contact without restrictions. Data and code availability—Original small-RNA sequencing datasets are publicly available in NCBI under the accession number BioProject: PRJNA874806. This study did not generate any new code, but the scripts used in the study are available from the lead contact upon request. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. EXPERIMENTAL MODEL AND SUBJECT DETAILS All the strains used in this study were derived from C. elegans Bristol N2 (CGC) and cultured on nematode growth media (NGM) plates with E. coli OP5043 or E. coli HT115 for RNAi experiments. Strains used in this study were generated by CRISPR-cas9 method or Cross (see Table S1 for details). METHOD DETAILS CRISPR-Cas9 genome editing—The Cas9 ribonucleoprotein (RNP) CRISPR strategy44 were used to edit the genome. Plasmid pRF4 containing rol-6 (su-1006) was used as co-injection marker. For short insertions like λN and deletion mutations, synthesized Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 13 single-strand DNAs were used as the donor; for long insertions like GFP, mCherry, and 5xBoxB, the annealed PCR products were used instead. The gRNA and donor sequences were listed in Table S2. The BoxB reporter strain was constructed based on a single copy insertion of Ppie-1:GFP::his-58:unc-54UTR (WM701). The 5xBoxB sequence amplified from a previously published strain JMC00222 was inserted before the unc-54 UTR. Live worm fluorescent image—Young adult worms were transferred to glass slide in M9 buffer with 0.4mM Tetramisole. Epifluorescence and differential interference contrast (DIC) microscopy were performed on a Zeiss Axio Imager M2 Microscope and images were processed with ZEN Microscopy Software (Zeiss). Confocal images were taken by a Andor Dragonfly Spinning Disk confocal microscope. Confocal images were processed with Imaris Microscopy Image Analysis Software. Quantifying reporter RNA using qPCR—Young adult worms were collected and washed with M9 for three times and ddH2O once. Total RNA was extracted with TRIZOL and treated with DNase I to remove DNA contamination. First strand cDNA was synthesized by Superscript IV with random hexamers. Quantitative PCR was performed on a Quant studio 5 Real-time PCR machine together with Fast SYBR Green Master Mix. Actin was used as internal reference (primer set S5265 and S527). Primer set of oYD826 and oYD827 were used for reporter. All primers used were listed in Table S2. CHIP-qPCR—A traditional worm CHIP method45 was applied to the young adult worm samples. Anti H3K9me3 antibody (Upstate 07523) and CHIP grade IgA/G magnetic beads were used for the immunoprecipitation. During elution, RNase A and Protease K were used to remove RNA and proteins. For qPCR, actin was used as internal reference. All primers used were listed in Table S2. Small RNA cloning and data analysis—Small RNA cloning was conducted as previously reported.6 Synchronized young adult worms were collected and total RNA were purified with Trizol. Two biological repeats were included for each strain. Small RNAs were enriched using a mirVana miRNA isolation kit. Homemade PIR-1 was used to remove the di or triphosphate at the 5′ to generate 5′ monophosphorylated small RNA. Adaptors of 3’ (DA35) and 5’ (DA4) were ligated to the small RNA by T4 RNA ligase 2 (NEB) and T4 ligase 1 (NEB) sequentially. Reverse transcription was performed with SuperScript III and RT primer (DA5). After PCR amplification, productions around 150 bp were separated by 12% SDS-PAGE and equally mixed. Libraries were sequenced on a NextSeq 550 sequencer with the illumina NextSeq 500/550 high output kit in 75bp single-end sequencing mod. Reads were trimmed by cutadapt and mapped using Bowtie2.42 For small RNAs mapped to the reporter, total reads with length longer than 16 nt were used to normalized between samples. Plots were generated by R and R studio. QUANTIFICATION AND STATISTICAL ANALYSIS To determine the genes with increased or decreased antisense small RNAs (Figures S4B and S4C), small RNAs were cloned and sequenced as described above with two biological Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 14 repeats for each strain. DEseq2 package in R was used to find out genes with 2-fold decrease of antisense small RNA (p value ≤ 0.05) in hrde-1(null) or hrde-1(Y669E) compared to WT. Structure prediction—The 3D structure of HRDE-1 was predicted by I-TASSER online server33 with default setting. HRDE-1 structure was aligned with hAgo2 by PyMOL46 and its domains were annotated based on the alignment. pUG RNA analysis—As previously reported,27 total RNAs were extracted with Trizol. SuperScript IV was used to generate the first strand DNA with reverse transcription primer oYD1001. A pair of outer primers (oYD998 and oYD1002) were used for the first round PCR amplification with Taq DNA polymerase. After 100-fold dilution, another round of PCR was performed with a pair of inner primers (oYD256 and oYD1003). PCR products were analyzed by 1.5% agarose gels. DNA bands were purified, cloned with TOPO TA Cloning Kit and sent for sanger sequencing. gsa-1 served as a control for pUG PCR analysis. RNA FISH—Worms at young adult stage were dissected in Happy Buffer (81mM HEPES pH 6.9, 42mM NaCl, 5mM KCl, 2mM MgCl2, 1mM EGTA) (From personal correspondence with James Priess). Dissected gonads were transferred to poly-lysine treated dish with 80 μl of Happy Buffer and fixed by adding equal volume of 5% formaldehyde in PBST (PBS+0.1% Tween 20) for 30 min. After one wash with PBST, gonads were treated with PBST-Triton (PBST+0.1% Triton) for 10 min, washed with PBST again and emerged in 70% ethanol for 30 min to overnight. Before hybridization, samples were washed with fresh wash buffer (2xSSC +10% formamide) for 5 min hybridization was performed at 37°C for 18 h to overnight in hybridization buffer (900 μl Stellaris RNA FISH Hybridization Buffer+ 100ul formamide) with 10 pmol RNA FISH probes. Samples were washed with wash buffer, once quick wash, one wash for 30 min at 37°C and two quick washes. Mounting medium with DAPI was added to preserve the signal. Confocal images were taken with an Andor Dragonfly Spinning Disk confocal microscope and processed with Fusion and Imaris. DNA FISH—Same to RNA FISH, gonads were dissected, fixed and washed with PBST and treated with 70% ethanol. Then, samples were washed with wash buffer three times, one at room temperature for 5 min, one at 95°C for 3 min, and one at 60°C for 20 min. Hybridization was performed in hybridization buffer (700 μl Stellaris RNA FISH Hybridization Buffer +300 μl formamide + primary probes (final 10 pmol) + detection probe (final 10 pmol)) at 95°C for 5 min and then transferred to 37°C for 3 h to overnight. After hybridization, samples were wash with 2xSSC for 20 min at 60°C, and then 2xSSCT (2xSSC +0.3% Triton X-100) for 5 min at 60°C and another 20 min at 60°C. After another wash with 2xSSCT for 5 min at room temperature, samples were preserved in the mounting medium with DAPI. Confocal images were taken with an Andor Dragonfly Spinning Disk confocal microscope and processed with Fusion and Imaris. Primary probes of DNA FISH were picked from the oligo lists generated by OligoMiner.47 RNAi experiments—Synchronous L1 worms of the reporter strain were plated on NGM plates for 48 h. Then the worms were collected and washed with M9. About 100 worms Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 15 were plated on every IPTG plate with the gfp RNAi food. After 24 h, worms were dissected for the FISH experiment. RNA FISH and DNA FISH were performed as described above. Supplementary Material Refer to Web version on PubMed Central for supplementary material. ACKNOWLEDGMENTS We thank members of Mello and Ambros labs for discussions; James Priess (Fred Hutchinson Cancer Center) for sharing the receipt of happy buffer and imaging experiences; Weifeng Gu (University of California, Riverside) for providing the PIR-1 protein for small RNA cloning; Ahmet Ozturk for building the small RNA analysis pipeline; Darryl Conte for critical comments and edits on the manuscript; and the RNA Therapeutics Institute for offering the Nextseq 550 sequencing machine. The work was supported by NIH funding (GM058800 and HD078253) to C.C.M. C.C.M. is a Howard Hughes Medical Institute Investigator. REFERENCES 1. Kasschau KD, Fahlgren N, Chapman EJ, Sullivan CM, Cumbie JS, Givan SA, and Carrington JC (2007). Genome-wide profiling and analysis of Arabidopsis siRNAs. PLoS Biol. 5, e57. [PubMed: 17298187] 2. Czech B, Malone CD, Zhou R, Stark A, Schlingeheyde C, Dus M, Perrimon N, Kellis M, Wohlschlegel JA, Sachidanandam R, et al. (2008). An endogenous small interfering RNA pathway in Drosophila. Nature 453, 798–802. [PubMed: 18463631] 3. Ghildiyal M, Seitz H, Horwich MD, Li C, Du T, Lee S, Xu J, Kittler ELW, Zapp ML, Weng Z, and Zamore PD (2008). Endogenous siRNAs derived from transposons and mRNAs in Drosophila somatic cells. Science 320, 1077–1081. [PubMed: 18403677] 4. Tam OH, Aravin AA, Stein P, Girard A, Murchison EP, Cheloufi S, Hodges E, Anger M, Sachidanandam R, Schultz RM, and Hannon GJ (2008). Pseudogene-derived small interfering RNAs regulate gene expression in mouse oocytes. Nature 453, 534–538. [PubMed: 18404147] 5. Watanabe T, Totoki Y, Toyoda A, Kaneda M, Kuramochi-Miyagawa S, Obata Y, Chiba H, Kohara Y, Kono T, Nakano T, et al. (2008). Endogenous siRNAs from naturally formed dsRNAs regulate transcripts in mouse oocytes. Nature 453, 539–543. [PubMed: 18404146] 6. Gu W, Shirayama M, Conte D Jr., Vasale J, Batista PJ, Claycomb JM, Moresco JJ, Youngman EM, Keys J, Stoltz MJ, et al. (2009). Distinct argonaute-mediated 22G-RNA pathways direct genome surveillance in the C. elegans germline. Mol. Cell 36, 231–244. [PubMed: 19800275] 7. Weick E-M, and Miska EA (2014). piRNAs: from biogenesis to function. Development 141, 3458– 3471. [PubMed: 25183868] 8. Shirayama M, Seth M, Lee H-C, Gu W, Ishidate T, Conte D Jr., and Mello CC (2012). piRNAs initiate an epigenetic memory of nonself RNA in the C. elegans germline. Cell 150, 65–77. [PubMed: 22738726] 9. Ashe A, Sapetschnig A, Weick E-M, Mitchell J, Bagijn MP, Cording AC, Doebley A-L, Goldstein LD, Lehrbach NJ, Le Pen J, et al. (2012). piRNAs can trigger a multigenerational epigenetic memory in the germline of C. elegans. Cell 150, 88–99. [PubMed: 22738725] 10. Fire A, Xu S, Montgomery MK, Kostas SA, Driver SE, and Mello CC (1998). Potent and specific genetic interference by double-stranded RNA in Caenorhabditis elegans. nature 391, 806–811. [PubMed: 9486653] 11. Makeyeva YV, Shirayama M, and Mello CC (2021). Cues from mRNA splicing prevent default Argonaute silencing in C. elegans. Dev. Cell 56, 2636–2648.e4. [PubMed: 34547227] 12. Lee HC, Gu W, Shirayama M, Youngman E, Conte D Jr., and Mello CC (2012). C. elegans piRNAs mediate the genome-wide surveillance of germline transcripts. Cell 150, 78–87. 10.1016/ j.cell.2012.06.016. [PubMed: 22738724] 13. Spracklin G, Fields B, Wan G, Becker D, Wallig A, Shukla A, and Kennedy S (2017). The RNAi inheritance machinery of Caenorhabditis elegans. Genetics 206, 1403–1416. [PubMed: 28533440] Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 16 14. Sapetschnig A, Sarkies P, Lehrbach NJ, and Miska EA (2015). Tertiary siRNAs mediate paramutation in C. elegans. PLoS Genet. 11, e1005078. 15. Buckley BA, Burkhart KB, Gu SG, Spracklin G, Kershner A, Fritz H, Kimble J, Fire A, and Kennedy S (2012). A nuclear Argonaute promotes multigenerational epigenetic inheritance and germline immortality. Nature 489, 447–451. [PubMed: 22810588] 16. Rechavi O, i-Ze’evi L, Anava S, Goh WSS, Kerk SY, Hannon GJ, and Hobert O (2014). Starvation-induced transgenerational inheritance of small RNAs in C. elegans. Cell 158, 277–287. [PubMed: 25018105] 17. Almeida MV, Andrade-Navarro MA, and Ketting RF (2019). Function and evolution of nematode RNAi pathways. Noncoding. RNA 5, 8. [PubMed: 30650636] 18. Kim H, Ding Y-H, Zhang G, Yan Y-H, Conte D Jr., Dong M-Q, and Mello CC (2021). HDAC1 SUMOylation promotes Argonaute-directed transcriptional silencing in C. elegans. Elife 10, e63299. 19. Towbin BD, González-Aguilera C, Sack R, Gaidatzis D, Kalck V, Meister P, Askjaer P, and Gasser SM (2012). Step-wise methylation of histone H3K9 positions heterochromatin at the nuclear periphery. Cell 150, 934–947. [PubMed: 22939621] 20. Luteijn MJ, Van Bergeijk P, Kaaij LJT, Almeida MV, Roovers EF, Berezikov E, and Ketting RF (2012). Extremely stable Piwi-induced gene silencing in Caenorhabditis elegans. The EMBO journal 31, 3422–3430. [PubMed: 22850670] 21. Baron-Benhamou J, Gehring NH, Kulozik AE, and Hentze MW (2004). Using the λN peptide to tether proteins to RNAs. In mRNA Processing and Metabolism (Springer), pp. 135–153. 22. Wedeles CJ, Wu MZ, and Claycomb JM (2013). Protection of germline gene expression by the C. elegans Argonaute CSR-1. Dev. Cell 27, 664–671. [PubMed: 24360783] 23. Aoki ST, Lynch TR, Crittenden SL, Bingman CA, Wickens M, and Kimble J (2021). C. elegans germ granules require both assembly and localized regulators for mRNA repression. Nat. Commun. 12, 996–1014. [PubMed: 33579952] 24. Cornes E, Bourdon L, Singh M, Mueller F, Quarato P, Wernersson E, Bienko M, Li B, and Cecere G (2022). piRNAs initiate transcriptional silencing of spermatogenic genes during C. elegans germline development. Dev. Cell 57, 180–196.e7. [PubMed: 34921763] 25. Bühler M, Verdel A, and Moazed D (2006). Tethering RITS to a nascent transcript initiates RNAi-and heterochromatin-dependent gene silencing. Cell 125, 873–886. [PubMed: 16751098] 26. Phillips CM, Montgomery TA, Breen PC, and Ruvkun G (2012). MUT-16 promotes formation of perinuclear mutator foci required for RNA silencing in the C. elegans germline. Genes Dev. 26, 1433–1444. [PubMed: 22713602] 27. Shukla A, Yan J, Pagano DJ, Dodson AE, Fei Y, Gorham J, Seidman JG, Wickens M, and Kennedy S (2020). Poly (UG)-tailed RNAs in genome protection and epigenetic inheritance. Nature 582, 283–288. [PubMed: 32499657] 28. Zhang C, Montgomery TA, Gabel HW, Fischer SEJ, Phillips CM, Fahlgren N, Sullivan CM, Carrington JC, and Ruvkun G (2011). mut-16 and other mutator class genes modulate 22G and 26G siRNA pathways in Caenorhabditis elegans. Proc. Natl. Acad. Sci. USA 108, 1201–1208. [PubMed: 21245313] 29. Guang S, Bochner AF, Burkhart KB, Burton N, Pavelec DM, and Kennedy S (2010). Small regulatory RNAs inhibit RNA polymerase II during the elongation phase of transcription. Nature 465, 1097–1101. [PubMed: 20543824] 30. Jiao AL, Perales R, Umbreit NT, Haswell JR, Piper ME, Adams BD, Pellman D, Kennedy S, and Slack FJ (2019). Human nuclear RNAi-defective 2 (NRDE2) is an essential RNA splicing factor. RNA 25, 352–363. [PubMed: 30538148] 31. Billi AC, Fischer SE, and Kim JK Endogenous RNAi pathways in C. elegans (May 7, 2014). WormBook, ed. The C. elegans Research Community, WormBook, 10.1895/wormbook.1.170.1, [http://www.wormbook.org]. 32. Rüdel S, Wang Y, Lenobel R, Körner R, Hsiao H-H, Urlaub H, Patel D, and Meister G (2011). Phosphorylation of human Argonaute proteins affects small RNA binding. Nucleic Acids Res. 39, 2330–2343. [PubMed: 21071408] Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 17 33. Yang J, and Zhang Y (2015). I-TASSER server: new development for protein structure and function predictions. Nucleic Acids Res. 43, W174–W181. [PubMed: 25883148] 34. Schirle NT, Sheu-Gruttadauria J, and MacRae IJ (2014). Structural basis for microRNA targeting. Science 346, 608–613. [PubMed: 25359968] 35. Uebel CJ, Anderson DC, Mandarino LM, Manage KI, Aynaszyan S, and Phillips CM (2018). Distinct regions of the intrinsically disordered protein MUT-16 mediate assembly of a small RNA amplification complex and promote phase separation of Mutator foci. PLoS Genet. 14, e1007542. 36. Holoch D, and Moazed D (2015). RNA-mediated epigenetic regulation of gene expression. Nat. Rev. Genet. 16, 71–84. [PubMed: 25554358] 37. Ouyang JPT, Zhang WL, and Seydoux G (2022). The conserved helicase ZNFX-1 memorializes silenced RNAs in perinuclear condensates. Nat. Cell Biol. 24, 1129–1140. [PubMed: 35739318] 38. Wan G, Yan J, Fei Y, Pagano DJ, and Kennedy S (2020). A conserved NRDE-2/MTR-4 complex mediates nuclear RNAi in Caenorhabditis elegans. Genetics 216, 1071–1085. [PubMed: 33055090] 39. Maine EM, Hauth J, Ratliff T, Vought VE, She X, and Kelly WG (2005). EGO-1, a putative RNA-dependent RNA polymerase, is required for heterochromatin assembly on unpaired DNA during C. elegans meiosis. Curr. Biol. 15, 1972–1978. [PubMed: 16271877] 40. Claycomb JM, Batista PJ, Pang KM, Gu W, Vasale JJ, van Wolf-swinkel JC, Chaves DA, Shirayama M, Mitani S, Ketting RF, et al. (2009). The Argonaute CSR-1 and its 22G-RNA cofactors are required for holocentric chromosome segregation. Cell 139, 123–134. [PubMed: 19804758] 41. Xu F, Feng X, Chen X, Weng C, Yan Q, Xu T, Hong M, and Guang S (2018). A cytoplasmic Argonaute protein promotes the inheritance of RNAi. Cell Rep. 23, 2482–2494. [PubMed: 29791857] 42. Langmead B, and Salzberg SL (2012). Fast gapped-read alignment with Bowtie 2. Nat. Methods 9, 357–359. [PubMed: 22388286] 43. Brenner S (1974). The genetics of Caenorhabditis elegans. Genetics 77, 71–94. [PubMed: 4366476] 44. Dokshin GA, Ghanta KS, Piscopo KM, and Mello CC (2018). Robust genome editing with short single-stranded and long, partially single-stranded DNA donors in Caenorhabditis elegans. Genetics 210, 781–787. [PubMed: 30213854] 45. Askjaer P, Ercan S, and Meister P (2014). Modern techniques for the analysis of chromatin and nuclear organization in C. elegans. WormBook, 1–35. 10.1895/wormbook.1.169.1. 46. Schrodinger L (2010). The PyMOL Molecular Graphics System, Version 2.4.0 (Schrodinger, L.). 47. Beliveau BJ, Kishi JY, Nir G, Sasaki HM, Saka SK, Nguyen SC, Wu C. t., and Yin P (2018). OligoMiner provides a rapid, flexible environment for the design of genome-scale oligonucleotide in situ hybridization probes. Proc. Natl. Acad. Sci. USA 115, E2183–E2192. [PubMed: 29463736] Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 18 Highlights • • • • Nuclear Argonaute HRDE-1 separately induces heterochromatin and small RNA production HRDE-1 induces target alleles to merge into a single focus of heterochromatin Transcription within heterochromatin feeds de novo loading of HRDE-1 with small RNAs HRDE-1 shuttles to nuage and promotes small RNA production via its N- terminal domain Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 19 Figure 1. HRDE-1 tethering caused reporter silencing and generated the silencing memory (A) Scheme of λN-BoxB tethering system. A sequence encoding five BoxB hairpins (5xBoxB) was inserted immediately after the coding region of the GFP::his-58(H2B) transgene and before the unc-54 3′ UTR. The reporter is driven by the pie-1 promoter (Ppie-1). The BoxB sites recruit λN::HRDE-1 or λN::NRDE-2 fusion proteins, thereby tethering HRDE-1 or NRDE-2 to the reporter RNA. (B) Representative fluorescence image of a syncytial germline (outlined by dashed lines) in the absence of tethering. The image represents 100% of worms scored, N > 30. (C) Representative fluorescence image in the presence of HRDE-1 tethering. The image represents 100% of worms scored, N > 30. (D) Representative fluorescence image in the presence of NRDE-2 tethering. The image represents 100% of worms scored, N > 30. (E and F) Analysis of inherited silencing triggered by λN::HRDE-1 tethering. After outcross to hrde-1 wild type (E) or hrde-1 null (F), reporter worms were scored for gfp expression for 13 generations after segregating away the λN::hrde-1 allele. The percentage of GFP+ (ON) or GFP– (OFF) worms is indicated, N > 30 worms scored in each generation. (G) Color chart showing genetic requirements of inherited silencing triggered by λN::HRDE-1 tethering. The λN::hrde-1; reporter worms were crossed to the indicated mutants. After segregating away λN::hrde-1, reporter worms homozygous for the indicated mutations were scored for GFP expression: ON or OFF, as indicated. N > 30 worms scored for each genotype. Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 20 Figure 2. HRDE-1 and NRDE-2 tethering induce heterochromatin formation (A and B) Quantification of H3K9me3 levels near the reporter in the presence or absence of HRDE-1 or NRDE-2 tethering, as determined by chromatin immunoprecipitation (ChIP)- qPCR. P1 and P4 primer sets analyze sequences 5 kb upstream or downstream of the reporter, and P2 and P3 analyze sequences within the reporter, as indicated in the schematic. All quantities were normalized to the level of P1 in reporter control samples. Error bars show the standard deviation from the mean. (C and D) Bar graphs showing the quantification of reporter RNA and pre-mRNA levels in response to HRDE-1 or NRDE-2 tethering, as determined by qPCR. The average quantities relative to wild type (WT) are indicated. Error bars show the standard deviation from the mean. (E and F) Color chart showing the genetic requirements of silencing in the presence of λN::NRDE-2 or λN::HRDE-1. Reporter worms homozygous for the indicated mutations were scored for GFP expression: ON or OFF, as indicated. N > 30 worms scored for each genotype. *GFP is ON, but signal is weak (see Figure S2H). Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 21 Figure 3. HRDE-1 and NRDE-2 tethering promote antisense small RNA production (A) Plot showing antisense small RNA reads (per million total reads) mapping to the reporter (indicated below the plot) in the absence of tethering. Only the first nucleotide is counted. Green boxes, GFP coding; blue box, H2B coding; pink boxes, BoxB hairpins. (B–E) Genetic requirements of small RNAs induced by NRDE-2 tethering. Plots showing antisense small RNA reads mapping to the reporter in the presence of λN::NRDE-2 in WT (B), nrde-4 (C), rde-3 (D), or hrde-1 (E) worms. Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 22 (F–I) Genetic requirements of small RNAs induced by HRDE-1 tethering. Plots showing antisense small RNA reads mapping to the reporter in the presence of λN::HRDE-1 in WT (F), nrde-2 (G), rde-3 (H), or mut-16 (I) worms. (J–M) Genetic requirements of inherited small RNAs induced by HRDE-1 tethering. Plots showing antisense small RNA reads mapping to the reporter in WT (F), nrde-2 (G), rde-3 (H), or mut-16 (I) worms after segregating λN::HRDE-1. Note that in (G), the y axis is compressed 50% compared with other plots to conserve space. Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 23 Figure 4. HRDE-1 guide RNA loading is not required for small RNA amplification (A) Western blot analysis to detect GFP::HRDE-1 and GFP::HRDE-1(Y669E) in worm lysates. Top panel: probed with anti-GFP antibody. GFP::HRDE-1 was indicated. Bottom panel: probed with anti-tubulin antibody as a loading control. (B) Confocal images showing the localization of GFP::HRDE-1(WT) or GFP::HRDE-1(Y669E) with mCherry::GLH-1 as P granule marker. The white dashed lines outline a gonadal arm of the germline. (C) Representative fluorescence (left panels) and differential interference contrast (DIC; right) images showing that λN::HRDE-1(Y669E) silences the BoxB reporter in WT worms (top panels, OFF) but not in rde-3 mutant worms (bottom panels, ON). The images represent 100% of the animals scored, N > 30. Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 24 (D) Bar graphs showing the quantification of reporter RNA and pre-mRNA levels in response to HRDE-1(Y669E) tethering, determined by qPCR. The average quantities relative to WT are indicated. Error bars show the standard deviation from the mean. (E and F) Plots showing antisense small RNA reads mapping to the reporter in the presence of λN::HRDE-1(Y669E) in WT (E) or rde-3 (F) worms. Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 25 Figure 5. HRDE-1 N-terminal domain promotes small RNA amplification and poly-UG modification (A) Schematic showing HRDE-1 linear domain structure and truncations tested. The subdomains are color coded based on human Ago2 (Figure S5A). The percentage of GFP+ worms (ON) is indicated, N > 30 worms scored in each test. (B) Predicted three-dimensional structures of HRDE-1 N-terminal domain (NTD) and C- terminal domain (CTD). Subdomains as in (A). (C) Color chart indicating the expression (ON) or silencing (OFF) of the reporter in the presence of λN::CTD or λN::NTD and the requirement of nrde-2 or rde-3. N > 30 worms scored for each genotype. (D and E) Plots showing antisense small RNA reads mapping to the reporter in the presence of λN::NTD in WT (D) or rde-3 worms (E). (F) Bar graphs showing the quantification of reporter RNA and pre-mRNA levels in response to NTD tethering, as determined by qPCR. The average quantities relative to the control are indicated. Error bars show the standard deviation from the mean. Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 26 (G and H) Analysis of poly-UG modification of reporter RNA in response to tethering in the indicated mutants. Poly-UG PCR products in (G) were cloned and sequenced to identify the precise positions of poly-UG addition (H), indicated by arrowheads. A gsa-1-specific PCR was used as loading control. Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 27 Figure 6. HRDE-1 localizes in Mutator foci, and HRDE-1 tethering caused peri-nuclear accumulation of reporter RNA in nuclear silencing mutants (A and B) Confocal image of live germ cells showing the co-localization of GFP::HRDE-1 with mCherry::GLH-1 (A) and MUT-16::mCherry (B). Each subpanel shows a projected view of a segment of the germline to the left, and the nucleus bounded by a dashed box is shown as a single-focal-plane image to the right. Yellow arrows point to peri-nuclear foci where HRDE-1 co-localizes with GLH-1 and MUT-16. (C and D) Confocal images of live germ cells showing the co-localization of GFP::HRDE-1(NTD) with mCherry::GLH-1 (C) and MUT-16::mCherry (D). As in (A) and Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 28 (B). (E and F) Confocal images of RNA FISH experiments showing the localization of reporter RNA with mCherry::GLH-1 (left) or MUT-16::mCherry (right) in control worms (E) or in worms exposed to gfp RNAi (F). Magenta, RNA; green, GLH-1 or MUT-16; and blue, DAPI. Each subpanel shows a projected view of a segment of a representative germline to the left, and the nucleus bounded by a dashed box is shown as a single-focal-plane image with DNA and GLH-1 or MUT-16 signals (center) or with DNA signal only (right). Yellow arrows point to nuclear RNA foci that likely correspond to transcription sites. (G and H) As in (F) but in hrde-1 (G) or nrde-2 (H) mutant worms. (I) Bar graphs showing the percentage of nuclei from (E)–(H) containing one reporter RNA focus (orange) or two or more reporter RNA foci (light green). Three independent germlines were measured for each condition. Error bars show the standard deviation from the mean. (J) Bar graphs showing the percentage of nuclei from DNA FISH (Figure S6E) containing one reporter DNA focus or two or more reporter DNA foci. Similar to (I). (K–N) Confocal images of RNA FISH showing the localization of reporter RNA with mCherry::GLH-1 (left) or MUT-16::mCherry (right) in the absence (K) or presence (L–N) of HRDE-1 tethering, as indicated. Details as in (E) and (F). (O) Bar graphs showing the percentage of nuclei from (J)–(N) containing one reporter RNA focus (peach) or two or more reporter RNA foci (light green). (P–S) Confocal images of DNA FISH experiments showing the localization of reporter DNA loci in the absence (P) or presence (Q–S) of HRDE-1 tethering, as indicated. Green, DNA FISH signal; blue, DAPI. A projected view of a segment of a representative germline is shown to the left, and the nucleus bounded by a dashed box is shown as a single-focal-plane image to the right. Yellow arrows point to the nuclear DNA signals. (T) Bar graphs showing the percentage of nuclei from DNA FISH experiments in (P)–(S) containing one reporter DNA focus or two or more reporter DNA foci. Details as in (I). Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 29 Figure 7. Model of HRDE-1-mediated self-enforcing mechanism (A and B) Confocal images showing the localization of GFP::HRDE-1 with mCherry::GLH-1 in WT worms (A) and mut-16 mutants (B). Green, GFP::HRDE-1 (left); magenta, mCherry::GLH-1 (middle); merge (right). Each subpanel shows a projected image of a representative pachytene region of the germline to the left, and the nucleus bounded by a dashed box is shown as a single-focal-plane image to the right. (C) Model (see Discussion). Cell Rep. Author manuscript; available in PMC 2023 August 22. Ding et al. Page 30 A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t 8 2 3 5 0 3 _ B A D I F R : ; 0 6 1 6 b a # t a C 7 8 6 0 1 3 _ B A D I F R : ; 3 2 5 7 0 # t a C 0 5 P O : e s a B m r o W 5 1 1 T H : e s a b m r o W m a c b A e t a t s p U C G C C G C S 6 5 9 2 # t a C g n i l a n g i S l l e C s n i a r t s s u r i v d n a l a i r e t c a B 0 5 P O oli: E. C 5 1 1 T H oli: E. C n i l u b u T - i t n A P F G - i t n A 3 e m 9 K 3 H - i t n A s e i d o b i t n A R E I F I T N E D I E C R U O S E C R U O S E R r o T N E G A E R E L B A T S E C R U O S E R Y E K 9 5 0 1 8 0 1 # t a C ) T D I ( s e i g o l o n h c e T A N D d e t a r g e t n I 3 V e s a e l c u N 9 s a C . p . S ® R - t l A s n i e t o r p t n a n i b m o c e r d n a , s e d i t p e p , s l a c i m e h C 4 3 5 2 7 0 1 # t a C 6 5 7 9 L # t a C 4 2 4 9 T # t a C S 3 0 3 0 M # t a C 0 0 2 1 9 0 8 1 # t a C 6 1 6 6 2 # t a C 2 6 1 6 2 # t a C 0 7 2 2 M A # t a C - O R K T O R P R # t a C 6 9 6 2 m a # t a C L 3 7 3 0 M # t a C S 4 0 2 0 M # t a C 5 8 0 0 8 0 8 1 # t a C 0 2 8 6 5 2 1 # t a C S 3 7 2 0 M # t a C 1 0 0 6 4 4 1 1 4 1 1 # t a C 0 0 5 – 7 3 3 P B h c i r d l A - a m g i S h c i r d l A - a m g i S B E N T D I s e i g o l o n h c e T e f i L s e i g o l o n h c e T e f i L e s a t p i r c s n a r T e s r e v e R V I ™ t p i r c S r e p u S x i M r e t s a M n e e r G R B Y S t s a F e d i r o l h c o r d y h e l o s i m a r t e T A N R r c a r t - 9 s a C R P S I R C ) L O Z I R T ( t n e g a e r I R T I e s a N D s e i g o l o n h c e T e f i L s d a e b c i t e n g a m G A g I / e d a r g P I H C e c r e i P s e i g o l o n h c e T e f i L h c i r d l A a m g i S n e g o r t i v n I s e i g o l o n h c e T e f i L c i f i t n e i c S r e h s i F h c i r d l A - a m g i S c i f i t n e i c S r e h s i F B E N B E N B E N e s a t p i r c s n a r T e s r e v e R I I I ™ t p i r c S r e p u S e d i l S k c i h T d e t s o r F n O - e t i R ™ a i d e r p E ) G T P I ( e d i s o t c a l a g o i h t - D β- - l y p o r p o s I e s a r e m y l o p A N D q a T 0 2 n e e w T Q K d e t a c n u r t , 2 e s a g i L A N R 4 T r o t i b i h n I e s a N R e s a R E P U S 1 e s a g i L A N R 4 T A e s a N R K e s a t o r P Cell Rep. Author manuscript; available in PMC 2023 August 22. Ding et al. Page 31 l m t h . n e z - s s i e z / e r a w t f o s / s t c u d o r p / n e / y p o c s o r c i m m o c . s s i e z . w w w / / / : s p t t h S S I E Z n o i t i d e e u l b n o i s r e V o r p 2 N E Z m o c . r o d n a . p l e h . n o i s u f / / : s p t t h s t n e m u r t s n I d r o f x O 4 4 . 0 . 3 . 2 n o i s r e V n o i s u F m o c . t s n i x o . s i r a m i / / : s p t t h s t n e m u r t s n I d r o f x O : e r a w t f o S s i s y l a n A e g a m I y p o c s o r c i M o i . s c o d e h t d a e r . t p a d a t u c / / : s p t t h g r o . n o h t y p . w w w / / : s p t t h t e n . j e g a m i / / : s p t t h g r o . n o h t y P S I B N H N I 1 . 7 . 9 n o i s r e V s i r a m I J e g a m I 3 n o h t y P 1 . 4 n o i s r e v t p a d a t u C L m 0 5 2 – 7 8 7 8 T # t a C R E I F I T N E D I 1 – 5 2 3 1 P B # t a C 9 4 5 2 5 2 # t a C h c i r d l A - a m g i S E C R U O S E C R U O S E R r o T N E G A E R 0 0 1 - X n o t i r T c i f i t n e i c S r e h s i F n o i t u l o S X 0 2 , ) C S S ( e t a r t i C m u i d o S e n i l a S h c i r d l A - a m g i S e d y h e d l a m r o F 0 1 – 1 B H - F M S # t a C d t L s c i m o n e G , h c r a e s o i B C G L r e f f u B n o i t a z i d i r b y H H S I F A N R ® s i r a l l e t S 0 1 – 0 0 2 1 - H # t a C s e i r o t a r o b a L r o t c e V g n i t n u o M e d a f i t n A ® D L E I H S A T C E V I P A D h t i w m u i d e M s y a s s a l a i c r e m m o c l a c i t i r C 9 2 0 0 – 0 1 9 5 # t a C 1 6 5 1 M A # t a C 6 0 9 4 2 0 0 2 # t a C 1 4 6 0 5 4 # t a C s e i g o l o n h c e T e f i L t i k n o i t a l o s I i A N R m ™ a n a V r i m s e i g o l o n h c e T e f i L t i K ™ g n i n o l C A T ™ O P O T . c n I , a n i m u l l I ) s e l c y C 5 7 ( 5 . 2 v t i K t u p t u O h g i H 0 5 5 / 0 0 5 q e S t x e N e r a C a r e S t i K g n i t t o l B t n e c s e n i m u l i m e h C P A ™ r o t c e t e D L P K a t a d d e t i s o p e D ) 6 0 8 4 7 8 A N J R P ( I B C N y d u t s s i h T a t a d g n i c n e u q e s A N R l l a m S 1 S e l b a T y d u t s s i h T 2 S e l b a T 3 S e l b a T 4 S e l b a T y d u t s s i h T y d u t s s i h T y d u t s s i h T s n i a r t s / s m s i n a g r O : s l e d o m l a t n e m i r e p x E s e b o r p H S I F A N R f o t s i L s e b o r p H S I F A N D f o t s i L s e d i t o e l c u n o g i l O f o t s i L s m h t i r o g l a d n a e r a w t f o S s n i a r t s s n a g e l e . C s e d i t o e l c u n o g i l O Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 32 l m t h . 2 q e S E D / l m t h / c o i b / 0 . 3 / s e g a k c a p / p j . n e k i r . r o t c u d n o c o i b / / : s p t t h / p o t k s e d - o i d u t s r / d a o l n w o d / o c . t i s o p / / : s p t t h / R E S S A T - I / g r o . p u o r g g n a h z / / : s p t t h g r o . l o m y p / / : s p t t h 2 e i t w o b / t e n . e g r o f e c r u o s . o i b - e i t w o b / / : s p t t h R E I F I T N E D I g r o . t c e j o r p - r . w w w / / : s p t t h 2 4 . l a t e d a e m g n a L E C R U O S g r o . t c e j o r p - R e r a w t f o S t i s o P r o t c u d n o c o i B 3 3 . l a t e g n a Y r e g n i d o r h c S E C R U O S E R r o T N E G A E R 4 5 5 d l i u B n o i s r e V o i d u t s R 0 . 6 2 . 1 n o i s r e v 2 q e S E D 5 . 4 . 2 n o i s r e V 2 e i t w o B 1 . 2 . 4 n o i s r e V R R E S S A T - I s c i h p a r G r a l u c e l o M ) M T ( L O M y P 0 . 5 . 2 n o i s r e V m e t s y S Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t
null
10.1088_1402-4896_ad0b51.pdf
Data availability statement The data cannot be made publicly available upon publication because they are not available in a format that is sufficiently accessible or reusable by other researchers. The data that support the findings of this study are available upon reasonable request from the authors.
Data availability statement The data cannot be made publicly available upon publication because they are not available in a format that is sufficiently accessible or reusable by other researchers. The data that support the findings of this study are available upon reasonable request from the authors.
RECEIVED 22 September 2023 REVISED 5 November 2023 ACCEPTED FOR PUBLICATION 9 November 2023 PUBLISHED 23 November 2023 Phys. Scr. 98 (2023) 125518 https://doi.org/10.1088/1402-4896/ad0b51 PAPER Efficient infrared nine-channel reflective polarization-dependent splitter , Bo Wang∗ Guoyu Liang School of Physics and Optoelectronic Engineering, Guangdong University of Technology, Guangzhou 510006, People’s Republic of China ∗ Author to whom any correspondence should be addressed. E-mail: wangb_wsx@yeah.net and Yuqing Xu Keywords: polarization sensitivity, addition-shaped grating, simulated annealing algorithm, nine-channel splitter. Abstract In order to meet the requirements of a multi-beam splitter in optical communication systems, an efficient infrared nine-channel reflective polarization-dependent beam splitter based on an addition- shaped ridge structure is proposed. All structural parameters of this polarization-dependent beam splitter are derived from the rigorous coupled-wave analysis. Upon the vertical entry of a plane wave with a wavelength of 800 nm into the grating, for transverse magnetic polarization, the diffraction efficiencies are 10.66%, 10.69%, 10.69%, 10.65%, and 10.67% at 0th, ±1st, ±2nd, ±3rd and ±4th orders, respectively. For transverse electric polarization, the diffraction efficiencies of the 0th, ±1st, ±2nd, ±3rd and ±4th orders are 10.79%, 10.86%, 10.88%, 10.84%, and 10.86%, respectively. In addition, the tolerance analysis in this paper reveals the practicality and efficiency of this beam splitter. Therefore, the addition-shaped ridge structure has a good performance of uniformity and broad application prospects in nine-channel reflective applications. 1. Introduction Beam splitter have a crucial role in various optical systems that can be widely used in the design of optical components [1–4], such as sensors [5, 6], interferometers [7–9], photonic crystal [10–14], and optical communication [15–19], etc. Reflective beam splitter is an important type of beam splitter [20–22]. Over the past few years, in the production and application field, multi-channel reflective beam splitters have gained significant attention from researchers due to the advancements in ultra-precision optical devices [23–28]. Lin et al developed a polarization-sensitive terahertz reflective multi-channel beam splitter [29]. The splitter utilizes resonance-domain diffraction gratings with periods similar to the incident wavelength, enabling effective multi- channel beam separation. Zhou et al introduced a two-dimensional (2D) reflective grating with remarkable polarization selectivity [30]. Through rigorous coupled wave analysis, they determined that the grating exhibited high sensitivity to the polarization states of incoming light. Consequently, this versatile 2D grating holds considerable potential for optical communication applications. Jin et al proposed a reflective polarizing beam splitter grating, which utilizes a multilayer metal-dielectric structure as its foundation [31]. The grating is optimized by the simplified simulated annealing method and has a high extinction ratio and high diffraction efficiency. Huang et al analyzed an analysis on a nanodisk array-based multi-port two-dimensional (2D) reflective grating, which demonstrates the capability for high-efficiency optical control at communication wavelengths, specifically for four-port and five-port configurations [32]. Although there are many multi- channel reflective grating studies, the nine-channel reflection polarization-dependent beam splitter is rare. In this paper, an addition-shaped ridge reflective polarization-dependent beam splitter is proposed. With a wavelength of 800 nm, through the optimization of rigorous coupled-wave analysis and simulated annealing algorithm, the polarization-dependent beam splitter with nine channels of uniform and efficient output can be obtained, and the total diffraction efficiency can reach more than 96%. And the reflective efficiency of each order under TM polarization is 10.66%, 10.69%, 10.69%, 10.65%, and 10.67%, respectively. Meanwhile, the paper also makes the tolerance analysis on the duty cycle, grating period, and grating ridge thickness of this © 2023 IOP Publishing Ltd Phys. Scr. 98 (2023) 125518 G Liang et al Figure 1. The (a) two-dimensional and (b) three-dimensional schematic diagram of a nine-channel reflective polarization-dependent beam splitter at normal incidence. polarization-dependent beam splitter. Through the analysis results, it can be concluded that each structural parameter of the grating has good manufacturing tolerance, so the structure can be well used in industrial applications. Finally, to verify the accuracy of the calculation results of the rigorous coupled-wave analysis and the simulated annealing algorithm, the paper also uses the finite element method (FEM) to solve the reflection efficiency of the polarization-dependent beam splitter at each order to enhance the credibility of the data in this paper. According to the performance of polarization-dependent optical devices, precision instruments such as interferometers [33, 34], and fiber optic sensors [35, 36] can be designed. 2. Structure of addition-shaped reflective beam splitter The two-dimensional and three-dimensional images of the polarization-dependent beam splitter with a cross- shaped grating ridge are shown in figures 1(a) and (b). As figure 1(a) shows, the grating ridge of the polarization- dependent beam splitter is designed to be an addition-shaped structure, which means that the duty cycle f1 of the first layer is the same as the third layer’s duty cycle f3. Meanwhile, f1 and f3 are all smaller than the second layer’s duty cycle f2. In addition, below the grating ridge is a reflective layer of Ag and the substrate is fused silica. In the z-axis direction, from top to bottom, the addition-shaped grating ridge is defined as the first layer, second layer, and third layer. Therefore, the thickness of the grating ridge is characterized as h1, h2, and h3. The thickness of the 2 Phys. Scr. 98 (2023) 125518 G Liang et al Figure 2. The schematic diagram of the etching region. Table 1. Parameters of nine-channel polarization-dependent beam splitter after optimization. TE polarization h1(nm) h2(nm) h3(nm) hm(nm) 753 1694 1049 100 TM polarization h1(nm) h2(nm) h3(nm) hm(nm) 1349 1213 1613 100 f1 0.4 f1 0.4 f2 0.6 f2 0.6 f3 0.4 f3 0.4 d(nm) 3743 d(nm) 3344 metallic mirror silver is defined as hm. The d in figure 1(a) is the grating period of the polarization-dependent beam splitter and the width of each layer of the grating ridge is defined as d1, d2, and d3. Hence, the duty cycle of each layer is defined as f1, f2, and f3 respectively. The duty cycle fi is di/d and i represents the corresponding layer of grating ridge. When the wavelength is 800 nm, the refractive index n2of the grating ridge, whose substance is resin, is 1.51, the refractive index n1 of the fused silica is 1.45 and the refractive index nm of the metallic mirror * silver is 0.469–9.32 i. Besides, the medium of the grating groove is air and the refractive index n0 is 1.00. According to [37, 38], the cost-effective and time-efficient fabrication of the addition-shaped structure mentioned in the study can be achieved through dry etching [39, 40] and HF etching [41, 42] methods. A layer of SiO2 is initially applied to the substrate, followed by the uniform deposit of resin and Ag. Under the protection of the photoresist, excess silica in region I of figure 2 can be removed by dry etching and HF etching. Then, a cross- shaped resin grating ridge can be obtained. The objective of this study is to create an infrared reflective nine-channel beam splitter that possesses an addition-shaped structure, exhibits high total efficiency, and demonstrates moderate efficiency uniformity. Therefore, the polarization-dependent beam splitter is optimized by rigorous coupled-wave analysis (RCWA) and simulated annealing algorithm (SAA). The RCWA is utilized in this study to solve Maxwell’s equations specifically for the grating layer. This enables the determination of the diffractive efficiency for each order. For a more comprehensive understanding of the calculation procedure, readers are recommended to consult [43, 44] for detailed information. These references provide a thorough explanation of the step-by-step calculations involved in the procedure. The SAA is a method of filtering data, which is used to screen the diffraction efficiency of each group calculated by RCWA to find the most uniform and efficient. The detailed optimization algorithm process of SAA is mentioned in [45]. The cost function for a beam splitter with nine ports can be expressed as follows: ( 4 =- i where the term Ii symbolizes the diffraction efficiency of the ith order, and Iav denotes the average efficiency for the nine-port beam splitter, given by: F = å av ) I I I , 4 i i - ( ) 1 4 =- 4 i å 2 I av 4 1 ⎛ å= ⎜ 9 ⎝ =- 4 i I i . ⎞ ⎟ ⎠ 3 ( ) 2 Phys. Scr. 98 (2023) 125518 G Liang et al Figure 3. The relationship between the efficiency of each order for TM polarization and the thickness: (a) the thickness of h1, (b) the thickness of h2, (c) the thickness of h3. 4 Phys. Scr. 98 (2023) 125518 G Liang et al Figure 4. The relationship between the efficiency of each order for TE polarization and the thickness: (a) the thickness of h1, (b) the thickness of h2, (c) the thickness of h3. The calculation of Ii is performed using the RCWA. The optimization parameters for the grating structure are acquired by achieving the minimum value of F. In addition, to obtain a nine-channel beam splitter, the grating period d and working wavelength λ should meet the following criteria, which have been pointed out 5 Phys. Scr. 98 (2023) 125518 G Liang et al Figure 5. Normalized field magnitude distribution of the nine-port beam splitter at normal incidence for (a) TE polarization and (b) TM polarization. in [46]: Meanwhile, the uniformity of beam splitters is a crucial feature. The uniformity of the nine-channel beam splitter can be defined by the following formula: l 4  d l 5 . U = h h max max - + h h min min ´ 100%, ( ) 3 ( ) 4 where hmax and hmin is the maximum and the minimum diffraction efficiency of each order for the beam splitter. 6 Phys. Scr. 98 (2023) 125518 G Liang et al Figure 6. Efficiencies in orders for the grating versus the duty cycle of the first layer f1 and the duty cycle of the second layer f2 for TM polarizations: (a) the 0th order for TM polarization, (b) the 1st order for TM polarization, (c) the 2nd order for TM polarization, (d) the 3rd order for TM polarization, (e) the 4th order for TM polarization. For normal incidence at the wavelength of 800 nm, the optimal structural parameters can be obtained by RCWA and SAA, which can be seen clearly in table 1. Before optimizing the beam splitter, the thickness of the Ag layer hm is set to 100 nm, which is enough for reflection [47]. Based on the conditions of these grating structure parameters, a nine-channel polarization-dependent beam 7 Phys. Scr. 98 (2023) 125518 G Liang et al Figure 6. (Continued.) Table 2. Diffraction efficiency of nine-channel reflection beam splitter through RCWA and FEM. 0th −1st/+1st −2nd/+2nd −3rd/+3rd −4th/+4th Total Uniformity TE(RCWA) TE(FEM) TM(RCWA) TM(FEM) 10.79% 10.95% 10.66% 10.63% 10.86% 10.64% 10.69% 10.65% 10.88% 11.08% 10.69% 10.62% 10.84% 10.69% 10.65% 10.73% 10.86% 10.87% 10.67% 10.48% 97.67% 97.51% 96.06% 95.75% 0.42% 2.02% 0.19% 1.18% splitter can be obtained, and the reflective efficiency of each order is shown in table 2. At the same time, the results of RCWA and SAA are also verified by the finite element method (FEM), and the reflective efficiency is shown in table 2. To verify the practical application prospects of polarization-dependent beam splitters, further data analysis is mentioned. Due to the symmetry of the transmission efficiency at all levels of the grating, in the following discussion, only one side of the transmission efficiency will be discussed, that is, only the transmission efficiency at the order of 0th, 1st, 2nd, 3rd, and 4th will be discussed. As shown in figure 3, it can be observed that when h1 varies in the range of 1345–1365 nm, h2 varies in the range of 1205–1228 nm, and h3 varies in the range of 1608–1624 nm, each order of TM polarization exhibits an efficiency of over 9.5%. For TE polarization, the relationship between reflection efficiency and thickness is shown in figure 4. When h1 is between 744 nm and 760 nm, h2 is between 1688 nm and 1700 nm, and h3 is between 1043 nm and 1055 nm, the reflection efficiency of each order is greater than 9.5%. Finally, through the RCWA and the SAA, it is found that when h1, h2, and h3 are equal to the values in table 1, an efficient and uniformly output reflective nine-channel polarization- dependent beam splitter is obtained. 8 Phys. Scr. 98 (2023) 125518 G Liang et al Figure 7. Efficiencies in orders for the grating versus the duty cycle of the first layer f1 and the duty cycle of the second layer f2 for TE polarizations: (a) the 0th order for TE polarization, (b) the 1st order for TE polarization, (c) the 2nd order for TE polarization, (d) the 3rd order for TE polarization, (e) the 4th order for TE polarization. 9 Phys. Scr. 98 (2023) 125518 G Liang et al Figure 7. (Continued.) Figure 8. Efficiency versus grating period under TM polarization with λ = 800 nm, f1 = 0.4, f2 = 0.6, f3 = 0.4, h1 = 1349 nm, h2 = 1213 nm and h3 = 1613 nm. 10 Phys. Scr. 98 (2023) 125518 G Liang et al Figure 9. Efficiency versus grating period under TE polarization with λ = 800 nm, f1 = 0.4, f2 = 0.6, f3 = 0.4, h1 = 753 nm, h2 = 1694 nm and h3 = 1049 nm. Figure 10. Efficiency versus operating band under TM polarization with d = 3344 nm, f1 = 0.4, f2 = 0.6, f3 = 0.4, h1 = 1349 nm, h2 = 1213 nm and h3 = 1613 nm. 3. Analysis and discussions To have a clearer understanding of the working state of the beam splitter at the wavelength of 800 nm, figure 5 shows the normalized electric field diagram at vertical incidence in this band. From the energy distribution inside the grating, it can be seen that most of the energy under TM polarization is mainly distributed at the centerline of the grating ridge. Moreover, the energy of the grating ridge is symmetrically distributed. For TE polarization, the energy of the grating ridge is symmetrically distributed, and the energy of the grating ridge is evenly distributed in each layer. If the beam splitter is to be applied in practical industries, attention should be paid to its manufacturing tolerances. When the duty cycle of the grating ridge changes, it will have a significant impact on the output efficiency of each port. Because the grating ridge is cross-shaped and the duty cycle of the first and third layers is the same, only the influence of duty cycles f1 and f2 is considered in the following tolerance analysis. As shown in figure 6, the relationship between the duty cycle of the grating ridge and the reflection efficiency of each order is 11 Phys. Scr. 98 (2023) 125518 G Liang et al Figure 11. Efficiency versus operating band under TE polarization with d = 3743 nm, f1 = 0.4, f2 = 0.6, f3 = 0.4, h1 = 753 nm, h2 = 1694 nm and h3 = 1049 nm. determined when all other structural parameters are optimal. Combining simulation data and figures 6(a)–(e), the following results can be obtained. Due to the symmetrical output efficiency of the grating, the following discussion will only focus on the positive order diffraction order. Under the vertical incidence of 800nm wavelength, the 0th order, 1st order, 2nd order, 3rd order, and 4th order reflection efficiency under the TM polarization is greater than 9.0%, when the duty cycle is 0.385 < f1 < 0.403, 0.595 < f2 < 0.610. In summary, when the reflection efficiency of each port remains the highest and uniform, f1 = f3 = 0.4, f2 = 0.6. For TE polarization, the impact of the change in the duty cycle is shown in figures 7(a)–(e). When f1 is in the range of 0.397–0.406 and f2 is in the range of 0.596–0.609, the reflection efficiency of each order is greater than 9%. In addition, the diffraction efficiency at all levels of the beam splitter is not only related to the width and thickness of the grating ridge but also to the grating period. Therefore, it is necessary to analyze the manufacturing tolerance of the grating period. As shown in figure 8, for TM polarization, when the grating period d changes in the range of 3325–3354 nm, the reflection efficiency of each order is above 9.0%. Meanwhile, for TE polarization, the reflection efficiencies of the 0th, 1st, 2nd, 3rd, and 4th orders are above 9.0% when the grating period d varies between 3732–3754 nm. Moreover, the working wavelength of the polarization-dependent beam splitter is 800 nm. However, in practical applications, the working wavelength may change. So, the paper analyzed the tolerance of working wavelength. Combining figure 10 with the optimized data, when the working wavelength is between 796–804 nm, the reflection efficiency of each level for TM polarization is above 9.0%. From figure 11, it can be observed that when the working wavelength λ is within the range of 798 to 801 nm, the reflection efficiency of each level is greater than 9%. 4. Conclusions A polarization-dependent reflection beam splitter of addition-shaped for equal nine-channel generated is proposed in this paper. The optimization parameters of the grating are obtained through RCWA and SAA. Under the optimization parameters in the second section, a nine-channel output for TM polarization with uniformity of 0.19% and a total diffraction efficiency of over 96% can be obtained. For TE polarization, the uniformity of the nine-channel beam splitter is less than 0.5%, and the total diffraction efficiency reaches over 97%. At the same time, the maximum difference between the diffraction efficiency of each order calculated by the finite element method and the results calculated by RCWA is less than 0.3%, which enhances the credibility of the article’s data. The analysis of the manufacturing tolerance and normalized energy distribution of the beam splitter in the y = 0 plane in the third section can guide the future manufacturing of nine-channel gratings with similar structures. It is believed that the proposed polarization-dependent reflection beam splitter has potential application value in many fields of optical application. 12 Phys. Scr. 98 (2023) 125518 Acknowledgments G Liang et al This work is supported by the Science and Technology Program of Guangzhou (202002030284). Data availability statement The data cannot be made publicly available upon publication because they are not available in a format that is sufficiently accessible or reusable by other researchers. The data that support the findings of this study are available upon reasonable request from the authors. ORCID iDs Guoyu Liang Bo Wang Yuqing Xu https://orcid.org/0009-0001-5868-210X https://orcid.org/0000-0002-0927-699X https://orcid.org/0009-0005-8784-8315 References [1] Torcal-Milla F J and Sanchez-Brea L M 2022 Optik 271 170119 [2] Ko M C, Kim N C, Choe H, Ri S R, Ryom J S and Kim S G 2021 Plasmonics 16 1991–8 [3] Radhakrishnan S, Raja G T and Kumar D S 2021 Plasmonics 16 493–500 [4] Chen P, Chen C, Xi J, Du X, Liang L, Mi J and Shi J 2021 Plasmonics 17 43–9 [5] Gao P, Liu Y, Zheng X and Wang Z 2023 Opt. Commun. 526 128942 [6] Wu J, Yang D, Huang X, Li Y and Xia Y 2021 Phys. Lett. A 389 127080 [7] Bera J, Halder B, Ghosh S, Lee R K and Roy U 2022 Phys. Lett. A 453 128484 [8] Olivares S 2021 Phys. Lett. A 418 127720 [9] Sengupta C and Bhattacharya K 2023 Phys. Scr. 98 045015 [10] Chen D, Wang Z, Zeng Y, Zhang X and Sun X 2022 Phys. Lett. A 432 128020 [11] Xu Y, Yuan J, Qu Y, Qiu S, Zhou X, Yan B, Wang K, Sang X and Yu C 2022 Opt. Eng. 61 057104 [12] Huang J, Xu X, Mu S, Zhang H, Liu Y and Zhai N 2022 Opt. Eng. 61 027103 [13] Hong P, Xu L, Ying C and Rahmani M 2022 Opt. Lett. 47 2326–9 [14] Yuan M, Han X, Xiao H, Nguyen T G, Boes A, Ren G, Hao Q, Xue J, Mitchell A and Tian Y 2023 Opt. Lett. 48 171–4 [15] Li X, Zhao Z, Tan J, Chen R, Wang J, Yu F, Chen J, Li G, Li Z and Chen X 2023 Plasmonics 18 485–92 [16] Dizaj L S, Abbasian K and Nurmohammadi T 2020 Plasmonics 15 2213–21 [17] Yu Z, Liu D, Cheng L, Meng Z, Zhao Z, Yuan X and Xu K 2022 Opt. Express 30 46822–37 [18] Shan M, Jin Q, Zhong Z and Lin L 2023 Phys. Scr. 98 045102 [19] Mai V and Kim H 2023 Opt. Commun. 527 128963 [20] Zhang H, Yang J and Zhang H 2020 Plasmonics 15 957–66 [21] Shen C, Huang Z, Chen X, Chen H, Wang Z, Li Y, Liu J and Zhou J 2022 J. Lightwave Technol. 40 6296–302 [22] Wang Y, Liu Y, Wu T, Li J, Sun Y and Ye H 2021 Opt. Eng. 60 115108 [23] Gao C and Wang B 2020 Phys. Scr. 95 085501 [24] Zhou Y, Li X, Wang B and Li L 2022 Optik 259 169014 [25] Xiong Z and Wang B 2023 Opt. Laser Technol. 168 109959 [26] Li J, Sun Y, Fan H, Wang X, Ye H and Liu Y 2023 Results Phys. 44 106181 [27] Huang Y and Wang B 2022 Opt. Eng. 61 035102 [28] Li J 2021 Laser Phys. 31 026203 [29] Lin Z, Wang B, Xing X, Zhang F, Xue J and Zhou J 2021 Results Phys. 28 104702 [30] Zhou B, Jia W, Xiang C, Xie Y, Zhang S, Wang J and Zhou C 2023 Opt. Laser Technol. 163 109332 [31] Jin G, Zhou C, Jia W, Xie Y, Zhou B and Wang J 2023 Optik 280 170789 [32] Huang Z and Wang B 2022 Opt. Laser Technol. 152 108102 [33] Shen C, Zhong C, You Y, Chu J, Zou X, Dong X, Jin Y, Wang J and Gong H 2012 Opt. Express 20 15406–17 [34] Tugolukov M, Levin Y and Vyatchanin S 2018 Phys. Lett. A 382 2181–5 [35] Wang M, Zhang N, Huang X, Yin B, Mu H, Han M and Chen D 2020 Opt. Lett. 45 4519–22 [36] Liu H, Zhu C, Wang Y, Tan C, Li H and Cheng D 2017 Opt. Eng. 56 057112 [37] Wang B, Zhou C, Feng J, Ru H and Zheng J 2008 Appl. Opt. 47 4004–8 [38] Wang S, Zhou C, Zhang Y and Ru H 2006 Appl. Opt. 45 2567–71 [39] Wilkinson C D W and Rahman M 2004 Phys. Eng. Sci. 362 125–38 [40] Morikawa K, Chen P, Tran H, Kazoe Y, Chen C and Kitamori T 2023 J. Micromech. Microeng. 33 047001 [41] Tamayo A, Rubio J, Rubio F and Rodriguez M A 2021 Materials 14 3276 [42] Zhao S and Liu H 2015 Nanotechnology 26 015301 [43] Moharam M G and Gaylord T K 1981 Appl. Opt. 20 240–4 [44] Moharam M G and Gaylord T K 1981 J. Opt. Soc. Am. 71 811–8 [45] Wu J, Zhou C, Cao H, Hu A, Yu J, Sun W and Jia W 2011 J. Opt. 13 115703 [46] Zheng J, Zhou C, Wang B and Feng J 2008 J. Opt. Soc. Am. A 25 1075–83 [47] Gao C, Wang B, Wen K, Meng Z, Nie Z, Xing X, Chen L, Lei L and Zhou J 2019 Opt. Commun. 452 395–9 13
null
10.1103_physrevd.107.035007.pdf
null
null
PHYSICAL REVIEW D 107, 035007 (2023) Neutrino nonstandard interactions with arbitrary couplings to u and d quarks New York University Abu Dhabi, P.O. Box 129188, Saadiyat Island, Abu Dhabi, United Arab Emirates Nicolás Bernal * Yasaman Farzan † School of Physics, Institute for Research in Fundamental Sciences (IPM), P.O. Box 19395-5531, Tehran, Iran (Received 5 December 2022; accepted 19 January 2023; published 7 February 2023) We introduce a model for nonstandard neutral current interaction (NSI) between neutrinos and the matter fields, with an arbitrary coupling to the up and down quarks. The model is based on a new Uð1Þ gauge symmetry with a light gauge boson that mixes with the photon. We show that the couplings to the u and d quarks can have a ratio such that the contribution from NSI to the coherent elastic neutrino-nucleus scattering (CEνNS) amplitude vanishes, relaxing the bound on the NSI from the CEνNS experiments. Additionally, the deviation of the measured value of the anomalous magnetic dipole moment of the muon from the standard-model prediction can be fitted. The most limiting constraints on our model come from the search for the decay of the new gauge boson to e−eþ and invisible particles, carried out by NA48=2 and NA64, respectively. We show that these bounds can be relaxed by opening up the decay of the new gauge boson to new light scalars that eventually decay into the e−eþ pairs. We show that there are ranges that can lead to both a solution to the ðg − 2Þμ anomaly and values of ϵμμ ¼ ϵττ large enough to be probed by future solar neutrino experiments. DOI: 10.1103/PhysRevD.107.035007 I. INTRODUCTION Within the Standard Model (SM) of the elementary particles, apart from gravity, the only interaction that neutrinos have is through the weak coupling. With the ever-increasing sensitivity of neutrino experiments, it is timely to ask whether there are any new subdominant interactions between neutrinos and matter fields. In recent years, a remarkable number of studies have been carried out on the impact of neutral current nonstandard interaction (NSI) on neutrino propagation in matter. The neutral current NSI can be parametrized as a four-fermion inter- action ffiffiffi p 2 2 GFεf αβ (cid:3) ¯ναγμ 1 − γ5 2 (cid:4) ð ¯fγμð1 þ κγ5ÞfÞ; νβ ð1Þ where f ∈ fu; d; eg. εf quantify the strength of the NSI, and the limit εf αβ are dimensionless parameters that αβ ¼ 0 *nicolas.bernal@nyu.edu † yasaman@theory.ipm.ac.ir Published by the American Physical Society under the terms of license. the Creative Commons Attribution 4.0 International Further distribution of this work must maintain attribution to the author(s) and the published article’s title, journal citation, and DOI. Funded by SCOAP3. corresponds to the standard coupling. In the case where jεf αβj ∼ 1, NSI becomes as strong as the weak interaction. It is straightforward to show that the axial part of NSI (i.e., the one proportional to κ) cannot induce matter effects for the propagation of neutrinos in an unpolarized medium such as Earth or the Sun. Moreover, the coherent elastic neutrino- scattering (CEνNS) experiments are mainly nucleus sensitive to the vector part of the NSI. However, the measurement of total solar neutrino flux by the Sudbury Neutrino Observatory (SNO) was sensitive only to the axial NSI with the quarks. That is, the SNO measurement of the Gamow-Teller process ν þ D → ν þ n þ p can constrain the products κ × εu αβ. Measurements of solar neutrino scattering off electrons can constrain both εe αβ by studying the dependence of the scattering cross section of the electron recoil energy. In this paper, we focus on model building for vectorlike NSI, so we fix κ ¼ 0. αβ rather than εu αβ and κ × εd αβ and κεe αβ and εd The effective Lagrangian shown in Eq. (1) can be obtained by integrating out a heavy UNEWð1Þ gauge boson that couples both to neutrinos and to matter fields. This idea has been pursued in several studies; see, e.g., Ref. [1]. Concerning the propagation of neutrinos in matter, only forward scattering with vanishing energy-momentum trans- fer (q2 → 0) is relevant, so that here, again, one can integrate out the mediator and use the effective action in 2470-0010=2023=107(3)=035007(10) 035007-1 Published by the American Physical Society NICOLÁS BERNAL and YASAMAN FARZAN PHYS. REV. D 107, 035007 (2023) Eq. (1), even if the energy of the neutrino beam in the rest frame of the medium is much larger than the media- tor mass.1 p ττ − εf μμ − εf ee ≃ εf In the presence of NSI, new degeneracies appear in the neutrino oscillation parameters. For example, the so-called generalized mass ordering degeneracy appears [3–7], which leads to an alternative solution to the solar neutrino anomaly known as the large mixing angle (LMA)-dark solution with θ12 > 45° and εf ee ∼ 1. As pointed out in Ref. [8], if we want to test the LMA-dark solution via only oscillation experiments, different media with different proton-to-neutron compositions are required. Furthermore, NSI with jεfj ≳ 0.1 can be tested, in princi- ple, in scattering experiments. There are, however, a few exceptions: (i) In scattering experiments, if the mediator mass mZ0 is smaller than the typical energy-momentum ffiffiffiffiffiffiffiffi jq2j transfer ( ), we cannot use the four-Fermi analysis and we should employ the whole propagator of the mediator Z0 − q2Þ that gives an amplitude proportional than to g2 Z0=m2 rather Z0, and hence a suppression of m2 Z0 − q2Þ. (ii) With a given target at CEνNS experi- ments, the contributions of NSI to the amplitudes of the scattering off the neutrons and protons of the target cancel out each other [9,10]. Motivated by this phenomenological consideration, we build a model for NSI with an arbitrary ratio of NSI couplings to the u and d quarks. The scenario is based on a flavor gauge model with a light gauge boson Z0, which mixes with the photon. We enumerate the relevant bounds on the parameters of the model. for certain ratios of εu Z0=ðm2 Z0=ðm2 αβ=εd αβ to g2 We focus on the allowed range of the parameter space that can (i) explain the ðg − 2Þμ anomaly [11,12], (ii) lead to large NSI, and (iii) yield ratios of εu=εd for which CEνNS bounds can be relaxed [9]. In our model, as in the case of B − L, the new gauge boson couples to electrons and neutrinos, so it can appear in the NA64 experiment as a missing energy on which there are strong bounds [13,14]. We discuss how the model can be augmented to suppress the invisible decay modes of Z0 and, therefore, open the parameter space to accommodate the solution to ðg − 2Þμ and a large NSI. The paper is organized as follows. In Sec. II, the model is presented. It is also shown how to augment the model to suppress the branching ratios of Z0 → e−eþ and Z0 → invisible in order to avoid the bounds from searches for these decay modes. In Sec. III, various observables that can test the model are discussed, and the relevant bounds are reviewed. Figures displaying the bounds on the param- eter space of our model are presented. The results are summarized in Sec. IV. II. THE MODEL We will augment the SM gauge group with a new local UNEWð1Þ to obtain NSI. We show the lepton and baryon numbers of the three generations with Lα and Bi, respec- tively. For any arbitrary real value of c, the combination of lepton and baryon numbers Lμ þ Lτ − cðB1 þ B2Þ − 2B3ð1 − cÞ ð2Þ is anomaly-free. The gauge boson of the UNEWð1Þ sym- metry is denoted by Z0, with a gauge coupling gZ0. Unless c ¼ 2=3, the UNEWð1Þ charges of the third generation of quarks are different from those of the first and second generations. As a result, on the quark-mass basis, left- handed down quarks can obtain a flavor-violating coupling to Z0. This feature has been invoked in Ref. [15] to address the so-called b anomalies observed at the LHCb.2 We shall comment on whether, in the range of parameters of our interest, the deviation of b → sμþμ− from the SM pre- diction is within the observed range or not. We have taken equal charges for the first and second generations of the quarks to respect the bounds from the neutral-kaon mixing. In the lepton sector, the charged lepton mass basis and the electroweak basis coincide, so we shall not have lepton flavor-violating coupling for the charged leptons, but we can have off-diagonal couplings in the neutrino mass basis μ ¯νiγμνj, where U is the Pontecorvo– like gZ0ðδij − UeiU(cid:2) Maki–Nakagawa–Sakata mixing matrix. This can lead to three-body decay of the heavy neutrinos to lighter ones, but with lifetimes much larger than the age of the Universe, an effect irrelevant for phenomenological purposes. Notice that we have set the new UNEWð1Þ charge of the first generation of leptons equal to zero. As a result, the strong limits of GEMMA on ¯νe þ e scattering can be relaxed [18,19]. The gauge symmetry in Eq. (2) induces equal couplings to the u and d quarks. We break this universality by introducing a kinetic mixing between Z0 and the photon parametrized by ϵ. The couplings of quarks to Z0 can then be written as ejÞZ0 (cid:5)(cid:3) − c 2 3 eϵ 3 gZ0 þ (cid:3) − c 3 gZ0 − þ (cid:4) X ¯uiγμui ui∈fu;cg (cid:4) X 1 3 eϵ di∈fd;sg (cid:6) ¯diγμdi Z0 μ and the couplings of leptons as ½ðgZ0 − eϵÞ¯μγμμ þ gZ0 ¯νμγμνμ þ ðgZ0 − eϵÞ¯τγμτ þ gZ0 ¯ντγμντ − eϵ¯eγμe(cid:3)Z0 μ; ð3Þ ð4Þ 1Indeed, as long as the mass of the mediator is larger than the inverse of the medium, we can integrate out the mediator and rely on the four-Fermi effective potential formalism [2]. the size of 2Very recently, the LHCb Collaboration reported measure- ments of the lepton flavor universality in b → slþl−, which for many years dominated the B-physics anomalies, compatible with the SM prediction [16,17]. 035007-2 NEUTRINO NONSTANDARD INTERACTIONS WITH ARBITRARY … PHYS. REV. D 107, 035007 (2023) where e and e, respectively, denote the electric charge and the electron field. As shown in the appendixes of Ref. [20], as long as there is no mass mixing between Z0 and the hypercharge boson, the kinetic mixing cannot induce an electric charge for neutrinos, on which there are extremely strong bounds [21]. Furthermore, in the absence of mass mixing in the Stückelberg mass term for the new gauge boson, the bounds from violation of atomic parity do not constrain ϵ [22]. Reference [23] has also invoked a Lμ − Lτ model with a gauge boson kinetically mixed with the photon that can explain the ðg − 2Þμ anomaly. Integrating out the Z0 boson, we can write the following effective couplings to quarks: μμ ¼ εu εu ττ ¼ ð2eϵ − cgZ0ÞgZ0 ffiffiffi GFm2 2 Z0 p 6 ; μμ ¼ εd εd ττ ¼ − ðeϵ þ cgZ0ÞgZ0 ffiffiffi p GFm2 2 6 Z0 ; ee ¼ εd εu ee ¼ 0; and to electrons ee ¼ 0; εe μμ ¼ εe εe ττ ¼ − gZ0eϵ ffiffiffi GFm2 2 Z0 p 2 : ð5Þ ð6Þ ð7Þ ð8Þ ð9Þ From Eqs. (5) and (6), one can obtain the effective couplings to neutrons and protons, εn μμ ¼ −cg2 ffiffiffi Z0 p GFm2 2 Z0 2 and εp μμ ¼ ðeϵ − cgZ0ÞgZ0 ffiffiffi p 2 2 GFm2 Z0 ; ð10Þ 2 3 Nu − 1 3 Nd − Ne ¼ 0: ð14Þ Taking Nn=Np ≃ 0.54 at the center of the Sun [24], we can translate the LMA-dark 2σ band found in Refs. [3,10] into 2 ≲ εmedium μμ ¼ εmedium ττ ≲ 3; ð15Þ which translates into gZ0 ¼ ð6.5 − 8.0Þ × 10−5 mZ0 10 MeV (cid:3) − (cid:4) 1 1=2 c : ð16Þ Of course, there is also the standard LMA solution with θ12 < π=4 that requires [3,10] −0.081 < εmedium μμ ¼ εmedium ττ < 1.422; ð17Þ which, for Nn=Np ≃ 0.54, corresponds to (cid:4) 2 (cid:3) −3 × 10−9 mZ0 10 MeV < cg2 Z0 < 1.7 × 10−10 (cid:3) mZ0 10 MeV (cid:4) 2 : ð18Þ In our model, the coupling to the muon is given by gμ ≡ gZ0 − eϵ, which can be rewritten as (cid:3) (cid:5) 1 − c ffiffiffi p GFm2 2 2 Z0 −c gμ ¼ gZ0 (cid:5) ¼ 1 − (cid:4)(cid:6) 1 tan η Ne Nn þ Np (cid:6)1=2(cid:5) (cid:3) εmedium μμ 1 − c 1 − (cid:4)(cid:6) : 1 tan η ð19Þ and their ratio tan η [9], tan η ¼ εn μμ εp μμ ¼ −cgZ0 eϵ − cgZ0 : In the limit jcj ≪ 1, gμ ≃ gZ0 as expected. To explain the ðg − 2Þμ anomaly, gμ should be in the range found in Ref. [25]. For example, if mZ0 ∼ 10 MeV, the 2σ band compatible with ðg − 2Þμ is ð11Þ In this model, the contribution from NSI to the effective form in matter takes the potential of neutrinos VNSI ¼ Diagð0; Vμ; VτÞ, with Vμ ¼ Vτ ¼ 2 ¼ 2 ffiffiffi p GFðNeεe 2 ; ffiffiffi p GFNeεmedium 2 μμ μμ þ Nuεu μμ þ Ndεd μμÞ in which εmedium μμ ¼ −cg2 ffiffiffi Z0 p GFm2 2 2 Z0 Nn þ Np Ne : ð12Þ ð13Þ Notice that we have used the fact that the medium is electrically neutral, so that gμ ¼ ð3.5–7Þ × 10−4: ð20Þ In the next section, we discuss the various bounds on the model and find the parameter range that can lead to interesting phenomenology. If ϵ does not vanish, Z0 can be produced by its coupling to electrons. Furthermore, if Z0 is lighter than 2mμ, the main Z0 decay modes are into νμ ¯νμ, ντ ¯ντ, and e−eþ. Up to corrections of order of ðme=mZ0Þ2, BrðZ0 → e−eþÞ ¼ BrðZ0 → invisibleÞ ¼ and ðeϵÞ2 ðeϵÞ2 þ g2 Z0 g2 Z0 ðeϵÞ2 þ g2 Z0 : ð21Þ 035007-3 NICOLÁS BERNAL and YASAMAN FARZAN PHYS. REV. D 107, 035007 (2023) As discussed in the next section, the NA48=4 experiment strongly constrains the scenario in which Z0 can be produced by the π0 decay with subsequent decay into a pair e−eþ. On the other hand, the NA64 experiment constrains Z0 that can be produced by its coupling to electrons and then decay invisibly. Motivated by saving the dark-photon solution to the ðg − 2Þμ anomaly, Ref. [26] suggests opening up a semivisible decay mode for Z0 to avoid these bounds. In the following, we suggest an alternative detour to these bounds by augmenting the model such that Z0 predominantly decays into a pair of intermediate scalars φ ¯φ that, in turn, decay to pairs e−eþ. We will see that this mechanism also gives mass to the Z0 boson. For φ lighter than 10 MeV decaying to a pair e−eþ, there are strong bounds from E774 and E141 [27], so we take φ to be heavier than 10 MeV. As a result, the mass of Z0 should be larger than 20 MeV. We assign a UNEWð1Þ charge cφ ≫ 1 to the φ scalars, obtaining the coupling For φ to decay into e−eþ, it should be coupled to electrons. A direct coupling would break both the UNEWð1Þ gauge symmetry and the electroweak symmetry. Therefore, we introduce a second φ0 with the same charge as φ and heavier than Z0. Furthermore, we add a new inert Higgs doublet Φ with a large coupling to e−eþ via the terms λφ†φ0H · Φ þ λe ¯eRΦ†Le; ð27Þ in which Le ¼ ðνeeLÞ. Since the SM Higgs coupling to electrons is very suppressed, we need this new Φ with a relatively large Yukawa coupling λe to ensure a fast decay of φ → e−eþ with τφ ∼ 10−14 sec. With such a short lifetime, the bounds from E177 can also be relaxed [28] because decays occur before φ or Z0 reach the detector. The vacuum expectation value of φ0 breaks the UNEWð1Þ symmetry and gives mass to the Z0 boson, mZ0 ¼ cφgZ0hφ0i: ð28Þ cφgZ0Z0 μ½iðφ(cid:2)∂μφÞ þ H:c:(cid:3); ð22Þ Furthermore, along with hHi, it leads to the mixing of φ with the neutral component of Φ given by which leads to a partial decay width ΓðZ0 → φ ¯φÞ ¼ φg2 c2 Z0mZ0 48π (cid:4) 3=2 (cid:3) 1 − 4 m2 φ m2 Z0 : ð23Þ It is important to note that, despite cφ being large, we are interested in a range of parameters where the coupling of φ to Z0, cφgZ0, is very small and well within the perturbative range. In the limit c2 Z0 ≫ e2ϵ2, the branching ratios can be rewritten as φg2 BrðZ0 → e−eþÞ ¼ BrðZ0 → invisibleÞ ¼ φg2 c2 8ðeϵÞ2 Z0ð1 − 4m2 8 φ=m2 φð1 − 4m2 c2 φ=m2 Z0Þ3=2 Z0Þ3=2 : and ð24Þ As we shall see in the next section, to avoid the NA64 bounds, a Z0 with the energy of ∼30 GeV should be able to decay to φ ¯φ before traveling a distance of ∼1 cm. Similarly, the φ produced should decay before traveling more than ∼1 cm. That is, τZ0 < 3 × 10−14 mZ0 τφ < 3 × 10−14 mφ 30 MeV 30 MeV 30 GeV EZ0 30 GeV Eφ sec and sec; ð25Þ and hence, cφgZ0 > 3.3 × 10−4 30 MeV mZ0 (cid:4) −3=4 (cid:3) 1 − 4 m2 φ m2 Z0 : ð26Þ sin β ¼ λhHihφ0i Φ0 − m2 m2 φ ð29Þ and, therefore, to an effective coupling of the form λφeφ† ¯ee with λφe ≡ λe sin β: ð30Þ should be of For τφ ∼ 10−14 sec, λφe the order of 4 × 10−4ð10 MeV=mφÞ1=2. Furthermore, Φ0 should be heavier than ∼400 GeV to avoid present bounds from direct searches at colliders. Taking λe ∼ 0.1, sin β should be of the order of 10−3. Note that such mixing is small enough not to cause an unnaturally large contribution to the φ mass: mφ ≫ mΦ sin2 β. We can then write λ ¼ 0.026 sin β 10−3 30 MeV mZ0 cφgZ0 8.5 × 10−4 m2 Φ0 ð400 GeVÞ2 : ð31Þ Finally, to allow λ to remain in the perturbative range, it is necessary that cφgZ0 < few × 10−2. III. THE BOUNDS For the values of the gZ0 coupling of interest for NSI or for ðg − 2Þμ, Z0 reaches thermal equilibrium in the early Universe with the plasma. If Z0 is lighter than ∼5 MeV, Z0 and/or its decay products can contribute significantly to the extra relativistic degrees of freedom in which there are strong bounds from Cosmic microwave background and big bang nucleosynthesis [29]. Therefore, we focus on the case where mZ0 > 5 MeV. Now, we present a compilation of the most stringent bounds relevant to the present scenario. 035007-4 NEUTRINO NONSTANDARD INTERACTIONS WITH ARBITRARY … PHYS. REV. D 107, 035007 (2023) A. Bounds from beam dump experiments, meson decays, and scattering experiments In the absence of φ, two regimes can be distinguished for mZ0 ∼ 10 MeV. (1) g2 Z0 ≫ e2ϵ2: In this case, Z0 decays mainly into νμ ¯νμ and ντ ¯ντ. Thus, Z0 would appear as missing energy in experiments such as BABAR [30] and NA64 [13,14], where Z0 can be produced by its coupling to electrons (for example, by e−eþ → γZ0 or electron bremsstrahlung). These experiments established an upper bound ϵ ≲ Oð10−5Þ for mZ0 ∼ 10 MeV. As invisible decay modes dominate over visible decay modes, the bounds of beam dump experiments on gZ0 and/or on ϵ are relaxed. The Z0 coupling to neutrinos can appear in meson decays such as Kþ → μþνZ0. Using the constraint of E949 on Kþ → μþ þ missing energy [31], an upper bound on the coupling of Z0 to νμ can be extracted [32]. With an improved constraint from NA62 on such decay modes [33], the bound for the mass range m2 Z0=m2 K can be rewritten as (cid:3) gZ0 < 0.003 mZ0 5 MeV (cid:4) : ð32Þ Moreover, from the bound on π0 → Z0γ, an upper bound of Oð10−3Þ on the coupling of Z0 to quarks is obtained [34,35]. Z0 ≪ e2ϵ2: In this case, a Z0 with mass mZ0 ∼ Oð10Þ MeV decays mainly into pairs e−eþ, relaxing the bound from NA64. Instead, the bounds from beam dump experiments apply. For mZ0 ∼ 10 MeV, the strongest upper bound on ϵ comes from the NA48=2 experiment [36]. The bound on ϵ versus mZ0 fluctuates violently between 5 × 10−4 and 10−3. the parameter space where For the time being, the ðg − 2Þμ anomaly can be fitted (that is, gZ0 ∼ 7 × 10−4) is experimentally allowed. Interest- ingly, such parameter space will be probed by future experiments such as MESA [37], VEPP-3 [38,39], and DARKLIGHT [40]. (2) g2 target. (ECAL) Opening the decay mode Z0 → φ ¯φ described at the end of the previous section, the bounds from NA64 and NA48=2 can be relaxed. In NA64, an electron beam of 100 (cid:4) 3 GeV [13] to an electromagnetic is sent calorimeter the energy deposited If within a few radiation lengths is less than 50 GeV, the signal is interpreted as e− þ nucleus → e−Z0X, with Z0 → missing energy. In our model, Z0 → φ ¯φ and φ → e−eþ within a few centimeters, so the entire energy of the initial e− entering the target at NA64 will be deposited at the ECAL within a few radiation lengths, so the NA64 bound will be relaxed. In NA48=2, the signal is e−eþγ from π0 decays, and events in which the invariant mass of the three final tracks significantly deviates from mπ0 are vetoed. is, when gZ0 ¼ 0), u þ q2 dÞ2 ¼ ð5=9Þ2e2ϵ2. and ϵ are nonzero, Thus, e−eþ from the φ decay will be vetoed. To recast the bound from BABAR [30] and NA64 [14] on the coupling of Z0 the to the electron, we should take into account expression of BrðZ0 → invisibleÞ in the present model. Similarly, the bound from NA48=2 [36] should be recasted by considering BrðZ0 → e−eþÞ in this model. In the simple the π0 kinetic mixing model (that decay rate to a photon and a dark photon is proportional to e2ϵ2ðq2 In our model, where both gZ0 it will be given by ½equðeϵqu þ cgZ0=3Þ þ eqdðeϵqd þ cgZ0=3Þ(cid:3)2. Thus, the branching ratio of a π0 decaying to a photon and a Z0 will be given by the same formula for pure kinetic mixing, replacing ϵ2 with ðϵ − cgZ0=ð5eÞÞ2. Furthermore, in our model, BrðZ0 → e−eþÞ is not one, so the bound on the square of mixing found by NA48=2 should be inter- preted as a limit on ½ϵ − cgZ0=ð5eÞ(cid:3)2 × BrðZ0 → e−eþÞ. In [14], BrðZ0 → invisibleÞjB−L ¼ BrðZ0 → ν¯νÞ=½BrðZ0 → e−eþÞ þBrðZ0 → ν¯νÞ(cid:3) ¼ 3=5. In NA64, Z0 is produced by its coupling to electrons, which for us is eϵ. As a result, the upper bound on the square of the B − L coupling found in Ref. [14] should be interpreted as an upper bound on ðeϵÞ2 × BrðZ0 → invisibleÞjours=BrðZ0 → invisibleÞjB−L. the B − L model A dedicated search using experiments such as BABAR, NA64, or NA48=2 may be able to test our model where Z0 production leads to the emission of two pairs (rather than one pair) of e−eþ. As mentioned above, the bounds from the E177, E774, and E141 beam dump experiments can be avoided in our model. Finally, we note that Z0 bosons can also be probed at the intensity and lifetime frontier experi- ments such as FASER, FASER2, DUNE, and the ILC [41]. considered in Ref. B. CEνNS experiments In our model, since the coupling of νe to Z0 is zero (i.e., εee ¼ 0), the reactor CEνNS experiments such as Dresden II [42] or CONUS do not constrain the model. However, we expect bounds from CEνNS experiments with a muon decay source such as COHERENT, as well as from direct dark matter search experiments sensitive to solar neutrinos. The cross section of the CEνNS process νμ þ nucleus → νμ þ nucleus is proportional to (cid:4) (cid:3) (cid:4) (cid:3) ; Q2 m2 m2 þ N ð33Þ μ ¼ Z n þ εn gV μμ p þ εp gV μμ m2 Z0 Z0 − t m2 Z0 Z0 − t p ¼ 1=2 − 2 sin2 θW with t being a Mandelstam variable. gV n ¼ −1=2 are the vector couplings of the standard Z and gV gauge boson to protons and neutrons, respectively. Furthermore, Z and N are the numbers of protons and neutrons in the target nucleus. With εn μμ (that is, tan η ¼ −Z=N), the NSI effect completely cancels out. these ratios are tan η ¼ −0.7 and For CsI and argon, tan η ¼ −0.8, respectively. The results of Ref. [9] confirm this argument. For our model, the allowed range of NSI can μμ ¼ −ðZ=NÞεp 035007-5 NICOLÁS BERNAL and YASAMAN FARZAN PHYS. REV. D 107, 035007 (2023) Z0=ðm2 be even larger due to the suppression jm2 Z0 − tÞj < 1. In our figures, we take the average for argon and CsI: tan η ¼ −0.75. For other target materials, this cancellation occurs at different values of tan η. For example, for silicon, tan η ¼ −Z=N ¼ −1. As a result, the change of the target material to silicon has the potential to test this degen- eracy [43]. the latter μμ ¼ −2ðZgV Note that there is also degeneracy under Qμ → −Qμ. For m2 Z0 ≫ t, transformation can take place for Zεp μμ þ Nεn p þ NgV [9] finds a fourfold degeneracy. However, for m2 Z0 ∼ jtj, the Qμ → −Qμ degeneracy (but not the εn=εp ¼ −Z=N degen- eracy) can be solved, in principle, by studying the depend- ence of the recoil energy. n Þ. As a result, Ref. C. Borexino results for the scattering of neutrinos off electrons μμ ¼ εe In our model, εe ττ ≠ 0, so the bounds from the Borexino experiment in Ref. [44] have to be taken into account. Rewriting Eq. (9) as (cid:3) (cid:4) εe μμ ¼ εe ττ ¼ εmedium μμ 1 − 1 tan η Ne Nn þ Np ; ð34Þ μμ ¼ εe it can be realized that the Borexino bound on jεe ττj < 2 [44] implies that the LMA-dark solution from tan η ¼ −0.8 to −0.7 is excluded regardless of the values of mZ0, c, and other parameters. Within our model, LMA-dark can be compatible with the Borexino bound only for tan η > 0.5 or for tan η < −37. However, a large NSI with εe ττ ∼ 1 still escapes the Borexino bound even at tan η ∼ −0.75. Such a large NSI will induce a significant deviation from the standard Mikheyev-Smirnov-Wolfenstein prediction for the low-energy part of the 8B solar neutrino spectrum, despite the vanishing contribution to CEνNS. μμ ¼ εe D. White dwarf cooling Large effective couplings between electrons and neu- trinos could lead to rapid cooling of white dwarfs [45]. As shown in Ref. [46], white dwarf cooling sets a bound 2g2 Z0ϵ 3m2 Z0 < 1.12 × 10−5 GeV−2; ð35Þ which is considerably weaker than the other relevant bounds discussed above. νμ ¯νμ; ντ ¯ντ → Z0 → νμ ¯νμ; ντ ¯ντ, E. Self-interaction of neutrinos in supernovae The gZ0 coupling could lead to resonant annihilation processes at mZ0 ∼ 30 MeV, even for gZ0 ∼ 10−5, the mean free path of neutrinos (antineutrinos) will be shorter than that of SM scattering off nucleons [47]. This consideration has been invoked in Ref. [48] to evaluate the duration of the burst such that using the simplified formula Δt ∼ R2 core=ðmean free pathÞ and to set a bound on the coupling gZ0 from the measured duration of the SN1987a neutrino burst. However, as shown in Ref. [49], when neutrinos are isotropically distributed, self-interactions cannot prolong the duration of neutrino bursts [47]. F. B physics As mentioned above, since in our model the UNEWð1Þ charges of the third generation of quarks are different from those of the first two generations, in the mass basis, the quarks obtain a flavor-changing neutral current (FCNC) coupling to Z0. After integrating out the Z0 boson, we obtain an effective coupling of the form Heff ¼ g2 Z0π 2m2 Z0 VtiV(cid:2) tj ð3c − 2Þ 3 ð ¯diγμPLdjÞð¯lγμlÞ; ð36Þ where l ∈ fμ; τg, di, dj ∈ fd; s; bg, and Vti and Vtj are the elements of the third row of the Cabibbo-Kobayashi- Maskawa matrix. Note that although we start with a nonchiral coupling of Z0 to fermions, the FCNC coupling in Eq. (36) is chiral because it originates from the quark- mass term, which mixes chiralities. That is, we have the freedom to choose a basis where the right-handed quark couplings to Z0 remain diagonal and attribute all FCNC to left-handed down quarks. Here, we use the common notation used in the literature of b anomalies [15], C9 ¼ − g2 Z0π ffiffiffi p m2 2 Z0 1 α EMGF 3c − 2 3 : ð37Þ In the case where C9 ∼ −1, the so-called b anomalies can be explained [50]. However, very recent LHCb results seem to be compatible with the SM, reducing the need for new physics [16,17]. We should note that, in our model mZ0 ≪ mb, and therefore the effective action formalism cannot be used to calculate b → sμþμ−. In fact, the contribution of our model to the amplitude of this process will be suppressed by a factor of m2 Z0 − q2Þ relative to C9, where q2 is the invariant mass of the final muon pair. In our model, Z0=ðm2 C9 × m2 Z0 Z0 − q2 ¼ −2 m2 (cid:3) gZ0 3.5 × 10−4 (cid:4) 2 GeV2 q2 3c − 2 −2 : ð38Þ Note that for gZ0 in the range that explains ðg − 2Þμ, the deviation in the low-energy bins of q2 can be significant. Taking c ¼ 2=3, the UNEWð1Þ charges of the quarks of all generations will be equal, so b → sμþμ− cancels out. The anomaly cancellation can also be fulfilled by adding more generations of fermions charged under UNEWð1Þ. Notice that in our model, the FCNC contribution to b → d is suppressed by one more order of magnitude, that is, by Vtd=Vts. 035007-6 NEUTRINO NONSTANDARD INTERACTIONS WITH ARBITRARY … PHYS. REV. D 107, 035007 (2023) FIG. 1. Bounds on the parameter space of the model for c ¼ −0.1 and tan η ¼ −0.75. The colored regions are excluded by various experiments as indicated in the legend and described in the text. The vertically dashed regions are favored by the ðg − 2Þμ anomaly. The LMA-dark solution to the solar neutrino anomaly, as well as the bounds on NSI from the neutrino oscillation data [3,10], are indicated by diagonally dashed lines. The right and left panels correspond to the variations of the model with and without φ, respectively (see Sec. II for a description). Right: we have taken cφ ¼ 40 and mφ ¼ mZ0 =3. With this ratio, for mZ0 < 30 MeV (i.e., to the left of the vertical line in the right panel), φ would be too light to avoid the bounds from E774 and E141 [27]. Figures 1–4 summarize all the relevant bounds discussed above. Figure 1 shows the bounds for c ¼ −0.1 and tan η ¼ −0.75 in the ½mZ0; gZ0(cid:3) plane. The value of tan η ¼ −0.75 is chosen because at this value the contribution of new physics to CEνNS cancels out. The colored regions show the excluded parameter ranges as follows: Borexino mea- surements of solar neutrino scattering off electrons (solid red), searches for Z0 decaying into eþe− at NA48=2 (dashed green), searches for invisible Z0 decays at NA64 (dotted blue), the upper bound mφ < 10 MeV from the combination of E774 with E141 (vertical dashed blue), and the region where the contribution of new physics to ðg − 2Þμ exceeds the observed deviation from the SM prediction (dash-dotted green). In the vertical dashed area, the ðg − 2Þμ our model provides an explanation for the diagonally dashed regions anomaly. Furthermore, FIG. 2. The same as Fig. 1, but projected in the ½εmedium solutions. The horizontal line depicts tan η ¼ −0.75 for which the contribution from the new physics to CEνNS is suppressed. ; tan η(cid:3) plane. The vertical bands correspond to the LMA and LMA-dark μμ 035007-7 NICOLÁS BERNAL and YASAMAN FARZAN PHYS. REV. D 107, 035007 (2023) FIG. 3. The same as Fig. 1, but for c ¼ 2=3. FIG. 4. The same as Fig. 2, but for c ¼ 2=3. μμ μμ < 1.422 and 2 < εmedium correspond to the LMA and LMA-dark solutions, for which −0.081 < εmedium < 3, respec- tively. The right panel of Fig. 1 shows the case with an additional scalar φ, assuming cφ ¼ 40 and mφ ¼ mZ0=3. With this value of cφ, the perturbativity limit discussed at the end of Sec. II is still satisfied. Similar information projected in the ½εmedium ; tan η(cid:3) plane is presented in Fig. 2. The vertical bands correspond to the LMA and LMA-dark solutions, and the horizontal line represents tan η ¼ −0.75, at which the contribution from new physics to CEνNS vanishes. As discussed above, for −37 ≲ tan η ≲ 0.5, the LMA-dark solution cannot be com- patible with the Borexino bound within our model, regard- less of the values of the other parameters. However, for μμ light Z0 [51,52]. As shown in Fig. 2, higher values of tan η, we can have the LMA-dark solution without conflict with other bounds. The CEνNS measure- ments will eventually test this solution with tan η > 0.5 even for for tan η > 0.5, both the LMA-dark and LMA bands have a significant overlap with the dashed area in which our model can explain the ðg − 2Þμ anomaly. Figure 2 also shows the conflict between LMA-dark and the Borexino bound for negative values of tan η. However, as seen in these figures for c ¼ −0.1 and mZ0 ∼ 40 MeV, values of εmedium ¼ εmedium ∼ 1 can be compatible with the Borexino bound ττ at tan η ∼ −0.7, with the values of gμ also explaining the ðg − 2Þμ anomaly. Without φ, the NA64 results exclude this interesting part of the parameter range but can be revived by μμ 035007-8 NEUTRINO NONSTANDARD INTERACTIONS WITH ARBITRARY … PHYS. REV. D 107, 035007 (2023) μμ introducing φ, as demonstrated in the right panels of Figs. 1 and 3. This solution to the ðg − 2Þμ anomaly can be tested by (i) searching for φ coupled to the electron in beam dump experiments, (ii) searching for the εmedium ¼ εmedium effects in the spectrum of solar neutrinos especially ττ around Eν ∼ 3 MeV to be probed by the THEIA detector, (iii) searching for new physics in b → sμþμ− with a signature enhanced in lower bins of the μþμ− invariant mass, and (iv) by a dedicated search for light Z0 producing two electron-positron pairs. From Eq. (13), we observe that, for c < 0 (c > 0), εmedium ¼ εmedium is positive (negative). The bound on the μμ ττ negative values of εmedium from the oscillation data is more stringent. For completeness, we have included Figs. 3 and 4 with a positive c: c ¼ þ2=3. At this value of c, the quarks of the three generations have the same UNEWð1Þ charges, leading to a vanishing new contribution to FCNC and therefore to b → sμþμ−. ¼ εmedium ττ μμ IV. CONCLUSIONS AND DISCUSSION large NSI In the literature, there is a class of models based on flavor gauge symmetries with a MeV-ish gauge boson that leads to nonstandard neutral current interaction between neutri- nos and quarks. By gauging the baryon number, the couplings of the u and d quarks are equal since they share the same baryon number. For relatively light Z0, the contribution to CEνNS is suppressed, so the present CEνNS for allow relatively bounds mZ0 < 30 MeV. However, these models can eventually be tested by improving the precision of the CEνNS experiments. As shown in Ref. [9], to hide NSI from CEνNS, the ratio of tan η ¼ εn=εp should have a certain value tan η ¼ −Z=N ≃ −0.75. In this paper, we have built a model that can produce NSI with arbitrary tan η. The model is based on gauging a combination of the lepton and baryon numbers of different generations with a light gauge boson Z0 that mixes with the photon. The mixing breaks the equality of the couplings of the up and down quarks because they have unequal electric charges. Within this framework, the NSI couplings are lepton flavor conserving. Since we do not gauge Le, the NSI for νe and ¯νe (that is, εee) remains zero, so the bounds from νe or ¯νe scattering (such as the ones from GEMMA [19]) can be evaded. However, μμ and εe because of the gauge boson mixing with the photon, nonstandard interactions between the muon and tau neu- trinos with the electron (i.e., εe ττ, respectively) are unavoidable. Thus, we expect an observable effect on the scattering of solar neutrinos off electrons at detectors such as Borexino. Within our model, the Borexino bound is not compatible with the LMA-dark solution for −37 < tan η < 0.5. However, we have found regions of the parameter space with tan η > 0.5 in which both LMA- dark and a solution to the ðg − 2Þμ anomaly can be this parameter space range can achieved. Interestingly, be tested by CEνNS experiments exploiting spallation neutron sources. μμ μμ ¼ εmedium ττ We have focused on regions of the parameter space for which tan η ≃ −0.75. In this range, even large values for εmedium can be hidden from CEνNS experiments. We have found that εmedium ∼ 1 and a solution to ðg − 2Þμ can be obtained simultaneously. If the invisible decay mode of Z0 dominates, the bound from NA64 rules out the tan η ∼ −0.75 range with large εmedium ∼ 1. However, it becomes viable once the decay mode Z0 → φ ¯φ → e−eþe−eþ is allowed. The light φ particles that decay into pairs e−eþ can be searched by beam dump experiments. Furthermore, εmedium ∼ 1 can be tested with future solar μμ neutrino experiments. If in the solar neutrino data evidence ∼ 1 is found without a corresponding for εmedium μμ signal at CEνNS, an interpretation would be tan η ¼ −0.7. Within our model, this also implies a distinct feature in the distribution of the invariant mass of the muon pair at b → sμþμ−, which can be tested. ¼ εmedium ττ ¼ εmedium ττ ACKNOWLEDGMENTS support N. B. received funding from the Spanish FEDER/MCIU- AEI under Grant No. FPA2017-84543-P. Y. F. has received financial from Saramadan under Contract No. ISEF/M/401439. She would like to acknowledge the support from the ICTP through the Associates Programme and from the Simons Foundation through Grant No. 284558FY19. This project has received funding and support from the European Union’s Horizon 2020 research and innovation program under the Marie Skłodowska- Curie Grant Agreement No. 860881 (H2020-MSCA- ITN-2019 HIDDeN). [1] Y. Farzan, Phys. Lett. B 748, 311 (2015); Y. Farzan and I. M. Shoemaker, J. High Energy Phys. 07 (2016) 033; Y. Farzan and J. Heeck, Phys. Rev. D 94, 053010 (2016); Y. Farzan, arXiv:1612.04971; Y. Farzan and M. Tortola, Front. Phys. 6, 10 (2018); J. Heeck, M. Lindner, W. Rodejohann, and S. Vogl, SciPost Phys. 6, 038 (2019); P. B. Denton, Y. Farzan, and I. M. Shoemaker, Phys. Rev. D 99, 035003 (2019); Y. Farzan, Phys. Lett. B 803, 135349 (2020); C. A. Argüelles, G. Barenboim, M. Bustamante, P. Coloma, P. B. Denton, I. Esteban, Y. Farzan, E. F. Martínez, D. V. Forero, A. M. Gago et al., arXiv:2203.10811; M. Hoferichter, J. Men´endez, and A. Schwenk, Phys. Rev. D 102, 074018 035007-9 NICOLÁS BERNAL and YASAMAN FARZAN PHYS. REV. D 107, 035007 (2023) (2020); P. Coloma, M. C. Gonzalez-Garcia, and M. Maltoni, J. High Energy Phys. 01 (2021) 114; A. Crivellin, M. Hoferichter, M. Kirk, C. A. Manzari, and L. Schnell, J. High Energy Phys. 10 (2021) 221. [2] A. Y. Smirnov and X. J. Xu, J. High Energy Phys. 12 (2019) 046. [3] M. C. Gonzalez-Garcia and M. Maltoni, J. High Energy Phys. 09 (2013) 152. [4] S. K. Agarwalla, P. Bagchi, D. V. Forero, and M. Tórtola, J. High Energy Phys. 07 (2015) 060. [5] P. Bakhti and Y. Farzan, J. High Energy Phys. 07 (2014) 064. [28] J. D. Bjorken, S. Ecklund, W. R. Nelson, A. Abashian, C. Church, B. Lu, L. W. Mo, T. A. Nunamaker, and P. Rassmann, Phys. Rev. D 38, 3375 (1988). [29] B. D. Fields, K. A. Olive, T. H. Yeh, and C. Young, J. Cosmol. Astropart. Phys. 03 (2020) 010; 11 (2020) E02; T. Kahniashvili, E. Clarke, J. Stepp, and A. Brandenburg, Phys. Rev. Lett. 128, 22 (2022); J. Coffey, L. Forestell, D. E. Morrissey, and G. White, J. High Energy Phys. 07 (2020) 179. [30] J. P. Lees et al. (BABAR Collaboration), Phys. Rev. Lett. 119, 131804 (2017). [31] A. V. Artamonov et al. (E949 Collaboration), Phys. Rev. D [6] P. Coloma and T. Schwetz, Phys. Rev. D 94, 055005 (2016); 94, 032012 (2016). 95, 079903(E) (2017). [7] J. Liao, D. Marfatia, and K. Whisnant, J. High Energy Phys. [32] P. Bakhti and Y. Farzan, Phys. Rev. D 95, 095008 (2017). [33] E. Cortina Gil et al. (NA62 Collaboration), Phys. Lett. B 01 (2017) 071. 816, 136259 (2021). [8] Y. Farzan and M. Tortola, Front. Phys. 6, 10 (2018). [9] M. Chaves and T. Schwetz, J. High Energy Phys. 05 (2021) [34] Y. Farzan and J. Heeck, Phys. Rev. D 94, 053010 (2016). [35] J. Altegoer et al. (NOMAD Collaboration), Phys. Lett. B 042. [10] I. Esteban, M. C. Gonzalez-Garcia, M. Maltoni, I. Martinez- Soler, and J. Salvado, J. High Energy Phys. 08 (2018) 180. [11] T. Aoyama, N. Asmussen, M. Benayoun, J. Bijnens, T. Blum, M. Bruno, I. Caprini, C. M. Carloni Calame, M. C`e, G. Colangelo et al., Phys. Rep. 887, 1 (2020). [12] B. Abi et al. (Muon g-2 Collaboration), Phys. Rev. Lett. 428, 197 (1998). [36] J. R. Batley et al. (NA48/2 Collaboration), Phys. Lett. B 746, 178 (2015). [37] T. Beranek, H. Merkel, and M. Vanderhaeghen, Phys. Rev. D 88, 015032 (2013). [38] B. Wojtsekhowski, AIP Conf. Proc. 1160, 149 (2009). [39] B. Wojtsekhowski, D. Nikolenko, and I. Rachek, 126, 141801 (2021). arXiv:1207.5089. [13] D. Banerjee, V. E. Burtsev, A. G. Chumakov, D. Cooke, P. Crivelli, E. Depero, A. V. Dermenev, S. V. Donskov, R. R. Dusaev, T. Enik et al., Phys. Rev. Lett. 123, 121801 (2019). [14] Y. M. Andreev, D. Banerjee, B. Banto-Oberhauser, J. Bernhard, P. Bisio, M. Bondi, V. Burtsev, A. Celentano, N. Charitonidis, A. G. Chumakov et al., Phys. Rev. Lett. 129, 161801 (2022). [15] A. Crivellin, G. D’Ambrosio, and J. Heeck, Phys. Rev. D 91, 075006 (2015). [16] LHCb Collaboration, arXiv:2212.09153. [17] LHCb Collaboration, arXiv:2212.09152. [18] R. Harnik, J. Kopp, and P. A. N. Machado, J. Cosmol. Astropart. Phys. 07 (2012) 026. [19] A. G. Beda, E. V. Demidova, A. S. Starostin, V. B. Brudanin, V. G. Egorov, D. V. Medvedev, M. V. Shirchenko, and T. Vylov, Phys. Part. Nucl. Lett. 7, 406 (2010). [20] D. Feldman, Z. Liu, and P. Nath, Phys. Rev. D 75, 115001 (2007). [21] M. Cadeddu, C. Giunti, K. A. Kouzakov, Y. Li, A. I. Studenikin, and Y. Zhang, Proc. Sci. EPS-HEP2019 (2020) 423 [arXiv:2001.02278]. [22] H. Davoudiasl, H. S. Lee, and W. J. Marciano, Phys. Rev. D 85, 115019 (2012). [23] A. Greljo, P. Stangl, A. E. Thomsen, and J. Zupan, J. High Energy Phys. 07 (2022) 098. [24] A. Serenelli, S. Basu, J. W. Ferguson, and M. Asplund, Astrophys. J. Lett. 705, L123 (2009). [25] D. W. P. Amaral, D. G. Cerdeno, A. Cheek, and P. Foldenauer, Eur. Phys. J. C 81, 861 (2021). [26] G. Mohlabeng, Phys. Rev. D 99, 115001 (2019). [27] A. Bross, M. Crisler, S. H. Pordes, J. Volk, S. Errede, and J. Wrbanek, Phys. Rev. Lett. 67, 2942 (1991). [40] M. Freytsis, G. Ovanesyan, and J. Thaler, J. High Energy Phys. 01 (2010) 111. [41] K. Asai, A. Das, J. Li, T. Nomura, and O. Seto, Phys. Rev. D 106, 095033 (2022). [42] P. B. Denton and J. Gehrlein, Phys. Rev. D 106, 015022 (2022). [43] D. Baxter, J. I. Collar, P. Coloma, C. E. Dahl, I. Esteban, P. Ferrario, J. J. Gomez-Cadenas, M. C. Gonzalez-Garcia, A. R. L. Kavner, C. M. Lewis et al., J. High Energy Phys. 02 (2020) 123. [44] P. Coloma, M. C. Gonzalez-Garcia, M. Maltoni, J. P. Pinheiro, and S. Urrea, J. High Energy Phys. 07 (2022) 138. [45] H. K. Dreiner, J. F. Fortin, J. Isern, and L. Ubaldi, Phys. Rev. D 88, 043517 (2013). [46] M. Bauer, P. Foldenauer, and J. Jaeckel, J. High Energy Phys. 07 (2018) 094. [47] D. G. Cerdeño, M. Cermeño, and Y. Farzan, arXiv:2301.00661. [48] A. Kamada and H. B. Yu, Phys. Rev. D 92, 113004 (2015). [49] D. A. Dicus, S. Nussinov, P. B. Pal, and V. L. Teplitz, Phys. Lett. B 218, 84 (1989). [50] M. Algueró, B. Capdevila, S. Descotes-Genon, J. Matias, and M. Novoa-Brunet, Eur. Phys. J. C 82, 326 (2022); W. Altmannshofer and P. Stangl, Eur. Phys. J. C 81, 952 (2021); M. Ciuchini, M. Fedele, E. Franco, A. Paul, L. Silvestrini, and M. Valli, arXiv:2110.10126; S. Neshatpour, T. Hurth, F. Mahmoudi, and D. Martinez Santos, Proc. Sci. EPS- HEP2021 (2022) 564 [arXiv:2112.08343]. [51] P. B. Denton, Y. Farzan, and I. M. Shoemaker, J. High Energy Phys. 07 (2018) 037. [52] V. De Romeri, O. G. Miranda, D. K. Papoulias, G. Sanchez Garcia, M. Tórtola, and J. W. F. Valle, arXiv:2211.11905. 035007-10
null
10.1371_journal.pone.0241692.pdf
Data Availability Statement: Data are contained within the paper.
Data are contained within the paper.
RESEARCH ARTICLE Understanding growth and age of red tree corals (Primnoa pacifica) in the North Pacific Ocean Emma Choy1, Kelly Watanabe1, Branwen WilliamsID Ellen Druffel4, Thomas Lorenson5, Mary Knaak1 1*, Robert Stone2, Peter Etnoyer3, 1 W.M. Keck Science Department of Claremont McKenna, Pitzer, and Scripps Colleges, Claremont, CA, United States of America, 2 Alaska Fisheries Science Center, National Marine Fisheries Service, NOAA, Juneau, AK, United States of America, 3 NOAA National Centers for Coastal Ocean Science, Charleston, SC, United States of America, 4 Department of Earth System Science, University of California Irvine, Irvine, CA, United States of America, 5 USGS Pacific Coastal and Marine Science Center, Santa Cruz, CA, United States of America * bwilliams@kecksci.claremont.edu Abstract Massive, long-lived deep-sea red tree corals (Primnoa pacifica) form a solid, layered axis comprised of calcite and gorgonin skeleton. They are abundant on the outer continental shelf and upper slope of the Northeast Pacific, providing habitat for fish and invertebrates. Yet, their large size and arborescent morphology makes them susceptible to disturbance from fishing activities. A better understanding of their growth patterns will facilitate in-situ estimates of population age structure and biomass. Here, we evaluated relationships between ages, growth rates, gross morphological characteristics, and banding patterns in 11 colonies collected from depths of ~141–335 m off the Alaskan coast. These corals ran- ged in age from 12 to 80 years old. They grew faster radially (0.33–0.74 mm year-1) and axi- ally (2.41–6.39 cm year-1) than in previously measured older colonies, suggesting that growth in P. pacifica declines slowly with age, and that basal diameter and axial height even- tually plateau. However, since coral morphology correlated with age in younger colonies (< century), we developed an in-situ age estimation technique for corals from the Northeast Pacific Ocean providing a non-invasive method for evaluating coral age without removing colonies from the population. Furthermore, we determined that annual bands provided the most accurate means for determining coral age in live-collected corals, relative to radiomet- ric dating. Taken together, this work provides insight into P. pacifica growth patterns to inform coastal managers about the demographics of this ecologically important species. With this new ability to estimate the age of red tree corals in-situ, we can readily determine the age-class structure and consequently, the maturity status of thickets, using non-invasive video survey techniques when coupled with mensuration systems such as lasers or stereo- cameras. Enhanced surveys could identify which populations are most vulnerable to distur- bance from human activities, and which should be highlighted for protection. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Choy E, Watanabe K, Williams B, Stone R, Etnoyer P, Druffel E, et al. (2020) Understanding growth and age of red tree corals (Primnoa pacifica) in the North Pacific Ocean. PLoS ONE 15(12): e0241692. https://doi.org/10.1371/journal. pone.0241692 Editor: Erik Caroselli, University of Bologna, ITALY Received: January 10, 2020 Accepted: October 20, 2020 Published: December 1, 2020 Copyright: This is an open access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication. Data Availability Statement: Data are contained within the paper. Funding: BW & PE NA08OAR4300817 NOAA’s West Coast & Polar Regions Undersea Research Center. Program no longer running. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. PLOS ONE | https://doi.org/10.1371/journal.pone.0241692 December 1, 2020 1 / 18 PLOS ONE Primnoidae age models Introduction Fishing practices, including bottom trawling and long lining, can disturb benthic ecosystems, particularly those where the seafloor is highly structured with large sedentary invertebrates such as corals and sponges [1–4]. Gorgonian corals are especially vulnerable to fishing prac- tices due to easy ensnarement of this large arborescent sea-fan, which adheres to the seafloor with a single holdfast. Some gorgonians are very long-lived and communities of older colonies compromised by fishing gear may take decades to centuries to recover [5–9]. Understanding how fast these corals grow and the relationship between size and age can provide estimates of recovery times of these communities [2, 10, 11]. This is critically important in areas vulnerable to fishing disturbance such as those in the Northeast Pacific Ocean where the importance of these habitats to fisheries has been documented [3, 4, 12]. Gorgonian coral communities are important components of the sea floor because they pro- vide habitat for a diversity of invertebrates and fishes [13, 14]. In fact, these corals can serve as habitat engineers [15]: in their absence, shallow-water assemblages shift from predominately corals and sponges to algae and turf-forming species [16]. Furthermore, some gorgonian corals are indicator species, and in addition to disturbance from fishing activities are sensitive to warming ocean temperatures [16, 17] and oil pollution [18, 19]. The conservation of these cor- als is thus critical for maintaining ecological diversity and community resilience. The gorgonian Primnoa pacifica [20], also known as the red tree coral, are ecologically important deep-sea corals in the North Pacific Ocean [4]. They have been referred to as “key- stone species,” “foundation species,” and “ecosystem engineers” [4, 21]. These animals are com- prised of an internal skeleton arising from the holdfast attached directly to hard substrate. Their skeleton is largely comprised of protein-rich organic gorgonin sometimes interspersed with cal- cite. The source of elements to the gorgonin skeleton is organic material produced in surface waters and transported to depth to be fed upon by the corals. In contrast, the calcite elements are sourced from ambient seawater at depth [22–24]. A thin layer of coenenchyme with polyps covers the entirety of the skeleton. The skeletal central axis grows axially and radially, such that through time the coral grows taller along its axial axis and adds layers to the outside of its skele- tal trunk increasing the trunk diameter. They can grow to massive size (greater than 2 m in height [25]), in part because of their long lifespans that can exceed a century or more [26]. In the central skeletal axis of some gorgonian corals, concentric couplets of gorgonin-calcite bands form annually, providing a means to determine the age of a colony; however, fine-scale bands of unknown periodicity are also present, indicating possible drivers of skeletal banding [9, 26–29]. The finer bands may reflect variations in the color of the organic skeleton (which in turn reflects the degree of protein cross-linkages during skeletal formation) and/or alternations of the gorgonin with calcite skeleton [9, 30, 31]. In addition to annual growth band counts, radiometric dating (14C and 210Pb) can provide estimates of coral age, albeit with potential uncertainties depending on the age and collection date of the coral [24, 26, 32, 33]. Previous studies using a combination of annual growth band counts and radiometric dating in P. pacifica yielded different estimates of radial growth rates ranging from 0.14 to 0.57 mm yr-1 [26, 34, 35]. Axial growth rates for this species have only been reported for two specimens, and those estimates ranged from 1.60 to 2.32 cm yr-1 [26]. Obtaining axial growth rates in a much larger sample of corals is key to determine the rate of recovery of disturbed communities. Therefore, the objectives of this study were to 1) determine the age and growth (radial and axial) for a suite of colonies collected in the Northeast Pacific Ocean with morphological data; 2) develop and evaluate age estimate calculations converting morphological data into age; 3) evaluate annual growth band counts and radiometric dating as age determination techniques; PLOS ONE | https://doi.org/10.1371/journal.pone.0241692 December 1, 2020 2 / 18 PLOS ONE Primnoidae age models and 4) examine the periodicity and potential drivers of sub-annual banding in the axial skele- ton. As a whole, this work provides critical insight into recovery times of P. pacifica which helps inform management policies of important deep-sea coral habitats. Materials and methods Sample sites and collection Three intact colonies were collected in 2013 using the remotely operated vehicle (ROV) H2000 deployed from the F/V Alaska Provider (Table 1; Figs 1 and 2). Seven intact colonies were col- lected in 2015 using the ROV Zeus II deployed from the R/V Dorado Discovery (Table 1, Fig 1). Additionally, a single specimen (GOA 004; Table 1) was collected with a research bottom trawl deployed from the F/V Alaska Provider just prior to the 2013 cruise. Fieldwork research was performed by NOAA’s Alaska Fisheries Science Center and Deep Sea Coral Research and Technology Program under the authority of the U.S. Department of Commerce. Colonies were air dried on board the vessels and morphological data were recorded, including maxi- mum height, basal diameter, wet weight, and distance from the base to first branch. Sample preparation In the laboratory, a diamond-edge saw was used to cut three adjacent 0.25–0.50 cm cross-sec- tion discs from the basal portion (holdfast) of each coral. One cross-section from each coral was mounted on a glass slide and polished for annual band counts and digital imaging, one cross-section was used for dissected band counts and radiocarbon dating, and one cross-sec- tion was used for 210Pb dating. Although growth rings were only counted in two of the three cross sections, it is assumed that all cross sections contain the same number of growth bands since they were cut from adjacent parts of the coral. The two cross-sections used for (1) dis- sected band counts and radiocarbon dating and (2) 210Pb dating were bathed in 100 ml of 5% HCl solution for a minimum of 10 days (up to four weeks). The HCl solution was refreshed every other week for four weeks so that all the calcite bands layered between the organic Table 1. Sample ID, collection information, and morphological data for the 11 coral colonies included in this study. Age estimates for all specimens are derived from annual growth bands. Dates are mm/dd/yyyy. Sample ID Collection date WPA 001 6/4/2015 WPA 002 6/5/2015 WPA 003 6/5/2015 WPA 004 6/7/2015 Locality Latitude Longitude Depth 58.2457 -138.9045 (m) 141 58.2458 -138.9044 142 58.2377 -138.9894 147 58.2047 -138.8175 163 Fairweather Grand Fairweather Grand Fairweather Grand Fairweather Grand WPB 005 6/9/2015 Dixon Entrance 54.6252 -132.8828 WPB 006 6/8/2015 Dixon Entrance 54.6345 -132.8510 WPB 007 6/8/2015 Dixon Entrance 54.6345 -132.8510 GOA 004 7/19/2013 Portlock Bank 58.3102 -149.5087 GOA 011 8/13/2013 Shutter Ridge 56.1749 -135.1165 GOA 022 8/13/2013 Shutter Ridge 56.1722 -135.1160 GOA 067 8/15/2013 Shutter Ridge 56.1784 -135.1180 335 164 164 147 191 203 214 Height (cm) Width (cm) 123 171 110 223 160 99 87 78 65 72 67 60 70 101 98 26 23 50 24 32 153 121 BaseD (mm) 32.5 25.5 45.0 53.0 49.0 19.0 15.0 15.0 10.0 16.0 45.0 DistToBr� (cm) Wet weight (kg) Age estimate (years) 7 62 30 2.24 1.25 n/a 29.5 20.45 29 20 17 16 8.5 12.3 32 16.6 1.14 0.64 1.08 0.62 1.13 n/a 26 ± 2 28 ± 2 19 ± 2 80 ± 1 67 ± 4 16 ± 2 15 ± 1 14 ± 2 12 ± 2 18 ± 1 31 ± 2 �Distance to first branch. https://doi.org/10.1371/journal.pone.0241692.t001 PLOS ONE | https://doi.org/10.1371/journal.pone.0241692 December 1, 2020 3 / 18 PLOS ONE Primnoidae age models Fig 1. Map showing the collection sites of Primnoa pacifica from the current project relative to previous studies (Andrews et al., 2002 [26]; Aranha et al., 2014 [34]; Williams et al., 2007 [35]). Map created by Michele Masuda. https://doi.org/10.1371/journal.pone.0241692.g001 skeleton had dissolved. After acidification, the sample was transferred to soak in Milli’Q water for band peeling. Photographed band counts Using the mounted and polished cross-sections (Fig 2), the number of annual bands along the longest radial transect were counted by two researchers under a light microscope for all speci- mens in this study. The radial growth rate (mm year-1) of each coral was obtained by dividing the base diameter of the coral by the age of the coral, as determined by annual growth bands. Similarly, axial growth rates (cm year-1) were calculated by dividing the maximum height of the coral by the age of the coral, as determined by annual growth bands. The cross sections were then imaged using a Nikon digital microscope with NIS-Elements, a Nikon microscope software package. Using these high-resolution images, the total number of sub-annual bands were counted along the same longest radial transect for four specimens (GOA 011, WPA 002, WPA 004, and WPB 005) by two researchers (S1 File). Four larger speci- mens were chosen to encompass a range in sizes and the ability to cut adjacent cross sections. The bands were sometimes difficult to distinguish; when major discrepancies in band counts were evident, researchers re-counted bands collaboratively, discussing the presence or absence of a band when uncertainties arose. Dissected band counts Using one of the HCl-bathed cross-sections, sub-annual growth bands were peeled and counted in sections from the outside to the center of the coral cross-section using forceps and PLOS ONE | https://doi.org/10.1371/journal.pone.0241692 December 1, 2020 4 / 18 PLOS ONE Primnoidae age models Fig 2. (A) Primnoa pacifica colonies on the Shutter Ridge at a depth of 180 m, (B) sub-section from specimen WPA 005 with a white line showing region of axis used to count growth bands, (C) image showing annual growth bands and (D) sub-annual growth bands. https://doi.org/10.1371/journal.pone.0241692.g002 working under a microscope at 30x magnification. Without tearing the bands, sections were peeled with the least number of bands possible, with each section containing approximately 1–20 bands depending on the ease of band separation. If possible, sections were peeled all the way around the circumference of the sample. The separated sections were air dried and pack- aged into labeled weigh paper packets. Radiocarbon analysis For radiocarbon sample preparation, a laminar flow hood workspace was cleaned with deion- ized water. All glassware was soaked in 10% HCl for a minimum of 1 hour. The glass vials used for radiocarbon dating were additionally combusted at 540 ˚C for 2 hours, while the vial caps were washed with soap and water, and then acid washed in 10% HCl for 30 seconds and rinsed with water. After sub-annual bands were dissected and counted, three of the four colonies (WPA 002, WPB 005, and WPA 004) were selected for radiocarbon analysis. Based on annual band counts, colony GOA 011 was too young for radiocarbon dating to be effective. Five milli- gram sub-samples reflecting no more than 10% of the entire sample and equidistant from each other based on dissected band count numbers were pulverized into acid washed glass vials. The outermost bands of each sample’s cross-section were analyzed to determine the Δ14C val- ues at the time of collection. The 14C was measured at the Keck Carbon Cycle Accelerator Mass Spectrometry at the University of California, Irvine. Five milligrams of a National Insti- tute of Standards and Technology (NIST) wood standard (Firi H) and a coal standard were also prepared as a reference for radiocarbon analysis and sample preparation backgrounds were subtracted based on these measurements. All results were corrected for isotopic fraction- ation according to the conventions of Stuiver and Polach (1977) [36]. Δ14C values were assigned a year using the following equation: Year ¼ ðD14C (cid:0) DClo AÞðYearhi A (cid:0) Yearlo AÞ DChi A (cid:0) D14Clo A þ Yearlo A PLOS ONE | https://doi.org/10.1371/journal.pone.0241692 December 1, 2020 ð1Þ 5 / 18 PLOS ONE Primnoidae age models where Year is the associated year calculated for this study, Δ14C is the measured value of the coral sub-section being dated, Δ14Clo A and Yearlo A are the lower estimates of Δ14C and calcu- lated year, and Δ14Chi A and Yearhi A are the upper estimates of Δ14C and calculated year (data from Andrews et al., 2013 [37]). Radiocarbon records from Andrews et al. (2013) [37] were constructed using Δ14C data from otoliths of Northeast Pacific yelloweye rockfish (Sebastes ruberrimus), Pacific halibut, and known-age abalone shell samples. A loess curve, including 95% confidence intervals, was fit to the Δ14C otolith data of Northeast Pacific yelloweye rock- fish to produce a radiocarbon bomb curve (S1 Fig in S1 File). The yelloweye rockfish are bot- tom dwelling fish that were collected in waters off southeast Alaska [38]. The rockfish chronology identified the initial rise in 14C in the late 1950s and peak values in the late 1960s/ early 1970s, which agrees with the 14C reconstruction by Roark et al. (2005) [23] for a bamboo coral from the northwestern coast of Canada. Lead-210 dating Measurements of 210Pb activity provided a third method for evaluating coral age for specimens WPA 002, WPA 004, and WPB 005. Specimens were rinsed thoroughly in Milli-Q water and then dried on a clean watch glass in a laminar flow hood. While still malleable enough to be manipulated, the specimens were split into three sections for sub-sampling (exterior, middle, and interior). Subsamples were dried, cooled in liquid nitrogen, and pulverized in a Genogrin- der. Subsamples were weighed (~ 1 g) and placed in clean, new 4-ml vials. Due to the smaller size of the cross-sections initially cut, it was necessary to cut a second cross-section for WPA 002 and WPA 004, such that the sub-samples for these specimens were a combination of two cross-sections. All prepared 210Pb samples were sent to the United States Geological Survey at the Pacific Coastal and Marine Science Center for 210Pb dating using a germanium gamma ray detector. 210Pb decays at a constant rate yet is also in secular equilibrium with 226Ra. To deter- mine the specimen age from 210Pb values, the excess amount of 210Pb (210Pbex), ultimately derived from deposition of atmospheric 222Rn, which in turn decays to 210Pb, is determined by subtracting measured 226Ra activity. 210Pbex decays according to the law of radioactive decay: Aex ¼ Aoe(cid:0) lt ð2Þ where Aex is the measured 210Pbex at time t, Ao is the initial 210Pbex at time 0, and λ is 0.0311 or the natural log of 2 divided by the half-life of 210Pb (22.3 years). Inputting the calculated 210Pbex values for each specimen into the equation above yields the age of each sample. Data analysis The relationship between specimen age derived from annual band counts and the morphologi- cal data was quantified by comparing linear and logarithmic fits to find the best fit model using R software. For height, width, and basal diameters, these equations represent a means to convert a measurement of a morphological parameter into specimen age. We primarily focus on basal diameter because this metric has most often been reported in corals that also have their ages determined. We validated these age estimation calculations in three ways: 1) We applied the calculations to ten previously collected and aged corals with available radii data that were less than a century old from the northeast Pacific Ocean ([34, 35]; Table 3). Basal diameter was determined from the radius in these corals (assumed diameter = 2�radius, although see discussion on this assumption below). 2) We developed and validated a second calculation relating basal diameter to age using all available colonies (< century in age) for the northeast Pacific Ocean. 3) We apply this regional age estimation to corals collected from the western Pacific Ocean [39]. For each of these comparisons, we evaluated how well the ages PLOS ONE | https://doi.org/10.1371/journal.pone.0241692 December 1, 2020 6 / 18 PLOS ONE Primnoidae age models derived from the basal estimates correspond to the ages derived from annual band counts or 210Pb dating. In the three specimens with 14C-dating (WPA 002, WPA 004, WPB 005), we calculated banding frequencies and sub-annual radial and axial growth rates. Banding frequencies (growth bands year-1) were calculated through time for each colony according to the following equation: Banding frequency ¼ Band countn (cid:0) Band countnþ1 Agen (cid:0) Agenþ1 ; n ¼ 1; 2; 3 . . . ð3Þ Where Band count is the average associated growth band count of the sub-section used for radiocarbon dating, Agen is the radiocarbon age of the older n sub-section of the coral, and n is the sub-section of the coral sent in for radiocarbon dating (a smaller n is closer to the center of the coral representing older skeleton). Results Morphological data Eleven intact specimens were collected during two research cruises in 2013 and 2015. Coral height (axial length) ranged from 65 to 223 cm, width ranged from 23 to 121 cm, and basal diameter ranged from 1 to 5.3 cm (Table 1). Many of the morphological parameters signifi- cantly correlated with each. Height and width were significantly correlated, such that taller specimens were wider (p = 0.004, r2 = 0.62, N = 11) (Table 2). Basal diameter significantly cor- related with height (p = 0.0003, r2 = 0.79, N = 11) and width (p<0.0001, r2 = 0.85, N = 11) (Tables 1 and 2). Height correlated with the distance to first branch (p = 0.03, r2 = 0.41, N = 11) (Table 2). The weight of the specimen significantly correlated with age, height, weight, and axial and radial growth rates (Table 2). Table 2. Statistics comparing morphological data, ages, and growth rates of Primnoa pacifica. Age estimate (years) Age estimate (years) Height (cm) Width (cm) BaseD (cm) DistToBr (cm) Wet weight (kg) (N = 9) Radial growth rates (mm yr-1) Axial growth rates (cm yr-1) Height (cm) 0.0001� 0.83 Width (cm) 0.0028� 0.65 BaseD (mm) 0.0000� 0.90 0.0041 0.0003 0.62 0.79 0.0001 0.85 DistToBr (cm) Wet weight (kg) (N = 9) 0.3475 0.10 0.0333 0.41 0.2020 0.17 0.2821 0.13 0.0000 0.97 0.0100 0.64 0.0021 0.76 0.0003 0.86 0.5543 0.05 Radial growth rate (mm yr-1) 0.0748 Axial growth rate (cm yr-1) 0.0008 0.31 0.3930 0.08 0.8160 0.01 0.6523 0.02 0.7496 0.01 0.0296 0.51 0.73 0.0866 0.29 0.0631 0.33 0.0089 0.55 0.8717 0.00 0.0022 0.76 0.0689 0.32 Age estimates for all specimens are derived from visual counts of annual growth bands. Reported is the p-value and r2 for each comparison. All statistics are based on a linear model, except those indicated with an asterisk, which are based on a logarithmic model. N = 11 unless otherwise noted. Significant relationships are in bold. https://doi.org/10.1371/journal.pone.0241692.t002 PLOS ONE | https://doi.org/10.1371/journal.pone.0241692 December 1, 2020 7 / 18 PLOS ONE Primnoidae age models Age estimates based on annual band counts ranged from 12 ± 2 to 80 ± 1 years. These age estimates correlated with all of the morphological data, except distance to first branch (Table 2). Using a logarithmic model, older specimens were taller (p<0.0001, r2 = 0.83, N = 11), wider (p = 0.0028, r2 = 0.65, N = 11), and had a larger basal diameter (p<0.0001, r2 = 0.90, N = 11) (Table 2; Fig 3). A logarithmic model provided the best fit between age and mor- phological data, excluding weight (Fig 3). Thus height, width, and basal diameter can be used to determine specimen age in years (± standard error) using the following equations: ð Age �21 Þ ¼ e Heightþ109:8 72:6 ð Age �21 Þ ¼ e Widthþ79:4 44:1 Age �5ð Þ ¼ e Basal Diameterþ45:8 23:0 ð4Þ ð5Þ ð6Þ Axial growth rates varied from 2.41 to 6.39 cm year-1 and radial growth rates varied from 0.33 to 0.74 mm year-1 (Table 3). Using a linear model, axial growth rates significantly inversely correlated with age (p<0.001, r2 = 0.73, N = 11) and basal diameter (p = 0.0089, r2 = 0.55, N = 11), and positively correlated with specimen weight (p = 0.0022, r2 = 0.76, N = 9) (Table 2). Radial growth rates only significantly varied with specimen weight (p = 0.030, r2 = 0.51, N = 9) (Table 2). Band count comparisons In four specimens ranging in age from 12 to 80 years, based on annual band counts, the num- ber of sub-annual dissected bands ranged from 290 ± 28 to 1589 ± 124, while the number of sub-annual photographed bands ranged from 152 ± 10 to 1131 ± 45 (Table 4). The number of sub-annual photographed growth bands positively correlated with the number of annual growth bands, such that there is an average of 14 ± 0.8 sub-annual bands for every annual Fig 3. Height (black circles), width (white circles), and basal diameter (grey circles) of Primnoa pacifica decrease logarithmically with age. Statistics are presented in Table 2, Eqs (4)–(6). Dashed lines represent 95% confidence intervals. https://doi.org/10.1371/journal.pone.0241692.g003 PLOS ONE | https://doi.org/10.1371/journal.pone.0241692 December 1, 2020 8 / 18 PLOS ONE Primnoidae age models Table 3. Reported ages, radial and axial growth rates of Primnoa pacifica. Radiometric age estimates for Andrews et al. (2002) [26] and Williams et al. (2007) [35] were derived from 210Pb. Radiometric age estimates for Aranha et al. (2014) [34] were derived from 14C. OCNMS is the Olympic Coast National Marine Sanctuary. Shiribeshi seamount is located in the Sea of Japan off the western coast of Hokkaido, Japan. Annual bands were ambiguous in specimens missing age estimates from growth bands (column 3). Location Colony name Age estimate Fairweather Ground Fairweather Ground Fairweather Ground WPA 001 WPA 002� WPA 003 Fairweather Ground WPA 004 Dixon Entrance Dixon Entrance Dixon Entrance Portlock Bank Shutter Ridge Shutter Ridge Shutter Ridge Dixon Entrance Dixon Entrance Dixon Entrance Dixon Entrance Dixon Entrance OCNMS OCNMS OCNMS OCNMS Dixon Entrance Dixon Entrance WPB 005 WPB 006 WPB 007 GOA 004 GOA 011 GOA 022 GOA 067 R1153-0003 R1155-0012 R1155-0013 R1156-0004 R1156-0016 R1162-0015 R1162-0016 R1162-0005 R1165-0002 Colony 1 Colony 2 Portlock Bank, Alaska Southeast Alaska Eastern Aleutian Islands PAL P88 P26 Shiribeshi Seamount Shiribeshi Seamount Shiribeshi Seamount Shiribeshi Seamount Shiribeshi Seamount Shiribeshi Seamount Shiribeshi Seamount Shiribeshi Seamount Shiribeshi Seamount 1 2 3 4 5 6 7 8 9 Shiribeshi Seamount 10 (annual bands) 26 28 19 80 67 16 15 14 12 18 31 46 34 30 22 16 45 31 40 21 124 5 8 5 9 12 14 15 18 29 40 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ## —— ± —— —— ± ± ± ± ± ± ± ± ± ± 2 2 2 1 4 2 1 2 2 1 2 4 2 3 2 1 2 2 2 4 3 3 2 2 2 2 3 2 6 2 2 Age estimate (radiometric) Radius (mm) Radial growth rate (mm year-1) Height (cm) Axial growth rate (cm year-1) Reference —— 42 ± 0.1 —— 66 ± 1 38 ± 1 —— 26 11 47 6 13 15 11 10 4 5 5 0.6 46 34 119 22 49 75 51 62 87 195 123 43 —— —— —— —— —— ± ± ± ± ± ± ± ± ± 112 —— ± ± ± —— —— —— —— —— —— —— —— —— —— 16.3 12.8 12.0 26.5 24.5 9.5 7.5 7.5 5.0 8.0 22.5 10.00 19.50 9.00 5.00 4.60 18.75 16.50 13.00 8.10 —— —— 17 23 16 0.7 0.9 1.0 1.2 1.3 1.7 2.0 2.2 2.8 5.4 0.64 ± 0.05 0.46 ± 0.04 0.65 ± 0.07 0.33 ± 0.00 0.37 ± 0.02 0.61 ± 0.08 0.52 ± 0.04 0.56 ± 0.08 0.43 ± 0.08 0.46 ± 0.03 0.74 ± 0.05 0.08 0.22 ± ± ± ± ± ± ± ± ± 0.13 0.07 0.05 0.08 0.06 0.06 0.03 0.08 0.002 ± —— 0.57 0.30 0.23 0.29 0.42 0.53 0.33 0.39 0.18 0.14 0.19 0.37 0.13 0.11 0.19 0.13 0.11 0.12 0.14 0.12 0.10 0.13 123 171 110 223 160 99 87 78 65 72 153 —— —— —— —— —— —— —— —— —— —— —— —— —— —— —— —— —— —— —— —— —— —— —— —— 4.82 ± 0.37 This study 6.22 ± 0.44 This study 5.95 ± 0.62 This study 2.79 ± 0.03 This study 2.41 ± 0.14 This study 6.39 ± 0.79 This study 6.00 ± 0.39 This study 5.78 ± 0.81 This study 5.65 ± 0.93 This study 4.11 ± 0.22 This study 5.02 ± 0.32 This study 1.74 2.32 —— —— —— —— —— —— —— —— —— ± ± —— —— —— —— —— —— —— —— —— —— —— —— —— Aranha et al., 2014 Aranha et al., 2014 Aranha et al., 2014 Aranha et al., 2014 Aranha et al., 2014 Aranha et al., 2014 Aranha et al., 2014 Aranha et al., 2014 Aranha et al., 2014 0.19 Andrews et al., 2002 0.09 Andrews et al., 2002 Williams et al., 2007 Williams et al., 2007 Williams et al., 2007 Matsumoto, 2007 Matsumoto, 2007 Matsumoto, 2007 Matsumoto, 2007 Matsumoto, 2007 Matsumoto, 2007 Matsumoto, 2007 Matsumoto, 2007 Matsumoto, 2007 Matsumoto, 2007 �14C results for WPA 002 could also yield an age of 49.1 years with radial growth rates of 0.31 mm yr-1 and axial growth rate of 4.12 cm yr-1. https://doi.org/10.1371/journal.pone.0241692.t003 band (p = 0.0002, r2 = 1.0, N = 4). Conversely, there is an average of 18 ± 5. Sub-annual dis- sected growth bands for every annual growth band, although this relationship was not statisti- cally significant (p = 0.096, r2 = 0.82, N = 4). The number of dissected band counts increased linearly with sub-annual photographed bands and in most cases was greater than the number of sub-annual photographed band counts, although this relationship was also not significant (p = 0.103, r2 = 0.81, N = 4) (Table 4). PLOS ONE | https://doi.org/10.1371/journal.pone.0241692 December 1, 2020 9 / 18 PLOS ONE Primnoidae age models Table 4. Annual, sub-annual dissected, and sub-annual photographed band counts for four Primnoa pacifica col- onies in this study. Sample ID Annual Band Count Sub-annual Dissected Band Count Sub-annual Photographed Band Count WPA 002 WPA 004 WPB 005 GOA 011 28 80 67 12 ± ± ± ± 2 1 4 2 477 1589 790 290 ± ± ± ± 64 124 93 28 399 1131 960 152 ± ± ± ± 36 45 72 10 https://doi.org/10.1371/journal.pone.0241692.t004 Radiometric age determination Measured Δ14C values ranged from -93.7 ± 1.3 to 89.4 ± 2.0 (Table 5) and were consistent with previously published regional Δ14C records from the North Pacific Ocean [23, 37] (S1 Fig in S1 File). The ages of the corals determined from the timing of the bomb-curve radiocarbon ranged in age from 38 ± 1 (WPB 005) to 66 ± 1 (WPA 004) years old. Thus, based on ages and collection year, the corals started growing between 1949 and 1977. The radiocarbon-derived coral ages were similar to, but consistently older, than the ages derived from the lead-210 dating (Table 5). WPA 002 has two data points for the oldest value (S1 Fig in S1 File) because the innermost Δ14C value in this specimen could reflect either rising or declining bomb Δ14C values. Based on growth rates and comparisons with the other corals, Table 5. Δ14C and 210Pbex values for radiocarbon and lead-210 dating, respectively, and resulting ages from each dating method for specimens WPA 002, WPA 004, and WPB 005. The banding frequency and growth rate for each specimen was calculated from the radiocarbon ages. The brackets around the 210Pbex values represent the possible range of bands associated with this measurement. Sample ID & sub-section Sub-annual dissected band count Δ14C (‰) Age from radiocarbon (years) Frequency (growth band year-1) Radial growth rate (mm yr-1) Vertical growth rate (cm yr-1) 210Pbex (dpm/g) Age from 210Pb (years) WPA 002 (Outer) 1 2 (Center) 3 (Center)� 3 WPA 004 (Outer) 1 2 3 4 5 6 7 (Center) 8 WPB 005 (Outer) 1 2 3 4 5 (Center) 6 6.5 ± 2 241 ± 56 477 ± 64 477 ± 64 13.5 ± 2 534 ± 38 712 ± 47 869 ± 19 1070 ± 81 1210 ± 6 1332 ± 33 1589 ± 124 23 ± 10 265 ± 52 348 ± 11 416 ± 28 534 ± 106 790 ± 93 -5.8 ± 2.0 31.5 ± 1.5 43.6 ± 1.7 43.6 ± 1.7 -12.2 ± 1.8 68.1 ± 1.6 80.0 ± 2.2 89.4 ± 2.0 75.3 ± 1.8 -22.5 ± 1.9 -56.1 ± 1.8 -93.7 ± 1.3 5.2 ± 1.8 87.2 ± 1.6 27.5 ± 1.9 -58.3 ± 1.7 -91.0 ± 1.4 -93.5 ± 1.6 0 38.2 ± 0.6 41.5 ± 0.1 49.1 ± 0.4 0 50.0 ± 0.3 47.7 ± 0.3 47.0 48.1 ± 0.1 51.7 ± 1.0 53.8 ± 0.2 65.7 ± 1.0 0 ± 0.7 19.5 22.4 ± 0.1 26.6 ± 0.2 35.4 ± 1.4 38.1 ± 1.3 0 6.1 16.1 53.6 0 11.6 81.5 203.5 177.1 39.5 56.4 17.5 0 12.4 29.3 16.1 13.3 83.6 — — 0.26 0.307 — — — — — — — 0.404 — — — — — 0.644 — — 3.49 4.12 — — — — — — — 3.4 — — — — — 4.2 2.52 ± 0.96 5.46 ± 1.28 2.82 ± 1.38 1.51 ± 0.73 — — — 0 0.90 ± 1.76 16.5 0.24 ± 0.69 59.7 3.57 ± 0.96 3.54 ± 1.28 0 0.3 1.58 ± 1.23 26.2 �WPA 002 center Δ14C value yielded two ages depending if the sample was placed on the rise or the decline of the bomb carbon. https://doi.org/10.1371/journal.pone.0241692.t005 PLOS ONE | https://doi.org/10.1371/journal.pone.0241692 December 1, 2020 10 / 18 PLOS ONE Primnoidae age models we estimate that the older value is more likely accurate and estimate the 14C-derived coral age as 42 ± 0.1 years old. 210Pbex values ranged from 0.24 ± 0.69 to 3.57 ± 0.96 dpm/g (desentigrations per minute / gram) and decreased as expected from the outside of the colony toward the inside for corals WPA 004 and WPB 005 (Table 5; S2 Fig in S1 File). Eq (2) calculated the age of WPA 004 to between 28.4 and 59.7 and WPB 005 to be 26.2 (+ 38.4/– 10.9) years old. 210Pb dating for sam- ple WPA 002 was inconclusive because 210Pbex activity increased from the outside of the coral toward the inside (Table 5; S2 Fig in S1 File). Age estimation validation In previously collected and aged corals (Table 2; corals from Aranha et al., 2104 [34] and Wil- liams et al., 2007 [35]), we find that the calculated ages using Eq (6) were similar in magnitude but underestimated age (on average ~ 10 years) relative to the reported ages (Fig 4). Thus, we revised the age estimate equation to include the corals with reported basal diame- ters from the Aranha et al. (2014) [34] and Williams et al. (2007) [35] for corals < century old: Age �8ð Þ ¼ e Basal Diameterþ41:9 20:5 ð7Þ that is inclusive of corals from the northeast Pacific Ocean (Fig 5; N = 21). Using this new regional equation, the average difference between age estimation from annual band counts or 210Pb dating versus basal diameter declined to only three years (Fig 6). We evaluated Eq (7) by calculating age in ten P. pacifica colonies collected from the Sea of Japan [39]. We find that the age estimated from basal diameter underestimated age compared to the annual band counts by an average of seven years, although this was largely driven by older corals with wider diameters (Fig 7). In coral IDs 1 through 7, the basal diameter underes- timated age by three years. Discussion The results of this analysis indicate a strong correlation between age and size (height, width, and basal diameter) of Primnoa pacifica (Table 2; Fig 3), and thus provide support for develop- ment of non-destructive techniques to estimate the age of younger (< century) specimens in situ. In this species of gorgonian coral, the growth rates declined logarithmically with age such Fig 4. Age estimations based on basal diameter (light grey bar) compared to age (annual bands; black line) of Primnoa pacifica <100 years old from previous studies (Aranha et. al. 2014 [34]; Williams et al. 2007 [35]). Age estimations were calculated using Eq 6 which translates basal diameter into a calculated age. Error bars show standard deviation. �Age based on lead-210 dating not annual growth bands. https://doi.org/10.1371/journal.pone.0241692.g004 PLOS ONE | https://doi.org/10.1371/journal.pone.0241692 December 1, 2020 11 / 18 PLOS ONE Primnoidae age models Fig 5. Basal diameter of Primnoa pacifica corals from this study (white circles), Aranha et al., 2014 [34] (black circles), and Williams et al., 2007 [35] (grey circles) decrease logarithmically with age (Eq (7), r2 = 0.66, p<0.001, N = 21). Corals over a century in age were excluded from this analysis. https://doi.org/10.1371/journal.pone.0241692.g005 that younger corals grew faster than older corals. This is consistent with other species of gorgo- nians [32, 40, 41]. Radial growth rates were similar to those reported in previous studies, albeit on the higher end of the reported ranges (Table 3). Axial growth rates were higher than previously reported for two colonies, both of which were older than a century (Table 3). Thus, there is significant variability in radial and axial growth rates, depending the age of the coral. The faster growth rates reported here likely reflect the younger age of the corals. Because growth logarithmically declines with age (Fig 3), eventually reaching a plateau–the age estimate techniques developed here are less useful for older corals and may underestimate the age of very old corals. Fig 6. Age estimations based on basal diameter (grey bar) compared to age (annual growth bands; black line) for 21 Primnoa pacifica colonies <100 years old from three studies (This study; Aranha et al. 2014 [34]; Williams et al. 2007[35]). Age estimations were calculated using Eq (7). Colonies are ordered from left to right based on increasing basal diameter measurements; diamond markers indicate a basal diameter of 39 mm or more. �Age based on lead-210 dating not annual growth bands. https://doi.org/10.1371/journal.pone.0241692.g006 PLOS ONE | https://doi.org/10.1371/journal.pone.0241692 December 1, 2020 12 / 18 PLOS ONE Primnoidae age models Fig 7. Age estimations based on basal diameter (grey bar) compared to age (annual growth bands; black line) for ten (n = 10) Primnoa pacifica colonies collected from the Sea of Japan [39]. Age estimations were calculated using Eq (7). Colonies are ordered from left to right based on increasing basal diameter measurements. https://doi.org/10.1371/journal.pone.0241692.g007 Age estimation validation Age estimated from basal diameter largely agreed with age from annual band counts in a suite of corals from the northeast Pacific (Fig 6). Discrepancies between the age estimation and annual band counts was the greatest for the oldest specimens with the largest diameters. We hypothesize that smaller corals represent colonies with minimal branching and symmetric radii. In contrast, the larger diameter colonies have more asymmetric basal growth (in which the basal diameter may not reflect 2 x radius) and often with many branches (e.g., specimen WPB 005, Fig 2). Thus Eq 7 is most effective for colonies with basal diameter < 39 mm to pro- vide approximate age estimates for P. pacifica corals in the North Pacific Ocean. When the northeast Pacific regional age estimation (Eq 7) was applied to corals from the western Pacific Ocean, basal diameter underestimated age in the oldest colonies. These colo- nies were, on average, collected from colder, deeper waters (350–505 m; 0.6–0.7˚C) and exhib- ited slower radial growth rates (0.1–0.19 mm year-1) than those from the northeast Pacific Ocean [39]. Thus, growth rates may plateau earlier in these corals than in the northeast Pacific Ocean. Some coral colonies in this study were collected remotely with fishing techniques, which tends to result in fragments, or portions, of a full colony, such that accurate height and width measurements are unavailable, so in-situ age estimations were not possible based on these properties. However, the equations can also be retroactively applied to previous studies that collected morphological data via video analysis but that did not collect corals for age analyses. A Stone (2014) [3] study used video with scaling lasers to categorize red tree corals into four size (height) categories: <0.5 m (small), 0.5–1 m (medium), 1–2 m (large), and >2 m (very large). The height categories can be transformed into the following approximate age ranges using Eq (4): 0–9 years, 9–18 years, 18–71 years, and 71+ years. Equs (4), (5) and (7) provide approximate age estimations for corals based solely on morphological data derived from in situ observations. The age estimates are most accurate for the smaller, younger corals (< century) because of the plateau in growth with age. Using this technique, it is possible to identify the locations of gorgonian thickets where recruitment is particularly high (many younger colonies), or colonies may serve as a source of the recruits (many larger colonies), and to locate climax communities where all age classes are well represented. PLOS ONE | https://doi.org/10.1371/journal.pone.0241692 December 1, 2020 13 / 18 PLOS ONE Primnoidae age models Insights into coral dating Three techniques determined the ages of a subset of the collected corals: annual growth bands were counted in 11 corals, and 14C and 210Pb dating techniques were performed on three cor- als. In some cases, results were ambiguous. In P. pacifica, growth bands form annually but are sometimes difficult to distinguish. Furthermore, radial growth does not always form concen- trically, meaning one direction of the coral can form more growth bands than other directions (Fig 2). Thus, most researchers count along the axes of maximum growth yielding the highest number of growth bands. In P. pacifica, annual growth bands form as clumps of gorgonin bands with some contributions of calcite. Older P. pacifica corals form significant calcite build- ups that mask growth bands; consequently, radiometric estimates of coral age typically exceed those of ages derived from growth bands for older corals [34]. The modern use of radiocarbon dating relies on the identification of the 14C bomb curve signature from thermonuclear bomb detonations in the 1950-60s. In the three corals analyzed for 14C content, identification of bomb-derived carbon provided age constraints on the timing of skeletal growth. The coral sample sizes analyzed for 14C typically included multiple years of growth, smoothing any seasonal variability in ocean 14C so that the coral 14C values should align with the regional 14C bomb curves. However, exact placement on that curve is subject to uncertainty, which is compounded by the analytical uncertainty of the 14C measurement. There are two instances in this study where these uncertainties impact age estimates. In WPA 002, the earliest 14C value measured from the core could either fall on the rising or declining limb of the bomb carbon curve (S1 Fig in S1 File). We assign it to the rise of the curve because otherwise radial growth rates are unreasonably high for the earliest part of this coral’s growth. In WPB 005, the coral was collected with no living tissue at the base of the specimen; thus, the outer layers of the skeleton were likely not formed immediately prior to collection. The 14C value for the outer layers is higher than the outer layers of the live collected WPA 002 and WPA 004, indicating it died some time prior to collection. However, because the slope of the decline of the 14C bomb curve flattens toward recent time, the placement of this sample on the curve has large uncertainty with time. Aligning the point exactly on the regional bomb curve gives an age of 38 years for the coral, which yields potentially unrealistically high growth rates. In this colony, annual band counts suggest a much older coral, with a death of only a few years prior to collection. As a result, the strength of 14C dating is most evident for specimens with known collection date, and with an age range that encompass the full bomb curve (extending prior to the mid-1950s). The 210Pb dating technique provides only rough age estimates. For example, 210Pb of a P. pacifica colony yielded an age range of 78 to 193 years with 95% confidence [26]. 210Pb dating can also yield inconclusive results, potentially in corals that are collected dead, although a mechanism of why is unknown [26]. Here, 210Pb yields an age for the dead colony WPB 005, albeit this age may be unrealistically young when viewed in the context of specimen size. In contrast, 210Pb yielded inconclusive results for the small live-collected coral WPA 002. Due to the large uncertainty around 210Pb, we propose that 14C and growth band counting are pre- ferred methods for determining coral age in this species. Sub-annual growth bands We closely examined the sub-annual banding using two techniques: growth bands counted through physical dissection of a cross-section and those visible in photographs of polished cross-sections. The number of dissected growth bands counted was in most cases much greater than the number counted in photographs, although the opposite was true for specimen WPB 004 (Table 4). The skeletal bands were difficult to count, which likely contributed to some of PLOS ONE | https://doi.org/10.1371/journal.pone.0241692 December 1, 2020 14 / 18 PLOS ONE Primnoidae age models the variation within each counting method–particularly with the photographed bands. Fused growth bands could, in some cases, be carefully peeled apart and counted, although more com- monly a peeled fused section included numerous bands that could not be separated. These bands were likely not individually visible in the photographed sections; thus, they were not included in the photographed band counts. We compared the number of sub-annual bands with the age of each coral to determine banding frequency. Overall, photographed bands average 14 ± 0.80 sub-annual bands year-1 and dissected bands average 18 ± 5.2 sub-annual bands year-1. We did not count the number of sub- annual bands per year directly because the annual groupings of sub-annual bands were not visible when physically dissecting the cross-sections nor at the lower magnification needed to count the annual bands in the photographs. However, when compared to ages derived from 14C-dating, the sub-annual banding frequency ranged from 200 bands year-1 to less than 15 bands year-1 for spec- imen WPA 004, 80 bands year-1 to less than 15 bands year-1 for specimen WPB 005, and over 15 bands year-1 to less than 10 bands year-1 for specimen WPA 002 (Table 5). Thus, the banding fre- quency can be highly variable through time. Spring tides (26 per year) have been proposed as a primary driver of sub-annual banding due to the influx of sedimentary organic layer [9]. How- ever, the variability in frequency and the number of bands per year is not consistent with only spring tides. We instead hypothesize that high seasonal and interannual variability of primary pro- ductivity [42] and/or energy allocation to reproduction [43, 44] combined with spring tides can influence the variable banding and growth pattern. Additional collections of colonies spanning the size range of the species coupled with time series in situ measurements of primary productiv- ity, specifically flux (POC or particulate organic carbon) delivery to the seafloor, and studies on reproductive seasonality of P. pacifica in the eastern Gulf of Alaska could elucidate the relationship between these factors and the sub-annual skeletal development. Growth rates Reported radial growth rates of Primnoa pacifica range from 0.14 to 0.74 mm year-1 (Table 3), with the corals in this study on the higher end of that range 0.33 to 0.74 mm year-1, compared with 0.22 to 0.57 mm year-1 (Aranha et al., 2014 [34]), 0.36 mm year-1 (Andrews et al., 2002 [26]), and 0.14 to 0.37 mm year-1 (Williams et al., 2007 [35]). Expanding the age–growth com- parisons to include all of these studies shows that radial growth rates are lower in older speci- mens (linear regression; p = 0.004, r2 = 0.35, N = 22). The decrease in radial growth rates overtime is also observed in North Atlantic Primnoa resedaeformis, a slower growing congener of P. pacifica [9, 32, 41]. Additionally, reported radial growth rates of P. resedaeformis (0.083 to 0.215 mm year-1) varied based on colony location; colonies in areas with a stronger tidal cur- rent exhibited faster radial growth than colonies in areas of weaker current [32]. There was no statistical difference between location and radial growth rates of the P. pacifica included in this study (one-way ANOVA; F4,17 = 1.15, p>0.05). Axial growth rates ranged from 2.41 to 6.39 cm year-1 (Table 3). This is substantially faster than previously reported axial growth rates of 1.74 and 2.32 cm year-1 (Andrews et al., 2002 [26]). This difference is easily explained by the decline in growth rates with age, and that the specimens measured by Andrews et al. (2002) [26] were substantially older than those in the present study (age of 114 years is reported for the specimen growing 1.74 cm year-1, with no age for the second colony). Age continues to have a strong, significant explanation of axial growth rates when we add this one specimen with both age and axial growth rates to our age– growth comparisons (linear regression; p<0.0001, r2 = 0.80, N = 12). A similar decrease in axial growth rates with age was reported in North Atlantic P. resedaeformis, where it was pro- posed that young corals (<30 years) grew four times as fast as older corals (>30 years) [41]. PLOS ONE | https://doi.org/10.1371/journal.pone.0241692 December 1, 2020 15 / 18 PLOS ONE Primnoidae age models Axial growth rates in these P. resedaeformis studies ranged from 1.00 to 2.61 cm year-1 (coral ages ranged from 18 years to 100 years) and were on the lower end compared to P. pacifica in this study [32, 41]. Implications for coral conservation With the ability to now estimate the age of red tree corals in situ we can readily determine how old corals are using non-invasive video survey techniques coupled with mensuration systems such as lasers or stereo-cameras [4, 6]. Such enhanced surveys could quickly determine the age-class structure and consequently maturity status of coral habitats. This information could be used by coastal managers to identify which aggregations are most vulnerable to disturbance from human activities, and which should be highlighted for protection. If age and growth char- acteristics are phylogenetically constrained, as has been suggested for some taxa [40], then the techniques developed and insights gained in this study could have broader application in the North Atlantic Ocean where another Primnoa species (e. g. P. resedaeformis; [13, 45]) also forms ecologically important habitats. Supporting information S1 File. (PDF) Acknowledgments We thank Ocean Science Services and the captain and crew of the FV Alaska Provider (2013) and Pelagic Research Services and the captain and crew of the RV Dorado Discovery (2015) for their support. Many thanks to Enrique Salgado and Robert McGuinn for assistance with specimen col- lection and field preparation. The findings and conclusions in this report are those of the author (s) and do not necessarily represent the views of the National Marine Fisheries Service (NOAA). The work was performed under the authority of the U.S. Department of Commerce. Author Contributions Conceptualization: Branwen Williams, Peter Etnoyer. Data curation: Emma Choy, Kelly Watanabe, Robert Stone, Thomas Lorenson, Mary Knaak. Formal analysis: Emma Choy, Kelly Watanabe, Ellen Druffel, Mary Knaak. Funding acquisition: Branwen Williams, Peter Etnoyer. Methodology: Branwen Williams. Project administration: Branwen Williams. Supervision: Robert Stone. Writing – original draft: Emma Choy, Kelly Watanabe, Branwen Williams, Robert Stone. Writing – review & editing: Emma Choy, Kelly Watanabe, Branwen Williams, Robert Stone, Ellen Druffel. References 1. Althaus F., Williams A., Schlacher T.A., Kloser R.J., Green M.A., Barker B.A., et al. 2009. Impacts of bottom trawling on deep-coral ecosystems of seamounts are long-lasting. Marine Ecology Progress Series 397, 279–294. PLOS ONE | https://doi.org/10.1371/journal.pone.0241692 December 1, 2020 16 / 18 PLOS ONE Primnoidae age models 2. Davies A.J., Roberts J.M. and Hall-Spencer J. 2007. Preserving Deep-Sea Natural Heritage: Emerging Issues in Offshore Conservation and Management. Biological Conservation 138, 299–312. 3. Stone R., 2014. The ecology of deep-sea coral and sponge habitats of the Aleutian Islands of Alaska. NOAA Profissional Paper NMFS 16, 52 pp, https://doi.org/10.7755/PP.7719 4. Stone R., Masuda M., Karinen J., 2014. Assessing the ecological importance of red tree coral thickets in the eastern Gulf of Alaska. ICES Journal of Marine Sciences, https://doi.org/10.1093/icesjms/ fsu1190 5. Clark M.R., Althaus F., Schlacher T., Williams A., Bowden D.A., and Rowden A.A. 2016. The Impacts of Deep-Sea Fisheries on Benthic Communities: A Review. ICES Journal of Marine Science 73, 51–69. 6. Etnoyer P., Wagner D., Fowle H., Poti M., Kinlan B., Georgian S., et al. 2018. Models of habitat suitabil- ity, size, and age-class structure for the deep-sea black coral Leiopathes glaberrima in the Gulf of Mexico. Deep Sea Research Part II: Topical Studies in Oceanography 150, 218–228. 7. Krieger K., Wing B., 2002. Megafaunal associations with deepwater corals (Primnoa sp.) in the Gulf of Alaska. Hydrobiologia 471, 83–90. 8. Morgan L., Etnoyer P., Scholz A., Mertens M., Powell M., 2005. Conservation and management impli- cations of deep-sea coral and fishing effort distributions in the Northeast Pacific Ocean. Cold-Water Corals and Ecosystems, 1171–1187. 9. Risk M., Heikoop J., Snow M., Beukens R., 2002. Lifespans and growth patterns of two deep-sea cor- als: Primnoa resedaeformis and Desmophyllum cristigalli. Hydrobiologia 471, 125–131. 10. Hapsari K.A., Biagioni S.A., Jennerjahn T.C., Reimer P., Saad A., Sabiham S., et al. 2018. Resilience of a Peatland in Central Sumatra, Indonesia to Past Anthropogenic Disturbance: Improving Conservation and Restoration Designs Using Palaeoecology. Journal of Ecology 106, 2473–2490. 11. Lovich J.E., and Bainbridge D., 1999. Anthropogenic Degradation of the Southern California Desert Ecosystem and Prospects for Natural Recovery and Restoration. Environmental Management 24, 309–326. https://doi.org/10.1007/s002679900235 PMID: 10486042 12. Du Preez C.D., and Tunnicliffe V., 2011. Shortspine thornyhead and rockfish (Scorpaenidae) distribu- tion in response to substratum, biogenic structures and trawling. Marine Ecology Progress Series 425, 217–231. 13. Risk M., McAllister D., Behnken L., 1998. Conservation of cold- and warm-water seafans: threatened ancient gorgonian groves. Sea Wind 12, 2–21. 14. Roberts S., Hirshfield M., 2004. Deep-sea corals: out of sight but no longer out of mind. Frontiers in Ecology and the Environment 2, 123–130. 15. Stone RP, Shotwell SK (2007) State of the deep coral ecosystems of the Alaska Region: Gulf of Alaska, Bering Sea, and the Aleutian Islands. In: Lumsden SE, Hourigan TF, Bruckner AW, Dorr G, editors. The State of Deep Coral Ecosystems of the United States, pp. 65–108. United States Department of Com- merce, NOAA Technical Memorandum CRCP-3. Silver Spring, MD. 16. Verdura J., Linares C., Ballesteros E., Coma R., Uriz M.J., Bensoussan N., et al. 2019. Biodiversity loss in a Mediterranean ecosystem due to an extreme warming event unveils the role of an engineering gor- gonian species. Scientific Reports 9, 5911. https://doi.org/10.1038/s41598-019-41929-0 PMID: 30976028 17. Gugliotti E.F., DeLorenzo M.E., Etnoyer P.J., 2019. Depth-dependent temperature variability in the Southern California bight with implications for the cold-water gorgonian octocoral Adelogorgia phyllo- sclera. Journal of Experimental Marine Biology and Ecology 514– 515, 118–126. 18. DeLeo D., Ruiz-Ramos D., Baums I., Cordes E., 2016. Response of deep-water corals to oil and chemi- cal dispersant exposure. Deep Sea Research Part II: Topical Studies in Oceanography 129, 137–147. 19. Frometa J., DeLorenzo M., Pisarski E., Etnoyer P., 2017. Toxicity of oil and dispersant on the deep water gorgonian octocoral Swiftia exserta, with implications for the effects of the Deepwater Horizon oil spill. Marine Pollution Bulletin 122, 91–99. https://doi.org/10.1016/j.marpolbul.2017.06.009 PMID: 28666594 20. Kinoshita K., 1907. Vorlaufige Mitteilung uber einige neue japanische Primnoidkorallen. Annotationes zoologicae Japonenses 6, 229–237. 21. Rossin A.H., Waller R.G., Stone R.P. 2019. The effects of in-vitro pH decrease on the gametogenesis of the red tree coral, Primnoa pacifica. PLoS ONE 14, e0203976. https://doi.org/10.1371/journal.pone. 0203976 PMID: 30998686 22. Griffin S., & Druffel E. (1989). Sources of Carbon to Deep-Sea Corals. Radiocarbon, 31(3), 533–543. 23. Roark E. B., Guilderson T. P., Flood-Page S., Dunbar R. B., Ingram B. L., Fallon S. J., et al. (2005), Radiocarbon-based ages and growth rates of bamboo corals from the Gulf of Alaska, Geophys. Res. Lett., 32, L04606, https://doi.org/10.1029/2004GL021919 PLOS ONE | https://doi.org/10.1371/journal.pone.0241692 December 1, 2020 17 / 18 PLOS ONE Primnoidae age models 24. Sherwood OA., Scott DB., Risk MJ., Guilderson, TP. (2005) Radiocarbon evidence for annual growth rings in the deep-sea octocoral Primnoa resedaeformis. Marine Ecology Progress Series. 301:129– 134. 25. Rooper C., Stone R., Etnoyer P., Conrath C., Reynolds J., Greene H.G., et al. 2017. Deep-Sea Coral Research and Technology Program: Alaska Deep-Sea Coral and Sponge Initiative Final Report. NOAA Tech. Memo. NMFS-OHC-2, 65 p. 26. Andrews A., Cordes E., Mahoney M., Munk K., Coale K., Cailliet G., et al. 2002. Age, growth and radio- metric age validation of a deep-sea, habitat-forming gorgonian (Primnoa resedaeformis) from the Gulf of Alaska, in: Watling L., Risk M. (Eds.), Hydrobiologia. Kluwer Academic Publishers, Netherlands. 27. Grigg R., 1974. Growth rings: Annual periodicity in two gorgonian corals. Ecology 55, 876–888. 28. Mistri M., Ceccherelli V., 1994. Growth and secondary production of the Mediterranean gorgonian Para- muricea clavata. Marine Ecology Progress Series 103, 291–296. 29. Mitchell N., Dardeau M., Schroeder W., 1993. Colony morphology, age structure, and relative growth of two gorgonian corals, Leptogorgia hebes (Verrill) and Leptogorgia virgulata (Lamarck), from the north- erm Gulf of Mexico. Coral Reefs 12, 65–70. 30. Goldberg W., 1974. Evidence of a sclerotized collagen from the skeleton of a gorgonian coral. Compar- ative Biochemistry and Physiology Part B: Comparative Biochemistry 49, 525–526. https://doi.org/10. 1016/0305-0491(74)90188-6 PMID: 4154179 31. Szmant-Froelich A., 1974. Structure, iodination and growth of the axial skeletons of Muricea californica and M. fruticosa (Coelenterata: Gorgonacea). Marine Biology 27, 299–306. 32. Sherwood OA, and Edinger E. (2009) Ages and Growth Rates of Some Deep-Sea Gorgonian and Anti- patharian Corals of Newfoundland and Labrador, Canadian Journal of Fisheries and Aquatic Sciences, 2009, 66(1): 142–152. 33. Noe´ S.U., Lembke-Jene L. & Dullo W. (2008) Varying growth rates in bamboo corals: sclerochronology and radiocarbon dating of a mid-Holocene deep-water gorgonian skeleton (Keratoisis sp.: Octocorallia) from Chatham Rise (New Zealand). Facies 54, 151–166 (2008). 34. Aranha R., Edinger E., Layne G., Piercey G., 2014. Growth rate variation and potential paleoceano- graphic proxies in Primnoa pacifica: Insights from high-resolution trace element microanalysis. Deep- Sea Research II 99, 213–226. 35. Williams B., Risk M., Stone R., Sinclair D., Ghaleb B., 2007. Oceanographic changes in the North Pacific Ocean over the past century recorded in deep-water gorgonian corals. Marine Ecology Progress Series 335, 85–94. 36. Stuiver M. and Polach H.A. (1977) Discussion Reporting of 14C Data. Radiocarbon. 19:355–363. 37. Andrews A., Leaf R., Rogers-Bennett L., Neuman M., Hawk H., Cailliet G. 2013. Bomb radiocarbon dat- ing of the endangered white abalone (Haliotis sorenseni): investigations of age, growth and lifespan. Marine and Freshwater Research 64, 1029–1039. 38. Kerr L.A., Andrews A.H., Frantz B.R., Coale K.H., Brown T.A., and Cailliet G.M. 2004. Radiocarbon in otoliths of yelloweye rockfish (Sebastes ruberrimus): a reference time series for the coastal waters of southeast Alaska. Canadian Journal of Fisheries and Aquatic Sciences 61, 443–451. 39. Matsumoto A., 2007. Effects of low water temperature on growth and magnesium carbonate concentra- tions in the cold-water gorgonian Primnoa pacifica. Bulletin of Marine Science 81: 423–435. 40. Stone R., Malecha P., Masuda M., 2017. A Five-Year, In Situ Growth Study on Shallow-Water Popula- tions of the Gorgonian Octocoral Calcigorgia spiculifera in the Gulf of Alaska. PLos ONE 12, e0169470. https://doi.org/10.1371/journal.pone.0169470 PMID: 28068374 41. Mortensen P.B. and Buhl-Mortensen L. 2005. Morphology and growth of the deep-water gorgonians Primnoa resedaeformis and Paragorgia arborea. Marine Biology. 147: 775–788. 42. Balcom B., Biggs D., Hu C., Montaga P., Stockwell D., 2011. A comparison of marine productivity among Outer Continental Shelf planning areas, in: Prepared by CSA International Inc. for the U. S. Dept. of the Interior, Bureau of Ocean Energy Management, Regulation and Enforcement (Eds.), Hern- don, VA., OCS Study BOEMRE 2011– 019, 195 pp. 43. Waller R.W., Stone R.P., Johnstone J., Mondragon J., 2014. Sexual reproduction and seasonality of the Alaskan red tree coral, Primnoa pacifica. PLos ONE 9, e90893. 908 https://doi.org/10.1371/journal. pone.0090893 PMID: 24770675 44. Waller R.W., Stone R. P., Rice L.N., Johnstone J., Rossin A.M., Hartill E., et al. 2019. Phenotypic plas- ticity or a reproductive dead end? Primnoa pacifica (Cnidaria: Alcyonacea) in the southeastern Alaska region. Frontiers in Marine Science 6, https://doi.org/103389/fmars.2019.00709 45. Auster P.J., Kilgour M., Packer D., Waller R., Auscavitch S. Watling L. 2013. Octocoral gardens in the Gulf of Maine (NW Atlantic). Biodiversity 14, 193–194. PLOS ONE | https://doi.org/10.1371/journal.pone.0241692 December 1, 2020 18 / 18 PLOS ONE
null
10.1021_acscentsci.3c00547.pdf
null
null
http://pubs.acs.org/journal/acscii This article is licensed under CC-BY 4.0 Article Molecular Crowding Facilitates Ribozyme-Catalyzed RNA Assembly Saurja DasGupta,* Stephanie Zhang, and Jack W. Szostak* Cite This: ACS Cent. Sci. 2023, 9, 1670−1678 Read Online ACCESS Metrics & More Article Recommendations *sı Supporting Information ABSTRACT: Catalytic RNAs or ribozymes are considered to be central to primordial biology. Most ribozymes require moderate to high concentrations of divalent cations such as Mg2+ to fold into their catalytically competent structures and perform catalysis. However, undesirable effects of Mg2+ such as hydrolysis of reactive RNA building blocks and degradation of RNA structures are likely to undermine its beneficial roles in ribozyme catalysis. Further, prebiotic cell-like compartments bounded by fatty acid membranes are destabilized in the presence of Mg2+, making ribozyme function inside prebiotically relevant protocells a significant challenge. Therefore, we sought to identify conditions that would enable ribozymes to retain activity at low concentrations of Mg2+. Inspired by the ability of ribozymes to function inside crowded cellular environments with <1 mM free Mg2+, we tested molecular crowding as a potential mechanism to lower the Mg2+ concentration required for ribozyme-catalyzed RNA assembly. Here, we show that the ribozyme-catalyzed ligation of phosphorimidazolide RNA substrates is significantly enhanced in the presence of the artificial crowding agent polyethylene glycol. We also found that molecular crowding preserves ligase activity under denaturing conditions such as alkaline pH and the presence of urea. Additionally, we show that crowding-induced stimulation of RNA-catalyzed RNA assembly is not limited to phosphorimidazolide ligation but extends to the RNA-catalyzed polymerization of nucleoside triphosphates. RNA-catalyzed RNA ligation is also stimulated by the presence of prebiotically relevant small molecules such as ethylene glycol, ribose, and amino acids, consistent with a role for molecular crowding in primordial ribozyme function and more generally in the emergence of RNA-based cellular life. ■ INTRODUCTION The catalytic repertoire of RNA lies at the foundation of the RNA world hypothesis, which posits that early life used RNA as both the genetic material and enzymes (ribozymes).1 The ability of single-stranded RNA molecules to assume a wide range of folded structures endows them with functions such as molecular recognition and catalysis, suggesting that folded RNA structures would have been essential to early life. RNA assembly processes (ligation and polymerization) that generate complex folded RNA structures were therefore likely to have played an important role in the propagation and evolution of the earliest living cells. Ribozymes usually require divalent cations like Mg2+ to access their functional folds and perform catalysis. Mg2+ facilitates RNA folding by partially neutralizing the negatively charged RNA backbone and often participates in catalytic interactions within the ribozyme active site.2−4 to RNA function, Mg2+ can also be Although essential detrimental. Mg2+ catalyzes RNA backbone hydrolysis, thereby disrupting functional the structures. hydrolysis of intrinsically reactive RNA building blocks such as phosphorimidazolides that would have been important for primordial RNA assembly.5−7 Additionally, Mg2+ is generally detrimental to the integrity of prebiotic cell-like compartments bounded by fatty acids, which are commonly used models of primordial cell membranes. This incompatibility between It also accelerates ribozyme function and the stability of protocell membranes poses a significant challenge for efficient RNA catalysis within fatty acid protocells.5 RNA assembly would have driven primordial genetics and generated the catalytic diversity required to sustain RNA-based primordial life; therefore, ribozymes that catalyze RNA ligation or polymerization were crucial to primordial biology. Such ribozymes have been identified through in vitro evolution.5 Ligase and polymerase ribozymes that use 5′-triphosphorylated oligoribonucleotides and nucleoside triphosphates as sub- respectively, exhibit high Mg2+ requirements. For strates, example, the Mg2+ concentration at which the half-maximum ligation rate was achieved, [Mg2+]1/2, of the first of its kind, class I ligase is 70−100 mM,8 and polymerase ribozymes derived from the class I ligase have an optimal [Mg2+] of ∼200 mM.9,10 We previously reported ribozymes that catalyze the ligation of RNA oligomers 5′ activated with a prebiotically plausible, 2-aminoimidazole (2AI) moiety. 2AI-activated RNA Received: May 1, 2023 Published: August 3, 2023 © 2023 The Authors. Published by American Chemical Society 1670 https://doi.org/10.1021/acscentsci.3c00547 ACS Cent. Sci. 2023, 9, 1670−1678 ACS Central Science http://pubs.acs.org/journal/acscii Article Figure 1. Stimulation of ribozyme activity of ligase 1 at 1−2 mM Mg2+ in the presence of ethylene glycol and PEGs. (A) Schematic of ribozyme- catalyzed ligation of a 2-aminoimidazole-activated RNA substrate. (B) Catalytic ligation is undetectable at 1 mM Mg2+ in a solution without any crowder but is rescued in crowded solutions. (C) Ligation yields after 3 h in the absence and presence of crowding agents at the indicated concentrations. (D) Ligation rates in the absence and presence of crowding agents. Ligation reactions contained 1 μM ribozyme, 1.2 μM RNA template, and 2 μM 2-AI-activated RNA substrate in 100 mM Tris-HCl pH 8.0 and the indicated concentrations of MgCl2. Reactions contained additives (EG, PEG 200−8000) as indicated. None indicates the absence of crowders. monomers/oligomers are useful substrates for nonenzymatic RNA assembly; therefore, these “2AI-ligase” ribozymes provide continuity between chemical and enzymatic RNA ligation.6 low Mg2+ Most of these 2AI-ligases were inefficient at concentrations (<4 mM); however, we identified a single ligase sequence that had a significantly lower Mg2+ requirement ([Mg2+]1/2 ≈ 0.9 mM).11 Although ribozymes with reduced Mg2+ requirements clearly exist, they are apparently relatively uncommon in the RNA sequence space. We have therefore searched for a more general solution that would have enabled ribozymes to operate in low-Mg2+ environments such as freshwater ponds or within protocells bounded by prebiotic fatty acids.12 Mechanisms that stimulate ribozyme activity at low [Mg2+] would lower the evolutionary threshold for the emergence of such molecules in the RNA world. Although ribozymes usually require moderate to high Mg2+ concentrations to function in vitro, naturally occurring ribozymes have evolved to function in the presence of 0.5−1 mM free Mg2+ within cellular environments.13 This lower Mg2+ requirement is thought to be a consequence of the crowded cellular environment. In addition to cellular structures like organelles, the intracellular milieu is crowded with molecules that range from biopolymers like nucleic acids and proteins to smaller molecules including amino acids, nucleotides, sugars, amines, and alcohols, which collectively occupy up to 30% of the cellular volume.13,14 The presence of these molecules introduces a variety of physical and chemical forces that alter the properties of cellular RNAs.15−17 Volume excluded by macromolecules decreases the conformational entropy of unfolded RNA (an effect commonly referred to as “macro- molecular crowding”) and consequently promotes RNA folding and RNA function. Unfavorable interactions between the solvent-exposed RNA backbone and low-MW species in the cellular milieu also induce folding to minimize these interactions. A decrease in dielectric constant may favor RNA− Mg2+ association due to the diminished solvation of free Mg2+, which can stimulate RNA folding and catalysis. A decrease in water activity caused by cosolutes may favor the formation of RNA folds with reduced solvent-exposed surface area that is accompanied by water release. Investigations into RNA structure and function in solutions artificially crowded with cosolutes like polyethylene glycol (PEG) have revealed favorable effects of crowding on RNA function.17 Biophysical studies using small-angle X-ray scattering (SAXS) and single-molecule Förster resonance energy transfer (smFRET) demonstrated that molecular crowding induces RNA folding. This effect is most pronounced in the low-Mg2+ regime, where folded structures are not usually predominant.18−20 Enhanced folding in crowded solutions is often reflected in modest to significant increases in catalytic rates.17 Ribozymes in the RNA world may have evolved in similarly crowded environments within either primitive cellular compartments or confined microspaces on the Earth’s surface, which may have allowed them to function at low concentrations of Mg2+.17,21 Here, we demonstrate the beneficial effects of molecular crowding on ribozyme-catalyzed RNA assembly, which includes the stimulation of ribozyme ligase activity at low millimolar concentrations of Mg2+ and the preservation of ribozyme activity under harsh reaction conditions such as alkaline pH or urea-induced denaturation. We propose that the stabilization of catalytic RNA folds in prebiotic crowded environments could provide a general means of enabling ribozyme-catalyzed RNA assembly in diverse environments including those with low availability of Mg2+. 1671 https://doi.org/10.1021/acscentsci.3c00547 ACS Cent. Sci. 2023, 9, 1670−1678 ACS Central Science http://pubs.acs.org/journal/acscii Article Figure 2. Crowding decreases the Mg2+ requirement for RNA-ligase activity. Mg2+ dependence on ligation rates of the ligase 1 ribozyme (A) in the absence of crowders and (B−D) in the presence of (B) 10% (w/v) EG, (C) 30% (w/v) PEG 200, and (D) 19% (w/v) PEG 1000. (E) Crowding agents reduce the [Mg2+]1/2 values for the rate of ribozyme ligation. Ligation reactions contained 1 μM ribozyme, 1.2 μM RNA template, and 2 μM 2-AI-activated RNA substrate in 100 mM Tris-HCl pH 8.0 and the indicated concentrations of MgCl2. Reactions contained additives (EG, PEG 200, or PEG 1000) as indicated. ■ RESULTS AND DISCUSSION Crowding Rescues RNA-Catalyzed RNA Ligation at Low Mg2+ Concentrations. To test the effect of crowding on RNA-catalyzed RNA assembly, we chose a ligase ribozyme (henceforth, ligase 1) (Figure 1A, Table S1), previously identified by in vitro selection, that catalyzes the template- directed ligation of a primer strand to a 2AI-activated oligonucleotide. This ribozyme exhibited significantly reduced product yields at Mg2+ concentrations below 4 mM.6 For example, ligation proceeded to ∼30% in 3 h at 4 mM Mg2+, but yields were reduced to 8%, 2%, and 1% at 3, 2, and 1 mM Mg2+, respectively. This ribozyme exhibits a corresponding reduction in activity in the low-Mg2+ regime with only 5−15- fold rate enhancement over background at 2−3 mM Mg2+ compared to the ∼300-fold enhancement observed at 10 mM Mg2+.6 We used polyethylene glycol (PEG) to generate a crowded environment in vitro. PEG is chemically inert and available in a wide range of MWs, which allowed us to simulate the presence of a variety of small molecules and biopolymers that could have been present in prebiotic milieus. We also included ethylene glycol (EG) in our studies in addition to PEGs of various MWs (PEG 200, PEG 400, PEG 1000, PEG 8000). EG can be synthesized abiotically22,23 and is one the larger molecules detected in interstellar medium.24,25 We first screened various concentrations of EG, PEG 200, PEG 400, PEG 1000, and PEG 8000 to identify optimal crowding conditions for ligase 1 activity in the presence of 1 mM Mg2+ and 100 mM Tris-HCl, pH 8 (Figure S1). We observed remarkable ligation rescue in the presence of EG and both low- and high-MW PEGs. Ligation yield rose from barely detectable levels in the absence of crowding agents to about 20% and 50% after 3 h at 1 and 2 mM Mg2+, respectively, in the presence of 10% (w/v) EG. Similar stimulation in ligation was observed in 30% (w/v) PEG 200, 30% (w/v) PEG 400, 19% (w/v) PEG 1000, and 19% (w/v) PEG 8000 at 1 mM Mg2+ with ∼50% ligation after 3 h, which is comparable to the 60% ligation observed in solution at 10 mM Mg2+ with no crowding agents (Figure 1B and 1C). Ligation rates in the presence of crowders at low Mg2+ (from 0.7 to 1.3 h−1) were also comparable to the rate observed in the absence of crowders at 10 mM Mg2+ (∼1.5 h−1) (Figure 1D). Ligation yield decreased with an increase in the concentration of EG. This trend is different from other PEG-based crowders which exhibit better ligation at higher concentrations (Figure S1). This difference between EG and PEGs could be due to the mechanism by which these crowders effect RNA structure. EG cannot exclude significant volume due to its small size and must act through direct interactions with the RNA backbone or through solvent effects which increase the association between RNA and Mg2+. Therefore, the crowding effects observed are likely enthalpic, in contrast to the entropic contributions from PEGs, especially ones with moderate to high MWs. To understand the attenuated Mg2+ dependence of ribozyme-ligase activity, we measured ligation rates as a function of Mg2+ concentration in the presence of 10% EG (low-MW additive), 30% PEG 200 (low-MW additive), and 19% PEG 1000 (high-MW additive). [Mg2+]1/2 was signifi- cantly lowered in the presence of crowding agents (Figure 2, Figure S2), consistent with the enhanced ligation yield observed at low Mg2+ concentrations. A 3-fold reduction in [Mg2+]1/2 was observed in 10% EG, while 30% PEG 200 and 19% PEG 1000 caused a ∼10-fold reduction (Figure 2E). While all three crowders (EG, PEG 200, PEG 1000) supported ligase 1 activity at lower concentrations of Mg2+, maximal rates were achieved at submillimolar Mg2+ with PEG 200 and PEG 1000 and at ∼2 mM Mg2+ with EG. Because EG shows optimal activity at 2 mM Mg2+, all experiments with EG (except for the screening experiment in Figure S1 and the ligation experiment at 55 °C) were performed at 2 mM Mg2+. Previous studies have found a decrease in Mg2+ requirement for ribozyme activity to accompany a decrease in Mg2+ requirement for folding in both the group II intron20 and 1672 https://doi.org/10.1021/acscentsci.3c00547 ACS Cent. Sci. 2023, 9, 1670−1678 ACS Central Science http://pubs.acs.org/journal/acscii Article Figure 3. Molecular crowding counteracts loss of ligase 3 ribozyme activity under denaturing conditions. Crowding rescues the loss of ligation activity induced by (A) molar concentrations of urea and (B) alkaline pH. Ligation reactions contained 1 μM ribozyme, 1.2 μM RNA template, and 2 μM 2-AI-activated RNA substrate in 100 mM Tris-HCl pH 8.0, 1 mM MgCl2. Reactions contained additives (PEG 200, PEG 1000, or PEG 8000) and urea (1 or 2.5 M) as indicated. the HDV26 ribozymes, supporting a role of crowding in facilitating the formation of catalytically relevant folds. An alternative explanation to the induction of RNA folding is that the addition of cosolutes like PEG may alter solution properties such as dielectric constant and water activity which result in greater association between RNA and Mg2+. An RNA-bound Mg2+ ion may activate the nucleophile at the site of ligation or stabilize the transition state.27 We tested nonenzymatic ligation in the presence of EG and various PEGs at 1 mM Mg2+ using a FAM-labeled primer (corresponding to the 3′ end sequence of the ligase downstream of the linker, Table S1), the 2AI-activated RNA substrate, and an appropriate RNA template. Ligation yields were unaffected in the presence of EG or PEGs (Figure S3), supporting the importance of ribozyme structure in crowding-induced rate enhancement. To ask if the crowding-induced stimulation of ribozyme- catalyzed phosphorimidazolide ligation was specific to ligase 1 or was more general, we tested the activity of another ligase ribozyme (henceforth, ligase 2) identified from our previous in vitro selection experiment (Table S1).6 Although distinct in sequence and structure, ligase 2 exhibited a similar response to crowding as ligase 1. While ∼6% ligation was observed in the absence of crowding agents after 3 h at 1 mM Mg2+, crowding increased ligation yields up to ∼60%, which was comparable to the ligation yield at 10 mM Mg2+ in the absence of crowding agents. The rates of ligase 2-catalyzed ligation followed a similar trend (Figure S4). Crowding Protects Ligase Ribozyme from Denatura- tion. Since crowding promotes the formation of compact RNA folds, we wondered if molecular crowding could protect ribozymes from unfolding under denaturing conditions at the low Mg2+ concentrations that are compatible with fatty acid- based protocell membranes. As ligase 1 and ligase 2 are inactive at low Mg2+ in the absence of crowding agents, these ribozymes cannot be used to capture the detrimental effects of denaturants or the protective effects of crowding in the presence of denaturants under these low-Mg2+ conditions. Therefore, we used a previously reported 2AI-ligase (hence- forth, ligase 3) that is functional under these conditions for the following experiments.11 First, we tested RNA ligation by ligase 3 in the presence of urea, which is an effective denaturant of RNA and also an important precursor molecule in the prebiotic syntheses of ribonucleotides and amino acids.28 As ligation rates in the background of 1 mM Mg2+ expected, respectively, decreased by ∼6-fold in the presence of 1 M urea, and ligation was further reduced in the presence of 2.5 M urea (Figure 3A, Figure S5A). Next, we tested the stabilizing effects of PEG 200, a low-MW crowder, and PEG 1000, a high-MW crowder, in the presence of urea. Ligation in 1 M urea was restored upon addition of 30% PEG 200 and 19% PEG 1000 (Figure 3A). Ligase 3 was even active in 2.5 M urea in the presence of PEG 200 and PEG 1000 with rate enhancements of 25-fold and 33- fold, relative to solutions without crowding agents. Interestingly, EG did not show any ligation rescue under these partially denaturing conditions (Figure S5B). Ribozyme activity in the presence of molar concentrations of urea is consistent with the stabilization of compact, solvent- excluded RNA tertiary structures by crowding agents.26 We suggest that polymeric crowders such as polypeptides or polyesters or even “proto-peptides” such as depsipeptides that contain a mixture of amide and ester linkages, if present in sufficient concentrations in prebiotic environments, could have shielded catalytic RNA structures from nonspecific denatura- tion by molecules such as urea and formamide.29 Alkaline pH, which can be beneficial for certain prebiotic processes such as the synthesis of sugars28 and RNA strand separation,30 is detrimental to the chemical stability of RNA. However, compact folded RNAs are more resistant to alkaline degradation than their unfolded counterparts. Encouraged by the protective effect of crowding in the presence of urea, we measured the activity of ligase 3 at pH 10 and pH 11 in crowded solutions. No ligation was observed at pH 11 in the presence or absence of crowders. A small amount of ligated product was detected at pH 10 in the absence of crowding agents with an 11-fold reduction in reaction rate relative to that at pH 8 (kobs values of 0.1 h−1 at pH 10 vs 1.1 h−1 at pH 8). We tested ligation at pH 10 with different crowders. Low-MW crowders like EG and PEG 200 showed no benefit; however, the loss of ligase activity at pH 10 was less pronounced in the presence of high-MW crowders PEG 1000 and PEG 8000 with only a 2.6-fold and 2.3-fold reduction in kobs, respectively, relative to their values at pH 8 (Figure 3B, Figure S6A). This represents a 4−5-fold rate enhancement ribozyme- catalyzed ligation at pH 10 upon crowding (Figure 3B). Although ligase activity was rescued in the presence of crowders, crowding had a minimal effect on the extent of RNA degradation at pH 10 or pH 11. Therefore, the beneficial effect of crowders may result from the protection of the catalytic fold from disruption at alkaline pH or by preserving for 1673 https://doi.org/10.1021/acscentsci.3c00547 ACS Cent. Sci. 2023, 9, 1670−1678 ACS Central Science http://pubs.acs.org/journal/acscii Article Figure 4. RNA ligation catalyzed by the ligase 1 ribozyme is stimulated in the presence of prebiotic molecules. (A) Ligation yields after 3 h at 1 mM Mg2+ in the presence of ribose and prebiotic amino acids. (B) Ligation rates at 1 mM Mg2+ in the presence of ribose and prebiotic amino acids. Ligation reactions contained 1 μM ribozyme, 1.2 μM RNA template, and 2 μM 2-AI-activated RNA substrate in 100 mM Tris-HCl pH 8.0 and 1 mM or 10 mM MgCl2. Reactions contained additives (3.8% (w/v) D-ribose, 2.5 mM individual amino acid, or 2.5 mM amino acid mixture) as indicated. base-pairing interactions between the substrate, template, and ribozyme. We asked whether crowding could have had a similar protective function during fluctuating temperature cycles on the early Earth. Only a modest enhancement in ligation rates by ligase 3 was observed at 55 °C in the presence of EG, PEG 400, PEG 1000, and PEG 8000 (Figure S6B and S6C). The lack of substantial benefit from crowding at high temperatures is consistent with UV melting experiments with the ligase 1 ribozyme, which revealed a negligible increase (ΔTm = 0.5 °C) in its thermal stability in the presence of high-MW crowder, PEG 1000 (Figure S7A and S7B). EG, on the other hand, caused a 4 °C decrease in Tm (Figure S7A and S7B). A similar decrease in Tm value in the presence of EG has been observed with the hammerhead ribozyme, which we speculate could be due to a destabilization of base-paired helices.31 Crowding Stimulates RNA-Catalyzed RNA Polymer- ization at Low Mg2+ Concentrations. Although ribozymes that catalyze the template-directed polymerization of nucleo- side phosphorimidazolides have not yet been reported, polymerase ribozymes that use NTPs as substrates have been evolved from the class I ligase ribozyme.9,10,32 These ribozymes generally require 50−200 mM Mg2+, which makes them incompatible with fatty acid vesicle-based models for primitive cells. Tagami et al. demonstrated modest polymerase function at 10 mM free Mg2+ in the presence of lysine decapeptide (K10), which enabled RNA-catalyzed RNA polymerization within Mg2+-resistant 1-palmitoyl-2-oleoylphosphatidylcholine (POPC) vesicles.33 Similarly, Takahashi et al. demonstrated the addition of up to 5 nucleotides by the tC9Y polymerase ribozyme in the presence of 10 mM Mg2+ upon addition of 20% PEG 200.34 We tested the ability of the 38−6 polymerase ribozyme10 to extend a 10 nt RNA primer on a 21 nt RNA template in the presence of 5 mM Mg2+ in solutions containing low- or high-MW PEGs. Negligible extension beyond +4 was observed in the absence of crowding agents; however, small amounts of full-length products (+11) were detected in the presence of PEG 200 or PEG 1000 after 24 h. The prominent +1 extension product increased from 24% without crowding agents to 33% and 40% in the presence of PEG 200 and PEG 1000, respectively. While only 26% of the primer was extended in the absence of crowding agents, 37% and 43% of the primer was extended in the presence of PEG 200 and PEG 1000, respectively (Figure S8). Enhancement of ribozyme polymer- ase activity at low millimolar Mg2+ underscores the generality of the beneficial effects of crowded environments on ribozyme- catalyzed RNA assembly. Interestingly, molecular crowding has also been found to enhance the polymerization of NTPs35 and dNTPs36 by biologically derived protein polymerases, which further supports the role of crowding in facilitating nucleic acid assembly. Prebiotically Relevant Small Molecules Enable Ribo- zyme-Catalyzed RNA Ligation at Low Mg2+. While our observations on the effects of molecular crowding agents on ribozyme activity are promising, the above results were obtained with prebiotically irrelevant synthetic PEG molecules with the exception of EG. Therefore, we explored the potential of prebiotically relevant small molecules for stimulating ribozyme-ligase activity. Considering the importance of simple sugars in a pre-RNA/RNA world and the stabilizing effect of ribose on fatty acid membranes, we decided to explore the effect of ribose on ligase 1 ribozyme activity.28,37,38 We also tested a subset of amino acids thought to be available on early Earth as products of prebiotic synthetic pathways such as the cyanosulfidic protometabolic reaction network.39 Ribose at 2% (w/v) and 3.8% (w/v) increased ligation yield from ∼1% to ∼11% and ∼26% after 3 h in the presence of 1 mM Mg2+ with kobs values of ∼0.4 and ∼0.5 h−1, respectively (Figure S9, Figure 4). We also screened the amino acids glycine, alanine, proline, leucine, serine, and aspartic acid at 2.5, 5, 10, and 20 mM concentrations for their ability to stimulate ligation at 1 mM Mg2+. All of the above amino acids were found to stimulate ligation regardless of their concentrations with yields of 25−35% after 3 h (Figure S10). As lower concentrations are prebiotically more likely in most microenvironments, we measured the yield and rate of ligase 1-catalyzed RNA ligation in the presence of 2.5 mM of each amino acid and a mixture of all six amino acids at a total concentration of 2.5 mM (Figure 4). The presence of amino acids both individually and as a mixture rescued ligation rates to within a factor of 1.4−3.8 of that observed at 10 mM Mg2+ without any additive (Figure 4B). As ligase 1 exhibits negligible ligation at 1 mM Mg2+ even in the presence of high concentrations of Na+ (300 mM),6 low in reactions concentrations of monovalent counterions containing 2.5−20 mM amino acids are unlikely to cause 1674 https://doi.org/10.1021/acscentsci.3c00547 ACS Cent. Sci. 2023, 9, 1670−1678 ACS Central Science http://pubs.acs.org/journal/acscii Article this pronounced rate stimulation, and the amino acids must be playing a direct role. The mechanism for ribozyme activation at low Mg2+ by ribose or amino acids is not clear. Aliphatic alcohols such as methanol, ethanol, propanol, 2-methoxyethanol, and propane- 1,3-diol stimulate hammerhead catalysis at 1 mM Mg2+ by decreasing the dielectric constant of the solution, thereby enhancing interactions between the ribozyme and Mg2+.31 Ribose-mediated enhancement of ribozyme-catalyzed RNA ligation could be a result of similar solution-level effects. The beneficial effect of amino acids toward ribozyme activity has It was been previously observed for RNA self-cleavage. proposed that the increase in ribozyme activity resulted from structural compaction of the RNA, which allowed greater sampling of its catalytic fold.40 This assertion was supported by thermal denaturation and SAXS studies. Amino acids may stimulate RNA folding by altering solvent properties like dielectric constant or water activity.17 Additionally, as amino acids can weakly chelate Mg2+, the chelated amino acids may form a layer on the RNA surface, increasing the local concentration of Mg2+, which may lead to improved folding and catalysis.40,41 Regardless, this ability of prebiotic small molecules to facilitate ribozyme-catalyzed RNA assembly presents a “systems”-level solution for lowering the Mg2+ requirement for this central process in primordial biochemis- try. ■ CONCLUSION results the emergence of RNA-based cellular The crucial role of Mg2+ in both nonenzymatic- and ribozyme- catalyzed RNA replication coupled with its ability to accelerate RNA degradation and destabilize fatty acid protocells presents a puzzle for life. Therefore, exploring scenarios that mitigate this “Mg2+ problem” is of critical importance. The low Mg2+ requirement for natural ribozymes that function within crowded cellular environments inspired us to study molecular crowding as a general solution to the Mg2+ problem in the context of ribozyme-catalyzed RNA assembly. Our show a dramatic stimulation of ribozyme-catalyzed assembly of 2AI- activated RNA oligomers and nucleoside triphosphates at low millimolar Mg2+ by prebiotically relevant amino acids, ribose, ethylene glycol, and polyethylene glycols of various MWs (200−8000). The beneficial effects of amino acids, ribose, and ethylene glycol are especially notable since these molecules can be synthesized abiotically and therefore were likely to have been present in early Earth environments. The 3−10-fold lower Mg2+ requirement for ligase ribozymes in the presence of such solutes likely stems from enhanced RNA folding in “crowded” solutions as the corresponding nonenzymatic ligation reaction was not affected by crowding. Stimulation of catalytic activity in the presence of molecular crowding has been reported for other ribozymes.17 Since the crowding- induced enhancement of RNA assembly was largely independent of crowder size, ribozyme folding could be favored by an interplay of both enthalpic forces arising from interactions between the RNA surface and the crowder and entropic forces arising from volume exclusion.17 In most cases where both low-MW and high-MW crowders affect macro- molecular function, it is extremely difficult to delineate the individual contributions of volume exclusion and the various enthalpic forces that are always at play.17 Further studies may help isolate the effects of these distinct thermodynamic forces. We demonstrated that in addition to enabling RNA ligation in the low-Mg2+ regime, crowding offers modest to significant protection to ligase ribozymes under various denaturing conditions relevant to early Earth environments. The ability to function under conditions that favor the disruption of RNA secondary structure could have been important for rapid RNA- catalyzed RNA replication, which requires the separation of newly synthesized RNA strands from their RNA templates while preserving catalytic RNA structures. Efficient RNA assembly, at low Mg2+ concentrations, presents a path to reconcile ribozyme function with the stability of protocell membranes made of fatty acids. Protocells crowded with prebiotic small molecules like sugars, alcohols, and amines and polymeric species such as short oligonucleo- tides or polypeptides could potentially support a wide range of the low Mg2+ concentrations ribozyme activities under is required for maintaining membrane integrity. This particularly interesting in the context of our earlier observation that prebiotically relevant small molecules including ribose also reduce RNA leakage from fatty acid vesicles.11 The combined effect of enhancing ribozyme function under low Mg2+ conditions and stabilizing protocell membranes against Mg2+ suggests a potential role for these prebiotic molecules that is separate from their roles as components of the building blocks of life. By providing a general mechanism to activate RNA low Mg2+ concentration, molecular crowding catalysis at expands the range of environments in which ribozymes can function to less salty environments such as freshwater bodies12 and increases the likelihood of the emergence of active ribozymes from the RNA sequence space.21 Suboptimal sequences that would otherwise not be selected in low-Mg2+ environments could emerge in crowded milieus, potentially creating neutral mutational pathways that would facilitate ribozyme evolution and therefore increase the catalytic diversity of the RNA world. ■ EXPERIMENTAL PROCEDURES the 3′ end of RNA Preparation and Substrate Activation. Ribozymes were prepared by in vitro transcription of PCR-generated dsDNA templates containing 2′-O-methyl modifications to reduce transcriptional heterogeneity at the RNA42 (Table S1). Transcription reactions contained 40 mM Tris-HCl (pH 8), 2 mM spermidine, 10 mM NaCl, 25 mM MgCl2, 10 mM dithiothreitol (DTT), 30 U/mL RNase inhibitor murine (NEB), 2.5 U/mL thermostable inorganic pyrophosphatase (TIPPase) (NEB), a 4 mM concentration of each NTP, 30 pmol/mL DNA template, and 1U/μL T7 RNA Polymerase (NEB) and were incubated for 3 h at 37 °C. DNA template was digested by DNase I (NEB) treatment, and RNA was extracted with phenol−chloroform−isoamyl alcohol (PCI), ethanol precipitated, and purified by denaturing PAGE. Ligation templates, FAM-labeled primers, and ssDNA were purchased from Integrated DNA Technologies. The 5′-monophosphorylated oligonucleotide corresponding to the substrate sequence was activated by incubating it with 0.2 M 1-ethyl-3-(3 dimethylaminopropyl) carbodiimide (HCl salt) and 0.6 M 2-aminoimidazole (HCl salt, pH adjusted to 6) for 2 h at room temperature. The reaction was washed with water in Amicon Ultra spin columns (3 kDa cutoff) 4−5 times (200 μL of water per wash) and purified by reverse-phase analytical HPLC using a gradient from 98% to 75% 20 mM TEAB (triethylamine bicarbonate, pH 8) versus acetonitrile over 40 min.6 1675 https://doi.org/10.1021/acscentsci.3c00547 ACS Cent. Sci. 2023, 9, 1670−1678 ACS Central Science http://pubs.acs.org/journal/acscii Article Ligation Assays. Ligation reactions contained 1 μM ribozyme, 1.2 μM RNA template, and 2 μM 2-AI-activated RNA substrate in 100 mM Tris-HCl pH 8.0, the indicated concentrations of MgCl2, and crowding agents. All reactions were performed at room temperature unless mentioned otherwise. Aliquots were quenched with 5 volumes of quench buffer (8 M urea, 100 mM Tris-Cl, 100 mM boric acid, 100 mM EDTA) and analyzed by denaturing PAGE. Gels were stained using SYBR Gold,43 imaged on an Amersham Typhoon RGB instrument (GE Healthcare), and analyzed in Image- Quant IQTL 8.1. Intensities corresponding to the ligated product were normalized to account for the difference in size between the 95 nt precursor band and the 111 nt product band. Kinetic data were nonlinearly fitted to the modified first- order rate equation, y = A(1 − e−kx), where A represents the fraction of active complex, k is the first-order rate constant, x is time, and y is the fraction of ligated product in GraphPad Prism 9. For nonenzymatic ligation, a 5′-FAM-labeled RNA primer corresponding to the last 8 nt of the ribozyme sequence was used instead of the ribozyme, and the gel was directly imaged. Ligation Assays under Denaturing Conditions. All ligation assays under denaturing conditions were performed with the ligase 3 ribozyme, which retains activity under low [Mg2+]. Ligation at High pH. Ribozyme and template were heated at 95 °C for 2 min in the absence of any buffer and cooled to room temperature. CAPS buffer (pH 10 or 11) was added to a final concentration of 100 mM in the absence or presence of crowding agents (19% PEG 1000 or 19% PEG 8000) and 1 mM MgCl2. The substrate was added immediately after the addition of MgCl2 to initiate ligation. Ligation at High Temperatures. Reactions with or without crowding agents (10% ethylene glycol, 30% PEG 400, 19% PEG 1000, or 19% PEG 8000) were incubated at 55 °C after initiating ligation by adding the substrate. Ligation in the Presence of Urea. A 10 M concentration of urea was added to final concentrations of 1 or 2.5 M after refolding in the presence of crowding agents (30% PEG 200 or 19% PEG 1000) and 1 mM MgCl2 to minimize degradation at high temperatures required for refolding. The substrate was added immediately after the addition of MgCl2 to initiate ligation. Ribozyme-Catalyzed NTP Polymerization Assays. A FAM-labeled RNA primer (80 nM), RNA template (100 nM), and polymerase ribozyme (100 nM) were heated in the absence and presence of crowding agents and 25 mM Tris·HCl pH 8 at 80 °C for 30 s and cooled to 17 °C at a gradient of 0.1 °C/s. MgCl2 was added to final concentrations of 5 and 200 followed by a 0.5 mM concentration of each NTP. mM, Reactions were incubated at 17 °C for 24 h, and 1 μL aliquots were quenched with 7 μL of quench buffer (8 M urea, 100 mM Tris-Cl, 100 mM boric acid, 100 mM EDTA containing 5 μM DNA oligo complementary to template). Reactions were analyzed by denaturing PAGE. Gels were imaged on an Amersham Typhoon RGB instrument (GE Healthcare) and analyzed in ImageQuant IQTL 8.1. UV Melting Analysis of Ligase Ribozyme. UV melting experiments were performed to determine the thermal stability of the ligase 1 ribozyme in the absence of presence of low- and high-MW crowding agents according to the protocol used by Struslon et al.44 Briefly, 0.5 μM ribozyme was incubated at 95 °C for 2 min in 10 mM sodium cacodylate buffer (pH 7) and refolded in the absence or presence of crowding agents (10% ethylene glycol or 19% PEG 1000) in the presence of 1 mM MgCl2 by heating the solution to 55 °C for 10 min followed by cooling to room temperature for 10 min. A Cary UV−vis multicell Peltier spectrophotometer was used for melting experiments. Absorbance was recorded at 260 nm every minute between 20 and 90 °C. Data was normalized with respect to “buffer only” sample in each case, which contained all components in the experimental sample except RNA. Derivative plots of normalized data (dA/dT) vs T) and melting temperatures (Tm) were obtained by the instrument’s default software. ■ ASSOCIATED CONTENT *sı Supporting Information The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acscentsci.3c00547. Supplementary table with RNA oligonucleotides used in this work; identification of optimal crowding conditions for ligase 1-catalyzed RNA ligation; representative gels illustrating the Mg2+ dependence of ligase 1 ribozyme- catalyzed RNA ligation in the absence and presence of crowding agents; nonenzymatic ligation is not influenced by molecular crowding; ligase 2 activity at low Mg2+ concentrations is rescued by crowding agents; effect of molecular crowding on ligase 3-catalyzed RNA ligation in the presence of urea; effect of molecular crowding on ligase 3-catalyzed RNA ligation under alkaline pH and high temperature; effect of molecular crowding on the thermal stability of the ligase 1 ribozyme; RNA-catalyzed polymerization of NTPs at low Mg2+ concentration; ribozyme-catalyzed RNA ligation is stimulated in the presence of ribose; ribozyme-catalyzed RNA ligation is stimulated in the presence of prebiotically relevant amino acids (PDF) Transparent Peer Review report available (PDF) ■ AUTHOR INFORMATION Corresponding Authors Saurja DasGupta − Department of Molecular Biology, Center for Computational and Integrative Biology, Massachusetts General Hospital, Boston, Massachusetts 02114, United States; Howard Hughes Medical Institute, Massachusetts General Hospital, Boston, Massachusetts 02114, United States; Department of Genetics, Harvard Medical School, Boston, Massachusetts 02115, United States; 0000-0002-9064-9131; Email: dasgupta@ molbio.mgh.harvard.edu orcid.org/ Jack W. Szostak − Department of Molecular Biology, Center for Computational and Integrative Biology, Massachusetts General Hospital, Boston, Massachusetts 02114, United States; Howard Hughes Medical Institute, Massachusetts General Hospital, Boston, Massachusetts 02114, United States; Department of Genetics, Harvard Medical School, Boston, Massachusetts 02115, United States; Department of Chemistry and Chemical Biology, Harvard University, Cambridge, Massachusetts 02138, United States; Present Address: Howard Hughes Medical Institute, Department of Chemistry, University of Chicago, Chicago, Illinois 60637, United States.; 1203; Email: jwszostak@uchicago.edu orcid.org/0000-0003-4131- 1676 https://doi.org/10.1021/acscentsci.3c00547 ACS Cent. Sci. 2023, 9, 1670−1678 http://pubs.acs.org/journal/acscii Article ACS Central Science Author Stephanie Zhang − Department of Molecular Biology, Center for Computational and Integrative Biology, Massachusetts General Hospital, Boston, Massachusetts 02114, United States; Department of Chemistry and Chemical Biology, Harvard University, Cambridge, Massachusetts 02138, United States; Present Address: Department of Pathology, Brigham and Women’s Hospital, Boston, Massachusetts 02115, United States Complete contact information is available at: https://pubs.acs.org/10.1021/acscentsci.3c00547 Notes The authors declare no competing financial interest. ■ ACKNOWLEDGMENTS J.W.S. is an Investigator of the Howard Hughes Medical Institute. This work was supported in part by a grant from the Simons Foundation (290363) to J.W.S. ■ REFERENCES (1) Gilbert, W. The RNA World. Nature 1986, 319, 618. (2) DasGupta, S.; Piccirilli, J. A. The Varkud Satellite Ribozyme: A Journey through Biochemistry, Crystallography, and Thirty-Year Computation. Acc. Chem. Res. 2021, 54 (11), 2591−2602. (3) Lee, K. Y.; Lee, B. J. Structural and Biochemical Properties of Novel Self-Cleaving Ribozymes. Molecules 2017, 22 (4), 678. (4) Ren, A.; Micura, R.; Patel, D. J. Structure-based mechanistic insights into catalysis by small self-cleaving ribozymes. Curr. Opin Chem. Biol. 2017, 41, 71−83. (5) Joyce, G. F.; Szostak, J. W. Protocells and RNA Self-Replication. Cold Spring Harb. Perspect. Biol. 2018, 10 (9), a034801. (6) Walton, T.; DasGupta, S.; Duzdevich, D.; Oh, S. S.; Szostak, J. W. In vitro selection of ribozyme ligases that use prebiotically plausible 2-aminoimidazole-activated substrates. Proc. Natl. Acad. Sci. U. S. A. 2020, 117 (11), 5741−5748. (7) AbouHaidar, M. G.; I. G. Non-enzymatic RNA Ivanov, hydrolysis promoted by the combined catalytic activity of buffers and magnesium ions. Z. Naturforsch C J. Biosci 1999, 54 (7−8), 542− 548. (8) Bartel, D. P.; Szostak, J. W. Isolation of new ribozymes from a large pool of random sequences [see comment]. Science 1993, 261 (5127), 1411−1418. (9) Attwater, J.; Wochner, A.; Holliger, P. In-ice evolution of RNA polymerase ribozyme activity. Nat. Chem. 2013, 5 (12), 1011−1018. (10) Tjhung, K. F.; Shokhirev, M. N.; Horning, D. P.; Joyce, G. F. An RNA polymerase ribozyme that synthesizes its own ancestor. Proc. Natl. Acad. Sci. U. S. A. 2020, 117 (6), 2906−2913. (11) DasGupta, S.; Zhang, S. J.; Smela, M. P.; Szostak, J. W. RNA- Catalyzed RNA Ligation within Prebiotically Plausible Model Protocells. Chem. Eur. J. 2023, No. e202301376. (12) Maurer, S. The Impact of Salts on Single Chain Amphiphile Membranes and Implications for the Location of the Origin of Life. Life (Basel) 2017, 7 (4), 44. (13) Leamy, K. A.; Assmann, S. M.; Mathews, D. H.; Bevilacqua, P. C. Bridging the gap between in vitro and in vivo RNA folding. Q. Rev. Biophys. 2016, 49, No. e10. (14) Fulton, A. B. How crowded is the cytoplasm? Cell 1982, 30 (2), 345−347. (15) Minton, A. P. The influence of macromolecular crowding and macromolecular confinement on biochemical reactions in physio- logical media. J. Biol. Chem. 2001, 276 (14), 10577−10580. (16) Zhou, H. X.; Rivas, G.; Minton, A. P. Macromolecular crowding and confinement: biochemical, biophysical, and potential physio- logical consequences. Annu. Rev. Biophys 2008, 37, 375−397. (17) DasGupta, S. Molecular crowding and RNA catalysis. Org. Biomol Chem. 2020, 18 (39), 7724−7739. (18) Dupuis, N. F.; Holmstrom, E. D.; Nesbitt, D. J. Molecular- crowding effects on single-molecule RNA folding/unfolding thermo- dynamics and kinetics. Proc. Natl. Acad. Sci. U. S. A. 2014, 111 (23), 8464−8469. (19) Paudel, B. P.; Rueda, D. Molecular crowding accelerates ribozyme docking and catalysis. J. Am. Chem. Soc. 2014, 136 (48), 16700−16703. (20) Paudel, B. P.; Fiorini, E.; Borner, R.; Sigel, R. K. O.; Rueda, D. S. Optimal molecular crowding accelerates group II intron folding and maximizes catalysis. Proc. Natl. Acad. Sci. U. S. A. 2018, 115 (47), 11917−11922. (21) Saha, R.; Pohorille, A.; Chen, I. A. Molecular crowding and early evolution. Orig Life Evol Biosph 2014, 44 (4), 319−324. (22) Karsili, T. N. V.; Fennimore, M. A.; Matsika, S. Electron- induced origins of prebiotic building blocks of sugars: mechanism of self-reactions of a methanol anion dimer. Phys. Chem. Chem. Phys. 2018, 20 (18), 12599−12607. (23) Liu, Z.; Wu, L. F.; Kufner, C. L.; Sasselov, D. D.; Fischer, W. W.; Sutherland, J. D. Prebiotic photoredox synthesis from carbon dioxide and sulfite. Nat. Chem. 2021, 13 (11), 1126−1132. (24) Rivilla, V. M.; Beltrán, M. T.; Cesaroni, R.; Fontani, F.; Codella, C.; Zhang, Q. Formation of ethylene glycol and other complex organic molecules in star-forming regions. Astron. Astrophys. 2017, 598, A59. (25) Fedoseev, G.; Cuppen, H. M.; Ioppolo, S.; Lamberts, T.; Linnartz, H. Experimental evidence for glycolaldehyde and ethylene glycol formation by surface hydrogenation of CO molecules under dense molecular cloud conditions. Mon. Not. R. Astron. Soc. 2015, 448 (2), 1288−1297. (26) Strulson, C. A.; Yennawar, N. H.; Rambo, R. P.; Bevilacqua, P. C. Molecular crowding favors reactivity of a human ribozyme under ionic conditions. Biochemistry 2013, 52 (46), 8187− physiological 8197. (27) Shechner, D. M.; Grant, R. A.; Bagby, S. C.; Koldobskaya, Y.; Piccirilli, J. A.; Bartel, D. P. Crystal structure of the catalytic core of an RNA-polymerase ribozyme. Science 2009, 326 (5957), 1271−1275. (28) Yadav, M.; Kumar, R.; Krishnamurthy, R. Chemistry of Abiotic Nucleotide Synthesis. Chem. Rev. 2020, 120 (11), 4766−4805. (29) Frenkel-Pinter, M.; Haynes, J. W.; Mohyeldin, A. M.; C, M.; Sargon, A. B.; Petrov, A. S.; Krishnamurthy, R.; Hud, N. V.; Williams, L. D.; Leman, L. J. Mutually stabilizing interactions between proto- peptides and RNA. Nat. Commun. 2020, 11 (1), 3137. (30) Mariani, A.; Bonfio, C.; Johnson, C. M.; Sutherland, J. D. pH- Driven RNA Strand Separation under Prebiotically Plausible Conditions. Biochemistry 2018, 57 (45), 6382−6386. (31) Nakano, S.; Kitagawa, Y.; Yamashita, H.; Miyoshi, D.; Sugimoto, N. Effects of Cosolvents on the Folding and Catalytic Activities of the Hammerhead Ribozyme. Chembiochem 2015, 16 (12), 1803−1810. (32) Horning, D. P.; Joyce, G. F. Amplification of RNA by an RNA polymerase ribozyme. Proc. Natl. Acad. Sci. U. S. A. 2016, 113 (35), 9786−9791. (33) Tagami, S.; Attwater, J.; Holliger, P. Simple peptides derived from the ribosomal core potentiate RNA polymerase ribozyme function. Nat. Chem. 2017, 9 (4), 325−332. (34) Takahashi, S.; Okura, H.; Sugimoto, N. Bisubstrate Function of RNA Polymerases Triggered by Molecular Crowding Conditions. Biochemistry 2019, 58 (8), 1081−1093. (35) Takahashi, S.; Okura, H.; Chilka, P.; Ghosh, S.; Sugimoto, N. Molecular crowding induces primer extension by RNA polymerase through base stacking beyond Watson-Crick rules. RSC Adv. 2020, 10 (55), 33052−33058. (36) Takahashi, S.; Herdwijn, P.; Sugimoto, N. Effect of Molecular Crowding on DNA Polymerase Reactions along Unnatural DNA Templates. Molecules 2020, 25 (18), 4120. (37) Furukawa, Y.; Chikaraishi, Y.; Ohkouchi, N.; Ogawa, N. O.; Glavin, D. P.; Dworkin, J. P.; Abe, C.; Nakamura, T. Extraterrestrial 1677 https://doi.org/10.1021/acscentsci.3c00547 ACS Cent. Sci. 2023, 9, 1670−1678 ACS Central Science http://pubs.acs.org/journal/acscii Article ribose and other sugars in primitive meteorites. Proc. Natl. Acad. Sci. U. S. A. 2019, 116 (49), 24440−24445. (38) Meinert, C.; Myrgorodska, I.; de Marcellus, P.; Buhse, T.; Nahon, L.; Hoffmann, S. V.; d’Hendecourt, L. L. S.; Meierhenrich, U. J. Ribose and related sugars from ultraviolet irradiation of interstellar ice analogs. Science 2016, 352 (6282), 208−212. (39) Wu, L. F.; Sutherland, J. D. Provisioning the origin and early evolution of life. Emerg Top Life Sci. 2019, 3 (5), 459−468. (40) Yamagami, R.; Bingaman, J. L.; Frankel, E. A.; Bevilacqua, P. C. Cellular conditions of weakly chelated magnesium ions strongly promote RNA stability and catalysis. Nat. Commun. 2018, 9 (1), 2149. (41) Yamagami, R.; Huang, R.; Bevilacqua, P. C. Cellular Concentrations of Nucleotide Diphosphate-Chelated Magnesium Ions Accelerate Catalysis by RNA and DNA Enzymes. Biochemistry 2019, 58 (38), 3971−3979. (42) Kao, C.; Zheng, M.; Rudisser, S. A simple and efficient method to reduce nontemplated nucleotide addition at the 3 terminus of RNAs transcribed by T7 RNA polymerase. RNA 1999, 5 (9), 1268− 1272. (43) Guillen, D.; Schievelbein, M.; Patel, K.; Jose, D.; Ouellet, J. A simple and affordable kinetic assay of nucleic acids with SYBR Gold gel staining. PLoS One 2020, 15 (3), No. e0229527. (44) Strulson, C. A.; Boyer, J. A.; Whitman, E. E.; Bevilacqua, P. C. Molecular crowders and cosolutes promote folding cooperativity of RNA under physiological ionic conditions. RNA 2014, 20 (3), 331− 347. 1678 https://doi.org/10.1021/acscentsci.3c00547 ACS Cent. Sci. 2023, 9, 1670−1678
null
10.1371_journal.pclm.0000184.pdf
journal.pclm.0000184 April 19, 2023 1 / 20 PLOS CLIMATE major data source of the study and are included in the submitted manuscript. Search terms and the extraction matrix used for the literature search to develop preliminary causal loop models are included as Supplementary material. Listing of the literature accessed and data extracted are lodged on the QMU eData repository: https://eresearch. qmu.ac.uk/handle/20.500.12289/12889.
null
RESEARCH ARTICLE Informing adaptation strategy through mapping the dynamics linking climate change, health, and other human systems: Case studies from Georgia, Lebanon, Mozambique and Costa Rica Giulia Loffreda1, Ivdity ChikovaniID Laura C. Blanco6, Liz Grant7, Alastair Ager1,8* 2, Ana O. Mocumbi3,4, Michele Kosremelli AsmarID 5, 1 Research Unit on Health in Situations of Fragility, Institute for Global Health and Development, Queen Margaret University, Edinburgh, United Kingdom, 2 Curatio International Foundation, Tbilisi, Georgia, 3 Instituto Nacional de Sau´ de, Marracuene, Mozambique, 4 Universidade Eduardo Mondlane, Maputo, Mozambique, 5 Institut Supe´rieur de Sante´ Publique, Saint Joseph University of Beirut, Beirut, Lebanon, 6 School of Economics, Universidad de Costa Rica, San Jose´, Costa Rica, 7 Global Health Academy, University of Edinburgh, Edinburgh, United Kingdom, 8 Department of Population and Family Health, Columbia Mailman School of Public Health, New York, NY, United States of America * alastair.ager@gmail.com Abstract While scientific research supporting mitigation of further global temperature rise remains a major priority, CoP26 and CoP27 saw increased recognition of the importance of research that informs adaptation to irreversible changes in climate and the increasing threats of extreme weather events. Such work is inevitably and appropriately contextual, but efforts to generalise principles that inform local strategies for adaptation and resilience are likely cru- cial. Systems approaches are particularly promising in this regard. This study adopted a sys- tem dynamics framing to consider linkages between climate change and population health across four low- and middle-income country settings with a view to identifying priority inter- sectoral adaptation measures in each. On the basis of a focused literature review in each setting, we developed preliminary causal loop diagrams (CLD) addressing dynamics operat- ing in Mozambique, Lebanon, Costa Rica, and Georgia. Participatory workshops in each setting convened technical experts from different disciplines to review and refine this causal loop analysis, and identify key drivers and leverage points for adaptation strategy. While analyses reflected the unique dynamics of each setting, common leverage points were iden- tified across sites. These comprised: i) early warning/preparedness regarding extreme events (thus mitigating risk exposure); ii) adapted agricultural practices (to sustain food security and community livelihoods in changing environmental conditions); iii) urban plan- ning (to strengthen the quality of housing and infrastructure and thus reduce population exposure to risks); iv) health systems resilience (to maintain access to quality healthcare for treatment of disease associated with increased risk exposure and other conditions for which access may be disrupted by extreme events); and v) social security (supporting the liveli- hoods of vulnerable communities and enabling their access to public services, including a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Loffreda G, Chikovani I, Mocumbi AO, Asmar MK, Blanco LC, Grant L, et al. (2023) Informing adaptation strategy through mapping the dynamics linking climate change, health, and other human systems: Case studies from Georgia, Lebanon, Mozambique and Costa Rica. PLOS Clim 2(4): e0000184. https://doi.org/10.1371/journal. pclm.0000184 Editor: Shouro Dasgupta, Centro Euro- Mediterraneo sui Cambiamenti Climatici, ITALY Received: November 12, 2022 Accepted: March 20, 2023 Published: April 19, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pclm.0000184 Copyright: © 2023 Loffreda et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: Causal loop diagrams refined during workshop discussion comprise the PLOS Climate | https://doi.org/10.1371/journal.pclm.0000184 April 19, 2023 1 / 20 PLOS CLIMATE major data source of the study and are included in the submitted manuscript. Search terms and the extraction matrix used for the literature search to develop preliminary causal loop models are included as Supplementary material. Listing of the literature accessed and data extracted are lodged on the QMU eData repository: https://eresearch. qmu.ac.uk/handle/20.500.12289/12889. Funding: This work was supported by an NIHR grant (16/136/100) to AA (PI) and through a CoP26 International Climate Change Network award by the Royal Society of Edinburgh (RSE) to AA and LG for the work of the Research Unit on Heath in Fragility at Queen Margaret University, Edinburgh. This grant funded GL’s role as research coordinator of the project. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Informing adaptation strategy through mapping the dynamics of linked systems healthcare). System dynamics modelling methods can provide a valuable mechanism for convening actors across multiple sectors to consider the development of adaptation strategies. Introduction Of all nations, low- and middle-income countries (LMIC) face the severest consequences of the climate crisis, despite having contributed the least to its occurrence [1, 2]. Climate change significantly threatens the major health gains witnessed across these settings over recent decades. Established direct and indirect pathways of influence [3] include: floods, increasing risk of water-borne disease; diminishing freshwater availability, eroding food security and san- itation; changes in temperature and rainfall impacting habitats and thus the spread of zoonotic and vector-borne disease; air pollution impacting pulmonary health and lung functions; land degradation and deforestation driving food insecurity and undernutrition; and environmental change compromising mental health [4]. Critically, highly inequitable, inefficient, and unsus- tainable patterns of resource consumption and technological development, together with pop- ulation growth, exacerbate these risks [5]. Addressing these pathways therefore requires an understanding of their interaction and linkage. Adaptation and resilience measures are actions to accommodate environmental changes anticipated as a result of projected increases in global temperature, complementing mitigation actions seeking to reduce drivers of further temperature increase (centrally through reduction of carbon emissions). Resilience, a crucial theme within environmental research, has also emerged as a central concept in the health systems literature [6]. Reflecting a broader engagement in systems thinking [4, 7], research in this field has come to increasingly focus on identifying system capacities for absorption, adaptation, and transformation developed from system dynamic analyses [8, 9]. In a similar fashion, the planetary health education framework highlights the importance of using system dynamics to understand how different factors inter- act as part of a complex system [10]. Adaptation and resilience became focal points for CoP26: the Glasgow Climate Pact agreed by 197 countries at its conclusion set out a way forward from the 2015 Paris Agreement, emphasising the urgency of scaling up action and support to enhance adaptive capacity, strengthen resilience and reduce vulnerability [11]. The launch of the Sharm-El-Sheikh Adaptation Agenda at CoP27 then outlined thirty adaptation outcomes which can enhance resilience for up to 4 billion people living in the most climate vulnerable communities by 2030 [12]. Steps were taken to initiate a Loss and Damage Fund to pay for climate related damage for vulnerable nations made increasingly vulnerable because of the rapidity of climate related adverse events, with a Transitional Committee set up to provide recommendations on types of financing, levels of vulnerability and what the fund should cover. Understanding the intercon- nected nature of loss (and the amplification of different losses, such as economic impacts on the loss of cultural heritage, or habitable land) will require new data and new tools to interpret this. Systems science has strong potential in this regard [5, 7, 10, 13]. This study addressed the linkages between climate change and health, by adopting a case study approach drawing on system science. The aim was to map the complex dynamics between climate change and population health across four settings linked to the Research Unit on Health in Situations of Fragility (RUHF) network [14]. By making more explicit the interre- lationships between the factors shaping climate and health in each context the aim was to PLOS Climate | https://doi.org/10.1371/journal.pclm.0000184 April 19, 2023 2 / 20 PLOS CLIMATE Informing adaptation strategy through mapping the dynamics of linked systems identify key entry-points and pathways for targeted adaptation and resilience measures. While other studies have used system dynamics to explore the dynamics of climate change and health, they have used it for specific settings or diseases [13]. To our knowledge, this is the first study to use a comparative case study design across different settings and consider the role of broader socio-political systems in connecting climate events and human health. Methods Theoretical framework The study adopted a socio-ecological and political ecology approach. The emerging field of planetary health reinforces the importance of the interconnections between environmental and human health and the relevance of considering these to formulate feasible solutions to the complex challenges of climate change [10]. We also drew on system thinking to better under- stand the non-linear relationships that exist among the complex systems under study and to address key adaptation and resilience measures. Research design We conducted case studies with partners in four low- and middle-income countries (LMICs): Mozambique, Lebanon, Georgia, and Costa Rica. These four settings each reflect some form of fragility as reflected in current OECD definitions [15], but exhibit diverse geographical, social, and political characteristics and forms of climate vulnerability. We adopted a mixed method approach incorporating a preliminary scoping literature review followed by group-based sys- tem dynamic modelling (Fig 1). Literature review. The search strategy for the preliminary scoping literature review included key terms such as climate change, country name, and adaptation or resilience. We intentionally kept our search approach wide to ensure retrieval of a sample of papers from Fig 1. Flow diagram of research process used to develop case studies. https://doi.org/10.1371/journal.pclm.0000184.g001 PLOS Climate | https://doi.org/10.1371/journal.pclm.0000184 April 19, 2023 3 / 20 PLOS CLIMATE Informing adaptation strategy through mapping the dynamics of linked systems different disciplines. No timeframe restrictions were applied. We searched peer-reviewed arti- cles and grey literature both in English and Spanish (for Costa Rica) in the following databases: PubMed/Medline, Google Scholar, WHO IRIS, World Bank. Based on our pre-defined inclu- sion criteria, we identified 36 papers. An additional six papers were shared by country partners and included for data extraction. The country specific literature was complemented and triangulated with key references from the global literature to assess accuracy of information on the more general issues. We piloted, revised, and finalised an extraction matrix covering the following information: biblio- graphic information; socio-ecological factors (such as climate, political, social stressors, human health, animal health) [16]; adaptation and resilience measures proposed; political ecology factors [17]; and other themes such as gender [18]. Participatory workshops and system dynamic modelling. We collated information from this preliminary scoping literature review–separately for each country—using a causal loop seed model (see Fig 2) suggested by the work of Proust and colleagues [13]. This spatially located variables identified in the reviewed literature with respect to three core domains: the state of the earth system; human made influence/activities, and human health/wellbeing. An initial causal loop diagram (CLD) was then elaborated for each country linking geographical, socio-political, health system, disease, and extreme weather event variables on the basis of the evidence presented by the reviewed literature and the research team consultations. CLDs were developed using the software package Vensim MLE. These CLDs were then refined during online consultations with collaborators in each set- ting. The consultations involved participatory workshops with health, climate and environ- ment specialists. Each workshop lasted approximately 2.5 hours and was conducted online between July and August 2021. A total of 18 participants took part across the four workshops. Participants, selected using a snowballing approach, were predominantly academics in differ- ent fields (climate science, health, forestry, economics etc.) and all based in the countries under study. Fig 2. Seed model adopted for the development of causal loop diagrams. Adapted from Proust et al. (2012) showing five key causal linkages between the state of the earth system, human influence and activities and human health and well-being. https://doi.org/10.1371/journal.pclm.0000184.g002 PLOS Climate | https://doi.org/10.1371/journal.pclm.0000184 April 19, 2023 4 / 20 PLOS CLIMATE Informing adaptation strategy through mapping the dynamics of linked systems Workshops involved confirmation of the key variables of relevance and negotiation–on the basis of the local multidisciplinary expertise and evidence available–of the core dynamics link- ing them. Participants were asked to confirm the relevance of each variable in turn, confirm or revise its pathway of connections and suggest additional variables linked to that pathway [7]. Feedback from participants was integrated into the CLDs in an iterative manner, editing the diagram on screen. While discussions and model development followed the lead of partici- pants (that is, the sequence of addressing variables and the pathways connecting them followed the flow of discussion of system dynamics by the group), it was ensured that all pathways were scrutinised at some stage of the workshop. Once the CLD reflected the inputs of all participants, the group was invited to indicate potential leverage points for instituting adaptation measures that would impact the dynamics mapped for their setting [13]. The implications for national and local adaptation strategies were then discussed. On the completion of each workshop the CLD was finalised by the research team, utilising a recording of the session to ensure that it reliably reflected the analysis of the group in terms of the directionality of connections and their polarity (i.e. whether they acted to increase or decrease the value of a connected variable) [7, 13]. Finally, after all four workshops were completed, researchers present at each conducted an integrative analysis, comparing the four models to identify common features and potential synergies regarding adaptation strategy. This analysis was shared and revised with input from the full research team. Ethics. Ethical approval for the research was granted through the Research Ethics Panel at Queen Margaret University (QMU). Results We discuss each country case study in turn by providing a brief introduction to the setting and then presenting the emerging themes from the analysis, including adaptation strategies priori- tized. We then present an integrative analysis noting commonalities in the dynamics observed across the four settings. Georgia case study Country profile. Georgia is a post-Soviet upper-middle-income country located in the South Caucasus, a region characterised by instability and economic challenges [19, 20]. It is rated as moderately fragile on the Fragile State Index (FSI) [21], with progressive erosion of state legitimacy and aspects of community cohesion. The country borders Armenia and Azer- baijan (now in conflict over the disputed region of Nagorno-Karabakh), Turkey and Russia. Armed conflict in 1990 in Tskhinvali region and Abkhazia, a history of civil war, rapid market- ization and hyperinflation following independence from the Soviet Union in 1991, have left Georgia in a state of economic collapse [22]. Since 1994, policy reforms and economic growth have improved the economic situation in the country [23]; however, signs of economic stress were again observed in 2008 due to the conflict between Georgia and Russia over Tskhinvali region. While recent decades have witnessed rapid economic development, socio-economic inequalities continue to pose a challenge, with one-fifth of the population living in relative poverty. The country’s disease profile is dominated by non-communicable diseases (NCD) which account for over 97% of all deaths and comprise 9 out of 10 conditions presenting for care, with significant prevalence of circulatory diseases, cancer, diabetes and respiratory diseases [22]. While Georgia has made progress on a number of indicators, such as maternal and infant PLOS Climate | https://doi.org/10.1371/journal.pclm.0000184 April 19, 2023 5 / 20 PLOS CLIMATE Informing adaptation strategy through mapping the dynamics of linked systems mortality, others remain above the regional average, with multi-drug resistant tuberculous (TB) a continuing threat and an increasing incidence of HIV. Since 2013 the government increased public spending on health to reduce financial barriers to access and use of services. As a result, the share of out-of-pocket payment in current spend- ing on health reduced to 48% in 2018. However public spending on health remains low (2.8% of GDP) and degree of financial hardship (impoverishing and catastrophic health spending) is among the highest in the European region [24]. Topographically, Georgia is characterized by the Great Caucasus Mountains in the north and the Lesser Caucasus in the south. Georgia has many natural resources and is highly depen- dent upon tourism, both of which are highly vulnerable to climate variability and change [25]. Almost half of the population lives in rural areas. In 2015, Georgia submitted its Nationally Determined Contribution (NDC) and has pledged to reduce its Green House Gas (GHG) emissions by 15% by 2030. Georgia’s National Adaptation Plan [26] includes the healthcare sector, although a lack of data is viewed as constraining progress in implementation. Georgia’s 2030 Climate Change Strategy and Action Plan [27] explicitly seeks to integrate healthcare needs into climate change adaptation and mitigation strategy. It highlights a number of respi- ratory conditions (e.g. chronic obstructive pulmonary disease and asthma) clearly associated with climate change and high greenhouse gas emissions. Cases of infectious and parasitic dis- eases also doubled between 2008 and 2017, with the influence of changing climate again implicated. Emerging themes and strategies. Workshop participants highlighted several dynamics characterising climate impact in the country (see Fig 3). One related to air pollution and cli- mate change and how they influence each other through complex interactions in the atmo- sphere and their consequences on health. Air pollution has been directly associated with cardiovascular and pulmonary related health issues. This has received political attention and is being recognised as a research priority with health impact assessments now underway. Heatwaves are becoming more common in the country and are associated with increased mortality due to cerebrovascular events, dehydration, and other health problems. Heatwaves additionally burden health services through increased strain on water, energy, and transporta- tion resources. High temperatures also raise the levels of ozone and other pollutants in the air that exacerbate cardiovascular and respiratory disease. Food and livelihood security is also be impacted when people lose their crops or livestock due to extreme heat. Extreme weather events (such as floods) are causing coastal erosion, which impacts the live- lihoods and mental health of people living in coastal areas; coastal erosion has also led to the displacement of communities. Despite most of the Georgian population having access to improved water supplies, participants added an additional pathway in relation to the availabil- ity of water resources and sanitation, potentially at risk with projected increases in extreme weather events. For many of these pathways of impact it was observed that risks fell dispropor- tionally on lower-income households, and act to increase socio-economic and health inequali- ties in the country. In terms of adaptation, capacity building was considered to be a key requirement. In the health sector, one participant highlighted the importance of planetary health advocacy targeted to medical students and health professionals. Setting up multi-sectoral collaborations and a ‘whole-of-society-approach’ was viewed as essential for political progress on, and effective implementation of, adaptation strategies. To achieve this necessary coordination across actors and stakeholders in tackling climate change networks or institutions needed to be established connecting civil society, non-governmental organisations and academics. In terms of practical measures to strengthen resilience, discussion focused on the establishment of alerts and early warning systems to protect populations from the risks of floods and poor air quality. PLOS Climate | https://doi.org/10.1371/journal.pclm.0000184 April 19, 2023 6 / 20 PLOS CLIMATE Informing adaptation strategy through mapping the dynamics of linked systems Fig 3. Georgia causal loop analysis. Key pathways exacerbating threat considered in the workshop flagged in light blue; potential foci of adaptation and resilience shown in red. https://doi.org/10.1371/journal.pclm.0000184.g003 Mozambique case study Country profile. Mozambique is located in sub-Saharan Africa, a region exposed to gen- erally high levels of economic and environmental risk. The OECD formally classifies Mozam- bique as fragile, with several dimensions of fragility flagging concern, including environmental risk [15]. Following independence from Portugal in 1975, Mozambique experienced a long- lasting civil war which damaged the country’s infrastructure and institutions, severely limiting the state’s capacity to provide essential services [28]. The country faces many development challenges, including widespread poverty, low life expectancy, and wide gaps in educational achievement. Provision of social sector services is heavily dependent upon donor contribu- tions, which have prevented greater deterioration of wellbeing of vulnerable groups [29]. Despite sustained economic growth and improvements in socio-economic indicators in recent years, Mozambique is still one of the poorest countries in the world [30]. Tropical cyclones Idai and Kenneth, which hit the country in 2019, massively damaged infrastructure and left 2.2 million people in need urgent assistance. Environmental, security and economic risks shape both resource availability for the health system and the burden of NCD in the country [31]. While communicable diseases (including HIV/AIDS) and maternal and neonatal condi- tions remain the greatest contributors to disease burden, 15 of the top 22 causes of loss of PLOS Climate | https://doi.org/10.1371/journal.pclm.0000184 April 19, 2023 7 / 20 PLOS CLIMATE Informing adaptation strategy through mapping the dynamics of linked systems disability-adjusted life-years (DALYs) relate to NCD, notably cardiovascular disease, neo- plasms, unintentional injuries and mental health disorders [32]. Over two–thirds of the population live and work in rural areas. The country is endowed with ample arable land, water, energy, as well as newly discovered natural gas and mineral resources offshore; three, deep seaports; and a relatively large potential pool of labour. Agricul- ture remains the pillar of Mozambique’s economy, contributing 28% of GDP and employing over 81% of the workforce. The majority of the country’s agricultural production is through small-scale subsistence farming, with 95% of food production is rain-fed. Through the Ministry for Coordination of Environmental Affairs (MICOA), the Govern- ment of Mozambique developed a national climate change strategy in 2011. This targeted increased resilience in communities and the national economy and the promotion of low-car- bon development and the green economy through integrating adaptation and mitigation strat- egies across multiple sectors. The Government of Mozambique subsequently defined its climate mitigation and adaptation commitments through Mozambique’s First Nationally Determined Contribution (NDC 1) 2020–30, which came into force in 2018 when the country formally became a party to the Paris Agreement [33]. WHO and the Ministry of Health devel- oped the 2022–2025 National Health Adaption Plan for Climate Change, which reflects an integrated and multisectoral approach. This was informed by a district-by-district health vul- nerability and adaptation to climate change assessment [34], utilising the WHO-recom- mended Health Vulnerability Index. This featured projections of the impact of climate change on the incidence of malaria and diarrhoea calculated considering the scenarios of low, medium and high emissions. Emerging themes and strategies. Workshop participants addressed several dynamics linking climate change and health (see Fig 4). Key threats were identified in relation to the increased intensity and frequency of extreme weather events. Participants highlighted that water resources were a particular focus of concern with regard to both floods (influenced by La Niña, in the north) and droughts (by El Niño, in the south). During floods, large amounts of water (including from neighbour countries) strained the ability of the country to effectively manage water resources, impacting water quality and sanitation and thus population health risk from water-borne disease. This pathway had not been identified from the literature review. Population health was acknowledged to also be impacted by the influence of restricted access to health services due to flooding. Due to its low-lying topography, rising sea level is a cause of coastal erosion, impacting both biodiversity and the livelihoods of the poor populations living in coastal zones depending on fishing and agriculture. With respect to such populations, the quality of housing was con- sidered an important factor in mediating the impacts of climate change. Poor housing exposed households to much greater risks regarding health and livelihoods, and was linked to a range of factors including migration, unplanned urbanisation and dependence on biomass fuels. Current governance of the health system, constraints on the health workforce due to migra- tion and damage to infrastructure due to extreme events were all contributing to greater fragil- ity of the health system, with implications for addressing the increasing burden of both non- communicable (including mental health) and communicable disease (including emerging infections and chronic infectious disease such as HIV and TB). Discussion on adaptation strategies focused particularly on issues of water management. Monitoring and surveillance systems needed to be strengthened, particularly in the coastal areas and to anticipate flooding. Given hydrological linkages with neighbouring countries, the political security of water needed be addressed when designing water management strategies. In this regard, stronger data collection and information systems would enable and support political decision-making as well as inform locally driven strategies. Strengthening the health PLOS Climate | https://doi.org/10.1371/journal.pclm.0000184 April 19, 2023 8 / 20 PLOS CLIMATE Informing adaptation strategy through mapping the dynamics of linked systems Fig 4. Mozambique causal loop analysis. Key pathways exacerbating threat considered in the workshop flagged in light blue; potential foci of adaptation and resilience shown in red. https://doi.org/10.1371/journal.pclm.0000184.g004 system–in terms of preparedness, capacity and resilience of infrastructure–was also identified as a key focus for action if the impacts of climate change were to be moderated. Lebanon case study Country profile. Lebanon is located on the eastern basin of the Mediterranean Sea. It is a LMIC with a population of approximately 6 million people [35]. In recent years, Lebanon has witnessed political instability, sectarian division, economic crises and recurring civil unrest [36] which has affected its ability to build consensus on political issues and develop equitable and effective policies. The World Bank characterises Lebanon as exhibiting high institutional and social fragility [15]. Even before considering the significant impacts of climate change, the stressors experienced by the country are substantive, including the need to accommodate the highest number of Syrian refugees per capita post 2011 [37], progressive economic collapse precipitated by high levels of unrest and limited economic growth [38, 39], and the devastating impacts of the August 4th 2020 explosion [40]. Lebanon struggles with an increased burden of NCD (including mental health) needs, pre- cipitated by the fragility-related risks it has navigated over time. These have limited the coun- try’s capacity to deliver primary care and related NCD services through its network of primary PLOS Climate | https://doi.org/10.1371/journal.pclm.0000184 April 19, 2023 9 / 20 PLOS CLIMATE Informing adaptation strategy through mapping the dynamics of linked systems health centres [35, 41, 42]. Current circumstances underscore the need to identify effective and affordable primary care-based services which can be sustainably financed by the diverse stakeholders active in Lebanon (e.g., Ministry of Health, World Bank, and UNHCR). Dominated by mountains, 67% of the country’s total land is arable and 24% is forest and other wooded lands. The economy is dominated by the service sector, which contributes 45% of the country’s GDP. Degraded sandy soils contribute to dust and sandstorms, which are haz- ardous to both humans and livestock. Signs of water shortages are evident due to increased demand from agriculture and industry. Weak institutional structures, policies and legislations, limited access to new technologies, skills and technical resources all hamper Lebanon’s ability to address the current challenges, especially in relation to water, agriculture, forests, and man- agement of coastal areas [35]. In 2013 Lebanon identified Nationally Appropriate Mitigation Actions (NAMAs) articulat- ing voluntary emission reduction proposals, and established working groups on the transport, energy, waste, forestry, and industry sectors. Lebanon signed the Paris Agreement in 2016 and submitted an update to its initial NDC in 2020 [43]. The country’s most recent WHO Health and Climate Change Country Profile [44] particularly highlights health risks due to heat stress, food safety and security, and water quantity and quality. Associated risks due to air pollution are also noted, with recent data indicating annual mean PM2.5 levels for major cities over five times greater than the WHO guideline value of 5 μg/m3. Emerging themes and strategies. A core focus of workshop discussion was the complex dynamics related to the environment and agricultural production which mediated between cli- mate and health (see Fig 5). Such variables were not initially included in the CLD but were highlighted by participants during the workshop. Harvesting of pine nuts, for example, is one of a number of important sources of livelihood threatened by changing climatic conditions. Irrigation to sustain horticulture through changing seasonal conditions is placing a strain on insecure water sources. Extension of dairy and cattle farming to meet local demand for food supply is further taxing water resources, as well as contributing to greenhouse gas emissions. Fig 5. Lebanon causal loop analysis. Key pathways exacerbating threat considered in the workshop flagged in light blue; potential foci of adaptation and resilience shown in red. https://doi.org/10.1371/journal.pclm.0000184.g005 PLOS Climate | https://doi.org/10.1371/journal.pclm.0000184 April 19, 2023 10 / 20 PLOS CLIMATE Informing adaptation strategy through mapping the dynamics of linked systems All these dynamics impact population health (e.g., through food security or availability of water) as well as upon household livelihoods and environmental conditions. Human displacement and population pressure were other factors considered to be shaping the dynamics of climate and health. War and conflict in the region have driven a cycle of envi- ronmental degradation and population movement. The influx of refugees has exacerbated pres- sure on land, urban settlements, food and water, adding to the direct impacts of climate change. The political and economic crisis facing the country drives further dynamics eroding popula- tion health and the capacity to moderate climate impacts. There are implications for food secu- rity and the sustainability of agricultural production. Economic conditions are also restricting access to vital commodities to support the operation of the health system. Together with popula- tion displacement involving outward migration of health workers, these trends are contributing to greater fragility of the health system, with major implications for population health. Potential adaptation strategies addressed include strengthening sustainable agricultural solutions (such as climate smart agriculture, agroforestry and greater use of small ruminants such as local goats and sheep) and developing sustainable water services. Although govern- ment policy can facilitate development, given the economic and governance challenges in the country, local community-based initiatives were considered crucial. Conflict- and climate- sensitive approaches were viewed as vital to sustain access to health services enabling universal health coverage (UHC). Greater cross-sectoral collaboration is required to ensure public health safety and disaster risk reduction are integrated into national health plans. Costa Rica case study Country profile. Costa Rica, situated between Nicaragua and Panama, has moderate pov- erty rates in comparison with other states within Latin America and the Caribbean. However, fiscal challenges and increasing income inequality are persistent pressing issues [45], with the Fragile State Index (FSI) noting escalating concerns on issues of security and resource distribu- tion [15]. The country is characterised by high rates of migration from across Central America, being one of the top ten countries in the world to receive asylum requests [46]. Evidence from 2015 suggests that the average disposable income of the 10% richest households was 32 times higher than that of the poorest 10% (c.f. OECD average of 9.6) [47]. The threat of economic recession leaves the Costa Rican population open to health-related risk. While UHC is formally guaranteed, more than one-third of the assets of the Caja Costar- ricense de Seguro Social (social security and health insurance agency) are owed to it by the State [48], itself struggling to raise revenues given rapid increases in unemployment, informal employment [49] and effects of COVID-19. The country’s disease profile is dominated by a high NCD burden, typically addressed by high-cost treatments at the level of secondary care. The country has a varied topography that includes coastal plains separated by rugged mountains, including over 100 volcanic cones. Even though Costa Rica constitutes less than 0.05 percent of the total Earth surface, its habitats represent around 5 percent of the planet’s biodiversity. Costa Rica is known worldwide for its conservation efforts and is a ‘hot spot’ for eco-tourism, with more than 26 percent of its land under protection. However, due to a combination of geographic and economic factors, Costa Rica is highly vulnerable to extreme climate events and natural hazards. Part of this vulnerability is a result of the presence of populations in areas prone to volcanic eruptions and in unstable lands, degraded by widespread cattle ranching, or in poorly planned settlements prone to landslides and flooding. Costa Rica’s National Climate Change Strategy (ENCC) and its Plan of Action, as well as advances in the Framework Law on Climate Change, frame policy objectives in this area. The ENCC prioritizes action on mitigation, adaptation, technology, education and PLOS Climate | https://doi.org/10.1371/journal.pclm.0000184 April 19, 2023 11 / 20 PLOS CLIMATE Informing adaptation strategy through mapping the dynamics of linked systems Fig 6. Costa Rica causal loop analysis. Key pathways exacerbating threat considered in the workshop flagged in light blue; potential foci of adaptation and resilience shown in red. https://doi.org/10.1371/journal.pclm.0000184.g006 finance with the goal to integrate climate change policy with the long-term competitiveness of the country and a strategy of sustainable development. The National Adaptation Policy (2018– 2030), the National Decarbonization Plan (2018–2050), and the country’s NDC [50] all affirm the country priorities and commitment to tackle climate change. The National Adaptation Plan to Climate Change 2022–2026 [51] makes a clear reference to the links between climate change and health, noting marked increases in the prevalence of infectious diseases such as Zika, malaria, dengue, and chikungunya. It also emphasises the increasing vulnerabilities of indigenous communities, women, and the elderly to climate change stressors. Emerging themes and strategies. With important changes in patterns of rainfall, a major focus of discussion amongst participants were the dynamics influencing water resources, whether directly through droughts, floods and salinization of aquifers or indirectly through the impact of forestry and agricultural practices (see Fig 6). A lack of safe water was seen as impacting economic growth (due to water cuts and rationing) and as a major contribution to compromised hygiene and increased risk of diarrhoeal disease. Floods contaminate freshwater supplies, heighten the risk of water-borne diseases, and create breeding grounds for disease vectors, for many of which climate change was lengthening the transmission season and geo- graphic range. Another major focus of discussion was the role of settlement on marginal land, poorly planned settlements and, more broadly, poverty and inequality on mediating the influences of climate change. Areas where there was significant population pressure on land and public infrastructure had poorer access to public services, which data confirmed affected both health and educational outcomes. These variables and associated pathways were elaborated during the workshop. Economic development which addressed deep inequalities was viewed as important to confront these sources of vulnerability. PLOS Climate | https://doi.org/10.1371/journal.pclm.0000184 April 19, 2023 12 / 20 PLOS CLIMATE Informing adaptation strategy through mapping the dynamics of linked systems Potential adaptation strategies discussed included the need to tackle the direct impacts of climate change via surveillance, monitoring and early warning systems. Strengthened social security strategies were considered of significance in reducing the multiple risks linked to pov- erty. Health systems need to expand traditional systems of healthcare delivery by integrating climate change considerations (e.g. control of climate-sensitive diseases), improving manage- ment of environmental determinants of health (such as water and sanitation, nutrition, and air quality), and establish emergency preparedness plans for extreme events. Urban and housing planning in marginal lands, coastal or flood-risk areas was also considered a key area of intervention. Integrative analysis. Causal loop analysis identified complex dynamics reflecting the unique characteristics of each setting. Modelling served a valuable function in collating evi- dence from multiple sources, convening consultations from researchers of varied disciplines, and identifying actions—and interactions—of relevance across multiple sectors. This approach to mapping the linkage of climate change, health, and other human systems such as agricul- ture, settlement, and livelihoods is thus perhaps best suited to local, contextual engagement of actors in identifying key leverage points for adaption strategy. However, while the causal loop analyses across these four settings reflect the unique characteristics of each setting, they also suggest some dynamics that are shared across these contexts. The causal loop diagram (Fig 7) seeks to represent some of the recurrent features from the country system dynamics. In all settings, mean temperature rise is leading to an increased fre- quent and intensity of extreme weather events that–whether through the means of floods, droughts, heatwaves etc.–expose populations to health risks. These risks exacerbate disease Fig 7. Causal loop diagram showing common dynamics across the four settings. Key foci of adaptation strategy indicated in red. https://doi.org/10.1371/journal.pclm.0000184.g007 PLOS Climate | https://doi.org/10.1371/journal.pclm.0000184 April 19, 2023 13 / 20 PLOS CLIMATE Informing adaptation strategy through mapping the dynamics of linked systems burden and undermine population health. This pathway of climate impact on population health is complemented by a pathway mediated by loss of agricultural production and reduced food security. Additional dynamics influencing the degree of impact of climate change are commonly mediated by the economic livelihoods of communities; migration (often related to conflict) and the resulting pressures on housing and infrastructure; and access to quality healthcare. There may–as illustrated in the country case studies—be multiple factors linking these path- ways, and this figure is not presented as an exhaustive analysis. Such interlinkage may be cru- cial in determining appropriate foci for local adaptation policy and practice (e.g. impact of government investment in healthcare on functional service access). However, the pathways highlighted in Fig 7 serve to signal broad classes of adaptation strategy operating with respect to factors highlighted in this integrative analysis. The five strategies are focused on i) early warning/preparedness regarding extreme events (thus mitigating exposure to risk); ii) adapted agricultural practices (to sustain food security and community livelihoods in changing environmental conditions); iii) urban planning (to strengthen the quality of housing and infrastructure and thus reduce population exposure to risks); iv) health systems resilience (to maintain access to quality healthcare both for the treat- ment of disease associated with increased risk exposure and for other conditions for which access may be disrupted by extreme events); and v) social security (supporting the livelihoods of communities vulnerable through the impact of climate change or otherwise) enabling their access to public services, including healthcare. Identification of key leverage points for intervention within a complex system of interac- tions is a valuable outcome of system dynamics analyses and an increasingly important focus of inter-disciplinary research focus [13, 52]. Discussion Climate change represents a significant threat globally, but particularly for LMIC and fragile settings. Linkages with health are increasingly recognised and becoming prioritised in the global health agenda [53]. While the Sharm-El-Sheikh Adaptation Agenda [12] does not list health as one of the ‘impact systems’ targeted for adaptation, its recognition of the importance that ‘actors across several sectors see . . . their actions and progress mutually reinforce to over- come obstacles, break silos, enhance synergies and create catalytic action’ has clear implica- tions for acknowledging the linkage of climate, health and other human systems. Indeed, the analyses presented illustrate how the ‘impact systems’ defined within the Sharm-El-Sheikh Adaptation Agenda –food and agriculture; water and nature; human settlement; coastal and ocean systems; infrastructure; planning; and finance–in practice richly interact with each other in shaping well-being. This research thus aimed to contribute to understanding by providing country specific findings and recommendation and by developing further the adoption of system thinking methodologies for use for climate and health research. We used a case study approach based on system dynamic modelling to identify adaptation strategies in four settings that present dif- ferent fragility features. The aim is ultimately to sustain the development of climate-resilient health systems, in line with the WHO operational framework [54]. The findings also speak directly to the interventions outlined in the WHO guidance for climate resilient and environ- mentally sustainable healthcare facilities [55] in providing evidence of the amplification of impacts through the interconnectedness of the challenges. This not only informs adaptation and mitigation measures required but also signals the co-benefits of investments in, for exam- ple, solar power, where transition from fossil fuels reduces carbon emissions, mitigates the PLOS Climate | https://doi.org/10.1371/journal.pclm.0000184 April 19, 2023 14 / 20 PLOS CLIMATE Informing adaptation strategy through mapping the dynamics of linked systems destabilising effects of energy systems facing outages because of adverse climate events, and reduces health risks through cleaner air. The causal loop diagrams presented in this paper act as useful starting points to identify fra- gility and leverage points that can support the policy development process. The use of system thinking has been recognised to be a key element to unpack climate change and build resilient health systems. Systems thinking, which stems from complexity theory, analyses the interac- tions between systems’ components to explain how and why they give rise to observed system outcomes and behaviours [56]. System thinking is particularly useful to support multi-sectoral collaboration through a shared understanding of the nexus between climate change and health and to foster political action by identifying effective strategies. For instance, the four models developed for this study highlighted the need to build surveillance and early warning systems. Key steps to reach these goals would include establishing key indicators [57], such as the ones suggested by The Lancet Countdown on health and climate change (e.g., risk exposures, vul- nerability factors, adaptation, planning, and resilience; mitigation and health co-benefits; eco- nomics; and political engagement) [58]. In this regard, risk assessment and health impact assessments should be integrated in routine assessments to quantify climate-driven health impacts. A system thinking approach to climate change and its impact on health is well suited to support health in all policy (HiAP) approach. HiAP is required to develop a comprehensive response to the risks presented by short-term climate variability and long-term climate change [59] and to define the health components of National Health Adaptation Plans (NHAPs) under the UN Framework Convention on Climate Change (UNFCCC). By identifying vulner- abilities in the health system as well as opportunities to increase the resilience of health systems to climate change, countries will be making important steps to achieve Universal Health Care (UHC). Climate-driven health outcomes should be included in the essential health services coverage by way of workforce training on climate–health relationships, financing, and increas- ing resilience of health care service delivery which may be disrupted during climate-related events (e.g., storms, and flooding). These can bolster UHC to address context-specific climate- driven health effects that are already being experienced and expected to worsen over time. Overall, more research and action are required to avoid the effects of climate change aggra- vating even further global health inequalities. A more profound question of justice is at play, whereby climate change interacts with existing social and economic disparities and exacerbates longstanding trends within and between countries. Finally, it is essential to incorporate differ- ent types of knowledge and an indigenous lens into the conceptualisation and implementation of planetary health [60]. Limitations To our knowledge, this is the first study that presented country case studies on the link between climate change and health using system thinking. Even though we used a robust methodological approach, some limitations need to be noted. Given the qualitative nature of the approach, we acknowledge that researcher perspectives may have influenced the work and findings; however, researchers from diverse backgrounds and from local contexts collaborated on the synthesis of the CLDs, bringing in diverse positions and perspectives. Conclusions Our research highlights five important lessons. First, system dynamics modelling methods, such as participatory group model building, provide a useful mechanism for convening actors across multiple sectors to consider the development of adaptation strategies. Consultations at national and local levels using approaches informed by systems dynamics should be used to PLOS Climate | https://doi.org/10.1371/journal.pclm.0000184 April 19, 2023 15 / 20 PLOS CLIMATE Informing adaptation strategy through mapping the dynamics of linked systems identify linkages that can promote–or, unattended, would undermine—coherent, cross-sec- toral action in support of adaptation. Second, in line with the OECD multi-dimensional analysis of fragility [61], climate-related environmental risks need to be increasingly factored into appraisal of state and regional fragil- ity, alongside issues of security and social, economic and political risks. Third, our modelling has highlighted how pathways of impact of climate change can dis- proportionally affect those with lower household incomes, exacerbating inequalities. Adapta- tion strategies need to consider a priori investments which prioritise social security of vulnerable communities and populations. Fourth, strategies focused on strengthening health systems resilience need to consider the relevant influences not only of national preparedness and early warning systems, but also of evolving agricultural (and wider livelihood) practices and patterns of settlement. Finally, fifth, effective data monitoring systems need to be prioritised at national level to integrate information from all relevant sectors, with datasets and analyses shared across all ministries. We consider these lessons to have important implications for conceptualizing adaptation both nationally and globally. In terms of the former, we have shared findings with governmen- tal partners regarding national climate adaptation strategy and, in Mozambique, are working with the National Institute for Health on a major prospective study of community adaptation measures in three locations at particular risk for extreme weather events. In terms of the latter, the lessons have been shared in a range of fora, ranging from fringe meetings in the context of CoP26 in Scotland to the multi-stakeholder policy forum of the 2023 Prince Mahidol Award Conference in Bangkok focused on ‘Setting a New Health Agenda: at the Nexus of Climate Change, Environment and Biodiversity’. By such means we aim for findings to foster the adop- tion of systems thinking in the formulation of adaptation strategies reflecting the dynamic linkages between climate change, health, and other human systems. Supporting information S1 Table. Search terms and inclusion criteria for literature review. (DOCX) S2 Table. Extraction template/matrix for literature review. (DOCX) S3 Table. List of included studies. (DOCX) Acknowledgments We are deeply grateful to our workshops participants who provided their knowledge, time and expertise to develop the case studies. These include: Dr Maia Uchaneishvili, Research Unit Director, Curatio International Foundation; Dr Nia Giuashvili, Environmental Health Expert, Advisor of the National Center for Disease Control and Public Health General Director on Environmental Health; Dr Mariam Maglakelidze, Head, Department of Institutional Culture Development, Petre Shotadze Tbilisi Medical Academy; Affiliate Scholar, Institute for Advanced Sustainability Studies, Potsdam, Germany; Ina Girard, Climate Change and Human Health Expert, WHO Focal Point on the Environmental Health Issues at the National Environ- mental Agency; Dr Tamar Kashibadze, Public Health Specialist, NCD Department, National Center for Disease Control and Public Health; Dr Tatiana Marrufo, Instituto Nacional de Sau´de (INS), National Health Observatory Technical Secretariat, Program Lead of PLOS Climate | https://doi.org/10.1371/journal.pclm.0000184 April 19, 2023 16 / 20 PLOS CLIMATE Informing adaptation strategy through mapping the dynamics of linked systems Environmental Health; Dr Fady Asmar, Forestry Expert, Lebanon; D.E.A. Pascal Girot, Head of the School of Geography, Universidad de Costa Rica; Dr Valeria Lentini, Lecturer, School of Economics, Universidad de Costa Rica; Dr Juan Robalino, Head of the Economics Research Institute, Universidad de Costa Rica; Dr Yanira Xirinachs-Salazar, Associate Professor, School of Economics, Universidad de Costa Rica; and Dr Paola Zu´ñiga-Brenes, Associate Professor, School of Economics, Universidad de Costa Rica. Author Contributions Conceptualization: Giulia Loffreda, Liz Grant, Alastair Ager. Data curation: Giulia Loffreda, Ivdity Chikovani, Ana O. Mocumbi, Michele Kosremelli Asmar, Laura C. Blanco. Formal analysis: Giulia Loffreda, Alastair Ager. Funding acquisition: Liz Grant, Alastair Ager. Investigation: Giulia Loffreda, Ivdity Chikovani, Ana O. Mocumbi, Michele Kosremelli Asmar, Laura C. Blanco, Alastair Ager. Methodology: Giulia Loffreda, Alastair Ager. Project administration: Alastair Ager. Resources: Ivdity Chikovani, Ana O. Mocumbi, Michele Kosremelli Asmar, Laura C. Blanco. Software: Giulia Loffreda. Supervision: Alastair Ager. Validation: Ivdity Chikovani, Ana O. Mocumbi, Michele Kosremelli Asmar, Laura C. Blanco. Writing – original draft: Giulia Loffreda. Writing – review & editing: Giulia Loffreda, Ivdity Chikovani, Ana O. Mocumbi, Michele Kosremelli Asmar, Laura C. Blanco, Liz Grant, Alastair Ager. References 1. United Nations Framework Convention on Climate Change. Copenhagen Accord. 2009. Available from: https://unfccc.int/resource/docs/2009/cop15/eng/l07.pdf 2. World Meteorological Organisation. State of the Climate in Africa. Avail. 2020. Available from: https:// library.wmo.int/doc_num.php?explnum_id=10421 3. Whitmee S, Haines A, Beyrer C, Boltz A., Capon A G, de Souza Dias B F et al. Safeguarding human health in the anthropocene epoch: Report of The Rockefeller Foundation–Lancet Commission on Plan- etary Health. Lancet, 2015; 386:1973–2028 https://doi.org/10.1016/S0140-6736(15)60901-1 PMID: 26188744 4. Berry H L, Waite T D, Dear K B G., Capon A G and Murray V. The case for systems thinking about cli- mate change and mental health. Nature Climate Change. 2018; 8, 282–290. https://doi.org/10.1038/ s41558-018-0102-4 5. Romanello M., Di Napoli C, Drummond P, Green C, Kennard H, Lampard P, et al. The 2022 report of the Lancet Countdown on health and climate change: health at the mercy of fossil fuels. Lancet, 2022; 400:1619–1654. https://doi.org/10.1016/S0140-6736(22)01540-9 PMID: 36306815 6. European Observatory on Health Systems and Policies, Thomas S, Sagan A, Larkin J, Cylus J. et al. Strengthening health systems resilience: key concepts and strategies. World Health Organization, Regional Office for Europe. 2020. Available from: https://apps.who.int/iris/handle/10665/332441 7. El-Sayed A M and Galea S Systems Science and Population Health. OUP: Oxford; 2017. 8. Jamal Z, Alameddine M, Diaconu K, Lough G, Witter S, Ager A et al. Health system resilience in the face of crisis: analysing the challenges, strategies and capacities for UNRWA in Syria, Health Policy and Planning, 2020; 35 (1), 26–35. https://doi.org/10.1093/heapol/czz129 PMID: 31625558 PLOS Climate | https://doi.org/10.1371/journal.pclm.0000184 April 19, 2023 17 / 20 PLOS CLIMATE Informing adaptation strategy through mapping the dynamics of linked systems 9. Blanchet K, Nam SL, Ramalingam B, Pozo-Martin F. Governance and Capacity to Manage Resilience of Health Systems: Towards a New Conceptual Framework. Int J Health Policy Manag. 2017; 6 (8):431–435. https://doi.org/10.15171/ijhpm.2017.36 PMID: 28812842; PMCID: PMC5553211. 10. Guzma´n CAF, Aguirre AA, Astle B, Barros E, Bayles B, Chimbari M, et al. A framework to guide plane- tary health education. Lancet Planet Health. 2021 May; 5(5):e253–e255. https://doi.org/10.1016/ S2542-5196(21)00110-8 Epub 2021 Apr 21. PMID: 33894134. 11. UNFCCC Report of the Conference of the Parties serving as the meeting of the Parties to the Paris Agreement on its third session, held in Glasgow from 31 October to 13 November 2021. Available from: https://unfccc.int/sites/default/files/resource/cma2021_10_add1_adv.pdf 12. Sharm-El-Sheikh Adaptation Agenda: The global transformations towards adaptive and resilient devel- opment. 2022. Available from: https://climatechampions.unfccc.int/wp-content/uploads/2022/11/SeS- Adaptation-Agenda_Complete-Report-COP27_FINAL-1.pdf 13. Proust K, Newell B, Brown H, Capon A, Browne C, Burton A, et al. Human health and climate change: leverage points for adaptation in urban environments. Int J Environ Res Public Health. 2012; 9 (6):2134–58. https://doi.org/10.3390/ijerph9062134 Epub 2012 Jun 6. PMID: 22829795; PMCID: PMC3397369. 14. Ager A, Saleh S, Wurie H, Witter S. Health systems research in fragile settings. WHO Bulletin, 2020, https://doi.org/http%3A//dx.doi.org/10.2471/BLT.19.233965 PMID: 31210671 15. OECD. States of Fragility 2022. OECD Publishing, Paris, 2022. Available from: https://doi.org/10.1787/ c7fedf5e-en 16. De Garine-Wichatitsky M, Binot A, Ward J, Caron A, Perrotton A, Ross H, et al. “Health in” and “Health of” Social-Ecological Systems: A Practical Framework for the Management of Healthy and Resilient Agricultural and Natural Ecosystems. Front. Public Health 2021; 8:616328. https://doi.org/10.3389/ fpubh.2020.616328 PMID: 33585387 17. Benjaminsen T A, Svarstad H. Political ecology. In: Encyclopedia of Ecology, 2nd ed. Volume 4. Else- vier; 2019. 18. United Nations Climate Change. Introduction to Gender and Climate Change. UNFCCC; 2020. Avail- able from: https://unfccc.int/gender 19. Markedonov, Sergey M., and Suchkov M. A. Russia and the United States in the Caucasus: cooperation and competition. Caucasus Survey. 2020; 8.2: 179–195. https://doi.org/10.1080/23761199.2020. 1732101 20. Rukhadze T. An overview of the health care system in Georgia: expert recommendations in the context of predictive, preventive and personalised medicine. EPMA J. 2013; 4(1):8. https://doi.org/10.1186/ 1878-5085-4-8 PMID: 23442219; PMCID: PMC3621519. 21. The Fund for Peace. Fragile States Index Report. 2021. Available from: https://fragilestatesindex.org/ wp-content/uploads/2021/05/fsi2021-report.pdf 22. World Health Organization. Regional Office for Europe, European Observatory on Health Systems and Policies. Georgia: health system review. World Health Organization. Regional Office for Europe, 2017. 23. World Health Organization. Regional Office for Europe, European Observatory on Health Systems and Policies, Gamkrelidze A, Atun R, Gotsadze G. et al. Health care systems in transition: Georgia. WHO, Regional Office for Europe; 2002. Available from: https://apps.who.int/iris/handle/10665/107402 24. Goginashvili K, Nadareishvili M, Habicht T. Can people afford to pay for health care? New evidence on financial protection in Georgia. Copenhagen: WHO Regional Office for Europe; 2021. 25. USAID. Climate risk profile Georgia, Factsheet, 2017. Available from: https://www.climatelinks.org/ sites/default/files/asset/document/2017_USAID%20ATLAS_Climate%20Change%20Risk%20Profile %20-%20Georgia.pdf 26. Fourth National Communication of Georgia under the UNFCCC, 2021. Available from: https://unfccc. int/sites/default/files/resource/4%20Final%20Report%20-%20English%202020%2030.03_0.pdf 27. Georgia’s 2030 Climate Change Strategy and Action Plan. Government of Georgia. Available from: https://mepa.gov.ge/En/Files/ViewFile/50123 28. Batley R, Bjørnestad L, Cumbi A. Joint Evaluation of General Budget Support 1994–2004: Mozambique Country Report. International Development Department School of Public Policy University of Birming- ham. 2006. Available from: https://www.oecd.org/countries/mozambique/43867765.pdf 29. Anselmi L, Lagarde M, Hanson K. Health service availability and health seeking behaviour in resource poor settings: evidence from Mozambique. Health Econ Rev. 2015; 5(1):62. https://doi.org/10.1186/ s13561-015-0062-6 Epub 2015 Sep 2. PMID: 26329425; PMCID: PMC4556719. 30. World Bank. The World Bank in Mozambique [Internet] Accessed 24 June 2022. Available from: https:// www.worldbank.org/en/country/mozambique/overview PLOS Climate | https://doi.org/10.1371/journal.pclm.0000184 April 19, 2023 18 / 20 PLOS CLIMATE Informing adaptation strategy through mapping the dynamics of linked systems 31. Bukhman G, Mocumbi AO, Atun R, Becker AE, Bhutta Z, Binagwaho A et al. Lancet NCDI Poverty Commission Study Group. The Lancet NCDI Poverty Commission: bridging a gap in universal health coverage for the poorest billion. Lancet. 2020; 3; 396(10256):991–1044. https://doi.org/10.1016/ S0140-6736(20)31907-3 Epub 2020 Sep 14. PMID: 32941823; PMCID: PMC7489932. 32. Mozambique. [internet] Institute for Health Metrics and Evaluation, Global Burden of Disease. Accessed 10 November 2022. Available from https://www.healthdata.org/mozambique 33. Update of the First Nationally Determined Contribution to the United Nations Framework Convention on Climate Change Mozambique Period: 2020–2025. Republic of Mozambique. 2021. Available from https://unfccc.int/sites/default/files/NDC/2022-06/NDC_EN_Final.pdf 34. Clim-Health Africa. Health vulnerability and adaptation to climate change assessment conducted in Mozambique. May 2020. Available from: https://climhealthafrica.org/news-mozambique-vulnerability- and-adaptation-assessment 35. World Health Organization and the United Nations Framework Convention on Climate Change. Leba- non Health and Climate Change Country Profile. WHO/UNFCCC, 2021. 36. Government of Lebanon and United Nations, Lebanon Crisis Response Plan 2017–2020: 2020 Update. Available from: https://lebanon.un.org/sites/default/files/2021-02/LCRP2020%20update_EN_Full_ 180122-035824.pdf 37. United Nation Refugee Agency (UNHCR). UNHCR Global Appeal, 2015 Update. Available from: https://www.unhcr.org/uk/publications/fundraising/5461e5ec3c/unhcr-global-appeal-2015-update- populations-concern-unhcr.html 38. Masri S., Srour I. Assessment of the impact of Syrian refugees in Lebanon and their employment profile. International Labour Organization. Geneva, 2014. Available from: https://www.ilo.org/beirut/ publications/WCMS_240134/lang—en/index.htm 39. Situation Syria Regional Refugee Response [internet] United Nation Refugee Agency (UNHCR). Accessed: 16 August 2021. Available from: https://data.unhcr.org/en/situations/syria 40. Devi S. Lebanon faces humanitarian emergency after blast. Lancet. 2020; 396(10249): 456. https://doi. org/10.1016/S0140-6736(20)31750-5 PMID: 32798477 41. Naja F., Shatila H., El Koussa M, Lokman M, Ghandour L, Saleh S. Burden of non-communicable dis- eases among Syrian refugees: a scoping review. BMC Public Health, 2019; 19, 637. https://doi.org/10. 1186/s12889-019-6977-9 PMID: 31126261 42. Noubani A, Diaconu K, Loffreda G, et al. Readiness to deliver person-focused care in a fragile situation: the case of Mental Health Services in Lebanon. Int J Ment Health Syst. 2021 15, 21. https://doi.org/10. 1186/s13033-021-00446-2 PMID: 33653392 43. 44. Lebanon’s Nationally Determined Contribution Updated 2020 Version. Government of Lebanon; 2020. Lebanon: WHO Health and Climate Change Country Profile 2021. WHO/UNFCCC; 2021. Available from: https://www.who.int/publications/i/item/WHO-HEP-ECH-CCH-21.01.09 45. Costa Rica [Internet] The World Bank. Accessed 10 November 2022. Available from: https://data. worldbank.org/country/CR 46. Costa Rica. [Internet] United Nations Refugee Agency (UNHCR). Accessed 10 November 2022. Avail- able from: https://www.unhcr.org/uk/costa-rica.html 47. OECD Costa Rica Policy Brief: Inequity. February 2016. Available from: https://www.oecd.org/policy- briefs/costa-rica-towards-a-more-inclusive-society.pdf 48. Caja Costarricense de Seguro Social. Estados financieros: Seguro de salud, San Jose´: CCSS. Geren- cia Financiera; 2020. 49. National Institute of Statistics and Census. Continuous Survey on Employment for the first quarter of 2020. General Results. San Jose´ : INEC; 2020. 50. Contribucio´ n Nacionalmente Determinada de Costa Rica. [Internet] Accessed 10 November 2022. Available: https://cambioclimatico.go.cr/contribucion-nacionalmente-determinada-ndc-de-costa-rica/ 51. Direccio´ n de Cambio Clima´ tico; Ministerio de Ambiente y Energı´a. Plan Nacional de Adaptacio´ n al Cambio Clima´ tico de Costa Rica, 2022–2026. San Jose´ , Costa Rica, 2022. 52. Schnitter R. and Berry P. The Climate Change, Food Security and Human Health Nexus in Canada: A Framework to Protect Population Health. International Journal of Environmental Research and Public Health 16(14):2531. https://doi.org/10.3390/ijerph16142531 PMID: 31315172 53. The 2022 Global Report of the Lancet Countdown: Tracking Progress on Health and Climate Change. Available from: https://www.lancetcountdown.org/2022-report/ 54. Operational framework for building climate resilient health systems, WHO, 2015. Available from: https:// www.who.int/publications/i/item/9789241565073 PLOS Climate | https://doi.org/10.1371/journal.pclm.0000184 April 19, 2023 19 / 20 PLOS CLIMATE Informing adaptation strategy through mapping the dynamics of linked systems 55. WHO guidance for climate-resilient and environmentally sustainable health care facilities. Geneva: World Health Organization, 2020. 56. Kwamie A, Ha S, Ghaffar A. Applied systems thinking: unlocking theory, evidence and practice for health policy and systems research, Health Policy and Planning. 2021; 36 (10), 1715–1717. https://doi. org/10.1093/heapol/czab062 PMID: 34131699 57. World Health Organization. WHO Guidance to Protect Health from Climate Change through Health Adaptation Planning; World Health Organization: Geneva, Switzerland, 2015. 58. Watts N, Amann M, Arnell N, Ayeb-Karlsson S, Beagley J. Belesova K. et al. The 2020 report of The Lancet Countdown on health and climate change: Responding to converging crises. Lancet 2020, 396, 129–170. 59. Pongsiri M J and Bassi A M. A Systems understanding underpins actions at the climate and health nexus. Int. J. Environ. Res. Public Health 2021, 18, 2398. https://doi.org/10.3390/ijerph18052398 PMID: 33804531 60. Redvers N, Celidwen Y, Schultz C, Horn O, Githaiga C, Vera M et al. The determinants of planetary health: an Indigenous consensus perspective. Lancet Planet Health. 2022 Feb; 6(2):e156–e163. https://doi.org/10.1016/S2542-5196(21)00354-5 PMID: 35150624. 61. Diaconu K, Falconer J, Vidal N, O’May F, Azasi E, Elimian K et al. Understanding fragility: implications for global health research and practice, Health Policy and Planning, 2020; 35 (2) 235–243. https://doi. org/10.1093/heapol/czz142 PMID: 31821487 PLOS Climate | https://doi.org/10.1371/journal.pclm.0000184 April 19, 2023 20 / 20 PLOS CLIMATE
null
10.1186_s12917-019-2170-8.pdf
Availability of data and materials The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.
Availability of data and materials The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.
Sirikaew et al. BMC Veterinary Research (2019) 15:419 https://doi.org/10.1186/s12917-019-2170-8 R E S E A R C H A R T I C L E Open Access Proinflammatory cytokines and lipopolysaccharides up regulate MMP-3 and MMP-13 production in Asian elephant (Elephas maximus) chondrocytes: attenuation by anti-arthritic agents Nutnicha Sirikaew1, Siriwadee Chomdej2, Siriwan Tangyuenyong3, Weerapongse Tangjitjaroen3, Chaleamchat Somgird3, Chatchote Thitaram3 and Siriwan Ongchai1* Abstract Background: Osteoarthritis (OA), the most common form of arthritic disease, results from destruction of joint cartilage and underlying bone. It affects animals, including Asian elephants (Elephas maximus) in captivity, leading to joint pain and lameness. However, publications regarding OA pathogenesis in this animal are still limited. Therefore, this study aimed to investigate the effect of proinflammatory cytokines, including interleukin-1 beta (IL- 1β), IL-17A, tumor necrosis factor-alpha (TNF-α), and oncostatin M (OSM), known mediators of OA pathogenesis, and lipopolysaccharides on the expression of cartilaginous degrading enzymes, matrix metalloproteinase (MMP)-3 and MMP-13, in elephant articular chondrocytes (ELACs) cultures. Anti-arthritic drugs and the active compounds of herbal plants were tested for their potential attenuation against overproduction of these enzymes. Results: Among the used cytokines, OSM showed the highest activation of MMP3 and MMP13 expression, especially when combined with IL-1β. The combination of IL-1β and OSM was found to activate phosphorylation of the mitogen-activated protein kinase (MAPK) pathway in ELACs. Lipopolysaccharides or cytokine-induced expressions were suppressed by pharmacologic agents used to treat OA, including dexamethasone, indomethacin, etoricoxib, and diacerein, and by three natural compounds, sesamin, andrographolide, and vanillylacetone. Conclusions: Our results revealed the cellular mechanisms underlying OA in elephant chondrocytes, which is triggered by proinflammatory cytokines or lipopolysaccharides and suppressed by common pharmacological or natural medications used to treat human OA. These results provide a more basic understanding of the pathogenesis of elephant OA, which could be useful for adequate medical treatment of OA in this animal. Keywords: Elephas maximus, Osteoarthritis, Proinflammatory cytokines, MMP-3, MMP-13 * Correspondence: siriwan.ongchai@cmu.ac.th 1Thailand Excellence Center for Tissue Engineering and Stem Cells, Department of Biochemistry, Faculty of Medicine, Chiang Mai University, 110 Intrawarorot Rd., Chiang Mai 50200, Thailand Full list of author information is available at the end of the article © The Author(s). 2019 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated. Sirikaew et al. BMC Veterinary Research (2019) 15:419 Page 2 of 13 Background Osteoarthritis (OA), the most prevalent arthritic disease, is characterized by cartilage degradation and consequent joint pain and disability [1, 2]. OA affects many species, including elephants, especially Asian elephants (Elephas maximus) kept in captivity. Excessive body weight along with the captive environment and trained behaviors are critical factors of OA pathogenesis in elephants [3, 4]. These factors disturb the equilibrium between the syn- thesis and degradation of the extracellular matrix (ECM) by chondrocytes, leading to further degradation of the ECM by matrix-degrading enzymes, especially matrix metalloproteinases (MMPs) [5]. The disturbance of this equilibrium is found particularly among captive ele- phants [6]. MMPs are a group of zinc-dependent endopeptidases that, when in excess, cause degeneration of the cartilage ECM. There has been a reported increase in MMP-3 and MMP-13 in humans and animals with OA, suggest- ing that these MMPs play a pivotal role in OA cartilage destruction [7–10]. It has previously been shown that the production of matrix-degrading enzymes is activated by proinflammatory cytokines, including interleukin-1 beta (IL-1β), IL-17A, tumor necrosis factor-alpha (TNF- α), and oncostatin M (OSM) [11–14]. In addition, the combination of OSM with other proinflammatory cyto- kines causes the greatest loss of cartilage matrix in OA [15–17]. Moreover, lipopolysaccharides (LPS), i.e., outer- membrane components of Gram-negative bacteria, con- tribute to septic arthritis and cartilage degeneration by upregulating the synthesis of catabolic factors, including proinflammatory cytokines and matrix-degrading en- zymes [18, 19]. In OA pathogenesis, cytokine-induced signal transduction involves the activation of several pathways, including those of the mitogen-activated pro- tein kinase (MAPK) family [20]. OA in elephants is caused by an imbalance of pressure on joints, which in turn is caused by a lack of exercise or an excessive body weight. This damages the cartilage, releasing inflammatory mediators and enzymes and, consequently, leading to joint inflammation. Affected el- ephants show signs of lameness and joint swelling and are reluctant to lay down because it will be difficult to stand up again. Swimming in a big pool to reduce weight bearing and administration of anti-inflammatory drugs are considered suitable treatments [21]. Current pharmacologic approaches for OA treatment aim at reducing inflammation and pain, improving joint function, and delaying disease progression. Commonly used medicines include steroids, non-steroidal anti- inflammatory drugs (NSAIDs), and disease-modifying OA drugs (DMOADs) [22], among which the most com- mon agents are dexamethasone, indomethacin, etori- coxib, and diacerein, which have been shown to inhibit the expression of MMPs such as MMP1, MMP2, MMP3, MMP9, and MMP13 [23–26]. However, these substances are associated with a high incidence of adverse effects, including gastrointestinal damage and heart failure [27]. Thus, natural product-derived compounds with anti- inflammatory activity and low toxicity have become alternative treatments for OA. Among such compounds, sesamin, andrographolide, and vanillylacetone or zinger- one have been reported to exhibit chondroprotective activity by inhibiting the expression of MMP1, MMP3, and MMP13 in chondrocytes [28–30]. It was reported that IL-1β stimulated the degradation of elephant cartilage in an explant culture model [31]. However, the existence of published studies on the cellular mechanisms of OA in elephants is limited. Therefore, the present study aimed to investigate the molecular mechanisms underlying the activation of ex- pression of MMP-3 and MMP-13 by proinflammatory cytokines and LPS in elephant articular chondrocytes (ELACs). Additionally, the ability of commonly used anti-OA medications and natural compounds to inhibit these mechanisms was investigated. The information gained from this study will be useful in improving the treatment of elephants with OA and in supporting fur- ther research on elephant degenerative arthritis, both of which are important for a better quality of life for the el- ephants and contribute to vital elephant conservation. Results Proinflammatory cytokines induced upregulation of MMP3 and MMP13 expression in ELACs culture Treatment with OSM alone resulted in a slight increase in MMP3 mRNA levels and a marked elevation of MMP13 levels. However, IL-1β, IL-17A, and TNF-α did not influence the expression of these genes in the mono- layer culture model (Fig. 1). The combination of cyto- kines OSM and TNF-α significantly induced MMP13 expression, whereas the combination of OSM and IL-1β or IL-17A tended to induce MMP3 expression. In the pellet culture model (Fig. 2), the results of individual cytokine treatments show that only TNF-α could signifi- cantly activate the expression of MMP13. Meanwhile, the results of treatments with combined cytokines dem- onstrate that OSM combined with IL-1β dramatically in- creased the expression of both MMP3 and MMP13, whereas OSM combined with TNF-α slightly induced the expression of MMP13 but not that of MMP3. Drugs and active compounds of medicinal plants inhibited cytokine-induced expression of MMP3 and MMP13 in ELACs culture The results show that medications used to treat OA in humans, such as diacerein, dexamethasone, indometh- acin, and etoricoxib, significantly attenuated MMP3 and Sirikaew et al. BMC Veterinary Research (2019) 15:419 Page 3 of 13 Fig. 1 Proinflammatory cytokines upregulate the mRNA expression of MMP3 (a) and MMP13 (b) in ELACs. The chondrocytes were treated with individual proinflammatory cytokines as follows: IL-1β (2.5 ng/mL); IL-17A (5 ng/mL); and TNF-α (5 ng/mL), or their combination with OSM (2 ng/ mL) or IL-17A (5 ng/mL), for 24 h. mRNA levels were assessed by real-time RT-PCR. Results are presented as mean ± SEM. * signifies statistical significance compared with control (*p < 0.05), whereas # signifies statistical significance in relation to single-cytokine treatment (#p < 0.05) MMP13 mRNA levels in the ELACs culture (Fig. 3a and b). Likewise, natural active compounds, including sesa- min, andrographolide, and vanillylacetone, significantly suppressed the MMP3 and MMP13 mRNA levels in a dose-dependent manner (Fig. 4a and b). LPS induced the expression of MMP3 and MMP13 along with proinflammatory cytokine genes in ELACs culture The results show that LPS at a 0.125 μg/mL concentra- tion significantly increased MMP3 and MMP13 mRNA levels as well as the levels of IL1B and IL6 while increas- ing the expression of the TNF-α gene (TNFA) at a con- centration of only 0.25 μg/mL (Fig. 5). Co-treatment with LPS and anti-arthritic drugs such indomethacin, and etori- as diacerein, dexamethasone, coxib significantly suppressed MMP3 and MMP13 mRNA levels in a dose-dependent manner (Fig. 6a and b). Figure 6c illustrates the LPS-induced increase of MMP-13 protein levels in the culture media, which was significantly suppressed by dexamethasone and indo- methacin. However, the level of MMP-3 in the culture media could not be assessed using a human MMP-3 CLIA kit (data not shown). Activation of the MAPK pathway in ELACs by IL-1β combined with OSM The MAPK pathway, one of the molecular mechanisms involved in OA pathogenesis, was activated in ELACs treated with a combination of IL-1β and OSM. The re- sults show that the combined proinflammatory cytokines activated the maximum phosphorylation of p38, ERK, and JNK from 5 to 10 min, followed by its gradual de- crease after 15 min (Fig. 7). Discussion OA is the most prevalent musculoskeletal disorder in both humans and animals. Most studies on OA have fo- cused on humans, with few reports available on animals, Sirikaew et al. BMC Veterinary Research (2019) 15:419 Page 4 of 13 the mechanisms underlying OA pathogenesis. Although the pellet culture, a three-dimensional culture model, mimicked the chondrocytes’ microenvironment within cartilage tissue more accurately [32], two-dimensional monolayer cultures are a faster and simpler model for cell-based studies. They allowed for quick evaluation of the effects of several proinflammatory cytokines known to be involved in OA pathogenesis on the expressions of MMP3 and MMP13 in ELACs. The present results clearly demonstrate that ELACs are sensitive to activation by proinflammatory cytokines. Among the proinflammatory cytokines, the treatment with OSM alone strongly induced the expression of MMP13 in the monolayer cultures; TNF-α, which has been previously reported to induce the expression of MMP1, MMP3, and MMP13 in equine chondrocytes [11], caused a significant upregulation of MMP13 in the elephant chondrocyte pellet culture. IL-17A, alone or in combination with IL-1β or TNF-α, did not alter the ex- pression of MMP3 or MMP13. The treatment with a combination of IL-17A and OSM caused a slight upreg- ulation of MMP3 with no effect on MMP13. This result is inconsistent with previous studies on human cartilage cultures, which showed that the combination of IL-17A with TNF-α and OSM synergistically upregulates the ex- pression of enzymes MMP-1 and MMP-13 [33]. This cytokine is known to be increased in the serum of OA patients, in human OA pathogenesis [34]. suggesting its involvement Although IL-1β has been reported to play a key role in the OA pathogenesis of large animals by upregulating the expression of MMP-1, MMP-3, and MMP-13 en- zymes [13, 35, 36], our results clearly demonstrate that in the elephant chondrocyte pellet culture model, this cytokine could only induce the expression of MMP3 and MMP13 in combination with OSM. This result is con- sistent with a recent report suggesting that IL-1α and IL-1β are not crucial mediators of murine OA, which may explain the lack of success of IL-1-targeted therap- ies for OA [37]. Nevertheless, a previous report by our team demonstrated a great loss of hyaluronan from elephant cartilage explants treated with human recom- binant IL-1β, suggesting the catabolic potential of this cytokine via accelerating the processes of cleavage and release of ECM biomolecules from the affected cartilage tissue, leading to degenerative cartilage in OA [31]. Fig. 2 IL-1β in combination with OSM stimulates expression of MMP3 (a) and MMP13 (b) in ELAC pellets culture. ELAC pellets were treated with IL-1β or TNF-α, alone or in combination with OSM, for 3 days. The mRNA levels were assessed by real-time RT-PCR. Results are presented as mean ± SEM. * signifies statistical significance compared with control (*p < 0.05), whereas # signifies statistical significance in relation to single-cytokine treatment (#p < 0.05) especially elephants. Asian elephants kept in captivity frequently suffer from OA caused primarily by residing in damp buildings and being overworked by humans as well as by restricted movement, which leads to cartilage degeneration and lameness [3, 4]. Reports on the mecha- nisms underlying OA in elephants are rare. The present study used monolayer and pellet cultures of elephant chondrocytes as in vitro models to investigate OSM, which belongs to the IL-6 family, is one of the proinflammatory cytokines that contribute to inflamma- tion and cartilage destruction in degenerative arthritis [38]. OSM induces the expression of MMP1, MMP3, and MMP13 in bovine chondrocytes [12]. This cytokine has also been reported to synergize the action of other proinflammatory cytokines such as IL-1β, TNF-α, and IL- 17A, resulting in acceleration of cartilage degeneration Sirikaew et al. BMC Veterinary Research (2019) 15:419 Page 5 of 13 Fig. 3 Anti-arthritic drugs decrease the cytokines-induced expressions of MMP3 (a) and MMP13 (b) in ELACs. Chondrocytes were pre-treated with a combination of IL-1β (2.5 ng/mL) and OSM (2 ng/mL) for 2 h, after which they were treated with various concentrations of DIA (diacerein; 2.5– 10 μM), DEX (dexamethasone; 5–20 nM), INDO (indomethacin; 2.5–10 μM), and ETORI (etoricoxib; 2.5–10 μM), for 24 h. mRNA levels were assessed by real-time RT-PCR. Results are presented as mean ± SEM. * signifies statistical significance compared with control (*p < 0.05), whereas # signifies statistical significance in relation to the cytokines treatment group (#p < 0.05) [15–17]. In this study, in elephant chondrocytes, the com- bination of OSM with IL-1β exerted the strongest induc- tion of MMP3 and MMP13 expression in both the monolayer and pellet culture models, whereas the com- bined OSM with TNF-α only influenced the expression of MMP13. Our results suggest a cell-type specificity in response to the activation of cytokines. Additionally, all cytokines used in the present study were human recom- binant proteins, implying that their actions on elephant chondrocytes may not represent the actions of species- specific cytokines. Nevertheless, the significant enhance- ment of MMP3 and MMP13 expression achieved by the combination of OSM and IL-1β provides important infor- mation regarding the action of these cytokines in the cata- bolic processes of elephant OA, which are similar to OA pathogenesis in other animals [17, 39]. Enzymes MMP-3 and MMP-13 are members of a zinc-dependent group of endopeptidases and considered crucial for the destruction process of cartilage ECM that occurs in OA [7–10]. The present study reveals that the expression of elephant MMP13 is more sensitive to in- duction by cytokines than MMP3. Among MMPs, most studies have focused on MMP-13, a collagenase-3, which is suggested to play a critical role in both the early stages and progression of OA [9, 40]. It is overexpressed in pa- tients with OA but not in healthy patients. MMP-13 in- volves in cartilage degradation and also acts as a regulatory factor. It has been suggested that it plays a key role in controlling the onset of OA by leading chon- drocytes from a normal to a pathological state [41]. MMP-3, stromelysin-1, is a matrix-degrading enzyme found to be increased in the serum and plasma of Sirikaew et al. BMC Veterinary Research (2019) 15:419 Page 6 of 13 Fig. 4 Natural active compounds reduce the cytokines-induced mRNA levels MMP3 (a) and MMP13 (b) in ELACs. The chondrocytes were pre- treated with a combination of IL-1β (2.5 ng/mL) and OSM (2 ng/mL) for 2 h, after which they were treated with various concentrations of SE (sesamin; 0.25–1 μM), AD (andrographolide; 1.25–5 μM), and VA (vanillylacetone; 20–80 μM), for 24 h. The mRNA levels were assessed by real-time RT-PCR. Results are presented as mean ± SEM. * signifies statistical significance compared with control (*p < 0.05), whereas # signifies statistical significance in relation to the cytokines treatment group (#p < 0.05) humans with OA, although its levels are not directly as- sociated with OA severity [42]. Immunohistochemical assay of the synovium tissue of OA shows a high expres- sion of MMP-3, which is positively correlated to the se- verity of the disease [10]. Likewise, in this study, the high expression of these enzymes in elephant chondrocytes was demonstrated under activation by the proinflammatory cytokines re- sponsible for OA pathogenesis. Our results suggest that these enzymes, especially MMP-13, which exerts a strong response to cytokine activation, may be one of the key catabolic enzymes involved in elephant cartilage degeneration. Cytokine-induced upregulation of MMP13 mRNA levels was accompanied by an increase of MMP- 13 protein levels in the culture media. This protein was successfully measured by a test kit designed to deter- mine the level of human MMP-13, suggesting that the structures of elephant and human MMP-13 is closely re- lated. However, another test kit designed to analyze hu- man MMP-3 levels could not successfully be applied to measure the level of MMP-3 protein in elephant chon- drocytes. Therefore, we postulate that the MMP-3 pro- tein structure similarity between humans and elephants falls below the threshold of the recognizable capability of the human MMP-3 monoclonal antibody provided with the test kit. Currently, scientific evidence on OA pathogenesis in elephants is limited. Expanding information regarding the biomechanisms of the disease as well as the effective- ness of drugs will support the development of thera- peutic interventions to treat elephant OA. As such, the present study selected four drugs commonly prescribed to treat OA in humans and indomethacin, other animals, namely, dexamethasone, that may be helpful Sirikaew et al. BMC Veterinary Research (2019) 15:419 Page 7 of 13 Fig. 5 LPS induces expression of MMP3 and MMP13 (a), and proinflammatory cytokines (b) in ELACs culture. The chondrocytes were treated with LPS at various concentrations (0.125–1 μg/mL) for 24 h, then mRNA levels were assessed by real-time RT-PCR. Results are presented as mean ± SEM. * signifies statistical significance compared with control (*p < 0.05) etoricoxib, and diacerein. Dexamethasone is a synthetic corticosteroid previously shown to inhibit the expression of MMP3 and MMP13 in IL-1α-induced bovine chon- drocytes and suppress cytokine-induced inhibition of matrix biosynthesis in bovine cartilage [26]. NSAIDs are generally used to reduce pain and inflammation in arth- ritis through inhibition of cyclooxygenase (COX) [43]. Indomethacin is a non-selective inhibitor, whereas etori- coxib is in the COX2 selective class of NSAIDs. The former has been reported to reduce the expression of MMP1 and MMP3 in IL-1α-induced bovine chondro- cytes [23], whereas the latter has been found to decrease the levels of MMP-2 and MMP-9 [25]. Diacerein, a DMOADs, has been reported to decrease the production of IL-1-converting enzyme and IL-1β in human osteo- arthritic cartilage [44] as well as suppress the expression of MMP1, MMP3, MMP13, ADAMTS-4, and ADAMTS- 5 in IL-1β-induced bovine chondrocytes [24]. Our re- sults show that these drugs effectively suppress the ex- pression of MMP3 and MMP13 induced by the combination of IL-1β and OSM or LPS, suggesting that they exhibited an anti-arthritic potential in the elephant chondrocytes culture model. Moreover, this study demonstrates the protective ef- fect of natural compounds previously reported to have anti-arthritic properties such as sesamin, andrographo- lide, and vanillylacetone against cytokine-induced ex- pression of MMP3 and MMP13 in elephants, suggesting similarities in human and elephant OA pathogenesis, which is ameliorated by the action of these natural com- pounds. The concentration ranges of the natural com- pounds used in this study did not cause cell mortality but still effectively reduced the expression of MMP3 and MMP13 and were selected based on the results of the Sirikaew et al. BMC Veterinary Research (2019) 15:419 Page 8 of 13 Fig. 6 Anti-arthritic drugs suppressed mRNA levels of MMP3 (a) and MMP13 (b) and decreasing MMP13 protein levels (c). The chondrocytes were pre-treated with 0.5 μg/mL LPS for 2 h, after which they were treated with various concentrations of DIA (diacerein; 2.5–10 μM), DEX (dexamethasone; 5–20 nM), INDO (indomethacin; 2.5–10 μM), and ETORI (etoricoxib; 2.5–10 μM) for 24 h. mRNA levels were then assessed by real- time RT-PCR. Results are presented as mean ± SEM. * signifies statistical significance compared with control (*p < 0.05), whereas # signifies statistical significance in relation to the cytokines treatment group (#p < 0.05) MTT cytotoxic assay [see Additional file 1]. However, the therapeutic dose of these agents on human or animal arthritis remains unclear. Therefore, the application of these agents to human or animal arthritis must be further investigated to achieve the maximum therapeutic effect. It was reported that supplementation of sesame seed in patients with knee OA at a dose of 40 g daily for 2 Sirikaew et al. BMC Veterinary Research (2019) 15:419 Page 9 of 13 Fig. 7 Activation of the MAPK pathway in ELACs by IL-1β combined with OSM. ELACs were stimulated by the combination of IL-1β (2.5 ng/mL) and OSM (2.5 ng/mL) at the indicated time points. Cell lysates were immunoblotted to investigate the total and phosphorylated molecular forms, which indicated an active MAPK pathway. Immunoblots are represented in (a) and bar graphs (b) show the proportion between the band intensities of phosphorylated p38, ERK, and JNK over their total forms. Results are presented as mean ± SEM. * signifies statistical significance compared with control (*p < 0.05) months, along with standard medical therapy, improved the disease activity by reducing serum IL-6 [45]. In papain-induced rat OA, intra-articular injection of 20 μL of 1 or 10 μM sesamin reduced cartilage distortion [28]. This compound is the most prominent lignan in sesame seed oil [46] and has been reported to exert anti- arthritic effects by reducing IL-1β-induced production of proinflammatory mediators and cartilage-degrading en- zymes MMP-1, MMP-3, and MMP-13, in human osteo- arthritic chondrocytes via suppressing phosphorylation of NF-κB p65 and IκB and activation of the Nrf2 signal- ing pathway [28, 47]. [48]. Vanillylacetone, also called zingerone, is the major component of ginger root and has known antioxidant and anti-inflammatory properties In cytokine- induced degradation of porcine cartilage explant, this compound decreased the release of MMP-13 and cartil- age matrix biomolecules into the culture media by sup- pressing the p38 and JNK MAPK signaling pathways [30]. Patients receiving one ginger extract capsule pre- pared from 2500 to 4000 mg of dried ginger rhizomes twice daily for 6 weeks showed a significant reduction of OA symptoms [49]. However, reports on the usage of vanillylacetone for anti-arthritic purposes in humans or animals are still limited. Andrographolide is a major bioactive compound of Andrographis paniculata (Burm.f.) that was found to in- hibit the expression of MMPs and inducible nitric oxide synthase in an IL-1β-induced OA model [29]. This agent reduced the productions of proinflammatory cytokines in vitro by suppressing the p38 MAPK and ERK1/2 pathways and alleviated arthritis severity in mice treated by oral administration of andrographolide 100 mg/kg/d [50]. It was reported that a combined administration of andrographolide 50 mg/kg/d and methotrexate 2 mg/kg/ week in rat arthritis induced by complete Freund’s adju- vant significantly attenuated inflammatory symptoms and reduced liver injury caused by methotrexate [51]. Andrographolide has been proposed as a new potential anti-arthritic agent [52]. Therefore, it is worth further investigating the optimal dose of this agent for arthritis treatments in animals or humans. LPS are known to in- duce infectious arthritis and contribute to low-grade inflammation in OA pathogenesis [19, 53, 54]. They en- hance the production of MMP-1, MMP-3, MMP-13, ni- tric oxide, and prostaglandin E2 in OA patients, leading to an increase in the area of cartilage destruction [55]. Likewise, the present study on elephant chondrocytes demonstrated a strong inducing effect of bacterial LPS on the expression of proinflammatory cytokine genes, including IL1B, TNFA, and IL6, together with matrix- degrading enzymes MMP3 and MMP13. These results shed light on the in vitro mechanisms of septic arthritis in an elephant chondrocyte culture model, which, when induced by LPS, showed an increased expression of pro- inflammatory cytokines and matrix-degrading enzymes. These effects were mitigated by dexamethasone, indo- methacin, etoricoxib, and diacerein. Our findings suggest Sirikaew et al. BMC Veterinary Research (2019) 15:419 Page 10 of 13 that these drugs attenuate LPS-induced inflammation and catabolic factors in both elephant and human chondrocytes. MAPK is one of the most important signaling path- ways regulating OA pathogenesis [56]. It is activated by including IL-1β and OSM proinflammatory cytokines, [12, 57], with consequent upregulation of cartilage- degrading enzyme production, including that of MMP-3 and MMP-13 [56, 58]. This study investigated the mech- anisms underlying elephant OA by treating elephant chondrocytes with a combination of IL-1β and OSM via a commercial test kit commonly used to detect cellular activation in human cells via the MAPK signaling path- way. The present study shows that this test kit was suc- cessful in revealing the effects of these cytokines on the activation of p38, ERK, and JNK phosphorylation within 5–10 min before the phosphorylated forms gradually weakened. Our results support the notion that signal transduction in elephants is similar to that in humans and that to elephant is chondrocytes. applicable test kit this Conclusions Overall, the findings of this study provide insight into in the molecular mechanisms of OA pathogenesis ELACs, which share similarities with those occurring in humans and other animals. In addition, anti-arthritic drugs commonly used to treat OA in humans and other animals were found to ameliorate the expression of fac- tors associated with arthritis, including proinflammatory cytokines and enzymes responsible for cartilage degener- ation. The present study provides data that contribute to the development of treatments for elephants with OA and support research into arthritis in this species. Methods Preparation of primary ELACs A stillborn elephant calf was caused by dystocia with no clinical appearance of joint disease in an elephant camp in Chiang Mai, Thailand. Cartilage samples from the femoral head of the stifle joint were aseptically collected within 6 h postmortem during the necropsy process, which was consented by the owner. Primary ELACs were isolated by overnight digestion with type II collagenase at 37 °C. The ELACs were washed with phosphate- buffered saline and grown in Dulbecco’s Modified Eagle Medium (DMEM) containing 10% v/v fetal calf serum (FCS), penicillin (100 U/mL), and streptomycin (100 μg/ mL) in a humidified incubator at 37 °C with 5% CO2 until confluence. Monolayer culture and cytokine treatment of ELACs ELACs at a 3 × 105 cells/well density were grown to con- fluence in DMEM containing 10% FCS. The ELACs were sustained in serum-free DMEM for 24 h, after which they were cytokines treated with proinflammatory (ProSpec, Rehovot, Israel), IL-1β (2.5 ng/mL), IL-17A (5 ng/mL), and TNF-α (5 ng/mL), either alone or in com- bination with OSM (2 ng/mL) for 24 h or with IL-17A (5 ng/mL) for 24 h. The ELACs were also treated with various concentrations of 0.125–1 μg/mL LPS (Sigma- Aldrich, U.S.A.). After 24 h, the cells were harvested, and the expression of MMP3 and MMP13 was investigated by real-time RT-PCR. Pellet culture and cytokine treatment of ELACs ELACs at 1 × 106 were centrifuged in 15 mL conical cul- ture tubes at 1500 rpm for 5 min. The pellets that formed at the bottom of the tube were cultured for seven days in 500 μl of chondrogenic medium (DMEM containing 10% FCS, 1X Insulin-Transferrin-Selenium − 7 M dexa- [59], 25 μg/mL ascorbic acid-2 phosphates, 10 methasone) in a humidified incubator at 37 °C and 5% CO2 to allow for spherical shape formation of each pel- let. The pellets were then further treated with IL-1β (5 ng/mL) and TNF-α (10 ng/mL), alone or in combination with OSM (4 ng/mL), for 3 days before being harvested for MMP3 and MMP13 mRNA expression analysis by real-time RT-PCR. Treatment with drugs and natural compounds ELACs in monolayer cultures were treated with a combin- ation of 2.5 ng/mL IL-1β and 2 ng/mL OSM or 0.5 μg/mL LPS for 2 h [60]. Following this, they were treated with drugs, including diacerein (2.5–10 μM; TRB Chemidica, Italy), dexamethasone (5–20 nM; Sigma-Aldrich, U.S.A.), indomethacin (2.5–10 μM; Sigma-Aldrich, U.S.A.), and etoricoxib (2.5–10 μM; Zuelling, Philippines) or with nat- ural bioactive compounds (Sigma-Aldrich, U.S.A.), includ- ing sesamin (0.25–1 μM), andrographolide (1.25–5 μM), and vanillylacetone (20–80 μM), for 24 h. The cells were then harvested to investigate the expression of MMP3 and MMP13 by real-time RT-PCR, and the culture media were analyzed for protein levels of MMP-3 and MMP-13. Real-time RT-PCR Total RNA was extracted from the ELACs obtained from the monolayer or pellet cultures using the Illustra RNAs- pin Mini RNA Isolation Kit (GE Healthcare Life Sciences, U.K.), according to the manufacturer’s protocol. The total (0.25 μg) the monolayer (0.5 μg) and pellet RNA of cultures was reverse transcribed into complementary DNA using the ReverTra Ace® qPCR RT Master Mix (TOYOBO, Japan). The elephant primer sequences were designed based on the NCBI Primer-BLAST tool in asso- ciation with GenBank accession numbers and synthesized by Bio Basic, Canada (Table 1). Real-time RT-PCR was performed using the SensiFAST™ SYBR No-ROX Kit Sirikaew et al. BMC Veterinary Research (2019) 15:419 Page 11 of 13 Table 1 Real-time RT-PCR primer sequences Gene MMP3 MMP13 IL1β IL6 TNFα GAPDH Primer sequence (5′-3′) Forward: AAAGGCAGGCATTTTTGGCG Reverse: AGGGTGAGGGTAGCTCTCG Forward: AGTTCCAAAGGCTACAACTT Reverse: CGCCAGAAGAATCTGTCTTT Forward: CTTGGTGCTTTCTGGTCCTTAT Reverse: AGACAAATCGCTTTTCCATCCT Forward: GGCACTGGCAGGAAACAATC Reverse: GCATTTGCAGTTGGGTCAGG Forward: ATCAGCCGTATCGCTGTCTC Reverse: CCAAAGTAGACCTGCCCAGA Forward: ATCACTGCCACCCAGAAGA Reverse: TTTCTCCAGGCGGCAGGTCAG (Bioline, U.K.). Gene expression quantification was −ΔΔCt method against the expression of based on the 2 the glyceraldehydes-3-phosphate dehydrogenase gene (GAPDH) as a housekeeping gene [61]. Measurement of MMP-3 and MMP-13 levels in the culture media The levels of MMP-3 and MMP-13 enzymes in the culture media were measured using human MMP-3 (catalog number: E-CL-H0931) and MMP-13 (catalog number: E-CL-H0127) sandwich ELISA kits (Elabscience, China), according to the manufacturer’s instructions. Briefly, 100 μl of MMP-3 or MMP-13 standard and sam- ple (culture media) was added to the monoclonal antibody against the proteins (MMP-3 or MMP-13) pre-coated mi- cro CLIA plate well, then incubated at 37 °C. After 90 min of incubation, the standard and sample were discarded, and 100 μl of a biotinylated detection antibody working solution was added to each well. The plate was incubated for 1 h at 37 °C, followed by three washings. A horseradish peroxidase conjugate (HRP) working solution was then added to each well (100 μl/well) and left to incubate at 37 °C for 30 min. After washing, 100 μl of substrate mix- ture solution was added to each well before being incu- bated in the dark for 5 min at 37 °C. The luminescence value was detected using a Synergy H4 hybrid multi-mode microplate reader (BioTek, U.S.A.), and the protein con- centrations were calculated by comparing the samples with standard curves. Western blot analysis of intracellular signaling molecules ELACs were treated with a combination of the cytokines IL-1β (2.5 ng/mL) and OSM (2.5 ng/mL) at various time points. To investigate the activation of the MAPK path- way, the cells were collected in a radioimmunoprecipita- tion assay buffer. The cell lysates were vortexed every few minutes before centrifugation at 14,000 g for 10 min at 4 °C, after which the supernatants of the cell lysate were transferred into new tubes. The cells were lysed with a sample buffer containing 5% mercaptoethanol. Equal amounts (25 μg protein) of the cell lysates were heated for 10 min at 95 °C then subjected to 13% SDS- PAGE and transferred to a nitrocellulose membrane. After blocking non-specific proteins with 5% skim milk in TBS containing 0.1% Tween 20 (TBS-T) for 1 h, the membranes were washed with TBS-T and probed with primary antibodies (Cell Signaling Technology, U.S.A.), including rabbit anti-phosphorylated-p38 MAPK anti- body, rabbit anti-phosphorylated-p44/42 MAPK anti- body, rabbit anti-phosphorylated-SAPK/JNK antibody, rabbit anti-p44/42 rabbit anti-p38 MAPK antibody, MAPK antibody, rabbit anti-SAPK/JNK antibody, and mouse anti-β-actin (Biolegend, CA), at 4 °C overnight. After being washed with TBS-T, the membranes were incubated for 1 h with the secondary antibody conju- gated with HRP anti-rabbit IgG or anti-mouse IgG at room temperature. The positive bands were visualized by enhanced chemiluminescence using the ChemiDoc system (Bio-Rad, U.S.A.). The intensity of the immuno- positive bands was calculated using the TotalLab TL120 software. Statistical analysis The results are presented as the mean ± standard error of the mean of three independent experiments. The stat- istical analysis was performed using one-way analysis of variance followed by LSD for post-hoc multiple compar- isons. A level of p < 0.05 was considered statistically significant. Supplementary information Supplementary information accompanies this paper at https://doi.org/10. 1186/s12917-019-2170-8. Additional file 1. The effect of natural compounds on elephant articular chondrocytes viability by using MTT assay. Abbreviations ELACs: Elephant articular chondrocytes; FCS: Fetal calf serum; IL- 17A: Interleukin-17A; IL-1β: Interleukin-1beta; LPS: Lipopolysaccharides; MAPK: the mitogen-activated protein kinase; MMP: Matrix metalloproteinase; NSAIDs: Non-steroidal anti-inflammatory drugs; OA: Osteoarthritis; OSM: Oncostatin M; TNF-α: Tumor necrosis factor-alpha Acknowledgements The authors gratefully acknowledge all general support throughout the research process from Thailand Excellence Center for Tissue Engineering and Stem Cells, Department of Biochemistry, Faculty of Medicine, Chiang Mai University, Thailand. In addition, we wish to thank Miss Pianghathai Yavirach for her valuable suggestions in some parts of molecular analysis. Authors‘contributions SO and SC designed the experiments and applying for Grants; S.O. contributed as a project administrator; ST, WT, CS, and CT worked out for the ethical approval and collected the animal tissues; N.S. performed the experiments; SO and N.S. analyzed the data; N.S. and S.O. prepared the original draft of manuscript; SC, ST, WT, CS, and CT, NS and SO revised and edited the manuscript. All the authors mentioned in this manuscript have Sirikaew et al. BMC Veterinary Research (2019) 15:419 Page 12 of 13 agreed for authorship, read and approved the manuscript, and given consent for submission and subsequent publication of the manuscript. Funding This research work was supported by Thailand and the National Research Council of Thailand (Government Budget; 2015). The funder had no role in study. Availability of data and materials The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request. Ethics approval Animal use and all procedures in the present study were approved by the Animal Care and Use Committee, Faculty of Veterinary Medicine, Chiang Mai University, Thailand (FVM–ACUC; Ref. No. R22/2559). We obtained written informed consent to use the deceased elephant from the owner of the elephant camp. Consent for publication Not applicable. Competing interests The authors declare that they have no competing interests. Author details 1Thailand Excellence Center for Tissue Engineering and Stem Cells, Department of Biochemistry, Faculty of Medicine, Chiang Mai University, 110 Intrawarorot Rd., Chiang Mai 50200, Thailand. 2Department of Biology, Faculty of Science, Chiang Mai University, Chiang Mai 50200, Thailand. 3Department of Companion Animal and Wildlife Clinic, Faculty of Veterinary Medicine, Chiang Mai University, Chiang Mai 50100, Thailand. Received: 23 November 2018 Accepted: 8 November 2019 References 1. Kapoor M, Martel-Pelletier J, Lajeunesse D, Pelletier JP, Fahmi H. Role of proinflammatory cytokines in the pathophysiology of osteoarthritis. Nat Rev Rheumatol. 2011;7:1. Loeser RF, Goldring SR, Scanzello CR, Goldring MB. Osteoarthritis: a disease of the joint as an organ. Arthritis Rheum. 2012;64:6. Hittmair KM, Vielgrader HD. Radiographic diagnosis of lameness in African elephants (Loxodonta africana). Vet Radiol Ultrasound. 2000;41:6. Clubb R, Mason G. A review of the welfare of zoo elephants in Europe: RSPCA Horsham, UK; 2002. Im HJ, Li X, Muddasani P, Kim GH, Davis F, Rangan J, et al. Basic fibroblast growth factor accelerates matrix degradation via a neuro-endocrine pathway in human adult articular chondrocytes. J Cell Physiol. 2008;215:2. Egger GF, Witter K, Weissengruber G, Forstenpointner G. Articular cartilage in the knee joint of the African elephant, Loxodonta africana, Blumenbach 1797. J Morphol. 2008;269:1. Brama PA, TeKoppele JM, Beekman B, van El B, Barneveld A, van Weeren PR. Influence of development and joint pathology on stromelysin enzyme activity in equine synovial fluid. Ann Rheum Dis. 2000;59:2. Clements DN, Carter SD, Innes JF, Ollier WE, Day PJ. Analysis of normal and osteoarthritic canine cartilage mRNA expression by quantitative polymerase chain reaction. Arthritis Res Ther. 2006;8:6. 2. 3. 4. 5. 6. 7. 8. 9. Wang M, Sampson ER, Jin H, Li J, Ke QH, Im HJ, et al. MMP13 is a critical target gene during the progression of osteoarthritis. Arthritis Res Ther. 2013;15:1. 10. Chen JJ, Huang JF, Du WX, Tong PJ. Expression and significance of MMP3 in synovium of knee joint at different stage in osteoarthritis patients. Asian Pac J Trop Med. 2014;7:4. 11. Richardson DW, Dodge GR. Effects of interleukin-1β and tumor necrosis 12. factor-α on expression of matrix-related genes by cultured equine articular chondrocytes. Am J Vet Res. 2000;61:6. Li WQ, Dehnade F, Zafarullah M. Oncostatin M-induced matrix metalloproteinase and tissue inhibitor of metalloproteinase-3 genes expression in chondrocytes requires Janus kinase/STAT signaling pathway. J Immunol. 2001;166:5. 13. 14. 15. Tung J, Fenton J, Arnold C, Alexander L, Yuzbasiyan-Gurkan V, Venta P, et al. Recombinant equine interleukin-1ß induces putative mediators of articular cartilage degradation in equine chondrocytes. Can J Vet Res. 2002;66. Sylvester J, Liacini A, Li WQ, Zafarullah M. Interleukin-17 signal transduction pathways implicated in inducing matrix metalloproteinase-3, −13 and aggrecanase-1 genes in articular chondrocytes. Cell Signal. 2004;16:4. Koshy PJ, Henderson N, Logan C, Life PF, Cawston TE, Rowan AD. Interleukin 17 induces cartilage collagen breakdown: novel synergistic effects in combination with proinflammatory cytokines. Ann Rheum Dis. 2002;61:8. 16. Hui W, Rowan AD, Richards CD, Cawston TE. Oncostatin M in combination with tumor necrosis factor alpha induces cartilage damage and matrix metalloproteinase expression in vitro and in vivo. Arthritis Rheum. 2003;48:12. 17. Durigova M, Roughley PJ, Mort JS. Mechanism of proteoglycan aggregate degradation in cartilage stimulated with oncostatin M. Osteoarthr Cartil. 2008;16:1. 18. Campo GM, Avenoso A, Campo S, D'Ascola A, Traina P, Sama D, et al. Glycosaminoglycans modulate inflammation and apoptosis in LPS-treated chondrocytes. J Cell Biochem. 2009;106:1. 19. Huang ZY, Stabler T, Pei FX, Kraus VB. Both systemic and local lipopolysaccharide (LPS) burden are associated with knee OA severity and inflammation. Osteoarthr Cartil. 2016;24:10. Loeser RF, Erickson EA, Long DL. Mitogen-activated protein kinases as therapeutic targets in osteoarthritis. Curr Opin Rheumatol. 2008;20:5. 20. 21. West G. Musculoskeletal system. Biology, medicine, and surgery of elephants Ames. IA: Blackwell Publishing; 2006. 22. Qvist P, Bay-Jensen AC, Christiansen C, Dam EB, Pastoureau P, Karsdal MA. 23. 24. The disease modifying osteoarthritis drug (DMOAD): is it in the horizon? Pharmacol Res. 2008;58:1. Sadowski T, Steinmeyer J. Effects of non-steroidal antiinflammatory drugs and dexamethasone on the activity and expression of matrix metalloproteinase-1, matrix metalloproteinase-3 and tissue inhibitor of metalloproteinases-1 by bovine articular chondrocytes. Osteoarthr Cartil. 2001;9:5. Legendre F, Bogdanowicz P, Martin G, Domagala F, Leclercq S, Pujol JP, et al. Rhein, a diacerhein-derived metabolite, modulates the expression of matrix degrading enzymes and the cell proliferation of articular chondrocytes by inhibiting ERK and JNK-AP-1 dependent pathways. Clin Exp Rheumatol. 2007;25:4. 25. Yang SF, Hsieh YS, Lue KH, Chu SC, Chang IC, Lu KH. Effects of nonsteroidal anti-inflammatory drugs on the expression of urokinase plasminogen activator and inhibitor and gelatinases in the early osteoarthritic knee of humans. Clin Biochem. 2008;41:1–2. Li Y, Wang Y, Chubinskaya S, Schoeberl B, Florine E, Kopesky P, et al. Effects of insulin-like growth factor-1 and dexamethasone on cytokine-challenged cartilage: relevance to post-traumatic osteoarthritis. Osteoarthr Cartil. 2015;23:2. 26. 27. Varga Z, Sabzwari SRA, Vargova V. Cardiovascular risk of nonsteroidal anti- inflammatory drugs: an under-recognized public health issue. Cureus. 2017;9:4. 28. Phitak T, Pothacharoen P, Settakorn J, Poompimol W, Caterson B, Kongtawelert P. Chondroprotective and anti-inflammatory effects of sesamin. Phytochemistry. 2012;80. 29. Ding QH, Ji XW, Cheng Y, Yu YQ, Qi YY, Wang XH. Inhibition of matrix metalloproteinases and inducible nitric oxide synthase by andrographolide in human osteoarthritic chondrocytes. Mod Rheumatol. 2013;23:6. 31. 30. Ruangsuriya J, Budprom P, Viriyakhasem N, Kongdang P, Chokchaitaweesuk C, Sirikaew N, et al. Suppression of cartilage degradation by Zingerone involving the p38 and JNK MAPK signaling pathway. Planta Med. 2017;83:3–04. Tangyuenyong S, Viriyakhasem N, Aungsuchawan S, Peansukmanee S, Thitaram C, Kongtawelert P, et al. Catabolism of Asian elephant cartilage matrix biomolecules in explant culture. KKU Vet J. 2012;22:2. Edmondson R, Broglie JJ, Adcock AF, Yang L. Three-dimensional cell culture systems and their applications in drug discovery and cell-based biosensors. Assay Drug Dev Technol. 2014;12:4. 32. 33. Moran EM, Mullan R, McCormick J, Connolly M, Sullivan O, Fitzgerald O, et al. Human rheumatoid arthritis tissue production of IL-17A drives matrix and cartilage degradation: synergy with tumour necrosis factor-alpha, Oncostatin M and response to biologic therapies. Arthritis Res Ther. 2009;11:4. 34. Askari A, Naghizadeh MM, Homayounfar R, Shahi A, Afsarian MH, Paknahad A, et al. Increased serum levels of IL-17A and IL-23 are associated with decreased vitamin D3 and increased pain in osteoarthritis. PLoS One. 2016;11:11. Sirikaew et al. BMC Veterinary Research (2019) 15:419 Page 13 of 13 inhibition of MAPK signaling pathways in rat articular chondrocytes. Drug Dev Res. 2017;78:8. 58. Hellman NE, Spector J, Robinson J, Zuo X, Saunier S, Antignac C, et al. Matrix metalloproteinase 13 (MMP13) and tissue inhibitor of matrix metalloproteinase 1 (TIMP1), regulated by the MAPK pathway, are both necessary for Madin-Darby canine kidney tubulogenesis. J Biol Chem. 2008;283:7. Su SC, Tanimoto K, Tanne Y, Kunimatsu R, Hirose N, Mitsuyoshi T, et al. Celecoxib exerts protective effects on extracellular matrix metabolism of mandibular condylar chondrocytes under excessive mechanical stress. Osteoarthritis Cartilage. 2014;22:6. 59. 60. Zheng X, Xia C, Chen Z, Huang J, Gao F, Li G, et al. Requirement of the phosphatidylinositol 3-kinase/Akt signaling pathway for the effect of nicotine on interleukin-1beta-induced chondrocyte apoptosis in a rat model of osteoarthritis. Biochem Biophys Res Commun. 2012;423:3. 61. Al-Shanti N, Saini A, Stewart CE. Two-step versus one-step RNA-to-CT 2-step and one-step RNA-to-CT 1-step: validity, sensitivity, and efficiency. J Biomol Tech. 2009;20:3. Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. 35. Dvorak LD, Cook JL, Kreeger JM, Kuroki K, Tomlinson JL. Effects of carprofen and dexamethasone on canine chondrocytes in a three-dimensional culture model of osteoarthritis. Am J Vet Res. 2002;63:10. 36. Cortial D, Gouttenoire J, Rousseau C, Ronzière M-C, Piccardi N, Msika P, et al. Activation by IL-1 of bovine articular chondrocytes in culture within a 3D collagen-based scaffold. An in vitro model to address the effect of compounds with therapeutic potential in osteoarthritis. Osteoarthr Cartil. 2006;14:7. 37. Nasi S, Ea H-K, So A, Busso N. Revisiting the role of Interleukin-1 pathway in osteoarthritis: interleukin-1α and-1β, and NLRP3 Inflammasome are not involved in the pathological features of the murine Menisectomy model of osteoarthritis. Front Pharmacol. 2017;8. 38. Plater-Zyberk C, Buckton J, Thompson S, Spaull J, Zanders E, Papworth J, et al. Amelioration of arthritis in two murine models using antibodies to Oncostatin M. Arthritis Rheum. 2001;44:11. 40. 39. Rowan AD, Hui W, Cawston TE, Richards CD. Adenoviral gene transfer of interleukin-1 in combination with oncostatin M induces significant joint damage in a murine model. Am J Pathol. 2003;162:6. Kamekura S, Hoshi K, Shimoaka T, Chung U, Chikuda H, Yamada T, et al. Osteoarthritis development in novel experimental mouse models induced by knee joint instability. Osteoarthr Cartil. 2005;13:7. Li H, Wang D, Yuan Y, Min J. New insights on the MMP-13 regulatory network in the pathogenesis of early osteoarthritis. Arthritis Res Ther. 2017;19:1. 41. 42. Naito K, Takahashi M, Kushida K, Suzuki M, Ohishi T, Miura M, et al. Measurement of matrix metalloproteinases (MMPs) and tissue inhibitor of metalloproteinases-1 (TIMP-1) in patients with knee osteoarthritis: comparison with generalized osteoarthritis. Rheumatology (Oxford). 1999;38:6. 43. Yoon JB, Kim SJ, Hwang SG, Chang S, Kang SS, Chun JS. Non-steroidal anti- inflammatory drugs inhibit nitric oxide-induced apoptosis and dedifferentiation of articular chondrocytes independent of cyclooxygenase activity. J Biol Chem. 2003;278:17. 44. Moldovan F, Pelletier J, Jolicoeur F-C, Cloutier J-M, Martel-Pelletier J. 45. Diacerhein and rhein reduce the ICE-induced IL-1β and IL-18 activation in human osteoarthritic cartilage. Osteoarthr Cartil. 2000;8:3. Khadem Haghighian M, Alipoor B, Malek Mahdavi A, Eftekhar Sadat B, Asghari Jafarabadi M, Moghaddam A. Effects of sesame seed supplementation on inflammatory factors and oxidative stress biomarkers in patients with knee osteoarthritis. Acta Med Iran. 2015;53:4. 46. Murata J, Matsumoto E, Morimoto K, Koyama T, Satake H. Generation of 47. 48. triple-transgenic Forsythia cell cultures as a platform for the efficient, stable, and sustainable production of Lignans. PLoS One. 2015;10:12. Kong P, Chen G, Jiang A, Wang Y, Song C, Zhuang J, et al. Sesamin inhibits IL-1beta-stimulated inflammatory response in human osteoarthritis chondrocytes by activating Nrf2 signaling pathway. Oncotarget. 2016;7:50. Kim MK, Chung SW, Kim DH, Kim JM, Lee EK, Kim JY, et al. Modulation of age-related NF-κB activation by dietary zingerone via MAPK pathway. Exp Gerontol. 2010;45:6. 49. Altman RD, Marcussen KC. Effects of a ginger extract on knee pain in 50. 51. patients with osteoarthritis. Arthritis Rheum. 2001;44:11. Li ZZ, Tan JP, Wang LL, Li QH. Andrographolide benefits rheumatoid arthritis via inhibiting MAPK pathways. Inflammation. 2017;40:5. Li F, Li H, Luo S, Ran Y, Xie X, Wang Y, et al. Evaluation of the effect of andrographolide and methotrexate combined therapy in complete Freund's adjuvant induced arthritis with reduced hepatotoxicity. Biomed Pharmacother. 2018;106. 52. Hidalgo MA, Hancke JL, Bertoglio JC, Burgos RA. Andrographolide a new potential drug for the long term treatment of rheumatoid arthritis disease. Innovative Rheumatology: IntechOpen; 2013. 53. Blasioli DJ, Kaplan DL. The roles of catabolic factors in the development of osteoarthritis. Tissue Eng Part B Rev. 2014;20:4. 54. Huang Z, Kraus VB. Does lipopolysaccharide-mediated inflammation have a 55. 56. 57. role in OA? Nat Rev Rheumatol. 2016;12:2. Kim HA, Cho ML, Choi HY, Yoon CS, Jhun JY, Oh HJ, et al. The catabolic pathway mediated by toll-like receptors in human osteoarthritic chondrocytes. Arthritis Rheum. 2006;54:7. Shi J, Zhang C, Yi Z, Lan C. Explore the variation of MMP3, JNK, p38 MAPKs, and autophagy at the early stage of osteoarthritis. IUBMB Life. 2016;68:4. Li X, Guo Y, Huang S, He M, Liu Q, Chen W, et al. Coenzyme Q10 prevents the Interleukin-1 Beta induced inflammatory response via
null
10.1088_1402-4896_ad0a2a.pdf
Data availability statement The data that support the findings of this study are openly available at the following URL/DOI: https://orcid. org/0000-0003-3389-9318.
Data availability statement The data that support the findings of this study are openly available at the following URL/DOI: https://orcid. org/0000-0003-3389-9318 .
Phys. Scr. 98 (2023) 125949 https://doi.org/10.1088/1402-4896/ad0a2a PAPER RECEIVED 21 July 2023 REVISED 27 October 2023 ACCEPTED FOR PUBLICATION 6 November 2023 PUBLISHED 17 November 2023 Electroplating of hydrophobic/hydrophilic ZnO nano-structural coatings on metallic substrates Zehira Belamri , Leila Boumaza and Smail Boudjadar Department of Physics, Phase Transformation Laboratory, Frères Mentouri_Constantine 1 University, Constantine 25000, Algeria E-mail: belamri.zehira@umc.edu.dz Keywords: hydrophobic, hydrophilic, wettability properties, contact angle, ZnO nanostructure Abstract In the present work, ZnO thin film is shown as a coating on an aluminum substrate. In order to synthesize ZnO thin films, electroplated Zn thin layers were thermally oxidized in atmospheric air for different times (1h–4h) at a fixed annealing temperature of 500 °C. The samples were characterized by scanning electron microscopy (FEG-SEM) equipped with energy dispersive x-ray analysis (EDX), a profilometer, x-ray diffraction (XRD), and Raman spectroscopy. The wettability properties of the synthesized films were evaluated by measuring the contact angle between the surface of the films and a deposited water drop (WCA). The FEG-SEM images show that the surface morphologies change throughout treatment time. The sample treated for 2 h shows flower-like microstructures with an average size of 100 μm, which are covered with spherical ZnO nanostructures with a size less than 50 nm. Measured surface roughness ranges from 5.800 μm to 6.560 μm. Layers thicknesses vary between 31 and 38 μm. Structural characterization by XRD demonstrates that the synthesized ZnO thin films were polycrystalline and have Wurtzite hexagonal structures, grown manly along the (101) plan. The estimated crystallite sizes are in the nanometric scale and reach their maximum value for the sample treated for 2 h. This annealing time corresponds to the low dislocation density (δ) and low lattice strain (ε), indicating fewer defects. The Raman analysis shows five normal vibrational modes, which correspond to the ZnO Wurtzite structure. It was possible to obtain both hydrophobic and hydrophilic surfaces; the shape and surface roughness of the as-prepared films had an impact on the results. The largest measured contact angle, of 97°, was obtained after annealing for 2 h at 500 °C. 1. Introduction Because of their higher mechanical and physicochemical characteristics, metals including steel, copper, and aluminum are frequently utilized in industrial applications and in daily activities [1–4]. However, it is easily corroded in the environment as chemical or electrochemical reactions occur on the metal’s surface to turn it into an oxidized or ionic state [5, 6]. Stress corrosion cracking and corrosion fatigue will occur in the metal material as a result, which will significantly reduce its mechanical properties. One of the best ways to prevent corrosion on metal surfaces is to form a passivation coating [2, 3]. The most important quality of the material surface is wettability, which deals with the surface’s affinity for water. It is widely known that a surface’s wettability depends mostly on its surface energy, surface roughness, or topography surface micro-nano structure and chemical composition [13–15]. Hydrophobic thin coatings have many applications, including making waterproof clothing, self-cleaning surfaces, anti-fog coatings, and anti- corrosion coatings to minimize the loss of efficiency in photovoltaic cells and biomedical devices [7–12]. As an accepted rule in a scientific community, surfaces with water contact angle θ < 90° are hydrophilic, those have contact angle θ > 90° are considered hydrophobic and superhydrophobic when θ > 145° [16]. Among these materials used as coating thin film is zinc oxide (ZnO), which is an extremely promising material because of its abundance and exceptional physicochemical properties, such as a large direct band gap (3.37 eV), high excitonic energy (60 meV), stability, and biocompatibility [17–19]. Zinc oxide (ZnO) can exist in © 2023 IOP Publishing Ltd Phys. Scr. 98 (2023) 125949 Z Belamri et al several different crystalline structures, also known as polymorphs, depending on the conditions of its synthesis and processing. The Wurtzite structure is the most stable and commonly observed form of ZnO. The zinc blend structure is a high-pressure phase, and the rocksalt structure is known as a high-temperature phase [20]. Several studies have been carried out on this subject. Depending on the wetting properties, roughness, durability, and anti-icing performance, water-repellent surfaces on aluminum substrates have been demonstrated [21]. The hydrophobic self-cleaning surfaces of transparent ZnO thin films have been reported to be controllable through surface homogeneity manipulations [22]. A successful method for inducing specially patterned PAA substrates to produce superhydrophobic lotus-leaf-like ZnO micro-nanostructured films with extremely strong adhesion forces was demonstrated [23]. Hydrophobic and hydrophilic ZnO thin films can be applied to self-cleaning glass [24]. The combination of superhydrophilic and hydrophobic photocatalytic effectivity makes ZnO a promising choice for the use of self-cleaning glass [25]. It was found that the structural properties of ZnO thin films have a major influence on wettability behavior as well as electrical, optical, and photocatalytic properties [26]. There are many methods for synthesizing ZnO thin films. However, the selection of the synthesis method depends on the specific requirements of the application, the desired thickness, uniformity of the film, and surface states, as well as the equipment and resources available. Sputtering, Molecular Beam Epitaxy (MBE), Pulse Laser Deposition (PLD), and Metal–Organic Chemical Vapor Deposition (MOCVD) are considered as complicated and expensive methods, which involve sophisticated experimental apparatus and specific conditions [27–32]. Soft chemistry methods such as spray pyrolysis, dip- and spin-coating, and electrodeposition are regarded as relatively easy and inexpensive techniques [33–37]. The objective of this research is to examine the impact of annealing time on the morphological and structural characteristics of zinc oxide thin films grown on aluminum cathode substrates using a low-cost, simple electroplating technique. Additionally, contact angle measurement will be used to determine the wettability (hydrophobic and hydrophilic) of the layers. 2. Experiment 2.1. Preparation of the substrate In this present work, pure aluminum is used as a substrate; it must undergo mechanical polishing with abrasive paper of 600 until obtaining a flat shape with a thickness of about 2 mm. Before use, the substrates have been cleaned in ultrasound baths for 15 min with distilled water and methanol, respectively. 2.2. Deposition of thin films To make an aqueous solution, 1.755 g of dehydrated zinc acetate (Zn (CH3COO)2−2H2O) precursors were dissolved in distilled water (0.2 M). Acetic acid (CH3COOH) was used as the complexing agent. A cleaned aluminum substrate is used as a cathode and the platinum plate as an anode; both electrodes were vertically immersed in the as-prepared solution and kept at a distance of 1.5 cm. A voltage of −10 V DC was applied for 15 min. The bath temperature is maintained at 50 °C to activate chemical reactions. After deposition, the thin Zn layers of 37 μm were annealed at different times at 500 °C (1 h, 2 h, 3 h, and 4 h). In order to ensure the complete oxidation of Zn, it became clear through our previous work [38] that temperatures below 500 °C are insufficient for generating a full ZnO compound. 2.3. Characterizations The morphological and elemental analyses were performed using a Field Emission Gun Scanning Electron Microscope (FEG-SEM, JEOL FEG JSM-7100F) equipped with an energy dispersive x-ray spectrometer (EDS). The thickness and roughness of deposited ZnO films were measured on a PCE-RT 1200 model profilometer. The crystallographic properties of the as-prepared samples were determined using a PANALYTICAL empyrean diffractometer (XRD, λCu = 1.540 Å). The data from XRD were analyzed using X’Pert High Score software. The Raman spectra were measured on a HORIBA LabRAM HR Evolution spectrometer at room temperature with a monochromatic light source of 473 nm. In this work, we used a commercial ZnO powder that was purchased from Fluka Analytical company in order to compare the vibration modes of this compound to those of the studied samples.The measurement of the water contact angle (WCA (θ)) has been used to determine the surface wettability of the elaborated ZnO thin films. A light source of the LEYBOLD type (6 V, 30 W) was used for lighting and projecting the drop’s image onto the sample, together with a projection lens that allowed the image to be magnified onto a transparent screen of dimensions 30 × 30 cm2. 2 Phys. Scr. 98 (2023) 125949 Z Belamri et al Figure 1. FEG-SEM image of electrodeposited Zn thin film (untreated sample). 3. Results and discussions 3.1. Morphological studies Figure 1 shows the FEG-SEM image of the electrodeposited Zn thin layer. The image shows that this layer has an overlapping structure of micrometric hexagonal sheets. The morphology of the surface of ZnO thin films electrodeposited on an Al substrate with different annealing times is presented in figures 2 (a)–(d). On the left, show the low and high magnification (inset) FEG-SEM images of ZnO thin films treated at 500 °C for different times (1 h, 2 h, 3 h, and 4) , and on the right, the corresponding EDS analysis. The images show that there are some changes in the morphologies of samples as treatment time increases. For figures (a), (b), and (c), at the micrometric scale, there are fissures, pores, flower-like structures, and digging on the surface of films. From figure 2(b), extending the thermal oxidation time to 2 h induced the formation of a rough surface with flower-like microstructures of about 100 μm. Each flower-like is consists of conical pores. These micro-flowers are distributed randomly, which increases the roughness of the surface. Similar morphology was observed in previous work [39]. The surface of these cones is covered by a large number of uniform spherical ZnO nanostructures of about 50 nm in size, as shown in the magnified images. For the sample that was heated to 500 °C for 4 h (figure 2(d)), the thin film’s surface exhibits a vertical mico- sheet structure with a needle-like structure at the nonmetric scale. To confirm the chemical composition of the ZnO films, an EDX analysis is performed (the right of figure 2). It shows that zinc and oxygen were the only elements present in the films. Height Zn and O peak intensities in the sample indicated preponderance of ZnO and within the EDX detection limit; no traces of contaminants were discovered. While Zn was normalized to 1 at the atomic percentage of Zn and O in ZnO, the O:Zn ratios were 1.3, 1.18, 1.24 and 1.01 in sample treated 1 h, 2 h, 3 h, and 4 h respectively. With the exception of the sample processed for four hours, where the stoichiometry is almost achieved (table 1), this might be a result of the diffusion of oxygen toward the bulk of the sample or the desorption phenomenon. As known, ZnO is the most stable metallic oxide compound; consequently, it is not affected by annealing time, where the stoichiometric characteristic is always verified. The thickness of the studied thin layers ranged between 31 and 38 μm, and the surface roughness increased with the increase in annealing time from 5.800 μm to 6.560 μm. The coexistence of surface roughness (micro- nanostructure) and low surface energy coating is crucial for surfaces that exhibit superhydrophobicity, as highlighted by a number of previously published works on superhydrophobic surfaces made using water [39, 40]. 3.2. Structural studies The identification of the substrate structure, electrodeposited Zn, and ZnO thin films was carried out by comparison with existing databases in the form of ICSD cards N° : 1109-00-00–/03-065- 3358/00-036-1451, respectively. After the designation of the substrate peaks, the x-ray diffraction spectrum of the untreated sample shows Zn peaks where they appear to crystallize well and orient in the most intense direction (figure 3 (b)).The 3 Phys. Scr. 98 (2023) 125949 Z Belamri et al Figure 2. FEG-SEM images of electrodeposited ZnO thin films annealed at 500 °C for 1 h (a), 2 h (b), 3 h (c) and 4 h. (d), (inset shows a high magnification FEG-SEM images of ZnO thin films) and on the right the corresponding EDX analysis. various diffraction peaks of Zn are observed at 2θ = 36.366°, 39.086°, 43.321°, 54.447°, 70.244°, 70.815°, and 77.233 which correspond to (002), (100), (101), (102), (103), (110), and (004) lattice planes, respectively (figure 3 (b)). 4 Phys. Scr. 98 (2023) 125949 Z Belamri et al Figure 3. XRD spectra of (a) substrate and (b)–(f) untreated and treated electrodeposited Zn thin films. Table 1. Oxygen–Zinc atomic ratio in all samples as function of treatment time. Treatment time, h O:Zn Atomic ratio 1h 1.3 2h 3h 4h 1.18 1.24 1.01 After annealing for1h at 500 °C, the different diffraction peaks are observed at 2θ = 31.770°, 34.422°, 36.253°, 47.539°, 56.603°, and 62.864°, which attributed respectively to the (100), (002), (101), (102), (110) and (103) lattice planes of the Wurtzite hexagonal structure (figures 3 (c)–(f)). However, there is a shift in the peaks position for the other spectra of the samples annealed for 2 h, 3 h, and 4 h, which will be discussed later. For all samples, the planes (100), (002), and (101) have the sharpest peaks with the greatest intensities, which is a sign of good crystallinity and a larger grain boundary dimension [41, 42]. Equation (1) was used to determine the highly oriented plane by calculating the relative peak intensity orientation (hkl) of the three main planes, which is equal to the ratio of the intensity of the (hkl) orientation to the sum of the intensities of the three dominant orientations (100), (002), and (101) in ZnO thin films [43]. Table 2 displays the results of the calculations. *( I hkl ) = ( ) I hkl ( I hkl i å i ) ( ) 1 The diffraction peak at the (101) plane clearly shows the highest intensity, demonstrating that growth is manifest along the (101) plane. The intensity reaches its maximum after two hours of treatment and then begins to decrease as time increases. Plan (002) is the lowest of all samples, indicative of weak preferred growth toward this plan [43]. This can be illustrated by the texture coefficient TC(hkl) calculated using the following relation, equation 2 [44]: 5 Phys. Scr. 98 (2023) 125949 Z Belamri et al Table 2. Intensity ratio of main peaks in ZnO thin films with treatment time. Treatment time at 500 °C *( ) I 100 *( ) I 002 *( ) I 101 1h 2h 3h 4h 0.183 0.110 0.278 0.259 0.155 0.102 0.219 0.228 0.662 0.787 0.502 0.512 Table 3. Crystallite sizes of elaborated ZnO thin films with treatment time. Treatment time at 500 °C 2 q, (101) peak D (nm) 1h 2h 3h 4h 54 93 65 65 36.288 36.233 36.240 36.310 I ( hkl ) TC ( hkl ) = 1 N hkl ) ( 0 I å I ( hkl ) N I 0 ( hkl ) ( ) 2 Where: TC(hkl): the texture coefficient of (hkl) plane, I(hkl): the XRD peak intensities obtained from the films, I0(hkl): the intensities of the standard diffraction pattern (JCPDS card 00-036-1451) N is the number of diffraction peaks considered. The obtained calculus shows that the texture coefficient of the (101) peak varies from one sample to another and takes the maximum value for the sample annealed for 2 h (2.044), whose maximum value of the thickness 39.00 μm. As the film thickens, more crystallites are formed via the gathering of more solute [44]. This shows that greater film thickness can enhances the crystallinity of the films. Therefore, the crystalline size can be estimated based on the Scherrer method [14] using the full width at half maximum (FWHM) of the highly oriented peak (101) located around 36.2° in the XRD spectrum, equation 3. D = l 0.9 b cos q ( ) 3 Where λ, θ, and β are the x-ray wavelength (0.1540 nm), Bragg diffraction angle, and FWHM, respectively. Table 3 summarizes the position of the (101) peak and estimated crystallite sizes. The obtained results reveal that the crystallite size is on the nanometric scale. The crystallite size increases from 54 nm for the ZnO thin film treated for 1 h at 500 °C to 93 nm for the sample treated for 2 h at 500 °C. This may be due to the energy input provided by the heat treatment and the interaction between atmospheric oxygen and the surface layer. During annealing, oxygen atoms captured by the surface of the layer can diffuse into the crystallite joints. This can decrease the oxygen-related defects in the film, increase the crystallite sizes, and improve the stœchiometry of thin films. As the treatment time increases to 3 h at 500 °C, the crystallite sizes decrease to 65 nm; this decrease is probably related to the recrystallization of the material following the refining of the crystallites. The observed change confirms that the ZnO layer properly needs a certain amount of energy to crystallize. There is also a shift of the preferentially oriented peak (101) towards higher values than the normalized one (2θ = 36.215°) after a treatment time of 4 h at 500 °C (figure 4). The shift toward a high diffraction angle may be due to decreasing crystallite sizes as treatment time increases. This is because the capillary pressure exerted by grain boundaries in the case of nanomaterials can result in peak shift, high dislocation density, and peak broadening [45]. In order to explain this shift, we can calculate the lattice parameter c using the following relations valid for the hexagonal structure [46–49]: a = l q sin 3 c = l sin q 6 ( ) 4 ( ) 5 Phys. Scr. 98 (2023) 125949 Z Belamri et al Figure 4. Superposition of XRD spectra for the ZnO (101) peak of samples treated at 500 °C for 1 h, 2 h, 3 h, and 4 h. Table 4. Values of a, c, ezz, σ as a function of heat treatment time at 500 °C for elaborated ZnO thin films. Treatment time at 500 °C 1 h 2 h 3 h 4 h a (nm) c (nm) 0.3249 0.3252 0.3252 0.3244 0.5199 0. 5203 0.5203 0.5196 ezz (%) −0.154 −0.077 −0.077 −0.211 σ (GPa) 0.349 0.179 0.179 0.489 4 3 1 2 d hkl = 2 d hkl 2 h ⎜ ⎛ ⎝ sin + hk 2 a ) + 2 k + ⎟ ⎞ ⎠ l= n 2 l c 2 ( ) 6 ( ) 7 ( q hkl Where : λ: The wavelength of the x-ray used (0.1540 nm) θ: The diffraction angle of the peak (100) for the parameter a and of the peak (002) for c parameter. dhkl: The interreticular distance The lattice parameters calculated for the ZnO thin films elaborated in this present work are given in table 4. These values are slightly different from those of the normalized ZnO, which are: a0 = 0.3253 nm and c0 = 0.5213 nm. This indicates that these layers are compressed parallel to their growth direction; this may be due to the difference in the thermal expansion coefficients between the ZnO thin film and the substrate [46]. It is known that the expansion coefficient (α) of aluminum substrate is 23 × 10 structure, its room temperature expansion coefficients α11 and α33 at room temperature are 6.05 × 10 −1 respectively [46]. So regardless of α value of ZnO, the expansion coefficient of the and 3.53 × 10 substrate used is greater than that of ZnO. Consequently, this maladjustment generates stresses in the deposited layer. −1. While ZnO has a hexagonal −1 −6 °C −6 °C −6 °C When the crystal lattices of the substrate and the thin film perfectly accommodate each other, a crystallographic relationship can appear at the interface. A deformation due to a disagreement between the lattice parameters of both materials can also be caused by this accommodation. This type of deformation generates coherence stresses in the two contact materials. The XRD spectra of an elaborated ZnO thin film can be used to determine the state of stress. The biaxial stress ezz along the c-axis direction perpendicular to the substrate is calculated from the following relationship [34]: C 0 ´ 100 ( ) 8 - film C 0 C e zz = 7 Phys. Scr. 98 (2023) 125949 Z Belamri et al Table 5. Structural parameters of elaborated ZnO thin films as a function of the heat treatment time at 500 °C. Treatment time at 500 °C Crystallite size D (nm) Dislocation density δ × 1014 (ligne/m2) Lattice deformation (ε x10 −3) 1 h 2 h 3 h 4 h 54 93 65 65 3,4 1,2 2,4 2,4 0,64 0,38 0,52 0,52 Where cfilm is the lattice parameter of the elaborated thin film and c0 is the lattice parameter of the unconstrained thin film (c0 = 0.5213 nm). We can confirm the type of stress by studying the sign of the ezz parameter. In this present work, the ezz values represented in table 4 are negative, which confirms that this film undergoes a compressive stress parallel to its growth direction. The residual stress parallel to the thin film surface is expressed as follows [50]: 2 13 - 2 C s = ( C C 33 C 2 13 With cij is the elastic constant for a monocrystalline structure of ZnO (c13 = 104.2 GPa, c33 = 213.8 GPa, c11 = 208.8 GPa and c12 = 119.7 GPa [35]). s film C = - - 233xe GPa C C C 12 11 x ( ) ) 0 0 + zz ( ) 9 ( ) 10 It was discovered that there is a relationship between the biaxial stress ezz and the residual stress; they have the opposite direction in the plane of the thin film-substrate interface [46]. The calculated (σ) values of samples treated at 500 °C for different times are presented in table 4. All these values have a positive sign, indicating that the elaborated ZnO thin films are under traction stress perpendicular to the c axis. The change in film stress values can be attributed to the variation in treatment time. When the treatment time increases to 2 h, ZnO crystallizes due to oxygen atom diffusion in the material through the crystallite joints. This could decrease the oxygen-related defects in the film and increase the crystallite size, leading to a decrease in stress. If the treatment time increases again, the stress increases due to the increase in defects, in particular in the crystallite joints after recrystallization. A study of other structural parameters, especially the dislocation density (δ) and the lattice strain (ε), was also carried out to better evaluate the quality state of thin films [47]. The obtained results are shown in table 5. d = b e = 1 2 D cos 4 q ( 10 a ) ( ) 11 Where; D is the average crystallite size. The obtained results show that the low value of (δ) is reached after a heat treatment of 2 h at 500 °C, which indicates the presence of fewer defects in the deposited ZnO thin film. This time corresponds to the best crystallization of the ZnO hexagonal phase. The increase in the crystallite size of this film is the origin of the decrease in stress. When the heat treatment time is sufficient, the dislocations become spontaneously mobile and a reorganization of the crystalline structure, accompanied by an increase in the stress that the elaborated thin film undergoes, which is known as recrystallization. The lattice strain (ε) is mainly due to the lattice shift between the film and the aluminum substrate. The minimum value of (ε) obtained for a treatment time of 2 h at 500 °C indicates very little lattice mismatch between the substrate and the deposited film, with fewer defects in the elaborated ZnO thin film. The decrease in lattice defects as crystallite size increases can be attributed to the presence of sufficiently thicker films in a less strained state. This, in turn, leads to a reduction in lattice strain and dislocation density, signifying an enhancement in crystallinity or a decrease in film intrinsic defects [44]. 3.3. Raman spectroscopy study Raman spectra are more sensitive to crystallinity, structural disorder, chemical composition of materials, and defects in nanostructures. Figures 5 (c)–(f) show the Raman spectra of the samples annealing at 500 °C for 1 h, −1. No vibration mode appears on the 2 h, 3 h, and 4 h, respectively, in wavenumber ranges of 50 to 900 cm Raman spectrum of the untreated sample. All spectra contain the characteristic Raman active modes for the hexagonal Wurtzite structure [48, 49]. The vibrational modes that are strongest are E2 high − E2 high at E2 −1 caused by the sublattice vibration of the oxygen atoms in the ZnO crystal [51, 52]. The high around 443 cm intensity of this peak reflects the crystallization quality of ZnO thin films with a hexagonal Wurtzite structure. Low at 100 cm −1, and E1 (LO) around 587 cm −1, A1 (TO) at 383 cm high at 443 cm Low at 329 cm −1, −1. E2 −1, E2 8 Phys. Scr. 98 (2023) 125949 Z Belamri et al Figure 5. Raman spectra of ZnO powder (a), untreated electrodeposited Zn thin film (b), ZnO thin films treated at 500 °C (c)-(f) for different times. high peak towards high wavenumber values (figure 5(c); This confirms the results obtained from the XRD analysis. In comparison to massive ZnO (figure 5(a), ZnO thin film treated for 1 h at 500 °C results in a shift of the E2 this could be the result of stress in the treated thin films. Other, more intense peaks were observed on the Raman scattering spectra of the samples treated at 500 °C (figure 5 (c)–(f)) around 587 cm E1(LO) modes, which is a Raman-active mode of hexagonal Wurtzite ZnO [51–54], caused by impurities and −1, appears on all the Raman formation defects such as oxygen vacancies. The E2 scattering spectra of the annealed samples and is associated with the vibration of the lattice of zinc atoms −1. The peak at 383 cm [52, 53, 56]. Another two low-intensity peaks appeared near 336 cm attributed to the A1 (TO) mode, which is a first-order optical mode of hexagonal Wurtzite ZnO, based on previous experimental studies. The observed peak at 336 cm characteristic of second order caused by the multiphonon process [54, 55]. −1 was attributed to the mode E2 −1, corresponding to the Low peak, located at 100 cm −1 and 383 cm low, which is a high-E2 −1 is Based on the findings of this investigation, no changes in Raman spectra were observed with an increase in annealing time. Therefore, the increase in annealing time has no effect on the vibration mode of elaborated ZnO. This can be due to several possible reasons: ZnO may have reached a stable crystal structure after a certain annealing time, which means that more annealing time will not result in significant changes in the crystal structure. The vibrations observed in the Raman spectrum can already be associated with all possible characteristics of ZnO under the given annealing conditions. Therefore, extending the annealing time does not add new vibrational characteristics. It is also possible that the annealing time of 2 h at 500 °C is already sufficient to achieve the desired properties of ZnO in this study, and extending the annealing time provides no additional benefit. 3.4. Study of the wettability The wetting characteristics of the elaborated ZnO thin film on the metallic Al substrate were analyzed by measuring the water contact angles (figures 6 (a)–(d)). Their contact angle and annealing time were correlated. The hydrophilic and hydrophobic characteristics of ZnO films were directly influenced by their micromorphology. In this work, the measurements reveal that the maximum value of the contact angles is 97.01°, which was found for a sample treated for 2 h at 500 °C. This indicates that the sample exhibits hydrophobic characteristics. 9 Phys. Scr. 98 (2023) 125949 Z Belamri et al Figure 6. Contact angle with water of ZnO thin films treated: 1 h (a), 2 h (b), 3 h (c) and 4 h (d). The state of the surface of the treated sample for 2 h at 500 °C (figure 2(b)) shows the coexistence of micro- nanostructure. The formation of conical ZnO micropores on the surface of the layers traps air on this surface and prevents water from adhering to the ZnO film, which leads to the hydrophobicity of the material. Additionally, the presence of nanostructures can trap air between them, resulting in air pockets on the surface that inhibit the uniform diffusion of water. Surface roughening can also contribute to an increased contact angle. Rough surfaces tend to trap air pockets, preventing the liquid from spreading and resulting in a higher contact angle. It is found that the surface roughness increases with the increase in annealing time from 5.800 μm to 6.560 μm, accompanied by an increase in contact angle. This may be due to the increased size and density of the micro-clusters of ZnO. Similar results were discussed previously [56], where the authors found that the water contact angle and the surface roughness of the elaborated ZnO thin films on aluminum substrate increased with the increase in bath temperature. The idea of the effect of roughness on the contact angle has been studied where liquid does not penetrate the grooves on a rough surface and leaves air gaps [57]. From the obtained results, it was found that the largest crystallite size corresponds to the largest thickness and roughness, which may be linked to the increased intensity of the (101) peak in the x-ray diffraction spectrum. Such observations are consistent with the decreasing FWHM of the (101) peak. These characteristics correspond to the sample annealed at 2 h, which presents the biggest contact angle. The extension of the treatment time up to 4 h at 500 °C leads to a decrease in the contact angle. This may be due to the decrease in crystallite size and the increase in defects in the ZnO film structure. Thus, the formation of ZnO nano- needles leads to a decrease in the contact angle, and the surface becomes hydrophilic. However, the samples heated to 500 °C for 1, 3, and 4 h exhibit hydrophilic characteristics, with contact angles of 66.1°, 73.37°, and 61.31°, respectively. These might result from the significant large pores and cracks on the surface of thin films and the high density of needles on the surface of the (d) sample. Extended annealing time may allow the ZnO material to reach a more stable or equilibrium state where its surface properties are different from those during the initial phase. Surface reactions, diffusion, or restructuring processes may continue to evolve, leading to a surface that is more amenable to wetting. As the behavior of fluids strongly depends on the hydrophilic nature of the surface, in this case, hydrophilic ZnO films can be used as microchannels. 10 Phys. Scr. 98 (2023) 125949 4. Conclusion Z Belamri et al Stable and stoichiometric ZnO thin films were obtained by simple thermal oxidation of electrodeposited Zn on aluminum substrates. The FEG-SEM analysis shows that the morphologies of ZnO thin layers depend on time treatment. There are fissures, pores, and flower-like structures of about 100 μm on the surface of the ZnO film treated for 2 h. Every flower-like structure has conical pores, which are covered by a large number of uniform nanospheres of about 50 nm in size. The thickness of the studied thin layers changes between 31 and 38 μm, and the surface roughness increases with the increase in annealing time from 5.800 μm to 6.560 μm. The conical flower-like morphology covered with ZnO nanostructures and micropores observed after 2 h at 500 °C can all contribute to the hydrophobicity of the surface by creating rough structures and areas of trapped air. The treatment for 2 h at 500 °C of electroplating Zn layers leads to the best crystallization of the ZnO nanostructure thin films. The obtained results show that the low value of dislocation density (δ) reached after this treatment accompanied by a low value of lattice strain (ε), which indicates the presence of fewer defects in the layers deposited during this treatment time. Results from Raman spectroscopy support those from the XRD analysis. The obtained spectra showed five normal vibrational modes that are consistent with the hexagonal ZnO Wurtzite structure. The study’s results indicate that increasing the annealing time does not alter the Raman spectra, implying that it has no impact on the vibration mode of the ZnO produced in this work. This may be attributed to several factors, such as the stability of the ZnO crystal structure, saturation of the ZnO properties, and optimal annealing conditions. Crystallite size increases, and crystallinity improves with increasing film thickness. These surface characteristics promote the formation of water drops with a high contact angle, indicating better hydrophobicity. The observed changes in the contact angle during treatment likely reflect the dynamic interplay of surface roughness, chemical transformations, and material stability. The initial increase in the contact angle may be due to surface changes that render the surface less wettable, while the subsequent decrease in the contact angle may result from further surface evolution that promotes wetting. It can be deduced that the sample produced with a deposition time of 15 min and a voltage of −10 V followed by treatment for 2 h at 500 °C has perfect hydrophobicity, which means that these conditions are optimal for the manufacture of ZnO hydrophobic coatings. Acknowledgments The authors warmly thank Prof D HAMANA, director of the Ecole Nationale Polytechnique de Constantine (ENPC), Algeria, for her precious help and support. Authors also would like to thank Mr K CHETTAH researcher in the mechanics research center (CRM) of Frères Mentouri-Constantine 1 university, for the roughness and thickness mesurement. Data availability statement The data that support the findings of this study are openly available at the following URL/DOI: https://orcid. org/0000-0003-3389-9318. ORCID iDs Zehira Belamri https://orcid.org/0000-0003-3178-9318 References [1] Liu L J, Xu F Y, Yu Z L and Dong P 2012 Facile fabrication of non-sticking superhydrophobic boehmite film on Al foil Appl. Surf. Sci. 258 8928e8933 [2] Feng L, Che Y, Liu Y, Yan Z, Wang Y and Qiang X 2016 One-step immersion method for fabricating superhydrophobic aluminum alloy with excellent corrosion resistance Surf. Interface Anal. 48 1320e1327 [3] Zheng S, Li F Q, Hu W, Xiang T, Wang Q, Du M, Liu X and Chen Z 2016 Development of stable superhydrophobic coatings on aluminum surface for corrosion-resistant, self-cleaning, and anti-icing applications Mater. Des. 93 261e270 [4] Zhang S, Huang J, Tang Y, Li S, Ge M, Chen Z, Zhang K and Lai Y 2017 Dynamic wettability: understanding the role of dynamic wettability for condensate microdrop self-propelling based on designed superhydrophobic TiO2 nanostructures Small 13 1600687 11 Phys. Scr. 98 (2023) 125949 Z Belamri et al [5] Wang S, Xu C, Hua Y, Ren X, Lu J, Li J, Chen X, Xiang Q and Li Y 2012 Anodic dissolution of aluminum in AlCl3-[BzMIM]Cl ionic liquid J. Electroanal. Chem. 900 115715 [6] Vargel C 2020 Chapter B.1—The Corrosion of Aluminium Corrosion of Aluminium (Second Edition) 1 Elsevier, Amsterdam, the Netherlands 41–61 [7] Loghin C, Ciobanu L, Ionesi D, Loghin E and Cristian I 2018 Introduction to waterproof and water repellent textiles Waterproof And Water Repellent Textiles And Clothing (Woodhead Publishing) pp 3–24 [8] Yilbas B S, Hassan G, Al-Qahtani H, Al-Aqeeli N, Al-Sharaf A, Al-Merbati A S, Baroud T N and Adukwu J A E 2019 Stretchable hydrophobic surfaces and self-cleaning applications Sci Rep. 9 14697 [9] Shi J, Xu L and Qiu D 2022 Effective antifogging coating from hydrophilic/hydrophobic polymer heteronetwork Adv. Sci. 9 2200072 [10] Williams J, Griffiths C, Dunlop T and Jewell E 2022 Improving the processability of a one-step hydrophobic coating for hot-dipped galvanised Steel for industrial applications Coatings 12 895 [11] Alamri H R, Rezk H, Elbary H A, Ziedan H A and Elnozahy A 2020 Experimental investigation to improve the energy efficiency of solar PV panels using hydrophobic SiO2 nanomaterial Coatings 10 503 [12] Falde E J, Yohe S T, Colson Y L and Grinstaff M W 2016 Superhydrophobic materials for biomedical applications Biomaterials 104 87 [13] Khan S A, Al-Hazmi F, Al-Heniti S, Faidah A and Al-Ghamdi A 2010 Effect of cadmium addition on the optical constants of thermally evaporated amorphous Se–S–Cd thin films Curr. Appl Phys. 10 145 [14] Kenanakis G, Stratakis E, Vlachou K, Vernardou D, Koudoumas E and Katsarakis N 2008 light-induction reversible hydrophylicity of ZnO structures grown by aqueous chemical growth Appl. Surf. Sci. 254 5695 [15] Lv J, Zhu J, Huang K, Meng F, Song X and Sun Z 2011 Tunable surface wettability of ZnO nanorods prepared by two-step method Appl. Surf. Sci. 257 7534 [16] Law K 2014 Definitions for hydrophilicity, hydrophobicity, and superhydrophobicity: getting the basics right J. Phys. Chem. Lett. 5 686 [17] Huang H W, Chang W C, Lin S J and Chueh Y L 2012 Growth of controllable ZnO film by atomic layer deposition technique via inductively coupled plasma treatment J. Appl. Phys. 112 124102 [18] Pilz J, Perrotta A, Christian P, Tazreiter M, Resel R, Leising G, Griesser T and Coclite A M 2018 Tuning of material properties of ZnO thin films grown by plasma-enhanced atomic layer deposition at room temperature J. Vac. Sci. Technol. A: Vacuum, Surfaces, and Films 3 01A109 [19] Zhou J, Xu N and Wang Z L 2006 Dissolving behavior and stability of ZnO wires in biofluids: a study on biodegradability and biocompatibility of ZnO nanostructures Adv. Mater. 18 2432 [20] Geurts J 2010 Crystal Structure, Chemical Binding, and Lattice Properties In: Zinc Oxide. Springer Series in Materials Science 120 7–37 [21] Montes Ruiz-Cabello F J, Ibañez-Ibañez P, Paz-Gomez G, Cabrerizo-Vilchez M and Angel Rodriguez-Valverde M 2018 Fabrication of superhydrophobic metal surfaces for anti-icing applications J. Vis. Exp. 138 e57635 [22] Patra S, Sarkar S, Bera S K, Ghosh R and Paul G K 2009 Hydrophobic self-cleaning surfaces of ZnO thin films synthesized by sol–gel technique J. Phys. D: Appl. Phys. 42 075301 [23] Li Y, Zheng M, Ma L, Zhong M and Shen W 2008 Fabrication of hierarchical ZnO architectures and their superhydrophobic surfaces with strong adhesive force Inorg. Chem. 47 3140 [24] Mufti N, Arista D, Diantoro M, Fuad A, Taufiq A and Sunaryono S 2017 Series: Materials Science and Engineering 202 012006 IOP Conf [25] Kenanakis G, Vernardou D and Katsarakis N 2012 Light-induced self-cleaning properties of ZnO nanowires grown at low temperatures Appl. Catal. Gen. 411 7–14 [26] Ghannam H, Chahboun A and Turmine M 2019 Wettability of zinc oxide nanorod surfaces RSC Adv. 9 38289 [27] Jiang F X, Tong R X, Yan Z and Xu L F J X H d-electron-dependent transparent conducting oxide of V-doped ZnO thin films J. Alloys Compd. 822 153706 [28] Kennedy O W 2019 MBE-grown ZnO-based Nanostructures for Electronics Applications, thesis University College London. [29] Kramer A, Engel S, Sangiorgi N, Sanson A, Bartolome J F, Graf S and Muller F A 2017 ZnO thin films on single carbon fibres fabricated by pulsed laser deposition (PLD) Appl. Surf. Sci. 399 282 [30] Wang Z et al 2019 Vacancy cluster in ZnO films grown by pulsed laser deposition Sci Rep. 9 3534 [31] Guerrero de León J A, Centeno A P, Rosas G G, Mariscal A, Serna R, Aranda M A S and Galván J G Q 219 Influence of the Zn plasma kinetics on the structural and optical properties of ZnO thin films grown by PLD SN Appl. Sci. 1 475 [32] Li J, Wu Z, Xu Y, Pei Y and Wang G 2019 Stability analysis of multi process parameters for metal-organic chemical vapor deposition reaction cavity Molecules 24 876 [33] Amlouk A, Boubaker K and Amlouk M 2010 A new procedure to prepare semiconducting ternary compounds from binary buffer materials and vacuum-deposited copper for photovoltaic applications Vacuum 85 60 [34] Karyaoui M, Jaballah A B, Mechiak R and Chtourou R 2012 The porous nature of ZnO thin films deposited by sol–gel Spin-Coating technique IOP Conf. SeriesMaterials Science and Engineering, IOP Publishing 012019 [35] Jacob A A, Balakrishnan L, Meher S, Shambavi K and Alex Z 2017 Structural, optical and photodetection characteristics of Cd alloyed ZnO thin film by spin coating J. Alloy. Compd. 695 3753 [36] Ganesh V, Salem G, Yahia I and Yakuphanoglu F 2018 Synthesis, optical and photoluminescence properties of Cu-doped ZnO nano- fibers thin films: nonlinear optics J. Electron. Mater. 47 1798 [37] Nithyaa Sree D, Paul Mary Deborrah S, Gopinathan C and Inbanathan S S R 2019 Enhanced UV light induced photocatalytic degradation of methyl orange by Fe doped spray pyrolysis deposited ZnO thin films Appl. Surf. Sci. 494 116 [38] Belamri Z, Darenfad W and Guermat N 2023 Impact of annealing temperature on surface reactivity of ZnO nanostructured thin films deposited on aluminum substrate journal of Nano- and Electronic Physics 15 02026 [39] Huang Y, Sarkar D K and Chen X-G 2011 Fabrication of superhydrophobic surfaces on aluminum alloy via electrodeposition of copper followed by electrochemical modification Nano-Micro Lett. 3 160–5 [40] Sarkar D K, Farzaneh M and Paynter R W 2008 Superhydrophobic properties of ultrathin rf-sputtered Teflon films coated etched aluminum surfaces Mater. Lett. 62 1226 [41] Perillo P M, Atia M N and Rodríguez D F 2018 Studies on the growth control of ZnO nanostructures synthesized by the chemical method Rev. Mater. 23 e-12133 [42] Malek M F, Mamat M H, Sahdan M Z, Zahidi M M, Khusaimi Z and Mahmood M R 2013 Influence of various sol concentrations on stress/strain and properties of ZnO thin films synthesised by sol–gel technique Thin Solid Films 527 102 [43] Singh G, Shrivastava S B, Jain D, Pandya S and Ganesan V 2009 Effect of molarity of precursor solution on nanocrystalline Zinc Oxide thin films Defect and Diffusion Forum 293 99 [44] Khedmi N, Ben Rabeh M, Abdelkadher D, Ousgi F and Kanzari M 2014 Effect of thickness on structural and optical properties of vacuum-deposited Sn2Sb2S5 thin films Cryst. Res. Technol. 1–8 12 Phys. Scr. 98 (2023) 125949 Z Belamri et al [45] Muhammed S P and Chandra A B 2015 Impact of crystalline defects and size on x-ray line broadening: a phenomenological approach for tetragonal SnO2 nanocrystals AIP Adv. 5 057137 [46] Srinivasulu T, Saritha K and Ramakrishna Reddy K T 2017 Synthesis and characterization of Fe-doped ZnO thin films deposited by chemical spray pyrolysis Mod. Electron. Mater. 3 76 [47] Zegadi C, Abderrahmane A, Djelloul A, Hamzaoui S, Adnane M, Chaumont D and Abdelkebir K 2015 Effects on structural and electro- optical properties of iron incorporation to p-Zinc Oxide (ZnO) thin films deposited by dip-coating process Int. Rev. Phys. 9 39 [48] Shakti N and Gupta P S 2010 Structural and optical properties of sol-gel prepared ZnO thin film Appl. Phys. Res. 2 19 [49] Chen M, Pei Z L, Sun C, Wen L S and Wang X 2000 Surface characterization of transparent conductive oxide Al-doped ZnO films J. Cryst. Growth 220 254 [50] Tüzemen E S, Eker S, Kavak H and Esen R 2009 Dependence of film thickness on the structural and optical properties of ZnO thin films Appl. Surf. Sci. 255 6195 [51] Miranda G G, Lucas de Sousa e Silva R, Veridiana dos Santos Pessoni H and Franco Jr A 2021 Raman spectroscopy study of Ga-doped ZnO ceramics: an estimative of the structural disorder degree Physica B 606 412726 [52] Gupta H, Joshi K, Gautam S K, Singh R G and Singh F 2020 Raman scattering from irradiated nanocrystalline zinc oxide thin films: perspective view on effects of energy loss, ion fluence, and ion flux Vacuum 181 109598 [53] Taziwa R, Meyer E, Katwire D and Ntozakhe L 2017 Influence of carbon modification on the morphological, structural, and optical properties of zinc oxide nanoparticles synthesized by pneumatic spray pyrolysis technique J. Nanomater. Article ID 9095301 1 [54] Ozgur U, Alivov Y I, Liu C, Teke A, Reshchikov M A, Dogan S, Avrutin V, Cho S J and Morkoc H 2005 A comprehensive review of ZnO materials and devices J. Appl. Phys. 98 1 [55] Cusco R, Alarcon-Llado E, Ibanez J, Artus L, Jimenez J, Wang B G and Callahan M J 2007 Temperature dependence of raman scattering in ZnO Phys. Rev. B 75 1 [56] Huang Y, Sarkar D K and Chen X-G 2015 Superhydrophobic nanostructured ZnO thin films on aluminum alloy substrates by electrophoretic deposition Appl. Surf. Sci. 327 327 [57] Cassie A B D and Baxter S 1944 Wettability of porous surfaces Trans. Faraday Soc. 40 546 13
null
10.1038/s41467-023-35891-9
Data availability The data that support the findings of this study are available from the corresponding authors upon request.
Data availability The data that support the findings of this study are available from the corresponding authors upon request.
Article https://doi.org/10.1038/s41467-023-35891-9 Universality of light thermalization in multimoded nonlinear optical systems Received: 28 July 2022 Accepted: 5 January 2023 Check for updates ; , : ) ( 0 9 8 7 6 5 4 3 2 1 ; , : ) ( 0 9 8 7 6 5 4 3 2 1 Qi Zhong1,5, Fan O. Wu 1,5, Absar U. Hassan1, Ramy El-Ganainy 2,3 & Demetrios N. Christodoulides 1,4 Recent experimental studies in heavily multimoded nonlinear optical systems have demonstrated that the optical power evolves towards a Rayleigh–Jeans (RJ) equilibrium state. To interpret these results, the notion of wave turbulence founded on four-wave mixing models has been invoked. Quite recently, a different paradigm for dealing with this class of problems has emerged based on thermodynamic principles. In this formalism, the RJ distribution arises solely because of ergodicity. This suggests that the RJ distribution has a more general origin than was earlier thought. Here, we verify this universality hypothesis by investigating various nonlinear light-matter coupling effects in physically accessible multimode platforms. In all cases, we find that the system evolves towards a RJ equilibrium—even when the wave-mixing paradigm completely fails. These observations, not only support a thermodynamic/ probabilistic interpretation of these results, but also provide the foundations to expand this thermodynamic formalism along other major disciplines in physics. Nonlinear optics plays a crucial role in a wide range of modern sci- ence and technologies. These include optical cavity microcombs1,2, high-power light sources3, cavity optomechanics4,5, nonlinear topo- logical and non-Hermitian photonics6–10, bioimaging11,12, as well as classic/quantum networks and signal processing13–16. While nonlinear interactions widely vary in strength and differ from one material system to another, their vast majority can still be described using an relies on perturbative framework that underlying theoretical analysis17. Particularly, by expressing the electric polarization vector as a Taylor series expansion in terms of the driving electric field, one can classify nonlinear optical effects into several, largely indepen- dent processes such as those associated with second harmonic and sum/difference frequency generation and multi-wave mixing interactions17. A few decades ago, this same paradigm was adopted by Zakharov and colleagues to study optical nonlinear propagation effects when an infinite number of Fourier components is involved—a field of research that is nowadays known as wave turbulence18. In this seminal work, it was shown that such a system can be described by a Boltzmann-like kinetic model that admits a steady-state solution in the form of a Rayleigh–Jeans (RJ) distribution. In this regard, it was conjectured that the RJ law results as a mere byproduct of the non- linear attractor dynamics taking place during multi-wave mixing19. In developing this model, several assumptions were made. Firstly, it was implicitly assumed that four-wave mixing dominates the interaction process. Secondly, the so-called random phase approximation20 was employed to omit off-resonant interaction terms. Meanwhile, recent progress in the general area of multimode fiber optics21–29 has enabled a new generation of nonlinear experimental setups where the RJ distribution (power allocation among modes) was successfully observed for the first time30–33. The clear demonstration of RJ ther- malization in such settings has been touted as evidence in support of the wave turbulence theory. While reaching such a conclusion does not seem to pose a problem from a practical point of view, it is unsettling at a more fundamental level. In essence, adopting the wave 1CREOL, College of Optics and Photonics, University of Central Florida, Orlando, FL 32816, USA. 2Department of Physics, Michigan Technological University, Houghton, MI 49931, USA. 3Henes Center for Quantum Phenomena, Michigan Technological University, Houghton, MI 49931, USA. 4Ming Hsieh Department of Electrical and Computer Engineering, University of Southern California, Los Angeles, CA 90089, USA. 5These authors contributed equally: Qi Zhong, Fan O. Wu. e-mail: ganainy@mtu.edu; demetri@creol.ucf.edu Nature Communications | (2023) 14:370 1 Article https://doi.org/10.1038/s41467-023-35891-9 turbulence hypothesis is to a great extent analogous to attempting to infer, for example, the nature of the interactions between gas molecules solely from the Maxwell–Boltzmann distribution. Even more importantly, while the laws of simple thermodynamic systems like gases can be developed from either classical (Newtonian) kinetic theories or quantum mechanical perspectives, this is by no means necessary, given that the corresponding equations of state can be derived from purely entropic principles—in total disregard to the underlying collisional mechanisms. So, the question naturally arises: is the RJ distribution an actual byproduct of multi-wave mixing pro- cesses or does it represent a much more general result that has little to do with the specifics of the inherent nonlinearity involved? Quite recently, a different approach for studying light thermali- zation was put forward on the basis of statistical mechanics and thermodynamics34–38. While this latter theoretical framework reaches similar conclusions to those derived from the aforementioned kinetic theories18,19 as far as the RJ distribution is concerned, its perspective of optical thermalization is fundamentally different. Being founded on notions from statistical mechanics, this paradigm34,35 allows one to predict and interpret the RJ law emerging in a microcanonical system from purely entropic considerations. In this regard, the RJ equilibrium state macroscopically manifests itself because it is ergodically asso- ciated with a largest number of microstates (in phase space) and thus it can be considered a byproduct of probability theory—an aspect that has little to do with the nature of the underlying nonlinearity involved. If this is indeed the case, then in analogy with statistical mechanics of gases, the RJ thermalization should occur in systems with more generic nonlinearities beyond the wave mixing paradigm as illustrated in Fig. 1. The situation is however more complex. Nonlinear optical systems often exhibit two constants of motion, i.e., the power and the Hamil- tonian. The first, which describes the conservation of optical power, is analogous to the number of particles in a gas system. The second, however, when expressed in the linear eigenbasis, involves both a linear and a nonlinear component. Thus, strictly speaking, such a system is not necessarily expected to relax to a RJ distribution. Only under the condition that the linear part is constant, the RJ distribution can be anticipated. In reality, however, even under weak nonlinear conditions, the linear part of the Hamiltonian is only quasi-conserved. In other words, the analogy between multimoded nonlinear optical arrangements and idealized thermodynamic systems involving two constants of motion is not formal, which further complicates the question about thermalization in nonlinear optical systems and the physical mechanism responsible for observing the RJ distribution. Non-equilibrium state Equilibrium state FWM SHG OM MWM w/o WM y c n a p u c c O Maximize the entropy y c n a p u c c O RJ Energy level Energy level Fig. 1 | Conceptual illustration of thermalization in a nonlinear multimode optical system. Similar to thermalization in matter, the nature of the interaction forces (like forces between gas molecules) is irrelevant. Here, we show that light thermalization into a Rayleigh–Jeans (RJ) distribution can take place under a wide range of nonlinear conditions beyond the traditional four-wave mixing (FWM) paradigm. These include second harmonic generation (SHG), multi-wave mixing (MWM), optomechanical (OM) cascaded interactions between optical and mechanical modes, and even scenarios where the system cannot be described by any wave mixing expansion (w/o WM). In this work, we critically examine the manner in which optical thermalization processes unfold in nonlinear environments with dif- ferent types of nonlinearities such as those arising from optomecha- interactions (where wave mixing interpretations are rather nical cumbersome) and those associated with photorefractive crystals (where above certain power thresholds, standard perturbative wave mixing expansions are not possible). In addition, we consider also artificial nonlinear systems with nonanalytic and discontinuous non- linear functions that cannot be described by any convergent poly- nomial and demonstrate that such set-ups can also reach the RJ equilibrium distribution. Our work thus establishes the universality of the thermalization towards the RJ state in nonlinear optical systems, and, in doing so, presents compelling evidences in favor of the more general entropic view of optical thermalization as opposed to the more restrictive four-wave mixing paradigm. Results Before we proceed, perhaps it would be useful to highlight some of the basic notions upon which the optical thermodynamic approach relies on. As in the case of standard statistical mechanics39, the entropy of the optical multimode arrangement can be built within a microcanonical ensemble formalism by accounting all possible microstates, each containing information as to the energy/power and phase distribution among all modes in the system. In defining the macrostates, the energy/power distribution is retained while the phase information is omitted40 (being superfluous given that it is uniformly distributed within the range 0 to 2π). In this respect, the nonlinear interaction acts merely as an agent that enables a chaotic reshuffling of optical energy among modes and therefore facilitates thermalization. On the other hand, the specifics of nonlinearity are inconsequential. Optical ther- modynamic equilibrium is then reached when entropy is maximized over all possible microstates under the constraints dictated by the two constants of motion35. Kerr nonlinearity We begin our analysis by first considering a Kerr nonlinear multimode tight-binding model—a one-dimensional photonic array comprised of M evanescently coupled single-mode waveguides with nearest neigh- bor coupling41,42 (a situation most relevant to experimental imple- mentations), as shown in Fig. 2a. Under these conditions, light propagation along z in such a lattice can be described by the following normalized discrete nonlinear Schrödinger equation43: i dam dz + am(cid:1)1 + am + 1 + ∣am∣2am = 0, ð1Þ (cid:2) m = 1 PM ∣am∣2 = PM j = 1 where am is the field amplitude at site m, and the last term denotes Kerr nonlinear effects. Equation (1) exhibits two constants of motion. The first invariant (denoting power conservation) is given by ∣cj∣2, where cj is the field amplitude component P = associated with supermode ∣ψji of the linear array (i.e., the normal modes obtained by diagonalizing Eq. (1) in the absence of the nonlinear term). The complex amplitudes cj at any distance z are obtained by projecting the state ∣ψ of the system on the linear supermodes as expressed in the local representation (i.e., in terms of am). The second invariant is associated with the optical Hamiltonian comprised of a linear HL and a nonlinear HNL component, i.e., H = HL + HNL where ∣am∣4, where aM+1 = 0 H because of the truncated boundary condition. Under weak nonlinear conditions, the contribution from the linear term HL dominates, and as a result one can define a quasi-invariant internal energy by εj∣cj∣2, where εj = 2 cosð jπ Þ are the eigenvalues U (cid:3) (cid:1)H M + 1 associated with the linear supermodes ∣ψji. As indicated above, by using purely entropic principles, one can show that light propagating in such a system evolves towards a thermal state obeying the RJ m + 1 + a* PM j = 1 Þ and H mam + 1 L = (cid:1) ðama* PM PM NL = L = m = 1 m = 1 1 2 Nature Communications | (2023) 14:370 2 Article https://doi.org/10.1038/s41467-023-35891-9 Fig. 2 | Thermalization of light in nonlinear waveguide arrays with different nonlinearities. Linear and nonlinear couplings in three optical lattices when acted upon by three different nonlinearities: a a Kerr nonlinearity, c cascade χ(2) process, and e optomechanical nonlinearities, as described by Eqs. (1), (3), and (4), respectively. Numerical simulations provide the modal occupancies after therma- lization in all these three scenarios, in good agreement with the predicted Rayleigh–Jeans (RJ) distributions (black lines), as shown in b, d, and f. The insets display a monotonic increase in entropy S as well as the invariants of the motion U and P. Note that in all cases, numerical simulations are performed over ensemble averages. The thermal fluctuations of quasi-invariants (when applicable) are indi- cated by gray lines, depicting the instantaneous values of U and P around their mean values. In all cases, the nonlinear array has M = 100 sites and the dashed lines represent the initial occupancies for the linear optical supermodes. distribution34,35: ∣cj∣2 = (cid:1) T μ + εj , ð2Þ where T and μ represents the optical temperature and chemical potential, respectively. In general, the equilibrium values of T, μ can be predicted from the initial conditions, i.e., from the invariants P and U34,36,38. For instance, for a lattice with M = 100 elements, an input excitation ∣cj∣2 = 0.05(εj + 2) (dashed line in Fig. 2b) leads to P = 10 and U = − 9.9, which in turn predicts T = 0.15 and μ = − 2.5 (see Supplemen- tary Note 1). The size of the systems considered in this study is large enough so as to guarantee the extensivity of the entropy and the self- consistency of the thermodynamic formulation used44. By numerically integrating Eq. (1), we find that the equilibrium modal occupancies ∣cj∣2 are consistent with the theoretically predicted RJ distribution (Fig. 2b). The inset panel in Fig. 2b shows that during propagation, the optical j = 1 lnð∣cj∣2Þ monotonically increases until it reaches a entropy S = maximum (as expected by the second law of thermodynamics) while the optical energy U remains quasi-invariant. PM Cascade second order χ(2) nonlinearity In order to demonstrate the universality of RJ thermalization, we now investigate a variety of scenarios. In this respect, we consider cascade second order χ(2) nonlinear processes unfolding in waveguide arrays, governed by the following normalized coupled evolution equations45–47: ða2 m = 1 mb* PM PM mbmÞ(cid:4) m = 1 (cid:1) 1 2 mam + 1 PM j = 1 m + 1 + a* where am and bm are the local site field amplitudes associated with the fundamental and the second-harmonic frequency, and Δ is the phase mismatch. Here the linear coupling among bm is neglected46, as illustrated in Fig. 2c. This system exhibits two constants of motion: the ð∣am∣2 + ∣bm∣2Þ and the Hamiltonian total optical power P = m + a*2 Δ∣bm∣2 + 1 ½ama* H = (see Sup- 2 plementary Note 2). Under weak nonlinear conditions, the field in the fundamental frequency am dominates, and therefore its power and ∣cj∣2, and energy can be regarded as quasi-invariants, i.e., Pa = Ua = (cid:1) εj∣cj∣2, where cj is the field amplitude of the corresponding supermode . If indeed this system can thermalize through the χ(2) process under these two invariants, one should then anticipate a RJ distribution once equilibrium is reached. To confirm this hypothesis, we numerically simulated Eq. (3) with Δ = 1, M = 100 when the first 30 modes in the fundamental frequency were evenly excited (dashed line in Fig. 2d). As shown in Fig. 2d, after a non-equilibrium prethermaliza- tion stage, the quantities Pa and Ua eventually settle to Pa = 8:3 and Ua = −15.1, i.e., they remain invariants. For this set of values, once thermal equilibrium is attained, our theory predicts T = 0.016 and μ = −2.007, in excellent agreement with our numerical simula- tions (Fig. 2d). PM j = 1 Optomechanical nonlinearity Next, we consider a lossless nonlinear optomechanical cavity array where the intracavity optical fields and the vibrational motions are described by the following evolution equations48: dam dz + am(cid:1)1 + am + 1 + a* i dbm i dz (cid:1) Δbm + a2 m = 0, mbm = 0, Nature Communications | (2023) 14:370 ð3Þ i i dam dt dbm dt (cid:1) ðam(cid:1)1 + am + 1 Þ + amðbm + b* mÞ = 0, (cid:1) Ωbm + ∣am∣2 = 0: ð4Þ 3 Article https://doi.org/10.1038/s41467-023-35891-9 m = 1 m = 1 PM ∣am∣2 = PM Þ + ∣am∣2ðbm + b* PM j = 1 ½(cid:1)ðama* Here am and bm stands for the optical field and the mechanical oscil- lation amplitude in cavity m, respectively (Fig. 2e), while the parameter Ω represents a normalized angular frequency of the mechanical resonance. Synchronization between driven optomechanical oscilla- tors have been investigated in earlier studies and it was shown that the synchronization dynamics follow the generic features of the Kur- amoto model49. Here, instead, we are interested in the nonlinear dynamics of coupled optomechanical oscillators in the absence of the driving force. We proceed by first noting that the above system exhibits two invariants: the number of “photons” in the cavities ∣cj∣2, and the overall Hamiltonian of the sys- Pa = tem H = mam + 1 m + 1 + a* (see Supplementary Note 3), where cj denotes the field amplitude of the jth optical supermode. As before, under weakly nonlinear conditions and when the normalized Ω is large, such as Ω = 8 in our numerical simulations, one finds that the Hamiltonian associated with the optical field is a quasi-invariant, εj∣cj∣2. Even in this more Ua = complex scenario, the RJ distribution emerges at thermal equilibrium as a result of ergodicity as can be seen in Fig. 2f. In all cases, a good agreement was found to exist between numerical simulations and the theoretically anticipated RJ distribution once Pa, Ua were specified by initial conditions. Note that in this case, it is impossible to associate a multi-wave mixing process to the optical nonlinearity—an aspect that dispels the wave turbulence paradigm. Interestingly, unlike their photon counterparts, the mechanical vibrations themselves do not display a pair of (quasi-)invariants P and U (see Supplementary Note 3), and therefore cannot thermalize to a RJ equilibrium state in the same manner. the linear part of mÞ (cid:1) Ω∣bm∣2(cid:4) m + 1 + a* PM j = 1 ½(cid:1)ðama* mam + 1 PM Þ(cid:4) = m = 1 Nonlinearity described by a smooth but nowhere analytic function So far, we have analyzed thermalization effects in multimode systems where the nonlinearities conform to standard Taylor series expan- sions. Naturally, one may ask whether the RJ thermalization process can indeed manifest itself in more general nonlinear settings. To address this question, we now consider optical lattices involving generalized intensity-dependent nonlinearities F(x) as described by50: i dam dz + am(cid:1)1 + am + 1 + Fð∣am∣2Þam = 0: ð5Þ PM j = 1 m = 1 PM ½ama* m + 1 + a* ∣cj∣2 as well as the Hamiltonian Here the optical power P = H = mam + 1 + Gð∣am∣2Þ(cid:4) of the system are still con- served, where G(x) is the antiderivative of F(x) (i.e., dG(x)/dx = F(x), and G(0) = 0). As before, in the weak nonlinear regime, i.e., F(x) ≪ 1, the linear part of the Hamiltonian U = (cid:1) εj∣cj∣2 is a quasi-invariant. PM j = 1 1 PN ðxÞ = First, we consider the case where F(x) is chosen to be a smooth (infinitely differentiable) function everywhere, yet nowhere analytic (i.e., it does not have a convergent Taylor series representation). This function, which we will henceforth denote as F1(x). For example, here we construct such a nonanalytic function via Fourier series n = (cid:1)N hn expði2πnxÞ, where the Fourier coefficients hn are F random variables chosen such that their amplitudes drop with n faster than the reciprocal of any polynomial but slower than exponential51–53 (see Supplementary Note 4). This condition guarantees that in the limit N → ∞, the function F1(x) is infinitely differentiable but nowhere ana- lytic. In other words, this function has a Taylor series but its radius of convergence tends to 0 as N → ∞. From a practical point of view, one can choose N to be large enough so as the function F1(x) does not have a proper Taylor series within the range of interest of the intensities involved in our simulations. Figure 3a shows one such possible func- tion F1(x) used in our computations. In this case, numerical simulations carried out on Eq. (5) clearly indicate that the RJ distribution still emerges upon thermalization, as shown in Fig. 3b. While these results clearly support the universality hypothesis for RJ thermalization, they still do not provide compelling evidence, mainly because the function F1(x) is continuous. In this case, the Stone–Weierstrass theorem54 guarantees that it can be still represented by a polynomial expansion, even though it does not correspond to its Taylor series. Thus, in this Fig. 3 | Thermalization of light in nonlinear lattices involving generalized intensity-dependent nonlinearities F(x). a An example of non-analytic function used in our simulations. b Corresponding Rayleigh–Jeans (RJ) distribution (T = 0.15, μ = −2.5) occurring after thermalization. c A discontinuous multi-step function used in our simulations. d Again this nonlinearity leads to a RJ distribution. e A saturable nonlinearity described by F 1 + x, and (f) its corresponding RJ distribution. In ðxÞ = x 3 (b) and (d), the initial excitation conditions are exactly the same and as a result they attain the same RJ allocation, an aspect indicating universality in thermaliza- tion.The insets have been plotted in a manner similar to Fig. 2. As before, here we used M = 100 and the initial mode occupancies are represented by the dashed lines. Nature Communications | (2023) 14:370 4 Article https://doi.org/10.1038/s41467-023-35891-9 scenario one could still argue that the underlying nonlinear interac- tions do arise from a series of higher-order wave mixing terms. A discontinuous nonlinearity function In order to assert the universality of RJ thermalization, i.e., being of a purely entropic (ergodic) origin that goes beyond the wave mixing picture, we next consider a nonlinearity that is described by a dis- continuous multi-step function55–57 such as that depicted in Fig. 3c, denoted as F2(x). Due its discontinuous nature, the function F2(x) cannot be analytically represented by a polynomial expansion across its entire domain. In other words, the wave mixing paradigm com- pletely fails in this case. Interestingly, even in this case, the system thermalizes and reaches a RJ equilibrium state as shown in Fig. 3d, in full accord with theoretically anticipated results. This latter example demonstrates once and for all that optical thermalization in multi- mode systems has a more fundamental origin—rooted in the system’s ergodicity rather than in the intricate nature of the nonlinear interac- tions involved. In other words, the onset of a RJ distribution does not necessarily require the presence of any multi-wave mixing mechan- isms. Instead, it is simply the outcome of the maximizing the entropy itself. Note that the simulations depicted in Fig. 3b, d were carried out for the same parameters and initial conditions (M = 100, P = 10, U = −9.9). Interestingly, despite the profound differences in their nonlinearity, they all settle exactly at the same RJ distribution with T = 0.15 and μ = −2.5. This further supports our hypothesis. In other words, as indicated before, one cannot infer the nature of the inter- molecular collision processes from the Maxwell–Boltzmann distribu- tion as manifested in actual gases. 3 ðxÞ = x Saturable nonlinearity We finally extend this discussion to more realistic material systems. For instance, consider photorefractive crystals where the nonlinearity is saturable57,58 F 1 + x, as shown in Fig. 3e. In the domain where x > 1, F3(x) does not have a Taylor representation but instead has a Laurent series expansion59: F3(x) = 1 − x−1 + x−2 − x−3 + . . . . Obviously, in this regime, the nonlinear interaction cannot be described by a simple wave mixing approach. Yet, assuming that ergodicity holds, and given that two invariants P and U still exist, as per our previous arguments, this should lead to RJ thermalization. This is verified using numerical simulations as shown in Fig. 3f. To ensure the validity of our conclu- sions, the values of the local intensities ∣am∣2 have been monitored during our simulations so as the F3(x) function was predominantly within the Laurent series expansion (see Supplementary Note 5). Discussion In conclusion, we have critically examined the manner in which optical thermalization processes unfold in nonlinear multimode environments and showed that the RJ distribution law is universal: it can manifest itself even in systems where the multi-wave mixing picture fails. These results extend the notion of wave thermalization beyond the original wave turbulence hypothesis that is founded on the premise of wave mixing interactions. In other words, through the use of counterexamples we demonstrated that nonlinear wave mix- ing may be sufficient but by no means necessary. Importantly, it would seem that, in some cases, these processes may not be in fact responsible for thermalization. Instead, our results suggest that RJ equilibrium is obtained because of ergodicity and entropy max- imization as expected by the second law of thermodynamics. These observations, not only support a thermodynamic/probabilistic interpretation of these results, but also provide appropriate foun- dations to expand the thermodynamic formalism in other physical settings governed by classical bosonic interactions. Finally, of inter- est would be to investigate the prospects of devising a formal proof that would dictate the universality of thermalization processes under general nonlinear conditions. Methods Numerical simulation All the simulation results in this work are obtained by numerically integrating the nonlinear equations of motion described by Eqs. (1), and (3)–(5). Due to the finite size of the system, the modal occupancies ∣cj∣2 fluctuate around their equilibrium values. Thus, the final equili- brium state can be evaluated either by calculating the time (distance) average or ensembles average. In this work, we adopted the latter strategy. In particular, for each simulation in Figs. 2 and 3, we have employed 400 ensembles, each of which corresponds to a random initial condition that have the same intensity profile (i.e., same values for ∣cj∣2) but a different phase distribution. Data availability The data that support the findings of this study are available from the corresponding authors upon request. References 1. 2. Bao, H. et al. Laser cavity-soliton microcombs. Nat. Photonics 13, 384–389 (2019). Kippenberg, T. J., Holzwarth, R. & Diddams, S. A. Microresonator-based optical frequency combs. Science 332, 555–559 (2011). 3. Wright, L. G., Christodoulides, D. N. & Wise, F. W. Spatio- temporal mode-locking in multimode fiber lasers. Science 358, 94–97 (2017). Kippenberg, T. J. & Vahala, K. J. Cavity optomechanics: back-action at the mesoscale. Science 321, 1172–1176 (2008). 4. 5. Aspelmeyer, M., Kippenberg, T. J. & Marquardt, F. Cavity opto- mechanics. Rev. Mod. Phys. 86, 1391–1452 (2014). 6. Xia, S. et al. Nontrivial coupling of light into a defect: the interplay of 7. nonlinearity and topology. Light Sci. Appl. 9, 147 (2020). Kirsch, M. S. et al. Nonlinear second-order photonic topological insulators. Nat. Phys. 17, 995–1000 (2021). 8. Wimmer, M. et al. Observation of optical solitons in PT-symmetric lattices. Nat. Commun. 6, 7782 (2015). 9. Xia, S. et al. Nonlinear tuning of PT symmetry and non-Hermitian topological states. Science 372, 72–76 (2021). 10. Bongiovanni, D. et al. Dynamically emerging topological phase transitions in nonlinear interacting soliton lattices. Phys. Rev. Lett. 127, 184101 (2021). 11. Bezryadina, A. et al. Nonlinear self-action of light through biological suspensions. Phys. Rev. Lett. 119, 058101 (2017). 12. Zhang, S. et al. Recent advances in nonlinear optics for bio-imaging applications. Opto-Electron. Adv. 3, 200003 (2020). 13. Tan, M. et al. RF and microwave photonic temporal signal proces- sing with Kerr micro-combs. Adv. Phys. X 6, 1838946 (2021). 14. Reimer, C. et al. High-dimensional one-way quantum processing implemented on d-level cluster states. Nat. Phys. 15, 148–153 (2019). 15. Ramos, A., Fernández-Alcázar, L., Kottos, T. & Shapiro, B. Optical phase transitions in photonic networks: a spin-system formulation. Phys. Rev. X 10, 031024 (2020). 16. Shi, C., Kottos, T. & Shapiro, B. Controlling optical beam thermali- zation via band-gap engineering. Phys. Rev. Res. 3, 033219 (2021). 17. Boyd, R. W. Nonlinear Optics (Academic Press, 2008). 18. Zakharov, V. E., L’vov, V. S. & Falkovich, G. Kolmogorov Spectra of Turbulence I: Wave Turbulence (Springer, 2012). 19. Picozzi, A. et al. Optical wave turbulence: towards a unified none- quilibrium thermodynamic formulation of statistical nonlinear optics. Phys. Rep. 542, 1–132 (2014). 20. Abarbanel, H. D., Rabinovich, M. I. & Sushchik, M. M. Introduction to Nonlinear Dynamics for Physicists (World Scientific, 1993). 21. Kibler, B. & Béjot, P. Discretized conical waves in multimode optical fibers. Phys. Rev. Lett. 126, 023902 (2021). Nature Communications | (2023) 14:370 5 Article https://doi.org/10.1038/s41467-023-35891-9 22. Mafi, A. Pulse propagation in a short nonlinear graded-index 43. Christodoulides, D. N. & Joseph, R. I. Discrete self-focusing in multimode optical fiber. J. Lightwave Technol. 30, 2803–2811 (2012). 23. Nazemosadat, E. & Mafi, A. Nonlinear multimodal interference and saturable absorption using a short graded-index multimode optical fiber. J. Opt. Soc. Am. B 30, 1357–1367 (2013). 24. Pourbeyram, H., Agrawal, G. P. & Mafi, A. Stimulated Raman scat- tering cascade spanning the wavelength range of 523 to 1750 nm using a graded-index multimode optical fiber. Appl. Phys. Lett. 102, 201107 (2013). 25. Teğin, U. et al. Controlling spatiotemporal nonlinearities in multi- mode fibers with deep neural networks. APL Photonics 5, 030804 (2020). 26. Rahmani, B., Loterie, D., Konstantinou, G., Psaltis, D. & Moser, C. Multimode optical fiber transmission with a deep learning network. Light Sci. Appl. 7, 69 (2018). 27. Mumtaz, S., Essiambre, R.-J. & Agrawal, G. P. Nonlinear propagation in multimode and multicore fibers: generalization of the Manakov equations. J. Lightwave Technol. 31, 398–406 (2013). 28. Chen, T. et al. All-fiber passively mode-locked laser using nonlinear multimode interference of step-index multimode fiber. Photonics Res. 6, 1033–1039 (2018). nonlinear arrays of coupled waveguides. Opt. Lett. 13, 794–796 (1988). 44. Selim, M. A., Wu, F. O., Ren, H., Khajavikhan, M. & Christodoulides, D. Thermodynamic description of the near- and far-field intensity patterns emerging from multimode nonlinear waveguide arrays. Phys. Rev. A 105, 013514 (2022). 45. Trillo, S. & Torruellas, W. Spatial Solitons (Springer, 2001). 46. Iwanow, R. et al. Observation of discrete quadratic solitons. Phys. Rev. Lett. 93, 113902 (2004). 47. Siviloglou, G. A. et al. Observation of discrete quadratic surface solitons. Opt. Express 14, 5508–5516 (2006). 48. Xiong, H., Gan, J. & Wu, Y. Kuznetsov-Ma soliton dynamics based on the mechanical effect of light. Phys. Rev. Lett. 119, 153901 (2017). 49. Heinrich, G., Ludwig, M., Qian, J., Kubala, B. & Marquardt, F. Col- lective dynamics in optomechanical arrays. Phys. Rev. Lett. 107, 043603 (2011). 50. Yang, J. Nonlinear Waves in Integrable and Nonintegrable Systems (SIAM, 2010). 51. Trefethen, N. Smoothies: nowhere analytic functions. https://www. chebfun.org/examples/stats/Smoothies.html (2020). 29. Poletti, F. & Horak, P. Dynamics of femtosecond super- 52. Trefethen, N. Notes of a numerical analyst: random smoothies. LMS continuum generation in multimode fibers. Opt. Express 17, 6134–6147 (2009). 30. Pourbeyram, H. et al. Direct observations of thermalization to a Rayleigh–Jeans distribution in multimode optical fibres. Nat. Phys. 18, 685–690 (2022). Newsletter 43 (2021). 53. Bilodeau, G. G. The origin and early development of non-analytic infinitely differentiable functions. Arch. Hist. Exact Sci. 27, 115–135 (1982). 54. Stone, M. H. The generalized weierstrass approximation theorem. 31. Mangini, F. et al. Statistical mechanics of beam self-cleaning in Math. Mag. 21, 237–254 (1948). GRIN multimode optical fibers. Opt. Express 30, 10850–10865 (2022). 32. Podivilov, E. V. et al. Thermalization of orbital angular momentum beams in multimode optical fibers. Phy. Rev. Lett. 128, 243901 (2022). 33. Baudin, K. et al. Classical Rayleigh-Jeans condensation of light waves: observation and thermodynamic characterization. Phys. Rev. Lett. 125, 244101 (2020). 34. Wu, F. O., Hassan, A. U. & Christodoulides, D. N. Thermodynamic theory of highly multimoded nonlinear optical systems. Nat. Pho- tonics 13, 776–782 (2019). 35. Makris, K. G., Wu, F. O., Jung, P. S. & Christodoulides, D. N. Statis- tical mechanics of weakly nonlinear optical multimode gases. Opt. Lett. 45, 1651–1654 (2020). 36. Parto, M., Wu, F. O., Jung, P. S., Makris, K. & Christodoulides, D. N. Thermodynamic conditions governing the optical temperature and chemical potential in nonlinear highly multimoded photonic sys- tems. Opt. Lett. 44, 3936–3939 (2019). 37. Wu, F. O., Jung, P. S., Parto, M., Khajavikhan, M. & Christodoulides, D. N. Entropic thermodynamics of nonlinear photonic chain net- works. Commun. Phys. 3, 216 (2020). 38. Efremidis, N. K. & Christodoulides, D. N. Fundamental entropic processes in the theory of optical thermodynamics. Phys. Rev. A 103, 043517 (2021). 39. Pathria, R. K. Statistical Mechanics (Elsevier, 2016). 40. Wu, F. O. et al. Thermalization of light’s orbital angular momentum in nonlinear multimode waveguide systems. Phys. Rev. Lett. 128, 123901 (2022). 41. Leblond, H., Kremer, D. & Mihalache, D. Few-cycle spatiotemporal optical solitons in waveguide arrays. Phys. Rev. A 95, 043839 (2017). 55. Musslimani, Z. H., Segev, M., Christodoulides, D. N. & Soljačić, M. Composite multihump vector solitons carrying topological charge. Phys. Rev. Lett. 84, 1164–1167 (2000). 56. Musslimani, Z. H., Segev, M. & Christodoulides, D. N. Multi- component two-dimensional solitons carrying topological charges. Opt. Lett. 25, 61–63 (2000). 57. Musslimani, Z. H., Soljačić, M., Segev, M. & Christodoulides, D. N. Interactions between two-dimensional composite vector solitons carrying topological charges. Phys. Rev. E 63, 066608 (2001). 58. Jia, P., Li, Z., Hu, Y., Chen, Z. & Xu, J. Visualizing a nonlinear response in a Schrödinger wave. Phys. Rev. Lett. 123, 234101 (2019). 59. Arfken, G., Weber, H. & Harris, F. Mathematical Methods for Physi- cists: A Comprehensive Guide (Academic Press, 2012). Acknowledgements This work was partially supported by ONR MURI (Grant No. N00014-20-1- 2789), AFOSR MURI (Grant Nos. FA9550-20-1-0322 and FA9550-21-1- 0202), National Science Foundation (Grant Nos. DMR-1420620 and EECS-1711230), MPS Simons Collaboration (Simons Grant No. 733682), W. M. Keck Foundation, U.S.-Israel Binational Science Foundation (Grant No. 2016381), and US Air Force Research Laboratory (Grant No. FA86511820019). Author contributions R.E.-G. and D.N.C. conceived the idea. Q.Z. and F.O.W. developed the theory and conducted the simulations with feedback from A.U.H. All the authors contributed in preparing the manuscript. 42. Pelinovsky, D. E., Sukhorukov, A. A. & Kivshar, Y. S. Bifurcations and stability of gap solitons in periodic potentials. Phys. Rev. E 70, 036618 (2004). Competing interests The authors declare no competing interests. Nature Communications | (2023) 14:370 6 Article https://doi.org/10.1038/s41467-023-35891-9 Additional information Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41467-023-35891-9. Correspondence and requests for materials should be addressed to Ramy El-Ganainy or Demetrios N. Christodoulides. Peer review information Nature Communications thanks Mario Ferraro, Mikko Huttunen, and the other, anonymous, reviewer(s) for their con- tribution to the peer review of this work. Reprints and permissions information is available at http://www.nature.com/reprints Publisher’s note Springer Nature remains neutral with regard to jur- isdictional claims in published maps and institutional affiliations. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/ licenses/by/4.0/. © The Author(s) 2023 Nature Communications | (2023) 14:370 7
null
10.1371_journal.pone.0262844.pdf
Data Availability Statement: All relevant data are within the paper.
All relevant data are within the paper.
RESEARCH ARTICLE Improvements following multimodal pelvic floor physical therapy in gynecological cancer survivors suffering from pain during sexual intercourse: Results from a one-year follow- up mixed-method study Marie-Pierre CyrID Paul Bessette2,5, Annick Pina6,7, Walter Henry GotliebID He´ lène Mayrand7,10, Me´ lanie Morin1,2* 1,2, Rosalie Dostie1,2, Chantal Camden1,2, Chantale DumoulinID 3,4, 8,9, Korine Lapointe-Milot2,5, Marie- 1 Faculty of Medicine and Health Sciences, School of Rehabilitation, University of Sherbrooke, Sherbrooke, Quebec, Canada, 2 Research Center of the Centre Hospitalier Universitaire de Sherbrooke, Sherbrooke, Quebec, Canada, 3 Faculty of Medicine, School of Rehabilitation, University of Montreal, Montreal, Quebec, Canada, 4 Research Center of the Institut Universitaire de Ge´ riatrie de Montre´al, Montreal, Quebec, Canada, 5 Faculty of Medicine and Health Sciences, Division of Gynecologic Oncology, Department of Obstetrics and Gynecology, University of Sherbrooke, Sherbrooke, Quebec, Canada, 6 Faculty of Medicine, Division of Gynecologic Oncology, Department of Obstetrics and Gynecology, University of Montreal, Montreal, Quebec, Canada, 7 Research Center of the Centre Hospitalier de l’Universite´ de Montre´al, Montreal, Quebec, Canada, 8 Faculty of Medicine, Division of Gynecologic Oncology, Department of Obstetrics and Gynecology, McGill University, Montreal, Quebec, Canada, 9 Lady Davis Institute of the Jewish General Hospital, Montreal, Quebec, Canada, 10 Faculty of Medicine, Departments of Obstetrics and Gynecology and Social and Preventive Medicine, University of Montreal, Montreal, Quebec, Canada * melanie.m.morin@usherbrooke.ca Abstract Background A large proportion of gynecological cancer survivors suffer from pain during sexual inter- course, also known as dyspareunia. Following a multimodal pelvic floor physical therapy (PFPT) treatment, a reduction in pain and improvement in psychosexual outcomes were found in the short term, but no study thus far has examined whether these changes are sus- tained over time. Purpose To examine the improvements in pain, sexual functioning, sexual distress, body image con- cerns, pain anxiety, pain catastrophizing, painful intercourse self-efficacy, depressive symp- toms and pelvic floor disorder symptoms in gynecological cancer survivors with dyspareunia after PFPT, and to explore women’s perceptions of treatment effects at one-year follow-up. Methods This mixed-method study included 31 gynecological cancer survivors affected by dyspareu- nia. The women completed a 12-week PFPT treatment comprising education, manual a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Cyr M-P, Dostie R, Camden C, Dumoulin C, Bessette P, Pina A, et al. (2022) Improvements following multimodal pelvic floor physical therapy in gynecological cancer survivors suffering from pain during sexual intercourse: Results from a one- year follow-up mixed-method study. PLoS ONE 17(1): e0262844. https://doi.org/10.1371/journal. pone.0262844 Editor: Diego Raimondo, Dipartimento di Scienze Mediche e Chirugiche (DIMEC), Orsola Hospital, ITALY Received: September 10, 2021 Accepted: January 6, 2022 Published: January 25, 2022 Copyright: © 2022 Cyr et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper. Funding: The Quebec Network for Research on Aging funded the current study. The Fonds de recherche du Que´bec – Sante´ granted a scholarship to Marie-Pierre Cyr and salary awards to Me´lanie Morin, Chantal Camden and Marie- He´lène Mayrand. The Canadian Research Chair Tier PLOS ONE | https://doi.org/10.1371/journal.pone.0262844 January 25, 2022 1 / 20 PLOS ONE II on Urogynecological Health and Aging supported Chantale Dumoulin. The laboratory infrastructures were funded by the Canadian Foundation for Innovation. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Multimodal pelvic floor physical therapy for cancer survivors suffering from pain during sexual intercourse therapy and pelvic floor muscle exercises. Quantitative data were collected using validated questionnaires at baseline, post-treatment and one-year follow-up. As for qualitative data, semi-structured interviews were conducted at one-year follow-up to better understand wom- en’s perception and experience of treatment effects. Results Significant improvements were found from baseline to one-year follow-up on all quantitative outcomes (P � 0.028). Moreover, no changes were found from post-treatment to one-year follow-up, supporting that the improvements were sustained at follow-up. Qualitative data highlighted that reduction in pain, improvement in sexual functioning and reduction in urinary symptoms were the most meaningful effects perceived by participants. Women expressed that these effects resulted from positive biological, psychological and social changes attrib- utable to multimodal PFPT. Adherence was also perceived to influence treatment outcomes. Conclusions Findings suggest that the short-term improvements following multimodal PFPT are sus- tained and meaningful for gynecological cancer survivors with dyspareunia one year after treatment. Introduction An increasing number of women live with the deleterious, long-term consequences of cancer [1,2]. Alongside urinary incontinence, chronic pain during sexual intercourse, also known as dyspareunia, is one of the most common sexual health issues, affecting more than half of gyne- cological cancer survivors [3,4]. Dyspareunia is recognized as resulting from the complex interaction of anatomical, physiological, psychological and relationship factors related to can- cer and oncological treatments [5], in line with the biopsychosocial model [6,7]. Vaginal steno- sis, impaired tissue flexibility, heightened pelvic floor muscle tone and contractility impairments as well as vaginal dryness [5,8] may contribute to experiencing pain during inter- course. These biological factors interplay with pain anxiety (i.e., fear of pain), pain catastro- phizing [9] and low pain self-efficacy [10], thereby intensifying the pain [11]. Gynecological cancer survivors are also at risk of depressive symptoms and body image concerns [12,13], which may disturb how they perceive themselves as women [14–16]. These pain and psycho- logical factors may contribute to sexual distress [17,18]. Moreover, women who have been treated for gynecological cancer are often affected by other sexual dysfunctions such as loss of libido or sexual desire [17]. All this can lead to relationship difficulties [12,13], disrupting their quality of life [19–21]. Despite the high prevalence of dyspareunia, there are limited treatment options supported by empirical evidence. Clinical survivorship guidelines suggest multimodal pelvic floor physi- cal therapy (PFPT) as a nonhormonal, non-pharmacological and non-invasive first-line treat- ment to alleviate dyspareunia in cancer survivors [22–24]. Through psychosexual education, manual therapy techniques and pelvic floor muscle exercises, PFPT targets the consequences of oncological treatments by restoring the pelvic floor tissues [8] while providing support and guidance to women to resume painless sexual activities [25,26]. So far, only one recent PLOS ONE | https://doi.org/10.1371/journal.pone.0262844 January 25, 2022 2 / 20 PLOS ONE Multimodal pelvic floor physical therapy for cancer survivors suffering from pain during sexual intercourse multicenter prospective study conducted by our team investigated a 12-week PFPT treatment in this population [27]. Significant changes in biological and psychosexual outcomes were found following treatment [27–29]. Using a comprehensive assessment combining intra-vagi- nal dynamometry and ultrasound imaging, pelvic floor muscle tone was significantly reduced while tissue flexibility, muscle contractile properties, control as well as endurance significantly improved immediately after treatment [28]. An increase in vaginal dimensions and a reduction in vaginal atrophy signs were also measured [28]. Concurrently, pain during intercourse, sex- ual distress, body image concerns, pain anxiety, pain catastrophizing, depressive symptoms, urinary symptoms, vaginal symptoms and sexual matters decreased while sexual functioning and pain self-efficacy improved after PFPT [27,29]. To date, no study has examined whether the short-term improvements following PFPT in gynecological cancer survivors with dyspareunia are sustained over time. Long-term treatment effects have important socioeconomic implications [30,31], and evaluating them may provide critical insights beyond those assessed in the short term [32]. More importantly, using only quantitative methods may not be sufficient to fully capture the extent of PFPT effects as these are multidimensional and likely depend on the interaction of multiple factors [6]. Further- more, it has been recently recognized that PFPT is not only a physical treatment but it is also a behavioral treatment, which emphasizes the relevance of investigating physical, cognitive and behavioral outcomes associated with PFPT [33]. Combining quantitative and qualitative meth- ods would therefore provide a better understanding of the treatment effects and how they influence each other considering the clinical context of multimodal PFPT [34,35]. This mixed- method study aimed to examine the improvements in pain, sexual functioning, sexual distress, body image concerns, pain anxiety, pain catastrophizing, pain self-efficacy, depressive symp- toms and pelvic floor disorder symptoms in gynecological cancer survivors with dyspareunia after multimodal PFPT, and to explore women’s perceptions of treatment effects at one-year follow-up. Materials and methods Design and methodology This study is a planned follow-up study of a multicenter prospective interventional study investigating the treatment effects of multimodal PFPT for gynecological cancer survivors with dyspareunia [27]. Our intent was to follow the whole cohort instead of a subsample in order to most closely match the primary trial (mainly in terms of participant characteristics and study outcomes) [32]. This research was conducted in Sherbrooke and Montreal (Can- ada). Changes from baseline to post-treatment have been published elsewhere [27–29], and changes from baseline and post-treatment to one-year follow-up will be the focus of the present manuscript. The participants underwent baseline, post-treatment and one-year fol- low-up assessments. Quantitative data were collected at all time points. To ascertain and advance our understanding of treatment effects at one-year follow-up, individual semi- structured telephone interviews were carried out to collect qualitative data [34,35]. This research was approved by the Ethics Review Board of the CIUSSS de l’Estrie–CHUS (MP- 31-2016-1322) and was registered on ClinicalTrials.gov (NCT03935698). Participants pro- vided written informed consent. Participants Women were included according to the following criteria: (i) all planned oncological treat- ments for either endometrial or cervical cancer (stages ranging from I to IV) completed for at least three months; (ii) in remission given the absence of disease on radiologic imaging PLOS ONE | https://doi.org/10.1371/journal.pone.0262844 January 25, 2022 3 / 20 PLOS ONE Multimodal pelvic floor physical therapy for cancer survivors suffering from pain during sexual intercourse for at least three months; (iii) moderate to severe vulvovaginal pain during sexual inter- course (i.e., pain at the entry of the vagina and at the mid-vagina, at the level of the pelvic floor muscles), corresponding to a pain intensity of 5 or more on a Numerical Rating Scale (NRS) ranging from 0 (no pain) to 10 (worst pain); (iv) vulvovaginal pain experienced in more than 80% of sexual intercourse for at least three months; (v) a stable sexual partner; and (vi) willingness to attempt vaginal penetrations. A gynecologic oncologist of our team at each site performed a standardized gynecological examination to rule out other condi- tions possibly causing dyspareunia (e.g., vaginitis, cystitis or dermatitis). Exclusion criteria were: (i) inability to communicate in French or English; (ii) dyspareunia prior to cancer or pelvic pain unrelated to intercourse; (iii) other pelvic conditions including urinary tract or vaginal infection, deep pelvic pain (i.e., pain experienced in the abdomen with deep pene- tration), chronic constipation, severe pelvic organ descent based on the Pelvic Organ Pro- lapse–Quantification system (stage III or more); (iv) other primary pelvic cancer or breast cancer; (v) any history of vulvar, vaginal or pelvic surgery unrelated to cancer; (vi) PFPT in the last year; (vii) changes in the use or dosage of menopausal hormone therapy in the last six months; (viii) a major medical or psychological condition likely to interfere with study procedures; or (ix) refusal to abstain from using other treatments for dyspareunia until the post-treatment assessment. Treatment content The treatment protocol was designed by a multidisciplinary team consisting of experts in gyne- cologic oncology, physical therapy, psychology and sexual health. The treatment included 12 weekly sessions of 60 minutes with a physical therapist certified and experienced in pelvic and women’s health. The treatment components were chosen to reflect practice in a clinical setting [36]. At each session, the physical therapist provided information, advice and support to women. She explained the underlying mechanisms of chronic pain experienced during sexual intercourse after gynecological cancer including the role of the pelvic floor muscles and how the treatment could help to reduce the pain. She gave additional information about how to manage chronic pain and other pelvic floor disorder symptoms (e.g., bladder training). The use of relaxation techniques using deep breathing as well as the application of vaginal lubri- cants and moisturizers were encouraged. The physical therapist also helped the participants gain more knowledge about sexual functioning (i.e., physiology of desire, excitation and orgasm) and guided them into resuming non-painful sexual activities with their partners. The latter was invited to participate in the treatment to help his partner in this process. Moreover, the physical therapist was available to further discuss topics with the participants who were invited to reflect on their sexual difficulties in order to overcome them with the help of their therapist. At each session, manual therapy techniques (i.e., stretching, myofascial release and tissue desensitization) and pelvic floor muscle exercises with electromyography biofeedback (i.e., relaxation, motor control, strength and endurance) using a small intra-vaginal probe were used. Women were also asked to perform home exercises resembling those performed under supervision five times per week as well as auto-insertion exercises with a finger or graded vaginal dilator in addition to desensitization techniques three times per week. Throughout the treatment, the physical therapist supervised each woman’s progress and pro- vided feedback. Additionally, modalities were intensified (e.g., more pressure applied to stretch the tissues, longer duration of the technique or exercise and greater dilator size) following each woman’s progress. At the end of the treatment, women were encouraged to pursue home exer- cises two to three times per week to maintain the effects of treatment. Further details of the treatment modalities are presented elsewhere [27]. PLOS ONE | https://doi.org/10.1371/journal.pone.0262844 January 25, 2022 4 / 20 PLOS ONE Multimodal pelvic floor physical therapy for cancer survivors suffering from pain during sexual intercourse Data collection Participants were assessed at baseline, post-treatment and one-year follow-up. Sample charac- teristics were collected at baseline. At each time point, quantitative outcomes were assessed using validated scales and questionnaires. After the collection of quantitative data at one-year follow-up, an individual semi-structured telephone interview was conducted in French or in English to further explore women’s perceptions of treatment effects. Participants were also asked if there were any changes regarding their health (e.g., cancer recurrence), if they were pursuing the home exercises, if they had attempted other treatments for pain or sexual dys- function and if their relationship status had changed during the follow-up period. Study outcomes Quantitative. The NRS was used to evaluate the average intensity of pain during inter- course [37]. The McGill Pain Questionnaire (MPQ) was used to qualify the pain according to its sensory, affective and evaluative dimensions, with higher scores corresponding to more sig- nificant pain [38]. The Female Sexual Function Index (FSFI) was used to examine sexual func- tioning including desire, arousal, lubrication, orgasm, satisfaction and pain, with higher total scores representing a better sexual function [39,40]. The Female Sexual Distress Scale-Revised (FSDS-R) was used to assess sexual distress, with higher scores relating to more sexual distress [41,42]. The Body Image Scale (BIS) was administered to evaluate body image concerns, with higher scores indicating greater concerns [43]. The Pain Anxiety Symptom Scale (PASS), which is an indirect measure of fear of pain during intercourse, was used to assess pain-related anxiety, with higher scores indicating more severe pain anxiety [44]. The Pain Catastrophizing Scale (PCS) was used to evaluate the exaggerated negative cognitions and emotions regarding pain, with higher scores pointing to greater pain catastrophizing [45]. The Painful Intercourse Self-Efficacy Scale (PISES) was used to assess pain self-efficacy associated with painful sexual intercourse, with higher scores representing better self-efficacy [46]. The Beck Depression Inventory-II (BDI-II) was used to evaluate depressive symptoms, with higher scores corre- sponding to higher severity of symptoms [37]. Pelvic floor disorder symptoms including uri- nary symptoms, vaginal symptoms and sexual matters were assessed with the International Consultation on Incontinence Questionnaire (ICIQ) modules. The ICIQ-Urinary Inconti- nence Short Form (ICIQ-UI SF) was used for urinary symptoms [47] and the ICIQ-Vaginal Symptoms (ICIQ-VS) for vaginal symptoms and sexual matters [48], with higher scores repre- senting more symptoms or sexual matters [47,48]. In addition, the Patient Global Impression of Change (PGIC) allowed the participants to self-report their perceived improvement (catego- ries ranging from very much improved to very much worse) [49]. Qualitative. Prior to their individual semi-structured telephone interview, participants were informed of the interview topics and invited to reflect on the treatment effects they per- ceived and how these effects evolved over time during the follow-up period. Each interview lasted approximately 70 minutes. The first author (MPC) underwent qualitative research train- ing to conduct all the interviews. She was not involved in participants’ care and was blinded to the participants’ responses in the questionnaires to avoid any preconceived ideas about the treatment effects. Before conducting the interviews, the interviewer reconfirmed the women’s consent to participate in the interviews and for recording the conversation. She used a non- judgmental approach and created a trustful and respectful relationship to ease the discussion of what could be perceived by participants as sensitive topics. Interviews followed a semi-struc- tured guide co-constructed by the first author (MPC), the principal investigator (MM) and another research team member who has extensive experience conducting qualitative research (CC) (see S1 File for the interview guide). The interview questions related to this manuscript’s PLOS ONE | https://doi.org/10.1371/journal.pone.0262844 January 25, 2022 5 / 20 PLOS ONE Multimodal pelvic floor physical therapy for cancer survivors suffering from pain during sexual intercourse research objective focused on women’s perceptions of treatment effects and their hypotheses about factors influencing these effects. Probing questions aimed to obtain in-depth informa- tion about participants’ perceptions of treatment effects, exploring short-term effects previ- ously documented in quantitative research [27–29] and using a biopsychosocial approach of health to explore any further effects and factors perceived to influence these effects [6,7]. The semi-structured guide was pilot-tested with a patient partner under the supervision of the principal investigator (MM) and the other research team member (CC). Sample size An a priori sample size was calculated for the multicenter prospective interventional study based on the proportion of completed home exercises (80%) as adherence was suggested as being important to perceive significant effects in physical therapy [50]. With a confidence level of 95%, an interval width of 30%, and to account for potential dropouts over time (15%), a total of 31 women were initially recruited for quantitative purposes (further details are avail- able elsewhere) [27]. All these women were invited to take part in an individual semi-struc- tured telephone interview to explore all of the various perceptions of treatment effects. Data analysis Quantitative data analysis was performed using IBM SPSS Statistics 27 (IBM Corporation, Armonk, N.Y., USA). Descriptive statistics were used to present baseline and one-year sample characteristics as well as PGIC results. Intention-to-treat analyses (i.e., all participants are included in the statistical analysis, regardless of their level of adherence) were conducted to explore whether the improvements in all outcomes were sustained at one-year follow-up. Out- comes at baseline and one-year follow-up as well as the changes from baseline and post-treat- ment to one-year follow-up are reported and expressed as mean estimated values (95% confidence interval) according to linear mixed modeling with Bonferroni correction [51–53]. Models included time as the fixed effect and random intercepts for each subject to account for repeated measures (i.e., to accommodate within-subject correlation). Statistical significance was set at p-value < 0.05 (two-tailed). Qualitative data analysis was based on the audio-recorded interviews which were transcribed and analyzed by the first author (MPC) using NVivo (version 12) software. A thematic analysis was adopted to ensure data-driven analyses and interpretations [54]. Specifically, an inductive approach was used when the first author (MPC) coded key ideas and started identifying emerg- ing themes. Subsequently, another team member (RD) reviewed the codes. Coding disagree- ments were discussed until a consensus was achieved. Codes were reviewed by two research team members (MM and CC), and several meetings were held to regroup codes into themes. Relationships between themes were explored by observing patterns across themes. As most of the original quotations used in this manuscript were in French, they were translated into English and revised by a certified translator. Field notes were used to explore researcher reflexivity and further support the data interpretation. It should be noted that results from quantitative and qualitative methods were integrated during the interpretation phase of the study. Results Participant characteristics Thirty-one women enrolled initially in this study. Fig 1 shows the flow of participants through the study. Additional details on screening and eligibility assessments are available elsewhere [27]. PLOS ONE | https://doi.org/10.1371/journal.pone.0262844 January 25, 2022 6 / 20 PLOS ONE Multimodal pelvic floor physical therapy for cancer survivors suffering from pain during sexual intercourse Fig 1. Flow of participants through the study. https://doi.org/10.1371/journal.pone.0262844.g001 Baseline sample characteristics (n = 31) are presented in Table 1. Before the multimodal PFPT treatment, women had an average pain intensity of 7.3 (6.7 to 8.0) on the NRS and the median duration of pain was approximately three years. Of the 29 women assessed at one-year follow-up, three reported having had a cancer recurrence or another cancer dur- ing the follow-up period, and one was recovering from a severe upper urinary tract infection. Study outcomes Quantitative. The quantitative outcomes assessed at baseline and one-year follow-up as well as the changes from baseline and post-treatment to one-year follow-up are presented in Table 2. Significant improvements were found from baseline to one-year follow-up on all out- comes (P � 0.028). Moreover, changes from post-treatment to one-year follow-up were statis- tically non-significant (P � 0.084), suggesting that the improvements were maintained over time. Of the 29 women assessed at one-year follow-up, 25 (86%) reported being very much or much improved. The others reported minimal improvements (7%), no changes (3%) or being minimally worse (3%) compared to baseline. Concerning the adherence to home exercises, 18 (62%) performed the pelvic floor muscle exercises during the follow-up period, with a median frequency of three times (two to eight) per month. Moreover, 10 (34%) participants performed the auto-insertion exercises, with a median frequency of three times (one to five) per month. PLOS ONE | https://doi.org/10.1371/journal.pone.0262844 January 25, 2022 7 / 20 PLOS ONE Multimodal pelvic floor physical therapy for cancer survivors suffering from pain during sexual intercourse Table 1. Sample characteristics at baseline. Characteristics Age (years), mean (SD) Body mass index (kg/m2), mean (SD) Cancer type, n (%) Endometrial Cervical Disease stage, n (%) I II III IV Time since oncological treatments (months), median (Q1 to Q3) Oncological treatments, n (%) Surgery alone Surgery + brachytherapy or external beam radiation therapy Surgery + brachytherapy + external beam radiation therapy + chemotherapy Surgery + chemotherapy Brachytherapy + external beam radiation therapy + chemotherapy SD, standard deviation; n, number of participants; Q1, first quartile; Q3, third quartile. https://doi.org/10.1371/journal.pone.0262844.t001 Value 55.9 (10.8) 28.5 (5.3) 20 (64.5) 11 (35.5) 19 (61) 6 (19) 5 (16) 1 (3) 38 (9 to 70) 9 (29) 6 (19) 7 (23) 2 (6) 7 (23) No women stated having attempted other treatments for pain or sexual dysfunction during this period, and only one reported being no longer with her partner at one-year follow-up. Qualitative. Three main themes were described by participants as the most meaning- ful treatment effects for them in terms of symptoms or functioning: (a) reduction in pain during intercourse; (b) improvement in sexual functioning; and (c) reduction in urinary symptoms. These themes are detailed below along with participants’ perceived modulating and contributing factors. Modulating factors were defined as the factors altering the mag- nitude of the main effects (e.g., adherence) while contributing factors were those described as other treatment effects which influenced positively the main effects (e.g., reduction in muscle tensions). Fig 2 illustrates how the main treatment effects (in black) interacted and were influenced by various biological, psychological or social factors (in grey). THEME 1. Reduction in pain during intercourse. All participants reported experiencing less pain during intercourse, with several stating having no pain at all since the end of the PFPT treatment. Although the majority expressed that this effect was maintained, a small number of women said that the pain reduction was attenuated at one-year follow-up. Among the poten- tial explanations, some of them suggested that discontinuing home exercises or stopping regu- lar sexual intercourse with vaginal penetration might have contributed to this depletion effect. “It fixed my pain problem and it lasted over time.”–C02 “I would say that it has deteriorated a bit since, but it’s my fault because I didn’t keep doing the exercises long enough. I know if I resumed the exercises it would get better. However, it [the pain] hasn’t come back to how it was before; in other words what has been done has been of benefit. Having sexual intercourse regularly helps to ensure these gains are maintained in a way.”–C12 PLOS ONE | https://doi.org/10.1371/journal.pone.0262844 January 25, 2022 8 / 20 PLOS ONE Multimodal pelvic floor physical therapy for cancer survivors suffering from pain during sexual intercourse Table 2. Outcomes at baseline (n = 31) and one-year follow-up (n = 29) and changes from baseline and post-treatment to one-year follow-up. Pain intensity NRS (0–10) 7.3 (6.7 to 8.0) 2.7 (2.0 to 3.3) -4.6 (-5.7 to -3.6) Baseline One-year follow-up Changes from baseline to follow-up Pain quality MPQ (0–78) 21.1 (17.6 to 24.6) 6.7 (3.1 to 10.4) -14.4 (-20.5 to -8.3) Sexual function FSFI (2–36) Sexual distress FSDS-R (0–52) 18.9 (16.3 to 21.4) (n = 20)b 26.7 (22.3 to 31.1) 23.4 (20.8 to 26.0) (n = 18)b 16.6 (12.1 to 21.1) 4.6 (1.0 to 8.1) -10.0 (-15.7 to -4.4) Body image concerns BIS (0–30) 6.4 (4.8 to 7.9) 3.0 (1.4 to 4.6) -3.4 (-5.4 to -1.3) Pain anxiety PASS (0–100) 37.5 (32.4 to 42.7) 23.7 (18.4 to 28.9) -13.9 (-21.6 to -6.2) Pain catastrophizing PCS (0–52) 20.9 (16.6 to 25.2) 8.3 (3.9 to 12.7) -12.6 (-18.1 to -7.1) Painful intercourse self-efficacy PISES (10–100) 63.6 (58.1 to 69.0) 80.6 (75.0 to 86.2) 17.1 (10.1 to 24.1) Depressive symptoms BDI-II (0–63) 10.9 (8.0 to 13.9) 7.5 (4.5 to 10.5) Urinary symptoms ICIQ-UI (0–21) 3.8 (2.5 to 5.2) 1.8 (0.4 to 3.3) Vaginal symptoms ICIQ-VS (0–53) 13.5 (11.5 to 15.4) 7.2 (5.2 to 9.2) -3.5 (-6.6 to -0.3) -2.0 (-3.3 to -0.6) -6.3 (-8.6 to -4.0) Sexual matters ICIQ-VS (0–58) 43.7 (37.7 to 49.7) (n = 24)c 20.9 (14.8 to 27.0) (n = 23)c -22.8 (-32.3 to -13.4) Pa < 0.001 < 0.001 0.009 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 0.028 0.002 < 0.001 < 0.001 Changes from post-treatment to follow-up 1.0 (-0.1 to 2.0) -0.5 (-6.7 to 5.6) -2.8 (-6.2 to 0.5) 2.7 (-2.9 to 8.4) 0.1 (-1.9 to 2.1) 2.8 (-5.0 to 10.5) 0.6 (-5.0 to 6.1) -6.3 (-13.4 to 0.7) 1.1 (-2.1 to 4.2) -0.5 (-1.8 to 0.9) -0.4 (-2.7 to 1.9) 1.2 (-8.0 to 10.3) Pa 0.084 1.000 0.119 0.708 1.000 1.000 1.000 0.095 1.000 1.000 1.000 1.000 The data shown are the mean estimated values (95% confidence interval) derived from the linear mixed models. a P-values extracted from the linear mixed modeling with Bonferroni correction. b Eleven women at baseline and 11 women at one-year follow-up did not engage in sexual activities including vaginal penetration in the last month and thereby, due to the one-month time frame used in the FSFI questionnaire, their total score could not be compilated. Reasons for not engaging in such activities at one-year follow-up: 4 = partner-related reasons including lack of sexual desire or medical problems such as erectile problems; 4 = participant-related reasons including lack of sexual desire (n = 2) or pain during intercourse (n = 2) although they reported a pain reduction of 4.5 and 5 on the NRS from baseline to one-year follow-up; 2 = relationship-related difficulties; 1 = medical indication to not engage due to vaginal bleeding unrelated to PFPT. c Seven participants at baseline and six at one-year follow-up did not engage in any form of sexual activities in the last month (time frame of ICIQ-VS for sexual matters). https://doi.org/10.1371/journal.pone.0262844.t002 Every participant associated the pain reduction with pelvic floor tissue changes. They noticed that the muscle tensions decreased while the tissue flexibility increased, attributing this to the manual techniques and the exercises. Some emphasized that relaxation techniques such as deep breathing promoted muscle relaxation, reduction of tensions, and hence, a pain relief. Overall, the women related these tissue changes to a less tense or deeper vagina, which allowed them to be more at ease and helped them to have a more complete and comfortable vaginal penetration with less or no pain. “All the exercises [contraction and stretching] I had done and what the physical therapist had done removed the tension and loosened me up. It felt good. Penetration was easier.”–C01 “The stretching we did reduced my pain because when it stretches better, it’s less painful. Oth- erwise, I felt like the skin inside wanted so badly to split because, before, it wouldn’t stretch.”– C18 “Breathing helps because I think when you calm down, it’s less contracted and there’s more flexibility for the activity.”–C16 PLOS ONE | https://doi.org/10.1371/journal.pone.0262844 January 25, 2022 9 / 20 PLOS ONE Multimodal pelvic floor physical therapy for cancer survivors suffering from pain during sexual intercourse Fig 2. Relationships between treatment effects that emerged from the interviews. https://doi.org/10.1371/journal.pone.0262844.g002 Many women also observed becoming more aware of the pelvic floor musculature and its relationship with pain. During the PFPT treatment, they recalled gaining control over their muscles and developing muscle awareness. Motor control was noted as being important by the participants to break a chain of events involving the pelvic floor muscles and pain. “When you are calmer, it [the pelvic floor muscle] is less contracted, so it is more flexible. [. . .] Before the treatments, I didn’t know how to do [relax my muscles], I was tense. Now, I have techniques that last over time. [. . .] I have gone from. . . not hysteria, but from an uncon- trolled fear to something more serene. I am calmer when considering having sex, I am more welcoming.”–C16 Our participants often mentioned being reassured knowing how to influence the pain. They frequently expressed being less afraid of pain because they understood what led to their symptoms and were taught relevant and effective tools to reduce it. “After cancer treatments, you feel diminished. Will it come back as before? I was starting to be afraid. With physical therapy, you feel less diminished. It seemed as if it was finally possible that things could get better. When I got into the program, it was another story as I realized it was possible to improve, and it was much less upsetting, less scary. It’s because we found PLOS ONE | https://doi.org/10.1371/journal.pone.0262844 January 25, 2022 10 / 20 PLOS ONE Multimodal pelvic floor physical therapy for cancer survivors suffering from pain during sexual intercourse where it hurt most. It’s about understanding. . . It’s partly confidence, partly the fear that’s gone.”–C124 Consequently, they explained that they were feeling more in control, self-efficient and hopeful while being less anxious about their pain. Some participants even emphasized that they were no longer afraid to undergo gynecological examinations. Experiencing less pain dur- ing intercourse also enhanced these feelings, which in turn amplified their self-esteem and confidence to engage in sexual activities. They felt less distressed, with several highlighting that fact they were less depressed and more positive in their everyday lives. “And what I also learned was that I felt that I could influence my pain. When it’s less painful, less tight, you are more relaxed, you have more confidence and you let go more easily. Psycho- logically, I could say that I felt I was moving further away from the operation and its negative side. I found that I was getting closer to a more normal life, as it was before, in a sense. . . with- out much difficulty. Yes, it’s vague, isn’t it? Well, normal life. . . having sex again, get away from the cancer thing.”–C115 THEME 2. Improvement in sexual functioning. All women reported improvement in their sexual functioning following PFPT. Although a low proportion of participants did not perceive changes in their sexual functioning in terms of lubrication and libido or sexual desire, the vast majority mentioned their vagina being less dry and more naturally lubricated during sexual activities. Among other things, several women emphasized not needing to use vaginal products anymore and reported being less stressed and more interested in engaging in sexual activities. “The lubrication. . . it all came basically together after the treatment. Sure, at first I needed some lubricant, but little by little, as I worked, it just faded so I didn’t need the lubricant any- more.”–C09 The perceptions relating to pain reduction described previously could also suggest how par- ticipants felt about sexuality. Many of them reported being more interested in engaging given the pain reduction and the positive emotions and thoughts they developed about their sexual identity. Some women associated their increased sexual desire to the improved perception of their body, which defined them as women. They grew to accept themselves, sensed that their body belonged to them and reclaimed it. Participants specified that this body re-appropriation helped them to express themselves sexually as women. They were able to have sexual inter- course with vaginal penetration rather than endure the barriers induced by cancer, which hampered them. Consequently, they referred to being complete women and having a more normal life. Participants related that regaining the capacity of having intercourse helped them initiate and engage in sexual activities, which in turn increased their femininity. “I could see that there were still defects in my body since the operation and all that, and psy- chologically it disturbed me. Now, I let myself go more. There is a connection that has been made with my body and my whole person. I participate more with my body now, which I didn’t before. I had an easier time opening up to sexuality. That’s why I say it really. . . changed my life. Physical therapy is beneficial, it is a psycho-unblocker.”–C100 “Knowing what to do to have intercourse and being able to have it [sexual intercourse] really made me feel like a woman. I am very happy to have learned to control my body better and to be able to have a more fulfilling sex life. It’s like. . . I feel like more of a complete woman, I don’t know. . . entirely a woman.”–C11 PLOS ONE | https://doi.org/10.1371/journal.pone.0262844 January 25, 2022 11 / 20 PLOS ONE Multimodal pelvic floor physical therapy for cancer survivors suffering from pain during sexual intercourse “Basically, sexuality is more about being a complete woman, [. . .] Now, if I feel like having sex, I can have it. [. . .] So, life for me is much more normal than it used to be. It changed my life, it gave me back intimacy. So, we’re less active than we were, but at least if we want to, we can! So that’s the difference.”–C10 Participants also recognized that they were more comfortable talking about sexuality. They stressed that this led them to communicate more about their feelings and difficulties to their partner. As a result, participants and their partners were more capable of adapting their behav- ior, and when considering physical intimacy, it was therefore less stressful and more pleasur- able. Furthermore, participants said that because they had less pain during intercourse, their partner was less afraid to hurt them, and this dynamic was helpful for the couple to be physi- cally intimate. “I was also able to talk about it [thoughts about intercourse] with my partner because I had not talked about it before. When I had intercourse before, it was because I felt obliged. It was very rare that we had any. With the study, it was like day and night, winter and summer. It was like having sex two or three times a week by the end of the study.”–C06 “I was no longer in pain. . . well, for sure in our intimate relationship and all that there was a letting go so that was really amazing. Less fear, less apprehension. Yes, I think it reassured my husband a lot to see that it was going well, that it was getting better. He was also less afraid of hurting me and he was more reassured that there were two of us in this sexual activity.”–C08 Because they were more communicative, most women acknowledged that they and their partner discussed their sexuality and intimacy more openly. Those who did not report any changes in this regard claimed their relationship was already strong and without issues before enrolling in the study. The former noticed that they and their partner were closer to each other, discovered and tried new ways to express their love. Several participants spoke of how it became more affectionate than sexually demonstrative with intercourse during the one-year follow-up period. For a handful of women, this was accentuated if there had been a significant event (e.g., cancer recurrence), low sexual desire, pain during intercourse or a medical condi- tion of the partner. “It helped me to understand how my body reacted to a lot of things, to understand that I was not alone and it helped me to accept myself and accept living my sex life in a different way. It [the treatment] allowed us to make different connections. There is a lot, really a lot of affec- tion. It starts slowly, and, in the end, it becomes intense. This is what is new, this is what we learned.”–C17B THEME 3. Reduction in urinary symptoms. Half of the sample experienced either stress uri- nary incontinence, urgency urinary incontinence or symptoms of urinary urgency before the study and all women reported significant improvements following PFPT. Participants observed that the pelvic floor muscle exercises in addition to bladder training increased their muscle awareness, strength and endurance to activate their pelvic floor muscles when needed. For instance, it gave them the means to delay the urge to urinate or to hold the urine for longer periods. “Before, I used to go to the bathroom. . . a lot! Almost every hour, and now I go like three or four times a day and that’s enough. So, for sure, there is a difference there as well.”–C14 PLOS ONE | https://doi.org/10.1371/journal.pone.0262844 January 25, 2022 12 / 20 PLOS ONE Multimodal pelvic floor physical therapy for cancer survivors suffering from pain during sexual intercourse “I used to go to the bathroom all the time, all the time, and she [the physical therapist] gave me some tips for the bladder and exercises, and it’s getting better in that respect too.”–C111 “All the exercises, the squeezing and all that helped. You squeeze and it calms your bladder. I didn’t think it would work. Listen, I can even hold my urine when I go to the bathroom. . . Before, when I saw the toilet, I had to run and when I saw the toilet bowl, I leaked two or three drops. But now, I am able to hold it. I know what to do.” –C10 Interestingly, two women said that having had painful urination and difficulty retaining high volumes of urine since the oncological treatments and they explained that, by releasing tensions in the pelvic area, the PFPT modalities such as manual therapy and auto-insertion exercises helped them to resolve these issues. “It was stiff near the bladder and it hurt. I felt the bladder was jammed, it was like there was no room for it to fill up. So, the physical therapy helped to relax the tensions and my bladder had more room so I needed to urinate less often. At night, I used to get up every three hours, I get up less now, so I sleep better. Everything is going in the right direction.”–C17B Discussion This mixed-method study provides evidence that the improvements in pain, sexual function- ing, sexual distress, body image concerns, pain anxiety, pain catastrophizing, painful inter- course self-efficacy, depressive symptoms, urinary symptoms, vaginal symptoms and sexual matters following multimodal PFPT can be sustained at one-year follow-up in gynecological cancer survivors with dyspareunia. Furthermore, reduction in pain during sexual intercourse, improvement in sexual functioning and reduction in urinary symptoms were reported by par- ticipants as the most meaningful effects during the interviews. In addition, participants expressed these treatment effects in relation to adherence. They also emphasized that the treat- ment led to positive biological, psychological and social changes which contributed to the improvements in dyspareunia and sexual functioning. This is the first study to examine whether the short-term improvements following multi- modal PFPT are maintained over time in gynecological cancer survivors affected by dyspareu- nia [55]. Interventional studies conducted to date in women who had been treated for gynecological cancer were not specific to dyspareunia (e.g., urinary incontinence, vaginal atro- phy or low sexual desire) [56–64]. To our knowledge, only a few cohort studies included a fol- low-up assessment beyond six months [60,62,65,66]. Improvement in sexual functioning have been seen following interventions integrating psychosexual education and unsupervised pelvic floor exercises in gynecological cancer survivors [60,62], which is consistent with the current study. However, their target population was different as women with or without symptoms were included immediately after oncological treatments. The experimental interventions were also designed to prevent or address common symptoms in gynecological cancer survivors while not specifically targeting dyspareunia [60,62]. In contrast, our sample was probably more affected at baseline as all women presented a pain intensity of more than 5 on the NRS for a median duration of three years, representing chronic moderate-to-severe dyspareunia [67,68]. Despite this chronicity and severity, it is noteworthy that participants still observed and reported sustained significant effects one year later. The women in the present study expressed meaningful improvements in pain during inter- course, sexual functioning and urinary symptoms that lasted one year after PFPT. Similar find- ings were found in studies investigating multimodal PFPT effects in younger women suffering PLOS ONE | https://doi.org/10.1371/journal.pone.0262844 January 25, 2022 13 / 20 PLOS ONE Multimodal pelvic floor physical therapy for cancer survivors suffering from pain during sexual intercourse from vulvar pain with no history of cancer, although the available data is limited to a six- month follow-up in this population [69,70]. Morin et al. [70] in a large multicenter random- ized controlled trial (n = 212) revealed reductions in pain and sexual distress with improved sexual functioning from baseline to six-month follow-up, compared to topical lidocaine, a fre- quent first-line treatment. Moreover, a recent Cochrane meta-analysis concluded that pelvic floor muscle training can reduce or cure urinary symptoms in women without a history of can- cer [71], which is in line with our results. It is worth noting that the majority of studies con- ducted in women affected by dyspareunia with no history of cancer applied quantitative methods to evaluate the effects of multimodal PFPT [69,70,72]. A quantitative research design could only provide a narrow view of PFPT effects, as demonstrated in the current study. Quantitative results combined with the participants’ inputs suggest that multimodal PFPT improved multiple dimensions of the biopsychological framework of dyspareunia [6,9,11,73], and these improvements remained at one-year follow-up. More precisely, the effects on pain during intercourse, sexual functioning and urinary symptoms were explained by gynecological cancer survivors through biological, psychological and social changes attributable to PFPT modalities. Gynecological cancer survivors emphasized the role of multimodal PFPT in the effects perceived and how it helped them to achieve pain-free sexual activities or improve their sexual functioning or behavior. It is notable that the treatment not only improved the pelvic floor tissues, as observed in short-term studies using objective tools [28,72,74], but also had a direct or indirect positive impact on psychological and social dimensions according to our cohort. Qualitative data suggested that performing PFPT exercices or having sexual inter- course regularly could be important to retain the biological changes related to pain for certain women. These details show that treatment effects over time could depend on adherence in the long term. Comparing our results to the studies conducted in women with no history of can- cer, only two studies [75,76] to date have investigated the improvements following myofascial release techniques [76] and multimodal PFPT at three-month follow-up using a shorter inter- view [75] for dyspareunia in young women. The latter study reported similar effects in regard to muscle awareness, knowledge and communication about pain, self-efficacy, self-esteem, sexual confidence, attitudes about sexuality and relationship with the partner [75]. However, it should be underlined that our group of participants was still experiencing substantial effects at one-year follow-up after PFPT even though they had been treated for cancer, were older and had had dyspareunia for a median duration of three years. As opposed to previous work [75,76], our study is the first to triangulate data from different methods and to present exten- sively qualitative findings about multimodal PFPT effects by reporting the participants’ inputs that supported our interpretation while providing a deeper understanding. Overall, our find- ings suggest multimodal PFPT as a biopsychosocial treatment for reducing dyspareunia and improving sexual functioning. The main strength of this study is the integration of quantitative and qualitative methods to allow data triangulation and complementarity to fully capture the treatment effects [77–79]. Validated scales and questionnaires were used to assess the quantitative outcomes. Intention- to-treat analyses were conducted and considered multiple comparisons as well as missing data. The high participation rate in qualitative interviews promoted a wide range of perspectives and shed light on how multimodal PFPT could have influenced dimensions other than the well-known biological dimension. The mixed-method design has allowed us to illustrate elo- quently the quantitative findings supported by statistics and through the perceptions of women. Our results should, however, be interpreted within the context of certain limitations. The absence of a control group limits the causal inference. Nonetheless, the women’s percep- tions support the role of PFPT in leading to these effects. They also did not attempt other treat- ments during the follow-up period. Moreover, it is unlikely that they would have improved PLOS ONE | https://doi.org/10.1371/journal.pone.0262844 January 25, 2022 14 / 20 PLOS ONE Multimodal pelvic floor physical therapy for cancer survivors suffering from pain during sexual intercourse without any treatment, given that they were suffering from dyspareunia for a median time of approximately three years and that sexual issues tend to persist over time [80,81]. Even though these aspects are suggestive of a causal inference of PFPT on outcomes, a randomized con- trolled trial is ultimately required to confirm the long-term efficacy of this treatment. As the PFPT treatment combined multiple modalities, it is difficult to isolate their respective effect on the outcomes. Moreover, determining precisely how the treatment effects (i.e., reduction in pain during intercourse, improvement in sexual functioning and reduction in urinary symp- toms) and their modulating and contributing factors (i.e., adherence as well as biological, psy- chological and social changes) interacted was not feasible. It is worth mentioning that it has frequently been reported that these may overlap and influence each other dynamically and dif- ferently among gynecological cancer survivors [18,82,83]. A biopsychosocial treatment approach could have contributed to the magnitude of the effects [26]. Conclusions Findings of this one-year follow-up mixed-method study suggest that the short-term improve- ments in pain during sexual intercourse, sexual functioning and urinary incontinence follow- ing PFPT can be sustained over time in gynecological cancer survivors with dyspareunia. Although a randomized controlled trial is still required to confirm the efficacy, multimodal PFPT showed beneficial effects of treating dyspareunia in this population through biological, psychological and social changes after one year. The study therefore supports the biopsychoso- cial role of multimodal PFPT in gynecological cancer survivors who are frequently affected by pain and other types of sexual dysfunction. This treatment could be implemented in multidis- ciplinary cancer care. Supporting information S1 File. Semi-structured interview guide. (DOCX) Acknowledgments We would like to extend our gratitude to the physical therapists involved in the treatments and assessments. We would also like to thank all the study participants for their support and dedi- cation to this research project. Author Contributions Conceptualization: Marie-Pierre Cyr, Chantal Camden, Chantale Dumoulin, Me´lanie Morin. Data curation: Marie-Pierre Cyr. Formal analysis: Marie-Pierre Cyr, Rosalie Dostie, Chantal Camden, Me´lanie Morin. Funding acquisition: Marie-Pierre Cyr, Chantale Dumoulin, Paul Bessette, Walter Henry Gotlieb, Me´lanie Morin. Investigation: Marie-Pierre Cyr, Chantale Dumoulin, Paul Bessette, Annick Pina, Walter Henry Gotlieb, Korine Lapointe-Milot, Me´lanie Morin. Methodology: Marie-Pierre Cyr, Chantal Camden, Chantale Dumoulin, Paul Bessette, Annick Pina, Walter Henry Gotlieb, Marie-He´lène Mayrand, Me´lanie Morin. Project administration: Marie-Pierre Cyr, Chantale Dumoulin, Me´lanie Morin. PLOS ONE | https://doi.org/10.1371/journal.pone.0262844 January 25, 2022 15 / 20 PLOS ONE Multimodal pelvic floor physical therapy for cancer survivors suffering from pain during sexual intercourse Resources: Marie-Pierre Cyr. Supervision: Chantale Dumoulin, Me´lanie Morin. Validation: Marie-Pierre Cyr, Me´lanie Morin. Visualization: Marie-Pierre Cyr. Writing – original draft: Marie-Pierre Cyr. Writing – review & editing: Marie-Pierre Cyr, Rosalie Dostie, Chantal Camden, Chantale Dumoulin, Paul Bessette, Annick Pina, Walter Henry Gotlieb, Korine Lapointe-Milot, Marie-He´lène Mayrand, Me´lanie Morin. References 1. Howlader N, Noone AM, Krapcho M, Miller D, Brest A, Yu M, et al. SEER Cancer Statistics Review, 1975–2017, National Cancer Institute. Bethesda, 2020. 2. Torre LA, Islami F, Siegel RL, Ward EM, Jemal A. Global cancer in women: burden and trends. Cancer Epidemiol Biomarkers Prev. 2017; 26:444–57. https://doi.org/10.1158/1055-9965.EPI-16-0858 PMID: 28223433 3. Rutledge TL, Heckman SR, Qualls C, Muller CY, Rogers RG. Pelvic floor disorders and sexual function in gynecologic cancer survivors: a cohort study. Am J Obstet Gynecol. 2010; 203:514 e1–7. https://doi. org/10.1016/j.ajog.2010.08.004 PMID: 20869691 4. Stinesen Kollberg K, Waldenstrom AC, Bergmark K, Dunberger G, Rossander A, Wilderang U, et al. Reduced vaginal elasticity, reduced lubrication, and deep and superficial dyspareunia in irradiated gynecological cancer survivors. Acta Oncol. 2015; 54:772–9. https://doi.org/10.3109/0284186X.2014. 1001036 PMID: 25761090 5. Coady D, Kennedy V. Sexual health in women affected by cancer: focus on sexual pain. Obstet Gyne- col. 2016; 128:775–91. https://doi.org/10.1097/AOG.0000000000001621 PMID: 27607852 6. Bergeron S, Corsini-Munt S, Aerts L, Rancourt K, Rosen NO. Female sexual pain disorders: a review of the literature on etiology and treatment. Curr Sex Health Rep. 2015; 7:159–69. 7. Engel GL. The need for a new medical model: a challenge for biomedicine. Science. 1977; 196:129–36. https://doi.org/10.1126/science.847460 PMID: 847460 8. Cyr MP, Dumoulin C, Bessette P, Pina A, Gotlieb WH, Lapointe-Milot K, et al. Characterizing pelvic floor muscle function and morphometry in survivors of gynecological cancer who have dyspareunia: a comparative cross-sectional study. Phys Ther. 2021;101. https://doi.org/10.1093/ptj/pzab042 PMID: 33522584 9. 10. Thomte´ n J, Linton SJ. A psychological view of sexual pain among women: applying the fear-avoidance model. Women’s health (London, England). 2013; 9:251–63. https://doi.org/10.2217/whe.13.19 PMID: 23638781 Lemieux AJ, Bergeron S, Steben M, Lambert B. Do romantic partners’ responses to entry dyspareunia affect women’s experience of pain? The roles of catastrophizing and self-efficacy. J Sex Med. 2013; 10:2274–84. https://doi.org/10.1111/jsm.12252 PMID: 23809759 11. Corsini-Munt S, Rancourt KM, Dube´ JP, Rossi MA, Rosen NO. Vulvodynia: a consideration of clinical and methodological research challenges and recommended solutions. J Pain Res. 2017; 10:2425–36. https://doi.org/10.2147/JPR.S126259 PMID: 29070953 12. Juraskova I, Butow P, Robertson R, Sharpe L, McLeod C, Hacker N. Post-treatment sexual adjustment following cervical and endometrial cancer: a qualitative insight. Psychooncology. 2003; 12:267–79. https://doi.org/10.1002/pon.639 PMID: 12673810 13. Cleary V, Hegarty J. Understanding sexuality in women with gynaecological cancer. Eur J Oncol Nurs. 2011; 15:38–45. https://doi.org/10.1016/j.ejon.2010.05.008 PMID: 20584629 14. Bowes H, Jones G, Thompson J, Alazzam M, Wood H, Hinchliff S, et al. Understanding the impact of the treatment pathway upon the health-related quality of life of women with newly diagnosed endome- trial cancer—a qualitative study. Eur J Oncol Nurs. 2014; 18:211–7. https://doi.org/10.1016/j.ejon.2013. 10.007 PMID: 24290535 15. Reis N, Beji NK, Coskun A. Quality of life and sexual functioning in gynecological cancer patients: results from quantitative and qualitative data. Eur J Oncol Nurs. 2010; 14:137–46. https://doi.org/10. 1016/j.ejon.2009.09.004 PMID: 19836305 PLOS ONE | https://doi.org/10.1371/journal.pone.0262844 January 25, 2022 16 / 20 PLOS ONE Multimodal pelvic floor physical therapy for cancer survivors suffering from pain during sexual intercourse 16. Sekse RJ, Raaheim M, Blaaka G, Gjengedal E. Life beyond cancer: women’s experiences 5 years after treatment for gynaecological cancer. Scand J Caring Sci. 2010; 24:799–807. https://doi.org/10.1111/j. 1471-6712.2010.00778.x PMID: 20487404 17. Vermeer WM, Bakker RM, Kenter GG, Stiggelbout AM, Ter Kuile MM. Cervical cancer survivors’ and partners’ experiences with sexual dysfunction and psychosexual support. Support Care Cancer. 2016; 24:1679–87. https://doi.org/10.1007/s00520-015-2925-0 PMID: 26412245 18. Bakker RM, Kenter GG, Creutzberg CL, Stiggelbout AM, Derks M, Mingelen W, et al. Sexual distress and associated factors among cervical cancer survivors: a cross-sectional multicenter observational study. Psychooncology. 2017; 26:1470–7. https://doi.org/10.1002/pon.4317 PMID: 27862635 19. Stabile C, Gunn A, Sonoda Y, Carter J. Emotional and sexual concerns in women undergoing pelvic surgery and associated treatment for gynecologic cancer. Transl Androl Urol. 2015; 4:169–85. https:// doi.org/10.3978/j.issn.2223-4683.2015.04.03 PMID: 26816823 20. Abbott-Anderson K, Kwekkeboom KL. A systematic review of sexual concerns reported by gynecologi- cal cancer survivors. Gynecol Oncol. 2012; 124:477–89. https://doi.org/10.1016/j.ygyno.2011.11.030 PMID: 22134375 21. Izycki D, Wozniak K, Izycka N. Consequences of gynecological cancer in patients and their partners from the sexual and psychological perspective. Prz Menopauzalny. 2016; 15:112–6. https://doi.org/10. 5114/pm.2016.61194 PMID: 27582686 22. Carter J, Lacchetti C, Andersen BL, Barton DL, Bolte S, Damast S, et al. Interventions to address sexual problems in people with cancer: American Society of Clinical Oncology Clinical Practice Guideline Adaptation of Cancer Care Ontario Guideline. J Clin Oncol. 2018; 36:492–511. https://doi.org/10.1200/ JCO.2017.75.8995 PMID: 29227723 23. Huffman LB, Hartenbach EM, Carter J, Rash JK, Kushner DM. Maintaining sexual health throughout gynecologic cancer survivorship: a comprehensive review and clinical guide. Gynecol Oncol. 2016; 140:359–68. https://doi.org/10.1016/j.ygyno.2015.11.010 PMID: 26556768 24. Crean-Tate KK, Faubion SS, Pederson HJ, Vencill JA, Batur P. Management of genitourinary syndrome of menopause in female cancer patients: a focus on vaginal hormonal therapy. Am J Obstet Gynecol. 2020; 222:103–13. https://doi.org/10.1016/j.ajog.2019.08.043 PMID: 31473229 25. Morin M, Carroll MS, Bergeron S. Systematic review of the effectiveness of physical therapy modalities in women with provoked vestibulodynia. Sex Med Rev. 2017; 5:295–322. https://doi.org/10.1016/j. sxmr.2017.02.003 PMID: 28363763 26. Wijma AJ, van Wilgen CP, Meeus M, Nijs J. Clinical biopsychosocial physiotherapy assessment of patients with chronic pain: the first step in pain neuroscience education. Physiother Theory Pract. 2016; 32:368–84. https://doi.org/10.1080/09593985.2016.1194651 PMID: 27351769 27. Cyr MP, Dumoulin C, Bessette P, Pina A, Gotlieb WH, Lapointe-Milot K, et al. Feasibility, acceptability and effects of multimodal pelvic floor physical therapy for gynecological cancer survivors suffering from painful sexual intercourse: a multicenter prospective interventional study. Gynecol Oncol. 2020; 159:778–84. https://doi.org/10.1016/j.ygyno.2020.09.001 PMID: 33010968 28. Cyr MP, Dumoulin C, Bessette P, Pina A, Gotlieb WH, Lapointe-Milot K, et al. Changes in pelvic floor morphometry and muscle function after multimodal physiotherapy for gynaecological cancer survivors suffering from dyspareunia: a prospective interventional study. Physiotherapy. 2021;In press. 29. Cyr MP, Dumoulin C, Bessette P, Pina A, Gotlieb WH, Lapointe-Milot K, et al. A prospective single-arm study evaluating the effects of a multimodal physical therapy intervention on psychosexual outcomes in women with dyspareunia after gynecologic cancer. J Sex Med. 2021; 18:946–54. https://doi.org/10. 1016/j.jsxm.2021.02.014 PMID: 33931347 30. Hlatky MA, Owens DK, Sanders GD. Cost-effectiveness as an outcome in randomized clinical trials. Clinical trials (London, England). 2006; 3:543–51. https://doi.org/10.1177/1740774506073105 PMID: 17170039 31. Edmunds K, Ling R, Shakeshaft A, Doran C, Searles A. Systematic review of economic evaluations of interventions for high risk young people. BMC Health Serv Res. 2018; 18:660. https://doi.org/10.1186/ s12913-018-3450-x PMID: 30139384 32. 33. Fitzpatrick T, Perrier L, Shakik S, Cairncross Z, Tricco AC, Lix L, et al. Assessment of long-term follow- up of randomized trial participants by linkage to routinely collected data: a scoping review and analysis. JAMA. 2018; 1:e186019-e. https://doi.org/10.1001/jamanetworkopen.2018.6019 PMID: 30646311 Frawley HC, Dean SG, Slade SC, Hay-Smith EJC. Is pelvic-floor muscle training a physical therapy or a behavioral therapy? A call to name and report the physical, cognitive, and behavioral elements. Phys Ther. 2017; 97:425–37. https://doi.org/10.1093/ptj/pzx006 PMID: 28499001 34. Abildgaard JS, Saksvik PØ, Nielsen K. How to measure the intervention process? An assessment of qualitative and quantitative approaches to data collection in the process evaluation of organizational interventions. Front Psychol. 2016; 7:1380. https://doi.org/10.3389/fpsyg.2016.01380 PMID: 27713707 PLOS ONE | https://doi.org/10.1371/journal.pone.0262844 January 25, 2022 17 / 20 PLOS ONE Multimodal pelvic floor physical therapy for cancer survivors suffering from pain during sexual intercourse 35. Blackman T, Wistow J, Byrne D. Using qualitative comparative analysis to understand complex policy problems. Evaluation. 2013; 19:126–40. 36. Hartmann D, Strauhal MJ, Nelson CA. Treatment of women in the United States with localized, pro- voked vulvodynia: practice survey of women’s health physical therapists. J Reprod Med. 2007; 52:48– 52. PMID: 17286069 37. Dworkin RH, Turk DC, Wyrwich KW, Beaton D, Cleeland CS, Farrar JT, et al. Interpreting the clinical importance of treatment outcomes in chronic pain clinical trials: IMMPACT recommendations. J Pain. 2008; 9:105–21. https://doi.org/10.1016/j.jpain.2007.09.005 PMID: 18055266 38. Melzack R. The McGill Pain Questionnaire: major properties and scoring methods. Pain. 1975; 1:277– 99. https://doi.org/10.1016/0304-3959(75)90044-5 PMID: 1235985 39. Wiegel M, Meston C, Rosen R. The female sexual function index (FSFI): cross-validation and develop- ment of clinical cutoff scores. J Sex Marital Ther. 2005; 31:1–20. https://doi.org/10.1080/ 00926230590475206 PMID: 15841702 40. Meyer-Bahlburg HF, Dolezal C. The female sexual function index: a methodological critique and sug- gestions for improvement. J Sex Marital Ther. 2007; 33:217–24. https://doi.org/10.1080/ 00926230701267852 PMID: 17454519 41. Santos-Iglesias P, Mohamed B, Walker LM. A systematic review of sexual distress measures. J Sex Med. 2018; 15:625–44. https://doi.org/10.1016/j.jsxm.2018.02.020 PMID: 29576431 42. Derogatis L, Clayton A, Lewis-D’Agostino D, Wunderlich G, Fu Y. Validation of the female sexual dis- tress scale-revised for assessing distress in women with hypoactive sexual desire disorder. J Sex Med. 2008; 5:357–64. https://doi.org/10.1111/j.1743-6109.2007.00672.x PMID: 18042215 43. Hopwood P, Fletcher I, Lee A, Al Ghazal S. A body image scale for use with cancer patients. Eur J Can- cer. 2001; 37:189–97. https://doi.org/10.1016/s0959-8049(00)00353-1 PMID: 11166145 44. McCracken LM, Dhingra L. A short version of the Pain Anxiety Symptoms Scale (PASS-20): preliminary development and validity. Pain Res Manag. 2002; 7:45–50. https://doi.org/10.1155/2002/517163 PMID: 16231066 45. Sullivan MJL, Bishop SR, Pivik J. The Pain Catastrophizing Scale: development and validation. Psychol Assess. 1995; 7:524–32. 46. Desrochers G, Bergeron S, Khalife S, Dupuis MJ, Jodoin M. Fear avoidance and self-efficacy in relation to pain and sexual impairment in women with provoked vestibulodynia. Clin J Pain. 2009; 25:520–7. https://doi.org/10.1097/AJP.0b013e31819976e3 PMID: 19542801 47. Avery K, Donovan J, Peters TJ, Shaw C, Gotoh M, Abrams P. ICIQ: a brief and robust measure for eval- uating the symptoms and impact of urinary incontinence. Neurourol Urodyn. 2004; 23:322–30. https:// doi.org/10.1002/nau.20041 PMID: 15227649 48. Price N, Jackson SR, Avery K, Brookes ST, Abrams P. Development and psychometric evaluation of the ICIQ Vaginal Symptoms Questionnaire: the ICIQ-VS. BJOG. 2006; 113:700–12. https://doi.org/10. 1111/j.1471-0528.2006.00938.x PMID: 16709214 49. Farrar JT, Young JP Jr., LaMoreaux L, Werth JL, Poole RM. Clinical importance of changes in chronic pain intensity measured on an 11-point numerical pain rating scale. Pain. 2001; 94:149–58. https://doi. org/10.1016/S0304-3959(01)00349-9 PMID: 11690728 50. Venegas M, Carrasco B, Casas-Cordero R. Factors influencing long-term adherence to pelvic floor exercises in women with urinary incontinence. Neurourol Urodyn. 2018; 37:1120–7. https://doi.org/10. 1002/nau.23432 PMID: 29095511 51. Schielzeth H, Dingemanse NJ, Nakagawa S, Westneat DF, Allegue H, Teplitsky C, et al. Robustness of linear mixed-effects models to violations of distributional assumptions. Methods Ecol Evol. 2020; 11:1141–52. 52. Ashbeck EL, Bell ML. Single time point comparisons in longitudinal randomized controlled trials: power and bias in the presence of missing data. BMC Med Serv Res. 2016; 16:43. https://doi.org/10.1186/ s12874-016-0144-0 PMID: 27068578 53. Harrell FJ. Regression modeling strategies: with applications to linear models, logistic and ordinal regression, and survival analysis. 2nd ed. New York: Springer; 2015. 54. Braun V, Clarke V. Using thematic analysis in psychology. Qualitative research in psychology. 2006; 3:77–101. 55. Brennen R, Lin KY, Denehy L, Frawley HC. The effect of pelvic floor muscle interventions on pelvic floor dysfunction after gynecological cancer treatment: a systematic review. Phys Ther. 2020; 100:1357–71. https://doi.org/10.1093/ptj/pzaa081 PMID: 32367126 56. Yang EJ, Lim JY, Rah UW, Kim YB. Effect of a pelvic floor muscle training program on gynecologic can- cer survivors with pelvic floor dysfunction: a randomized controlled trial. Gynecol Oncol. 2012; 125:705–11. https://doi.org/10.1016/j.ygyno.2012.03.045 PMID: 22472463 PLOS ONE | https://doi.org/10.1371/journal.pone.0262844 January 25, 2022 18 / 20 PLOS ONE Multimodal pelvic floor physical therapy for cancer survivors suffering from pain during sexual intercourse 57. Rutledge TL, Rogers R, Lee SJ, Muller CY. A pilot randomized control trial to evaluate pelvic floor mus- cle training for urinary incontinence among gynecologic cancer survivors. Gynecol Oncol. 2014; 132:154–8. https://doi.org/10.1016/j.ygyno.2013.10.024 PMID: 24183730 58. Bober SL, Recklitis CJ, Michaud AL, Wright AA. Improvement in sexual function after ovarian cancer: effects of sexual therapy and rehabilitation after treatment for ovarian cancer. Cancer. 2018; 124:176– 82. https://doi.org/10.1002/cncr.30976 PMID: 28881456 59. Damast S, Jeffery DD, Son CH, Hasan Y, Carter J, Lindau ST, et al. Literature review of vaginal steno- sis and dilator use in radiation oncology. Pract Radiat Oncol. 2019; 9:479–91. https://doi.org/10.1016/j. prro.2019.07.001 PMID: 31302301 60. Bakker RM, Mens JW, de Groot HE, Tuijnman-Raasveld CC, Braat C, Hompus WC, et al. A nurse-led sexual rehabilitation intervention after radiotherapy for gynecological cancer. Support Care Cancer. 2017; 25:729–37. https://doi.org/10.1007/s00520-016-3453-2 PMID: 27787681 61. Carter J, Goldfarb S, Baser RE, Goldfrank DJ, Seidel B, Milli L, et al. A single-arm clinical trial investigat- ing the effectiveness of a non-hormonal, hyaluronic acid-based vaginal moisturizer in endometrial can- cer survivors. Gynecol Oncol. 2020; 158:366–74. https://doi.org/10.1016/j.ygyno.2020.05.025 PMID: 32522420 62. Carter J, Stabile C, Seidel B, Baser RE, Goldfarb S, Goldfrank DJ. Vaginal and sexual health treatment strategies within a female sexual medicine program for cancer patients and survivors. J Cancer Surviv. 2017; 11:274–83. https://doi.org/10.1007/s11764-016-0585-9 PMID: 27868156 63. 64. Li J, Huang J, Zhang J, Li Y. A home-based, nurse-led health program for postoperative patients with early-stage cervical cancer: a randomized controlled trial. Eur J Oncol Nurs. 2016; 21:174–80. https:// doi.org/10.1016/j.ejon.2015.09.009 PMID: 26482004 Zhu G, Li X, Yang S. Effect of postoperative intervention on the quality of life of patients with cervical cancer. 2016; 28:819–22. 65. Bahng AY, Dagan A, Bruner DW, Lin LL. Determination of prognostic factors for vaginal mucosal toxicity associated with intravaginal high-dose rate brachytherapy in patients with endometrial cancer. Int J Radiat Oncol Biol Phys. 2012; 82:667–73. https://doi.org/10.1016/j.ijrobp.2010.10.071 PMID: 21300451 66. Gondi V, Bentzen SM, Sklenar KL, Dunn EF, Petereit DG, Tannehill SP, et al. Severe late toxicities fol- lowing concomitant chemoradiotherapy compared to radiotherapy alone in cervical cancer: an inter-era analysis. Int J Radiat Oncol Biol Phys. 2012; 84:973–82. https://doi.org/10.1016/j.ijrobp.2012.01.064 PMID: 22898381 67. Turner JA, Franklin G, Heagerty PJ, Wu R, Egan K, Fulton-Kehoe D, et al. The association between pain and disability. Pain. 2004; 112:307–14. https://doi.org/10.1016/j.pain.2004.09.010 PMID: 15561386 68. Dworkin RH, Turk DC, Peirce-Sandner S, Baron R, Bellamy N, Burke LB, et al. Research design consid- erations for confirmatory chronic pain clinical trials: IMMPACT recommendations. Pain. 2010; 149:177– 93. https://doi.org/10.1016/j.pain.2010.02.018 PMID: 20207481 69. Goldfinger C, Pukall CF, Thibault-Gagnon S, McLean L, Chamberlain S. Effectiveness of cognitive- behavioral therapy and physical therapy for provoked vestibulodynia: a randomized pilot study. J Sex Med. 2016; 13:88–94. https://doi.org/10.1016/j.jsxm.2015.12.003 PMID: 26755091 70. Morin M, Dumoulin C, Bergeron S, Mayrand MH, Khalife´ S, Waddell G, et al. Multimodal physical ther- apy versus topical lidocaine for provoked vestibulodynia: a multicenter, randomized trial. Am J Obstet Gynecol. 2021; 224:189.e1–.e12. https://doi.org/10.1016/j.ajog.2020.08.038 PMID: 32818475 71. Dumoulin C, Cacciari LP, Hay-Smith EJC. Pelvic floor muscle training versus no treatment, or inactive control treatments, for urinary incontinence in women. Cochrane Database Syst Rev. 2018; 10: Cd005654. https://doi.org/10.1002/14651858.CD005654.pub4 PMID: 30288727 72. Gentilcore-Saulnier E, McLean L, Goldfinger C, Pukall CF, Chamberlain S. Pelvic floor muscle assess- ment outcomes in women with and without provoked vestibulodynia and the impact of a physical ther- apy program. J Sex Med. 2010; 7:1003–22. https://doi.org/10.1111/j.1743-6109.2009.01642.x PMID: 20059663 73. Thomte´ n J, Lundahl R, Stigenberg K, Linton S. Fear avoidance and pain catastrophizing among women with sexual pain. Women’s health (London, England). 2014; 10:571–81. https://doi.org/10.2217/whe. 14.51 PMID: 25482484 74. Mercier J, Morin M, Tang A, Reichetzer B, Lemieux MC, Samir K, et al. Pelvic floor muscle training: mechanisms of action for the improvement of genitourinary syndrome of menopause. Climacteric. 2020; 23:468–73. https://doi.org/10.1080/13697137.2020.1724942 PMID: 32105155 75. Goldfinger C, Pukall CF, Gentilcore-Saulnier E, McLean L, Chamberlain S. A prospective study of pelvic floor physical therapy: pain and psychosexual outcomes in provoked vestibulodynia. J Sex Med. 2009; 6:1955–68. https://doi.org/10.1111/j.1743-6109.2009.01304.x PMID: 19453890 PLOS ONE | https://doi.org/10.1371/journal.pone.0262844 January 25, 2022 19 / 20 PLOS ONE Multimodal pelvic floor physical therapy for cancer survivors suffering from pain during sexual intercourse 76. Mackenzie N. A phenomenological study of women who presented to a physiotherapy-led continence service with dyspareunia and were treated with trigger point massage. J Assoc Chart Physiother Wom- en’s Health. 2009; 105:24–39. 77. Schoonenboom J, Johnson RB. How to construct a mixed methods research design. Kolner Z Soz Sozpsychol. 2017; 69:107–31. https://doi.org/10.1007/s11577-017-0454-1 PMID: 28989188 78. Irvine FE, Clark MT, Efstathiou N, Herber OR, Howroyd F, Gratrix L, et al. The state of mixed methods research in nursing: a focused mapping review and synthesis. J Adv Nurs. 2020; 76:2798–809. https:// doi.org/10.1111/jan.14479 PMID: 32896959 79. O’Cathain A, Murphy E, Nicholl J. The quality of mixed methods studies in health services research. J Health Serv Res Policy. 2008; 13:92–8. https://doi.org/10.1258/jhsrp.2007.007074 PMID: 18416914 80. Pieterse QD, Kenter GG, Maas CP, de Kroon CD, Creutzberg CL, Trimbos JB, et al. Self-reported sex- ual, bowel and bladder function in cervical cancer patients following different treatment modalities: longi- tudinal prospective cohort study. Int J Gynecol Cancer. 2013; 23:1717–25. https://doi.org/10.1097/IGC. 0b013e3182a80a65 PMID: 24172106 81. DeSimone M, Spriggs E, Gass JS, Carson SA, Krychman ML, Dizon DS. Sexual dysfunction in female cancer survivors. Am J Clin Oncol. 2014; 37:101–6. https://doi.org/10.1097/COC.0b013e318248d89d PMID: 22643563 82. Carpenter KM, Andersen BL, Fowler JM, Maxwell GL. Sexual self schema as a moderator of sexual and psychological outcomes for gynecologic cancer survivors. Arch Sex Behav. 2009; 38:828–41. https://doi.org/10.1007/s10508-008-9349-6 PMID: 18418707 83. Andersen BL, Woods XA, Copeland LJ. Sexual self-schema and sexual morbidity among gynecologic cancer survivors. J Consult Clin Psychol. 1997; 65:221–9. https://doi.org/10.1037//0022-006x.65.2.221 PMID: 9086685 PLOS ONE | https://doi.org/10.1371/journal.pone.0262844 January 25, 2022 20 / 20 PLOS ONE
null
10.3389_fmicb.2021.711073.pdf
DATA AVAILABILITY STATEMENT The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author/s.
DATA AVAILABILITY STATEMENT The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author/s.
fmicb-12-711073 September 7, 2021 Time: 15:42 # 1 ORIGINAL RESEARCH published: 10 September 2021 doi: 10.3389/fmicb.2021.711073 Ratio of Electron Donor to Acceptor Influences Metabolic Specialization and Denitrification Dynamics in Pseudomonas aeruginosa in a Mixed Carbon Medium Irene H. Zhang1,2, Susan Mullen1†, Davide Ciccarese1, Diana Dumit1, Donald E. Martocello III1,3, Masanori Toyofuku4, Nobuhiko Nomura4, Steven Smriga1 and Andrew R. Babbin1* 1 Department of Earth, Atmospheric and Planetary Sciences, Massachusetts Institute of Technology, Cambridge, MA, United States, 2 Program in Microbiology, Massachusetts Institute of Technology, Cambridge, MA, United States, 3 Department of Marine Chemistry and Geochemistry, Woods Hole Oceanographic Institution, Woods Hole, MA, United States, 4 Faculty of Life and Environmental Sciences, Microbiology Research Center for Sustainability, University of Tsukuba, Tsukuba, Japan −) to nitrite (NO2 −), NO, N2O, and Denitrifying microbes sequentially reduce nitrate (NO3 N2 through enzymes encoded by nar, nir, nor, and nos. Some denitrifiers maintain the whole four-gene pathway, but others possess partial pathways. Partial denitrifiers may evolve through metabolic specialization whereas complete denitrifiers may adapt −) utilization. Both exist toward greater metabolic flexibility in nitrogen oxide (NOx within natural environments, but we lack an understanding of selective pressures driving the evolution toward each lifestyle. Here we investigate differences in growth rate, growth yield, denitrification dynamics, and the extent of intermediate metabolite accumulation under varying nutrient conditions between the model complete denitrifier Pseudomonas aeruginosa and a community of engineered specialists with deletions in the denitrification genes nar or nir. Our results in a mixed carbon medium indicate a growth rate vs. yield tradeoff between complete and partial denitrifiers, which varies −. We found that with total nutrient availability and ratios of organic carbon to NOx the cultures of both complete and partial denitrifiers accumulated nitrite and that the metabolic lifestyle coupled with nutrient conditions are responsible for the extent of nitrite accumulation. Keywords: Pseudomonas aeruginosa, denitrification, rate-yield tradeoff, specialization, nitrite INTRODUCTION Microbial assemblages in natural environments perform diverse biogeochemical transformations that drive global nutrient cycling and serve key ecological functions (Flemming and Wuertz, 2019). Among these, denitrification is a generally microbially mediated process that balances the nitrogen budget in terrestrial and marine ecosystems (Arrigo, 2005). Denitrifying microbes −) as terminal electron acceptors under oxygen-limiting conditions, use nitrogen oxides (NOx −), nitric oxide (NO), nitrous oxide sequentially reducing in turn nitrate (NO3 −) to nitrite (NO2 Edited by: Harold J. Schreier, University of Maryland, Baltimore County, United States Reviewed by: Julian Damashek, Utica College, United States Richard Villemur, Université du Québec, Canada *Correspondence: Andrew R. Babbin babbin@mit.edu †Present address: Susan Mullen, Department of Earth and Planetary Science, University of California, Berkeley, Berkeley, CA, United States Specialty section: This article was submitted to Microbial Physiology and Metabolism, a section of the journal Frontiers in Microbiology Received: 17 May 2021 Accepted: 09 August 2021 Published: 10 September 2021 Citation: Zhang IH, Mullen S, Ciccarese D, Dumit D, Martocello DE III, Toyofuku M, Nomura N, Smriga S and Babbin AR (2021) Ratio of Electron Donor to Acceptor Influences Metabolic Specialization and Denitrification Dynamics in Pseudomonas aeruginosa in a Mixed Carbon Medium. Front. Microbiol. 12:711073. doi: 10.3389/fmicb.2021.711073 Frontiers in Microbiology | www.frontiersin.org 1 September 2021 | Volume 12 | Article 711073 fmicb-12-711073 September 7, 2021 Time: 15:42 # 2 Zhang et al. Denitrifier Specialization and Nitrite Dynamics (N2O), and finally N2 through reductase enzymes encoded by the genes nar or nap, nir, nor, and nos, respectively (Zumft, 1997). As each step of denitrification yields free energy for the cell by coupling the reduction of nitrogen species to the oxidation of carbon, microbes theoretically harness the most energy for growth by performing the entire pathway. However, molecular surveys have revealed that many denitrifiers possess only partial denitrifying potential, whereas others contain the full suite of four genes (Zumft, 1997; Graf et al., 2014; Marchant et al., 2018). In addition, the polyphyletic distribution of denitrifying capabilities across diverse taxonomic groups and unique evolutionary history of each denitrification gene indicate the independent loss, gain, or horizontal transfer of these genes between microbes (Jones et al., 2008). Selective pressure to minimize the metabolic costs of enzyme biosynthesis, along with genome streamlining, may lead to the loss of individual denitrification genes (Mira et al., 2001; Giovannoni, 2005). Horizontal transfer may lead to the acquisition of genes that confer the ability to reduce available forms of inorganic nitrogen (Jones et al., 2008; Alvarez et al., 2011). The modularity of denitrification genes, whether as a cause or function of the fragmentation of the denitrification pathway, points to possible metabolic specialization within these communities (Lycus et al., 2018). The phenomenon of metabolic specialization in microbial communities has been well- established (Johnson et al., 2012; Wong et al., 2015; D’Souza et al., 2018; Thommes et al., 2019; Meijer et al., 2020). Specialization may manifest as members of coexisting populations diversify to fill available niches defined by nutrient availability, spatial structure, temporal variability, or other factors. Laboratory experiments with model denitrifying organisms have determined that different genes involved in the denitrification process activate under distinct environmental cues and display unique dynamics (Lycus et al., 2018). Within denitrifying ecosystems, the availability of multiple inorganic nitrogen species may lead to the diversification of microbes into populations of − consumers. NO3 Specialization can also evolve if community members construct new niches through the release of metabolic byproducts which then become substrates for the growth of other members (Kinnersley et al., 2009; Lilja and Johnson, 2019). Additionally, partial denitrifiers may have evolved unique functions beyond canonical denitrification, such as detoxification by nir and nor of toxic chemical intermediates and cellular regulation and signaling using NO (Sasaki et al., 2016; Vázquez-Torres − results in and Bäumler, 2016). As the reduction of NO3 −, NO, the production of the intermediate metabolites NO2 −- and N2O, which are released into the environment, NO3 reducing microbes may create new metabolic niches for specialist populations that perform downstream denitrification steps. Over time, a community of complementary specialists relying on substrate cross-feeding of intermediate metabolites may arise. − producers) and NO2 − consumers (NO2 The accumulation of intermediate metabolites may drive specialization by forming new ecological niches. Denitrification enzymes form dynamic, membrane-bound complexes via protein-protein interactions, which maximizes electron transfer efficiency (Borrero-de Acuña et al., 2017). Despite this tight intermediate relationship between denitrification proteins, −, accumulate in both culture- metabolites, particularly NO2 based denitrification systems (Matsubara and Zumft, 1982; Granger and Ward, 2003; Bergaust et al., 2010) and in natural environments where denitrification occurs such as marine oxygen deficient zones (ODZs) (Brandhorst, 1959; Ulloa et al., 2012). This accumulation of metabolic intermediates may indicate a spatial separation of denitrification steps through partitioning different metabolic steps into separate cells or a temporal separation in the transcription of individual genes or the activity of individual enzymes. A multitude of −) ratios result in significant NO2 factors have been shown to influence the accumulation of intermediate denitrification metabolites. − Previous studies indicate that lower organic carbon to NOx − accumulation in an (C: NOx aquatic system (Chen et al., 2017). Other possible explanations invoke competition between denitrification enzymes for co- factors, membrane space, biosynthetic building blocks, or other intracellular resources (Almeida et al., 1995; Lilja and Johnson, 2016). The involvement of transporters may contribute to the accumulation of metabolites prior to movement across a membrane. Moreover, the metabolic costs of enzyme biosynthesis create a tradeoff between maintaining and activating the full denitrification pathway and specializing in only one or several steps (Pfeiffer and Bonhoeffer, 2004; Costa et al., 2006; Wortel et al., 2018). Minimizing biosynthesis costs in multi-enzyme pathways can lead to intermediate metabolite accumulation, giving rise to multiple specialist populations even upon a single resource (Treves et al., 1998; Pfeiffer and Bonhoeffer, 2004). Therefore, this tradeoff is a key element for the evolution and coexistence of species. Complete denitrification may occur either through full − to N2 within one or several independent reduction of NO3 complete denitrifiers or as a community process between complementary partial denitrifiers. Here, we use laboratory cultures of model complete and partial denitrifiers to examine the tradeoffs involved in these two lifestyles and the effects of each lifestyle upon denitrification and growth dynamics. To eliminate the confounding factor of strain or species differences in comparing metabolic lifestyles, we use the wild- type complete denitrifier Pseudomonas aeruginosa and knockout strains with either a deletion in the gene for nitrite reductase ((cid:49)nir) or a deletion in the gene for membrane-bound nitrate reductase ((cid:49)nar), the respiratory nitrate reductase in canonical denitrification. P. aeruginosa occurs widely in marine, aquatic, and soil ecosystems and is attractive as a model organism due to its genetic tractability (Schreiber et al., 2007). We define the wild- type P. aeruginosa as a generalist in the context of denitrification, as it possesses the capability to utilize diverse oxidized nitrogen species as electron acceptors for energy. Conversely, we define the isogenic mutants as specialists since they possess only a defined subset of metabolic capabilities. We compare the growth and denitrification dynamics of these two model specialists in co-culture against their parent wild-type P. aeruginosa under varying nutrient conditions. As environmental denitrifiers utilize heterogeneous organic carbon and inorganic nitrogen sources, we test four nutrient regimes Frontiers in Microbiology | www.frontiersin.org 2 September 2021 | Volume 12 | Article 711073 fmicb-12-711073 September 7, 2021 Time: 15:42 # 3 Zhang et al. Denitrifier Specialization and Nitrite Dynamics − and NO2 −, specifically NO3 characterized by differing ratios of mixed organic carbon to −. We further use a varied NOx organic medium with many compound classes more akin to natural systems than a medium with a single carbon source. Both total nutrient availability and carbon to nitrogen oxide ratios −) have been demonstrated to impact denitrification (C: NOx processes and the metabolic division of labor within communities (Blaszczyk, 1993; Ge et al., 2012; Chen et al., 2017), so we expect these regimes to exert different selective pressures on influencing our model specialist vs. generalist communities, their growth, denitrification dynamics, and accumulation of the intermediate NO2 −. MATERIALS AND METHODS Strains and Culture Methods For the Pseudomonas aeruginosa (cid:49)nir mutant, the region from nirS to nirN was deleted, while for (cid:49)nar the narG gene was deleted (Toyofuku and Sawada, 2014). Isogenic mutants were constructed as follows: PCR primers listed in Supplementary Table 1 were used to amplify DNA fragments upstream and downstream of either narG or nirS-N with overlap extension PCR. The amplified fragments were cloned into a multicloning site in pG19II. The pG19-(cid:49)nar or pG19-(cid:49)nir plasmids were conjugated from E. coli S17-1 into wild-type P. aeruginosa PAO1 and deletion mutants were generated with allelic exchange. Deletions were confirmed with PCR and phenotypic analysis. Pseudomonas aeruginosa wild-type PAO1 and mutant strains were inoculated into 25 mL of either 100% Luria-Bertani (LB) Broth (for regimes with 100% LB) or 10% LB Broth diluted with phosphate-buffered saline (PBS; for regimes with 10% LB). LB Broth was chosen due to its varied and complex carbon sources, which may more closely resemble conditions in natural systems preferred by heterotrophic bacteria in which various carbon sources derive from complex cellular metabolites. Through the use of LB, we hoped to avoid growth dynamics that depend on and are specific to the choice of an individual carbon compound. LB Broth also contains an abundance of reduced, organic nitrogen species for − or NO2 − assimilatory anabolism, enabling supplemental NO3 to be used primarily for dissimilatory energetic pathways. We additionally performed a control experiment in M9 minimal − media supplemented with approximately 10 or 1 mM NO3 and 50 or 5 mM citrate (a C6 compound) as the sole carbon source to confirm our results are specific to mixed carbon media. M9 minimal media contains 9.35 mM NH4 − to be for nitrogen assimilation, allowing supplemental NOx used primarily for dissimilatory reduction. Additionally, M9 minimal media was supplemented with 4.1 nM biotin, 3.8 nM thiamin, 31 µM FeCl3, 6.2 µM ZnCl2, 0.76 µM CuCl2, 0.42 µM CoCl2, 1.62 µM H3BO3, and 0.081 µM MnCl2. Cultures were incubated overnight until reaching stationary phase at 37◦C with shaking within 125 mL foil-covered Erlenmeyer flasks under oxic conditions. This was used as the starting culture for inoculating into anoxic media. − or NO2 − (∼1 mM NOx − (∼10 mM NOx −, 10% LB), low NOx Media Preparation and Sampling Anoxic media was prepared in 150 mL serum bottles. In total, 50 mL of sterile 100% LB or 10% LB in PBS were amended − in serum with various concentrations of sterile NO3 bottles and purged of oxygen. Four nutrient regimes were −, 100% LB), low tested: high carbon and NOx −, − (∼1 mM NOx carbon (∼10 mM NOx −, 10% 100% LB), and low carbon and NOx LB). LB concentrations lower than 10% LB did not result in measurable culture growth after 24 h of incubation under the − regime, therefore 10% LB was chosen to low carbon and NOx represent the low carbon regime. Within each regime, four initial − ratios were tested: 10:0, 9:1, 5:5, stoichiometric NO3 and 1:9. Two replicate bottles were prepared for stoichiometric ratios 9:1, 5:5, and 1:9 under each nutrient regime, totaling eight bottles for each along with one abiotic control. For the 10:0 stoichiometric ratio, four replicates were performed, with two sets of bottles sampled on different dates for each nutrient condition. These two sets of bottles were denoted as run 1 and run 2, with the goal to assess reproducibility in growth and denitrification dynamics. Bottles were capped with a butyl rubber stopper and crimped with an aluminum ring to create an airtight seal. Each bottle was purged prior to culture inoculation with N2 gas for 2 h at 100 mL min−1, resulting in ∼80 volume turnovers. Prepared anoxic bottles were incubated overnight at 37◦C without shaking to confirm the sterility of the media prior to inoculation. −/NO2 Inoculation and sampling were performed with 10 mL syringes which were purged each time prior to insertion into bottles. Purging was performed with N2 gas three times as follows: needles were inserted into a capped, sealed empty serum bottle connected to N2 gas flowing at 1,000 mL min−1. After syringes were filled fully with N2 gas, they were removed from the bottle and N2 gas was discharged. Holding each syringe stopper down to prevent oxygen from entering the syringe, needles were reinserted into the N2 serum bottle and allowed to refill with N2 gas. This process ensured that any residual oxygen within each syringe and needle was removed and no oxygen contaminated anaerobic cultures. With the last purge, prior to insertion of the needle into media, 2 mL of N2 gas was retained within the syringe and injected into the serum bottle to maintain pressure inside the incubation bottles after sampling. The optical density (OD) of each overnight aerobic bacterial culture was measured at 600 nm on a Nanodrop OneC spectrophotometer using a 1 cm tte. For (cid:49)nir and (cid:49)nar cultures, overnight cultures were combined in a 1:1 cellular ratio within 50 mL Falcon tubes. Either this 1:1 culture mixture or the wild- type culture was added to each bottle to achieve a starting inoculum OD of 0.05. Serum bottles were then placed within a 37◦C incubator with shaking. A diagram summarizing the experimental setup, sampling scheme, and analysis methods is included as Figure 1. Sampling was performed every hour using needles purged as described above. For each sample, 2 mL of media was removed from each bottle. In total, 1 mL of media was preserved in a −, 1.5 mL microcentrifuge tube for analysis of NOx − and NO2 Frontiers in Microbiology | www.frontiersin.org 3 September 2021 | Volume 12 | Article 711073 fmicb-12-711073 September 7, 2021 Time: 15:42 # 4 Zhang et al. Denitrifier Specialization and Nitrite Dynamics while the other 1 mL was measured directly for OD. The tube was centrifuged to pellet cells and the supernatant was transferred to a second tube and frozen at −20◦C until analysis for inorganic nitrogen. Bottles were removed from the 37◦C incubator only during sampling to minimize time at room temperature, and the total duration of sampling for all bottles at each timepoint was approximately 5 min. Sampling was terminated when no NO2 − remained within any cultures. A diagnostic test was performed upon each sample by adding 10 µL of Griess reagent to cuvettes used for − remained, media within cuvettes OD measurement. If NO2 − was fully consumed, media developed a pink hue, while if NO2 − concentrations within preserved tubes remained clear. NO2 were also determined using this method, the Griess colorimetric assay (Strickland and Parsons, 1972). Absorbance was measured on a plate reader at 543 nm using a reference absorption baseline −, was determined by at 750 nm. Total NOx chemical reduction to NO with hot acidified vanadium (III) and measured via chemiluminescence with a NOx analyzer (Garside, 1982; Braman and Hendrix, 1989). The detection limit for the − method was <0.10 µM. Initial and final chemiluminescent NOx pH was taken from a replicate under the same nutrient regimes and culture conditions, with initial pH measured prior to culture inoculation and final pH measured after culture had reached stationary phase. −, or NO3 − + NO2 Data Analysis Logistic growth curves were fit to each OD time course and evaluated for goodness of fit. From this analysis, maximum growth rates, saturation points, and lag times were calculated for each replicate in each condition (Zwietering et al., 1990). Growth yields were approximated using fold-change differences between the initial inoculum of each culture and final OD at saturation. −, NO3 − include a degree of noisiness, From NOx − = NOx −, NO3 − measurements, colorimetric NO2 − data, we calculated NO3 −–NO2 −, and NO2 − concentrations as − for each timepoint. Measurements NO3 − were smoothed with a 2nd- for NOx degree polynomial Savitsky–Golay filter, which is widely used to filter time series data (Savitzky and Golay, 1964). As − concentrations, and NOx this filter calculated NO3 minimizes the influence of noise upon calculated DNRN and −, and denitrification rates. Rates of change for NOx − with time were determined by differentiating with time NO2 each curve for each regime, condition, and replicate. DNRN −/dt and denitrification rates were calculated by DNRN = –dNO3 −/dt. Maximum rates were calculated as denitrification = –dNOx DNRN and denitrification rates were found for each trial. For temporal dynamics for DNRN and denitrification, we delineate three broad categories in our data: synchronous, asynchronous, and contemporaneous. We define the activation of DNRN and denitrification as “synchronous” when the peaks for DNRN and denitrification rates are concurrent, i.e., the second derivatives of concentration with respect to time share the same sign and the maximum rates for DNRN and denitrification occur simultaneously. “Asynchronous” activation is defined as when DNRN rates and denitrification rates do not have maxima at approximately the same time, and rates do not follow the same temporal pattern of change (i.e., the second derivatives of concentration with respect to time have opposite signs). Behaviors in which the curves for DNRN and denitrification rates follow similar upward or downward trends over similar time periods, but do not peak at the same time point are termed “contemporaneous.” − − The defined max–NO2 initial)/NO3 index was accumulation as nitrite − NAI = (NO2 initial. Analyses were performed in MATLAB release R2018a. We used paired 1-sided and 2-sided t-tests as appropriate to evaluate the statistical significance of differences in growth rate, growth yield, nitrite accumulation indices, DNRN rates, and denitrification rates for each nutrient regime for generalists vis-à-vis specialists. The paired t-test was used to compare across all stoichiometric − for generalists vs. specialists under each ratios of NO3 nutrient regime. −/NO2 RESULTS − or NO2 − but not NO2 − but not NO3 − but not under NO3 To confirm that the Pseudomonas aeruginosa PAO1 (cid:49)nar mutant −, the (cid:49)nir mutant could could respire NO2 −, and the wild-type (WT) strain respire NO3 could respire both, we grew all strains axenically under anoxic conditions for 27 h in LB media supplemented with 10 mM of −. As anticipated, the (cid:49)nar mutant could only NO3 − (Supplementary grow under 10 mM NO2 −, Figure 1A). The (cid:49)nir mutant could not grow under NO2 −. WT grew under both conditions, but could grow under NO3 − did not inhibit its growth, and grew indicating that 10 mM NO2 − as it could harness the additional energy of the better given NO3 first denitrification step. In addition, the (cid:49)nar mutant reached the same optical density (OD) as the wild-type under 10 mM − did not inhibit its growth NO2 either. These results show that the (cid:49)nar mutant did indeed lose the function of the nar gene responsible for nitrate reductase but maintained the remainder of the denitrification pathway. Likewise, the (cid:49)nir mutant lost the function of the nir gene responsible for nitrite reductase but retained the function of nar and likely nor and nos. As all mutants reached an OD of ∼0.5 or higher within 27 h, and lag times for co-cultures (Supplementary Figure 2) approach those of axenic wild-type in several nutrient conditions, we do not expect growth or regulatory defects from these gene deletions to substantially impact our results. −, indicating that 10 mM NO2 To test whether the (cid:49)nar and (cid:49)nir mutants performed substrate cross-feeding in co-culture, we compared the growth of axenic (cid:49)nar and (cid:49)nir cultures against a co-culture of (cid:49)nar and (cid:49)nir ((cid:49) + (cid:49)) under anoxic conditions in LB supplemented −. We found the (cid:49) + (cid:49) co-culture with 1, 10, or 100 mM NO3 − conditions, while the axenic (cid:49)nar grew under all initial NO3 did not grow (Supplementary Figure 1B). The final co-culture OD surpassed the final OD of the axenic (cid:49)nir strain under all conditions, indicating that growth was not simply due to the (cid:49)nir strain within the co-culture but that the two strains performed metabolite cross-feeding. The (cid:49) + (cid:49) co-culture surpassed the −, so it is growth of the axenic (cid:49)nir strain under 1 mM NO3 unlikely that this result was due to the toxicity of accumulated Frontiers in Microbiology | www.frontiersin.org 4 September 2021 | Volume 12 | Article 711073 fmicb-12-711073 September 7, 2021 Time: 15:42 # 5 Zhang et al. Denitrifier Specialization and Nitrite Dynamics − concentrations. The dynamics of NOx − within the media. In addition, the axenic (cid:49)nir reached − than −, indicating its ability to tolerate higher NO2 a higher OD under 10 mM NO3 under 1 mM NO3 NO2 − and 100 mM NO3 − consumption, NO2 − accumulation, and growth over the time course of sampling for each nutrient − ratio is −/NO2 regime for two replicates for the 10:0 NO3 − regime displayed in Figure 2. Under a high carbon and NOx − and 100% LB), the OD of all cultures reached (10 mM NOx greater than 0.5, but when available carbon was reduced by a factor of 10, growth decreased to an OD of 0.2–0.4 (Figure 2), indicating that growth was reduced by the lower amount of carbon. When carbon availability was kept high with 100% LB −, growth but available nitrogen was decreased to 1 mM NOx decreased to an OD of 0.15–0.25 (Figure 2), demonstrating −. When both carbon and nitrogen growth limitation by NOx − regime availability were low in the low carbon and NOx − and 10% LB), the final OD was further depressed (1 mM NOx − regimes, to an OD compared to both low carbon and low NOx of approximately 0.1 (Figure 2), indicating growth was depressed −. The control experiments in by starvation for carbon and NOx M9 minimal media showed no decrease in OD between 50 mM citrate (300 mM carbon) and 5 mM citrate (30 mM carbon) for − concentrations (Supplementary Figure 3). This the same NO3 may be due to specifics of carbon metabolic processing or the total bioavailability of labile carbon derived from cellular material compared to citrate (Rojo, 2010; Dolan et al., 2020). For each culture, we calculated the maximum growth rate and the approximate growth yield, represented in Figure 3. We approximate growth yields by taking the fold-change between the cell density of the starting culture and the maximum cell − and low density based on OD. Under high carbon and NOx carbon regimes, in which nitrogen was high, (cid:49) + (cid:49) co-cultures −: n = 8, achieved a higher growth yield (high carbon and NOx paired 1-sided t-test, p = 0.0002; low carbon: n = 8, paired 1- sided t-test, p = 0.002), while WT exhibited a higher maximum −: n = 8, paired 1-sided growth rate (high carbon and NOx low carbon: n = 8, paired 1-sided t-test, t-test, p = 0.001; p = 6.5 × 10−7) (Figure 3). However, when NOx − was low at 1 mM, the relationship between growth rate and growth yield for the (cid:49) + (cid:49) co-cultures compared to WT changed. Under − regimes, the differences between growth rate and low NOx growth yield for each culture were non-significant (p > 0.01 as determined by a 1-sided t-test). Under low carbon and − regimes, WT had higher maximum growth rates (n = 8, NOx paired 1-sided t-test, p = 0.009) but growth yields were not significantly different (n = 8, paired 1-sided t-test, p = 0.08). For −/NO2 − these statistical tests, we used 10:0, 9:1, and 5:5 NO3 stoichiometric ratios and excluded the 1:9 stoichiometric ratio as − availability could not support substantial growth the low NO3 − producer within the co-culture. Growth of the obligate NO2 effects for the 1:9 ratio more likely result from the lower initial − in the (cid:49) + (cid:49) inoculum sizes of cells capable of utilizing NO2 co-cultures compared to the WT, and little potential for cross- feeding exists between the two mutants, particularly in the 1 mM − regimes. Previous studies indicate the precise context in NOx terms of the type of carbon compound is key (Rojo, 2010; Dolan −) et al., 2020), and our control experiments with a single carbon source also suggest a possible growth yield vs. growth rate tradeoff between WT and (cid:49) + (cid:49) under all nutrient regimes, warranting follow-up study to investigate species specific responses to carbon affinity (Supplementary Figure 4A). further − to NO2 − to NO2 reduced NO2 − and NO2 − and the loss of NO2 From total nitrogen oxyanion (NOx concentrations − from measured for each time point, we found a loss of NOx both the WT and (cid:49) + (cid:49) co-cultures (Figure 2). For the − (cid:49) + (cid:49) co-cultures, this demonstrates that the obligate NO2 −, and the obligate producer ((cid:49)nir) reduced NO3 −. Distinct − consumer ((cid:49)nar) NO2 −) consumption temporal dynamics of nitrogen oxyanion (NOx distinguish (cid:49) + (cid:49) co-cultures and axenic WT. In addition − − over time, we also measured NO2 to measuring total NOx −, calculated over time and, from the curves of NOx − over time. Assuming that both mutants could respire NO3 NO and N2O, we focused our analysis on the reduction of −, which we respectively NO3 differentiate as DNRN and denitrification. Although DNRN canonically represents the initial reaction of the denitrification − remains pathway, fixed nitrogen is not lost as the resulting NO2 − to gaseous forms bioavailable. However, the reduction of NO2 of nitrogen results in the loss of bioavailable nitrogen from the system, and this step is considered the defining reaction of denitrification. We found the rates of DNRN and denitrification in all conditions, as shown in Figure 4. DNRN rates did not vary between WT or co-culture in any condition (n = 40, paired 2-sided t-test, p = 0.3) (Figure 4A). Notably, DNRN rates − regimes compared to were 10-fold higher in the 10 mM NOx − conditions − regimes. At higher initial NO3 the 1 mM NOx (10:0, 9:1), DNRN rates were highest, whereas DNRN rates −, condition. This reveals were lowest under the 1:9 NO3 that the major determinant of DNRN rate is the amount of − available and that both WT and the (cid:49) + (cid:49) co-culture NO3 − with equal speed. In contrast, denitrification reduce NO3 statistically indistinguishable between cultures rates were growing in 100% LB (n = 20, paired 2-sided t-test, p = 0.2) but denitrification rates were lower for (cid:49) + (cid:49) co-cultures compared to wild-type in 10% LB (n = 20, paired 1-sided t-test, p = 0.006) (Figure 4B). This difference was observed for all stoichiometric NO3 − was reduced quantitatively to NO2 − accumulation differs between − accumulation specialists and generalists, we compared the NO2 index (NAI) for each nutrient condition. NAI = 1 indicates − before NO2 − all NO3 reduction commenced whereas NAI = 0 reflects no transient − accumulation. For the high carbon and NOx − regime, NO2 − accumulates to a moderate extent in the 10:0 ratio NO2 (NAI = 0–0.5), and (cid:49) + (cid:49) co-cultures and WT do not differ significantly from each other (n = 8, paired 2-sided t-test, p = 0.9). The highest NAI values occurred in the (cid:49) + (cid:49) co-cultures under low carbon, reaching almost 100% of the initial nitrogen loading, significantly higher than WT under the same conditions (n = 8, paired 1-sided t-test, p = 0.002) (Figure 5). For the −, high carbon regime, WT cultures generally reached low NOx higher NAI than (cid:49) + (cid:49) co-cultures (n = 8, paired 1-sided t-test, p = 0.01). In the single carbon control, WT cultures did − ratios in LB. To examine whether the NO2 −/NO2 −/NO2 Frontiers in Microbiology | www.frontiersin.org 5 September 2021 | Volume 12 | Article 711073 fmicb-12-711073 September 7, 2021 Time: 15:42 # 6 Zhang et al. Denitrifier Specialization and Nitrite Dynamics FIGURE 1 | Diagram of experimental setup, sampling scheme, and analysis methods. P. aeruginosa strains for inoculation are depicted in orange (wild-type), cyan ((cid:49)nar), and red ((cid:49)nir). Nutrient regimes are shown in large serum bottles, and stoichiometric ratios for each nutrient regime are shown in smaller serum bottles. Number of replicates for each stoichiometric ratio within each nutrient regime is indicated next to smaller serum bottles. The sampling scheme for each replicate is depicted within the dashed rectangle, along with the analyses performed on each sample. to − in any condition whereas not accumulate measurable NO2 (cid:49) + (cid:49) co-cultures consistently accumulated this intermediate (Supplementary Figure 4B). However, high variability was observed in NAI between replicates started from different inocula for the 10:0 ratio, indicating additional controls on NAI beyond nutrient condition or metabolic lifestyle. regime according either nutrient In addition to maximum rates and NO2 − changes, we explored the temporal dynamics of DNRN and denitrification during growth. We did not find clear partitioning of or synchronicity specialist vs. generalist cultures (Supplementary Figure 5). The observed patterns of synchronicity follow the extent of − accumulation. For cultures in LB exhibiting asynchronous NO2 − accumulates. DNRN and denitrification dynamics, more NO2 exhibiting synchronous DNRN and Conversely, cultures −, and contemporaneous denitrification accumulate less NO2 − but not to levels as high cultures accumulate some NO2 as asynchronous cultures. In all conditions, DNRN always proceeds prior to denitrification. These patterns suggest that synchronicity, rather than maximal DNRN or denitrification rates, drives the accumulation of NO2 In general, (cid:49) + (cid:49) co-cultures required a longer time to begin growth, perform DNRN and denitrification, and reach the stationary phase. Lag times, defined as the period prior to cell division and exponential growth, are generally higher for (cid:49) + (cid:49) co-cultures vs. the WT PAO1 across all regimes (Supplementary Figure 2). The onset of DNRN and denitrification in (cid:49) + (cid:49) was generally slower than in WT, possibly reflecting the lower initial density of denitrification- capable cells ((cid:49)nar). The duration of time from inoculation to − was consistently longer in (cid:49) + (cid:49) the total consumption of NOx −. than in WT (Supplementary Figures 6–9), reflective of the same growth rate v. yield tradeoff between WT and (cid:49) + (cid:49) cultures. OD curves indicate that the (cid:49) + (cid:49) co-cultures require longer to reach saturation, which is consistent with the observation that logarithmic growth corresponds with the period of DNRN and denitrification activity. DISCUSSION The results from this study, examining a high carbon and − regime for the axenic Pseudomonas aeruginosa PAO1 NOx wild-type (WT) generalist compared to co-cultured DNRN and denitrification ((cid:49) + (cid:49)) specialists, are consistent with a growth rate vs. growth yield tradeoff. Our results in both LB and M9 media offer evidence for this tradeoff. Growth rate and growth yield, a proxy for net growth efficiency, are two fundamental traits of microbes that influence community function, evolution, and species coexistence (Lipson, 2015). This rate vs. yield tradeoff has been found within laboratory evolution experiments using diverse organisms such as E. coli (Novak et al., 2006), Lactobacillus lactis (Bachmann et al., 2013), and yeasts (Weusthuis et al., 1994), as well as in natural microbial communities (Sorokin et al., 2003; Lipson et al., 2009). However, to our knowledge, this tradeoff has not previously been experimentally demonstrated using model generalists and specialists within the same strain in the context of varying nutrient regimes. Previous numerical approaches have shown that the rate vs. yield tradeoff is not universal, but depends upon environmental conditions (Lipson et al., 2009; Beardmore et al., 2011). We Frontiers in Microbiology | www.frontiersin.org 6 September 2021 | Volume 12 | Article 711073 fmicb-12-711073 September 7, 2021 Time: 15:42 # 7 Zhang et al. Denitrifier Specialization and Nitrite Dynamics −/NO2 FIGURE 2 | Time series data under all LB nutrient regimes for 10:0 NO3 − (green), NO2 − and (G,H) low carbon, and NOx carbon, (E,F) low NOx wild-type culture whereas right hand panels depict the mutant co-culture. NOx variability introduced by chemiluminescent measurements and variations in initial NOx with error bars indicating ranges. Curves for NOx Selected plots are presented for brevity; analogous plots for other replicates and initial NO3 − and NO2 −. NOx − ratios for the first run. The regimes are: (A,B) high carbon and NOx − (purple), and bacterial growth (black) are shown. Left hand panels correspond to the − data are normalized to the initial NOx − concentration during plotting to reduce − (C,D) low − loading. Data points are the means of two biological replicates per condition, − were smoothed with a Savitsky-Golay filter, while OD600 curves were fit to a logistic growth model. −/NO2 − ratios can be found in Supplementary Figures 6–9. Frontiers in Microbiology | www.frontiersin.org 7 September 2021 | Volume 12 | Article 711073 fmicb-12-711073 September 7, 2021 Time: 15:42 # 8 Zhang et al. Denitrifier Specialization and Nitrite Dynamics FIGURE 3 | Growth rate vs. growth yield for generalists and specialists in LB. Results for three stoichiometric NO3 dark purple; 9:1, red; 5:5, green) are plotted together for each nutrient regime as follows: (A) high carbon and NOx carbon and NOx WT and (cid:49) + (cid:49), n = 4 for the 10:0 NO3 depicted ratios, omitting 1:9. −/NO2 −. Maximum growth rates (µmax) vs. growth yields (Cmax/C0) are plotted separately for wild-type (WT) and (cid:49)nar + (cid:49)nir co-culture ((cid:49) + (cid:49)). For both − ratio and n = 2 for the 9:1 and 5:5 ratio for each nutrient regime, and n = 8 for each nutrient regime including all − ratios (10:0 run 1, lavender; 10:0 run 2, −/NO2 −, (B) low carbon, (C) low NOx −, and (D) low Frontiers in Microbiology | www.frontiersin.org 8 September 2021 | Volume 12 | Article 711073 fmicb-12-711073 September 7, 2021 Time: 15:42 # 9 Zhang et al. Denitrifier Specialization and Nitrite Dynamics −/NO2 −/NO2 − ratio (10:0 run 1, lavender; 10:0 run 2, dark purple; 9:1, red; 5:5, green; 1:9, blue). For both WT and (cid:49) + (cid:49), n = 4 for the 10:0 − ratio and n = 2 for the 9:1 and 5:5 ratio for each nutrient regime, and n = 10 for each nutrient regime including all depicted ratios. There is no significant FIGURE 4 | Maximum DNRN and denitrification rates for WT and (cid:49) + (cid:49) cultures in LB. Maximum DNRN rates are plotted for each culture under each nutrient condition and colored by NO3 NO3 difference in DNRN rates when comparing between WT or (cid:49) + (cid:49) for any treatment. DNRN rates are 10-fold higher under 10-fold higher NOx influenced by carbon availability. (B) Maximum denitrification rates for cultures and conditions, as in panel (A). For 10% LB regimes, denitrification rates decrease in (cid:49) + (cid:49) cultures, while there is no difference between denitrification rates when comparing cultures for 100% LB regimes. NO3 affect denitrification rates. − ratios did not significantly −, but are not −/NO2 find this tradeoff to be consistent for nutrient replete and − (electron low carbon conditions. When the amount of NOx acceptor) is decreased, some (cid:49) + (cid:49) co-cultures reach higher growth rates compared to WT and some WT cultures reach higher growth yields than (cid:49) + (cid:49) co-cultures (Figure 3). Metabolic savings alone are unlikely to explain this, so it may be useful to interpret this result through the lens of ecological interactions within each culture. Within axenic WT, each cell − and carbon. However, within a competes with others for NOx − is reduced (cid:49) + (cid:49) co-culture, competitive pressure for NOx − while as only a portion of the population can use NO3 −. In addition, obligate the other portion can only use NO2 − consumers engage in a commensal relationship with NO2 − producers. Previous studies have demonstrated obligate NO2 Frontiers in Microbiology | www.frontiersin.org 9 September 2021 | Volume 12 | Article 711073 fmicb-12-711073 September 7, 2021 Time: 15:42 # 10 Zhang et al. Denitrifier Specialization and Nitrite Dynamics − accumulation index (NAI) for generalist and specialist cultures in LB. Stoichiometric ratios are plotted separately for (cid:49) + (cid:49) co-cultures and WT for − ratio and n = 2 for FIGURE 5 | NO2 each nutrient regime (10:0 run 1, lavender; 10:0 run 2, dark purple; 9:1, red; 5:5, green). For both WT and (cid:49) + (cid:49), n = 4 for the 10:0 NO3 the 9:1 and 5:5 ratio for each nutrient regime, and n = 8 for each nutrient regime including all depicted ratios, omitting 1:9. NAI relates to synchronicity, with synchronous cultures exhibiting low NAI and asynchronous cultures exhibiting high NAI. (A) The NAI is low to moderate for high carbon and NOx is generally higher for (cid:49) + (cid:49) for low carbon regimes, but variable between replicates; (C) NAI is generally higher for WT for low NOx high values for low carbon and NOx − but is variable between replicates. − regimes; (D) NAI can reach − regimes; (B) NAI −/NO2 the differential effects of competition or commensalism on the spatial arrangements and communal behaviors of interacting denitrifiers (Hibbing et al., 2010; Faust and Raes, 2012; Lilja and − regime, Johnson, 2019; Ciccarese et al., 2020). Under a low NOx − may complicate the terms of the rate competition for NOx vs. yield tradeoff. Additionally, in nutrient-depleted conditions, low cell density may impede efficient substrate exchange between separate specialist populations. Further experiments are required to pinpoint the conditions under which the rate vs. yield tradeoff changes and the mechanisms, ecological or physiological, underlying this change. The presence of metabolic division of labor generally − correlates with increased potential accumulation of NO2 and other metabolic intermediates in natural environments. − may also occur in complete However, accumulation of NO2 denitrifiers under and displays high certain conditions, variability even for cultures with the same genetic content growing under the same nutrient conditions. The accumulation − also changes for cultures growing in mixed vs. of NO2 single carbon sources, as complexity in carbon resources will likely modify thermodynamic and kinetic stimuli (Rojo, interactions in the 2010). Over the course of the cultures, co-culture may lead to varied growth dynamics of each mutant. These specific dynamics, while not captured in our scheme, may clarify this variability between experimental cultures and serves as a basis future experimental for work. Additionally, further studies on the regulation of the denitrification pathway, the link between denitrification and carbon metabolism in generalist and specialist species, and the individual growth dynamics of each specialist within co-cultures are required to determine the exact drivers of intermediate accumulation and its impacts on denitrifying community behavior. The accumulation of NO2 − does not track with differences in maximal DNRN or denitrification rates between cultures − producers do not or nutrient conditions. Obligate NO2 − conversion compared to wild- exhibit lower rates of NO3 type generalists under any nutrient condition (Figure 4A). In − consumers maintain decreased rates contrast, obligate NO2 − reduction compared to generalists only when carbon of NO2 is low (Figure 4B). This indicates different sensitivities or regulations of nar and nir toward nutrient availability and − between type. However, differential accumulation of NO2 − conditions, which WT and co-cultures occurred in low NOx displayed little difference in DNRN or denitrification rates. This may be explained by the temporal dynamics of DNRN and denitrification. Cultures exhibiting low NAI are more synchronous than cultures exhibiting high NAI (Supplementary Figure 5). For the WT, this synchronicity points to simultaneous activation of both portions of the denitrification pathway, but for the co-culture synchronous DNRN and denitrification indicates simultaneous metabolism by both specialists. Asynchronous behavior in the WT reveals a temporal delay between DNRN and denitrification, possibly due to regulatory differences in gene expression between nar and nir (Körner and Zumft, 1989; Schreiber et al., 2007). In the (cid:49) + (cid:49) co-culture, asynchrony indicates population succession, with the obligate − − producer growing first followed later by the obligate NO2 NO2 consumer. The only conditions under which specialists exhibit −. generally more synchronous behavior than WT are low NOx The ability of both populations to grow non-exclusively points to a commensal, rather than competitive, interaction for a scarce Frontiers in Microbiology | www.frontiersin.org 10 September 2021 | Volume 12 | Article 711073 fmicb-12-711073 September 7, 2021 Time: 15:42 # 11 Zhang et al. Denitrifier Specialization and Nitrite Dynamics nutrient. This commensal interaction correlates with the only −) in which NAI is lower for the (cid:49) + (cid:49) regime (low NOx co-culture than for the WT (Figure 5). Broadly, these results suggest that the temporal dynamics rather than the maximal rates of the individual steps of denitrification may drive the extent of intermediate accumulation. Further work on individual dynamics of growth for each mutant and timing of nar and nir transcription may shed light upon the mechanisms underlying these behaviors. Pseudomonas aeruginosa, along with many other denitrifying and non-denitrifying organisms, possesses an additional nitrate reductase system, periplasmic nitrate reductase enzyme Nap encoded by the nap gene (Alst et al., 2009). As nap was not deleted − reduction. in our system, this may potentially influence NO3 However, since Nap cannot generate a proton motive force for ATP synthesis and growth, it is unlikely that Nap played a large role in the growth dynamics observed. Additionally, nap, along with fermentative processes, has been shown to activate mostly in the stationary phase, while nar is expressed during active growth (Alst et al., 2009; Schiessl et al., 2019). As the dynamics we observe are based upon pre-stationary phase growth and metabolism under anoxic conditions, we do not expect either of these processes to be a substantial influence. However, the regulation of the denitrification pathway is complex, so the influence of nap cannot be ruled out and requires further investigation. The role of nap in denitrification dynamics, denitrifier evolution, and metabolic niche differentiation is an exciting complementary research avenue. Using engineered strains of Pseudomonas aeruginosa PAO1, we compare the behavior of complete denitrifiers against a community in which the denitrification pathway has been − producers and consumers. partitioned between obligate NO2 Our results indicate a growth rate vs. growth yield tradeoff between complete denitrifiers, or generalists, and partial − denitrifiers, or specialists under nutrient replete and high NOx conditions. While few studies have surveyed complete vs. partial denitrifiers across various environments, several studies on denitrifying communities reveal a high prevalence of partial denitrifiers in soils and wetlands (Roco et al., 2017). A study of metagenome-assembled genomes from various environments also discovered a higher ratio of complete:partial denitrifiers in built environments and in marine and brackish systems (Hester et al., 2019). Relatively richer nutrient conditions and spatial segregation in soils and wetlands may select for metabolic specialization, while more nutrient-limited environments may select for complete denitrifiers. However, more work is required to link the prevalence of complete vs. partial denitrifiers across environments and their nutrient contexts. We find that nutrient availability, relative amounts of carbon −, and the composition of metabolic lifestyles within to NOx REFERENCES Almeida, J. S., Reis, M. A. M., and Carrondo, M. J. T. (1995). Competition between nitrate and nitrite reduction in denitrification by Pseudomonas fluorescens. Biotechnol. Bioeng. 46, 476–484. a denitrifying system play key roles in driving the rate vs. − consumption, and the yield tradeoff, the dynamics of NOx accumulation of chemical intermediates. Our data provide evidence of the differences in the growth and denitrification behavior between a community of specialists and generalists, but variability between replicates in relation to the extent of − accumulation indicates a complexity in the denitrification NO2 pathway that remains to be resolved. Denitrification regulation, bacterial specific thermodynamics driving complete vs. partial denitrification, and the ecological and chemical interactions among denitrifying microbes are likely to be fruitful avenues of future investigation. carbon and nitrogen metabolism, the DATA AVAILABILITY STATEMENT The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author/s. AUTHOR CONTRIBUTIONS IZ, SM, DC, SS, and AB analyzed the data and wrote the manuscript. SM conducted the experiments. IZ, SM, DD, and DM analyzed nitrogen samples. MT and NN provided knockout strains for this research. AB designed the study and supervised the project. All authors contributed to the article and approved the submitted version. FUNDING Funding for this work was provided by Simons Foundation award 622065 and an MIT Environmental Solutions Initiative seed grant to AB. Additional support was received by the MIT Ferry Fund. ACKNOWLEDGMENTS We would like to thank Sarah Schwartz for her preliminary experimental the directions for this study. results which helped define have SUPPLEMENTARY MATERIAL for this article can be found at: https://www.frontiersin.org/articles/10.3389/fmicb. The Supplementary Material online 2021.711073/full#supplementary-material Alst, V. E. N., Sherrill, L. A., Iglewski, B. H., and Haidaris, C. G. (2009). Compensatory periplasmic nitrate reductase activity supports anaerobic growth of Pseudomonas aeruginosa PAO1 in the absence of membrane nitrate doi: 10.1139/ w09-065 J. Microbiol. 55, 1133–1144. reductase. Can. Frontiers in Microbiology | www.frontiersin.org 11 September 2021 | Volume 12 | Article 711073 fmicb-12-711073 September 7, 2021 Time: 15:42 # 12 Zhang et al. Denitrifier Specialization and Nitrite Dynamics Alvarez, L., Bricio, C., Gómez, M. J., and Berenguer, J. (2011). Lateral transfer of the denitrification pathway genes among Thermus thermophilus strains. Appl. Environ. Microbiol. 77, 1352–1358. doi: 10.1128/aem.02048-10 Hester, E. R., Jetten, M. S. M., Welte, C. U., and Lücker, S. (2019). Metabolic overlap in environmentally diverse microbial communities. Front. Genet. 10:989. doi: 10.3389/fgene.2019.00989 Arrigo, K. R. (2005). Marine microorganisms and global nutrient cycles. Nature 437, 349–355. doi: 10.1038/nature04159 Bachmann, H., Fischlechner, M., Rabbers, I., Barfa, N., Branco, F., Santos, D., et al. (2013). Availability of public goods shapes the evolution of competing metabolic strategies. Proc. Natl. Acad. Sci. U.S.A. 110, 14302–14307. doi: 10.1073/pnas. 1308523110 Beardmore, R. E., Gudelj, I., Lipson, D. A., and Hurst, L. D. (2011). Metabolic trade- offs and the maintenance of the fittest and the flattest. Nature 472, 342–346. doi: 10.1038/nature09905 Bergaust, L., Mao, Y., Bakken, L. R., and Frostegård, Å (2010). Denitrification response patterns during respiration and posttranscriptional effects of suboptimal pH on Nitrogen oxide reductase in Paracoccus denitrificans. Appl. Environ. Microbiol. 76, 6387–6396. doi: 10.1128/aem.00608-10 transition to anoxic the Blaszczyk, M. (1993). Effect of medium composition on the denitrification of nitrate by Paracoccus denitrificans. Appl. Environ. Microbiol. 59, 3951–3953. doi: 10.1128/aem.59.11.3951-3953.1993 Hibbing, M. E., Fuqua, C., Parsek, M. R., and Peterson, S. B. (2010). Bacterial competition: surviving and thriving in the microbial jungle. Nat. Rev. Microbiol. 8, 15–25. doi: 10.1038/nrmicro2259 Johnson, D. R., Goldschmidt, F., Lilja, E. E., and Ackermann, M. (2012). Metabolic specialization and the assembly of microbial communities. ISME J. 6, 1985– 1991. doi: 10.1038/ismej.2012.46 Jones, C. M., Stres, B., Rosenquist, M., and Hallin, S. (2008). Phylogenetic analysis of nitrite, nitric oxide, and nitrous oxide respiratory enzymes reveal a complex evolutionary history for denitrification. Mol. Biol. Evol. 25, 1955–1966. doi: 10.1093/molbev/msn146 Kinnersley, M. A., Holben, W. E., and Rosenzweig, F. (2009). E unibus plurum: genomic analysis of an experimentally evolved polymorphism in Escherichia coli. PLoS Genet. 5:e1000713. doi: 10.1371/journal.pgen.1000713 Körner, H., and Zumft, W. G. (1989). Expression of denitrification enzymes in response to the dissolved oxygen level and respiratory substrate in continuous culture of Pseudomonas stutzeri. Appl. Environ. Microbiol. 55, 1670–1676. doi: 10.1128/aem.55.7.1670-1676.1989 Borrero-de Acuña, J. M., Timmis, K. N., Jahn, M., and Jahn, D. (2017). Protein complex formation during denitrification by Pseudomonas aeruginosa. Microb. Biotechnol. 10, 1523–1534. Lilja, E. E., and Johnson, D. R. (2016). Segregating metabolic processes into different microbial cells accelerates the consumption of inhibitory substrates. ISME J. 10, 1568–1578. doi: 10.1038/ismej.2015.243 Braman, R. S., and Hendrix, S. A. (1989). Nanogram nitrite and nitrate determination in environmental and biological materials by vanadium(III) reduction with chemiluminescence detection. Anal. Chem. 61, 2715–2718. doi: 10.1021/ac00199a007 Brandhorst, W. (1959). Nitrification and denitrification in the eastern tropical North Pacific. ICES J. Mar. Sci. 25, 3–20. doi: 10.1093/icesjms/25.1.3 Chen, R., Deng, M., He, X., and Hou, J. (2017). Enhancing nitrate removal from freshwater pond by regulating carbon/nitrogen ratio. Front. Microbiol. 8:1712. doi: 10.3389/fmicb.2017.01712 Ciccarese, D., Zuidema, A., Merlo, V., and Johnson, D. R. (2020). Interaction- dependent effects of surface structure on microbial spatial self-organization. Philos. Trans. R. Soc. Lond. B Biol. Sci. 375:20190246. doi: 10.1098/rstb.2019. 0246 Costa, E., Pérez, J., and Kreft, J. U. (2006). Why is metabolic labour divided doi: 10.1016/j.tim.2006. in nitrification? Trends Microbiol. 14, 213–219. 03.006 D’Souza, G., Shitut, S., Preussger, D., Yousif, G., Waschina, S., and Kost, C. (2018). Ecology and evolution of metabolic cross-feeding interactions in bacteria. Nat. Prod. Rep. 35, 455–488. doi: 10.1039/c8np00009c Dolan, S. K., Kohlstedt, M., Trigg, S., Vallejo Ramirez, P., Kaminski, C. F., Wittmann, C., et al. (2020). Contextual flexibility in Pseudomonas aeruginosa central carbon metabolism during growth in single carbon sources. mBio 11, e2684–e2719. Faust, K., and Raes, J. (2012). Microbial interactions: from networks to models. Nat. Rev. Microbiol. 10, 538–550. doi: 10.1038/nrmicro2832 Flemming, H. C., and Wuertz, S. (2019). Bacteria and archaea on Earth and their abundance in biofilms. Nat. Rev. Microbiol. 17, 247–260. doi: 10.1038/s41579- 019-0158-9 Garside, C. (1982). A chemiluminescent technique for the determination of nanomolar concentrations of nitrate and nitrite in seawater. Mar. Chem. 11, 159–167. doi: 10.1016/0304-4203(82)90039-1 Ge, S., Peng, Y., Wang, S., Lu, C., Cao, X., and Zhu, Y. (2012). Nitrite accumulation under constant temperature in anoxic denitrification process: the effects of carbon sources and COD/NO3-N. Bioresour. Technol. 114, 137–143. doi: 10. 1016/j.biortech.2012.03.016 Giovannoni, S. J. (2005). Genome streamlining in a cosmopolitan oceanic bacterium. Science 309, 1242–1245. doi: 10.1126/science.1114057 Graf, D. R. H., Jones, C. M., and Hallin, S. (2014). Intergenomic comparisons highlight modularity of the denitrification pathway and underpin the importance of community structure for N2O emissions. PLoS One 9:e114118. doi: 10.1371/journal.pone.0114118 Lilja, E. E., and Johnson, D. R. (2019). Substrate cross-feeding affects the speed and trajectory of molecular evolution within a synthetic microbial assemblage. BMC Evol. Biol. 19:129. doi: 10.1186/s12862-019-1458-4 Lipson, D. A. (2015). The complex relationship between microbial growth rate and yield and its implications for ecosystem processes. Front. Microbiol. 6:615. doi: 10.3389/fmicb.2015.00615 Lipson, D. A., Monson, R. K., Schmidt, S. K., and Weintraub, M. N. (2009). The trade-off between growth rate and yield in microbial communities and the consequences for under-snow soil respiration in a high elevation coniferous forest. Biogeochemistry 95, 23–35. doi: 10.1007/s10533-008-9252-1 Lycus, P., Soriano-Laguna, M. J., Kjos, M., Richardson, D. J., Gates, A. J., Milligan, D. A., et al. (2018). A bet-hedging strategy for denitrifying bacteria curtails their release of N2O. Proc. Natl. Acad. Sci.U.S.A. 115, 11820–11825. doi: 10.1073/ pnas.1805000115 Marchant, H. K., Tegetmeyer, H. E., Ahmerkamp, S., Holtappels, M., Lavik, G., Graf, J., et al. (2018). Metabolic specialization of denitrifiers in permeable sediments controls N2O emissions. Environ. Microbiol. 12, 4486–4502. doi: 10.1111/1462-2920.14385 Matsubara, T., and Zumft, W. G. (1982). Identification of a copper protein as part of the nitrous oxide-reducing system in nitrite-respiring (denitrifying) pseudomonads. Arch. Microbiol. 132, 322–328. doi: 10.1007/bf00413383 Meijer, J., van Dijk, B., and Hogeweg, P. (2020). Contingent evolution of alternative metabolic network topologies determines whether cross-feeding evolves. Commun. Biol. 3:401. Mira, A., Ochman, H., and Moran, N. A. (2001). Deletional bias and the evolution of bacterial genomes. Trends Genet. 17, 589–596. doi: 10.1016/s0168-9525(01) 02447-7 Novak, M., Pfeiffer, T., Lenski, R. E., Sauer, U., and Bonhoeffer, S. (2006). Experimental tests for an evolutionary trade-off between growth rate and yield in E. coli. Am. Nat. 168, 242–251. doi: 10.2307/3844729 Pfeiffer, T., and Bonhoeffer, S. (2004). Evolution of cross-feeding in microbial populations. Am. Nat. 163, E126–E135. Roco, C. A., Bergaust, L. L., Bakken, L. R., Yavitt, J. B., and Shapleigh, J. P. (2017). Modularity of nitrogen-oxide reducing soil bacteria: linking phenotype to genotype. Environ. Microbiol. 19, 2507–2519. doi: 10.1111/1462-2920.13250 Rojo, F. (2010). Carbon catabolite repression in Pseudomonas: optimizing metabolic versatility and interactions with the environment. FEMS Microbiol. Rev. 34, 658–684. doi: 10.1111/j.1574-6976.2010.00218.x Sasaki, Y., Oguchi, H., Kobayashi, T., Kusama, S., Sugiura, R., Moriya, K., et al. (2016). Nitrogen oxide cycle regulates nitric oxide levels and bacterial cell signaling. Sci. Rep. 6:22038. Granger, J., and Ward, B. B. (2003). Accumulation of nitrogen oxides in copper- limited cultures of denitrifying bacteria. Limnol. Oceanogr. 48, 313–318. doi: 10.4319/lo.2003.48.1.0313 Savitzky, A., and Golay, M. J. E. (1964). Smoothing and differentiation of data by simplified least squares procedures. Anal. Chem. 36, 1627–1639. doi: 10.1021/ ac60214a047 Frontiers in Microbiology | www.frontiersin.org 12 September 2021 | Volume 12 | Article 711073 fmicb-12-711073 September 7, 2021 Time: 15:42 # 13 Zhang et al. Denitrifier Specialization and Nitrite Dynamics Schiessl, K. T., Hu, F., Jo, J., Nazia, S. Z., Wang, B., Price-Whelan, A., et al. (2019). Phenazine production promotes antibiotic tolerance and metabolic heterogeneity in Pseudomonas aeruginosa biofilms. Nat. Commun. 10:762. Schreiber, K., Krieger, R., Benkert, B., Eschbach, M., Arai, H., Schobert, M., et al. (2007). The anaerobic regulatory network required for Pseudomonas aeruginosa nitrate respiration. J. Bacteriol. 189, 4310–4314. doi: 10.1128/jb.00240-07 Sorokin, D. Y., Banciu, H., van Loosdrecht, M., and Kuenen, J. G. (2003). Growth physiology and competitive interaction of obligately chemolithoautotrophic, haloalkaliphilic, sulfur-oxidizing bacteria from soda lakes. Extremophiles 7, 195–203. doi: 10.1007/s00792-002-0313-4 Strickland, J. D. H., and Parsons, T. R. (1972). A Practical Handbook of Seawater Analysis. Ottawa, ON: Fisheries Research Board Of Canada. Thommes, M., Wang, T., Zhao, Q., Paschalidis, I. C., and Segrè, D. (2019). Designing metabolic division of labor in microbial communities. mSystems 4, e263–e218. Toyofuku, M., and Sawada, I. (2014). Membrane vesicle formation is associated with pyocin production under denitrifying conditions in Pseudomonas aeruginosa PAO1. Environ. Microbiol. 16, 2927–2938. doi: 10.1111/1462-2920. 12260 Treves, D. S., Manning, S., and Adams, J. (1998). Repeated evolution of an acetate- crossfeeding polymorphism in long-term populations of Escherichia coli. Mol. Biol. Evol. 15, 789–797. doi: 10.1093/oxfordjournals.molbev.a025984 Ulloa, O., Canfield, D. E., DeLong, E. F., Letelier, R. M., and Stewart, F. J. (2012). Microbial oceanography of anoxic oxygen minimum zones. Proc. Natl. Acad. Sci. U.S.A. 109, 15996–16003. doi: 10.1073/pnas.1205009109 Vázquez-Torres, A., and Bäumler, A. (2016). Nitrate, nitrite and nitric oxide reductases: from the last universal common ancestor to modern bacterial pathogens. Curr. Opin. Microbiol. 29, 1–8. doi: 10.1016/j.mib.2015.09.002 Weusthuis, R. A., Pronk, J. T., van den Broek, P. J., and van Dijken, J. P. (1994). Chemostat cultivation as a tool for studies on sugar transport in yeasts. Microbiol. Mol. Biol. Rev. 58, 616–630. doi: 10.1128/mmbr.58.4.616-630.1994 Wong, H. L., Smith, D. L., Visscher, P. T., and Burns, B. P. (2015). Niche differentiation of bacterial communities at a millimeter scale in Shark Bay microbial mats. Sci. Rep. 5, 1–17. Wortel, M. T., Noor, E., Ferris, M., Bruggeman, F. J., and Liebermeister, W. (2018). Metabolic enzyme cost explains variable trade-offs between microbial growth rate and yield. PLoS Comput. Biol. 14:e1006010. doi: 10.1371/journal. pcbi.1006010 Zumft, W. G. (1997). Cell biology and molecular basis of denitrification. doi: 10.1128/.61.4.533-616. Microbiol. Mol. Biol. Rev. 61, 533–616. 1997 Zwietering, M. H., Jongenburger, I., Rombouts, F. M., and Van’t Riet, K. (1990). Modeling of the bacterial growth curve. Appl. Environ. Microbiol. 56, 1875– 1881. Conflict of Interest: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Publisher’s Note: All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. Copyright © 2021 Zhang, Mullen, Ciccarese, Dumit, Martocello, Toyofuku, Nomura, Smriga and Babbin. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms. Frontiers in Microbiology | www.frontiersin.org 13 September 2021 | Volume 12 | Article 711073
null
10.1038_s42003-021-02716-8.pdf
Data availability Several public databases were used in this study, including Immune Epitope Database and Analysis Resource (IEDB) (https://www.iedb.org/) for experimental measurements, UniProt (https://ftp.uniprot.org/pub/databases/uniprot/current_release/knowledgebase/ complete/uniprot_sprot.fasta.gz) for decoy peptides, and IPD-IMGT/HLA (https:// github.com/ANHIG/IMGTHLA/tree/3410) for MHC-I allele sequences. Research data files supporting this study, including the peptide-binding cleft sequence of MHC-I alleles; the training, validation, and benchmark datasets; the prediction of the validation and benchmark datasets; and the prediction of the allele expansion are available from Mendeley Data (https://doi.org/10.17632/c249p8gdzd.3)33. Source data for all figures are provided in Supplementary Data. Code availability The source code of the research and the MHCfovea’s predictor are freely available at GitHub (https://github.com/kohanlee1995/MHCfovea) and Mendeley Data33 for academic non-commercial research purposes. All source codes are based on Python (v3.6.9) and its packages, including numpy (v1.18.2), pandas (v1.0.3), scikit-learn (v0.22.2), pytorch (v1.4.0), matplotlib (v3.2.1), seaborn (v0.10.0), logomaker (v0.8). Numpy, pandas, and scikit-learn, are used for data analysis; pytorch is used for deep learning; matplotlib, seaborn, and logomaker are used for visualization. The website for the summarization of MHCfovea is available at https://mhcfovea.ailabs.tw.
Data availability Several public databases were used in this study, including Immune Epitope Database and Analysis Resource (IEDB) ( https://www.iedb.org/ ) for experimental measurements, UniProt ( https://ftp.uniprot.org/pub/databases/uniprot/current_release/knowledgebase/ complete/uniprot_sprot.fasta.gz ) for decoy peptides, and IPD-IMGT/HLA ( https:// github.com/ANHIG/IMGTHLA/tree/3410 ) for MHC-I allele sequences. Research data files supporting this study, including the peptide-binding cleft sequence of MHC-I alleles; the training, validation, and benchmark datasets; the prediction of the validation and benchmark datasets; and the prediction of the allele expansion are available from Mendeley Data ( https://doi.org/10.17632/c249p8gdzd.3 ) 33 . Source data for all figures are provided in Supplementary Data. Code availability The source code of the research and the MHCfovea's predictor are freely available at GitHub ( https://github.com/kohanlee1995/MHCfovea ) and Mendeley Data 33 for academic non-commercial research purposes. All source codes are based on Python (v3.6.9) and its packages, including numpy (v1.18.2), pandas (v1.0.3), scikit-learn (v0.22.2), pytorch (v1.4.0), matplotlib (v3.2.1), seaborn (v0.10.0), logomaker (v0.8). Numpy, pandas, and scikit-learn, are used for data analysis; pytorch is used for deep learning; matplotlib, seaborn, and logomaker are used for visualization. The website for the summarization of MHCfovea is available at https://mhcfovea.ailabs.tw .
ARTICLE https://doi.org/10.1038/s42003-021-02716-8 OPEN Connecting MHC-I-binding motifs with HLA alleles via deep learning Ko-Han Lee Chien-Yu Chen 1,6✉ 1, Yu-Chuan Chang1, Ting-Fu Chen1, Hsueh-Fen Juan 1,2,3,4, Huai-Kuang Tsai 1,5 & ; , : ) ( 0 9 8 7 6 5 4 3 2 1 The selection of peptides presented by MHC molecules is crucial for antigen discovery. Previously, several predictors have shown impressive performance on binding affinity. However, the decisive MHC residues and their relation to the selection of binding peptides are still unrevealed. Here, we connected HLA alleles with binding motifs via our deep learning-based framework, MHCfovea. MHCfovea expanded the knowledge of MHC-I- binding motifs from 150 to 13,008 alleles. After clustering N-terminal and C-terminal sub- motifs on both observed and unobserved alleles, MHCfovea calculated the hyper-motifs and the corresponding allele signatures on the important positions to disclose the relation between binding motifs and MHC-I sequences. MHCfovea delivered 32 pairs of hyper-motifs and allele signatures (HLA-A: 13, HLA-B: 12, and HLA-C: 7). The paired hyper-motifs and allele signatures disclosed the critical polymorphic residues that determine the binding preference, which are believed to be valuable for antigen discovery and vaccine design when allele specificity is concerned. 1 Taiwan AI Labs, Taipei 10351, Taiwan. 2 Graduate Institute of Biomedical Electronics and Bioinformatics, National Taiwan University, Taipei 10617, Taiwan. 3 Department of Life Science, National Taiwan University, Taipei 10617, Taiwan. 4 Center for Computational and Systems Biology, National Taiwan University, Taipei 10617, Taiwan. 5 Institute of Information Science, Academia Sinica, Taipei 11529, Taiwan. 6 Department of Biomechatronics Engineering, National Taiwan University, Taipei 10617, Taiwan. email: chienyuchen@ntu.edu.tw ✉ COMMUNICATIONS BIOLOGY | (2021) 4:1194 | https://doi.org/10.1038/s42003-021-02716-8 | www.nature.com/commsbio 1 ARTICLE COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-021-02716-8 for Antigens are essential the induction of adaptive immunity to respond to threats, such as infectious dis- eases or cancer1. Most antigens are short non-self-pep- tides; however, not all peptides are antigenic1. Researchers have been committed to the development of peptide-based vaccines to prevent or treat numerous diseases2–5. For instance, tumor neoantigens, derived from proteins with nonsynonymous somatic mutations, may be suitable cancer therapeutic vaccines6–8. In order to choose good antigens, it is important to understand the process of antigen presentation. Major histocompatibility complex class I (MHC-I) molecules are cell surface proteins essential for antigen presentation1. MHC- I encoded by three gene loci (HLA-A, -B, and -C) are composed of a polymorphic heavy α-chain and an invariant β-2 micro- globulin light chain9. The α1- and α2-domains form the peptide- binding cleft, a highly polymorphic region, contributing to the diversity of MHC-I-binding motifs9. There are >13,000 MHC-I alleles on a four-digit level (e.g., A*02:01) recorded in the IPD- IMGT/HLA database10, a particular protein representing sequence. Thus, it is difficult to select antigens from numerous peptides for each MHC allele via experiments. In order to facilitate the process of antigen discovery, several predictors have been developed and shown accurate performance on MHC-I–peptide binding affinity11,12. Owing to the similarity of polymorphic regions in MHC-I alleles, researchers tended to build a single pan-allele predictor rather than numerous allele- specific predictors13; of note, a pan-allele predictor takes both MHC-I and peptide sequences as the input. A pan-allele predictor is thought to disclose the connection among different alleles via the consensus pattern in polymorphic regions13. Nevertheless, the relation between MHC-I sequences and their binding motifs is still unspecified. In the past years, a few studies have discussed the similarity between MHC-I-binding motifs14–16. Some key residues of MHC-I molecules determine the binding motifs that can be clustered into several groups14; the types of key residues within allele clusters and motif clusters are consistent to some extent15. In addition, the similarity between binding motifs can be used to improve the performance of binding prediction16. However, it is difficult to specify the key residues of each motif group from the limited number of alleles with experimental measurements. In this regard, we developed a deep learning-based framework, MHCfovea, that incorporates supervised binding prediction with unsupervised summarization to connect important residues to binding preference. As exemplified in Fig. 1, this study explored the binding potential of billions of peptide–allele pairs via the prediction module; only qualified binding pairs were sent to the summarization module to infer the relation between binding motifs and MHC-I sequences. In the end, the resultant pairs of hyper-motifs and allele signatures can be easily queried through a web interface (https://mhcfovea.ailabs.tw). Results Overview of MHCfovea. MHCfovea integrates a supervised prediction module and an unsupervised summarization module to connect important residues to binding motifs (Fig. 1). The predictor in the prediction module is constructed of an ensemble model based on convolutional neural networks (CNN) (Supplementary Fig. 1) embedded with ScoreCAM17, a class activation mapping (CAM)-based18 approach, to highlight the important positions of the input MHC-I sequences. As for the summarization module, to infer the relation between the important residues and the binding motifs, we made predictions on unobserved alleles to expand our knowledge from 150 to 13,008 alleles followed by clustering all N- and C-terminal binding motifs, respectively. Then the corresponding signatures of MHC-I sequences on the important positions were generated to reveal the relation between MHC-I sequences and their binding motifs. In the following subsections, we first demon- strate the performance of MHCfovea’s predictor using 150 alleles with experimental data. Second, we introduce the important positions highlighted by ScoreCAM embedded in MHCfovea’s predictor. Finally, we present the summarization results on 13,008 alleles in the groups of HLA-A, -B, and -C, respectively. Additionally, alleles from the same HLA group but falling into different clusters are identified to disclose the critical residues that determine the binding preference beyond the HLA groups. Performance evaluation of MHCfovea’s predictor. The pre- dictor of MHCfovea takes an MHC-I-binding cleft sequence with 182 amino acids (a.a.) and a peptide sequence with 8–15 a.a.19 to predict the binding probability. We trained the predictor using 150 alleles with either binding assay data or ligand elution data and then tested it on an independent ligand elution dataset built by Sarkizova et al.15. We adopted a large number of in silico decoy peptides in parallel with in vivo free peptides (not present on MHC-I molecules) to train and test the predictor; of note, we took NetMHCpan4.120 as a reference to set the ratio of decoy peptides to eluted peptides (decoy-eluted ratio (D-E ratio)) at 30 in the benchmark (testing) dataset. The data sources used are characterized in Supplementary Table 1 and Supplementary Data 1. The number of decoy peptides is notably higher than that of eluted peptides, meaning that MHC-I–peptide binding prediction is an extremely imbalanced classification process. the imbalance among classes is a common issue in machine learning, and some methods have been developed to deal with it21. In MHCfovea, we used the ensemble strategy with downsampling22–24 to resolve such an imbalanced learning task (Fig. 2a). In fact, Next, to evaluate the effect of the D-E ratio in the overall training dataset (denoted as A in Fig. 2a) and the D-E ratio in each downsized dataset (denoted as B in Fig. 2a), we trained models with five different D-E ratios (B = 1, 5, 10, 15, and 30) in each downsized dataset and three different D-E ratios (A = 30, 60, and 90) in the training dataset. Of note, all experimental data were shared in each downsized dataset, and the decoys were non- overlapping between each downsized dataset to make sure all the decoys were used in the ensemble model eventually. Figure 2b depicts the performance of the validation dataset (Supplementary Tables 2 and 3). The best model was with D-E ratios of B = 5 and A = 90, showing an average precision (AP) of 0.898 and an area under the receiver operating characteristic (ROC) curve (AUC) of 0.991. Therefore, we used the ensemble model with 18 (=90/5) CNN models (the best performance on the validation dataset) as the predictor of MHCfovea. To compare MHCfovea’s predictor with other well-known including NetMHCpan4.120, MHCflurry2.025, and predictors, MixMHCpred2.116, we adopted an independent benchmark dataset from NetMHCpan4.1. Even though the testing data (benchmark) are the same in the comparison of this study, the training data of different predictors are not consistent. Supplementary Table 4 and Supplementary Fig. 2 summarized the training dataset used by each predictor. Both of NetMHCpan4.1 or MHCflurry2.0 used more alleles and a larger number of experimental measurements (positive peptides) than MHCfovea. To be specific, only one peptide is unique for MixMHCpred2.1, MHCfovea as well as the other two predictors used more alleles and peptides than it owing to the paucity of public data when MixMHCpred2.1 was published. (Supplementary Fig. in MHCfovea 2b). As 2 COMMUNICATIONS BIOLOGY | (2021) 4:1194 | https://doi.org/10.1038/s42003-021-02716-8 | www.nature.com/commsbio COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-021-02716-8 ARTICLE Fig. 1 An overview of MHCfovea. MHCfovea, a deep learning-based framework, contains a prediction module and a summarization module that infers the relation between MHC-I sequences and peptide-binding motifs. First, the predictor, an ensemble model of multiple convolutional neural networks (CNN models), was trained on 150 observed alleles. In the predictor, 42 important positions were highlighted from MHC-I sequence (182 a.a.) using ScoreCAM. Next, we made predictions on 150 observed alleles and 12,858 unobserved alleles against a peptide dataset (number: 254,742) and extracted positive predictions (score >0.9) to generate the binding motif of an allele. Then, after clustering the N-terminal and C-terminal sub-motifs, we built hyper-motifs and the corresponding allele signatures based on 42 important positions to reveal the relation between binding motifs and MHC-I sequences. On the benchmark dataset, MHCfovea showed an AUC of 0.977 (Fig. 2c and Supplementary Table 5) and an AP of 0.841 (Supplementary Fig. 3a and Supplementary Table 5), both better than those obtained with the other predictors. The primitive output of MHCfovea is the estimated probability of allele–peptide binding. For the threshold of predicting an input pair as positive, setting a threshold at 0.68 reaches a maximal F1 score of 0.837 on the validation dataset. This threshold is suggested when adopting MHCfovea as a binary predictor. Apart from the whole bench- mark dataset, we also evaluated the performance on every allele. MHCfovea showed a median AUC value of 0.984. For 82 of the 92 (89%) alleles, the AUC is at least 0.95. MHCfovea performed significantly better than the other predictors with respect to the AUC and AP metrics (Fig. 2d, Supplementary Fig. 3c, and Supplementary Data 2). Next, the performance of our pan-allele model was carefully examined in the context of 16 unobserved alleles (with no listed in experimental measurements in the training dataset), there is no significant Supplementary Table 6. Importantly, difference between the AUC and AP of unobserved alleles and of the observed alleles (Fig. 2e, Supplementary Fig. 3e, and Supplementary Data 3), suggesting that MHCfovea shows good performance not only toward alleles present in the training data but also in the context of unobserved alleles. Furthermore, when compared with other predictors on the ten commonly unobserved alleles across all the predictors, listed in Supplementary Table 6, MHCfovea also has slightly better performance (Supplementary Fig. 3f, g and Supplementary Data 3). The high similarity of sequences between alleles in the same HLA group was regarded as a reason for the good performance on unobserved alleles. Nevertheless, B*55:02 is an unobserved allele with an AUC of 0.993, while no alleles in the group B*55 are present in the training dataset, giving an example of MHCfovea’s good accuracy on the alleles of a rarely observed HLA group. To further evaluate the reliability of the MHCfovea’s predictor on unseen peptides, we took the sets of similar and dissimilar peptides in the benchmark dataset into consideration, where similar peptides denote a peptide in the testing data is identical or near-identical (one peptide is another peptide’s substring) to any peptides in the training or validation data. Because most experimental data were conducted on normal human cells, it is possible to have identical or near-identical peptides in the benchmark dataset even when we require that no identical allele–peptide pairs are present in the benchmark and training (or validation) data simultaneously. Finally, benchmark data were partitioned into four groups (1) unobserved alleles paired with dissimilar peptides; (2) unobserved alleles paired with similar peptides; (3) observed alleles paired with dissimilar peptides; and (4) observed alleles paired with similar peptides. Figure 2f and Supplementary Fig. 3h (Supplementary Data 4) provide the results on the metrics of AUC and AP, respectively. For each group, MHCfovea outperformed the other predictors in the respect of AUC and has better AP than MHCflurry2.0 and MixMHCpred2.1. Undeniably, similar peptides have better performance than dissimilar peptides in MHCfovea, and this phenomenon did not appear in other predictors because the definition of similar and dissimilar peptides might not applicable on them because the training data of each predictor (Supplementary Table 7): COMMUNICATIONS BIOLOGY | (2021) 4:1194 | https://doi.org/10.1038/s42003-021-02716-8 | www.nature.com/commsbio 3 ARTICLE COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-021-02716-8 Fig. 2 The framework and performance of the MHCfovea’s predictor. a The ensemble framework with the partitioning strategy. We first adopted the training dataset with a decoy–eluted ratio (D-E ratio) of A. The decoy dataset was partitioned into A/B downsized decoy datasets with D-E ratio of B. Then A/B CNN models were trained on one downsized decoy dataset along with the experimental dataset. Finally, the mean of results was calculated as the prediction score. b AP and AUC scores on the validation dataset of the ensemble model trained under different D-E ratios in the overall training dataset, including A = 30, 60, and 90, against different D-E ratios in the downsized decoy dataset, including B = 1, 5, 10, 15, and 30. The x-axis represents the D-E ratio in the training dataset, and the y-axis represents the metric score. Source data are provided in Supplementary Tables 2 and 3. c–f The following performances are all applied on the benchmark dataset. c The ROC curves with AUC depict the comparison between predictors. d The violin plot shows the distribution of AUC of each predictor by alleles (n = 91, one allele was removed because it is unavailable in MixMHCpred2.1). e Comparison of the AUC between observed (n = 76) and unobserved (n = 16) alleles. f The comparison of AUC on the four groups split from the benchmark dataset between predictors. In violin plots, boxplots depict the median value with a white dot, the 75th and 25th percentile upper and lower hinges, respectively, and whiskers with 1.5× interquartile ranges. P values (two-tailed independent t test) are shown as **P ≤ 0.01 and ****P ≤ 0.0001. Source data and details of the statistical analysis are provided in Supplementary Data 2, 3, and 4. are different. It is reasonable for a machine learning task to have better performance on the groups of similar peptides than of dissimilar ones. Of note, MHCfovea still has better performance than the other predictors on the dissimilar groups. Selection of important MHC-I residues. The MHC-I-binding cleft is a sequence of 182 a.a., some of which occupy highly polymorphic sites considered as decisive for epitope binding. investigated the Therefore, we important positions using ScoreCAM17, a kind of CAM algorithm. First, we applied Scor- eCAM on positive peptides to illustrate how ScoreCAM works, since it has been widely considered that the second and last residues of peptides are anchor positions for most alleles26. Fig- ure 3a (Supplementary Data 5) depicts that the anchor positions have higher mask scores than other residues, which reveals that ScoreCAM is capable of highlighting important positions in the peptide sequences. 4 COMMUNICATIONS BIOLOGY | (2021) 4:1194 | https://doi.org/10.1038/s42003-021-02716-8 | www.nature.com/commsbio COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-021-02716-8 ARTICLE Fig. 3 Selection of the important positions. a A clustering heatmap of the peptide mask on each peptide position of each allele. b A stack plot of the position importance of HLA genes at each MHC-I residue and a heatmap of allele masks derived from ScoreCAM results with clustering on alleles. These two plots are aligned by MHC-I-binding cleft sequences, to better demonstrate the distribution of mask scores. In the stack plot, different HLA genes were counted independently due to the number of alleles with variation as well as the divergent patterns of conserved or polymorphic sequences (Supplementary Fig. 4). As for the heatmap clustering in a, b, we used Euclidean distance and unweighted average linkage for clustering mask scores, and the row color is used to label the HLA gene. c A scatterplot with linear correlation shows the relationship between polymorphism and importance of each polymorphic MHC- I residue (n = 80). Information entropy (−ΣP × ln(P), where P is the amino acid frequency) is used to represent the degree of polymorphism. The important positions selected using ScoreCAM are colored in red, and the 34 residues derived from NetMHCpan4.1 are cross-marked. The blue band represents the 95% confidence interval of the regression fit, and the line represents the estimated regression. d A Venn diagram shows the intersection of the important position set from each HLA gene and the polymorphic residue sets. Residues in the set of “(A ∪ B ∪ C) ∩ polymorphism” are selected as the 42 important positions of MHCfovea. Source data are provided in Supplementary Data 5 and 6. Next, we focused on the positive predictions of the training dataset and obtained allele masks; briefly, every position has a mask score representing the relative importance across the 182 a.a. Figure 3b (Supplementary Data 5 and 6) shows the stack plot importance of each HLA gene at each position and the of heatmap clustering of allele masks. The importance of each position was quantified by the proportion of alleles with a mask score of >0.4. Importantly, alleles from identical HLA genes were mostly grouped together in the heatmap, consistent with the divergence of importance between different HLA genes in the stack plot. This result indicates that our model not only learned the differences between HLA-A, -B, and -C but also focused on different positions in different HLA genes. Additionally, to evaluate the consistency of polymorphism and mask score of each position, we applied linear regression analysis on the degree of polymorphism and importance. The degree of polymorphism was calculated by the information entropy of a.a. frequency. Owing to the divergence between HLA genes in Fig. 3b, the importance scores of HLA-A, -B, and -C were calculated separately, and the maximum one was chosen as the final importance. The activation maps derived from CAM-based approaches are not sharp enough; residues next to the real important residue could be highlighted simultaneously. This explains why some non-polymorphic positions also have high importance; therefore, before applying linear regression, we removed all non-polymorphic positions. Figure 3c (Supplemen- tary Data 6) presents a Pearson’s correlation of 0.67 (P < 0.05) between polymorphism and importance and reveals that highly polymorphic sites play a more important role in the predictor. Polymorphic positions with importance >0.4 were chosen as important positions. Figure 3d presents the Venn diagram of position selection. In the end, 42 important positions were COMMUNICATIONS BIOLOGY | (2021) 4:1194 | https://doi.org/10.1038/s42003-021-02716-8 | www.nature.com/commsbio 5 ARTICLE COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-021-02716-8 Fig. 4 The relation between MHC-I sequences and MHC-I-binding motifs. A summarization table of HLA-B. The MHC-I-binding motifs are divided into N- and C-terminal sub-motifs; sub-motifs are clustered by agglomerative hierarchical clustering. Hyper-motifs and the corresponding allele signatures are calculated from each sub-motif cluster. In each cluster, the number of alleles and the HLA groups with the number of alleles ≥25 are recorded in the last two columns. Source data are provided in Supplementary Data 7. selected, and 13 of them were important in all HLA genes (Supplementary Data 6). (the residues pseudo-sequence We compared the selected residues (42 residues) with 34 contact in NetMHCpan4.1)20 in Fig. 3c. Some highly polymorphic sites are not included in the pseudo-sequence but have high importance, suggesting that some residues other than the 34 contact residues are essential for epitope binding, such as position 65 and 71. applied Expansion and summarization of MHC-I-binding motifs. Each MHC-I allele has its own binding motif owing to the distinct MHC-I sequence. To further explore the pattern among different alleles, we computed the binding motif of alleles in the training dataset. Since the length of epitopes ranges from 8 to 15 and the important residues are usually located at the second and last positions, we focused on the first four (N-terminal) and last four (C-terminal) residues to construct an 8-a.a.-long motif for pep- tides bound by each allele26. Supplementary Fig. 5 depicts the hierarchical clustering of the binding motifs of HLA-B alleles. Some alleles, especially those of the identical HLA group (e.g., B*44), have similar binding motifs and are grouped together; however, some alleles with similar N-terminal sub-motifs have dissimilar C-terminal sub-motifs. For example, both HLA- B*40:01 and HLA-B*41:01 have an E-dominant N-terminal sub- motif, but the former has an L-dominant C-terminal sub-motif and the latter has an A-dominant one. This motivated MHCfovea to cluster the N-terminal and C-terminal sub-motifs separately. When exploring the relation between HLA sequences and MHC-I-binding motifs/sub-motifs, we noticed that the number of alleles in a cluster is too small to form meaningful signatures. The training dataset has only 150 alleles, a fraction of the 13,008 MHC-I alleles recorded in the IPD-IMGT/HLA database10; it is difficult to obtain notable MHC-I sequence patterns from such an insufficient number of alleles. Therefore, we made predictions on all available alleles to generate more binding motifs, relying on the good performance of the MHCfovea’s predictor. In total, we obtained 4158 HLA-A-binding motifs, 4985 HLA-B-binding motifs, and 3865 HLA-C-binding motifs. We then retrieved N- and C-terminal sub-motifs and clustered them into several clusters. Figure 4 (Supplementary Data 7) shows the clustering of N- and C-terminal sub-motifs of all HLA- B alleles, with 7 N-terminal and 5 C-terminal sub-motif clusters where minor clusters that have <50 alleles are neglected. For each sub-motif cluster, we calculated the hyper-motif and the corresponding allele signature to represent the preference of binding motifs and a.a. at the important positions (Fig. 4, Supplementary Figs. 6 and 7, and Supplementary Data 7). Of in each cluster, 50 alleles from each HLA group were note, 6 COMMUNICATIONS BIOLOGY | (2021) 4:1194 | https://doi.org/10.1038/s42003-021-02716-8 | www.nature.com/commsbio COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-021-02716-8 ARTICLE Fig. 5 The combination map of N- and C-terminal hyper-motifs. a The binding motif of an allele is a combination of an N-terminal and a C-terminal hyper- motif. After allocating all the alleles into the combination map, the cell color is determined by log10(number of alleles in the cell). In each cell with an allele number >10, the maximal HLA group and HLA groups with an allele number ≥25 or with a proportion (the allele number in the cell to the overall number of an allele group) >0.1 are listed. b The relation of a combination to its hyper-motifs. Four combinations are used as an example to illustrate the consistent signatures across different cells in the same column or row. The header column and header row consist of two N-terminal and two C-terminal clusters, respectively. Then, alleles of a cell, the combination of the N-terminal (column) and C-terminal (row) clusters, are used to generate the corresponding hyper-motif and allele signature. The color boxes are used to highlight the similar part of allele signatures. Source data are provided in Supplementary Data 8. randomly sampled to construct the allele signature to reduce the imbalance between different HLA groups. Notably, the pattern of binding motifs and allele signatures are partly interpretable with the property of a.a. In Fig. 4, the first cluster of C-terminal hyper- motifs is composed of aromatic residues (e.g., Y and F), whereas the second and third clusters are composed of aliphatic a.a. (e.g., the fifth and sixth clusters of L, V, I, and A). Moreover, N-terminal hyper-motifs dominated by basic a.a. (H and R) with similar allele signatures, indicating that MHC-I–peptide binding depends on physicochemical properties to some extent. To investigate the distribution of allele groups with respect to the combinations of N- and C-terminal clusters, we plotted the combination heatmap in Fig. 5a (Supplementary Fig. 8 for HLA- A and -C and Supplementary Data 8), which in total has 35 combinations (7 N-terminus × 5 C-terminus) for HLA-B. Inter- estingly, five unobserved combinations, not present in the training dataset, were discovered by MHCfovea via the pattern learned from the observed combinations. In Fig. 5b, we presented four combinations of N- and C-terminal clusters. The noticeable residues of N- and C-terminal hyper-motifs are mostly located in the first half and last half part of allele signatures, respectively, which is consistent with the binding structure of MHC-I molecules27. For example, the E-dominant cluster has noticeable residues in the first half part of the allele signature; these residues are highly conserved in not only different combinations but also the cluster, which enhances confidence of the key residues highlighted in the allele signature. Disclosure of the HLA groups falling into multiple sub-motif clusters. Overall, alleles within the same HLA group were clus- tered into the same sub-motif cluster. However, Fig. 4 shows that some HLA groups, such as B*15 and B*56, fell into multiple sub- motif clusters. An HLA group is defined as a multi-cluster HLA group if its alleles fall into multiple clusters and the second large cluster contains the number of alleles ≥25 or the ratio to the total allele number of this group >0.1; MHCfovea identified 27 multi- cluster HLA groups, listed in Supplementary Table 8. Here we used the important positions and expanded alleles to further investigate the multi-cluster HLA groups. Figure 6a (Supplementary Data 9) shows that the difference in polymorph- ism between multi-cluster and mono-cluster HLA groups is COMMUNICATIONS BIOLOGY | (2021) 4:1194 | https://doi.org/10.1038/s42003-021-02716-8 | www.nature.com/commsbio 7 ARTICLE COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-021-02716-8 Fig. 6 Characteristics of the HLA groups falling into multiple sub-motif clusters. a Polymorphism on all 182 amino acids or the important positions of mono-cluster (group number = 44) or multi-cluster (group number = 27) HLA-groups. b AUC and AP of unobserved alleles grouped by mono-cluster (allele number = 7) or multi-cluster (allele number = 9) HLA-groups. c, d The hyper-motifs and highlighted allele signatures of the N-terminal sub-motif clusters of B*15 (c) and the C-terminal sub-motif clusters of B*56 (d). The box colored in gray is used to highlight the polymorphic sites. Boxplots depict the median value with a middle line, the 75th and 25th percentile upper and lower hinges, respectively, whiskers with 1.5× interquartile ranges, and points as outliers. P values (two-tailed independent t test) are shown as “ns” no significance and **P ≤ 0.01. Source data and details of the statistical analysis are provided in Supplementary Data 9 and 10. significant considering the important positions, but not all 182 a.a. Figure 6b (Supplementary Data 10) shows that MHCfovea has good performance with respect to unobserved alleles for both the mono- and multi-cluster HLA groups. Figure 6c, d demonstrate hyper-motifs and highlighted allele signatures of multi-cluster HLA groups. Figure 6c shows three major N-terminal sub-motif clusters of B*15; the gray box highlights the highly polymorphic sites, especially position 67, which may contribute to different MHC-I-binding motifs. Additionally, position 65 and 71, not selected in the pseudo-sequence of NetMHCpan4.1 (Fig. 3c), are highlighted in the second cluster of Fig. 6c, supporting that some important positions beyond 34 contact residues are also decisive for the binding motif. On the other hand, Fig. 6d shows three major C-terminal sub-motif clusters of B*56; in the B*56 HLA group, only B*56:01 was present in the training dataset, which reveals that another two clusters were discovered by MHCfovea after allele expansion. In summary, these results demonstrate some notable patterns of MHC-I sequences beyond HLA groups, corresponding to some specific sub-motifs. Discussion Antigen discovery is composed of two major steps, antigen pre- sentation and T cell recognition1; several researches have built especially antigen accurate presentation, predictors for MHC–peptide binding12. However, the decisive residues of MHC sequences for peptide binding are still unspecified. A few studies have explored the pattern of MHC sequences and peptides14–16; nevertheless, owing to the limited number of alleles with experimental measurements, it is hard to conclude the relation of MHC sequences and binding motifs from all MHC alleles. Here we developed MHCfovea for predicting binding prob- ability and providing the connection between MHC-I sequences and binding motifs. MHCfovea’s predictor outperformed the other predictors via an ensemble framework with downsampling to solve the data imbalance between decoy and eluted peptides. To focus on the important positions determining the binding motifs, MHCfovea selected 42 a.a. of MHC-I sequences based on 150 observed alleles using ScoreCAM. After expanding the knowledge from observed alleles to unobserved alleles (total number: 13,008), MHCfovea delivered 32 pairs (HLA-A: 13, HLA-B: 12, and HLA-C: 7) of hyper-motifs and allele signatures on 42 important positions to reveal the relation of MHC-I sequences and binding motifs. In addition, MHCfovea discovered some unobserved combinations of N- and C-terminal sub-motifs with the support from high similarity between allele signatures. Finally, MHCfovea disclosed some multi-cluster HLA groups, such as B*15 and B*56, and highlighted the key residues to determine the different binding motifs. 8 COMMUNICATIONS BIOLOGY | (2021) 4:1194 | https://doi.org/10.1038/s42003-021-02716-8 | www.nature.com/commsbio COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-021-02716-8 ARTICLE Since the positive allele–peptide pairs in the benchmark data have a high ratio of peptides (32.8%) that were present in the training data, it is not clear if the good performance of MHCfovea came from the memorization of the peptides in the positive pairs of the training data. To clarify this point, we built an artificial dataset by pairing all the alleles in the benchmark dataset and the positive peptides in the training dataset. Supplementary Fig. 9 depicts the distribution of the artificial dataset, which is close to the negative data in the benchmark. In other words, the artificial pairs are recognized as negative samples mostly in the MHCfo- vea’s predictor. This result indicated that the MHCfovea’s pre- dictor actually recognized the binding peptides via the sequence patterns rather than memorizing all the positive peptides in the training data. Some limitations of MHCfovea are addressed here. First, the unobserved binding motifs are derived from predictions. Although MHCfovea has an accurate performance in the context of unobserved alleles, the total number of alleles with experi- mental data is a small fraction of available MHC-I alleles. Second, sub-motifs with a dominant a.a. can be clustered notably. In contrast, sub-motifs of HLA-C mostly with no dominant a.a. have neither obvious clusters nor indistinguishable allele signatures; therefore, it is difficult to determine the relation between binding motifs and MHC-I sequences on such alleles. Additionally, the number of clusters is fixed once summarization is completed. In this study, some minor clusters with <50 alleles were neglected, and in the end 32 major clusters are presented in our summar- ization. Most alleles (12,919 in 13,008, 99%) belong to one N-terminal and one C-terminal cluster within these 32 clusters. If new alleles are appended in the future, the process of allele extension and summarization can be reperformed to generate a new set of clusters. As for the binding prediction, the testing dataset is the same for each predictor, but is not. Although the training dataset MHCfovea has no advantage on the numbers of alleles and peptides when compared with NetMHCpan4.1 or MHCflurry2.0 (Supplementary Table 4 and Supplementary Fig. 2), the lack of a public training dataset is still a limitation for comparison between different algorithms. Furthermore, MHCfovea is only trained on mono-allelic measurements; adding multi-allelic data to the training dataset increases not only the number of peptides but also the diversity of MHC-I alleles. Alvarez et al.28 designed a semi-supervised method to associate each ligand to its MHC-I allele, which can potentially deal with the ambiguous annotation on multi-allelic data. In the future, we will incorporate this method with MHCfovea to enlarge the number of observed alleles; we anticipate increasing the number of experimental data can further improve model performance and the quality of the summarization of MHCfovea. Furthermore, a complete immune response depends on the recognition of MHC-I–peptide com- plexes by T cells. Building a model for T cell immunogenicity following MHCfovea is expected to promote the contribution of computational approaches on antigen discovery. In summary, MHCfovea successfully connects MHC-I alleles with binding motifs via deep learning. MHCfovea’s predictor expanded the knowledge of MHC-I-binding motifs from 150 alleles to 13,008, which were further summarized into pairs of hyper-motifs and allele signatures. The large number of allele sequences realized the generalization of allele signatures con- nected to distinct binding motifs correspondingly. Antigen dis- covery and vaccine design can be facilitated by knowing such clustered alleles and their key residues. Additionally, MHCfovea reveals some multi-cluster HLA groups, which provided addi- tional examination for allele similarity beyond the allele group, based on the 42 important positions of MHC-I uncovered by MHCfovea. Methods Preparation of MHC-I sequences. We used the IPD-IMGT/HLA database (ver- sion 3.41.0)10 as a reference for MHC-I sequences and used peptide-binding clefts annotated in the UniProt database29 as the target binding region. Of note, the peptide-binding cleft, composed of α-1 and α-2 regions, is a protein sequence with 182 a.a. and is critical for epitope presentation9. We used the alignment file from the IPD-IMGT/HLA database and obtained the corresponding sequences to build a peptide-binding domain database of all MHC-I alleles for the development of the proposed pan-allele-binding predictor adopted by MHCfovea. Preparation of peptide data. Experimental data of binding and ligand elution assays, especially mass spectrometry (MS), were collected from Immune Epitope Database and Analysis Resource (IEDB)30, the most comprehensive immuno- peptidome database. Because MHCfovea is a binary classifier for MHC-I–peptide binding, all measurements were labeled with 0 and 1. For the binding assays, an IC50 of 500 nM was set as the upper bound for the positive label. As for ligand elution assay, all samples were labeled as positive. The binding assay dataset generated in 2013 was directly downloaded from IEDB. To focus on the prediction of four-digit human MHC-I alleles (for example, A*01:01), non-human, mutant, and digital-insufficient MHC-I alleles were excluded. The peptides were restricted to 8–15-mers and this setting covered most epitopes19. The MS dataset was exported from IEDB on 2020/07/01; the following filters were used: linear epitopes, human species, MHC class I, and positive MHC ligand assay. Both 4-digit human alleles and peptides with a length of 8–15 a.a. were selected, following the same selection strategy as above. After filtration, the dataset consisted of 515,110 measurements across 150 alleles. Separation of the training, validation, and benchmark datasets. To build an isolated testing benchmark, we considered a single experimental reference selected from the previous ligand elution assay dataset. The MHC-I immunopeptidome built by Sarkizova et al.15 is the largest mono-allelic MS dataset, comprising 127,371 measurements across 92 alleles and was, therefore, chosen as the testing benchmark in this study. The binding assay dataset and the MS dataset excluding the experimental data used in the benchmark were combined to build the training dataset (95%) and the validation dataset (5%). In addition, to avoid duplication between training and benchmark datasets, we excluded peptides with identical allele and peptide sequences from the training and validation datasets and retained them in the benchmark dataset. Preparation of decoy peptides. As the MS data only provide positive results, we prepared a decoy dataset to be used as negative results. We created two types of decoy peptides, “protein decoy” and “random decoy,” both extracted from the UniProt proteome. “Protein decoy” refers to the peptides that were generated from the same protein as an eluted peptide, whereas “random decoy” refers to the peptides that were randomly extracted from the UniProt proteome. For each eluted peptide in the benchmark and validation datasets, we created two protein decoy peptides and two random decoy peptides for each length of 8–15 (a.a.). Duplicated peptides with identical allele and peptide sequence were excluded. In the end, both benchmark and validation datasets had a D-E ratio of 30, which is close to that of the dataset in NetMHCpan4.120. On the other hand, in the training dataset, to evaluate the effect of D-E ratio on model performance, we generated decoy pep- tides with D-E ratios >30. For each eluted peptide, we created two protein decoy peptides and ten random decoy peptides for each length of 8–15 (a.a.). We only enlarged the number of random decoy peptides, because it was difficult to select more different unique peptides from a single protein (protein decoy peptides) with a short length. In the end, the training dataset had a D-E ratio of 90, which is three times that in the validation and benchmark datasets. The number of data instances in each dataset is listed in Supplementary Table 1, and the data number by alleles of the training, validation, and benchmark datasets is recorded in Supplementary Data 1. CNN model architecture. The predictor adopted by MHCfovea is an ensemble model of multiple CNN model. A CNN model takes the allele (182 a.a.) and peptide (8–15 a.a.) sequences as input; both sequences are encoded with a one-hot encoder of a.a. The CNN model architecture is shown in Supplementary Fig. 1. Before concatenation, the encoded vectors are passed through several convolution blocks separately. The convolution block is composed of a 1D convolution layer with kernel size 3, stride 1, and zero-padding 1; a batch normalization layer; a ReLU activation layer; as well as a max-pooling layer. In the allele part, sequences are passed through four convolution blocks and downsized to a 15-long matrix. In the peptide part, all sequences are padded with “X” as an unknown a.a. to 15-long at the end of the sequence, the maximal length of peptides, and three convolution blocks are applied. After concatenation in the dimension of filters, the matrix is passed through another two convolution blocks with the replacement of the last max-pooling layer by a global max-pooling layer followed by a fully connected layer and a sigmoid operator. Finally, a prediction score is obtained to represent the binding probability of MHC-I and peptide sequences. COMMUNICATIONS BIOLOGY | (2021) 4:1194 | https://doi.org/10.1038/s42003-021-02716-8 | www.nature.com/commsbio 9 ARTICLE COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-021-02716-8 Model training. MHCfovea uses binary cross entropy as its loss function and the Adam optimization algorithm with the weight decay of 10−4 as the optimizer. The number of training epochs was set to 30, and the best model state was chosen after epoch 24 via the loss of the validation dataset to avoid overfitting. The hyper- parameters, including the batch size and learning rate, were selected via the grid search optimizer based on AP of the validation dataset. The batch size of 32 from the options [16, 32, 64] and the learning rate of 10−4 from the options [10−5, 10−4, 10−3] were selected (Supplementary Table 9). In addition, the learning rate sche- duler was used to adjust the learning rate during the training process. Of note, the learning rate was reduced to 10−5 after epoch 15 and to 10−6 after epoch 24. Prediction of all alleles. With the good performance of MHCfovea’s predictor on unobserved alleles, we predicted the binding probability of each allele against 254,742 peptides (including all ligand elution data and some decoy peptides whose number was the same as ligand elution data of the benchmark dataset). In total, 3.3 billion pairs of peptide–allele were tested. The peptides with a prediction score >0.9 (~78 million peptides) were sent to the summarization module to calculate the binding motif for each allele. Each MHC-I-binding motif with 8 a.a. was composed of the first four residues (N-terminal) and the last four residues (C-terminal). In total, we obtained 4158 HLA-A-binding motifs, 4985 HLA-B-binding motifs, and 3865 HLA-C-binding motifs in the summarization step. Performance metrics. We used four metrics, AUC, AUC0.1, AP, and PPV, to evaluate the performance of our model as well as that of other predictors. The AUC is a curve of the true positive rate (TPR) against the false positive rate (FPR). AUC0.1 has a restriction of the FPR under 0.1. AP is the area under the precision- recall curve created by plotting the precision against TPR, also called recall. PPV (positive predictive value) is defined as Eq. (1), where N is the number of positive measurements. Sequence motifs. The sequence motif is the pattern of a set sequences. There are some types of matrices, including position probability matrix (PPM), position weight matrix, and information content matrix (ICM), used to represent the sequence motif. In this study, we used PPM to calculate the MHC-I sequence motif and ICM to calculate the MHC-I-binding motif. From a set S of M aligned sequences of length L, the elements of the PPM are calculated from Eq. (2), where I is an indicator function. PPV ¼ positive predictions within top N predictions N ð1Þ In addition, we calculated these metrics in the context of every allele to evaluate the distribution of allelic performance. Threshold and %rank for the prediction score. To explain the prediction score more explicitly, a positive threshold and the %rank score, the percentage of ranking among background peptides, were provided. The positive threshold was set according to the maximal F1 score on the validation dataset. As for the %rank, 10,000 random peptides extracted from the UniProt database were built as the background peptides to calculate the %rank of each prediction score. When a prediction receives a %rank of 0.5, it means a peptide binds to MHC-I more probably than 99.5% random peptides. Comparison with other predictors. NetMHCpan4.120, MHCflurry2.025, and MixMHCpred2.116, well-known MHC-I–peptide binding predictors, were com- pared with the MHCfovea’s predictor. For MHCflurry2.0, we used the variant model of MHCflurry2.0-BA, the only one trained without our benchmark dataset. Both NetMHCpan4.1 and MHCflurry2.0 are compatible with all kinds of a.a. and 8–15 length peptides; however, for MHCflurry2.0, we had to replace a.a. beyond 20 human-required a.a. with “X” as an unknown a.a. On the other hand, MixMHCpred2.1 only allows 8–14 length peptides and sequences within 20 a.a.; therefore, for accurate comparison, we removed peptides with other a.a. or those >14 a.a. First, we tested all models directly on the benchmark dataset and calculated performance metrics for comparison. The output of MHCflurry2.0 was the IC50 of the binding affinity; therefore, we used a function (1 − log50,000(x)) to transform the binding affinity into binding probability. Then the performances of these models were tested by allele to evaluate the confidence between different alleles. In total, there are two types of results because of the peptide availability of MixMHCpred2.1 depicted in Supplementary Data 2 and 4. Class activation mapping. We applied CAM on our model for interpretation purposes. CAM-based approaches provide the explanation for a single input with activation maps from a convolution layer. There are several CAM-based methods, including CAM18, GradCAM31, GradCAM++32, and ScoreCAM17. ScoreCAM was chosen due to its stability on the former convolution layer. We applied ScoreCAM on the second convolution block before the max-pooling layer of the MHC part (Supplementary Fig. 1). We focused on positive predictions with pre- diction scores >0.9. The mean of ScoreCAM scores from positive predictions of a single allele was calculated as the final result called “epitope mask” for epitope part and “allele mask” for allele part used in Fig. 3. Of note, in epitope and allele masks, every position has a score representing the relative importance across the 8-a.a.- long and 182-a.a.-long sequences, respectively. Selection of the important positions. The training dataset with 150 alleles composed of 46 HLA-A, 85 HLA-B, as well as 19 HLA-C alleles was used to select the important positions, and both allele masks and a.a. polymorphism were taken into consideration. Owing to the divergence between HLA genes, positions from different HLA genes were chosen separately. First, we calculated the importance of each position for each HLA gene. The importance of a position was quantified as the proportion of alleles with mask scores >0.4 (set heuristically). Then, for each HLA gene, residues with importance >0.4 (also set heuristically) were selected; however, those with no polymorphism (all alleles had the same a.a.) were dropped. Then we combined the selected positions from each HLA gene as important positions of our model. In total, we selected 42 important positions, including positions 1, 9, 11, 12, 24, 31, 32, 43, 44, 45, 62, 63, 65, 66, 67, 69, 70, 71, 73, 74, 76, 77, 79, 80, 94, 95, 97, 98, 109, 114, 116, 127, 131, 138, 142, 143, 144, 145, 152, 156, 163, and 180 of the MHC-I peptide-binding cleft sequence (182 a.a.). PPMi;j ¼ 1 M M ∑ k¼1 IðSk;j ¼ iÞ; i ¼ f20 amino acidsg j ¼ 1; :::; L ð2Þ The ICM is used to correct PPM with background frequencies and highlight more important residues. The elements of the ICM are calculated from Eq. (3), where the background frequency B is 0.05 (=1/20) for each a.a. 2 (cid:1) (cid:3) 3 ICMi;j ¼ PPMi;j ∑ g m2 20 amino acids f 4 PPMm;j ´ log2 PPMm;j B 5 ð3Þ Sub-motif clustering. An MHC-I-binding motif with 8 a.a. was split into an N-terminal sub-motif with the first 4 residues of the binding motif and a C-terminal sub-motif with the last 4 residues of the binding motif. Consequently, a sub-motif is represented by a 4 × 20 (the number of a.a.) ICM. Before clustering, the pairwise distance of each sub-motif was calculated via cosine metric. Then we used agglomerative hierarchical clustering with cosine metrics and maximum linkage to cluster the pairwise distance. Different numbers of clusters were set for different HLA genes and termini manually, and minor clusters with <50 alleles were neglected. Hyper-motifs and allele signatures. Hyper-motifs and allele signatures are both used to demonstrate the characteristics of a specific group of alleles. Hyper-motifs representing the MHC-I-binding motif of alleles were calculated from the element- wise mean of motif or sub-motif matrices. Allele signatures disclose the preference of a.a. at important positions. For each sub-motif cluster, we sampled 50 alleles from each HLA group on a two-digit level to balance the allele number of each group because of two reasons. First, alleles with the same HLA group have similar MHC-I sequences, which may lead to similar binding motifs (Supplementary Fig. 5). Second, there is a huge variation among the allele number of different HLA groups. For example, HLA-B*07 has 394 alleles, but HLA-B*56 only has 69 alleles. Of note, if the allele number of an HLA group was <50, all alleles were selected. Afterward, to generate the allele signature matrix (ASM), we had to calculate a background PPM (PPMbackground) from all sampled alleles of an HLA gene and a PPM of sampled alleles from a specific sub-motif cluster (PPMcluster). On the other hand, to evaluate the sequence pattern of HLA groups, we also calculated the PPM of alleles from an HLA group in a specific sub-motif cluster (PPMgroup). The ASMcluster was defined as the positive part of the difference between PPMcluster and PPMbackground in Eq. (4); the ASMgroup was defined as the difference between PPMgroup and PPMbackground in Eq. (5), where Iverson bracket was used to set positive elements and the others as 1 and 0, respectively. ASMcluster ¼ f þ PPMcluster (cid:2) PPMbackground (cid:4) (cid:4) (cid:5) (cid:5) ; f þðxÞ ¼ max f xð Þ; 0 (cid:4) (cid:5) ð4Þ ð5Þ ASMgroup ¼ g PPMgroup (cid:2) PPMbackground ; gðxÞ ¼ x > 0 (cid:3) For instance, we used 1790 sampled HLA-B alleles to generate the ½ PPMbackground of HLA-B and 502 sampled alleles of the P-dominant sub-motif cluster (Fig. 4) to produce the PPMcluster. The allele signature of the P-dominant N- terminal sub-motif in the header row of Fig. 4 was calculated from the difference of these two probability matrices. In addition, the highlighted allele signature demonstrated in Fig. 6c, d was used to highlight the similarity of allele signatures of the specific alleles and of the corresponding sub-motif clusters. We implemented the element-wise product to get the highlighted allele signatures in Eq. (6), where L is the sequence length and HASM is the matrix of the highlighted allele signature. (cid:4) HASMi;j ¼ ASMgroup (cid:4) ASMcluster (cid:5) ; i;j i ¼ f20 amino acidsg j ¼ 1; :::; L ð6Þ Statistics and reproducibility. For all comparisons of performance, we used two- tailed independent t test and set the criterion for statistical significance as P < 0.05. 10 COMMUNICATIONS BIOLOGY | (2021) 4:1194 | https://doi.org/10.1038/s42003-021-02716-8 | www.nature.com/commsbio COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-021-02716-8 ARTICLE For the relation between polymorphism and importance of each polymorphic MHC-I residue, we fit a linear regression along with 95% confidence interval (Fig. 3c). Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article. Data availability Several public databases were used in this study, including Immune Epitope Database and Analysis Resource (IEDB) (https://www.iedb.org/) for experimental measurements, UniProt (https://ftp.uniprot.org/pub/databases/uniprot/current_release/knowledgebase/ complete/uniprot_sprot.fasta.gz) for decoy peptides, and IPD-IMGT/HLA (https:// github.com/ANHIG/IMGTHLA/tree/3410) for MHC-I allele sequences. Research data files supporting this study, including the peptide-binding cleft sequence of MHC-I alleles; the training, validation, and benchmark datasets; the prediction of the validation and benchmark datasets; and the prediction of the allele expansion are available from Mendeley Data (https://doi.org/10.17632/c249p8gdzd.3)33. Source data for all figures are provided in Supplementary Data. Code availability The source code of the research and the MHCfovea’s predictor are freely available at GitHub (https://github.com/kohanlee1995/MHCfovea) and Mendeley Data33 for academic non-commercial research purposes. All source codes are based on Python (v3.6.9) and its packages, including numpy (v1.18.2), pandas (v1.0.3), scikit-learn (v0.22.2), pytorch (v1.4.0), matplotlib (v3.2.1), seaborn (v0.10.0), logomaker (v0.8). Numpy, pandas, and scikit-learn, are used for data analysis; pytorch is used for deep learning; matplotlib, seaborn, and logomaker are used for visualization. The website for the summarization of MHCfovea is available at https://mhcfovea.ailabs.tw. Received: 14 February 2021; Accepted: 24 September 2021; References 1. Blum, J. S., Wearsch, P. A. & Cresswell, P. Pathways of antigen processing. Annu. Rev. Immunol. 31, 443–473 (2013). Purcell, A. W., McCluskey, J. & Rossjohn, J. More than one reason to rethink the use of peptides in vaccine design. Nat. Rev. Drug Discov. 6, 404–414 (2007). Sette, A. & Rappuoli, R. Reverse vaccinology: developing vaccines in the era of genomics. Immunity 33, 530–541 (2010). Sahin, U. & Türeci, Ö. Personalized vaccines for cancer immunotherapy. Science 359, 1355–1360 (2018). 2. 3. 4. 5. Malonis, R. J., Lai, J. R. & Vergnolle, O. Peptide-based vaccines: current progress and future challenges. Chem. Rev. 120, 3210–3229 (2020). Schumacher, T. N. & Schreiber, R. D. Neoantigens in cancer immunotherapy. Science 348, 69–74 (2015). 6. 7. Keskin, D. B. et al. Neoantigen vaccine generates intratumoral T cell responses in phase Ib glioblastoma trial. Nature 565, 234–239 (2019). 8. Han, X.-J. et al. Progress in neoantigen targeted cancer immunotherapies. Front. Cell Dev. Biol. 8, 728 (2020). 9. Wieczorek, M. et al. Major histocompatibility complex (MHC) class I and MHC class II proteins: conformational plasticity in antigen presentation. Front. Immunol. 8, 292 (2017). 10. Robinson, J. et al. IPD-IMGT/HLA Database. Nucleic Acids Res. 48, D948–D955 (2020). 11. Finotello, F., Rieder, D., Hackl, H. & Trajanoski, Z. Next-generation computational tools for interrogating cancer immunity. Nat. Rev. Genet. 20, 724–746 (2019). 12. Mei, S. et al. A comprehensive review and performance evaluation of bioinformatics tools for HLA class I peptide-binding prediction. Brief. Bioinform. 21, 1119–1135 (2020). 13. Zhang, L., Udaka, K., Mamitsuka, H. & Zhu, S. Toward more accurate pan- specific MHC-peptide binding prediction: a review of current methods and tools. Brief. Bioinform. 13, 350–364 (2012). 14. Sidney, J., Peters, B., Frahm, N., Brander, C. & Sette, A. HLA class I supertypes: a revised and updated classification. BMC Immunol. 9, 1 (2008). 15. Sarkizova, S. et al. A large peptidome dataset improves HLA class I epitope prediction across most of the human population. Nat. Biotechnol. 38, 199–209 (2020). 16. Bassani-Sternberg, M. et al. Deciphering HLA-I motifs across HLA peptidomes improves neo-antigen predictions and identifies allostery regulating HLA specificity. PLoS Comput. Biol. 13, e1005725 (2017). 17. Wang, H. et al. Score-CAM: score-weighted visual explanations for convolutional neural networks. In 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) 111–119 (IEEE, 2020). 18. Zhou, B., Khosla, A., Lapedriza, A., Oliva, A. & Torralba, A. Learning deep features for discriminative localization. In 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2921–2929 (IEEE, 2016). 19. Trolle, T. et al. The length distribution of class I–restricted T cell epitopes is determined by both peptide supply and MHC allele–specific binding preference. J. Immunol. 196, 1480–1487 (2016). 20. Reynisson, B., Alvarez, B., Paul, S., Peters, B. & Nielsen, M. NetMHCpan-4.1 and NetMHCIIpan-4.0: improved predictions of MHC antigen presentation by concurrent motif deconvolution and integration of MS MHC eluted ligand data. Nucleic Acids Res. 48, 449–454 (2020). 21. He, H. & Garcia, E. A. Learning from imbalanced data. IEEE Trans. Knowl. Data Eng. 21, 1263–1284 (2009). 22. Liu, X.-Y., Wu, J. & Zhou, Z.-H. Exploratory undersampling for class- imbalance learning. IEEE Trans. Syst. Man Cybern. Part B 39, 539–550 (2009). 23. Schubach, M., Re, M., Robinson, P. N. & Valentini, G. Imbalance-Aware Machine Learning for Predicting Rare and Common Disease-Associated Non- Coding Variants. Sci. Rep. 7, 2959 (2017). 24. Chuang, K.-W. & Chen, C.-Y. Predicting pathogenic non-coding variants on imbalanced data set using cluster ensemble sampling. in 2019 IEEE 19th International Conference on Bioinformatics and Bioengineering (BIBE) 850–855 (IEEE, 2019). 25. O’Donnell, T. J., Rubinsteyn, A. & Laserson, U. MHCflurry 2.0: improved pan-allele prediction of MHC class I-presented peptides by incorporating antigen processing. Cell Syst. 11, 42–48.e7 (2020). 26. Bassani-Sternberg, M. & Gfeller, D. Unsupervised HLA peptidome deconvolution improves ligand prediction accuracy and predicts cooperative effects in peptide–HLA interactions. J. Immunol. 197, 2492–2499 (2016). 27. van Deutekom, H. W. M. & Keşmir, C. Zooming into the binding groove of HLA molecules: which positions and which substitutions change peptide binding most? Immunogenetics 67, 425–436 (2015). 28. Alvarez, B. et al. NNAlign-MA; MHC peptidome deconvolution for accurate MHC binding motif characterization and improved t-cell epitope predictions. Mol. Cell. Proteomics 18, 2459–2477 (2019). 29. Bateman, A. UniProt: a worldwide hub of protein knowledge. Nucleic Acids Res. 47, D506–D515 (2019). 30. Vita, R. et al. The Immune Epitope Database (IEDB): 2018 update. Nucleic Acids Res. 47, D339–D343 (2019). 31. Selvaraju, R. R. et al. Grad-CAM: visual explanations from deep networks via gradient-based localization. Int. J. Computer Vis. 128, 336–359 (2020). 32. Chattopadhay, A., Sarkar, A., Howlader, P. & Balasubramanian, V. N. Grad- CAM++: generalized gradient-based visual explanations for deep convolutional networks. in 2018 IEEE Winter Conference on Applications of Computer Vision (WACV) 839–847 (IEEE, 2018). 33. Lee, K.-H. et al. Data for: Connecting MHC-I-binding motifs with HLA alleles via deep learning. Mendeley Data https://doi.org/10.17632/c249p8gdzd.3 (2021). Acknowledgements We thank Tsung-Ting Hsieh and Hung-Ching Chang from Taiwan AI Labs for pro- viding computational and mathematical advices. We acknowledge support from the Ministry of Science and Technology, Taiwan (MOST 109-2221-e-002-161-MY3). Author contributions K.-H.L., Y.-C.C. and C.-Y.C. designed the study. K.-H.L. prepared and analyzed the data; developed, validated, and interpreted the predictor; summarized the predicted results; and wrote the manuscript. Y.-C.C. designed the figures of the MHCfovea overview and CNN model framework. T.-F.C. built the website. H.-F.J., H.-K.T. and C.-Y.C. revised the manuscript. Competing interests The authors declare no competing interests. Additional information Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s42003-021-02716-8. Correspondence and requests for materials should be addressed to Chien-Yu Chen. Peer review information Communications Biology thanks Claudio Mirabello and the other anonymous reviewer(s) for their contribution to the peer review of this work. Primary handling editor: George Inglis. Peer reviewer reports are available. This article has been peer reviewed as part of Springer Nature’s Guided Open Access initiative. Reprints and permission information is available at http://www.nature.com/reprints Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. COMMUNICATIONS BIOLOGY | (2021) 4:1194 | https://doi.org/10.1038/s42003-021-02716-8 | www.nature.com/commsbio 11 ARTICLE COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-021-02716-8 Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/ licenses/by/4.0/. © The Author(s) 2021 12 COMMUNICATIONS BIOLOGY | (2021) 4:1194 | https://doi.org/10.1038/s42003-021-02716-8 | www.nature.com/commsbio
null